commit 8a731e2eb4586ae50681732ee3da3a9d5df50ec2 Author: Fabio Micheluz Date: Thu Oct 1 12:43:56 2026 +0200 Initial validated RK3588 OpenCL face recognition engine diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..ad1ab0d --- /dev/null +++ b/.gitignore @@ -0,0 +1,29 @@ +# Build output +/build/ +/bin/ + +# Runtime libraries and models +/lib/*.so +/lib/*.so.* +/models/*.rknn + +# Local backups +*.before_* +*.before-* +*.bak +*~ + +# Runtime output +*.tmp +*.stderr +*.log + +# Generated result files +*-cpu-raw.json +*-opencl-raw.json +*-cpp-cpu.json +*-cpp-opencl.json + +# Editors +.vscode/ +.idea/ diff --git a/Makefile b/Makefile new file mode 100644 index 0000000..de64934 --- /dev/null +++ b/Makefile @@ -0,0 +1,99 @@ +CXX ?= g++ + +ROOT := $(abspath .) + +CXXFLAGS := \ + -std=c++17 \ + -O2 \ + -Wall \ + -Wextra \ + -Wpedantic \ + -Iinclude \ + -DFACE_CLUSTERING_ROOT=\"$(ROOT)\" + +LDFLAGS := \ + -L$(ROOT)/lib \ + -Wl,-rpath,'$$ORIGIN/../lib' + +LDLIBS := \ + -lrknnrt \ + -ldl \ + -lpthread + +BIN_DIR := bin +BUILD_DIR := build + +ENGINE_SOURCE := src/face_engine.cc +ENGINE_OBJECT := $(BUILD_DIR)/face_engine.o + +FACE_RECOGNITION_SOURCE := src/face_recognition.cc +FACE_RECOGNITION_OBJECT := $(BUILD_DIR)/face_recognition.o +FACE_RECOGNITION_BINARY := $(BIN_DIR)/face_recognition + +FACE_SCAN_SOURCE := src/face_scan.cc +FACE_SCAN_OBJECT := $(BUILD_DIR)/face_scan.o +FACE_SCAN_BINARY := $(BIN_DIR)/face_scan + +.PHONY: all clean test info + +all: \ + $(FACE_RECOGNITION_BINARY) \ + $(FACE_SCAN_BINARY) + +$(BUILD_DIR): + mkdir -p $(BUILD_DIR) + +$(BIN_DIR): + mkdir -p $(BIN_DIR) + +$(ENGINE_OBJECT): $(ENGINE_SOURCE) include/face_engine.h | $(BUILD_DIR) + $(CXX) $(CXXFLAGS) -c $< -o $@ + +$(FACE_RECOGNITION_OBJECT): \ + $(FACE_RECOGNITION_SOURCE) \ + include/face_engine.h \ + include/face_config.h \ + | $(BUILD_DIR) + $(CXX) $(CXXFLAGS) -c $< -o $@ + +$(FACE_SCAN_OBJECT): \ + $(FACE_SCAN_SOURCE) \ + include/face_engine.h \ + include/face_config.h \ + | $(BUILD_DIR) + $(CXX) $(CXXFLAGS) -c $< -o $@ + +$(FACE_RECOGNITION_BINARY): \ + $(ENGINE_OBJECT) \ + $(FACE_RECOGNITION_OBJECT) \ + | $(BIN_DIR) + $(CXX) $^ $(LDFLAGS) $(LDLIBS) -o $@ + +$(FACE_SCAN_BINARY): \ + $(ENGINE_OBJECT) \ + $(FACE_SCAN_OBJECT) \ + | $(BIN_DIR) + $(CXX) $^ $(LDFLAGS) $(LDLIBS) -o $@ + +test: all + ./bin/face_recognition \ + test/test.jpg \ + test/test3f.jpg + ./bin/face_scan \ + test/test3f.jpg \ + > /tmp/face_scan_test.json + python3 -m json.tool \ + /tmp/face_scan_test.json \ + > /dev/null + @echo "Tests completed successfully" + +info: + @echo "ROOT=$(ROOT)" + @echo "CXX=$(CXX)" + @echo "CXXFLAGS=$(CXXFLAGS)" + @echo "LDFLAGS=$(LDFLAGS)" + +clean: + rm -rf $(BUILD_DIR) + rm -f $(FACE_RECOGNITION_BINARY) + rm -f $(FACE_SCAN_BINARY) diff --git a/README.md b/README.md new file mode 100644 index 0000000..82b7ddf --- /dev/null +++ b/README.md @@ -0,0 +1,481 @@ +# Face Recognition RK3588 OpenCL + +Motore C++ per rilevamento facciale, generazione di embedding e riconoscimento kNN accelerato su Rockchip RK3588. + +Il progetto usa: + +- NPU tramite RKNN per SCRFD e SFace; +- GPU Mali-G610 tramite OpenCL per kNN; +- CPU C++ come riferimento numerico e fallback. + +## Funzionalità + +- rilevamento volti con SCRFD; +- confidence, bounding box e cinque landmark; +- non-maximum suppression; +- allineamento volto 112 x 112; +- embedding SFace float32 a 128 dimensioni; +- normalizzazione L2; +- cosine similarity tramite prodotto scalare; +- kNN CPU e OpenCL; +- conteggi dinamici, TOPK configurabile e chunking; +- leave-one-out tramite self-map; +- JSON su stdout e diagnostica su stderr. + +## Architettura + +```text +Immagine + | + v +SCRFD tramite RKNN / NPU + | + v +Bounding box e landmark + | + v +Allineamento volto 112 x 112 + | + v +SFace tramite RKNN / NPU + | + v +Embedding float32[128] normalizzato L2 + | + +-----------------------------+ + | | + v v +kNN CPU C++ kNN OpenCL / Mali-G610 + | | + +-------------+---------------+ + | + v + TOPK prototype + | + v + Leave-one-out opzionale + | + v + Soglia di riconoscimento +``` + +## Formato degli embedding + +```text +Dimensione: 128 +Tipo: float32 +Dimensione binaria: 512 byte +Ordine byte: little-endian +Normalizzazione: L2 +Norma attesa: circa 1.0 +``` + +Poiché gli embedding sono normalizzati: + +```text +cosine_similarity(a, b) = dot(a, b) +``` + +## Componenti + +### Face engine RKNN + +```text +include/face_config.h +include/face_engine.h +include/rknn_api.h +src/face_engine.cc +src/face_recognition.cc +src/face_scan.cc +``` + +### Motore kNN + +```text +include/knn_opencl.h +src/knn_opencl.cc +src/knn_opencl_backend.cc +src/face_knn.cc +kernels/knn_float4_topk.cl +kernels/knn_reference.cl +``` + +Il kernel produttivo viene compilato a runtime, ad esempio con: + +```text +-cl-fast-relaxed-math +-D DIM=128 +-D TOPK=5 +-D QUERY_COUNT=21 +-D LANES_PER_QUERY=8 +``` + +## Requisiti + +- Linux AArch64; +- Rockchip RK3588 o piattaforma compatibile; +- compilatore C++17 e GNU Make; +- runtime RKNN; +- OpenCL ICD loader e driver Mali; +- modelli SCRFD e SFace in formato RKNN. + +Su Debian: + +```bash +sudo apt-get install \ + build-essential \ + opencl-headers \ + ocl-icd-opencl-dev +``` + +Verifica OpenCL: + +```bash +clinfo 2>/dev/null \ + | grep -E 'Platform Name|Device Name|Device Type|Device Version' +``` + +## File locali richiesti + +Non sono versionati: + +```text +models/SCRFD_500M_KPS_640.rknn +models/face_recognition_sface_2021dec.rknn +lib/librknnrt.so +``` + +## Compilazione + +Dalla root del repository: + +```bash +make clean +make -j"$(nproc)" +``` + +Binari previsti: + +```text +bin/face_recognition +bin/face_scan +bin/face_knn +``` + +Verifica del binario OpenCL: + +```bash +file bin/face_knn +ldd bin/face_knn | grep -Ei 'opencl|stdc|not found' +``` + +## Utilizzo dei binari + +### face_recognition + +Programma dimostrativo con output destinato alla lettura umana. + +Analisi di una singola fotografia: + +```bash +./bin/face_recognition test/test.jpg +``` + +Confronto tra due fotografie: + +```bash +./bin/face_recognition \ + test/test.jpg \ + test/test3f.jpg +``` + +Usarlo per verificare modelli e runtime RKNN, detection, landmark, embedding e similarità. Non è il binario consigliato per integrazioni applicative. + +### face_scan + +Elabora una fotografia e produce JSON con immagine, volti, confidence, bounding box, landmark ed embedding. + +```bash +./bin/face_scan test/test3f.jpg +``` + +Separazione di risultato e diagnostica: + +```bash +./bin/face_scan test/test3f.jpg \ + > /tmp/face-scan.json \ + 2> /tmp/face-scan.stderr +``` + +Validazione: + +```bash +python3 -m json.tool /tmp/face-scan.json > /dev/null \ + && echo "OK: JSON valido" +``` + +Exit code: + +```text +0 scansione completata +2 argomenti non validi +3 inizializzazione FaceEngine fallita +4 caricamento o elaborazione immagine fallita +5 risultato facciale o embedding non valido +``` + +Usarlo per integrare il face engine con Node.js, Python o altri linguaggi e per costruire query destinate a `face_knn`. + +### face_knn + +Confronta embedding query con un database di prototype. Backend disponibili: + +```text +cpu +opencl +``` + +Il programma va normalmente eseguito dalla root del repository, perché il kernel viene caricato da: + +```text +kernels/knn_float4_topk.cl +``` + +Help: + +```bash +./bin/face_knn --help +``` + +Backend CPU: + +```bash +./bin/face_knn \ + --queries queries.bin \ + --database prototypes.bin \ + --query-count 21 \ + --database-count 5 \ + --dimension 128 \ + --top-k 5 \ + --threshold 0.50 \ + --backend cpu +``` + +Backend OpenCL: + +```bash +./bin/face_knn \ + --queries queries.bin \ + --database prototypes.bin \ + --query-count 21 \ + --database-count 5 \ + --dimension 128 \ + --top-k 5 \ + --threshold 0.50 \ + --backend opencl +``` + +Argomenti: + +```text +--queries file binario degli embedding query +--database file binario degli embedding prototype +--query-count numero di query +--database-count numero di prototype +--dimension componenti per embedding, 128 per SFace +--top-k numero massimo di vicini +--threshold soglia minima di riconoscimento +--self-map esclusioni leave-one-out opzionali +--backend cpu oppure opencl +``` + +Il TOPK effettivo è: + +```text +effective_top_k = min(top_k, database_count) +``` + +#### Leave-one-out + +Quando una query è anche un prototype, il confronto con sé stessa va escluso. La self-map contiene una riga per query: + +```text +-1 +1 +3 +0 +4 +-1 +``` + +`-1` indica nessuna esclusione; un valore da zero in su indica l'indice del prototype da escludere. + +```bash +./bin/face_knn \ + --queries queries.bin \ + --database prototypes.bin \ + --query-count 21 \ + --database-count 5 \ + --dimension 128 \ + --top-k 5 \ + --threshold 0.50 \ + --self-map self-map.txt \ + --backend opencl +``` + +#### Output e diagnostica + +```bash +./bin/face_knn \ + --queries queries.bin \ + --database prototypes.bin \ + --query-count 21 \ + --database-count 5 \ + --dimension 128 \ + --top-k 5 \ + --threshold 0.50 \ + --self-map self-map.txt \ + --backend opencl \ + > /tmp/face-knn.json \ + 2> /tmp/face-knn.stderr +``` + +```text +stdout → JSON valido +stderr → diagnostica OpenCL e driver Mali +``` + +Validazione: + +```bash +python3 -m json.tool /tmp/face-knn.json > /dev/null \ + && echo "OK: JSON valido" +``` + +Campi principali del JSON: + +```text +backend, device_name, query_count, database_count, dimension, top_k, +threshold, leave_one_out, elapsed_ms, results, recognized_count, +below_threshold_count, coverage_percent +``` + +Per ogni query: + +```text +query_index, excluded_index, recognized, best_index, best_score, neighbors +``` + +### test_knn_cpu + +Sorgente: + +```text +test/knn/test_knn_cpu.cc +``` + +Esecuzione, se compilato: + +```bash +./bin/test_knn_cpu +``` + +Output atteso: + +```text +OK: backend CPU dinamico +query_count=2 +database_count=3 +top_k=2 +``` + +## Scelta del backend + +Usare CPU per database piccoli, riferimento numerico o fallback. Usare OpenCL per database grandi o molte query. Con cinque prototype l'overhead OpenCL può superare il costo del calcolo; con 350000 prototype e 256 query è stato misurato uno speedup di circa 7 volte. + +## Flusso completo + +```text +face_scan +→ estrae gli embedding +→ applicazione seleziona i prototype +→ face_knn calcola TOPK +→ applicazione associa gli indici alle identità +→ applicazione applica soglia e margine +→ applicazione salva il risultato +``` + +## Prestazioni sintetiche validate + +```text +Database: 350000 +Query: 256 +Dimensione: 128 +TOPK: 32 +Chunk: 50000 +GPU: Mali-G610 r0p0 +CPU: 9663.226 ms +OpenCL: 1373.797 ms +Speedup: 7.034x +TOPK identico: 8192 / 8192 +``` + +Variante float4 min-replace: + +```text +Kernel: 1244.576 ms +Throughput: 71.992 M vector/s +Dot rate: 18.430 GFLOP/s +Traffico effettivo: 36.860 GB/s +``` + +## Validazione reale + +```text +Query: 21 +Prototype: 5 +Dimensione: 128 +Soglia: 0.50 +Leave-one-out: attivo +Riconosciuti: 20 / 21 +Copertura: 95.2% +Caso sotto soglia: IMG_0183.JPG +Score: 0.466428 +Miglior prototype: IMG_0143.JPG +``` + +Confronti: + +```text +Python vs C++ CPU: 21 / 21 query coerenti +CPU vs OpenCL dopo leave-one-out: 100 / 100 posizioni +CPU vs OpenCL TOPK grezzo: 105 / 105 posizioni +Errore massimo CPU/OpenCL: 3.58e-7 +``` + +Dettagli: + +```text +VALIDATION_JESSICA.md +README1.md +modules/knn-opencl/reference/VALIDATION.md +``` + +## Integrazione applicativa + +Un'applicazione deve leggere e validare i prototype, generare i file binari e i metadati, invocare `face_knn`, associare gli indici alle identità, applicare soglia e margine e salvare il risultato. La prima integrazione dovrebbe funzionare in modalità shadow, confrontando il risultato corrente con OpenCL prima dell'attivazione definitiva. + +## Stato + +```text +Face engine RKNN: validato +Backend CPU: validato +Backend OpenCL: validato +Mali-G610: validata +TOPK dinamico: validato +Chunking: validato +Leave-one-out: validato +Output JSON puro: validato +Dataset reale: validato +Integrazione applicativa: da completare +``` diff --git a/README1.md b/README1.md new file mode 100644 index 0000000..ca6a58e --- /dev/null +++ b/README1.md @@ -0,0 +1,239 @@ +# Note tecniche e cronologia del kNN OpenCL + +Questo documento raccoglie il percorso sperimentale che ha portato al motore kNN OpenCL. Per installazione, compilazione e uso dei binari consultare `README.md`. + +## Obiettivo + +Input: + +```text +float32[N][128] +``` + +Gli embedding sono normalizzati L2, quindi cosine similarity e prodotto scalare coincidono. Output: + +```text +indices[query_count][TOPK] +scores[query_count][TOPK] +``` + +## Baseline + +```text +DIM=128 +TOPK=32 +QUERIES=256 +LANES_PER_QUERY=8 +CHUNK_SIZE=50000 +``` + +## Risultati sperimentali principali + +### Numero di lane + +```text +LANES=4: 9.60 GOP/s +LANES=8: 10.57 GOP/s +LANES=16: 4.04 GOP/s +LANES=32: 1.49 GOP/s +``` + +Otto lane sono risultate la configurazione più equilibrata. + +### Stage benchmark + +```text +DOT: 72.79 M vector/s +DOT+threshold: 83.47 M vector/s +TOPK score: 77.94 M vector/s +TOPK+indici: 76.28 M vector/s +kNN completo: 75.22 M vector/s +``` + +### Vettorizzazione + +```text +scalar: 74.37 M vector/s +float4: 84.14 M vector/s +float8: 83.49 M vector/s +``` + +Database e query entrambi float4 hanno raggiunto 84.13 M vector/s. + +### Accumulatori + +```text +F4 singolo accumulatore: 74.53 M vector/s +F4 quattro accumulatori: 83.47 M vector/s +F4 quattro acc + unroll2: 70.53 M vector/s +F8 quattro accumulatori: 82.76 M vector/s +``` + +L'unrolling esplicito ha peggiorato il code generation del compilatore Mali. + +### Pressione sui registri + +Una query mantenuta in `private float4 query[DIM4]` ha ridotto le prestazioni a 17.85 M vector/s. Rimuovendo l'array privato si è risaliti a circa 65 M vector/s. + +### Strategie TOPK + +```text +sorted insertion: 78.37 M vector/s +min-replace: 81.99 M vector/s +threshold + min-replace: 82.02 M vector/s +``` + +### Dimensione TOPK + +```text +TOPK=8: 73.87 M vector/s +TOPK=16: 84.60 M vector/s +TOPK=32: 82.53 M vector/s +TOPK=64: 78.81 M vector/s +``` + +## Benchmark da 350000 embedding + +```text +Database: 350000 +Query: 256 +Dimensione: 128 +TOPK: 32 +Chunk: 50000 +Confronti: 89600000 +``` + +Variante min-replace: + +```text +KNN kernel: 1244.576 ms +Merge GPU: 8.022 ms +Merge CPU globale: 14.039 ms +Tempo complessivo: 1266.638 ms +Vector rate: 71.992 M vector/s +Dot rate: 18.430 GFLOP/s +Traffico: 36.860 GB/s +Indici non validi: 0 +``` + +Benchmark parametrico: + +```text +CPU: 9663.226 ms +OpenCL: 1373.797 ms +Speedup: 7.034x +Posizioni TOPK identiche: 8192 / 8192 +Exact ratio: 100% +``` + +## Trasformazione in motore dinamico + +Il kernel produttivo usa definizioni configurabili: + +```c +#ifndef DIM +#define DIM 128 +#endif + +#ifndef TOPK +#define TOPK 32 +#endif + +#ifndef QUERY_COUNT +#define QUERY_COUNT 1 +#endif + +#ifndef LANES_PER_QUERY +#define LANES_PER_QUERY 8 +#endif +``` + +Il limite fisso `q >= 256` è diventato `q >= QUERY_COUNT`. Le definizioni vengono passate a `clBuildProgram` insieme a `-cl-fast-relaxed-math`. + +## Backend CPU + +È stato generalizzato per conteggi e dimensione dinamici, TOPK sicuro, ordinamento deterministico, rifiuto di valori non finiti e layout identico a OpenCL. + +## Backend OpenCL + +Il backend: + +1. enumera piattaforme e GPU; +2. seleziona preferibilmente Mali-G610; +3. carica il kernel da file; +4. compila con parametri runtime; +5. alloca buffer riutilizzabili; +6. copia il database a chunk; +7. esegue il kernel; +8. legge indici e score; +9. converte indici locali in globali; +10. effettua il merge TOPK; +11. rilascia le risorse tramite RAII. + +## Diagnostica Mali + +Il runtime Mali emette `arm_release_ver` direttamente su stdout. La CLI devia temporaneamente stdout verso stderr durante OpenCL e lo ripristina prima della scrittura JSON. + +## Validazione reale + +Prototype: + +```text +0 IMG_0145.JPG cluster_medoid +1 IMG_0141.JPG cluster_medoid +2 IMG_0193.JPG cluster_medoid +3 IMG_0143.JPG greedy_leave_one_out_gain +4 IMG_0146.JPG greedy_leave_one_out_gain +``` + +Risultato: + +```text +Query: 21 +Prototype: 5 +Riconosciuti: 20 +Sotto soglia: 1 +Copertura: 95.2% +Caso sotto soglia: IMG_0183.JPG +Score: 0.466428 +Miglior prototype: IMG_0143.JPG +``` + +Python contro C++ CPU: + +```text +Query confrontate: 21 +Mismatch top-1: 0 +Mismatch decisione: 0 +Errore massimo: 2.643e-7 +``` + +C++ CPU contro OpenCL: + +```text +Dopo leave-one-out: 100 / 100 posizioni identiche +TOPK grezzo: 105 / 105 posizioni identiche +Errore massimo: 3.58e-7 +``` + +## File di riferimento + +```text +modules/knn-opencl/reference/knn_benchmark.cpp +modules/knn-opencl/reference/knn_float4_minreplace_validate_benchmark.cpp +modules/knn-opencl/reference/validation-opencl-350k.txt +modules/knn-opencl/reference/validation-minreplace-350k.txt +modules/knn-opencl/reference/VALIDATION.md +VALIDATION_JESSICA.md +``` + +## Sviluppi futuri + +```text +backend auto +cache del programma OpenCL +indice persistente dei prototype +margine fra prima e seconda identità +modalità shadow applicativa +metriche operative +``` diff --git a/SHA256SUMS b/SHA256SUMS new file mode 100644 index 0000000..2529dda --- /dev/null +++ b/SHA256SUMS @@ -0,0 +1,9 @@ +9fb33ec13abd1714832d6bc81a3b3df553d80577195e54ff8bfdbc33220a6566 README.md +7f9373326a666d36dcd26fcf9d2326fa56198b5aab448d53651fe5aff379be57 README1.md +c6b144c487af817ea1578f3190e968d81a09dcee5d4cdc9f8f5e664b6cd922b5 VALIDATION_JESSICA.md +f0f057dda1dd0bb8d3e097698da7b8ea956a01217b47349a99936729aefd2f3d include/knn_opencl.h +ca211d9e68b4c15b1fcbf8a2607ba398b845f7fab7db417302642f5ed2e0413e src/knn_opencl.cc +918bed774c805cafb2645214ab239852b96ed4eb104fcdf78ac2488d64c19737 src/knn_opencl_backend.cc +4561c83dcec5015e09ee5e7990698c50eed9d2251dc56349fe51cbc579d011a3 src/face_knn.cc +2721831e9cb285c3ef12aedb59fb66ca35e067321f29e67582a39e58e47d283b kernels/knn_reference.cl +6c4f502438aad3dd1600a00cb7988e0e35e56b2e8e4620d44ffcb0f3610a7226 kernels/knn_float4_topk.cl diff --git a/VALIDATION_JESSICA.md b/VALIDATION_JESSICA.md new file mode 100644 index 0000000..5b3df1c --- /dev/null +++ b/VALIDATION_JESSICA.md @@ -0,0 +1,49 @@ +# Validazione kNN reale + +## Dataset + +- Persona: Jessica +- Query: 21 +- Prototype: 5 +- Dimensione embedding: 128 float32 +- Soglia: 0.50 +- Leave-one-out: attivo + +## Python vs C++ CPU + +- Query confrontate: 21 +- Mismatch top-1: 0 +- Mismatch decisione: 0 +- Errore massimo score: 2.643e-7 + +## C++ CPU vs OpenCL dopo leave-one-out + +- Posizioni confrontate: 100 +- Indici identici: 100 +- Exact ratio: 100% +- Mismatch decisione: 0 +- Errore massimo score: 3.58e-7 + +## C++ CPU vs OpenCL TOPK grezzo + +- Posizioni confrontate: 105 +- Indici identici: 105 +- Exact ratio: 100% +- Mismatch score oltre 1e-5: 0 +- Errore massimo score: 3.58e-7 + +## Risultato riconoscimento + +- Riconosciuti: 20/21 +- Copertura: 95.2% +- Unico caso sotto soglia: IMG_0183.JPG +- Best similarity: 0.466428 +- Miglior prototype: IMG_0143.JPG + +## Backend OpenCL + +- Device: ARM / Mali-G610 r0p0 +- DIM: 128 +- TOPK: 5 +- QUERY_COUNT: 21 +- LANES_PER_QUERY: 8 diff --git a/docs/architecture.md b/docs/architecture.md new file mode 100644 index 0000000..82a4c2f --- /dev/null +++ b/docs/architecture.md @@ -0,0 +1,144 @@ +# Architettura + +Il progetto separa il motore di elaborazione facciale dai programmi che lo +utilizzano. + +## Componenti principali + +### face_engine.h + +Definisce l'interfaccia pubblica C++: + +- `Point`; +- `Face`; +- `FaceEngine`; +- inizializzazione dei modelli; +- elaborazione di una singola immagine. + +### face_engine.cc + +Implementa: + +- caricamento delle immagini tramite stb_image; +- resize bilineare; +- conversione float32 e float16; +- inferenza SCRFD tramite RKNN; +- decodifica di score, bounding box e landmark; +- non-maximum suppression; +- trasformazione similarity; +- allineamento facciale; +- inferenza SFace tramite RKNN; +- conversione dell'output; +- normalizzazione L2 dell'embedding. + +### face_recognition.cc + +Programma dimostrativo con output leggibile. + +Può: + +- elaborare una singola immagine; +- elaborare due immagini; +- confrontare tutti i volti rilevati tramite cosine similarity. + +### face_scan.cc + +Programma destinato all'integrazione con altri software. + +Caratteristiche: + +- elabora una singola immagine; +- scrive esclusivamente JSON su standard output; +- scrive diagnostica ed errori su standard error; +- valida la dimensione degli embedding; +- rifiuta valori NaN o infiniti; +- restituisce exit code documentati. + +## Pipeline runtime + +```text +File immagine + | + v +Caricamento RGB + | + v +Resize proporzionale + | + v +Input SCRFD 640 x 640 FLOAT16 NHWC + | + v +SCRFD su NPU + | + v +Decodifica livelli stride 8, 16 e 32 + | + v +Score threshold + | + v +Non-maximum suppression + | + v +Bounding box e cinque landmark + | + v +Similarity transform + | + v +Volto allineato 112 x 112 + | + v +SFace su NPU + | + v +Embedding 128D + | + v +Normalizzazione L2 +``` + +## Riutilizzo + +Le applicazioni C++ possono utilizzare direttamente `FaceEngine`. + +Applicazioni scritte in altri linguaggi possono eseguire +`face_scan` come processo figlio e analizzare il JSON prodotto. + +Il motore non dipende: + +- dal server della galleria; +- dal database; +- dall'interfaccia utente; +- da uno specifico framework applicativo. + +## Estensione futura + +Gli embedding prodotti da `FaceEngine` saranno utilizzati dalla +pipeline di clustering: + +```text +FaceEngine + | + v +Embedding float32[N][128] + | + v +kNN OpenCL TOPK=32 + | + v +Core distance + | + v +Mutual reachability + | + v +Minimum spanning tree + | + v +HDBSCAN + | + v +Cluster ID per embedding +``` diff --git a/docs/json-format.md b/docs/json-format.md new file mode 100644 index 0000000..235187a --- /dev/null +++ b/docs/json-format.md @@ -0,0 +1,113 @@ +# Formato JSON di face_scan + +`face_scan` scrive un singolo oggetto JSON su standard output. + +Gli errori e i messaggi diagnostici vengono scritti su standard error. + +## Invocazione + +```bash +./bin/face_scan /percorso/immagine.jpg +``` + +## Esempio + +```json +{ + "image": { + "path": "/percorso/immagine.jpg", + "width": 1500, + "height": 2000 + }, + "faces": [ + { + "face_index": 0, + "score": 0.8671875, + "bbox": { + "x": 1079.3457, + "y": 489.648438, + "width": 131.591797, + "height": 182.275391 + }, + "landmarks": [ + [1107.8125, 558.325195], + [1170.99609, 559.423828], + [1134.17969, 594.500732], + [1106.22559, 615.83252], + [1168.34717, 616.967773] + ], + "embedding": [ + -0.0426222458, + 0.0678814054 + ] + } + ] +} +``` + +L'array `embedding` reale contiene sempre 128 valori. + +## Oggetto image + +Campi disponibili: + +- `path`: percorso ricevuto dal programma; +- `width`: larghezza originale dell'immagine; +- `height`: altezza originale dell'immagine. + +## Oggetto face + +Campi disponibili: + +- `face_index`: indice sequenziale del volto; +- `score`: confidence SCRFD; +- `bbox.x`: coordinata sinistra; +- `bbox.y`: coordinata superiore; +- `bbox.width`: larghezza; +- `bbox.height`: altezza; +- `landmarks`: cinque coppie di coordinate; +- `embedding`: vettore L2-normalizzato di 128 elementi. + +## Immagini senza volti + +Un'immagine elaborata correttamente ma senza volti produce: + +```json +{ + "image": { + "path": "image.jpg", + "width": 1920, + "height": 1080 + }, + "faces": [] +} +``` + +L'exit code resta `0`. + +## Exit code + +```text +0 scansione completata +2 argomenti non validi +3 inizializzazione FaceEngine fallita +4 caricamento o elaborazione immagine fallita +5 risultato facciale o embedding non valido +``` + +## Requisiti dell'embedding + +Per ogni volto: + +```text +Dimensione 128 +Tipo logico float32 +Norma L2 circa 1.0 +Valori tutti finiti +``` + +Quando viene salvato in formato binario float32: + +```text +128 x 4 byte = 512 byte +``` diff --git a/include/face_config.h b/include/face_config.h new file mode 100644 index 0000000..bd1f0f5 --- /dev/null +++ b/include/face_config.h @@ -0,0 +1,5 @@ +#pragma once + +#ifndef FACE_CLUSTERING_ROOT +#define FACE_CLUSTERING_ROOT "." +#endif diff --git a/include/face_engine.h b/include/face_engine.h new file mode 100644 index 0000000..b646335 --- /dev/null +++ b/include/face_engine.h @@ -0,0 +1,47 @@ +#pragma once + +#include +#include + +struct Point { + float x = 0.0f; + float y = 0.0f; +}; + +struct Face { + float score = 0.0f; + + float x1 = 0.0f; + float y1 = 0.0f; + float x2 = 0.0f; + float y2 = 0.0f; + + Point kps[5]{}; + + // SFace: embedding L2-normalizzato da 128 float32. + std::vector embedding; +}; + +class FaceEngine { +public: + FaceEngine(); + ~FaceEngine(); + + FaceEngine(const FaceEngine&) = delete; + FaceEngine& operator=(const FaceEngine&) = delete; + + bool init( + const std::string& scrfd_model, + const std::string& sface_model); + + bool process_image( + const std::string& image_path, + std::vector& faces, + int& image_width, + int& image_height, + int rotation_degrees = 0); + +private: + struct Impl; + Impl* impl_; +}; diff --git a/include/knn_opencl.h b/include/knn_opencl.h new file mode 100644 index 0000000..db9722d --- /dev/null +++ b/include/knn_opencl.h @@ -0,0 +1,121 @@ +#ifndef FACE_CLUSTERING_KNN_OPENCL_H +#define FACE_CLUSTERING_KNN_OPENCL_H + +#include +#include +#include +#include + +namespace face_knn { + +constexpr std::uint32_t invalid_index = + 0xffffffffu; + + +struct Neighbor { + float score = 0.0f; + + std::uint32_t index = + invalid_index; +}; + + +struct SearchOptions { + std::uint32_t dimension = 128; + + std::uint32_t top_k = 1; + + std::uint32_t lanes_per_query = 8; + + std::size_t chunk_size = 50000; + + std::string kernel_path; +}; + + +struct SearchOutput { + std::size_t query_count = 0; + + std::size_t database_count = 0; + + std::uint32_t dimension = 0; + + std::uint32_t top_k = 0; + + std::string backend; + + std::string device_name; + + double elapsed_ms = 0.0; + + /* + * Layout: + * + * neighbors[ + * query_index * top_k + * + neighbor_position + * ] + */ + std::vector neighbors; +}; + + +struct ValidationResult { + std::size_t positions = 0; + + std::size_t exact_index_matches = 0; + + std::size_t decision_matches = 0; + + float maximum_score_error = 0.0f; + + bool exact_indices = false; + + bool decisions_equal = false; +}; + + +void validate_input( + const std::vector& database, + const std::vector& queries, + std::size_t database_count, + std::size_t query_count, + const SearchOptions& options); + + +SearchOutput search_cpu( + const std::vector& database, + const std::vector& queries, + std::size_t database_count, + std::size_t query_count, + const SearchOptions& options); + + +SearchOutput search_opencl( + const std::vector& database, + const std::vector& queries, + std::size_t database_count, + std::size_t query_count, + const SearchOptions& options); + + +ValidationResult compare_outputs( + const SearchOutput& expected, + const SearchOutput& actual, + float score_tolerance, + float recognition_threshold); + + +std::vector query_neighbors( + const SearchOutput& output, + std::size_t query_index); + + +std::vector exclude_neighbor( + const std::vector& neighbors, + std::uint32_t excluded_index); + + +} // namespace face_knn + +#endif diff --git a/include/rknn_api.h b/include/rknn_api.h new file mode 100644 index 0000000..7400e4a --- /dev/null +++ b/include/rknn_api.h @@ -0,0 +1,804 @@ +/**************************************************************************** +* +* Copyright (c) 2017 - 2022 by Rockchip Corp. All rights reserved. +* +* The material in this file is confidential and contains trade secrets +* of Rockchip Corporation. This is proprietary information owned by +* Rockchip Corporation. No part of this work may be disclosed, +* reproduced, copied, transmitted, or used in any way for any purpose, +* without the express written permission of Rockchip Corporation. +* +*****************************************************************************/ + + +#ifndef _RKNN_API_H +#define _RKNN_API_H + +#ifdef __cplusplus +extern "C" { +#endif + +#include + +/* + Definition of extended flag for rknn_init. +*/ +/* set high priority context. */ +#define RKNN_FLAG_PRIOR_HIGH 0x00000000 + +/* set medium priority context */ +#define RKNN_FLAG_PRIOR_MEDIUM 0x00000001 + +/* set low priority context. */ +#define RKNN_FLAG_PRIOR_LOW 0x00000002 + +/* asynchronous mode. + when enable, rknn_outputs_get will not block for too long because it directly retrieves the result of + the previous frame which can increase the frame rate on single-threaded mode, but at the cost of + rknn_outputs_get not retrieves the result of the current frame. + in multi-threaded mode you do not need to turn this mode on. */ +#define RKNN_FLAG_ASYNC_MASK 0x00000004 + +/* collect performance mode. + when enable, you can get detailed performance reports via rknn_query(ctx, RKNN_QUERY_PERF_DETAIL, ...), + but it will reduce the frame rate. */ +#define RKNN_FLAG_COLLECT_PERF_MASK 0x00000008 + +/* allocate all memory in outside, includes weight/internal/inputs/outputs */ +#define RKNN_FLAG_MEM_ALLOC_OUTSIDE 0x00000010 + +/* weight sharing with the same network structure */ +#define RKNN_FLAG_SHARE_WEIGHT_MEM 0x00000020 + +/* send fence fd from outside */ +#define RKNN_FLAG_FENCE_IN_OUTSIDE 0x00000040 + +/* get fence fd from inside */ +#define RKNN_FLAG_FENCE_OUT_OUTSIDE 0x00000080 + +/* dummy init flag: could only get total_weight_size and total_internal_size by rknn_query*/ +#define RKNN_FLAG_COLLECT_MODEL_INFO_ONLY 0x00000100 + +/* allocate internal memory in outside */ +#define RKNN_FLAG_INTERNAL_ALLOC_OUTSIDE 0x00000200 + +/* set GPU as the preferred execution backend When the operator is not supported by the NPU */ +#define RKNN_FLAG_EXECUTE_FALLBACK_PRIOR_DEVICE_GPU 0x00000400 + +/* enable allocate sram type buffers */ +#define RKNN_FLAG_ENABLE_SRAM 0x00000800 + +/* sram type buffers are shared among different contexts */ +#define RKNN_FLAG_SHARE_SRAM 0x00001000 + +/* default nice -19, this flag can disable default priority */ +#define RKNN_FLAG_DISABLE_PROC_HIGH_PRIORITY 0x00002000 + +/* don't flush input buffer cache, the user must ensure that the input tensor has flushed the cache before calling rknn_run. +!!! Don't use this flags when you call rknn_inputs_set() to set input data. */ +#define RKNN_FLAG_DISABLE_FLUSH_INPUT_MEM_CACHE 0x00004000 + +/* Don't invalid output buffer cache. + Users cannot directly access output_mem->virt_addr, + which will cause cache consistency problems. + If you want to use output_mem->virt_addr, + you must use rknn_mem_sync (ctx, mem, RKNN_MEMORY_SYNC_FROM_DEVICE) to flush the cache. + This flags is generally used when the output data of the NPU is not accessed by the CPU, + but is accessed by the GPU or RGA to reduce the time required to flush the cache. + !!! Don't use this flags when you call rknn_outputs_get() to get output data.*/ +#define RKNN_FLAG_DISABLE_FLUSH_OUTPUT_MEM_CACHE 0x00008000 + +/* This flag is used when the model data buffer is allocated by NPU, and can be accessed by NPU directly. */ +#define RKNN_FLAG_MODEL_BUFFER_ZERO_COPY 0x00010000 + +/* This flag is a memory allocation flag, which is used in rknn_create_mem2() when no context is available. */ +#define RKNN_MEM_FLAG_ALLOC_NO_CONTEXT 0x00020000 + + +/* + Error code returned by the RKNN API. +*/ +#define RKNN_SUCC 0 /* execute succeed. */ +#define RKNN_ERR_FAIL -1 /* execute failed. */ +#define RKNN_ERR_TIMEOUT -2 /* execute timeout. */ +#define RKNN_ERR_DEVICE_UNAVAILABLE -3 /* device is unavailable. */ +#define RKNN_ERR_MALLOC_FAIL -4 /* memory malloc fail. */ +#define RKNN_ERR_PARAM_INVALID -5 /* parameter is invalid. */ +#define RKNN_ERR_MODEL_INVALID -6 /* model is invalid. */ +#define RKNN_ERR_CTX_INVALID -7 /* context is invalid. */ +#define RKNN_ERR_INPUT_INVALID -8 /* input is invalid. */ +#define RKNN_ERR_OUTPUT_INVALID -9 /* output is invalid. */ +#define RKNN_ERR_DEVICE_UNMATCH -10 /* the device is unmatch, please update rknn sdk + and npu driver/firmware. */ +#define RKNN_ERR_INCOMPATILE_PRE_COMPILE_MODEL -11 /* This RKNN model use pre_compile mode, but not compatible with current driver. */ +#define RKNN_ERR_INCOMPATILE_OPTIMIZATION_LEVEL_VERSION -12 /* This RKNN model set optimization level, but not compatible with current driver. */ +#define RKNN_ERR_TARGET_PLATFORM_UNMATCH -13 /* This RKNN model set target platform, but not compatible with current platform. */ + +/* + Definition for tensor +*/ +#define RKNN_MAX_DIMS 16 /* maximum dimension of tensor. */ +#define RKNN_MAX_NUM_CHANNEL 15 /* maximum channel number of input tensor. */ +#define RKNN_MAX_NAME_LEN 256 /* maximum name lenth of tensor. */ +#define RKNN_MAX_DYNAMIC_SHAPE_NUM 512 /* maximum number of dynamic shape for each input. */ + +#ifdef __arm__ +typedef uint32_t rknn_context; +#else +typedef uint64_t rknn_context; +#endif + + +/* + The query command for rknn_query +*/ +typedef enum _rknn_query_cmd { + RKNN_QUERY_IN_OUT_NUM = 0, /* query the number of input & output tensor. */ + RKNN_QUERY_INPUT_ATTR = 1, /* query the attribute of input tensor. */ + RKNN_QUERY_OUTPUT_ATTR = 2, /* query the attribute of output tensor. */ + RKNN_QUERY_PERF_DETAIL = 3, /* query the detail performance, need set + RKNN_FLAG_COLLECT_PERF_MASK when call rknn_init, + this query needs to be valid after rknn_outputs_get. */ + RKNN_QUERY_PERF_RUN = 4, /* query the time of run, + this query needs to be valid after rknn_outputs_get. */ + RKNN_QUERY_SDK_VERSION = 5, /* query the sdk & driver version */ + + RKNN_QUERY_MEM_SIZE = 6, /* query the weight & internal memory size */ + RKNN_QUERY_CUSTOM_STRING = 7, /* query the custom string */ + + RKNN_QUERY_NATIVE_INPUT_ATTR = 8, /* query the attribute of native input tensor. */ + RKNN_QUERY_NATIVE_OUTPUT_ATTR = 9, /* query the attribute of native output tensor. */ + + RKNN_QUERY_NATIVE_NC1HWC2_INPUT_ATTR = 8, /* query the attribute of native input tensor. */ + RKNN_QUERY_NATIVE_NC1HWC2_OUTPUT_ATTR = 9, /* query the attribute of native output tensor. */ + + RKNN_QUERY_NATIVE_NHWC_INPUT_ATTR = 10, /* query the attribute of native input tensor. */ + RKNN_QUERY_NATIVE_NHWC_OUTPUT_ATTR = 11, /* query the attribute of native output tensor. */ + + RKNN_QUERY_DEVICE_MEM_INFO = 12, /* query the attribute of rknn memory information. */ + + RKNN_QUERY_INPUT_DYNAMIC_RANGE = 13, /* query the dynamic shape range of rknn input tensor. */ + RKNN_QUERY_CURRENT_INPUT_ATTR = 14, /* query the current shape of rknn input tensor, only valid for dynamic rknn model*/ + RKNN_QUERY_CURRENT_OUTPUT_ATTR = 15, /* query the current shape of rknn output tensor, only valid for dynamic rknn model*/ + + RKNN_QUERY_CURRENT_NATIVE_INPUT_ATTR = 16, /* query the current native shape of rknn input tensor, only valid for dynamic rknn model*/ + RKNN_QUERY_CURRENT_NATIVE_OUTPUT_ATTR = 17, /* query the current native shape of rknn output tensor, only valid for dynamic rknn model*/ + + + RKNN_QUERY_CMD_MAX +} rknn_query_cmd; + +/* + the tensor data type. +*/ +typedef enum _rknn_tensor_type { + RKNN_TENSOR_FLOAT32 = 0, /* data type is float32. */ + RKNN_TENSOR_FLOAT16, /* data type is float16. */ + RKNN_TENSOR_INT8, /* data type is int8. */ + RKNN_TENSOR_UINT8, /* data type is uint8. */ + RKNN_TENSOR_INT16, /* data type is int16. */ + RKNN_TENSOR_UINT16, /* data type is uint16. */ + RKNN_TENSOR_INT32, /* data type is int32. */ + RKNN_TENSOR_UINT32, /* data type is uint32. */ + RKNN_TENSOR_INT64, /* data type is int64. */ + RKNN_TENSOR_BOOL, + RKNN_TENSOR_INT4, + RKNN_TENSOR_BFLOAT16, + + RKNN_TENSOR_TYPE_MAX +} rknn_tensor_type; + +inline static const char* get_type_string(rknn_tensor_type type) +{ + switch(type) { + case RKNN_TENSOR_FLOAT32: return "FP32"; + case RKNN_TENSOR_FLOAT16: return "FP16"; + case RKNN_TENSOR_INT8: return "INT8"; + case RKNN_TENSOR_UINT8: return "UINT8"; + case RKNN_TENSOR_INT16: return "INT16"; + case RKNN_TENSOR_UINT16: return "UINT16"; + case RKNN_TENSOR_INT32: return "INT32"; + case RKNN_TENSOR_UINT32: return "UINT32"; + case RKNN_TENSOR_INT64: return "INT64"; + case RKNN_TENSOR_BOOL: return "BOOL"; + case RKNN_TENSOR_INT4: return "INT4"; + case RKNN_TENSOR_BFLOAT16: return "BF16"; + default: return "UNKNOW"; + } +} + +/* + the quantitative type. +*/ +typedef enum _rknn_tensor_qnt_type { + RKNN_TENSOR_QNT_NONE = 0, /* none. */ + RKNN_TENSOR_QNT_DFP, /* dynamic fixed point. */ + RKNN_TENSOR_QNT_AFFINE_ASYMMETRIC, /* asymmetric affine. */ + + RKNN_TENSOR_QNT_MAX +} rknn_tensor_qnt_type; + +inline static const char* get_qnt_type_string(rknn_tensor_qnt_type type) +{ + switch(type) { + case RKNN_TENSOR_QNT_NONE: return "NONE"; + case RKNN_TENSOR_QNT_DFP: return "DFP"; + case RKNN_TENSOR_QNT_AFFINE_ASYMMETRIC: return "AFFINE"; + default: return "UNKNOW"; + } +} + +/* + the tensor data format. +*/ +typedef enum _rknn_tensor_format { + RKNN_TENSOR_NCHW = 0, /* data format is NCHW. */ + RKNN_TENSOR_NHWC, /* data format is NHWC. */ + RKNN_TENSOR_NC1HWC2, /* data format is NC1HWC2. */ + RKNN_TENSOR_UNDEFINED, + + RKNN_TENSOR_FORMAT_MAX +} rknn_tensor_format; + +/* + the mode of running on target NPU core. +*/ +typedef enum _rknn_core_mask { + RKNN_NPU_CORE_AUTO = 0, /* default, run on NPU core randomly. */ + RKNN_NPU_CORE_0 = 1, /* run on NPU core 0. */ + RKNN_NPU_CORE_1 = 2, /* run on NPU core 1. */ + RKNN_NPU_CORE_2 = 4, /* run on NPU core 2. */ + RKNN_NPU_CORE_0_1 = RKNN_NPU_CORE_0 | RKNN_NPU_CORE_1, /* run on NPU core 0 and core 1. */ + RKNN_NPU_CORE_0_1_2 = RKNN_NPU_CORE_0_1 | RKNN_NPU_CORE_2, /* run on NPU core 0 and core 1 and core 2. */ + RKNN_NPU_CORE_ALL = 0xffff, /* auto choice, run on NPU cores depending on platform */ + + RKNN_NPU_CORE_UNDEFINED, +} rknn_core_mask; + +inline static const char* get_format_string(rknn_tensor_format fmt) +{ + switch(fmt) { + case RKNN_TENSOR_NCHW: return "NCHW"; + case RKNN_TENSOR_NHWC: return "NHWC"; + case RKNN_TENSOR_NC1HWC2: return "NC1HWC2"; + case RKNN_TENSOR_UNDEFINED: return "UNDEFINED"; + default: return "UNKNOW"; + } +} + +/* + the information for RKNN_QUERY_IN_OUT_NUM. +*/ +typedef struct _rknn_input_output_num { + uint32_t n_input; /* the number of input. */ + uint32_t n_output; /* the number of output. */ +} rknn_input_output_num; + +/* + the information for RKNN_QUERY_INPUT_ATTR / RKNN_QUERY_OUTPUT_ATTR. +*/ +typedef struct _rknn_tensor_attr { + uint32_t index; /* input parameter, the index of input/output tensor, + need set before call rknn_query. */ + + uint32_t n_dims; /* the number of dimensions. */ + uint32_t dims[RKNN_MAX_DIMS]; /* the dimensions array. */ + char name[RKNN_MAX_NAME_LEN]; /* the name of tensor. */ + + uint32_t n_elems; /* the number of elements. */ + uint32_t size; /* the bytes size of tensor. */ + + rknn_tensor_format fmt; /* the data format of tensor. */ + rknn_tensor_type type; /* the data type of tensor. */ + rknn_tensor_qnt_type qnt_type; /* the quantitative type of tensor. */ + int8_t fl; /* fractional length for RKNN_TENSOR_QNT_DFP. */ + int32_t zp; /* zero point for RKNN_TENSOR_QNT_AFFINE_ASYMMETRIC. */ + float scale; /* scale for RKNN_TENSOR_QNT_AFFINE_ASYMMETRIC. */ + + uint32_t w_stride; /* the stride of tensor along the width dimention of input, + Note: it is read-only, 0 means equal to width. */ + uint32_t size_with_stride; /* the bytes size of tensor with stride. */ + + uint8_t pass_through; /* pass through mode, for rknn_set_io_mem interface. + if TRUE, the buf data is passed directly to the input node of the rknn model + without any conversion. the following variables do not need to be set. + if FALSE, the buf data is converted into an input consistent with the model + according to the following type and fmt. so the following variables + need to be set.*/ + uint32_t h_stride; /* the stride along the height dimention of input, + Note: it is write-only, if it was set to 0, h_stride = height. */ +} rknn_tensor_attr; + +typedef struct _rknn_input_range { + uint32_t index; /* input parameter, the index of input/output tensor, + need set before call rknn_query. */ + uint32_t shape_number; /* the number of shape. */ + rknn_tensor_format fmt; /* the data format of tensor. */ + char name[RKNN_MAX_NAME_LEN]; /* the name of tensor. */ + uint32_t dyn_range[RKNN_MAX_DYNAMIC_SHAPE_NUM][RKNN_MAX_DIMS]; /* the dynamic input dimensions range. */ + uint32_t n_dims; /* the number of dimensions. */ + +} rknn_input_range; + +/* + the information for RKNN_QUERY_PERF_DETAIL. +*/ +typedef struct _rknn_perf_detail { + char* perf_data; /* the string pointer of perf detail. don't need free it by user. */ + uint64_t data_len; /* the string length. */ +} rknn_perf_detail; + +/* + the information for RKNN_QUERY_PERF_RUN. +*/ +typedef struct _rknn_perf_run { + int64_t run_duration; /* real inference time (us) */ +} rknn_perf_run; + +/* + the information for RKNN_QUERY_SDK_VERSION. +*/ +typedef struct _rknn_sdk_version { + char api_version[256]; /* the version of rknn api. */ + char drv_version[256]; /* the version of rknn driver. */ +} rknn_sdk_version; + +/* + the information for RKNN_QUERY_MEM_SIZE. +*/ +typedef struct _rknn_mem_size { + uint32_t total_weight_size; /* the weight memory size */ + uint32_t total_internal_size; /* the internal memory size, exclude inputs/outputs */ + uint64_t total_dma_allocated_size; /* total dma memory allocated size */ + uint32_t total_sram_size; /* total system sram size reserved for rknn */ + uint32_t free_sram_size; /* free system sram size reserved for rknn */ + uint32_t reserved[10]; /* reserved */ +} rknn_mem_size; + +/* + the information for RKNN_QUERY_CUSTOM_STRING. +*/ +typedef struct _rknn_custom_string { + char string[1024]; /* the string of custom, lengths max to 1024 bytes */ +} rknn_custom_string; + +/* + The flags of rknn_tensor_mem. +*/ +typedef enum _rknn_tensor_mem_flags { + RKNN_TENSOR_MEMORY_FLAGS_ALLOC_INSIDE = 1, /*Used to mark in rknn_destroy_mem() whether it is necessary to release the "mem" pointer itself. + If the flag RKNN_TENSOR_MEMORY_FLAGS_ALLOC_INSIDE is set, rknn_destroy_mem() will call free(mem).*/ + RKNN_TENSOR_MEMORY_FLAGS_FROM_FD = 2, /*Used to mark in rknn_create_mem_from_fd() whether it is necessary to release the "mem" pointer itself. + If the flag RKNN_TENSOR_MEMORY_FLAGS_FROM_FD is set, rknn_destroy_mem() will call free(mem).*/ + RKNN_TENSOR_MEMORY_FLAGS_FROM_PHYS = 3, /*Used to mark in rknn_create_mem_from_phys() whether it is necessary to release the "mem" pointer itself. + If the flag RKNN_TENSOR_MEMORY_FLAGS_FROM_PHYS is set, rknn_destroy_mem() will call free(mem).*/ + RKNN_TENSOR_MEMORY_FLAGS_UNKNOWN +} rknn_tensor_mem_flags; + +/* + The mode to allocate rknn memory. +*/ +typedef enum _rknn_mem_alloc_flags { + RKNN_FLAG_MEMORY_FLAGS_DEFAULT = 0 << 0, /* Same with RKNN_FLAG_MEMORY_CACHEABLE */ + RKNN_FLAG_MEMORY_CACHEABLE = 1 << 0, /* Create Cacheable memory. */ + RKNN_FLAG_MEMORY_NON_CACHEABLE = 1 << 1, /* Create NON-Cacheable memory. */ + RKNN_FLAG_MEMORY_TRY_ALLOC_SRAM = 1 << 2, /* Try to allocate memory in SRAM if possible. if SRAM is not enough, allocate rest memory in DRAM. */ +} rknn_mem_alloc_flags; + +/* + The mode to sync cacheable rknn memory. +*/ +typedef enum _rknn_mem_sync_mode { + RKNN_MEMORY_SYNC_TO_DEVICE = 0x1, /* the mode used for consistency of device access after CPU accesses data. */ + RKNN_MEMORY_SYNC_FROM_DEVICE = 0x2, /* the mode used for consistency of CPU access after device accesses data. */ + RKNN_MEMORY_SYNC_BIDIRECTIONAL = RKNN_MEMORY_SYNC_TO_DEVICE | RKNN_MEMORY_SYNC_FROM_DEVICE, /* the mode used for consistency of data access + between device and CPU in both directions. */ +} rknn_mem_sync_mode; + +/* + the memory information of tensor. +*/ +typedef struct _rknn_tensor_memory { + void* virt_addr; /* the virtual address of tensor buffer. */ + uint64_t phys_addr; /* the physical address of tensor buffer. */ + int32_t fd; /* the fd of tensor buffer. */ + int32_t offset; /* indicates the offset of the memory. */ + uint32_t size; /* the size of tensor buffer. */ + uint32_t flags; /* the flags of tensor buffer, reserved */ + void * priv_data; /* the private data of tensor buffer. */ +} rknn_tensor_mem; + +/* + the input information for rknn_input_set. +*/ +typedef struct _rknn_input { + uint32_t index; /* the input index. */ + void* buf; /* the input buf for index. */ + uint32_t size; /* the size of input buf. */ + uint8_t pass_through; /* pass through mode. + if TRUE, the buf data is passed directly to the input node of the rknn model + without any conversion. the following variables do not need to be set. + if FALSE, the buf data is converted into an input consistent with the model + according to the following type and fmt. so the following variables + need to be set.*/ + rknn_tensor_type type; /* the data type of input buf. */ + rknn_tensor_format fmt; /* the data format of input buf. + currently the internal input format of NPU is NCHW by default. + so entering NCHW data can avoid the format conversion in the driver. */ +} rknn_input; + +/* + the output information for rknn_outputs_get. +*/ +typedef struct _rknn_output { + uint8_t want_float; /* want transfer output data to float */ + uint8_t is_prealloc; /* whether buf is pre-allocated. + if TRUE, the following variables need to be set. + if FALSE, the following variables do not need to be set. */ + uint32_t index; /* the output index. */ + void* buf; /* the output buf for index. + when is_prealloc = FALSE and rknn_outputs_release called, + this buf pointer will be free and don't use it anymore. */ + uint32_t size; /* the size of output buf. */ +} rknn_output; + +/* + the extend information for rknn_init. +*/ +typedef struct _rknn_init_extend { + rknn_context ctx; /* rknn context */ + int32_t real_model_offset; /* real rknn model file offset, only valid when init context with rknn file path and zero-copy model model */ + uint32_t real_model_size; /* real rknn model file size, only valid when init context with rknn file path and zero-copy model model */ + int32_t model_buffer_fd; /* the fd of model buffer. */ + uint32_t model_buffer_flags; /* the flags of model_buffer */ + uint8_t reserved[112]; /* reserved */ +} rknn_init_extend; + +/* + the extend information for rknn_run. +*/ +typedef struct _rknn_run_extend { + uint64_t frame_id; /* output parameter, indicate current frame id of run. */ + int32_t non_block; /* block flag of run, 0 is block else 1 is non block */ + int32_t timeout_ms; /* timeout for block mode, in milliseconds */ + int32_t fence_fd; /* fence fd from other unit */ +} rknn_run_extend; + +/* + the extend information for rknn_outputs_get. +*/ +typedef struct _rknn_output_extend { + uint64_t frame_id; /* output parameter, indicate the frame id of outputs, corresponds to + struct rknn_run_extend.frame_id.*/ +} rknn_output_extend; + + +/* rknn_init + + initial the context and load the rknn model. + + input: + rknn_context* context the pointer of context handle. + void* model if size > 0, pointer to the rknn model, if size = 0, filepath to the rknn model. + uint32_t size the size of rknn model. + uint32_t flag extend flag, see the define of RKNN_FLAG_XXX_XXX. + rknn_init_extend* extend the extend information of init. + return: + int error code. +*/ +int rknn_init(rknn_context* context, void* model, uint32_t size, uint32_t flag, rknn_init_extend* extend); + +/* rknn_dup_context + + initial the context and load the rknn model. + + input: + rknn_context* context_in the pointer of context in handle. + rknn_context* context_out the pointer of context out handle. + return: + int error code. +*/ +int rknn_dup_context(rknn_context* context_in, rknn_context* context_out); + +/* rknn_destroy + + unload the rknn model and destroy the context. + + input: + rknn_context context the handle of context. + return: + int error code. +*/ +int rknn_destroy(rknn_context context); + + +/* rknn_query + + query the information about model or others. see rknn_query_cmd. + + input: + rknn_context context the handle of context. + rknn_query_cmd cmd the command of query. + void* info the buffer point of information. + uint32_t size the size of information. + return: + int error code. +*/ +int rknn_query(rknn_context context, rknn_query_cmd cmd, void* info, uint32_t size); + + +/* rknn_inputs_set + + set inputs information by input index of rknn model. + inputs information see rknn_input. + + input: + rknn_context context the handle of context. + uint32_t n_inputs the number of inputs. + rknn_input inputs[] the arrays of inputs information, see rknn_input. + return: + int error code +*/ +int rknn_inputs_set(rknn_context context, uint32_t n_inputs, rknn_input inputs[]); + +/* + rknn_set_batch_core_num + + set rknn batch core_num. + + input: + rknn_context context the handle of context. + int core_num the core number. + return: + int error code. + +*/ +int rknn_set_batch_core_num(rknn_context context, int core_num); + +/* rknn_set_core_mask + + set the core mask for the model.(only supported on multi-core NPU platform) + + RKNN_NPU_CORE_AUTO: auto mode, default value + RKNN_NPU_CORE_0: core 0 mode + RKNN_NPU_CORE_1: core 1 mode + RKNN_NPU_CORE_2: core 2 mode + RKNN_NPU_CORE_0_1: combine core 0/1 mode + RKNN_NPU_CORE_0_1_2: combine core 0/1/2 mode + RKNN_NPU_CORE_ALL: auto mode, select multiple npu cores to run depending on platform + + + input: + rknn_context context the handle of context. + rknn_core_mask core_mask the core mask. + return: + int error code. +*/ +int rknn_set_core_mask(rknn_context context, rknn_core_mask core_mask); + +/* rknn_run + + run the model to execute inference. + + input: + rknn_context context the handle of context. + rknn_run_extend* extend the extend information of run. + return: + int error code. +*/ +int rknn_run(rknn_context context, rknn_run_extend* extend); + + +/* rknn_wait + + wait the model after execute inference. + + input: + rknn_context context the handle of context. + rknn_run_extend* extend the extend information of run. + return: + int error code. +*/ +int rknn_wait(rknn_context context, rknn_run_extend* extend); + + +/* rknn_outputs_get + + wait the inference to finish and get the outputs. + this function will block until inference finish. + the results will set to outputs[]. + + input: + rknn_context context the handle of context. + uint32_t n_outputs the number of outputs. + rknn_output outputs[] the arrays of output, see rknn_output. + rknn_output_extend* the extend information of output. + return: + int error code. +*/ +int rknn_outputs_get(rknn_context context, uint32_t n_outputs, rknn_output outputs[], rknn_output_extend* extend); + + +/* rknn_outputs_release + + release the outputs that get by rknn_outputs_get. + after called, the rknn_output[x].buf get from rknn_outputs_get will + also be free when rknn_output[x].is_prealloc = FALSE. + + input: + rknn_context context the handle of context. + uint32_t n_ouputs the number of outputs. + rknn_output outputs[] the arrays of output. + return: + int error code +*/ +int rknn_outputs_release(rknn_context context, uint32_t n_ouputs, rknn_output outputs[]); + + +/* new api for zero copy */ + +/* rknn_create_mem_from_phys (memory allocated outside) + + initialize tensor memory from physical address. + + input: + rknn_context ctx the handle of context. + uint64_t phys_addr physical address. + void *virt_addr virtual address. + uint32_t size the size of tensor buffer. + return: + rknn_tensor_mem the pointer of tensor memory information. +*/ +rknn_tensor_mem* rknn_create_mem_from_phys(rknn_context ctx, uint64_t phys_addr, void *virt_addr, uint32_t size); + + +/* rknn_create_mem_from_fd (memory allocated outside) + + initialize tensor memory from file description. + + input: + rknn_context ctx the handle of context. + int32_t fd file description. + void *virt_addr virtual address. + uint32_t size the size of tensor buffer. + int32_t offset indicates the offset of the memory (virt_addr without offset). + return: + rknn_tensor_mem the pointer of tensor memory information. +*/ +rknn_tensor_mem* rknn_create_mem_from_fd(rknn_context ctx, int32_t fd, void *virt_addr, uint32_t size, int32_t offset); + + +/* rknn_create_mem_from_mb_blk (memory allocated outside) + + create tensor memory from mb_blk. + + input: + rknn_context ctx the handle of context. + void *mb_blk mb_blk allocate from system api. + int32_t offset indicates the offset of the memory. + return: + rknn_tensor_mem the pointer of tensor memory information. +*/ +rknn_tensor_mem* rknn_create_mem_from_mb_blk(rknn_context ctx, void *mb_blk, int32_t offset); + + +/* rknn_create_mem (memory allocated inside) + + create tensor memory. + + input: + rknn_context ctx the handle of context. + uint32_t size the size of tensor buffer. + return: + rknn_tensor_mem the pointer of tensor memory information. +*/ +rknn_tensor_mem* rknn_create_mem(rknn_context ctx, uint32_t size); + +/* rknn_create_mem2 (memory allocated inside) + + create tensor memory. + + input: + rknn_context ctx the handle of context. + uint64_t size the size of tensor buffer. + uint64_t alloc_flags memory allocation flags. + return: + rknn_tensor_mem the pointer of tensor memory information. +*/ +rknn_tensor_mem* rknn_create_mem2(rknn_context ctx, uint64_t size, uint64_t alloc_flags); + +/* rknn_destroy_mem (support allocate inside and outside) + + destroy tensor memory. + + input: + rknn_context ctx the handle of context. + rknn_tensor_mem *mem the pointer of tensor memory information. + return: + int error code +*/ +int rknn_destroy_mem(rknn_context ctx, rknn_tensor_mem *mem); + + +/* rknn_set_weight_mem + + set the weight memory. + + input: + rknn_context ctx the handle of context. + rknn_tensor_mem *mem the array of tensor memory information + return: + int error code. +*/ +int rknn_set_weight_mem(rknn_context ctx, rknn_tensor_mem *mem); + + +/* rknn_set_internal_mem + + set the internal memory. + + input: + rknn_context ctx the handle of context. + rknn_tensor_mem *mem the array of tensor memory information + return: + int error code. +*/ +int rknn_set_internal_mem(rknn_context ctx, rknn_tensor_mem *mem); + + +/* rknn_set_io_mem + + set the input and output tensors buffer. + + input: + rknn_context ctx the handle of context. + rknn_tensor_mem *mem the array of tensor memory information. + rknn_tensor_attr *attr the attribute of input or output tensor buffer. + return: + int error code. +*/ +int rknn_set_io_mem(rknn_context ctx, rknn_tensor_mem *mem, rknn_tensor_attr *attr); + +/* rknn_set_input_shape(deprecated) + + set the input tensor shape (only valid for dynamic shape rknn model). + + input: + rknn_context ctx the handle of context. + rknn_tensor_attr *attr the attribute of input or output tensor buffer. + return: + int error code. +*/ +int rknn_set_input_shape(rknn_context ctx, rknn_tensor_attr* attr); + +/* rknn_set_input_shapes + + set all the input tensor shapes. graph will run under current set of input shapes after rknn_set_input_shapes.(only valid for dynamic shape rknn model). + + input: + rknn_context ctx the handle of context. + uint32_t n_inputs the number of inputs. + rknn_tensor_attr attr[] the attribute array of all input tensors. + return: + int error code. +*/ +int rknn_set_input_shapes(rknn_context ctx, uint32_t n_inputs, rknn_tensor_attr attr[]); + +/* rknn_mem_sync + + sync cacheable rknn memory when both cpu and device access data. + + input: + rknn_context context the handle of context. + rknn_tensor_mem *mem the pointer of tensor memory information. + rknn_mem_sync_mode mode the mode of sync cache. + return: + int error code. +*/ +int rknn_mem_sync(rknn_context context, rknn_tensor_mem* mem, rknn_mem_sync_mode mode); + +#ifdef __cplusplus +} //extern "C" +#endif + +#endif //_RKNN_API_H diff --git a/include/stb_image.h b/include/stb_image.h new file mode 100644 index 0000000..accef48 --- /dev/null +++ b/include/stb_image.h @@ -0,0 +1,7762 @@ +/* stb_image - v2.26 - public domain image loader - http://nothings.org/stb + no warranty implied; use at your own risk + + Do this: + #define STB_IMAGE_IMPLEMENTATION + before you include this file in *one* C or C++ file to create the implementation. + + // i.e. it should look like this: + #include ... + #include ... + #include ... + #define STB_IMAGE_IMPLEMENTATION + #include "stb_image.h" + + You can #define STBI_ASSERT(x) before the #include to avoid using assert.h. + And #define STBI_MALLOC, STBI_REALLOC, and STBI_FREE to avoid using malloc,realloc,free + + + QUICK NOTES: + Primarily of interest to game developers and other people who can + avoid problematic images and only need the trivial interface + + JPEG baseline & progressive (12 bpc/arithmetic not supported, same as stock IJG lib) + PNG 1/2/4/8/16-bit-per-channel + + TGA (not sure what subset, if a subset) + BMP non-1bpp, non-RLE + PSD (composited view only, no extra channels, 8/16 bit-per-channel) + + GIF (*comp always reports as 4-channel) + HDR (radiance rgbE format) + PIC (Softimage PIC) + PNM (PPM and PGM binary only) + + Animated GIF still needs a proper API, but here's one way to do it: + http://gist.github.com/urraka/685d9a6340b26b830d49 + + - decode from memory or through FILE (define STBI_NO_STDIO to remove code) + - decode from arbitrary I/O callbacks + - SIMD acceleration on x86/x64 (SSE2) and ARM (NEON) + + Full documentation under "DOCUMENTATION" below. + + +LICENSE + + See end of file for license information. + +RECENT REVISION HISTORY: + + 2.26 (2020-07-13) many minor fixes + 2.25 (2020-02-02) fix warnings + 2.24 (2020-02-02) fix warnings; thread-local failure_reason and flip_vertically + 2.23 (2019-08-11) fix clang static analysis warning + 2.22 (2019-03-04) gif fixes, fix warnings + 2.21 (2019-02-25) fix typo in comment + 2.20 (2019-02-07) support utf8 filenames in Windows; fix warnings and platform ifdefs + 2.19 (2018-02-11) fix warning + 2.18 (2018-01-30) fix warnings + 2.17 (2018-01-29) bugfix, 1-bit BMP, 16-bitness query, fix warnings + 2.16 (2017-07-23) all functions have 16-bit variants; optimizations; bugfixes + 2.15 (2017-03-18) fix png-1,2,4; all Imagenet JPGs; no runtime SSE detection on GCC + 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs + 2.13 (2016-12-04) experimental 16-bit API, only for PNG so far; fixes + 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes + 2.11 (2016-04-02) 16-bit PNGS; enable SSE2 in non-gcc x64 + RGB-format JPEG; remove white matting in PSD; + allocate large structures on the stack; + correct channel count for PNG & BMP + 2.10 (2016-01-22) avoid warning introduced in 2.09 + 2.09 (2016-01-16) 16-bit TGA; comments in PNM files; STBI_REALLOC_SIZED + + See end of file for full revision history. + + + ============================ Contributors ========================= + + Image formats Extensions, features + Sean Barrett (jpeg, png, bmp) Jetro Lauha (stbi_info) + Nicolas Schulz (hdr, psd) Martin "SpartanJ" Golini (stbi_info) + Jonathan Dummer (tga) James "moose2000" Brown (iPhone PNG) + Jean-Marc Lienher (gif) Ben "Disch" Wenger (io callbacks) + Tom Seddon (pic) Omar Cornut (1/2/4-bit PNG) + Thatcher Ulrich (psd) Nicolas Guillemot (vertical flip) + Ken Miller (pgm, ppm) Richard Mitton (16-bit PSD) + github:urraka (animated gif) Junggon Kim (PNM comments) + Christopher Forseth (animated gif) Daniel Gibson (16-bit TGA) + socks-the-fox (16-bit PNG) + Jeremy Sawicki (handle all ImageNet JPGs) + Optimizations & bugfixes Mikhail Morozov (1-bit BMP) + Fabian "ryg" Giesen Anael Seghezzi (is-16-bit query) + Arseny Kapoulkine + John-Mark Allen + Carmelo J Fdez-Aguera + + Bug & warning fixes + Marc LeBlanc David Woo Guillaume George Martins Mozeiko + Christpher Lloyd Jerry Jansson Joseph Thomson Blazej Dariusz Roszkowski + Phil Jordan Dave Moore Roy Eltham + Hayaki Saito Nathan Reed Won Chun + Luke Graham Johan Duparc Nick Verigakis the Horde3D community + Thomas Ruf Ronny Chevalier github:rlyeh + Janez Zemva John Bartholomew Michal Cichon github:romigrou + Jonathan Blow Ken Hamada Tero Hanninen github:svdijk + Laurent Gomila Cort Stratton github:snagar + Aruelien Pocheville Sergio Gonzalez Thibault Reuille github:Zelex + Cass Everitt Ryamond Barbiero github:grim210 + Paul Du Bois Engin Manap Aldo Culquicondor github:sammyhw + Philipp Wiesemann Dale Weiler Oriol Ferrer Mesia github:phprus + Josh Tobin Matthew Gregan github:poppolopoppo + Julian Raschke Gregory Mullen Christian Floisand github:darealshinji + Baldur Karlsson Kevin Schmidt JR Smith github:Michaelangel007 + Brad Weinberger Matvey Cherevko [reserved] + Luca Sas Alexander Veselov Zack Middleton [reserved] + Ryan C. Gordon [reserved] [reserved] + DO NOT ADD YOUR NAME HERE + + To add your name to the credits, pick a random blank space in the middle and fill it. + 80% of merge conflicts on stb PRs are due to people adding their name at the end + of the credits. +*/ + +#ifndef STBI_INCLUDE_STB_IMAGE_H +#define STBI_INCLUDE_STB_IMAGE_H + +// DOCUMENTATION +// +// Limitations: +// - no 12-bit-per-channel JPEG +// - no JPEGs with arithmetic coding +// - GIF always returns *comp=4 +// +// Basic usage (see HDR discussion below for HDR usage): +// int x,y,n; +// unsigned char *data = stbi_load(filename, &x, &y, &n, 0); +// // ... process data if not NULL ... +// // ... x = width, y = height, n = # 8-bit components per pixel ... +// // ... replace '0' with '1'..'4' to force that many components per pixel +// // ... but 'n' will always be the number that it would have been if you said 0 +// stbi_image_free(data) +// +// Standard parameters: +// int *x -- outputs image width in pixels +// int *y -- outputs image height in pixels +// int *channels_in_file -- outputs # of image components in image file +// int desired_channels -- if non-zero, # of image components requested in result +// +// The return value from an image loader is an 'unsigned char *' which points +// to the pixel data, or NULL on an allocation failure or if the image is +// corrupt or invalid. The pixel data consists of *y scanlines of *x pixels, +// with each pixel consisting of N interleaved 8-bit components; the first +// pixel pointed to is top-left-most in the image. There is no padding between +// image scanlines or between pixels, regardless of format. The number of +// components N is 'desired_channels' if desired_channels is non-zero, or +// *channels_in_file otherwise. If desired_channels is non-zero, +// *channels_in_file has the number of components that _would_ have been +// output otherwise. E.g. if you set desired_channels to 4, you will always +// get RGBA output, but you can check *channels_in_file to see if it's trivially +// opaque because e.g. there were only 3 channels in the source image. +// +// An output image with N components has the following components interleaved +// in this order in each pixel: +// +// N=#comp components +// 1 grey +// 2 grey, alpha +// 3 red, green, blue +// 4 red, green, blue, alpha +// +// If image loading fails for any reason, the return value will be NULL, +// and *x, *y, *channels_in_file will be unchanged. The function +// stbi_failure_reason() can be queried for an extremely brief, end-user +// unfriendly explanation of why the load failed. Define STBI_NO_FAILURE_STRINGS +// to avoid compiling these strings at all, and STBI_FAILURE_USERMSG to get slightly +// more user-friendly ones. +// +// Paletted PNG, BMP, GIF, and PIC images are automatically depalettized. +// +// =========================================================================== +// +// UNICODE: +// +// If compiling for Windows and you wish to use Unicode filenames, compile +// with +// #define STBI_WINDOWS_UTF8 +// and pass utf8-encoded filenames. Call stbi_convert_wchar_to_utf8 to convert +// Windows wchar_t filenames to utf8. +// +// =========================================================================== +// +// Philosophy +// +// stb libraries are designed with the following priorities: +// +// 1. easy to use +// 2. easy to maintain +// 3. good performance +// +// Sometimes I let "good performance" creep up in priority over "easy to maintain", +// and for best performance I may provide less-easy-to-use APIs that give higher +// performance, in addition to the easy-to-use ones. Nevertheless, it's important +// to keep in mind that from the standpoint of you, a client of this library, +// all you care about is #1 and #3, and stb libraries DO NOT emphasize #3 above all. +// +// Some secondary priorities arise directly from the first two, some of which +// provide more explicit reasons why performance can't be emphasized. +// +// - Portable ("ease of use") +// - Small source code footprint ("easy to maintain") +// - No dependencies ("ease of use") +// +// =========================================================================== +// +// I/O callbacks +// +// I/O callbacks allow you to read from arbitrary sources, like packaged +// files or some other source. Data read from callbacks are processed +// through a small internal buffer (currently 128 bytes) to try to reduce +// overhead. +// +// The three functions you must define are "read" (reads some bytes of data), +// "skip" (skips some bytes of data), "eof" (reports if the stream is at the end). +// +// =========================================================================== +// +// SIMD support +// +// The JPEG decoder will try to automatically use SIMD kernels on x86 when +// supported by the compiler. For ARM Neon support, you must explicitly +// request it. +// +// (The old do-it-yourself SIMD API is no longer supported in the current +// code.) +// +// On x86, SSE2 will automatically be used when available based on a run-time +// test; if not, the generic C versions are used as a fall-back. On ARM targets, +// the typical path is to have separate builds for NEON and non-NEON devices +// (at least this is true for iOS and Android). Therefore, the NEON support is +// toggled by a build flag: define STBI_NEON to get NEON loops. +// +// If for some reason you do not want to use any of SIMD code, or if +// you have issues compiling it, you can disable it entirely by +// defining STBI_NO_SIMD. +// +// =========================================================================== +// +// HDR image support (disable by defining STBI_NO_HDR) +// +// stb_image supports loading HDR images in general, and currently the Radiance +// .HDR file format specifically. You can still load any file through the existing +// interface; if you attempt to load an HDR file, it will be automatically remapped +// to LDR, assuming gamma 2.2 and an arbitrary scale factor defaulting to 1; +// both of these constants can be reconfigured through this interface: +// +// stbi_hdr_to_ldr_gamma(2.2f); +// stbi_hdr_to_ldr_scale(1.0f); +// +// (note, do not use _inverse_ constants; stbi_image will invert them +// appropriately). +// +// Additionally, there is a new, parallel interface for loading files as +// (linear) floats to preserve the full dynamic range: +// +// float *data = stbi_loadf(filename, &x, &y, &n, 0); +// +// If you load LDR images through this interface, those images will +// be promoted to floating point values, run through the inverse of +// constants corresponding to the above: +// +// stbi_ldr_to_hdr_scale(1.0f); +// stbi_ldr_to_hdr_gamma(2.2f); +// +// Finally, given a filename (or an open file or memory block--see header +// file for details) containing image data, you can query for the "most +// appropriate" interface to use (that is, whether the image is HDR or +// not), using: +// +// stbi_is_hdr(char *filename); +// +// =========================================================================== +// +// iPhone PNG support: +// +// By default we convert iphone-formatted PNGs back to RGB, even though +// they are internally encoded differently. You can disable this conversion +// by calling stbi_convert_iphone_png_to_rgb(0), in which case +// you will always just get the native iphone "format" through (which +// is BGR stored in RGB). +// +// Call stbi_set_unpremultiply_on_load(1) as well to force a divide per +// pixel to remove any premultiplied alpha *only* if the image file explicitly +// says there's premultiplied data (currently only happens in iPhone images, +// and only if iPhone convert-to-rgb processing is on). +// +// =========================================================================== +// +// ADDITIONAL CONFIGURATION +// +// - You can suppress implementation of any of the decoders to reduce +// your code footprint by #defining one or more of the following +// symbols before creating the implementation. +// +// STBI_NO_JPEG +// STBI_NO_PNG +// STBI_NO_BMP +// STBI_NO_PSD +// STBI_NO_TGA +// STBI_NO_GIF +// STBI_NO_HDR +// STBI_NO_PIC +// STBI_NO_PNM (.ppm and .pgm) +// +// - You can request *only* certain decoders and suppress all other ones +// (this will be more forward-compatible, as addition of new decoders +// doesn't require you to disable them explicitly): +// +// STBI_ONLY_JPEG +// STBI_ONLY_PNG +// STBI_ONLY_BMP +// STBI_ONLY_PSD +// STBI_ONLY_TGA +// STBI_ONLY_GIF +// STBI_ONLY_HDR +// STBI_ONLY_PIC +// STBI_ONLY_PNM (.ppm and .pgm) +// +// - If you use STBI_NO_PNG (or _ONLY_ without PNG), and you still +// want the zlib decoder to be available, #define STBI_SUPPORT_ZLIB +// +// - If you define STBI_MAX_DIMENSIONS, stb_image will reject images greater +// than that size (in either width or height) without further processing. +// This is to let programs in the wild set an upper bound to prevent +// denial-of-service attacks on untrusted data, as one could generate a +// valid image of gigantic dimensions and force stb_image to allocate a +// huge block of memory and spend disproportionate time decoding it. By +// default this is set to (1 << 24), which is 16777216, but that's still +// very big. + +#ifndef STBI_NO_STDIO +#include +#endif // STBI_NO_STDIO + +#define STBI_VERSION 1 + +enum +{ + STBI_default = 0, // only used for desired_channels + + STBI_grey = 1, + STBI_grey_alpha = 2, + STBI_rgb = 3, + STBI_rgb_alpha = 4 +}; + +#include +typedef unsigned char stbi_uc; +typedef unsigned short stbi_us; + +#ifdef __cplusplus +extern "C" { +#endif + +#ifndef STBIDEF +#ifdef STB_IMAGE_STATIC +#define STBIDEF static +#else +#define STBIDEF extern +#endif +#endif + +////////////////////////////////////////////////////////////////////////////// +// +// PRIMARY API - works on images of any type +// + +// +// load image by filename, open file, or memory buffer +// + +typedef struct +{ + int (*read) (void *user,char *data,int size); // fill 'data' with 'size' bytes. return number of bytes actually read + void (*skip) (void *user,int n); // skip the next 'n' bytes, or 'unget' the last -n bytes if negative + int (*eof) (void *user); // returns nonzero if we are at end of file/data +} stbi_io_callbacks; + +//////////////////////////////////// +// +// 8-bits-per-channel interface +// + +STBIDEF stbi_uc *stbi_load_from_memory (stbi_uc const *buffer, int len , int *x, int *y, int *channels_in_file, int desired_channels); +STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk , void *user, int *x, int *y, int *channels_in_file, int desired_channels); + +#ifndef STBI_NO_STDIO +STBIDEF stbi_uc *stbi_load (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels); +STBIDEF stbi_uc *stbi_load_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels); +// for stbi_load_from_file, file pointer is left pointing immediately after image +#endif + +#ifndef STBI_NO_GIF +STBIDEF stbi_uc *stbi_load_gif_from_memory(stbi_uc const *buffer, int len, int **delays, int *x, int *y, int *z, int *comp, int req_comp); +#endif + +#ifdef STBI_WINDOWS_UTF8 +STBIDEF int stbi_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input); +#endif + +//////////////////////////////////// +// +// 16-bits-per-channel interface +// + +STBIDEF stbi_us *stbi_load_16_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels); +STBIDEF stbi_us *stbi_load_16_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels); + +#ifndef STBI_NO_STDIO +STBIDEF stbi_us *stbi_load_16 (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels); +STBIDEF stbi_us *stbi_load_from_file_16(FILE *f, int *x, int *y, int *channels_in_file, int desired_channels); +#endif + +//////////////////////////////////// +// +// float-per-channel interface +// +#ifndef STBI_NO_LINEAR + STBIDEF float *stbi_loadf_from_memory (stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels); + STBIDEF float *stbi_loadf_from_callbacks (stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels); + + #ifndef STBI_NO_STDIO + STBIDEF float *stbi_loadf (char const *filename, int *x, int *y, int *channels_in_file, int desired_channels); + STBIDEF float *stbi_loadf_from_file (FILE *f, int *x, int *y, int *channels_in_file, int desired_channels); + #endif +#endif + +#ifndef STBI_NO_HDR + STBIDEF void stbi_hdr_to_ldr_gamma(float gamma); + STBIDEF void stbi_hdr_to_ldr_scale(float scale); +#endif // STBI_NO_HDR + +#ifndef STBI_NO_LINEAR + STBIDEF void stbi_ldr_to_hdr_gamma(float gamma); + STBIDEF void stbi_ldr_to_hdr_scale(float scale); +#endif // STBI_NO_LINEAR + +// stbi_is_hdr is always defined, but always returns false if STBI_NO_HDR +STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user); +STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len); +#ifndef STBI_NO_STDIO +STBIDEF int stbi_is_hdr (char const *filename); +STBIDEF int stbi_is_hdr_from_file(FILE *f); +#endif // STBI_NO_STDIO + + +// get a VERY brief reason for failure +// on most compilers (and ALL modern mainstream compilers) this is threadsafe +STBIDEF const char *stbi_failure_reason (void); + +// free the loaded image -- this is just free() +STBIDEF void stbi_image_free (void *retval_from_stbi_load); + +// get image dimensions & components without fully decoding +STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp); +STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp); +STBIDEF int stbi_is_16_bit_from_memory(stbi_uc const *buffer, int len); +STBIDEF int stbi_is_16_bit_from_callbacks(stbi_io_callbacks const *clbk, void *user); + +#ifndef STBI_NO_STDIO +STBIDEF int stbi_info (char const *filename, int *x, int *y, int *comp); +STBIDEF int stbi_info_from_file (FILE *f, int *x, int *y, int *comp); +STBIDEF int stbi_is_16_bit (char const *filename); +STBIDEF int stbi_is_16_bit_from_file(FILE *f); +#endif + + + +// for image formats that explicitly notate that they have premultiplied alpha, +// we just return the colors as stored in the file. set this flag to force +// unpremultiplication. results are undefined if the unpremultiply overflow. +STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply); + +// indicate whether we should process iphone images back to canonical format, +// or just pass them through "as-is" +STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert); + +// flip the image vertically, so the first pixel in the output array is the bottom left +STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip); + +// as above, but only applies to images loaded on the thread that calls the function +// this function is only available if your compiler supports thread-local variables; +// calling it will fail to link if your compiler doesn't +STBIDEF void stbi_set_flip_vertically_on_load_thread(int flag_true_if_should_flip); + +// ZLIB client - used by PNG, available for other purposes + +STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen); +STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header); +STBIDEF char *stbi_zlib_decode_malloc(const char *buffer, int len, int *outlen); +STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, const char *ibuffer, int ilen); + +STBIDEF char *stbi_zlib_decode_noheader_malloc(const char *buffer, int len, int *outlen); +STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen); + + +#ifdef __cplusplus +} +#endif + +// +// +//// end header file ///////////////////////////////////////////////////// +#endif // STBI_INCLUDE_STB_IMAGE_H + +#ifdef STB_IMAGE_IMPLEMENTATION + +#if defined(STBI_ONLY_JPEG) || defined(STBI_ONLY_PNG) || defined(STBI_ONLY_BMP) \ + || defined(STBI_ONLY_TGA) || defined(STBI_ONLY_GIF) || defined(STBI_ONLY_PSD) \ + || defined(STBI_ONLY_HDR) || defined(STBI_ONLY_PIC) || defined(STBI_ONLY_PNM) \ + || defined(STBI_ONLY_ZLIB) + #ifndef STBI_ONLY_JPEG + #define STBI_NO_JPEG + #endif + #ifndef STBI_ONLY_PNG + #define STBI_NO_PNG + #endif + #ifndef STBI_ONLY_BMP + #define STBI_NO_BMP + #endif + #ifndef STBI_ONLY_PSD + #define STBI_NO_PSD + #endif + #ifndef STBI_ONLY_TGA + #define STBI_NO_TGA + #endif + #ifndef STBI_ONLY_GIF + #define STBI_NO_GIF + #endif + #ifndef STBI_ONLY_HDR + #define STBI_NO_HDR + #endif + #ifndef STBI_ONLY_PIC + #define STBI_NO_PIC + #endif + #ifndef STBI_ONLY_PNM + #define STBI_NO_PNM + #endif +#endif + +#if defined(STBI_NO_PNG) && !defined(STBI_SUPPORT_ZLIB) && !defined(STBI_NO_ZLIB) +#define STBI_NO_ZLIB +#endif + + +#include +#include // ptrdiff_t on osx +#include +#include +#include + +#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) +#include // ldexp, pow +#endif + +#ifndef STBI_NO_STDIO +#include +#endif + +#ifndef STBI_ASSERT +#include +#define STBI_ASSERT(x) assert(x) +#endif + +#ifdef __cplusplus +#define STBI_EXTERN extern "C" +#else +#define STBI_EXTERN extern +#endif + + +#ifndef _MSC_VER + #ifdef __cplusplus + #define stbi_inline inline + #else + #define stbi_inline + #endif +#else + #define stbi_inline __forceinline +#endif + +#ifndef STBI_NO_THREAD_LOCALS + #if defined(__cplusplus) && __cplusplus >= 201103L + #define STBI_THREAD_LOCAL thread_local + #elif defined(__GNUC__) && __GNUC__ < 5 + #define STBI_THREAD_LOCAL __thread + #elif defined(_MSC_VER) + #define STBI_THREAD_LOCAL __declspec(thread) + #elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 201112L && !defined(__STDC_NO_THREADS__) + #define STBI_THREAD_LOCAL _Thread_local + #endif + + #ifndef STBI_THREAD_LOCAL + #if defined(__GNUC__) + #define STBI_THREAD_LOCAL __thread + #endif + #endif +#endif + +#ifdef _MSC_VER +typedef unsigned short stbi__uint16; +typedef signed short stbi__int16; +typedef unsigned int stbi__uint32; +typedef signed int stbi__int32; +#else +#include +typedef uint16_t stbi__uint16; +typedef int16_t stbi__int16; +typedef uint32_t stbi__uint32; +typedef int32_t stbi__int32; +#endif + +// should produce compiler error if size is wrong +typedef unsigned char validate_uint32[sizeof(stbi__uint32)==4 ? 1 : -1]; + +#ifdef _MSC_VER +#define STBI_NOTUSED(v) (void)(v) +#else +#define STBI_NOTUSED(v) (void)sizeof(v) +#endif + +#ifdef _MSC_VER +#define STBI_HAS_LROTL +#endif + +#ifdef STBI_HAS_LROTL + #define stbi_lrot(x,y) _lrotl(x,y) +#else + #define stbi_lrot(x,y) (((x) << (y)) | ((x) >> (32 - (y)))) +#endif + +#if defined(STBI_MALLOC) && defined(STBI_FREE) && (defined(STBI_REALLOC) || defined(STBI_REALLOC_SIZED)) +// ok +#elif !defined(STBI_MALLOC) && !defined(STBI_FREE) && !defined(STBI_REALLOC) && !defined(STBI_REALLOC_SIZED) +// ok +#else +#error "Must define all or none of STBI_MALLOC, STBI_FREE, and STBI_REALLOC (or STBI_REALLOC_SIZED)." +#endif + +#ifndef STBI_MALLOC +#define STBI_MALLOC(sz) malloc(sz) +#define STBI_REALLOC(p,newsz) realloc(p,newsz) +#define STBI_FREE(p) free(p) +#endif + +#ifndef STBI_REALLOC_SIZED +#define STBI_REALLOC_SIZED(p,oldsz,newsz) STBI_REALLOC(p,newsz) +#endif + +// x86/x64 detection +#if defined(__x86_64__) || defined(_M_X64) +#define STBI__X64_TARGET +#elif defined(__i386) || defined(_M_IX86) +#define STBI__X86_TARGET +#endif + +#if defined(__GNUC__) && defined(STBI__X86_TARGET) && !defined(__SSE2__) && !defined(STBI_NO_SIMD) +// gcc doesn't support sse2 intrinsics unless you compile with -msse2, +// which in turn means it gets to use SSE2 everywhere. This is unfortunate, +// but previous attempts to provide the SSE2 functions with runtime +// detection caused numerous issues. The way architecture extensions are +// exposed in GCC/Clang is, sadly, not really suited for one-file libs. +// New behavior: if compiled with -msse2, we use SSE2 without any +// detection; if not, we don't use it at all. +#define STBI_NO_SIMD +#endif + +#if defined(__MINGW32__) && defined(STBI__X86_TARGET) && !defined(STBI_MINGW_ENABLE_SSE2) && !defined(STBI_NO_SIMD) +// Note that __MINGW32__ doesn't actually mean 32-bit, so we have to avoid STBI__X64_TARGET +// +// 32-bit MinGW wants ESP to be 16-byte aligned, but this is not in the +// Windows ABI and VC++ as well as Windows DLLs don't maintain that invariant. +// As a result, enabling SSE2 on 32-bit MinGW is dangerous when not +// simultaneously enabling "-mstackrealign". +// +// See https://github.com/nothings/stb/issues/81 for more information. +// +// So default to no SSE2 on 32-bit MinGW. If you've read this far and added +// -mstackrealign to your build settings, feel free to #define STBI_MINGW_ENABLE_SSE2. +#define STBI_NO_SIMD +#endif + +#if !defined(STBI_NO_SIMD) && (defined(STBI__X86_TARGET) || defined(STBI__X64_TARGET)) +#define STBI_SSE2 +#include + +#ifdef _MSC_VER + +#if _MSC_VER >= 1400 // not VC6 +#include // __cpuid +static int stbi__cpuid3(void) +{ + int info[4]; + __cpuid(info,1); + return info[3]; +} +#else +static int stbi__cpuid3(void) +{ + int res; + __asm { + mov eax,1 + cpuid + mov res,edx + } + return res; +} +#endif + +#define STBI_SIMD_ALIGN(type, name) __declspec(align(16)) type name + +#if !defined(STBI_NO_JPEG) && defined(STBI_SSE2) +static int stbi__sse2_available(void) +{ + int info3 = stbi__cpuid3(); + return ((info3 >> 26) & 1) != 0; +} +#endif + +#else // assume GCC-style if not VC++ +#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16))) + +#if !defined(STBI_NO_JPEG) && defined(STBI_SSE2) +static int stbi__sse2_available(void) +{ + // If we're even attempting to compile this on GCC/Clang, that means + // -msse2 is on, which means the compiler is allowed to use SSE2 + // instructions at will, and so are we. + return 1; +} +#endif + +#endif +#endif + +// ARM NEON +#if defined(STBI_NO_SIMD) && defined(STBI_NEON) +#undef STBI_NEON +#endif + +#ifdef STBI_NEON +#include +// assume GCC or Clang on ARM targets +#define STBI_SIMD_ALIGN(type, name) type name __attribute__((aligned(16))) +#endif + +#ifndef STBI_SIMD_ALIGN +#define STBI_SIMD_ALIGN(type, name) type name +#endif + +#ifndef STBI_MAX_DIMENSIONS +#define STBI_MAX_DIMENSIONS (1 << 24) +#endif + +/////////////////////////////////////////////// +// +// stbi__context struct and start_xxx functions + +// stbi__context structure is our basic context used by all images, so it +// contains all the IO context, plus some basic image information +typedef struct +{ + stbi__uint32 img_x, img_y; + int img_n, img_out_n; + + stbi_io_callbacks io; + void *io_user_data; + + int read_from_callbacks; + int buflen; + stbi_uc buffer_start[128]; + int callback_already_read; + + stbi_uc *img_buffer, *img_buffer_end; + stbi_uc *img_buffer_original, *img_buffer_original_end; +} stbi__context; + + +static void stbi__refill_buffer(stbi__context *s); + +// initialize a memory-decode context +static void stbi__start_mem(stbi__context *s, stbi_uc const *buffer, int len) +{ + s->io.read = NULL; + s->read_from_callbacks = 0; + s->callback_already_read = 0; + s->img_buffer = s->img_buffer_original = (stbi_uc *) buffer; + s->img_buffer_end = s->img_buffer_original_end = (stbi_uc *) buffer+len; +} + +// initialize a callback-based context +static void stbi__start_callbacks(stbi__context *s, stbi_io_callbacks *c, void *user) +{ + s->io = *c; + s->io_user_data = user; + s->buflen = sizeof(s->buffer_start); + s->read_from_callbacks = 1; + s->callback_already_read = 0; + s->img_buffer = s->img_buffer_original = s->buffer_start; + stbi__refill_buffer(s); + s->img_buffer_original_end = s->img_buffer_end; +} + +#ifndef STBI_NO_STDIO + +static int stbi__stdio_read(void *user, char *data, int size) +{ + return (int) fread(data,1,size,(FILE*) user); +} + +static void stbi__stdio_skip(void *user, int n) +{ + int ch; + fseek((FILE*) user, n, SEEK_CUR); + ch = fgetc((FILE*) user); /* have to read a byte to reset feof()'s flag */ + if (ch != EOF) { + ungetc(ch, (FILE *) user); /* push byte back onto stream if valid. */ + } +} + +static int stbi__stdio_eof(void *user) +{ + return feof((FILE*) user) || ferror((FILE *) user); +} + +static stbi_io_callbacks stbi__stdio_callbacks = +{ + stbi__stdio_read, + stbi__stdio_skip, + stbi__stdio_eof, +}; + +static void stbi__start_file(stbi__context *s, FILE *f) +{ + stbi__start_callbacks(s, &stbi__stdio_callbacks, (void *) f); +} + +//static void stop_file(stbi__context *s) { } + +#endif // !STBI_NO_STDIO + +static void stbi__rewind(stbi__context *s) +{ + // conceptually rewind SHOULD rewind to the beginning of the stream, + // but we just rewind to the beginning of the initial buffer, because + // we only use it after doing 'test', which only ever looks at at most 92 bytes + s->img_buffer = s->img_buffer_original; + s->img_buffer_end = s->img_buffer_original_end; +} + +enum +{ + STBI_ORDER_RGB, + STBI_ORDER_BGR +}; + +typedef struct +{ + int bits_per_channel; + int num_channels; + int channel_order; +} stbi__result_info; + +#ifndef STBI_NO_JPEG +static int stbi__jpeg_test(stbi__context *s); +static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_PNG +static int stbi__png_test(stbi__context *s); +static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp); +static int stbi__png_is16(stbi__context *s); +#endif + +#ifndef STBI_NO_BMP +static int stbi__bmp_test(stbi__context *s); +static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_TGA +static int stbi__tga_test(stbi__context *s); +static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_PSD +static int stbi__psd_test(stbi__context *s); +static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc); +static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp); +static int stbi__psd_is16(stbi__context *s); +#endif + +#ifndef STBI_NO_HDR +static int stbi__hdr_test(stbi__context *s); +static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_PIC +static int stbi__pic_test(stbi__context *s); +static void *stbi__pic_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_GIF +static int stbi__gif_test(stbi__context *s); +static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static void *stbi__load_gif_main(stbi__context *s, int **delays, int *x, int *y, int *z, int *comp, int req_comp); +static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +#ifndef STBI_NO_PNM +static int stbi__pnm_test(stbi__context *s); +static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri); +static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp); +#endif + +static +#ifdef STBI_THREAD_LOCAL +STBI_THREAD_LOCAL +#endif +const char *stbi__g_failure_reason; + +STBIDEF const char *stbi_failure_reason(void) +{ + return stbi__g_failure_reason; +} + +#ifndef STBI_NO_FAILURE_STRINGS +static int stbi__err(const char *str) +{ + stbi__g_failure_reason = str; + return 0; +} +#endif + +static void *stbi__malloc(size_t size) +{ + return STBI_MALLOC(size); +} + +// stb_image uses ints pervasively, including for offset calculations. +// therefore the largest decoded image size we can support with the +// current code, even on 64-bit targets, is INT_MAX. this is not a +// significant limitation for the intended use case. +// +// we do, however, need to make sure our size calculations don't +// overflow. hence a few helper functions for size calculations that +// multiply integers together, making sure that they're non-negative +// and no overflow occurs. + +// return 1 if the sum is valid, 0 on overflow. +// negative terms are considered invalid. +static int stbi__addsizes_valid(int a, int b) +{ + if (b < 0) return 0; + // now 0 <= b <= INT_MAX, hence also + // 0 <= INT_MAX - b <= INTMAX. + // And "a + b <= INT_MAX" (which might overflow) is the + // same as a <= INT_MAX - b (no overflow) + return a <= INT_MAX - b; +} + +// returns 1 if the product is valid, 0 on overflow. +// negative factors are considered invalid. +static int stbi__mul2sizes_valid(int a, int b) +{ + if (a < 0 || b < 0) return 0; + if (b == 0) return 1; // mul-by-0 is always safe + // portable way to check for no overflows in a*b + return a <= INT_MAX/b; +} + +#if !defined(STBI_NO_JPEG) || !defined(STBI_NO_PNG) || !defined(STBI_NO_TGA) || !defined(STBI_NO_HDR) +// returns 1 if "a*b + add" has no negative terms/factors and doesn't overflow +static int stbi__mad2sizes_valid(int a, int b, int add) +{ + return stbi__mul2sizes_valid(a, b) && stbi__addsizes_valid(a*b, add); +} +#endif + +// returns 1 if "a*b*c + add" has no negative terms/factors and doesn't overflow +static int stbi__mad3sizes_valid(int a, int b, int c, int add) +{ + return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) && + stbi__addsizes_valid(a*b*c, add); +} + +// returns 1 if "a*b*c*d + add" has no negative terms/factors and doesn't overflow +#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) +static int stbi__mad4sizes_valid(int a, int b, int c, int d, int add) +{ + return stbi__mul2sizes_valid(a, b) && stbi__mul2sizes_valid(a*b, c) && + stbi__mul2sizes_valid(a*b*c, d) && stbi__addsizes_valid(a*b*c*d, add); +} +#endif + +#if !defined(STBI_NO_JPEG) || !defined(STBI_NO_PNG) || !defined(STBI_NO_TGA) || !defined(STBI_NO_HDR) +// mallocs with size overflow checking +static void *stbi__malloc_mad2(int a, int b, int add) +{ + if (!stbi__mad2sizes_valid(a, b, add)) return NULL; + return stbi__malloc(a*b + add); +} +#endif + +static void *stbi__malloc_mad3(int a, int b, int c, int add) +{ + if (!stbi__mad3sizes_valid(a, b, c, add)) return NULL; + return stbi__malloc(a*b*c + add); +} + +#if !defined(STBI_NO_LINEAR) || !defined(STBI_NO_HDR) +static void *stbi__malloc_mad4(int a, int b, int c, int d, int add) +{ + if (!stbi__mad4sizes_valid(a, b, c, d, add)) return NULL; + return stbi__malloc(a*b*c*d + add); +} +#endif + +// stbi__err - error +// stbi__errpf - error returning pointer to float +// stbi__errpuc - error returning pointer to unsigned char + +#ifdef STBI_NO_FAILURE_STRINGS + #define stbi__err(x,y) 0 +#elif defined(STBI_FAILURE_USERMSG) + #define stbi__err(x,y) stbi__err(y) +#else + #define stbi__err(x,y) stbi__err(x) +#endif + +#define stbi__errpf(x,y) ((float *)(size_t) (stbi__err(x,y)?NULL:NULL)) +#define stbi__errpuc(x,y) ((unsigned char *)(size_t) (stbi__err(x,y)?NULL:NULL)) + +STBIDEF void stbi_image_free(void *retval_from_stbi_load) +{ + STBI_FREE(retval_from_stbi_load); +} + +#ifndef STBI_NO_LINEAR +static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp); +#endif + +#ifndef STBI_NO_HDR +static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp); +#endif + +static int stbi__vertically_flip_on_load_global = 0; + +STBIDEF void stbi_set_flip_vertically_on_load(int flag_true_if_should_flip) +{ + stbi__vertically_flip_on_load_global = flag_true_if_should_flip; +} + +#ifndef STBI_THREAD_LOCAL +#define stbi__vertically_flip_on_load stbi__vertically_flip_on_load_global +#else +static STBI_THREAD_LOCAL int stbi__vertically_flip_on_load_local, stbi__vertically_flip_on_load_set; + +STBIDEF void stbi_set_flip_vertically_on_load_thread(int flag_true_if_should_flip) +{ + stbi__vertically_flip_on_load_local = flag_true_if_should_flip; + stbi__vertically_flip_on_load_set = 1; +} + +#define stbi__vertically_flip_on_load (stbi__vertically_flip_on_load_set \ + ? stbi__vertically_flip_on_load_local \ + : stbi__vertically_flip_on_load_global) +#endif // STBI_THREAD_LOCAL + +static void *stbi__load_main(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc) +{ + memset(ri, 0, sizeof(*ri)); // make sure it's initialized if we add new fields + ri->bits_per_channel = 8; // default is 8 so most paths don't have to be changed + ri->channel_order = STBI_ORDER_RGB; // all current input & output are this, but this is here so we can add BGR order + ri->num_channels = 0; + + #ifndef STBI_NO_JPEG + if (stbi__jpeg_test(s)) return stbi__jpeg_load(s,x,y,comp,req_comp, ri); + #endif + #ifndef STBI_NO_PNG + if (stbi__png_test(s)) return stbi__png_load(s,x,y,comp,req_comp, ri); + #endif + #ifndef STBI_NO_BMP + if (stbi__bmp_test(s)) return stbi__bmp_load(s,x,y,comp,req_comp, ri); + #endif + #ifndef STBI_NO_GIF + if (stbi__gif_test(s)) return stbi__gif_load(s,x,y,comp,req_comp, ri); + #endif + #ifndef STBI_NO_PSD + if (stbi__psd_test(s)) return stbi__psd_load(s,x,y,comp,req_comp, ri, bpc); + #else + STBI_NOTUSED(bpc); + #endif + #ifndef STBI_NO_PIC + if (stbi__pic_test(s)) return stbi__pic_load(s,x,y,comp,req_comp, ri); + #endif + #ifndef STBI_NO_PNM + if (stbi__pnm_test(s)) return stbi__pnm_load(s,x,y,comp,req_comp, ri); + #endif + + #ifndef STBI_NO_HDR + if (stbi__hdr_test(s)) { + float *hdr = stbi__hdr_load(s, x,y,comp,req_comp, ri); + return stbi__hdr_to_ldr(hdr, *x, *y, req_comp ? req_comp : *comp); + } + #endif + + #ifndef STBI_NO_TGA + // test tga last because it's a crappy test! + if (stbi__tga_test(s)) + return stbi__tga_load(s,x,y,comp,req_comp, ri); + #endif + + return stbi__errpuc("unknown image type", "Image not of any known type, or corrupt"); +} + +static stbi_uc *stbi__convert_16_to_8(stbi__uint16 *orig, int w, int h, int channels) +{ + int i; + int img_len = w * h * channels; + stbi_uc *reduced; + + reduced = (stbi_uc *) stbi__malloc(img_len); + if (reduced == NULL) return stbi__errpuc("outofmem", "Out of memory"); + + for (i = 0; i < img_len; ++i) + reduced[i] = (stbi_uc)((orig[i] >> 8) & 0xFF); // top half of each byte is sufficient approx of 16->8 bit scaling + + STBI_FREE(orig); + return reduced; +} + +static stbi__uint16 *stbi__convert_8_to_16(stbi_uc *orig, int w, int h, int channels) +{ + int i; + int img_len = w * h * channels; + stbi__uint16 *enlarged; + + enlarged = (stbi__uint16 *) stbi__malloc(img_len*2); + if (enlarged == NULL) return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory"); + + for (i = 0; i < img_len; ++i) + enlarged[i] = (stbi__uint16)((orig[i] << 8) + orig[i]); // replicate to high and low byte, maps 0->0, 255->0xffff + + STBI_FREE(orig); + return enlarged; +} + +static void stbi__vertical_flip(void *image, int w, int h, int bytes_per_pixel) +{ + int row; + size_t bytes_per_row = (size_t)w * bytes_per_pixel; + stbi_uc temp[2048]; + stbi_uc *bytes = (stbi_uc *)image; + + for (row = 0; row < (h>>1); row++) { + stbi_uc *row0 = bytes + row*bytes_per_row; + stbi_uc *row1 = bytes + (h - row - 1)*bytes_per_row; + // swap row0 with row1 + size_t bytes_left = bytes_per_row; + while (bytes_left) { + size_t bytes_copy = (bytes_left < sizeof(temp)) ? bytes_left : sizeof(temp); + memcpy(temp, row0, bytes_copy); + memcpy(row0, row1, bytes_copy); + memcpy(row1, temp, bytes_copy); + row0 += bytes_copy; + row1 += bytes_copy; + bytes_left -= bytes_copy; + } + } +} + +#ifndef STBI_NO_GIF +static void stbi__vertical_flip_slices(void *image, int w, int h, int z, int bytes_per_pixel) +{ + int slice; + int slice_size = w * h * bytes_per_pixel; + + stbi_uc *bytes = (stbi_uc *)image; + for (slice = 0; slice < z; ++slice) { + stbi__vertical_flip(bytes, w, h, bytes_per_pixel); + bytes += slice_size; + } +} +#endif + +static unsigned char *stbi__load_and_postprocess_8bit(stbi__context *s, int *x, int *y, int *comp, int req_comp) +{ + stbi__result_info ri; + void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 8); + + if (result == NULL) + return NULL; + + // it is the responsibility of the loaders to make sure we get either 8 or 16 bit. + STBI_ASSERT(ri.bits_per_channel == 8 || ri.bits_per_channel == 16); + + if (ri.bits_per_channel != 8) { + result = stbi__convert_16_to_8((stbi__uint16 *) result, *x, *y, req_comp == 0 ? *comp : req_comp); + ri.bits_per_channel = 8; + } + + // @TODO: move stbi__convert_format to here + + if (stbi__vertically_flip_on_load) { + int channels = req_comp ? req_comp : *comp; + stbi__vertical_flip(result, *x, *y, channels * sizeof(stbi_uc)); + } + + return (unsigned char *) result; +} + +static stbi__uint16 *stbi__load_and_postprocess_16bit(stbi__context *s, int *x, int *y, int *comp, int req_comp) +{ + stbi__result_info ri; + void *result = stbi__load_main(s, x, y, comp, req_comp, &ri, 16); + + if (result == NULL) + return NULL; + + // it is the responsibility of the loaders to make sure we get either 8 or 16 bit. + STBI_ASSERT(ri.bits_per_channel == 8 || ri.bits_per_channel == 16); + + if (ri.bits_per_channel != 16) { + result = stbi__convert_8_to_16((stbi_uc *) result, *x, *y, req_comp == 0 ? *comp : req_comp); + ri.bits_per_channel = 16; + } + + // @TODO: move stbi__convert_format16 to here + // @TODO: special case RGB-to-Y (and RGBA-to-YA) for 8-bit-to-16-bit case to keep more precision + + if (stbi__vertically_flip_on_load) { + int channels = req_comp ? req_comp : *comp; + stbi__vertical_flip(result, *x, *y, channels * sizeof(stbi__uint16)); + } + + return (stbi__uint16 *) result; +} + +#if !defined(STBI_NO_HDR) && !defined(STBI_NO_LINEAR) +static void stbi__float_postprocess(float *result, int *x, int *y, int *comp, int req_comp) +{ + if (stbi__vertically_flip_on_load && result != NULL) { + int channels = req_comp ? req_comp : *comp; + stbi__vertical_flip(result, *x, *y, channels * sizeof(float)); + } +} +#endif + +#ifndef STBI_NO_STDIO + +#if defined(_MSC_VER) && defined(STBI_WINDOWS_UTF8) +STBI_EXTERN __declspec(dllimport) int __stdcall MultiByteToWideChar(unsigned int cp, unsigned long flags, const char *str, int cbmb, wchar_t *widestr, int cchwide); +STBI_EXTERN __declspec(dllimport) int __stdcall WideCharToMultiByte(unsigned int cp, unsigned long flags, const wchar_t *widestr, int cchwide, char *str, int cbmb, const char *defchar, int *used_default); +#endif + +#if defined(_MSC_VER) && defined(STBI_WINDOWS_UTF8) +STBIDEF int stbi_convert_wchar_to_utf8(char *buffer, size_t bufferlen, const wchar_t* input) +{ + return WideCharToMultiByte(65001 /* UTF8 */, 0, input, -1, buffer, (int) bufferlen, NULL, NULL); +} +#endif + +static FILE *stbi__fopen(char const *filename, char const *mode) +{ + FILE *f; +#if defined(_MSC_VER) && defined(STBI_WINDOWS_UTF8) + wchar_t wMode[64]; + wchar_t wFilename[1024]; + if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, filename, -1, wFilename, sizeof(wFilename))) + return 0; + + if (0 == MultiByteToWideChar(65001 /* UTF8 */, 0, mode, -1, wMode, sizeof(wMode))) + return 0; + +#if _MSC_VER >= 1400 + if (0 != _wfopen_s(&f, wFilename, wMode)) + f = 0; +#else + f = _wfopen(wFilename, wMode); +#endif + +#elif defined(_MSC_VER) && _MSC_VER >= 1400 + if (0 != fopen_s(&f, filename, mode)) + f=0; +#else + f = fopen(filename, mode); +#endif + return f; +} + + +STBIDEF stbi_uc *stbi_load(char const *filename, int *x, int *y, int *comp, int req_comp) +{ + FILE *f = stbi__fopen(filename, "rb"); + unsigned char *result; + if (!f) return stbi__errpuc("can't fopen", "Unable to open file"); + result = stbi_load_from_file(f,x,y,comp,req_comp); + fclose(f); + return result; +} + +STBIDEF stbi_uc *stbi_load_from_file(FILE *f, int *x, int *y, int *comp, int req_comp) +{ + unsigned char *result; + stbi__context s; + stbi__start_file(&s,f); + result = stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp); + if (result) { + // need to 'unget' all the characters in the IO buffer + fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR); + } + return result; +} + +STBIDEF stbi__uint16 *stbi_load_from_file_16(FILE *f, int *x, int *y, int *comp, int req_comp) +{ + stbi__uint16 *result; + stbi__context s; + stbi__start_file(&s,f); + result = stbi__load_and_postprocess_16bit(&s,x,y,comp,req_comp); + if (result) { + // need to 'unget' all the characters in the IO buffer + fseek(f, - (int) (s.img_buffer_end - s.img_buffer), SEEK_CUR); + } + return result; +} + +STBIDEF stbi_us *stbi_load_16(char const *filename, int *x, int *y, int *comp, int req_comp) +{ + FILE *f = stbi__fopen(filename, "rb"); + stbi__uint16 *result; + if (!f) return (stbi_us *) stbi__errpuc("can't fopen", "Unable to open file"); + result = stbi_load_from_file_16(f,x,y,comp,req_comp); + fclose(f); + return result; +} + + +#endif //!STBI_NO_STDIO + +STBIDEF stbi_us *stbi_load_16_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *channels_in_file, int desired_channels) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__load_and_postprocess_16bit(&s,x,y,channels_in_file,desired_channels); +} + +STBIDEF stbi_us *stbi_load_16_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *channels_in_file, int desired_channels) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *)clbk, user); + return stbi__load_and_postprocess_16bit(&s,x,y,channels_in_file,desired_channels); +} + +STBIDEF stbi_uc *stbi_load_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp); +} + +STBIDEF stbi_uc *stbi_load_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user); + return stbi__load_and_postprocess_8bit(&s,x,y,comp,req_comp); +} + +#ifndef STBI_NO_GIF +STBIDEF stbi_uc *stbi_load_gif_from_memory(stbi_uc const *buffer, int len, int **delays, int *x, int *y, int *z, int *comp, int req_comp) +{ + unsigned char *result; + stbi__context s; + stbi__start_mem(&s,buffer,len); + + result = (unsigned char*) stbi__load_gif_main(&s, delays, x, y, z, comp, req_comp); + if (stbi__vertically_flip_on_load) { + stbi__vertical_flip_slices( result, *x, *y, *z, *comp ); + } + + return result; +} +#endif + +#ifndef STBI_NO_LINEAR +static float *stbi__loadf_main(stbi__context *s, int *x, int *y, int *comp, int req_comp) +{ + unsigned char *data; + #ifndef STBI_NO_HDR + if (stbi__hdr_test(s)) { + stbi__result_info ri; + float *hdr_data = stbi__hdr_load(s,x,y,comp,req_comp, &ri); + if (hdr_data) + stbi__float_postprocess(hdr_data,x,y,comp,req_comp); + return hdr_data; + } + #endif + data = stbi__load_and_postprocess_8bit(s, x, y, comp, req_comp); + if (data) + return stbi__ldr_to_hdr(data, *x, *y, req_comp ? req_comp : *comp); + return stbi__errpf("unknown image type", "Image not of any known type, or corrupt"); +} + +STBIDEF float *stbi_loadf_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__loadf_main(&s,x,y,comp,req_comp); +} + +STBIDEF float *stbi_loadf_from_callbacks(stbi_io_callbacks const *clbk, void *user, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user); + return stbi__loadf_main(&s,x,y,comp,req_comp); +} + +#ifndef STBI_NO_STDIO +STBIDEF float *stbi_loadf(char const *filename, int *x, int *y, int *comp, int req_comp) +{ + float *result; + FILE *f = stbi__fopen(filename, "rb"); + if (!f) return stbi__errpf("can't fopen", "Unable to open file"); + result = stbi_loadf_from_file(f,x,y,comp,req_comp); + fclose(f); + return result; +} + +STBIDEF float *stbi_loadf_from_file(FILE *f, int *x, int *y, int *comp, int req_comp) +{ + stbi__context s; + stbi__start_file(&s,f); + return stbi__loadf_main(&s,x,y,comp,req_comp); +} +#endif // !STBI_NO_STDIO + +#endif // !STBI_NO_LINEAR + +// these is-hdr-or-not is defined independent of whether STBI_NO_LINEAR is +// defined, for API simplicity; if STBI_NO_LINEAR is defined, it always +// reports false! + +STBIDEF int stbi_is_hdr_from_memory(stbi_uc const *buffer, int len) +{ + #ifndef STBI_NO_HDR + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__hdr_test(&s); + #else + STBI_NOTUSED(buffer); + STBI_NOTUSED(len); + return 0; + #endif +} + +#ifndef STBI_NO_STDIO +STBIDEF int stbi_is_hdr (char const *filename) +{ + FILE *f = stbi__fopen(filename, "rb"); + int result=0; + if (f) { + result = stbi_is_hdr_from_file(f); + fclose(f); + } + return result; +} + +STBIDEF int stbi_is_hdr_from_file(FILE *f) +{ + #ifndef STBI_NO_HDR + long pos = ftell(f); + int res; + stbi__context s; + stbi__start_file(&s,f); + res = stbi__hdr_test(&s); + fseek(f, pos, SEEK_SET); + return res; + #else + STBI_NOTUSED(f); + return 0; + #endif +} +#endif // !STBI_NO_STDIO + +STBIDEF int stbi_is_hdr_from_callbacks(stbi_io_callbacks const *clbk, void *user) +{ + #ifndef STBI_NO_HDR + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) clbk, user); + return stbi__hdr_test(&s); + #else + STBI_NOTUSED(clbk); + STBI_NOTUSED(user); + return 0; + #endif +} + +#ifndef STBI_NO_LINEAR +static float stbi__l2h_gamma=2.2f, stbi__l2h_scale=1.0f; + +STBIDEF void stbi_ldr_to_hdr_gamma(float gamma) { stbi__l2h_gamma = gamma; } +STBIDEF void stbi_ldr_to_hdr_scale(float scale) { stbi__l2h_scale = scale; } +#endif + +static float stbi__h2l_gamma_i=1.0f/2.2f, stbi__h2l_scale_i=1.0f; + +STBIDEF void stbi_hdr_to_ldr_gamma(float gamma) { stbi__h2l_gamma_i = 1/gamma; } +STBIDEF void stbi_hdr_to_ldr_scale(float scale) { stbi__h2l_scale_i = 1/scale; } + + +////////////////////////////////////////////////////////////////////////////// +// +// Common code used by all image loaders +// + +enum +{ + STBI__SCAN_load=0, + STBI__SCAN_type, + STBI__SCAN_header +}; + +static void stbi__refill_buffer(stbi__context *s) +{ + int n = (s->io.read)(s->io_user_data,(char*)s->buffer_start,s->buflen); + s->callback_already_read += (int) (s->img_buffer - s->img_buffer_original); + if (n == 0) { + // at end of file, treat same as if from memory, but need to handle case + // where s->img_buffer isn't pointing to safe memory, e.g. 0-byte file + s->read_from_callbacks = 0; + s->img_buffer = s->buffer_start; + s->img_buffer_end = s->buffer_start+1; + *s->img_buffer = 0; + } else { + s->img_buffer = s->buffer_start; + s->img_buffer_end = s->buffer_start + n; + } +} + +stbi_inline static stbi_uc stbi__get8(stbi__context *s) +{ + if (s->img_buffer < s->img_buffer_end) + return *s->img_buffer++; + if (s->read_from_callbacks) { + stbi__refill_buffer(s); + return *s->img_buffer++; + } + return 0; +} + +#if defined(STBI_NO_JPEG) && defined(STBI_NO_HDR) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM) +// nothing +#else +stbi_inline static int stbi__at_eof(stbi__context *s) +{ + if (s->io.read) { + if (!(s->io.eof)(s->io_user_data)) return 0; + // if feof() is true, check if buffer = end + // special case: we've only got the special 0 character at the end + if (s->read_from_callbacks == 0) return 1; + } + + return s->img_buffer >= s->img_buffer_end; +} +#endif + +#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) +// nothing +#else +static void stbi__skip(stbi__context *s, int n) +{ + if (n == 0) return; // already there! + if (n < 0) { + s->img_buffer = s->img_buffer_end; + return; + } + if (s->io.read) { + int blen = (int) (s->img_buffer_end - s->img_buffer); + if (blen < n) { + s->img_buffer = s->img_buffer_end; + (s->io.skip)(s->io_user_data, n - blen); + return; + } + } + s->img_buffer += n; +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_TGA) && defined(STBI_NO_HDR) && defined(STBI_NO_PNM) +// nothing +#else +static int stbi__getn(stbi__context *s, stbi_uc *buffer, int n) +{ + if (s->io.read) { + int blen = (int) (s->img_buffer_end - s->img_buffer); + if (blen < n) { + int res, count; + + memcpy(buffer, s->img_buffer, blen); + + count = (s->io.read)(s->io_user_data, (char*) buffer + blen, n - blen); + res = (count == (n-blen)); + s->img_buffer = s->img_buffer_end; + return res; + } + } + + if (s->img_buffer+n <= s->img_buffer_end) { + memcpy(buffer, s->img_buffer, n); + s->img_buffer += n; + return 1; + } else + return 0; +} +#endif + +#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_PSD) && defined(STBI_NO_PIC) +// nothing +#else +static int stbi__get16be(stbi__context *s) +{ + int z = stbi__get8(s); + return (z << 8) + stbi__get8(s); +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD) && defined(STBI_NO_PIC) +// nothing +#else +static stbi__uint32 stbi__get32be(stbi__context *s) +{ + stbi__uint32 z = stbi__get16be(s); + return (z << 16) + stbi__get16be(s); +} +#endif + +#if defined(STBI_NO_BMP) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) +// nothing +#else +static int stbi__get16le(stbi__context *s) +{ + int z = stbi__get8(s); + return z + (stbi__get8(s) << 8); +} +#endif + +#ifndef STBI_NO_BMP +static stbi__uint32 stbi__get32le(stbi__context *s) +{ + stbi__uint32 z = stbi__get16le(s); + return z + (stbi__get16le(s) << 16); +} +#endif + +#define STBI__BYTECAST(x) ((stbi_uc) ((x) & 255)) // truncate int to byte without warnings + +#if defined(STBI_NO_JPEG) && defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM) +// nothing +#else +////////////////////////////////////////////////////////////////////////////// +// +// generic converter from built-in img_n to req_comp +// individual types do this automatically as much as possible (e.g. jpeg +// does all cases internally since it needs to colorspace convert anyway, +// and it never has alpha, so very few cases ). png can automatically +// interleave an alpha=255 channel, but falls back to this for other cases +// +// assume data buffer is malloced, so malloc a new one and free that one +// only failure mode is malloc failing + +static stbi_uc stbi__compute_y(int r, int g, int b) +{ + return (stbi_uc) (((r*77) + (g*150) + (29*b)) >> 8); +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_BMP) && defined(STBI_NO_PSD) && defined(STBI_NO_TGA) && defined(STBI_NO_GIF) && defined(STBI_NO_PIC) && defined(STBI_NO_PNM) +// nothing +#else +static unsigned char *stbi__convert_format(unsigned char *data, int img_n, int req_comp, unsigned int x, unsigned int y) +{ + int i,j; + unsigned char *good; + + if (req_comp == img_n) return data; + STBI_ASSERT(req_comp >= 1 && req_comp <= 4); + + good = (unsigned char *) stbi__malloc_mad3(req_comp, x, y, 0); + if (good == NULL) { + STBI_FREE(data); + return stbi__errpuc("outofmem", "Out of memory"); + } + + for (j=0; j < (int) y; ++j) { + unsigned char *src = data + j * x * img_n ; + unsigned char *dest = good + j * x * req_comp; + + #define STBI__COMBO(a,b) ((a)*8+(b)) + #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b) + // convert source image with img_n components to one with req_comp components; + // avoid switch per pixel, so use switch per scanline and massive macros + switch (STBI__COMBO(img_n, req_comp)) { + STBI__CASE(1,2) { dest[0]=src[0]; dest[1]=255; } break; + STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break; + STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0]; dest[3]=255; } break; + STBI__CASE(2,1) { dest[0]=src[0]; } break; + STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break; + STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0]; dest[3]=src[1]; } break; + STBI__CASE(3,4) { dest[0]=src[0];dest[1]=src[1];dest[2]=src[2];dest[3]=255; } break; + STBI__CASE(3,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break; + STBI__CASE(3,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); dest[1] = 255; } break; + STBI__CASE(4,1) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); } break; + STBI__CASE(4,2) { dest[0]=stbi__compute_y(src[0],src[1],src[2]); dest[1] = src[3]; } break; + STBI__CASE(4,3) { dest[0]=src[0];dest[1]=src[1];dest[2]=src[2]; } break; + default: STBI_ASSERT(0); STBI_FREE(data); STBI_FREE(good); return stbi__errpuc("unsupported", "Unsupported format conversion"); + } + #undef STBI__CASE + } + + STBI_FREE(data); + return good; +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD) +// nothing +#else +static stbi__uint16 stbi__compute_y_16(int r, int g, int b) +{ + return (stbi__uint16) (((r*77) + (g*150) + (29*b)) >> 8); +} +#endif + +#if defined(STBI_NO_PNG) && defined(STBI_NO_PSD) +// nothing +#else +static stbi__uint16 *stbi__convert_format16(stbi__uint16 *data, int img_n, int req_comp, unsigned int x, unsigned int y) +{ + int i,j; + stbi__uint16 *good; + + if (req_comp == img_n) return data; + STBI_ASSERT(req_comp >= 1 && req_comp <= 4); + + good = (stbi__uint16 *) stbi__malloc(req_comp * x * y * 2); + if (good == NULL) { + STBI_FREE(data); + return (stbi__uint16 *) stbi__errpuc("outofmem", "Out of memory"); + } + + for (j=0; j < (int) y; ++j) { + stbi__uint16 *src = data + j * x * img_n ; + stbi__uint16 *dest = good + j * x * req_comp; + + #define STBI__COMBO(a,b) ((a)*8+(b)) + #define STBI__CASE(a,b) case STBI__COMBO(a,b): for(i=x-1; i >= 0; --i, src += a, dest += b) + // convert source image with img_n components to one with req_comp components; + // avoid switch per pixel, so use switch per scanline and massive macros + switch (STBI__COMBO(img_n, req_comp)) { + STBI__CASE(1,2) { dest[0]=src[0]; dest[1]=0xffff; } break; + STBI__CASE(1,3) { dest[0]=dest[1]=dest[2]=src[0]; } break; + STBI__CASE(1,4) { dest[0]=dest[1]=dest[2]=src[0]; dest[3]=0xffff; } break; + STBI__CASE(2,1) { dest[0]=src[0]; } break; + STBI__CASE(2,3) { dest[0]=dest[1]=dest[2]=src[0]; } break; + STBI__CASE(2,4) { dest[0]=dest[1]=dest[2]=src[0]; dest[3]=src[1]; } break; + STBI__CASE(3,4) { dest[0]=src[0];dest[1]=src[1];dest[2]=src[2];dest[3]=0xffff; } break; + STBI__CASE(3,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break; + STBI__CASE(3,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); dest[1] = 0xffff; } break; + STBI__CASE(4,1) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); } break; + STBI__CASE(4,2) { dest[0]=stbi__compute_y_16(src[0],src[1],src[2]); dest[1] = src[3]; } break; + STBI__CASE(4,3) { dest[0]=src[0];dest[1]=src[1];dest[2]=src[2]; } break; + default: STBI_ASSERT(0); STBI_FREE(data); STBI_FREE(good); return (stbi__uint16*) stbi__errpuc("unsupported", "Unsupported format conversion"); + } + #undef STBI__CASE + } + + STBI_FREE(data); + return good; +} +#endif + +#ifndef STBI_NO_LINEAR +static float *stbi__ldr_to_hdr(stbi_uc *data, int x, int y, int comp) +{ + int i,k,n; + float *output; + if (!data) return NULL; + output = (float *) stbi__malloc_mad4(x, y, comp, sizeof(float), 0); + if (output == NULL) { STBI_FREE(data); return stbi__errpf("outofmem", "Out of memory"); } + // compute number of non-alpha components + if (comp & 1) n = comp; else n = comp-1; + for (i=0; i < x*y; ++i) { + for (k=0; k < n; ++k) { + output[i*comp + k] = (float) (pow(data[i*comp+k]/255.0f, stbi__l2h_gamma) * stbi__l2h_scale); + } + } + if (n < comp) { + for (i=0; i < x*y; ++i) { + output[i*comp + n] = data[i*comp + n]/255.0f; + } + } + STBI_FREE(data); + return output; +} +#endif + +#ifndef STBI_NO_HDR +#define stbi__float2int(x) ((int) (x)) +static stbi_uc *stbi__hdr_to_ldr(float *data, int x, int y, int comp) +{ + int i,k,n; + stbi_uc *output; + if (!data) return NULL; + output = (stbi_uc *) stbi__malloc_mad3(x, y, comp, 0); + if (output == NULL) { STBI_FREE(data); return stbi__errpuc("outofmem", "Out of memory"); } + // compute number of non-alpha components + if (comp & 1) n = comp; else n = comp-1; + for (i=0; i < x*y; ++i) { + for (k=0; k < n; ++k) { + float z = (float) pow(data[i*comp+k]*stbi__h2l_scale_i, stbi__h2l_gamma_i) * 255 + 0.5f; + if (z < 0) z = 0; + if (z > 255) z = 255; + output[i*comp + k] = (stbi_uc) stbi__float2int(z); + } + if (k < comp) { + float z = data[i*comp+k] * 255 + 0.5f; + if (z < 0) z = 0; + if (z > 255) z = 255; + output[i*comp + k] = (stbi_uc) stbi__float2int(z); + } + } + STBI_FREE(data); + return output; +} +#endif + +////////////////////////////////////////////////////////////////////////////// +// +// "baseline" JPEG/JFIF decoder +// +// simple implementation +// - doesn't support delayed output of y-dimension +// - simple interface (only one output format: 8-bit interleaved RGB) +// - doesn't try to recover corrupt jpegs +// - doesn't allow partial loading, loading multiple at once +// - still fast on x86 (copying globals into locals doesn't help x86) +// - allocates lots of intermediate memory (full size of all components) +// - non-interleaved case requires this anyway +// - allows good upsampling (see next) +// high-quality +// - upsampled channels are bilinearly interpolated, even across blocks +// - quality integer IDCT derived from IJG's 'slow' +// performance +// - fast huffman; reasonable integer IDCT +// - some SIMD kernels for common paths on targets with SSE2/NEON +// - uses a lot of intermediate memory, could cache poorly + +#ifndef STBI_NO_JPEG + +// huffman decoding acceleration +#define FAST_BITS 9 // larger handles more cases; smaller stomps less cache + +typedef struct +{ + stbi_uc fast[1 << FAST_BITS]; + // weirdly, repacking this into AoS is a 10% speed loss, instead of a win + stbi__uint16 code[256]; + stbi_uc values[256]; + stbi_uc size[257]; + unsigned int maxcode[18]; + int delta[17]; // old 'firstsymbol' - old 'firstcode' +} stbi__huffman; + +typedef struct +{ + stbi__context *s; + stbi__huffman huff_dc[4]; + stbi__huffman huff_ac[4]; + stbi__uint16 dequant[4][64]; + stbi__int16 fast_ac[4][1 << FAST_BITS]; + +// sizes for components, interleaved MCUs + int img_h_max, img_v_max; + int img_mcu_x, img_mcu_y; + int img_mcu_w, img_mcu_h; + +// definition of jpeg image component + struct + { + int id; + int h,v; + int tq; + int hd,ha; + int dc_pred; + + int x,y,w2,h2; + stbi_uc *data; + void *raw_data, *raw_coeff; + stbi_uc *linebuf; + short *coeff; // progressive only + int coeff_w, coeff_h; // number of 8x8 coefficient blocks + } img_comp[4]; + + stbi__uint32 code_buffer; // jpeg entropy-coded buffer + int code_bits; // number of valid bits + unsigned char marker; // marker seen while filling entropy buffer + int nomore; // flag if we saw a marker so must stop + + int progressive; + int spec_start; + int spec_end; + int succ_high; + int succ_low; + int eob_run; + int jfif; + int app14_color_transform; // Adobe APP14 tag + int rgb; + + int scan_n, order[4]; + int restart_interval, todo; + +// kernels + void (*idct_block_kernel)(stbi_uc *out, int out_stride, short data[64]); + void (*YCbCr_to_RGB_kernel)(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step); + stbi_uc *(*resample_row_hv_2_kernel)(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs); +} stbi__jpeg; + +static int stbi__build_huffman(stbi__huffman *h, int *count) +{ + int i,j,k=0; + unsigned int code; + // build size list for each symbol (from JPEG spec) + for (i=0; i < 16; ++i) + for (j=0; j < count[i]; ++j) + h->size[k++] = (stbi_uc) (i+1); + h->size[k] = 0; + + // compute actual symbols (from jpeg spec) + code = 0; + k = 0; + for(j=1; j <= 16; ++j) { + // compute delta to add to code to compute symbol id + h->delta[j] = k - code; + if (h->size[k] == j) { + while (h->size[k] == j) + h->code[k++] = (stbi__uint16) (code++); + if (code-1 >= (1u << j)) return stbi__err("bad code lengths","Corrupt JPEG"); + } + // compute largest code + 1 for this size, preshifted as needed later + h->maxcode[j] = code << (16-j); + code <<= 1; + } + h->maxcode[j] = 0xffffffff; + + // build non-spec acceleration table; 255 is flag for not-accelerated + memset(h->fast, 255, 1 << FAST_BITS); + for (i=0; i < k; ++i) { + int s = h->size[i]; + if (s <= FAST_BITS) { + int c = h->code[i] << (FAST_BITS-s); + int m = 1 << (FAST_BITS-s); + for (j=0; j < m; ++j) { + h->fast[c+j] = (stbi_uc) i; + } + } + } + return 1; +} + +// build a table that decodes both magnitude and value of small ACs in +// one go. +static void stbi__build_fast_ac(stbi__int16 *fast_ac, stbi__huffman *h) +{ + int i; + for (i=0; i < (1 << FAST_BITS); ++i) { + stbi_uc fast = h->fast[i]; + fast_ac[i] = 0; + if (fast < 255) { + int rs = h->values[fast]; + int run = (rs >> 4) & 15; + int magbits = rs & 15; + int len = h->size[fast]; + + if (magbits && len + magbits <= FAST_BITS) { + // magnitude code followed by receive_extend code + int k = ((i << len) & ((1 << FAST_BITS) - 1)) >> (FAST_BITS - magbits); + int m = 1 << (magbits - 1); + if (k < m) k += (~0U << magbits) + 1; + // if the result is small enough, we can fit it in fast_ac table + if (k >= -128 && k <= 127) + fast_ac[i] = (stbi__int16) ((k * 256) + (run * 16) + (len + magbits)); + } + } + } +} + +static void stbi__grow_buffer_unsafe(stbi__jpeg *j) +{ + do { + unsigned int b = j->nomore ? 0 : stbi__get8(j->s); + if (b == 0xff) { + int c = stbi__get8(j->s); + while (c == 0xff) c = stbi__get8(j->s); // consume fill bytes + if (c != 0) { + j->marker = (unsigned char) c; + j->nomore = 1; + return; + } + } + j->code_buffer |= b << (24 - j->code_bits); + j->code_bits += 8; + } while (j->code_bits <= 24); +} + +// (1 << n) - 1 +static const stbi__uint32 stbi__bmask[17]={0,1,3,7,15,31,63,127,255,511,1023,2047,4095,8191,16383,32767,65535}; + +// decode a jpeg huffman value from the bitstream +stbi_inline static int stbi__jpeg_huff_decode(stbi__jpeg *j, stbi__huffman *h) +{ + unsigned int temp; + int c,k; + + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + + // look at the top FAST_BITS and determine what symbol ID it is, + // if the code is <= FAST_BITS + c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1); + k = h->fast[c]; + if (k < 255) { + int s = h->size[k]; + if (s > j->code_bits) + return -1; + j->code_buffer <<= s; + j->code_bits -= s; + return h->values[k]; + } + + // naive test is to shift the code_buffer down so k bits are + // valid, then test against maxcode. To speed this up, we've + // preshifted maxcode left so that it has (16-k) 0s at the + // end; in other words, regardless of the number of bits, it + // wants to be compared against something shifted to have 16; + // that way we don't need to shift inside the loop. + temp = j->code_buffer >> 16; + for (k=FAST_BITS+1 ; ; ++k) + if (temp < h->maxcode[k]) + break; + if (k == 17) { + // error! code not found + j->code_bits -= 16; + return -1; + } + + if (k > j->code_bits) + return -1; + + // convert the huffman code to the symbol id + c = ((j->code_buffer >> (32 - k)) & stbi__bmask[k]) + h->delta[k]; + STBI_ASSERT((((j->code_buffer) >> (32 - h->size[c])) & stbi__bmask[h->size[c]]) == h->code[c]); + + // convert the id to a symbol + j->code_bits -= k; + j->code_buffer <<= k; + return h->values[c]; +} + +// bias[n] = (-1<code_bits < n) stbi__grow_buffer_unsafe(j); + + sgn = (stbi__int32)j->code_buffer >> 31; // sign bit is always in MSB + k = stbi_lrot(j->code_buffer, n); + if (n < 0 || n >= (int) (sizeof(stbi__bmask)/sizeof(*stbi__bmask))) return 0; + j->code_buffer = k & ~stbi__bmask[n]; + k &= stbi__bmask[n]; + j->code_bits -= n; + return k + (stbi__jbias[n] & ~sgn); +} + +// get some unsigned bits +stbi_inline static int stbi__jpeg_get_bits(stbi__jpeg *j, int n) +{ + unsigned int k; + if (j->code_bits < n) stbi__grow_buffer_unsafe(j); + k = stbi_lrot(j->code_buffer, n); + j->code_buffer = k & ~stbi__bmask[n]; + k &= stbi__bmask[n]; + j->code_bits -= n; + return k; +} + +stbi_inline static int stbi__jpeg_get_bit(stbi__jpeg *j) +{ + unsigned int k; + if (j->code_bits < 1) stbi__grow_buffer_unsafe(j); + k = j->code_buffer; + j->code_buffer <<= 1; + --j->code_bits; + return k & 0x80000000; +} + +// given a value that's at position X in the zigzag stream, +// where does it appear in the 8x8 matrix coded as row-major? +static const stbi_uc stbi__jpeg_dezigzag[64+15] = +{ + 0, 1, 8, 16, 9, 2, 3, 10, + 17, 24, 32, 25, 18, 11, 4, 5, + 12, 19, 26, 33, 40, 48, 41, 34, + 27, 20, 13, 6, 7, 14, 21, 28, + 35, 42, 49, 56, 57, 50, 43, 36, + 29, 22, 15, 23, 30, 37, 44, 51, + 58, 59, 52, 45, 38, 31, 39, 46, + 53, 60, 61, 54, 47, 55, 62, 63, + // let corrupt input sample past end + 63, 63, 63, 63, 63, 63, 63, 63, + 63, 63, 63, 63, 63, 63, 63 +}; + +// decode one 64-entry block-- +static int stbi__jpeg_decode_block(stbi__jpeg *j, short data[64], stbi__huffman *hdc, stbi__huffman *hac, stbi__int16 *fac, int b, stbi__uint16 *dequant) +{ + int diff,dc,k; + int t; + + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + t = stbi__jpeg_huff_decode(j, hdc); + if (t < 0) return stbi__err("bad huffman code","Corrupt JPEG"); + + // 0 all the ac values now so we can do it 32-bits at a time + memset(data,0,64*sizeof(data[0])); + + diff = t ? stbi__extend_receive(j, t) : 0; + dc = j->img_comp[b].dc_pred + diff; + j->img_comp[b].dc_pred = dc; + data[0] = (short) (dc * dequant[0]); + + // decode AC components, see JPEG spec + k = 1; + do { + unsigned int zig; + int c,r,s; + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1); + r = fac[c]; + if (r) { // fast-AC path + k += (r >> 4) & 15; // run + s = r & 15; // combined length + j->code_buffer <<= s; + j->code_bits -= s; + // decode into unzigzag'd location + zig = stbi__jpeg_dezigzag[k++]; + data[zig] = (short) ((r >> 8) * dequant[zig]); + } else { + int rs = stbi__jpeg_huff_decode(j, hac); + if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG"); + s = rs & 15; + r = rs >> 4; + if (s == 0) { + if (rs != 0xf0) break; // end block + k += 16; + } else { + k += r; + // decode into unzigzag'd location + zig = stbi__jpeg_dezigzag[k++]; + data[zig] = (short) (stbi__extend_receive(j,s) * dequant[zig]); + } + } + } while (k < 64); + return 1; +} + +static int stbi__jpeg_decode_block_prog_dc(stbi__jpeg *j, short data[64], stbi__huffman *hdc, int b) +{ + int diff,dc; + int t; + if (j->spec_end != 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG"); + + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + + if (j->succ_high == 0) { + // first scan for DC coefficient, must be first + memset(data,0,64*sizeof(data[0])); // 0 all the ac values now + t = stbi__jpeg_huff_decode(j, hdc); + if (t == -1) return stbi__err("can't merge dc and ac", "Corrupt JPEG"); + diff = t ? stbi__extend_receive(j, t) : 0; + + dc = j->img_comp[b].dc_pred + diff; + j->img_comp[b].dc_pred = dc; + data[0] = (short) (dc << j->succ_low); + } else { + // refinement scan for DC coefficient + if (stbi__jpeg_get_bit(j)) + data[0] += (short) (1 << j->succ_low); + } + return 1; +} + +// @OPTIMIZE: store non-zigzagged during the decode passes, +// and only de-zigzag when dequantizing +static int stbi__jpeg_decode_block_prog_ac(stbi__jpeg *j, short data[64], stbi__huffman *hac, stbi__int16 *fac) +{ + int k; + if (j->spec_start == 0) return stbi__err("can't merge dc and ac", "Corrupt JPEG"); + + if (j->succ_high == 0) { + int shift = j->succ_low; + + if (j->eob_run) { + --j->eob_run; + return 1; + } + + k = j->spec_start; + do { + unsigned int zig; + int c,r,s; + if (j->code_bits < 16) stbi__grow_buffer_unsafe(j); + c = (j->code_buffer >> (32 - FAST_BITS)) & ((1 << FAST_BITS)-1); + r = fac[c]; + if (r) { // fast-AC path + k += (r >> 4) & 15; // run + s = r & 15; // combined length + j->code_buffer <<= s; + j->code_bits -= s; + zig = stbi__jpeg_dezigzag[k++]; + data[zig] = (short) ((r >> 8) << shift); + } else { + int rs = stbi__jpeg_huff_decode(j, hac); + if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG"); + s = rs & 15; + r = rs >> 4; + if (s == 0) { + if (r < 15) { + j->eob_run = (1 << r); + if (r) + j->eob_run += stbi__jpeg_get_bits(j, r); + --j->eob_run; + break; + } + k += 16; + } else { + k += r; + zig = stbi__jpeg_dezigzag[k++]; + data[zig] = (short) (stbi__extend_receive(j,s) << shift); + } + } + } while (k <= j->spec_end); + } else { + // refinement scan for these AC coefficients + + short bit = (short) (1 << j->succ_low); + + if (j->eob_run) { + --j->eob_run; + for (k = j->spec_start; k <= j->spec_end; ++k) { + short *p = &data[stbi__jpeg_dezigzag[k]]; + if (*p != 0) + if (stbi__jpeg_get_bit(j)) + if ((*p & bit)==0) { + if (*p > 0) + *p += bit; + else + *p -= bit; + } + } + } else { + k = j->spec_start; + do { + int r,s; + int rs = stbi__jpeg_huff_decode(j, hac); // @OPTIMIZE see if we can use the fast path here, advance-by-r is so slow, eh + if (rs < 0) return stbi__err("bad huffman code","Corrupt JPEG"); + s = rs & 15; + r = rs >> 4; + if (s == 0) { + if (r < 15) { + j->eob_run = (1 << r) - 1; + if (r) + j->eob_run += stbi__jpeg_get_bits(j, r); + r = 64; // force end of block + } else { + // r=15 s=0 should write 16 0s, so we just do + // a run of 15 0s and then write s (which is 0), + // so we don't have to do anything special here + } + } else { + if (s != 1) return stbi__err("bad huffman code", "Corrupt JPEG"); + // sign bit + if (stbi__jpeg_get_bit(j)) + s = bit; + else + s = -bit; + } + + // advance by r + while (k <= j->spec_end) { + short *p = &data[stbi__jpeg_dezigzag[k++]]; + if (*p != 0) { + if (stbi__jpeg_get_bit(j)) + if ((*p & bit)==0) { + if (*p > 0) + *p += bit; + else + *p -= bit; + } + } else { + if (r == 0) { + *p = (short) s; + break; + } + --r; + } + } + } while (k <= j->spec_end); + } + } + return 1; +} + +// take a -128..127 value and stbi__clamp it and convert to 0..255 +stbi_inline static stbi_uc stbi__clamp(int x) +{ + // trick to use a single test to catch both cases + if ((unsigned int) x > 255) { + if (x < 0) return 0; + if (x > 255) return 255; + } + return (stbi_uc) x; +} + +#define stbi__f2f(x) ((int) (((x) * 4096 + 0.5))) +#define stbi__fsh(x) ((x) * 4096) + +// derived from jidctint -- DCT_ISLOW +#define STBI__IDCT_1D(s0,s1,s2,s3,s4,s5,s6,s7) \ + int t0,t1,t2,t3,p1,p2,p3,p4,p5,x0,x1,x2,x3; \ + p2 = s2; \ + p3 = s6; \ + p1 = (p2+p3) * stbi__f2f(0.5411961f); \ + t2 = p1 + p3*stbi__f2f(-1.847759065f); \ + t3 = p1 + p2*stbi__f2f( 0.765366865f); \ + p2 = s0; \ + p3 = s4; \ + t0 = stbi__fsh(p2+p3); \ + t1 = stbi__fsh(p2-p3); \ + x0 = t0+t3; \ + x3 = t0-t3; \ + x1 = t1+t2; \ + x2 = t1-t2; \ + t0 = s7; \ + t1 = s5; \ + t2 = s3; \ + t3 = s1; \ + p3 = t0+t2; \ + p4 = t1+t3; \ + p1 = t0+t3; \ + p2 = t1+t2; \ + p5 = (p3+p4)*stbi__f2f( 1.175875602f); \ + t0 = t0*stbi__f2f( 0.298631336f); \ + t1 = t1*stbi__f2f( 2.053119869f); \ + t2 = t2*stbi__f2f( 3.072711026f); \ + t3 = t3*stbi__f2f( 1.501321110f); \ + p1 = p5 + p1*stbi__f2f(-0.899976223f); \ + p2 = p5 + p2*stbi__f2f(-2.562915447f); \ + p3 = p3*stbi__f2f(-1.961570560f); \ + p4 = p4*stbi__f2f(-0.390180644f); \ + t3 += p1+p4; \ + t2 += p2+p3; \ + t1 += p2+p4; \ + t0 += p1+p3; + +static void stbi__idct_block(stbi_uc *out, int out_stride, short data[64]) +{ + int i,val[64],*v=val; + stbi_uc *o; + short *d = data; + + // columns + for (i=0; i < 8; ++i,++d, ++v) { + // if all zeroes, shortcut -- this avoids dequantizing 0s and IDCTing + if (d[ 8]==0 && d[16]==0 && d[24]==0 && d[32]==0 + && d[40]==0 && d[48]==0 && d[56]==0) { + // no shortcut 0 seconds + // (1|2|3|4|5|6|7)==0 0 seconds + // all separate -0.047 seconds + // 1 && 2|3 && 4|5 && 6|7: -0.047 seconds + int dcterm = d[0]*4; + v[0] = v[8] = v[16] = v[24] = v[32] = v[40] = v[48] = v[56] = dcterm; + } else { + STBI__IDCT_1D(d[ 0],d[ 8],d[16],d[24],d[32],d[40],d[48],d[56]) + // constants scaled things up by 1<<12; let's bring them back + // down, but keep 2 extra bits of precision + x0 += 512; x1 += 512; x2 += 512; x3 += 512; + v[ 0] = (x0+t3) >> 10; + v[56] = (x0-t3) >> 10; + v[ 8] = (x1+t2) >> 10; + v[48] = (x1-t2) >> 10; + v[16] = (x2+t1) >> 10; + v[40] = (x2-t1) >> 10; + v[24] = (x3+t0) >> 10; + v[32] = (x3-t0) >> 10; + } + } + + for (i=0, v=val, o=out; i < 8; ++i,v+=8,o+=out_stride) { + // no fast case since the first 1D IDCT spread components out + STBI__IDCT_1D(v[0],v[1],v[2],v[3],v[4],v[5],v[6],v[7]) + // constants scaled things up by 1<<12, plus we had 1<<2 from first + // loop, plus horizontal and vertical each scale by sqrt(8) so together + // we've got an extra 1<<3, so 1<<17 total we need to remove. + // so we want to round that, which means adding 0.5 * 1<<17, + // aka 65536. Also, we'll end up with -128 to 127 that we want + // to encode as 0..255 by adding 128, so we'll add that before the shift + x0 += 65536 + (128<<17); + x1 += 65536 + (128<<17); + x2 += 65536 + (128<<17); + x3 += 65536 + (128<<17); + // tried computing the shifts into temps, or'ing the temps to see + // if any were out of range, but that was slower + o[0] = stbi__clamp((x0+t3) >> 17); + o[7] = stbi__clamp((x0-t3) >> 17); + o[1] = stbi__clamp((x1+t2) >> 17); + o[6] = stbi__clamp((x1-t2) >> 17); + o[2] = stbi__clamp((x2+t1) >> 17); + o[5] = stbi__clamp((x2-t1) >> 17); + o[3] = stbi__clamp((x3+t0) >> 17); + o[4] = stbi__clamp((x3-t0) >> 17); + } +} + +#ifdef STBI_SSE2 +// sse2 integer IDCT. not the fastest possible implementation but it +// produces bit-identical results to the generic C version so it's +// fully "transparent". +static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64]) +{ + // This is constructed to match our regular (generic) integer IDCT exactly. + __m128i row0, row1, row2, row3, row4, row5, row6, row7; + __m128i tmp; + + // dot product constant: even elems=x, odd elems=y + #define dct_const(x,y) _mm_setr_epi16((x),(y),(x),(y),(x),(y),(x),(y)) + + // out(0) = c0[even]*x + c0[odd]*y (c0, x, y 16-bit, out 32-bit) + // out(1) = c1[even]*x + c1[odd]*y + #define dct_rot(out0,out1, x,y,c0,c1) \ + __m128i c0##lo = _mm_unpacklo_epi16((x),(y)); \ + __m128i c0##hi = _mm_unpackhi_epi16((x),(y)); \ + __m128i out0##_l = _mm_madd_epi16(c0##lo, c0); \ + __m128i out0##_h = _mm_madd_epi16(c0##hi, c0); \ + __m128i out1##_l = _mm_madd_epi16(c0##lo, c1); \ + __m128i out1##_h = _mm_madd_epi16(c0##hi, c1) + + // out = in << 12 (in 16-bit, out 32-bit) + #define dct_widen(out, in) \ + __m128i out##_l = _mm_srai_epi32(_mm_unpacklo_epi16(_mm_setzero_si128(), (in)), 4); \ + __m128i out##_h = _mm_srai_epi32(_mm_unpackhi_epi16(_mm_setzero_si128(), (in)), 4) + + // wide add + #define dct_wadd(out, a, b) \ + __m128i out##_l = _mm_add_epi32(a##_l, b##_l); \ + __m128i out##_h = _mm_add_epi32(a##_h, b##_h) + + // wide sub + #define dct_wsub(out, a, b) \ + __m128i out##_l = _mm_sub_epi32(a##_l, b##_l); \ + __m128i out##_h = _mm_sub_epi32(a##_h, b##_h) + + // butterfly a/b, add bias, then shift by "s" and pack + #define dct_bfly32o(out0, out1, a,b,bias,s) \ + { \ + __m128i abiased_l = _mm_add_epi32(a##_l, bias); \ + __m128i abiased_h = _mm_add_epi32(a##_h, bias); \ + dct_wadd(sum, abiased, b); \ + dct_wsub(dif, abiased, b); \ + out0 = _mm_packs_epi32(_mm_srai_epi32(sum_l, s), _mm_srai_epi32(sum_h, s)); \ + out1 = _mm_packs_epi32(_mm_srai_epi32(dif_l, s), _mm_srai_epi32(dif_h, s)); \ + } + + // 8-bit interleave step (for transposes) + #define dct_interleave8(a, b) \ + tmp = a; \ + a = _mm_unpacklo_epi8(a, b); \ + b = _mm_unpackhi_epi8(tmp, b) + + // 16-bit interleave step (for transposes) + #define dct_interleave16(a, b) \ + tmp = a; \ + a = _mm_unpacklo_epi16(a, b); \ + b = _mm_unpackhi_epi16(tmp, b) + + #define dct_pass(bias,shift) \ + { \ + /* even part */ \ + dct_rot(t2e,t3e, row2,row6, rot0_0,rot0_1); \ + __m128i sum04 = _mm_add_epi16(row0, row4); \ + __m128i dif04 = _mm_sub_epi16(row0, row4); \ + dct_widen(t0e, sum04); \ + dct_widen(t1e, dif04); \ + dct_wadd(x0, t0e, t3e); \ + dct_wsub(x3, t0e, t3e); \ + dct_wadd(x1, t1e, t2e); \ + dct_wsub(x2, t1e, t2e); \ + /* odd part */ \ + dct_rot(y0o,y2o, row7,row3, rot2_0,rot2_1); \ + dct_rot(y1o,y3o, row5,row1, rot3_0,rot3_1); \ + __m128i sum17 = _mm_add_epi16(row1, row7); \ + __m128i sum35 = _mm_add_epi16(row3, row5); \ + dct_rot(y4o,y5o, sum17,sum35, rot1_0,rot1_1); \ + dct_wadd(x4, y0o, y4o); \ + dct_wadd(x5, y1o, y5o); \ + dct_wadd(x6, y2o, y5o); \ + dct_wadd(x7, y3o, y4o); \ + dct_bfly32o(row0,row7, x0,x7,bias,shift); \ + dct_bfly32o(row1,row6, x1,x6,bias,shift); \ + dct_bfly32o(row2,row5, x2,x5,bias,shift); \ + dct_bfly32o(row3,row4, x3,x4,bias,shift); \ + } + + __m128i rot0_0 = dct_const(stbi__f2f(0.5411961f), stbi__f2f(0.5411961f) + stbi__f2f(-1.847759065f)); + __m128i rot0_1 = dct_const(stbi__f2f(0.5411961f) + stbi__f2f( 0.765366865f), stbi__f2f(0.5411961f)); + __m128i rot1_0 = dct_const(stbi__f2f(1.175875602f) + stbi__f2f(-0.899976223f), stbi__f2f(1.175875602f)); + __m128i rot1_1 = dct_const(stbi__f2f(1.175875602f), stbi__f2f(1.175875602f) + stbi__f2f(-2.562915447f)); + __m128i rot2_0 = dct_const(stbi__f2f(-1.961570560f) + stbi__f2f( 0.298631336f), stbi__f2f(-1.961570560f)); + __m128i rot2_1 = dct_const(stbi__f2f(-1.961570560f), stbi__f2f(-1.961570560f) + stbi__f2f( 3.072711026f)); + __m128i rot3_0 = dct_const(stbi__f2f(-0.390180644f) + stbi__f2f( 2.053119869f), stbi__f2f(-0.390180644f)); + __m128i rot3_1 = dct_const(stbi__f2f(-0.390180644f), stbi__f2f(-0.390180644f) + stbi__f2f( 1.501321110f)); + + // rounding biases in column/row passes, see stbi__idct_block for explanation. + __m128i bias_0 = _mm_set1_epi32(512); + __m128i bias_1 = _mm_set1_epi32(65536 + (128<<17)); + + // load + row0 = _mm_load_si128((const __m128i *) (data + 0*8)); + row1 = _mm_load_si128((const __m128i *) (data + 1*8)); + row2 = _mm_load_si128((const __m128i *) (data + 2*8)); + row3 = _mm_load_si128((const __m128i *) (data + 3*8)); + row4 = _mm_load_si128((const __m128i *) (data + 4*8)); + row5 = _mm_load_si128((const __m128i *) (data + 5*8)); + row6 = _mm_load_si128((const __m128i *) (data + 6*8)); + row7 = _mm_load_si128((const __m128i *) (data + 7*8)); + + // column pass + dct_pass(bias_0, 10); + + { + // 16bit 8x8 transpose pass 1 + dct_interleave16(row0, row4); + dct_interleave16(row1, row5); + dct_interleave16(row2, row6); + dct_interleave16(row3, row7); + + // transpose pass 2 + dct_interleave16(row0, row2); + dct_interleave16(row1, row3); + dct_interleave16(row4, row6); + dct_interleave16(row5, row7); + + // transpose pass 3 + dct_interleave16(row0, row1); + dct_interleave16(row2, row3); + dct_interleave16(row4, row5); + dct_interleave16(row6, row7); + } + + // row pass + dct_pass(bias_1, 17); + + { + // pack + __m128i p0 = _mm_packus_epi16(row0, row1); // a0a1a2a3...a7b0b1b2b3...b7 + __m128i p1 = _mm_packus_epi16(row2, row3); + __m128i p2 = _mm_packus_epi16(row4, row5); + __m128i p3 = _mm_packus_epi16(row6, row7); + + // 8bit 8x8 transpose pass 1 + dct_interleave8(p0, p2); // a0e0a1e1... + dct_interleave8(p1, p3); // c0g0c1g1... + + // transpose pass 2 + dct_interleave8(p0, p1); // a0c0e0g0... + dct_interleave8(p2, p3); // b0d0f0h0... + + // transpose pass 3 + dct_interleave8(p0, p2); // a0b0c0d0... + dct_interleave8(p1, p3); // a4b4c4d4... + + // store + _mm_storel_epi64((__m128i *) out, p0); out += out_stride; + _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p0, 0x4e)); out += out_stride; + _mm_storel_epi64((__m128i *) out, p2); out += out_stride; + _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p2, 0x4e)); out += out_stride; + _mm_storel_epi64((__m128i *) out, p1); out += out_stride; + _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p1, 0x4e)); out += out_stride; + _mm_storel_epi64((__m128i *) out, p3); out += out_stride; + _mm_storel_epi64((__m128i *) out, _mm_shuffle_epi32(p3, 0x4e)); + } + +#undef dct_const +#undef dct_rot +#undef dct_widen +#undef dct_wadd +#undef dct_wsub +#undef dct_bfly32o +#undef dct_interleave8 +#undef dct_interleave16 +#undef dct_pass +} + +#endif // STBI_SSE2 + +#ifdef STBI_NEON + +// NEON integer IDCT. should produce bit-identical +// results to the generic C version. +static void stbi__idct_simd(stbi_uc *out, int out_stride, short data[64]) +{ + int16x8_t row0, row1, row2, row3, row4, row5, row6, row7; + + int16x4_t rot0_0 = vdup_n_s16(stbi__f2f(0.5411961f)); + int16x4_t rot0_1 = vdup_n_s16(stbi__f2f(-1.847759065f)); + int16x4_t rot0_2 = vdup_n_s16(stbi__f2f( 0.765366865f)); + int16x4_t rot1_0 = vdup_n_s16(stbi__f2f( 1.175875602f)); + int16x4_t rot1_1 = vdup_n_s16(stbi__f2f(-0.899976223f)); + int16x4_t rot1_2 = vdup_n_s16(stbi__f2f(-2.562915447f)); + int16x4_t rot2_0 = vdup_n_s16(stbi__f2f(-1.961570560f)); + int16x4_t rot2_1 = vdup_n_s16(stbi__f2f(-0.390180644f)); + int16x4_t rot3_0 = vdup_n_s16(stbi__f2f( 0.298631336f)); + int16x4_t rot3_1 = vdup_n_s16(stbi__f2f( 2.053119869f)); + int16x4_t rot3_2 = vdup_n_s16(stbi__f2f( 3.072711026f)); + int16x4_t rot3_3 = vdup_n_s16(stbi__f2f( 1.501321110f)); + +#define dct_long_mul(out, inq, coeff) \ + int32x4_t out##_l = vmull_s16(vget_low_s16(inq), coeff); \ + int32x4_t out##_h = vmull_s16(vget_high_s16(inq), coeff) + +#define dct_long_mac(out, acc, inq, coeff) \ + int32x4_t out##_l = vmlal_s16(acc##_l, vget_low_s16(inq), coeff); \ + int32x4_t out##_h = vmlal_s16(acc##_h, vget_high_s16(inq), coeff) + +#define dct_widen(out, inq) \ + int32x4_t out##_l = vshll_n_s16(vget_low_s16(inq), 12); \ + int32x4_t out##_h = vshll_n_s16(vget_high_s16(inq), 12) + +// wide add +#define dct_wadd(out, a, b) \ + int32x4_t out##_l = vaddq_s32(a##_l, b##_l); \ + int32x4_t out##_h = vaddq_s32(a##_h, b##_h) + +// wide sub +#define dct_wsub(out, a, b) \ + int32x4_t out##_l = vsubq_s32(a##_l, b##_l); \ + int32x4_t out##_h = vsubq_s32(a##_h, b##_h) + +// butterfly a/b, then shift using "shiftop" by "s" and pack +#define dct_bfly32o(out0,out1, a,b,shiftop,s) \ + { \ + dct_wadd(sum, a, b); \ + dct_wsub(dif, a, b); \ + out0 = vcombine_s16(shiftop(sum_l, s), shiftop(sum_h, s)); \ + out1 = vcombine_s16(shiftop(dif_l, s), shiftop(dif_h, s)); \ + } + +#define dct_pass(shiftop, shift) \ + { \ + /* even part */ \ + int16x8_t sum26 = vaddq_s16(row2, row6); \ + dct_long_mul(p1e, sum26, rot0_0); \ + dct_long_mac(t2e, p1e, row6, rot0_1); \ + dct_long_mac(t3e, p1e, row2, rot0_2); \ + int16x8_t sum04 = vaddq_s16(row0, row4); \ + int16x8_t dif04 = vsubq_s16(row0, row4); \ + dct_widen(t0e, sum04); \ + dct_widen(t1e, dif04); \ + dct_wadd(x0, t0e, t3e); \ + dct_wsub(x3, t0e, t3e); \ + dct_wadd(x1, t1e, t2e); \ + dct_wsub(x2, t1e, t2e); \ + /* odd part */ \ + int16x8_t sum15 = vaddq_s16(row1, row5); \ + int16x8_t sum17 = vaddq_s16(row1, row7); \ + int16x8_t sum35 = vaddq_s16(row3, row5); \ + int16x8_t sum37 = vaddq_s16(row3, row7); \ + int16x8_t sumodd = vaddq_s16(sum17, sum35); \ + dct_long_mul(p5o, sumodd, rot1_0); \ + dct_long_mac(p1o, p5o, sum17, rot1_1); \ + dct_long_mac(p2o, p5o, sum35, rot1_2); \ + dct_long_mul(p3o, sum37, rot2_0); \ + dct_long_mul(p4o, sum15, rot2_1); \ + dct_wadd(sump13o, p1o, p3o); \ + dct_wadd(sump24o, p2o, p4o); \ + dct_wadd(sump23o, p2o, p3o); \ + dct_wadd(sump14o, p1o, p4o); \ + dct_long_mac(x4, sump13o, row7, rot3_0); \ + dct_long_mac(x5, sump24o, row5, rot3_1); \ + dct_long_mac(x6, sump23o, row3, rot3_2); \ + dct_long_mac(x7, sump14o, row1, rot3_3); \ + dct_bfly32o(row0,row7, x0,x7,shiftop,shift); \ + dct_bfly32o(row1,row6, x1,x6,shiftop,shift); \ + dct_bfly32o(row2,row5, x2,x5,shiftop,shift); \ + dct_bfly32o(row3,row4, x3,x4,shiftop,shift); \ + } + + // load + row0 = vld1q_s16(data + 0*8); + row1 = vld1q_s16(data + 1*8); + row2 = vld1q_s16(data + 2*8); + row3 = vld1q_s16(data + 3*8); + row4 = vld1q_s16(data + 4*8); + row5 = vld1q_s16(data + 5*8); + row6 = vld1q_s16(data + 6*8); + row7 = vld1q_s16(data + 7*8); + + // add DC bias + row0 = vaddq_s16(row0, vsetq_lane_s16(1024, vdupq_n_s16(0), 0)); + + // column pass + dct_pass(vrshrn_n_s32, 10); + + // 16bit 8x8 transpose + { +// these three map to a single VTRN.16, VTRN.32, and VSWP, respectively. +// whether compilers actually get this is another story, sadly. +#define dct_trn16(x, y) { int16x8x2_t t = vtrnq_s16(x, y); x = t.val[0]; y = t.val[1]; } +#define dct_trn32(x, y) { int32x4x2_t t = vtrnq_s32(vreinterpretq_s32_s16(x), vreinterpretq_s32_s16(y)); x = vreinterpretq_s16_s32(t.val[0]); y = vreinterpretq_s16_s32(t.val[1]); } +#define dct_trn64(x, y) { int16x8_t x0 = x; int16x8_t y0 = y; x = vcombine_s16(vget_low_s16(x0), vget_low_s16(y0)); y = vcombine_s16(vget_high_s16(x0), vget_high_s16(y0)); } + + // pass 1 + dct_trn16(row0, row1); // a0b0a2b2a4b4a6b6 + dct_trn16(row2, row3); + dct_trn16(row4, row5); + dct_trn16(row6, row7); + + // pass 2 + dct_trn32(row0, row2); // a0b0c0d0a4b4c4d4 + dct_trn32(row1, row3); + dct_trn32(row4, row6); + dct_trn32(row5, row7); + + // pass 3 + dct_trn64(row0, row4); // a0b0c0d0e0f0g0h0 + dct_trn64(row1, row5); + dct_trn64(row2, row6); + dct_trn64(row3, row7); + +#undef dct_trn16 +#undef dct_trn32 +#undef dct_trn64 + } + + // row pass + // vrshrn_n_s32 only supports shifts up to 16, we need + // 17. so do a non-rounding shift of 16 first then follow + // up with a rounding shift by 1. + dct_pass(vshrn_n_s32, 16); + + { + // pack and round + uint8x8_t p0 = vqrshrun_n_s16(row0, 1); + uint8x8_t p1 = vqrshrun_n_s16(row1, 1); + uint8x8_t p2 = vqrshrun_n_s16(row2, 1); + uint8x8_t p3 = vqrshrun_n_s16(row3, 1); + uint8x8_t p4 = vqrshrun_n_s16(row4, 1); + uint8x8_t p5 = vqrshrun_n_s16(row5, 1); + uint8x8_t p6 = vqrshrun_n_s16(row6, 1); + uint8x8_t p7 = vqrshrun_n_s16(row7, 1); + + // again, these can translate into one instruction, but often don't. +#define dct_trn8_8(x, y) { uint8x8x2_t t = vtrn_u8(x, y); x = t.val[0]; y = t.val[1]; } +#define dct_trn8_16(x, y) { uint16x4x2_t t = vtrn_u16(vreinterpret_u16_u8(x), vreinterpret_u16_u8(y)); x = vreinterpret_u8_u16(t.val[0]); y = vreinterpret_u8_u16(t.val[1]); } +#define dct_trn8_32(x, y) { uint32x2x2_t t = vtrn_u32(vreinterpret_u32_u8(x), vreinterpret_u32_u8(y)); x = vreinterpret_u8_u32(t.val[0]); y = vreinterpret_u8_u32(t.val[1]); } + + // sadly can't use interleaved stores here since we only write + // 8 bytes to each scan line! + + // 8x8 8-bit transpose pass 1 + dct_trn8_8(p0, p1); + dct_trn8_8(p2, p3); + dct_trn8_8(p4, p5); + dct_trn8_8(p6, p7); + + // pass 2 + dct_trn8_16(p0, p2); + dct_trn8_16(p1, p3); + dct_trn8_16(p4, p6); + dct_trn8_16(p5, p7); + + // pass 3 + dct_trn8_32(p0, p4); + dct_trn8_32(p1, p5); + dct_trn8_32(p2, p6); + dct_trn8_32(p3, p7); + + // store + vst1_u8(out, p0); out += out_stride; + vst1_u8(out, p1); out += out_stride; + vst1_u8(out, p2); out += out_stride; + vst1_u8(out, p3); out += out_stride; + vst1_u8(out, p4); out += out_stride; + vst1_u8(out, p5); out += out_stride; + vst1_u8(out, p6); out += out_stride; + vst1_u8(out, p7); + +#undef dct_trn8_8 +#undef dct_trn8_16 +#undef dct_trn8_32 + } + +#undef dct_long_mul +#undef dct_long_mac +#undef dct_widen +#undef dct_wadd +#undef dct_wsub +#undef dct_bfly32o +#undef dct_pass +} + +#endif // STBI_NEON + +#define STBI__MARKER_none 0xff +// if there's a pending marker from the entropy stream, return that +// otherwise, fetch from the stream and get a marker. if there's no +// marker, return 0xff, which is never a valid marker value +static stbi_uc stbi__get_marker(stbi__jpeg *j) +{ + stbi_uc x; + if (j->marker != STBI__MARKER_none) { x = j->marker; j->marker = STBI__MARKER_none; return x; } + x = stbi__get8(j->s); + if (x != 0xff) return STBI__MARKER_none; + while (x == 0xff) + x = stbi__get8(j->s); // consume repeated 0xff fill bytes + return x; +} + +// in each scan, we'll have scan_n components, and the order +// of the components is specified by order[] +#define STBI__RESTART(x) ((x) >= 0xd0 && (x) <= 0xd7) + +// after a restart interval, stbi__jpeg_reset the entropy decoder and +// the dc prediction +static void stbi__jpeg_reset(stbi__jpeg *j) +{ + j->code_bits = 0; + j->code_buffer = 0; + j->nomore = 0; + j->img_comp[0].dc_pred = j->img_comp[1].dc_pred = j->img_comp[2].dc_pred = j->img_comp[3].dc_pred = 0; + j->marker = STBI__MARKER_none; + j->todo = j->restart_interval ? j->restart_interval : 0x7fffffff; + j->eob_run = 0; + // no more than 1<<31 MCUs if no restart_interal? that's plenty safe, + // since we don't even allow 1<<30 pixels +} + +static int stbi__parse_entropy_coded_data(stbi__jpeg *z) +{ + stbi__jpeg_reset(z); + if (!z->progressive) { + if (z->scan_n == 1) { + int i,j; + STBI_SIMD_ALIGN(short, data[64]); + int n = z->order[0]; + // non-interleaved data, we just need to process one block at a time, + // in trivial scanline order + // number of blocks to do just depends on how many actual "pixels" this + // component has, independent of interleaved MCU blocking and such + int w = (z->img_comp[n].x+7) >> 3; + int h = (z->img_comp[n].y+7) >> 3; + for (j=0; j < h; ++j) { + for (i=0; i < w; ++i) { + int ha = z->img_comp[n].ha; + if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0; + z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data); + // every data block is an MCU, so countdown the restart interval + if (--z->todo <= 0) { + if (z->code_bits < 24) stbi__grow_buffer_unsafe(z); + // if it's NOT a restart, then just bail, so we get corrupt data + // rather than no data + if (!STBI__RESTART(z->marker)) return 1; + stbi__jpeg_reset(z); + } + } + } + return 1; + } else { // interleaved + int i,j,k,x,y; + STBI_SIMD_ALIGN(short, data[64]); + for (j=0; j < z->img_mcu_y; ++j) { + for (i=0; i < z->img_mcu_x; ++i) { + // scan an interleaved mcu... process scan_n components in order + for (k=0; k < z->scan_n; ++k) { + int n = z->order[k]; + // scan out an mcu's worth of this component; that's just determined + // by the basic H and V specified for the component + for (y=0; y < z->img_comp[n].v; ++y) { + for (x=0; x < z->img_comp[n].h; ++x) { + int x2 = (i*z->img_comp[n].h + x)*8; + int y2 = (j*z->img_comp[n].v + y)*8; + int ha = z->img_comp[n].ha; + if (!stbi__jpeg_decode_block(z, data, z->huff_dc+z->img_comp[n].hd, z->huff_ac+ha, z->fast_ac[ha], n, z->dequant[z->img_comp[n].tq])) return 0; + z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*y2+x2, z->img_comp[n].w2, data); + } + } + } + // after all interleaved components, that's an interleaved MCU, + // so now count down the restart interval + if (--z->todo <= 0) { + if (z->code_bits < 24) stbi__grow_buffer_unsafe(z); + if (!STBI__RESTART(z->marker)) return 1; + stbi__jpeg_reset(z); + } + } + } + return 1; + } + } else { + if (z->scan_n == 1) { + int i,j; + int n = z->order[0]; + // non-interleaved data, we just need to process one block at a time, + // in trivial scanline order + // number of blocks to do just depends on how many actual "pixels" this + // component has, independent of interleaved MCU blocking and such + int w = (z->img_comp[n].x+7) >> 3; + int h = (z->img_comp[n].y+7) >> 3; + for (j=0; j < h; ++j) { + for (i=0; i < w; ++i) { + short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w); + if (z->spec_start == 0) { + if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n)) + return 0; + } else { + int ha = z->img_comp[n].ha; + if (!stbi__jpeg_decode_block_prog_ac(z, data, &z->huff_ac[ha], z->fast_ac[ha])) + return 0; + } + // every data block is an MCU, so countdown the restart interval + if (--z->todo <= 0) { + if (z->code_bits < 24) stbi__grow_buffer_unsafe(z); + if (!STBI__RESTART(z->marker)) return 1; + stbi__jpeg_reset(z); + } + } + } + return 1; + } else { // interleaved + int i,j,k,x,y; + for (j=0; j < z->img_mcu_y; ++j) { + for (i=0; i < z->img_mcu_x; ++i) { + // scan an interleaved mcu... process scan_n components in order + for (k=0; k < z->scan_n; ++k) { + int n = z->order[k]; + // scan out an mcu's worth of this component; that's just determined + // by the basic H and V specified for the component + for (y=0; y < z->img_comp[n].v; ++y) { + for (x=0; x < z->img_comp[n].h; ++x) { + int x2 = (i*z->img_comp[n].h + x); + int y2 = (j*z->img_comp[n].v + y); + short *data = z->img_comp[n].coeff + 64 * (x2 + y2 * z->img_comp[n].coeff_w); + if (!stbi__jpeg_decode_block_prog_dc(z, data, &z->huff_dc[z->img_comp[n].hd], n)) + return 0; + } + } + } + // after all interleaved components, that's an interleaved MCU, + // so now count down the restart interval + if (--z->todo <= 0) { + if (z->code_bits < 24) stbi__grow_buffer_unsafe(z); + if (!STBI__RESTART(z->marker)) return 1; + stbi__jpeg_reset(z); + } + } + } + return 1; + } + } +} + +static void stbi__jpeg_dequantize(short *data, stbi__uint16 *dequant) +{ + int i; + for (i=0; i < 64; ++i) + data[i] *= dequant[i]; +} + +static void stbi__jpeg_finish(stbi__jpeg *z) +{ + if (z->progressive) { + // dequantize and idct the data + int i,j,n; + for (n=0; n < z->s->img_n; ++n) { + int w = (z->img_comp[n].x+7) >> 3; + int h = (z->img_comp[n].y+7) >> 3; + for (j=0; j < h; ++j) { + for (i=0; i < w; ++i) { + short *data = z->img_comp[n].coeff + 64 * (i + j * z->img_comp[n].coeff_w); + stbi__jpeg_dequantize(data, z->dequant[z->img_comp[n].tq]); + z->idct_block_kernel(z->img_comp[n].data+z->img_comp[n].w2*j*8+i*8, z->img_comp[n].w2, data); + } + } + } + } +} + +static int stbi__process_marker(stbi__jpeg *z, int m) +{ + int L; + switch (m) { + case STBI__MARKER_none: // no marker found + return stbi__err("expected marker","Corrupt JPEG"); + + case 0xDD: // DRI - specify restart interval + if (stbi__get16be(z->s) != 4) return stbi__err("bad DRI len","Corrupt JPEG"); + z->restart_interval = stbi__get16be(z->s); + return 1; + + case 0xDB: // DQT - define quantization table + L = stbi__get16be(z->s)-2; + while (L > 0) { + int q = stbi__get8(z->s); + int p = q >> 4, sixteen = (p != 0); + int t = q & 15,i; + if (p != 0 && p != 1) return stbi__err("bad DQT type","Corrupt JPEG"); + if (t > 3) return stbi__err("bad DQT table","Corrupt JPEG"); + + for (i=0; i < 64; ++i) + z->dequant[t][stbi__jpeg_dezigzag[i]] = (stbi__uint16)(sixteen ? stbi__get16be(z->s) : stbi__get8(z->s)); + L -= (sixteen ? 129 : 65); + } + return L==0; + + case 0xC4: // DHT - define huffman table + L = stbi__get16be(z->s)-2; + while (L > 0) { + stbi_uc *v; + int sizes[16],i,n=0; + int q = stbi__get8(z->s); + int tc = q >> 4; + int th = q & 15; + if (tc > 1 || th > 3) return stbi__err("bad DHT header","Corrupt JPEG"); + for (i=0; i < 16; ++i) { + sizes[i] = stbi__get8(z->s); + n += sizes[i]; + } + L -= 17; + if (tc == 0) { + if (!stbi__build_huffman(z->huff_dc+th, sizes)) return 0; + v = z->huff_dc[th].values; + } else { + if (!stbi__build_huffman(z->huff_ac+th, sizes)) return 0; + v = z->huff_ac[th].values; + } + for (i=0; i < n; ++i) + v[i] = stbi__get8(z->s); + if (tc != 0) + stbi__build_fast_ac(z->fast_ac[th], z->huff_ac + th); + L -= n; + } + return L==0; + } + + // check for comment block or APP blocks + if ((m >= 0xE0 && m <= 0xEF) || m == 0xFE) { + L = stbi__get16be(z->s); + if (L < 2) { + if (m == 0xFE) + return stbi__err("bad COM len","Corrupt JPEG"); + else + return stbi__err("bad APP len","Corrupt JPEG"); + } + L -= 2; + + if (m == 0xE0 && L >= 5) { // JFIF APP0 segment + static const unsigned char tag[5] = {'J','F','I','F','\0'}; + int ok = 1; + int i; + for (i=0; i < 5; ++i) + if (stbi__get8(z->s) != tag[i]) + ok = 0; + L -= 5; + if (ok) + z->jfif = 1; + } else if (m == 0xEE && L >= 12) { // Adobe APP14 segment + static const unsigned char tag[6] = {'A','d','o','b','e','\0'}; + int ok = 1; + int i; + for (i=0; i < 6; ++i) + if (stbi__get8(z->s) != tag[i]) + ok = 0; + L -= 6; + if (ok) { + stbi__get8(z->s); // version + stbi__get16be(z->s); // flags0 + stbi__get16be(z->s); // flags1 + z->app14_color_transform = stbi__get8(z->s); // color transform + L -= 6; + } + } + + stbi__skip(z->s, L); + return 1; + } + + return stbi__err("unknown marker","Corrupt JPEG"); +} + +// after we see SOS +static int stbi__process_scan_header(stbi__jpeg *z) +{ + int i; + int Ls = stbi__get16be(z->s); + z->scan_n = stbi__get8(z->s); + if (z->scan_n < 1 || z->scan_n > 4 || z->scan_n > (int) z->s->img_n) return stbi__err("bad SOS component count","Corrupt JPEG"); + if (Ls != 6+2*z->scan_n) return stbi__err("bad SOS len","Corrupt JPEG"); + for (i=0; i < z->scan_n; ++i) { + int id = stbi__get8(z->s), which; + int q = stbi__get8(z->s); + for (which = 0; which < z->s->img_n; ++which) + if (z->img_comp[which].id == id) + break; + if (which == z->s->img_n) return 0; // no match + z->img_comp[which].hd = q >> 4; if (z->img_comp[which].hd > 3) return stbi__err("bad DC huff","Corrupt JPEG"); + z->img_comp[which].ha = q & 15; if (z->img_comp[which].ha > 3) return stbi__err("bad AC huff","Corrupt JPEG"); + z->order[i] = which; + } + + { + int aa; + z->spec_start = stbi__get8(z->s); + z->spec_end = stbi__get8(z->s); // should be 63, but might be 0 + aa = stbi__get8(z->s); + z->succ_high = (aa >> 4); + z->succ_low = (aa & 15); + if (z->progressive) { + if (z->spec_start > 63 || z->spec_end > 63 || z->spec_start > z->spec_end || z->succ_high > 13 || z->succ_low > 13) + return stbi__err("bad SOS", "Corrupt JPEG"); + } else { + if (z->spec_start != 0) return stbi__err("bad SOS","Corrupt JPEG"); + if (z->succ_high != 0 || z->succ_low != 0) return stbi__err("bad SOS","Corrupt JPEG"); + z->spec_end = 63; + } + } + + return 1; +} + +static int stbi__free_jpeg_components(stbi__jpeg *z, int ncomp, int why) +{ + int i; + for (i=0; i < ncomp; ++i) { + if (z->img_comp[i].raw_data) { + STBI_FREE(z->img_comp[i].raw_data); + z->img_comp[i].raw_data = NULL; + z->img_comp[i].data = NULL; + } + if (z->img_comp[i].raw_coeff) { + STBI_FREE(z->img_comp[i].raw_coeff); + z->img_comp[i].raw_coeff = 0; + z->img_comp[i].coeff = 0; + } + if (z->img_comp[i].linebuf) { + STBI_FREE(z->img_comp[i].linebuf); + z->img_comp[i].linebuf = NULL; + } + } + return why; +} + +static int stbi__process_frame_header(stbi__jpeg *z, int scan) +{ + stbi__context *s = z->s; + int Lf,p,i,q, h_max=1,v_max=1,c; + Lf = stbi__get16be(s); if (Lf < 11) return stbi__err("bad SOF len","Corrupt JPEG"); // JPEG + p = stbi__get8(s); if (p != 8) return stbi__err("only 8-bit","JPEG format not supported: 8-bit only"); // JPEG baseline + s->img_y = stbi__get16be(s); if (s->img_y == 0) return stbi__err("no header height", "JPEG format not supported: delayed height"); // Legal, but we don't handle it--but neither does IJG + s->img_x = stbi__get16be(s); if (s->img_x == 0) return stbi__err("0 width","Corrupt JPEG"); // JPEG requires + if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + c = stbi__get8(s); + if (c != 3 && c != 1 && c != 4) return stbi__err("bad component count","Corrupt JPEG"); + s->img_n = c; + for (i=0; i < c; ++i) { + z->img_comp[i].data = NULL; + z->img_comp[i].linebuf = NULL; + } + + if (Lf != 8+3*s->img_n) return stbi__err("bad SOF len","Corrupt JPEG"); + + z->rgb = 0; + for (i=0; i < s->img_n; ++i) { + static const unsigned char rgb[3] = { 'R', 'G', 'B' }; + z->img_comp[i].id = stbi__get8(s); + if (s->img_n == 3 && z->img_comp[i].id == rgb[i]) + ++z->rgb; + q = stbi__get8(s); + z->img_comp[i].h = (q >> 4); if (!z->img_comp[i].h || z->img_comp[i].h > 4) return stbi__err("bad H","Corrupt JPEG"); + z->img_comp[i].v = q & 15; if (!z->img_comp[i].v || z->img_comp[i].v > 4) return stbi__err("bad V","Corrupt JPEG"); + z->img_comp[i].tq = stbi__get8(s); if (z->img_comp[i].tq > 3) return stbi__err("bad TQ","Corrupt JPEG"); + } + + if (scan != STBI__SCAN_load) return 1; + + if (!stbi__mad3sizes_valid(s->img_x, s->img_y, s->img_n, 0)) return stbi__err("too large", "Image too large to decode"); + + for (i=0; i < s->img_n; ++i) { + if (z->img_comp[i].h > h_max) h_max = z->img_comp[i].h; + if (z->img_comp[i].v > v_max) v_max = z->img_comp[i].v; + } + + // compute interleaved mcu info + z->img_h_max = h_max; + z->img_v_max = v_max; + z->img_mcu_w = h_max * 8; + z->img_mcu_h = v_max * 8; + // these sizes can't be more than 17 bits + z->img_mcu_x = (s->img_x + z->img_mcu_w-1) / z->img_mcu_w; + z->img_mcu_y = (s->img_y + z->img_mcu_h-1) / z->img_mcu_h; + + for (i=0; i < s->img_n; ++i) { + // number of effective pixels (e.g. for non-interleaved MCU) + z->img_comp[i].x = (s->img_x * z->img_comp[i].h + h_max-1) / h_max; + z->img_comp[i].y = (s->img_y * z->img_comp[i].v + v_max-1) / v_max; + // to simplify generation, we'll allocate enough memory to decode + // the bogus oversized data from using interleaved MCUs and their + // big blocks (e.g. a 16x16 iMCU on an image of width 33); we won't + // discard the extra data until colorspace conversion + // + // img_mcu_x, img_mcu_y: <=17 bits; comp[i].h and .v are <=4 (checked earlier) + // so these muls can't overflow with 32-bit ints (which we require) + z->img_comp[i].w2 = z->img_mcu_x * z->img_comp[i].h * 8; + z->img_comp[i].h2 = z->img_mcu_y * z->img_comp[i].v * 8; + z->img_comp[i].coeff = 0; + z->img_comp[i].raw_coeff = 0; + z->img_comp[i].linebuf = NULL; + z->img_comp[i].raw_data = stbi__malloc_mad2(z->img_comp[i].w2, z->img_comp[i].h2, 15); + if (z->img_comp[i].raw_data == NULL) + return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory")); + // align blocks for idct using mmx/sse + z->img_comp[i].data = (stbi_uc*) (((size_t) z->img_comp[i].raw_data + 15) & ~15); + if (z->progressive) { + // w2, h2 are multiples of 8 (see above) + z->img_comp[i].coeff_w = z->img_comp[i].w2 / 8; + z->img_comp[i].coeff_h = z->img_comp[i].h2 / 8; + z->img_comp[i].raw_coeff = stbi__malloc_mad3(z->img_comp[i].w2, z->img_comp[i].h2, sizeof(short), 15); + if (z->img_comp[i].raw_coeff == NULL) + return stbi__free_jpeg_components(z, i+1, stbi__err("outofmem", "Out of memory")); + z->img_comp[i].coeff = (short*) (((size_t) z->img_comp[i].raw_coeff + 15) & ~15); + } + } + + return 1; +} + +// use comparisons since in some cases we handle more than one case (e.g. SOF) +#define stbi__DNL(x) ((x) == 0xdc) +#define stbi__SOI(x) ((x) == 0xd8) +#define stbi__EOI(x) ((x) == 0xd9) +#define stbi__SOF(x) ((x) == 0xc0 || (x) == 0xc1 || (x) == 0xc2) +#define stbi__SOS(x) ((x) == 0xda) + +#define stbi__SOF_progressive(x) ((x) == 0xc2) + +static int stbi__decode_jpeg_header(stbi__jpeg *z, int scan) +{ + int m; + z->jfif = 0; + z->app14_color_transform = -1; // valid values are 0,1,2 + z->marker = STBI__MARKER_none; // initialize cached marker to empty + m = stbi__get_marker(z); + if (!stbi__SOI(m)) return stbi__err("no SOI","Corrupt JPEG"); + if (scan == STBI__SCAN_type) return 1; + m = stbi__get_marker(z); + while (!stbi__SOF(m)) { + if (!stbi__process_marker(z,m)) return 0; + m = stbi__get_marker(z); + while (m == STBI__MARKER_none) { + // some files have extra padding after their blocks, so ok, we'll scan + if (stbi__at_eof(z->s)) return stbi__err("no SOF", "Corrupt JPEG"); + m = stbi__get_marker(z); + } + } + z->progressive = stbi__SOF_progressive(m); + if (!stbi__process_frame_header(z, scan)) return 0; + return 1; +} + +// decode image to YCbCr format +static int stbi__decode_jpeg_image(stbi__jpeg *j) +{ + int m; + for (m = 0; m < 4; m++) { + j->img_comp[m].raw_data = NULL; + j->img_comp[m].raw_coeff = NULL; + } + j->restart_interval = 0; + if (!stbi__decode_jpeg_header(j, STBI__SCAN_load)) return 0; + m = stbi__get_marker(j); + while (!stbi__EOI(m)) { + if (stbi__SOS(m)) { + if (!stbi__process_scan_header(j)) return 0; + if (!stbi__parse_entropy_coded_data(j)) return 0; + if (j->marker == STBI__MARKER_none ) { + // handle 0s at the end of image data from IP Kamera 9060 + while (!stbi__at_eof(j->s)) { + int x = stbi__get8(j->s); + if (x == 255) { + j->marker = stbi__get8(j->s); + break; + } + } + // if we reach eof without hitting a marker, stbi__get_marker() below will fail and we'll eventually return 0 + } + } else if (stbi__DNL(m)) { + int Ld = stbi__get16be(j->s); + stbi__uint32 NL = stbi__get16be(j->s); + if (Ld != 4) return stbi__err("bad DNL len", "Corrupt JPEG"); + if (NL != j->s->img_y) return stbi__err("bad DNL height", "Corrupt JPEG"); + } else { + if (!stbi__process_marker(j, m)) return 0; + } + m = stbi__get_marker(j); + } + if (j->progressive) + stbi__jpeg_finish(j); + return 1; +} + +// static jfif-centered resampling (across block boundaries) + +typedef stbi_uc *(*resample_row_func)(stbi_uc *out, stbi_uc *in0, stbi_uc *in1, + int w, int hs); + +#define stbi__div4(x) ((stbi_uc) ((x) >> 2)) + +static stbi_uc *resample_row_1(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + STBI_NOTUSED(out); + STBI_NOTUSED(in_far); + STBI_NOTUSED(w); + STBI_NOTUSED(hs); + return in_near; +} + +static stbi_uc* stbi__resample_row_v_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // need to generate two samples vertically for every one in input + int i; + STBI_NOTUSED(hs); + for (i=0; i < w; ++i) + out[i] = stbi__div4(3*in_near[i] + in_far[i] + 2); + return out; +} + +static stbi_uc* stbi__resample_row_h_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // need to generate two samples horizontally for every one in input + int i; + stbi_uc *input = in_near; + + if (w == 1) { + // if only one sample, can't do any interpolation + out[0] = out[1] = input[0]; + return out; + } + + out[0] = input[0]; + out[1] = stbi__div4(input[0]*3 + input[1] + 2); + for (i=1; i < w-1; ++i) { + int n = 3*input[i]+2; + out[i*2+0] = stbi__div4(n+input[i-1]); + out[i*2+1] = stbi__div4(n+input[i+1]); + } + out[i*2+0] = stbi__div4(input[w-2]*3 + input[w-1] + 2); + out[i*2+1] = input[w-1]; + + STBI_NOTUSED(in_far); + STBI_NOTUSED(hs); + + return out; +} + +#define stbi__div16(x) ((stbi_uc) ((x) >> 4)) + +static stbi_uc *stbi__resample_row_hv_2(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // need to generate 2x2 samples for every one in input + int i,t0,t1; + if (w == 1) { + out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2); + return out; + } + + t1 = 3*in_near[0] + in_far[0]; + out[0] = stbi__div4(t1+2); + for (i=1; i < w; ++i) { + t0 = t1; + t1 = 3*in_near[i]+in_far[i]; + out[i*2-1] = stbi__div16(3*t0 + t1 + 8); + out[i*2 ] = stbi__div16(3*t1 + t0 + 8); + } + out[w*2-1] = stbi__div4(t1+2); + + STBI_NOTUSED(hs); + + return out; +} + +#if defined(STBI_SSE2) || defined(STBI_NEON) +static stbi_uc *stbi__resample_row_hv_2_simd(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // need to generate 2x2 samples for every one in input + int i=0,t0,t1; + + if (w == 1) { + out[0] = out[1] = stbi__div4(3*in_near[0] + in_far[0] + 2); + return out; + } + + t1 = 3*in_near[0] + in_far[0]; + // process groups of 8 pixels for as long as we can. + // note we can't handle the last pixel in a row in this loop + // because we need to handle the filter boundary conditions. + for (; i < ((w-1) & ~7); i += 8) { +#if defined(STBI_SSE2) + // load and perform the vertical filtering pass + // this uses 3*x + y = 4*x + (y - x) + __m128i zero = _mm_setzero_si128(); + __m128i farb = _mm_loadl_epi64((__m128i *) (in_far + i)); + __m128i nearb = _mm_loadl_epi64((__m128i *) (in_near + i)); + __m128i farw = _mm_unpacklo_epi8(farb, zero); + __m128i nearw = _mm_unpacklo_epi8(nearb, zero); + __m128i diff = _mm_sub_epi16(farw, nearw); + __m128i nears = _mm_slli_epi16(nearw, 2); + __m128i curr = _mm_add_epi16(nears, diff); // current row + + // horizontal filter works the same based on shifted vers of current + // row. "prev" is current row shifted right by 1 pixel; we need to + // insert the previous pixel value (from t1). + // "next" is current row shifted left by 1 pixel, with first pixel + // of next block of 8 pixels added in. + __m128i prv0 = _mm_slli_si128(curr, 2); + __m128i nxt0 = _mm_srli_si128(curr, 2); + __m128i prev = _mm_insert_epi16(prv0, t1, 0); + __m128i next = _mm_insert_epi16(nxt0, 3*in_near[i+8] + in_far[i+8], 7); + + // horizontal filter, polyphase implementation since it's convenient: + // even pixels = 3*cur + prev = cur*4 + (prev - cur) + // odd pixels = 3*cur + next = cur*4 + (next - cur) + // note the shared term. + __m128i bias = _mm_set1_epi16(8); + __m128i curs = _mm_slli_epi16(curr, 2); + __m128i prvd = _mm_sub_epi16(prev, curr); + __m128i nxtd = _mm_sub_epi16(next, curr); + __m128i curb = _mm_add_epi16(curs, bias); + __m128i even = _mm_add_epi16(prvd, curb); + __m128i odd = _mm_add_epi16(nxtd, curb); + + // interleave even and odd pixels, then undo scaling. + __m128i int0 = _mm_unpacklo_epi16(even, odd); + __m128i int1 = _mm_unpackhi_epi16(even, odd); + __m128i de0 = _mm_srli_epi16(int0, 4); + __m128i de1 = _mm_srli_epi16(int1, 4); + + // pack and write output + __m128i outv = _mm_packus_epi16(de0, de1); + _mm_storeu_si128((__m128i *) (out + i*2), outv); +#elif defined(STBI_NEON) + // load and perform the vertical filtering pass + // this uses 3*x + y = 4*x + (y - x) + uint8x8_t farb = vld1_u8(in_far + i); + uint8x8_t nearb = vld1_u8(in_near + i); + int16x8_t diff = vreinterpretq_s16_u16(vsubl_u8(farb, nearb)); + int16x8_t nears = vreinterpretq_s16_u16(vshll_n_u8(nearb, 2)); + int16x8_t curr = vaddq_s16(nears, diff); // current row + + // horizontal filter works the same based on shifted vers of current + // row. "prev" is current row shifted right by 1 pixel; we need to + // insert the previous pixel value (from t1). + // "next" is current row shifted left by 1 pixel, with first pixel + // of next block of 8 pixels added in. + int16x8_t prv0 = vextq_s16(curr, curr, 7); + int16x8_t nxt0 = vextq_s16(curr, curr, 1); + int16x8_t prev = vsetq_lane_s16(t1, prv0, 0); + int16x8_t next = vsetq_lane_s16(3*in_near[i+8] + in_far[i+8], nxt0, 7); + + // horizontal filter, polyphase implementation since it's convenient: + // even pixels = 3*cur + prev = cur*4 + (prev - cur) + // odd pixels = 3*cur + next = cur*4 + (next - cur) + // note the shared term. + int16x8_t curs = vshlq_n_s16(curr, 2); + int16x8_t prvd = vsubq_s16(prev, curr); + int16x8_t nxtd = vsubq_s16(next, curr); + int16x8_t even = vaddq_s16(curs, prvd); + int16x8_t odd = vaddq_s16(curs, nxtd); + + // undo scaling and round, then store with even/odd phases interleaved + uint8x8x2_t o; + o.val[0] = vqrshrun_n_s16(even, 4); + o.val[1] = vqrshrun_n_s16(odd, 4); + vst2_u8(out + i*2, o); +#endif + + // "previous" value for next iter + t1 = 3*in_near[i+7] + in_far[i+7]; + } + + t0 = t1; + t1 = 3*in_near[i] + in_far[i]; + out[i*2] = stbi__div16(3*t1 + t0 + 8); + + for (++i; i < w; ++i) { + t0 = t1; + t1 = 3*in_near[i]+in_far[i]; + out[i*2-1] = stbi__div16(3*t0 + t1 + 8); + out[i*2 ] = stbi__div16(3*t1 + t0 + 8); + } + out[w*2-1] = stbi__div4(t1+2); + + STBI_NOTUSED(hs); + + return out; +} +#endif + +static stbi_uc *stbi__resample_row_generic(stbi_uc *out, stbi_uc *in_near, stbi_uc *in_far, int w, int hs) +{ + // resample with nearest-neighbor + int i,j; + STBI_NOTUSED(in_far); + for (i=0; i < w; ++i) + for (j=0; j < hs; ++j) + out[i*hs+j] = in_near[i]; + return out; +} + +// this is a reduced-precision calculation of YCbCr-to-RGB introduced +// to make sure the code produces the same results in both SIMD and scalar +#define stbi__float2fixed(x) (((int) ((x) * 4096.0f + 0.5f)) << 8) +static void stbi__YCbCr_to_RGB_row(stbi_uc *out, const stbi_uc *y, const stbi_uc *pcb, const stbi_uc *pcr, int count, int step) +{ + int i; + for (i=0; i < count; ++i) { + int y_fixed = (y[i] << 20) + (1<<19); // rounding + int r,g,b; + int cr = pcr[i] - 128; + int cb = pcb[i] - 128; + r = y_fixed + cr* stbi__float2fixed(1.40200f); + g = y_fixed + (cr*-stbi__float2fixed(0.71414f)) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000); + b = y_fixed + cb* stbi__float2fixed(1.77200f); + r >>= 20; + g >>= 20; + b >>= 20; + if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; } + if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; } + if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; } + out[0] = (stbi_uc)r; + out[1] = (stbi_uc)g; + out[2] = (stbi_uc)b; + out[3] = 255; + out += step; + } +} + +#if defined(STBI_SSE2) || defined(STBI_NEON) +static void stbi__YCbCr_to_RGB_simd(stbi_uc *out, stbi_uc const *y, stbi_uc const *pcb, stbi_uc const *pcr, int count, int step) +{ + int i = 0; + +#ifdef STBI_SSE2 + // step == 3 is pretty ugly on the final interleave, and i'm not convinced + // it's useful in practice (you wouldn't use it for textures, for example). + // so just accelerate step == 4 case. + if (step == 4) { + // this is a fairly straightforward implementation and not super-optimized. + __m128i signflip = _mm_set1_epi8(-0x80); + __m128i cr_const0 = _mm_set1_epi16( (short) ( 1.40200f*4096.0f+0.5f)); + __m128i cr_const1 = _mm_set1_epi16( - (short) ( 0.71414f*4096.0f+0.5f)); + __m128i cb_const0 = _mm_set1_epi16( - (short) ( 0.34414f*4096.0f+0.5f)); + __m128i cb_const1 = _mm_set1_epi16( (short) ( 1.77200f*4096.0f+0.5f)); + __m128i y_bias = _mm_set1_epi8((char) (unsigned char) 128); + __m128i xw = _mm_set1_epi16(255); // alpha channel + + for (; i+7 < count; i += 8) { + // load + __m128i y_bytes = _mm_loadl_epi64((__m128i *) (y+i)); + __m128i cr_bytes = _mm_loadl_epi64((__m128i *) (pcr+i)); + __m128i cb_bytes = _mm_loadl_epi64((__m128i *) (pcb+i)); + __m128i cr_biased = _mm_xor_si128(cr_bytes, signflip); // -128 + __m128i cb_biased = _mm_xor_si128(cb_bytes, signflip); // -128 + + // unpack to short (and left-shift cr, cb by 8) + __m128i yw = _mm_unpacklo_epi8(y_bias, y_bytes); + __m128i crw = _mm_unpacklo_epi8(_mm_setzero_si128(), cr_biased); + __m128i cbw = _mm_unpacklo_epi8(_mm_setzero_si128(), cb_biased); + + // color transform + __m128i yws = _mm_srli_epi16(yw, 4); + __m128i cr0 = _mm_mulhi_epi16(cr_const0, crw); + __m128i cb0 = _mm_mulhi_epi16(cb_const0, cbw); + __m128i cb1 = _mm_mulhi_epi16(cbw, cb_const1); + __m128i cr1 = _mm_mulhi_epi16(crw, cr_const1); + __m128i rws = _mm_add_epi16(cr0, yws); + __m128i gwt = _mm_add_epi16(cb0, yws); + __m128i bws = _mm_add_epi16(yws, cb1); + __m128i gws = _mm_add_epi16(gwt, cr1); + + // descale + __m128i rw = _mm_srai_epi16(rws, 4); + __m128i bw = _mm_srai_epi16(bws, 4); + __m128i gw = _mm_srai_epi16(gws, 4); + + // back to byte, set up for transpose + __m128i brb = _mm_packus_epi16(rw, bw); + __m128i gxb = _mm_packus_epi16(gw, xw); + + // transpose to interleave channels + __m128i t0 = _mm_unpacklo_epi8(brb, gxb); + __m128i t1 = _mm_unpackhi_epi8(brb, gxb); + __m128i o0 = _mm_unpacklo_epi16(t0, t1); + __m128i o1 = _mm_unpackhi_epi16(t0, t1); + + // store + _mm_storeu_si128((__m128i *) (out + 0), o0); + _mm_storeu_si128((__m128i *) (out + 16), o1); + out += 32; + } + } +#endif + +#ifdef STBI_NEON + // in this version, step=3 support would be easy to add. but is there demand? + if (step == 4) { + // this is a fairly straightforward implementation and not super-optimized. + uint8x8_t signflip = vdup_n_u8(0x80); + int16x8_t cr_const0 = vdupq_n_s16( (short) ( 1.40200f*4096.0f+0.5f)); + int16x8_t cr_const1 = vdupq_n_s16( - (short) ( 0.71414f*4096.0f+0.5f)); + int16x8_t cb_const0 = vdupq_n_s16( - (short) ( 0.34414f*4096.0f+0.5f)); + int16x8_t cb_const1 = vdupq_n_s16( (short) ( 1.77200f*4096.0f+0.5f)); + + for (; i+7 < count; i += 8) { + // load + uint8x8_t y_bytes = vld1_u8(y + i); + uint8x8_t cr_bytes = vld1_u8(pcr + i); + uint8x8_t cb_bytes = vld1_u8(pcb + i); + int8x8_t cr_biased = vreinterpret_s8_u8(vsub_u8(cr_bytes, signflip)); + int8x8_t cb_biased = vreinterpret_s8_u8(vsub_u8(cb_bytes, signflip)); + + // expand to s16 + int16x8_t yws = vreinterpretq_s16_u16(vshll_n_u8(y_bytes, 4)); + int16x8_t crw = vshll_n_s8(cr_biased, 7); + int16x8_t cbw = vshll_n_s8(cb_biased, 7); + + // color transform + int16x8_t cr0 = vqdmulhq_s16(crw, cr_const0); + int16x8_t cb0 = vqdmulhq_s16(cbw, cb_const0); + int16x8_t cr1 = vqdmulhq_s16(crw, cr_const1); + int16x8_t cb1 = vqdmulhq_s16(cbw, cb_const1); + int16x8_t rws = vaddq_s16(yws, cr0); + int16x8_t gws = vaddq_s16(vaddq_s16(yws, cb0), cr1); + int16x8_t bws = vaddq_s16(yws, cb1); + + // undo scaling, round, convert to byte + uint8x8x4_t o; + o.val[0] = vqrshrun_n_s16(rws, 4); + o.val[1] = vqrshrun_n_s16(gws, 4); + o.val[2] = vqrshrun_n_s16(bws, 4); + o.val[3] = vdup_n_u8(255); + + // store, interleaving r/g/b/a + vst4_u8(out, o); + out += 8*4; + } + } +#endif + + for (; i < count; ++i) { + int y_fixed = (y[i] << 20) + (1<<19); // rounding + int r,g,b; + int cr = pcr[i] - 128; + int cb = pcb[i] - 128; + r = y_fixed + cr* stbi__float2fixed(1.40200f); + g = y_fixed + cr*-stbi__float2fixed(0.71414f) + ((cb*-stbi__float2fixed(0.34414f)) & 0xffff0000); + b = y_fixed + cb* stbi__float2fixed(1.77200f); + r >>= 20; + g >>= 20; + b >>= 20; + if ((unsigned) r > 255) { if (r < 0) r = 0; else r = 255; } + if ((unsigned) g > 255) { if (g < 0) g = 0; else g = 255; } + if ((unsigned) b > 255) { if (b < 0) b = 0; else b = 255; } + out[0] = (stbi_uc)r; + out[1] = (stbi_uc)g; + out[2] = (stbi_uc)b; + out[3] = 255; + out += step; + } +} +#endif + +// set up the kernels +static void stbi__setup_jpeg(stbi__jpeg *j) +{ + j->idct_block_kernel = stbi__idct_block; + j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_row; + j->resample_row_hv_2_kernel = stbi__resample_row_hv_2; + +#ifdef STBI_SSE2 + if (stbi__sse2_available()) { + j->idct_block_kernel = stbi__idct_simd; + j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd; + j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd; + } +#endif + +#ifdef STBI_NEON + j->idct_block_kernel = stbi__idct_simd; + j->YCbCr_to_RGB_kernel = stbi__YCbCr_to_RGB_simd; + j->resample_row_hv_2_kernel = stbi__resample_row_hv_2_simd; +#endif +} + +// clean up the temporary component buffers +static void stbi__cleanup_jpeg(stbi__jpeg *j) +{ + stbi__free_jpeg_components(j, j->s->img_n, 0); +} + +typedef struct +{ + resample_row_func resample; + stbi_uc *line0,*line1; + int hs,vs; // expansion factor in each axis + int w_lores; // horizontal pixels pre-expansion + int ystep; // how far through vertical expansion we are + int ypos; // which pre-expansion row we're on +} stbi__resample; + +// fast 0..255 * 0..255 => 0..255 rounded multiplication +static stbi_uc stbi__blinn_8x8(stbi_uc x, stbi_uc y) +{ + unsigned int t = x*y + 128; + return (stbi_uc) ((t + (t >>8)) >> 8); +} + +static stbi_uc *load_jpeg_image(stbi__jpeg *z, int *out_x, int *out_y, int *comp, int req_comp) +{ + int n, decode_n, is_rgb; + z->s->img_n = 0; // make stbi__cleanup_jpeg safe + + // validate req_comp + if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error"); + + // load a jpeg image from whichever source, but leave in YCbCr format + if (!stbi__decode_jpeg_image(z)) { stbi__cleanup_jpeg(z); return NULL; } + + // determine actual number of components to generate + n = req_comp ? req_comp : z->s->img_n >= 3 ? 3 : 1; + + is_rgb = z->s->img_n == 3 && (z->rgb == 3 || (z->app14_color_transform == 0 && !z->jfif)); + + if (z->s->img_n == 3 && n < 3 && !is_rgb) + decode_n = 1; + else + decode_n = z->s->img_n; + + // resample and color-convert + { + int k; + unsigned int i,j; + stbi_uc *output; + stbi_uc *coutput[4] = { NULL, NULL, NULL, NULL }; + + stbi__resample res_comp[4]; + + for (k=0; k < decode_n; ++k) { + stbi__resample *r = &res_comp[k]; + + // allocate line buffer big enough for upsampling off the edges + // with upsample factor of 4 + z->img_comp[k].linebuf = (stbi_uc *) stbi__malloc(z->s->img_x + 3); + if (!z->img_comp[k].linebuf) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); } + + r->hs = z->img_h_max / z->img_comp[k].h; + r->vs = z->img_v_max / z->img_comp[k].v; + r->ystep = r->vs >> 1; + r->w_lores = (z->s->img_x + r->hs-1) / r->hs; + r->ypos = 0; + r->line0 = r->line1 = z->img_comp[k].data; + + if (r->hs == 1 && r->vs == 1) r->resample = resample_row_1; + else if (r->hs == 1 && r->vs == 2) r->resample = stbi__resample_row_v_2; + else if (r->hs == 2 && r->vs == 1) r->resample = stbi__resample_row_h_2; + else if (r->hs == 2 && r->vs == 2) r->resample = z->resample_row_hv_2_kernel; + else r->resample = stbi__resample_row_generic; + } + + // can't error after this so, this is safe + output = (stbi_uc *) stbi__malloc_mad3(n, z->s->img_x, z->s->img_y, 1); + if (!output) { stbi__cleanup_jpeg(z); return stbi__errpuc("outofmem", "Out of memory"); } + + // now go ahead and resample + for (j=0; j < z->s->img_y; ++j) { + stbi_uc *out = output + n * z->s->img_x * j; + for (k=0; k < decode_n; ++k) { + stbi__resample *r = &res_comp[k]; + int y_bot = r->ystep >= (r->vs >> 1); + coutput[k] = r->resample(z->img_comp[k].linebuf, + y_bot ? r->line1 : r->line0, + y_bot ? r->line0 : r->line1, + r->w_lores, r->hs); + if (++r->ystep >= r->vs) { + r->ystep = 0; + r->line0 = r->line1; + if (++r->ypos < z->img_comp[k].y) + r->line1 += z->img_comp[k].w2; + } + } + if (n >= 3) { + stbi_uc *y = coutput[0]; + if (z->s->img_n == 3) { + if (is_rgb) { + for (i=0; i < z->s->img_x; ++i) { + out[0] = y[i]; + out[1] = coutput[1][i]; + out[2] = coutput[2][i]; + out[3] = 255; + out += n; + } + } else { + z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n); + } + } else if (z->s->img_n == 4) { + if (z->app14_color_transform == 0) { // CMYK + for (i=0; i < z->s->img_x; ++i) { + stbi_uc m = coutput[3][i]; + out[0] = stbi__blinn_8x8(coutput[0][i], m); + out[1] = stbi__blinn_8x8(coutput[1][i], m); + out[2] = stbi__blinn_8x8(coutput[2][i], m); + out[3] = 255; + out += n; + } + } else if (z->app14_color_transform == 2) { // YCCK + z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n); + for (i=0; i < z->s->img_x; ++i) { + stbi_uc m = coutput[3][i]; + out[0] = stbi__blinn_8x8(255 - out[0], m); + out[1] = stbi__blinn_8x8(255 - out[1], m); + out[2] = stbi__blinn_8x8(255 - out[2], m); + out += n; + } + } else { // YCbCr + alpha? Ignore the fourth channel for now + z->YCbCr_to_RGB_kernel(out, y, coutput[1], coutput[2], z->s->img_x, n); + } + } else + for (i=0; i < z->s->img_x; ++i) { + out[0] = out[1] = out[2] = y[i]; + out[3] = 255; // not used if n==3 + out += n; + } + } else { + if (is_rgb) { + if (n == 1) + for (i=0; i < z->s->img_x; ++i) + *out++ = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]); + else { + for (i=0; i < z->s->img_x; ++i, out += 2) { + out[0] = stbi__compute_y(coutput[0][i], coutput[1][i], coutput[2][i]); + out[1] = 255; + } + } + } else if (z->s->img_n == 4 && z->app14_color_transform == 0) { + for (i=0; i < z->s->img_x; ++i) { + stbi_uc m = coutput[3][i]; + stbi_uc r = stbi__blinn_8x8(coutput[0][i], m); + stbi_uc g = stbi__blinn_8x8(coutput[1][i], m); + stbi_uc b = stbi__blinn_8x8(coutput[2][i], m); + out[0] = stbi__compute_y(r, g, b); + out[1] = 255; + out += n; + } + } else if (z->s->img_n == 4 && z->app14_color_transform == 2) { + for (i=0; i < z->s->img_x; ++i) { + out[0] = stbi__blinn_8x8(255 - coutput[0][i], coutput[3][i]); + out[1] = 255; + out += n; + } + } else { + stbi_uc *y = coutput[0]; + if (n == 1) + for (i=0; i < z->s->img_x; ++i) out[i] = y[i]; + else + for (i=0; i < z->s->img_x; ++i) { *out++ = y[i]; *out++ = 255; } + } + } + } + stbi__cleanup_jpeg(z); + *out_x = z->s->img_x; + *out_y = z->s->img_y; + if (comp) *comp = z->s->img_n >= 3 ? 3 : 1; // report original components, not output + return output; + } +} + +static void *stbi__jpeg_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + unsigned char* result; + stbi__jpeg* j = (stbi__jpeg*) stbi__malloc(sizeof(stbi__jpeg)); + STBI_NOTUSED(ri); + j->s = s; + stbi__setup_jpeg(j); + result = load_jpeg_image(j, x,y,comp,req_comp); + STBI_FREE(j); + return result; +} + +static int stbi__jpeg_test(stbi__context *s) +{ + int r; + stbi__jpeg* j = (stbi__jpeg*)stbi__malloc(sizeof(stbi__jpeg)); + j->s = s; + stbi__setup_jpeg(j); + r = stbi__decode_jpeg_header(j, STBI__SCAN_type); + stbi__rewind(s); + STBI_FREE(j); + return r; +} + +static int stbi__jpeg_info_raw(stbi__jpeg *j, int *x, int *y, int *comp) +{ + if (!stbi__decode_jpeg_header(j, STBI__SCAN_header)) { + stbi__rewind( j->s ); + return 0; + } + if (x) *x = j->s->img_x; + if (y) *y = j->s->img_y; + if (comp) *comp = j->s->img_n >= 3 ? 3 : 1; + return 1; +} + +static int stbi__jpeg_info(stbi__context *s, int *x, int *y, int *comp) +{ + int result; + stbi__jpeg* j = (stbi__jpeg*) (stbi__malloc(sizeof(stbi__jpeg))); + j->s = s; + result = stbi__jpeg_info_raw(j, x, y, comp); + STBI_FREE(j); + return result; +} +#endif + +// public domain zlib decode v0.2 Sean Barrett 2006-11-18 +// simple implementation +// - all input must be provided in an upfront buffer +// - all output is written to a single output buffer (can malloc/realloc) +// performance +// - fast huffman + +#ifndef STBI_NO_ZLIB + +// fast-way is faster to check than jpeg huffman, but slow way is slower +#define STBI__ZFAST_BITS 9 // accelerate all cases in default tables +#define STBI__ZFAST_MASK ((1 << STBI__ZFAST_BITS) - 1) + +// zlib-style huffman encoding +// (jpegs packs from left, zlib from right, so can't share code) +typedef struct +{ + stbi__uint16 fast[1 << STBI__ZFAST_BITS]; + stbi__uint16 firstcode[16]; + int maxcode[17]; + stbi__uint16 firstsymbol[16]; + stbi_uc size[288]; + stbi__uint16 value[288]; +} stbi__zhuffman; + +stbi_inline static int stbi__bitreverse16(int n) +{ + n = ((n & 0xAAAA) >> 1) | ((n & 0x5555) << 1); + n = ((n & 0xCCCC) >> 2) | ((n & 0x3333) << 2); + n = ((n & 0xF0F0) >> 4) | ((n & 0x0F0F) << 4); + n = ((n & 0xFF00) >> 8) | ((n & 0x00FF) << 8); + return n; +} + +stbi_inline static int stbi__bit_reverse(int v, int bits) +{ + STBI_ASSERT(bits <= 16); + // to bit reverse n bits, reverse 16 and shift + // e.g. 11 bits, bit reverse and shift away 5 + return stbi__bitreverse16(v) >> (16-bits); +} + +static int stbi__zbuild_huffman(stbi__zhuffman *z, const stbi_uc *sizelist, int num) +{ + int i,k=0; + int code, next_code[16], sizes[17]; + + // DEFLATE spec for generating codes + memset(sizes, 0, sizeof(sizes)); + memset(z->fast, 0, sizeof(z->fast)); + for (i=0; i < num; ++i) + ++sizes[sizelist[i]]; + sizes[0] = 0; + for (i=1; i < 16; ++i) + if (sizes[i] > (1 << i)) + return stbi__err("bad sizes", "Corrupt PNG"); + code = 0; + for (i=1; i < 16; ++i) { + next_code[i] = code; + z->firstcode[i] = (stbi__uint16) code; + z->firstsymbol[i] = (stbi__uint16) k; + code = (code + sizes[i]); + if (sizes[i]) + if (code-1 >= (1 << i)) return stbi__err("bad codelengths","Corrupt PNG"); + z->maxcode[i] = code << (16-i); // preshift for inner loop + code <<= 1; + k += sizes[i]; + } + z->maxcode[16] = 0x10000; // sentinel + for (i=0; i < num; ++i) { + int s = sizelist[i]; + if (s) { + int c = next_code[s] - z->firstcode[s] + z->firstsymbol[s]; + stbi__uint16 fastv = (stbi__uint16) ((s << 9) | i); + z->size [c] = (stbi_uc ) s; + z->value[c] = (stbi__uint16) i; + if (s <= STBI__ZFAST_BITS) { + int j = stbi__bit_reverse(next_code[s],s); + while (j < (1 << STBI__ZFAST_BITS)) { + z->fast[j] = fastv; + j += (1 << s); + } + } + ++next_code[s]; + } + } + return 1; +} + +// zlib-from-memory implementation for PNG reading +// because PNG allows splitting the zlib stream arbitrarily, +// and it's annoying structurally to have PNG call ZLIB call PNG, +// we require PNG read all the IDATs and combine them into a single +// memory buffer + +typedef struct +{ + stbi_uc *zbuffer, *zbuffer_end; + int num_bits; + stbi__uint32 code_buffer; + + char *zout; + char *zout_start; + char *zout_end; + int z_expandable; + + stbi__zhuffman z_length, z_distance; +} stbi__zbuf; + +stbi_inline static int stbi__zeof(stbi__zbuf *z) +{ + return (z->zbuffer >= z->zbuffer_end); +} + +stbi_inline static stbi_uc stbi__zget8(stbi__zbuf *z) +{ + return stbi__zeof(z) ? 0 : *z->zbuffer++; +} + +static void stbi__fill_bits(stbi__zbuf *z) +{ + do { + if (z->code_buffer >= (1U << z->num_bits)) { + z->zbuffer = z->zbuffer_end; /* treat this as EOF so we fail. */ + return; + } + z->code_buffer |= (unsigned int) stbi__zget8(z) << z->num_bits; + z->num_bits += 8; + } while (z->num_bits <= 24); +} + +stbi_inline static unsigned int stbi__zreceive(stbi__zbuf *z, int n) +{ + unsigned int k; + if (z->num_bits < n) stbi__fill_bits(z); + k = z->code_buffer & ((1 << n) - 1); + z->code_buffer >>= n; + z->num_bits -= n; + return k; +} + +static int stbi__zhuffman_decode_slowpath(stbi__zbuf *a, stbi__zhuffman *z) +{ + int b,s,k; + // not resolved by fast table, so compute it the slow way + // use jpeg approach, which requires MSbits at top + k = stbi__bit_reverse(a->code_buffer, 16); + for (s=STBI__ZFAST_BITS+1; ; ++s) + if (k < z->maxcode[s]) + break; + if (s >= 16) return -1; // invalid code! + // code size is s, so: + b = (k >> (16-s)) - z->firstcode[s] + z->firstsymbol[s]; + if (b >= sizeof (z->size)) return -1; // some data was corrupt somewhere! + if (z->size[b] != s) return -1; // was originally an assert, but report failure instead. + a->code_buffer >>= s; + a->num_bits -= s; + return z->value[b]; +} + +stbi_inline static int stbi__zhuffman_decode(stbi__zbuf *a, stbi__zhuffman *z) +{ + int b,s; + if (a->num_bits < 16) { + if (stbi__zeof(a)) { + return -1; /* report error for unexpected end of data. */ + } + stbi__fill_bits(a); + } + b = z->fast[a->code_buffer & STBI__ZFAST_MASK]; + if (b) { + s = b >> 9; + a->code_buffer >>= s; + a->num_bits -= s; + return b & 511; + } + return stbi__zhuffman_decode_slowpath(a, z); +} + +static int stbi__zexpand(stbi__zbuf *z, char *zout, int n) // need to make room for n bytes +{ + char *q; + unsigned int cur, limit, old_limit; + z->zout = zout; + if (!z->z_expandable) return stbi__err("output buffer limit","Corrupt PNG"); + cur = (unsigned int) (z->zout - z->zout_start); + limit = old_limit = (unsigned) (z->zout_end - z->zout_start); + if (UINT_MAX - cur < (unsigned) n) return stbi__err("outofmem", "Out of memory"); + while (cur + n > limit) { + if(limit > UINT_MAX / 2) return stbi__err("outofmem", "Out of memory"); + limit *= 2; + } + q = (char *) STBI_REALLOC_SIZED(z->zout_start, old_limit, limit); + STBI_NOTUSED(old_limit); + if (q == NULL) return stbi__err("outofmem", "Out of memory"); + z->zout_start = q; + z->zout = q + cur; + z->zout_end = q + limit; + return 1; +} + +static const int stbi__zlength_base[31] = { + 3,4,5,6,7,8,9,10,11,13, + 15,17,19,23,27,31,35,43,51,59, + 67,83,99,115,131,163,195,227,258,0,0 }; + +static const int stbi__zlength_extra[31]= +{ 0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,0,0,0 }; + +static const int stbi__zdist_base[32] = { 1,2,3,4,5,7,9,13,17,25,33,49,65,97,129,193, +257,385,513,769,1025,1537,2049,3073,4097,6145,8193,12289,16385,24577,0,0}; + +static const int stbi__zdist_extra[32] = +{ 0,0,0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,8,8,9,9,10,10,11,11,12,12,13,13}; + +static int stbi__parse_huffman_block(stbi__zbuf *a) +{ + char *zout = a->zout; + for(;;) { + int z = stbi__zhuffman_decode(a, &a->z_length); + if (z < 256) { + if (z < 0) return stbi__err("bad huffman code","Corrupt PNG"); // error in huffman codes + if (zout >= a->zout_end) { + if (!stbi__zexpand(a, zout, 1)) return 0; + zout = a->zout; + } + *zout++ = (char) z; + } else { + stbi_uc *p; + int len,dist; + if (z == 256) { + a->zout = zout; + return 1; + } + z -= 257; + len = stbi__zlength_base[z]; + if (stbi__zlength_extra[z]) len += stbi__zreceive(a, stbi__zlength_extra[z]); + z = stbi__zhuffman_decode(a, &a->z_distance); + if (z < 0) return stbi__err("bad huffman code","Corrupt PNG"); + dist = stbi__zdist_base[z]; + if (stbi__zdist_extra[z]) dist += stbi__zreceive(a, stbi__zdist_extra[z]); + if (zout - a->zout_start < dist) return stbi__err("bad dist","Corrupt PNG"); + if (zout + len > a->zout_end) { + if (!stbi__zexpand(a, zout, len)) return 0; + zout = a->zout; + } + p = (stbi_uc *) (zout - dist); + if (dist == 1) { // run of one byte; common in images. + stbi_uc v = *p; + if (len) { do *zout++ = v; while (--len); } + } else { + if (len) { do *zout++ = *p++; while (--len); } + } + } + } +} + +static int stbi__compute_huffman_codes(stbi__zbuf *a) +{ + static const stbi_uc length_dezigzag[19] = { 16,17,18,0,8,7,9,6,10,5,11,4,12,3,13,2,14,1,15 }; + stbi__zhuffman z_codelength; + stbi_uc lencodes[286+32+137];//padding for maximum single op + stbi_uc codelength_sizes[19]; + int i,n; + + int hlit = stbi__zreceive(a,5) + 257; + int hdist = stbi__zreceive(a,5) + 1; + int hclen = stbi__zreceive(a,4) + 4; + int ntot = hlit + hdist; + + memset(codelength_sizes, 0, sizeof(codelength_sizes)); + for (i=0; i < hclen; ++i) { + int s = stbi__zreceive(a,3); + codelength_sizes[length_dezigzag[i]] = (stbi_uc) s; + } + if (!stbi__zbuild_huffman(&z_codelength, codelength_sizes, 19)) return 0; + + n = 0; + while (n < ntot) { + int c = stbi__zhuffman_decode(a, &z_codelength); + if (c < 0 || c >= 19) return stbi__err("bad codelengths", "Corrupt PNG"); + if (c < 16) + lencodes[n++] = (stbi_uc) c; + else { + stbi_uc fill = 0; + if (c == 16) { + c = stbi__zreceive(a,2)+3; + if (n == 0) return stbi__err("bad codelengths", "Corrupt PNG"); + fill = lencodes[n-1]; + } else if (c == 17) { + c = stbi__zreceive(a,3)+3; + } else if (c == 18) { + c = stbi__zreceive(a,7)+11; + } else { + return stbi__err("bad codelengths", "Corrupt PNG"); + } + if (ntot - n < c) return stbi__err("bad codelengths", "Corrupt PNG"); + memset(lencodes+n, fill, c); + n += c; + } + } + if (n != ntot) return stbi__err("bad codelengths","Corrupt PNG"); + if (!stbi__zbuild_huffman(&a->z_length, lencodes, hlit)) return 0; + if (!stbi__zbuild_huffman(&a->z_distance, lencodes+hlit, hdist)) return 0; + return 1; +} + +static int stbi__parse_uncompressed_block(stbi__zbuf *a) +{ + stbi_uc header[4]; + int len,nlen,k; + if (a->num_bits & 7) + stbi__zreceive(a, a->num_bits & 7); // discard + // drain the bit-packed data into header + k = 0; + while (a->num_bits > 0) { + header[k++] = (stbi_uc) (a->code_buffer & 255); // suppress MSVC run-time check + a->code_buffer >>= 8; + a->num_bits -= 8; + } + if (a->num_bits < 0) return stbi__err("zlib corrupt","Corrupt PNG"); + // now fill header the normal way + while (k < 4) + header[k++] = stbi__zget8(a); + len = header[1] * 256 + header[0]; + nlen = header[3] * 256 + header[2]; + if (nlen != (len ^ 0xffff)) return stbi__err("zlib corrupt","Corrupt PNG"); + if (a->zbuffer + len > a->zbuffer_end) return stbi__err("read past buffer","Corrupt PNG"); + if (a->zout + len > a->zout_end) + if (!stbi__zexpand(a, a->zout, len)) return 0; + memcpy(a->zout, a->zbuffer, len); + a->zbuffer += len; + a->zout += len; + return 1; +} + +static int stbi__parse_zlib_header(stbi__zbuf *a) +{ + int cmf = stbi__zget8(a); + int cm = cmf & 15; + /* int cinfo = cmf >> 4; */ + int flg = stbi__zget8(a); + if (stbi__zeof(a)) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec + if ((cmf*256+flg) % 31 != 0) return stbi__err("bad zlib header","Corrupt PNG"); // zlib spec + if (flg & 32) return stbi__err("no preset dict","Corrupt PNG"); // preset dictionary not allowed in png + if (cm != 8) return stbi__err("bad compression","Corrupt PNG"); // DEFLATE required for png + // window = 1 << (8 + cinfo)... but who cares, we fully buffer output + return 1; +} + +static const stbi_uc stbi__zdefault_length[288] = +{ + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, + 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, + 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, + 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, + 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, + 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,8,8,8,8,8,8,8,8 +}; +static const stbi_uc stbi__zdefault_distance[32] = +{ + 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5 +}; +/* +Init algorithm: +{ + int i; // use <= to match clearly with spec + for (i=0; i <= 143; ++i) stbi__zdefault_length[i] = 8; + for ( ; i <= 255; ++i) stbi__zdefault_length[i] = 9; + for ( ; i <= 279; ++i) stbi__zdefault_length[i] = 7; + for ( ; i <= 287; ++i) stbi__zdefault_length[i] = 8; + + for (i=0; i <= 31; ++i) stbi__zdefault_distance[i] = 5; +} +*/ + +static int stbi__parse_zlib(stbi__zbuf *a, int parse_header) +{ + int final, type; + if (parse_header) + if (!stbi__parse_zlib_header(a)) return 0; + a->num_bits = 0; + a->code_buffer = 0; + do { + final = stbi__zreceive(a,1); + type = stbi__zreceive(a,2); + if (type == 0) { + if (!stbi__parse_uncompressed_block(a)) return 0; + } else if (type == 3) { + return 0; + } else { + if (type == 1) { + // use fixed code lengths + if (!stbi__zbuild_huffman(&a->z_length , stbi__zdefault_length , 288)) return 0; + if (!stbi__zbuild_huffman(&a->z_distance, stbi__zdefault_distance, 32)) return 0; + } else { + if (!stbi__compute_huffman_codes(a)) return 0; + } + if (!stbi__parse_huffman_block(a)) return 0; + } + } while (!final); + return 1; +} + +static int stbi__do_zlib(stbi__zbuf *a, char *obuf, int olen, int exp, int parse_header) +{ + a->zout_start = obuf; + a->zout = obuf; + a->zout_end = obuf + olen; + a->z_expandable = exp; + + return stbi__parse_zlib(a, parse_header); +} + +STBIDEF char *stbi_zlib_decode_malloc_guesssize(const char *buffer, int len, int initial_size, int *outlen) +{ + stbi__zbuf a; + char *p = (char *) stbi__malloc(initial_size); + if (p == NULL) return NULL; + a.zbuffer = (stbi_uc *) buffer; + a.zbuffer_end = (stbi_uc *) buffer + len; + if (stbi__do_zlib(&a, p, initial_size, 1, 1)) { + if (outlen) *outlen = (int) (a.zout - a.zout_start); + return a.zout_start; + } else { + STBI_FREE(a.zout_start); + return NULL; + } +} + +STBIDEF char *stbi_zlib_decode_malloc(char const *buffer, int len, int *outlen) +{ + return stbi_zlib_decode_malloc_guesssize(buffer, len, 16384, outlen); +} + +STBIDEF char *stbi_zlib_decode_malloc_guesssize_headerflag(const char *buffer, int len, int initial_size, int *outlen, int parse_header) +{ + stbi__zbuf a; + char *p = (char *) stbi__malloc(initial_size); + if (p == NULL) return NULL; + a.zbuffer = (stbi_uc *) buffer; + a.zbuffer_end = (stbi_uc *) buffer + len; + if (stbi__do_zlib(&a, p, initial_size, 1, parse_header)) { + if (outlen) *outlen = (int) (a.zout - a.zout_start); + return a.zout_start; + } else { + STBI_FREE(a.zout_start); + return NULL; + } +} + +STBIDEF int stbi_zlib_decode_buffer(char *obuffer, int olen, char const *ibuffer, int ilen) +{ + stbi__zbuf a; + a.zbuffer = (stbi_uc *) ibuffer; + a.zbuffer_end = (stbi_uc *) ibuffer + ilen; + if (stbi__do_zlib(&a, obuffer, olen, 0, 1)) + return (int) (a.zout - a.zout_start); + else + return -1; +} + +STBIDEF char *stbi_zlib_decode_noheader_malloc(char const *buffer, int len, int *outlen) +{ + stbi__zbuf a; + char *p = (char *) stbi__malloc(16384); + if (p == NULL) return NULL; + a.zbuffer = (stbi_uc *) buffer; + a.zbuffer_end = (stbi_uc *) buffer+len; + if (stbi__do_zlib(&a, p, 16384, 1, 0)) { + if (outlen) *outlen = (int) (a.zout - a.zout_start); + return a.zout_start; + } else { + STBI_FREE(a.zout_start); + return NULL; + } +} + +STBIDEF int stbi_zlib_decode_noheader_buffer(char *obuffer, int olen, const char *ibuffer, int ilen) +{ + stbi__zbuf a; + a.zbuffer = (stbi_uc *) ibuffer; + a.zbuffer_end = (stbi_uc *) ibuffer + ilen; + if (stbi__do_zlib(&a, obuffer, olen, 0, 0)) + return (int) (a.zout - a.zout_start); + else + return -1; +} +#endif + +// public domain "baseline" PNG decoder v0.10 Sean Barrett 2006-11-18 +// simple implementation +// - only 8-bit samples +// - no CRC checking +// - allocates lots of intermediate memory +// - avoids problem of streaming data between subsystems +// - avoids explicit window management +// performance +// - uses stb_zlib, a PD zlib implementation with fast huffman decoding + +#ifndef STBI_NO_PNG +typedef struct +{ + stbi__uint32 length; + stbi__uint32 type; +} stbi__pngchunk; + +static stbi__pngchunk stbi__get_chunk_header(stbi__context *s) +{ + stbi__pngchunk c; + c.length = stbi__get32be(s); + c.type = stbi__get32be(s); + return c; +} + +static int stbi__check_png_header(stbi__context *s) +{ + static const stbi_uc png_sig[8] = { 137,80,78,71,13,10,26,10 }; + int i; + for (i=0; i < 8; ++i) + if (stbi__get8(s) != png_sig[i]) return stbi__err("bad png sig","Not a PNG"); + return 1; +} + +typedef struct +{ + stbi__context *s; + stbi_uc *idata, *expanded, *out; + int depth; +} stbi__png; + + +enum { + STBI__F_none=0, + STBI__F_sub=1, + STBI__F_up=2, + STBI__F_avg=3, + STBI__F_paeth=4, + // synthetic filters used for first scanline to avoid needing a dummy row of 0s + STBI__F_avg_first, + STBI__F_paeth_first +}; + +static stbi_uc first_row_filter[5] = +{ + STBI__F_none, + STBI__F_sub, + STBI__F_none, + STBI__F_avg_first, + STBI__F_paeth_first +}; + +static int stbi__paeth(int a, int b, int c) +{ + int p = a + b - c; + int pa = abs(p-a); + int pb = abs(p-b); + int pc = abs(p-c); + if (pa <= pb && pa <= pc) return a; + if (pb <= pc) return b; + return c; +} + +static const stbi_uc stbi__depth_scale_table[9] = { 0, 0xff, 0x55, 0, 0x11, 0,0,0, 0x01 }; + +// create the png data from post-deflated data +static int stbi__create_png_image_raw(stbi__png *a, stbi_uc *raw, stbi__uint32 raw_len, int out_n, stbi__uint32 x, stbi__uint32 y, int depth, int color) +{ + int bytes = (depth == 16? 2 : 1); + stbi__context *s = a->s; + stbi__uint32 i,j,stride = x*out_n*bytes; + stbi__uint32 img_len, img_width_bytes; + int k; + int img_n = s->img_n; // copy it into a local for later + + int output_bytes = out_n*bytes; + int filter_bytes = img_n*bytes; + int width = x; + + STBI_ASSERT(out_n == s->img_n || out_n == s->img_n+1); + a->out = (stbi_uc *) stbi__malloc_mad3(x, y, output_bytes, 0); // extra bytes to write off the end into + if (!a->out) return stbi__err("outofmem", "Out of memory"); + + if (!stbi__mad3sizes_valid(img_n, x, depth, 7)) return stbi__err("too large", "Corrupt PNG"); + img_width_bytes = (((img_n * x * depth) + 7) >> 3); + img_len = (img_width_bytes + 1) * y; + + // we used to check for exact match between raw_len and img_len on non-interlaced PNGs, + // but issue #276 reported a PNG in the wild that had extra data at the end (all zeros), + // so just check for raw_len < img_len always. + if (raw_len < img_len) return stbi__err("not enough pixels","Corrupt PNG"); + + for (j=0; j < y; ++j) { + stbi_uc *cur = a->out + stride*j; + stbi_uc *prior; + int filter = *raw++; + + if (filter > 4) + return stbi__err("invalid filter","Corrupt PNG"); + + if (depth < 8) { + if (img_width_bytes > x) return stbi__err("invalid width","Corrupt PNG"); + cur += x*out_n - img_width_bytes; // store output to the rightmost img_len bytes, so we can decode in place + filter_bytes = 1; + width = img_width_bytes; + } + prior = cur - stride; // bugfix: need to compute this after 'cur +=' computation above + + // if first row, use special filter that doesn't sample previous row + if (j == 0) filter = first_row_filter[filter]; + + // handle first byte explicitly + for (k=0; k < filter_bytes; ++k) { + switch (filter) { + case STBI__F_none : cur[k] = raw[k]; break; + case STBI__F_sub : cur[k] = raw[k]; break; + case STBI__F_up : cur[k] = STBI__BYTECAST(raw[k] + prior[k]); break; + case STBI__F_avg : cur[k] = STBI__BYTECAST(raw[k] + (prior[k]>>1)); break; + case STBI__F_paeth : cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(0,prior[k],0)); break; + case STBI__F_avg_first : cur[k] = raw[k]; break; + case STBI__F_paeth_first: cur[k] = raw[k]; break; + } + } + + if (depth == 8) { + if (img_n != out_n) + cur[img_n] = 255; // first pixel + raw += img_n; + cur += out_n; + prior += out_n; + } else if (depth == 16) { + if (img_n != out_n) { + cur[filter_bytes] = 255; // first pixel top byte + cur[filter_bytes+1] = 255; // first pixel bottom byte + } + raw += filter_bytes; + cur += output_bytes; + prior += output_bytes; + } else { + raw += 1; + cur += 1; + prior += 1; + } + + // this is a little gross, so that we don't switch per-pixel or per-component + if (depth < 8 || img_n == out_n) { + int nk = (width - 1)*filter_bytes; + #define STBI__CASE(f) \ + case f: \ + for (k=0; k < nk; ++k) + switch (filter) { + // "none" filter turns into a memcpy here; make that explicit. + case STBI__F_none: memcpy(cur, raw, nk); break; + STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k-filter_bytes]); } break; + STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break; + STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k-filter_bytes])>>1)); } break; + STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],prior[k],prior[k-filter_bytes])); } break; + STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k-filter_bytes] >> 1)); } break; + STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k-filter_bytes],0,0)); } break; + } + #undef STBI__CASE + raw += nk; + } else { + STBI_ASSERT(img_n+1 == out_n); + #define STBI__CASE(f) \ + case f: \ + for (i=x-1; i >= 1; --i, cur[filter_bytes]=255,raw+=filter_bytes,cur+=output_bytes,prior+=output_bytes) \ + for (k=0; k < filter_bytes; ++k) + switch (filter) { + STBI__CASE(STBI__F_none) { cur[k] = raw[k]; } break; + STBI__CASE(STBI__F_sub) { cur[k] = STBI__BYTECAST(raw[k] + cur[k- output_bytes]); } break; + STBI__CASE(STBI__F_up) { cur[k] = STBI__BYTECAST(raw[k] + prior[k]); } break; + STBI__CASE(STBI__F_avg) { cur[k] = STBI__BYTECAST(raw[k] + ((prior[k] + cur[k- output_bytes])>>1)); } break; + STBI__CASE(STBI__F_paeth) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],prior[k],prior[k- output_bytes])); } break; + STBI__CASE(STBI__F_avg_first) { cur[k] = STBI__BYTECAST(raw[k] + (cur[k- output_bytes] >> 1)); } break; + STBI__CASE(STBI__F_paeth_first) { cur[k] = STBI__BYTECAST(raw[k] + stbi__paeth(cur[k- output_bytes],0,0)); } break; + } + #undef STBI__CASE + + // the loop above sets the high byte of the pixels' alpha, but for + // 16 bit png files we also need the low byte set. we'll do that here. + if (depth == 16) { + cur = a->out + stride*j; // start at the beginning of the row again + for (i=0; i < x; ++i,cur+=output_bytes) { + cur[filter_bytes+1] = 255; + } + } + } + } + + // we make a separate pass to expand bits to pixels; for performance, + // this could run two scanlines behind the above code, so it won't + // intefere with filtering but will still be in the cache. + if (depth < 8) { + for (j=0; j < y; ++j) { + stbi_uc *cur = a->out + stride*j; + stbi_uc *in = a->out + stride*j + x*out_n - img_width_bytes; + // unpack 1/2/4-bit into a 8-bit buffer. allows us to keep the common 8-bit path optimal at minimal cost for 1/2/4-bit + // png guarante byte alignment, if width is not multiple of 8/4/2 we'll decode dummy trailing data that will be skipped in the later loop + stbi_uc scale = (color == 0) ? stbi__depth_scale_table[depth] : 1; // scale grayscale values to 0..255 range + + // note that the final byte might overshoot and write more data than desired. + // we can allocate enough data that this never writes out of memory, but it + // could also overwrite the next scanline. can it overwrite non-empty data + // on the next scanline? yes, consider 1-pixel-wide scanlines with 1-bit-per-pixel. + // so we need to explicitly clamp the final ones + + if (depth == 4) { + for (k=x*img_n; k >= 2; k-=2, ++in) { + *cur++ = scale * ((*in >> 4) ); + *cur++ = scale * ((*in ) & 0x0f); + } + if (k > 0) *cur++ = scale * ((*in >> 4) ); + } else if (depth == 2) { + for (k=x*img_n; k >= 4; k-=4, ++in) { + *cur++ = scale * ((*in >> 6) ); + *cur++ = scale * ((*in >> 4) & 0x03); + *cur++ = scale * ((*in >> 2) & 0x03); + *cur++ = scale * ((*in ) & 0x03); + } + if (k > 0) *cur++ = scale * ((*in >> 6) ); + if (k > 1) *cur++ = scale * ((*in >> 4) & 0x03); + if (k > 2) *cur++ = scale * ((*in >> 2) & 0x03); + } else if (depth == 1) { + for (k=x*img_n; k >= 8; k-=8, ++in) { + *cur++ = scale * ((*in >> 7) ); + *cur++ = scale * ((*in >> 6) & 0x01); + *cur++ = scale * ((*in >> 5) & 0x01); + *cur++ = scale * ((*in >> 4) & 0x01); + *cur++ = scale * ((*in >> 3) & 0x01); + *cur++ = scale * ((*in >> 2) & 0x01); + *cur++ = scale * ((*in >> 1) & 0x01); + *cur++ = scale * ((*in ) & 0x01); + } + if (k > 0) *cur++ = scale * ((*in >> 7) ); + if (k > 1) *cur++ = scale * ((*in >> 6) & 0x01); + if (k > 2) *cur++ = scale * ((*in >> 5) & 0x01); + if (k > 3) *cur++ = scale * ((*in >> 4) & 0x01); + if (k > 4) *cur++ = scale * ((*in >> 3) & 0x01); + if (k > 5) *cur++ = scale * ((*in >> 2) & 0x01); + if (k > 6) *cur++ = scale * ((*in >> 1) & 0x01); + } + if (img_n != out_n) { + int q; + // insert alpha = 255 + cur = a->out + stride*j; + if (img_n == 1) { + for (q=x-1; q >= 0; --q) { + cur[q*2+1] = 255; + cur[q*2+0] = cur[q]; + } + } else { + STBI_ASSERT(img_n == 3); + for (q=x-1; q >= 0; --q) { + cur[q*4+3] = 255; + cur[q*4+2] = cur[q*3+2]; + cur[q*4+1] = cur[q*3+1]; + cur[q*4+0] = cur[q*3+0]; + } + } + } + } + } else if (depth == 16) { + // force the image data from big-endian to platform-native. + // this is done in a separate pass due to the decoding relying + // on the data being untouched, but could probably be done + // per-line during decode if care is taken. + stbi_uc *cur = a->out; + stbi__uint16 *cur16 = (stbi__uint16*)cur; + + for(i=0; i < x*y*out_n; ++i,cur16++,cur+=2) { + *cur16 = (cur[0] << 8) | cur[1]; + } + } + + return 1; +} + +static int stbi__create_png_image(stbi__png *a, stbi_uc *image_data, stbi__uint32 image_data_len, int out_n, int depth, int color, int interlaced) +{ + int bytes = (depth == 16 ? 2 : 1); + int out_bytes = out_n * bytes; + stbi_uc *final; + int p; + if (!interlaced) + return stbi__create_png_image_raw(a, image_data, image_data_len, out_n, a->s->img_x, a->s->img_y, depth, color); + + // de-interlacing + final = (stbi_uc *) stbi__malloc_mad3(a->s->img_x, a->s->img_y, out_bytes, 0); + for (p=0; p < 7; ++p) { + int xorig[] = { 0,4,0,2,0,1,0 }; + int yorig[] = { 0,0,4,0,2,0,1 }; + int xspc[] = { 8,8,4,4,2,2,1 }; + int yspc[] = { 8,8,8,4,4,2,2 }; + int i,j,x,y; + // pass1_x[4] = 0, pass1_x[5] = 1, pass1_x[12] = 1 + x = (a->s->img_x - xorig[p] + xspc[p]-1) / xspc[p]; + y = (a->s->img_y - yorig[p] + yspc[p]-1) / yspc[p]; + if (x && y) { + stbi__uint32 img_len = ((((a->s->img_n * x * depth) + 7) >> 3) + 1) * y; + if (!stbi__create_png_image_raw(a, image_data, image_data_len, out_n, x, y, depth, color)) { + STBI_FREE(final); + return 0; + } + for (j=0; j < y; ++j) { + for (i=0; i < x; ++i) { + int out_y = j*yspc[p]+yorig[p]; + int out_x = i*xspc[p]+xorig[p]; + memcpy(final + out_y*a->s->img_x*out_bytes + out_x*out_bytes, + a->out + (j*x+i)*out_bytes, out_bytes); + } + } + STBI_FREE(a->out); + image_data += img_len; + image_data_len -= img_len; + } + } + a->out = final; + + return 1; +} + +static int stbi__compute_transparency(stbi__png *z, stbi_uc tc[3], int out_n) +{ + stbi__context *s = z->s; + stbi__uint32 i, pixel_count = s->img_x * s->img_y; + stbi_uc *p = z->out; + + // compute color-based transparency, assuming we've + // already got 255 as the alpha value in the output + STBI_ASSERT(out_n == 2 || out_n == 4); + + if (out_n == 2) { + for (i=0; i < pixel_count; ++i) { + p[1] = (p[0] == tc[0] ? 0 : 255); + p += 2; + } + } else { + for (i=0; i < pixel_count; ++i) { + if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2]) + p[3] = 0; + p += 4; + } + } + return 1; +} + +static int stbi__compute_transparency16(stbi__png *z, stbi__uint16 tc[3], int out_n) +{ + stbi__context *s = z->s; + stbi__uint32 i, pixel_count = s->img_x * s->img_y; + stbi__uint16 *p = (stbi__uint16*) z->out; + + // compute color-based transparency, assuming we've + // already got 65535 as the alpha value in the output + STBI_ASSERT(out_n == 2 || out_n == 4); + + if (out_n == 2) { + for (i = 0; i < pixel_count; ++i) { + p[1] = (p[0] == tc[0] ? 0 : 65535); + p += 2; + } + } else { + for (i = 0; i < pixel_count; ++i) { + if (p[0] == tc[0] && p[1] == tc[1] && p[2] == tc[2]) + p[3] = 0; + p += 4; + } + } + return 1; +} + +static int stbi__expand_png_palette(stbi__png *a, stbi_uc *palette, int len, int pal_img_n) +{ + stbi__uint32 i, pixel_count = a->s->img_x * a->s->img_y; + stbi_uc *p, *temp_out, *orig = a->out; + + p = (stbi_uc *) stbi__malloc_mad2(pixel_count, pal_img_n, 0); + if (p == NULL) return stbi__err("outofmem", "Out of memory"); + + // between here and free(out) below, exitting would leak + temp_out = p; + + if (pal_img_n == 3) { + for (i=0; i < pixel_count; ++i) { + int n = orig[i]*4; + p[0] = palette[n ]; + p[1] = palette[n+1]; + p[2] = palette[n+2]; + p += 3; + } + } else { + for (i=0; i < pixel_count; ++i) { + int n = orig[i]*4; + p[0] = palette[n ]; + p[1] = palette[n+1]; + p[2] = palette[n+2]; + p[3] = palette[n+3]; + p += 4; + } + } + STBI_FREE(a->out); + a->out = temp_out; + + STBI_NOTUSED(len); + + return 1; +} + +static int stbi__unpremultiply_on_load = 0; +static int stbi__de_iphone_flag = 0; + +STBIDEF void stbi_set_unpremultiply_on_load(int flag_true_if_should_unpremultiply) +{ + stbi__unpremultiply_on_load = flag_true_if_should_unpremultiply; +} + +STBIDEF void stbi_convert_iphone_png_to_rgb(int flag_true_if_should_convert) +{ + stbi__de_iphone_flag = flag_true_if_should_convert; +} + +static void stbi__de_iphone(stbi__png *z) +{ + stbi__context *s = z->s; + stbi__uint32 i, pixel_count = s->img_x * s->img_y; + stbi_uc *p = z->out; + + if (s->img_out_n == 3) { // convert bgr to rgb + for (i=0; i < pixel_count; ++i) { + stbi_uc t = p[0]; + p[0] = p[2]; + p[2] = t; + p += 3; + } + } else { + STBI_ASSERT(s->img_out_n == 4); + if (stbi__unpremultiply_on_load) { + // convert bgr to rgb and unpremultiply + for (i=0; i < pixel_count; ++i) { + stbi_uc a = p[3]; + stbi_uc t = p[0]; + if (a) { + stbi_uc half = a / 2; + p[0] = (p[2] * 255 + half) / a; + p[1] = (p[1] * 255 + half) / a; + p[2] = ( t * 255 + half) / a; + } else { + p[0] = p[2]; + p[2] = t; + } + p += 4; + } + } else { + // convert bgr to rgb + for (i=0; i < pixel_count; ++i) { + stbi_uc t = p[0]; + p[0] = p[2]; + p[2] = t; + p += 4; + } + } + } +} + +#define STBI__PNG_TYPE(a,b,c,d) (((unsigned) (a) << 24) + ((unsigned) (b) << 16) + ((unsigned) (c) << 8) + (unsigned) (d)) + +static int stbi__parse_png_file(stbi__png *z, int scan, int req_comp) +{ + stbi_uc palette[1024], pal_img_n=0; + stbi_uc has_trans=0, tc[3]={0}; + stbi__uint16 tc16[3]; + stbi__uint32 ioff=0, idata_limit=0, i, pal_len=0; + int first=1,k,interlace=0, color=0, is_iphone=0; + stbi__context *s = z->s; + + z->expanded = NULL; + z->idata = NULL; + z->out = NULL; + + if (!stbi__check_png_header(s)) return 0; + + if (scan == STBI__SCAN_type) return 1; + + for (;;) { + stbi__pngchunk c = stbi__get_chunk_header(s); + switch (c.type) { + case STBI__PNG_TYPE('C','g','B','I'): + is_iphone = 1; + stbi__skip(s, c.length); + break; + case STBI__PNG_TYPE('I','H','D','R'): { + int comp,filter; + if (!first) return stbi__err("multiple IHDR","Corrupt PNG"); + first = 0; + if (c.length != 13) return stbi__err("bad IHDR len","Corrupt PNG"); + s->img_x = stbi__get32be(s); + s->img_y = stbi__get32be(s); + if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + z->depth = stbi__get8(s); if (z->depth != 1 && z->depth != 2 && z->depth != 4 && z->depth != 8 && z->depth != 16) return stbi__err("1/2/4/8/16-bit only","PNG not supported: 1/2/4/8/16-bit only"); + color = stbi__get8(s); if (color > 6) return stbi__err("bad ctype","Corrupt PNG"); + if (color == 3 && z->depth == 16) return stbi__err("bad ctype","Corrupt PNG"); + if (color == 3) pal_img_n = 3; else if (color & 1) return stbi__err("bad ctype","Corrupt PNG"); + comp = stbi__get8(s); if (comp) return stbi__err("bad comp method","Corrupt PNG"); + filter= stbi__get8(s); if (filter) return stbi__err("bad filter method","Corrupt PNG"); + interlace = stbi__get8(s); if (interlace>1) return stbi__err("bad interlace method","Corrupt PNG"); + if (!s->img_x || !s->img_y) return stbi__err("0-pixel image","Corrupt PNG"); + if (!pal_img_n) { + s->img_n = (color & 2 ? 3 : 1) + (color & 4 ? 1 : 0); + if ((1 << 30) / s->img_x / s->img_n < s->img_y) return stbi__err("too large", "Image too large to decode"); + if (scan == STBI__SCAN_header) return 1; + } else { + // if paletted, then pal_n is our final components, and + // img_n is # components to decompress/filter. + s->img_n = 1; + if ((1 << 30) / s->img_x / 4 < s->img_y) return stbi__err("too large","Corrupt PNG"); + // if SCAN_header, have to scan to see if we have a tRNS + } + break; + } + + case STBI__PNG_TYPE('P','L','T','E'): { + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if (c.length > 256*3) return stbi__err("invalid PLTE","Corrupt PNG"); + pal_len = c.length / 3; + if (pal_len * 3 != c.length) return stbi__err("invalid PLTE","Corrupt PNG"); + for (i=0; i < pal_len; ++i) { + palette[i*4+0] = stbi__get8(s); + palette[i*4+1] = stbi__get8(s); + palette[i*4+2] = stbi__get8(s); + palette[i*4+3] = 255; + } + break; + } + + case STBI__PNG_TYPE('t','R','N','S'): { + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if (z->idata) return stbi__err("tRNS after IDAT","Corrupt PNG"); + if (pal_img_n) { + if (scan == STBI__SCAN_header) { s->img_n = 4; return 1; } + if (pal_len == 0) return stbi__err("tRNS before PLTE","Corrupt PNG"); + if (c.length > pal_len) return stbi__err("bad tRNS len","Corrupt PNG"); + pal_img_n = 4; + for (i=0; i < c.length; ++i) + palette[i*4+3] = stbi__get8(s); + } else { + if (!(s->img_n & 1)) return stbi__err("tRNS with alpha","Corrupt PNG"); + if (c.length != (stbi__uint32) s->img_n*2) return stbi__err("bad tRNS len","Corrupt PNG"); + has_trans = 1; + if (z->depth == 16) { + for (k = 0; k < s->img_n; ++k) tc16[k] = (stbi__uint16)stbi__get16be(s); // copy the values as-is + } else { + for (k = 0; k < s->img_n; ++k) tc[k] = (stbi_uc)(stbi__get16be(s) & 255) * stbi__depth_scale_table[z->depth]; // non 8-bit images will be larger + } + } + break; + } + + case STBI__PNG_TYPE('I','D','A','T'): { + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if (pal_img_n && !pal_len) return stbi__err("no PLTE","Corrupt PNG"); + if (scan == STBI__SCAN_header) { s->img_n = pal_img_n; return 1; } + if ((int)(ioff + c.length) < (int)ioff) return 0; + if (ioff + c.length > idata_limit) { + stbi__uint32 idata_limit_old = idata_limit; + stbi_uc *p; + if (idata_limit == 0) idata_limit = c.length > 4096 ? c.length : 4096; + while (ioff + c.length > idata_limit) + idata_limit *= 2; + STBI_NOTUSED(idata_limit_old); + p = (stbi_uc *) STBI_REALLOC_SIZED(z->idata, idata_limit_old, idata_limit); if (p == NULL) return stbi__err("outofmem", "Out of memory"); + z->idata = p; + } + if (!stbi__getn(s, z->idata+ioff,c.length)) return stbi__err("outofdata","Corrupt PNG"); + ioff += c.length; + break; + } + + case STBI__PNG_TYPE('I','E','N','D'): { + stbi__uint32 raw_len, bpl; + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if (scan != STBI__SCAN_load) return 1; + if (z->idata == NULL) return stbi__err("no IDAT","Corrupt PNG"); + // initial guess for decoded data size to avoid unnecessary reallocs + bpl = (s->img_x * z->depth + 7) / 8; // bytes per line, per component + raw_len = bpl * s->img_y * s->img_n /* pixels */ + s->img_y /* filter mode per row */; + z->expanded = (stbi_uc *) stbi_zlib_decode_malloc_guesssize_headerflag((char *) z->idata, ioff, raw_len, (int *) &raw_len, !is_iphone); + if (z->expanded == NULL) return 0; // zlib should set error + STBI_FREE(z->idata); z->idata = NULL; + if ((req_comp == s->img_n+1 && req_comp != 3 && !pal_img_n) || has_trans) + s->img_out_n = s->img_n+1; + else + s->img_out_n = s->img_n; + if (!stbi__create_png_image(z, z->expanded, raw_len, s->img_out_n, z->depth, color, interlace)) return 0; + if (has_trans) { + if (z->depth == 16) { + if (!stbi__compute_transparency16(z, tc16, s->img_out_n)) return 0; + } else { + if (!stbi__compute_transparency(z, tc, s->img_out_n)) return 0; + } + } + if (is_iphone && stbi__de_iphone_flag && s->img_out_n > 2) + stbi__de_iphone(z); + if (pal_img_n) { + // pal_img_n == 3 or 4 + s->img_n = pal_img_n; // record the actual colors we had + s->img_out_n = pal_img_n; + if (req_comp >= 3) s->img_out_n = req_comp; + if (!stbi__expand_png_palette(z, palette, pal_len, s->img_out_n)) + return 0; + } else if (has_trans) { + // non-paletted image with tRNS -> source image has (constant) alpha + ++s->img_n; + } + STBI_FREE(z->expanded); z->expanded = NULL; + // end of PNG chunk, read and skip CRC + stbi__get32be(s); + return 1; + } + + default: + // if critical, fail + if (first) return stbi__err("first not IHDR", "Corrupt PNG"); + if ((c.type & (1 << 29)) == 0) { + #ifndef STBI_NO_FAILURE_STRINGS + // not threadsafe + static char invalid_chunk[] = "XXXX PNG chunk not known"; + invalid_chunk[0] = STBI__BYTECAST(c.type >> 24); + invalid_chunk[1] = STBI__BYTECAST(c.type >> 16); + invalid_chunk[2] = STBI__BYTECAST(c.type >> 8); + invalid_chunk[3] = STBI__BYTECAST(c.type >> 0); + #endif + return stbi__err(invalid_chunk, "PNG not supported: unknown PNG chunk type"); + } + stbi__skip(s, c.length); + break; + } + // end of PNG chunk, read and skip CRC + stbi__get32be(s); + } +} + +static void *stbi__do_png(stbi__png *p, int *x, int *y, int *n, int req_comp, stbi__result_info *ri) +{ + void *result=NULL; + if (req_comp < 0 || req_comp > 4) return stbi__errpuc("bad req_comp", "Internal error"); + if (stbi__parse_png_file(p, STBI__SCAN_load, req_comp)) { + if (p->depth <= 8) + ri->bits_per_channel = 8; + else if (p->depth == 16) + ri->bits_per_channel = 16; + else + return stbi__errpuc("bad bits_per_channel", "PNG not supported: unsupported color depth"); + result = p->out; + p->out = NULL; + if (req_comp && req_comp != p->s->img_out_n) { + if (ri->bits_per_channel == 8) + result = stbi__convert_format((unsigned char *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y); + else + result = stbi__convert_format16((stbi__uint16 *) result, p->s->img_out_n, req_comp, p->s->img_x, p->s->img_y); + p->s->img_out_n = req_comp; + if (result == NULL) return result; + } + *x = p->s->img_x; + *y = p->s->img_y; + if (n) *n = p->s->img_n; + } + STBI_FREE(p->out); p->out = NULL; + STBI_FREE(p->expanded); p->expanded = NULL; + STBI_FREE(p->idata); p->idata = NULL; + + return result; +} + +static void *stbi__png_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + stbi__png p; + p.s = s; + return stbi__do_png(&p, x,y,comp,req_comp, ri); +} + +static int stbi__png_test(stbi__context *s) +{ + int r; + r = stbi__check_png_header(s); + stbi__rewind(s); + return r; +} + +static int stbi__png_info_raw(stbi__png *p, int *x, int *y, int *comp) +{ + if (!stbi__parse_png_file(p, STBI__SCAN_header, 0)) { + stbi__rewind( p->s ); + return 0; + } + if (x) *x = p->s->img_x; + if (y) *y = p->s->img_y; + if (comp) *comp = p->s->img_n; + return 1; +} + +static int stbi__png_info(stbi__context *s, int *x, int *y, int *comp) +{ + stbi__png p; + p.s = s; + return stbi__png_info_raw(&p, x, y, comp); +} + +static int stbi__png_is16(stbi__context *s) +{ + stbi__png p; + p.s = s; + if (!stbi__png_info_raw(&p, NULL, NULL, NULL)) + return 0; + if (p.depth != 16) { + stbi__rewind(p.s); + return 0; + } + return 1; +} +#endif + +// Microsoft/Windows BMP image + +#ifndef STBI_NO_BMP +static int stbi__bmp_test_raw(stbi__context *s) +{ + int r; + int sz; + if (stbi__get8(s) != 'B') return 0; + if (stbi__get8(s) != 'M') return 0; + stbi__get32le(s); // discard filesize + stbi__get16le(s); // discard reserved + stbi__get16le(s); // discard reserved + stbi__get32le(s); // discard data offset + sz = stbi__get32le(s); + r = (sz == 12 || sz == 40 || sz == 56 || sz == 108 || sz == 124); + return r; +} + +static int stbi__bmp_test(stbi__context *s) +{ + int r = stbi__bmp_test_raw(s); + stbi__rewind(s); + return r; +} + + +// returns 0..31 for the highest set bit +static int stbi__high_bit(unsigned int z) +{ + int n=0; + if (z == 0) return -1; + if (z >= 0x10000) { n += 16; z >>= 16; } + if (z >= 0x00100) { n += 8; z >>= 8; } + if (z >= 0x00010) { n += 4; z >>= 4; } + if (z >= 0x00004) { n += 2; z >>= 2; } + if (z >= 0x00002) { n += 1;/* >>= 1;*/ } + return n; +} + +static int stbi__bitcount(unsigned int a) +{ + a = (a & 0x55555555) + ((a >> 1) & 0x55555555); // max 2 + a = (a & 0x33333333) + ((a >> 2) & 0x33333333); // max 4 + a = (a + (a >> 4)) & 0x0f0f0f0f; // max 8 per 4, now 8 bits + a = (a + (a >> 8)); // max 16 per 8 bits + a = (a + (a >> 16)); // max 32 per 8 bits + return a & 0xff; +} + +// extract an arbitrarily-aligned N-bit value (N=bits) +// from v, and then make it 8-bits long and fractionally +// extend it to full full range. +static int stbi__shiftsigned(unsigned int v, int shift, int bits) +{ + static unsigned int mul_table[9] = { + 0, + 0xff/*0b11111111*/, 0x55/*0b01010101*/, 0x49/*0b01001001*/, 0x11/*0b00010001*/, + 0x21/*0b00100001*/, 0x41/*0b01000001*/, 0x81/*0b10000001*/, 0x01/*0b00000001*/, + }; + static unsigned int shift_table[9] = { + 0, 0,0,1,0,2,4,6,0, + }; + if (shift < 0) + v <<= -shift; + else + v >>= shift; + STBI_ASSERT(v < 256); + v >>= (8-bits); + STBI_ASSERT(bits >= 0 && bits <= 8); + return (int) ((unsigned) v * mul_table[bits]) >> shift_table[bits]; +} + +typedef struct +{ + int bpp, offset, hsz; + unsigned int mr,mg,mb,ma, all_a; + int extra_read; +} stbi__bmp_data; + +static void *stbi__bmp_parse_header(stbi__context *s, stbi__bmp_data *info) +{ + int hsz; + if (stbi__get8(s) != 'B' || stbi__get8(s) != 'M') return stbi__errpuc("not BMP", "Corrupt BMP"); + stbi__get32le(s); // discard filesize + stbi__get16le(s); // discard reserved + stbi__get16le(s); // discard reserved + info->offset = stbi__get32le(s); + info->hsz = hsz = stbi__get32le(s); + info->mr = info->mg = info->mb = info->ma = 0; + info->extra_read = 14; + + if (info->offset < 0) return stbi__errpuc("bad BMP", "bad BMP"); + + if (hsz != 12 && hsz != 40 && hsz != 56 && hsz != 108 && hsz != 124) return stbi__errpuc("unknown BMP", "BMP type not supported: unknown"); + if (hsz == 12) { + s->img_x = stbi__get16le(s); + s->img_y = stbi__get16le(s); + } else { + s->img_x = stbi__get32le(s); + s->img_y = stbi__get32le(s); + } + if (stbi__get16le(s) != 1) return stbi__errpuc("bad BMP", "bad BMP"); + info->bpp = stbi__get16le(s); + if (hsz != 12) { + int compress = stbi__get32le(s); + if (compress == 1 || compress == 2) return stbi__errpuc("BMP RLE", "BMP type not supported: RLE"); + stbi__get32le(s); // discard sizeof + stbi__get32le(s); // discard hres + stbi__get32le(s); // discard vres + stbi__get32le(s); // discard colorsused + stbi__get32le(s); // discard max important + if (hsz == 40 || hsz == 56) { + if (hsz == 56) { + stbi__get32le(s); + stbi__get32le(s); + stbi__get32le(s); + stbi__get32le(s); + } + if (info->bpp == 16 || info->bpp == 32) { + if (compress == 0) { + if (info->bpp == 32) { + info->mr = 0xffu << 16; + info->mg = 0xffu << 8; + info->mb = 0xffu << 0; + info->ma = 0xffu << 24; + info->all_a = 0; // if all_a is 0 at end, then we loaded alpha channel but it was all 0 + } else { + info->mr = 31u << 10; + info->mg = 31u << 5; + info->mb = 31u << 0; + } + } else if (compress == 3) { + info->mr = stbi__get32le(s); + info->mg = stbi__get32le(s); + info->mb = stbi__get32le(s); + info->extra_read += 12; + // not documented, but generated by photoshop and handled by mspaint + if (info->mr == info->mg && info->mg == info->mb) { + // ?!?!? + return stbi__errpuc("bad BMP", "bad BMP"); + } + } else + return stbi__errpuc("bad BMP", "bad BMP"); + } + } else { + int i; + if (hsz != 108 && hsz != 124) + return stbi__errpuc("bad BMP", "bad BMP"); + info->mr = stbi__get32le(s); + info->mg = stbi__get32le(s); + info->mb = stbi__get32le(s); + info->ma = stbi__get32le(s); + stbi__get32le(s); // discard color space + for (i=0; i < 12; ++i) + stbi__get32le(s); // discard color space parameters + if (hsz == 124) { + stbi__get32le(s); // discard rendering intent + stbi__get32le(s); // discard offset of profile data + stbi__get32le(s); // discard size of profile data + stbi__get32le(s); // discard reserved + } + } + } + return (void *) 1; +} + + +static void *stbi__bmp_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + stbi_uc *out; + unsigned int mr=0,mg=0,mb=0,ma=0, all_a; + stbi_uc pal[256][4]; + int psize=0,i,j,width; + int flip_vertically, pad, target; + stbi__bmp_data info; + STBI_NOTUSED(ri); + + info.all_a = 255; + if (stbi__bmp_parse_header(s, &info) == NULL) + return NULL; // error code already set + + flip_vertically = ((int) s->img_y) > 0; + s->img_y = abs((int) s->img_y); + + if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + mr = info.mr; + mg = info.mg; + mb = info.mb; + ma = info.ma; + all_a = info.all_a; + + if (info.hsz == 12) { + if (info.bpp < 24) + psize = (info.offset - info.extra_read - 24) / 3; + } else { + if (info.bpp < 16) + psize = (info.offset - info.extra_read - info.hsz) >> 2; + } + if (psize == 0) { + STBI_ASSERT(info.offset == s->callback_already_read + (int) (s->img_buffer - s->img_buffer_original)); + if (info.offset != s->callback_already_read + (s->img_buffer - s->buffer_start)) { + return stbi__errpuc("bad offset", "Corrupt BMP"); + } + } + + if (info.bpp == 24 && ma == 0xff000000) + s->img_n = 3; + else + s->img_n = ma ? 4 : 3; + if (req_comp && req_comp >= 3) // we can directly decode 3 or 4 + target = req_comp; + else + target = s->img_n; // if they want monochrome, we'll post-convert + + // sanity-check size + if (!stbi__mad3sizes_valid(target, s->img_x, s->img_y, 0)) + return stbi__errpuc("too large", "Corrupt BMP"); + + out = (stbi_uc *) stbi__malloc_mad3(target, s->img_x, s->img_y, 0); + if (!out) return stbi__errpuc("outofmem", "Out of memory"); + if (info.bpp < 16) { + int z=0; + if (psize == 0 || psize > 256) { STBI_FREE(out); return stbi__errpuc("invalid", "Corrupt BMP"); } + for (i=0; i < psize; ++i) { + pal[i][2] = stbi__get8(s); + pal[i][1] = stbi__get8(s); + pal[i][0] = stbi__get8(s); + if (info.hsz != 12) stbi__get8(s); + pal[i][3] = 255; + } + stbi__skip(s, info.offset - info.extra_read - info.hsz - psize * (info.hsz == 12 ? 3 : 4)); + if (info.bpp == 1) width = (s->img_x + 7) >> 3; + else if (info.bpp == 4) width = (s->img_x + 1) >> 1; + else if (info.bpp == 8) width = s->img_x; + else { STBI_FREE(out); return stbi__errpuc("bad bpp", "Corrupt BMP"); } + pad = (-width)&3; + if (info.bpp == 1) { + for (j=0; j < (int) s->img_y; ++j) { + int bit_offset = 7, v = stbi__get8(s); + for (i=0; i < (int) s->img_x; ++i) { + int color = (v>>bit_offset)&0x1; + out[z++] = pal[color][0]; + out[z++] = pal[color][1]; + out[z++] = pal[color][2]; + if (target == 4) out[z++] = 255; + if (i+1 == (int) s->img_x) break; + if((--bit_offset) < 0) { + bit_offset = 7; + v = stbi__get8(s); + } + } + stbi__skip(s, pad); + } + } else { + for (j=0; j < (int) s->img_y; ++j) { + for (i=0; i < (int) s->img_x; i += 2) { + int v=stbi__get8(s),v2=0; + if (info.bpp == 4) { + v2 = v & 15; + v >>= 4; + } + out[z++] = pal[v][0]; + out[z++] = pal[v][1]; + out[z++] = pal[v][2]; + if (target == 4) out[z++] = 255; + if (i+1 == (int) s->img_x) break; + v = (info.bpp == 8) ? stbi__get8(s) : v2; + out[z++] = pal[v][0]; + out[z++] = pal[v][1]; + out[z++] = pal[v][2]; + if (target == 4) out[z++] = 255; + } + stbi__skip(s, pad); + } + } + } else { + int rshift=0,gshift=0,bshift=0,ashift=0,rcount=0,gcount=0,bcount=0,acount=0; + int z = 0; + int easy=0; + stbi__skip(s, info.offset - info.extra_read - info.hsz); + if (info.bpp == 24) width = 3 * s->img_x; + else if (info.bpp == 16) width = 2*s->img_x; + else /* bpp = 32 and pad = 0 */ width=0; + pad = (-width) & 3; + if (info.bpp == 24) { + easy = 1; + } else if (info.bpp == 32) { + if (mb == 0xff && mg == 0xff00 && mr == 0x00ff0000 && ma == 0xff000000) + easy = 2; + } + if (!easy) { + if (!mr || !mg || !mb) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); } + // right shift amt to put high bit in position #7 + rshift = stbi__high_bit(mr)-7; rcount = stbi__bitcount(mr); + gshift = stbi__high_bit(mg)-7; gcount = stbi__bitcount(mg); + bshift = stbi__high_bit(mb)-7; bcount = stbi__bitcount(mb); + ashift = stbi__high_bit(ma)-7; acount = stbi__bitcount(ma); + if (rcount > 8 || gcount > 8 || bcount > 8 || acount > 8) { STBI_FREE(out); return stbi__errpuc("bad masks", "Corrupt BMP"); } + } + for (j=0; j < (int) s->img_y; ++j) { + if (easy) { + for (i=0; i < (int) s->img_x; ++i) { + unsigned char a; + out[z+2] = stbi__get8(s); + out[z+1] = stbi__get8(s); + out[z+0] = stbi__get8(s); + z += 3; + a = (easy == 2 ? stbi__get8(s) : 255); + all_a |= a; + if (target == 4) out[z++] = a; + } + } else { + int bpp = info.bpp; + for (i=0; i < (int) s->img_x; ++i) { + stbi__uint32 v = (bpp == 16 ? (stbi__uint32) stbi__get16le(s) : stbi__get32le(s)); + unsigned int a; + out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mr, rshift, rcount)); + out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mg, gshift, gcount)); + out[z++] = STBI__BYTECAST(stbi__shiftsigned(v & mb, bshift, bcount)); + a = (ma ? stbi__shiftsigned(v & ma, ashift, acount) : 255); + all_a |= a; + if (target == 4) out[z++] = STBI__BYTECAST(a); + } + } + stbi__skip(s, pad); + } + } + + // if alpha channel is all 0s, replace with all 255s + if (target == 4 && all_a == 0) + for (i=4*s->img_x*s->img_y-1; i >= 0; i -= 4) + out[i] = 255; + + if (flip_vertically) { + stbi_uc t; + for (j=0; j < (int) s->img_y>>1; ++j) { + stbi_uc *p1 = out + j *s->img_x*target; + stbi_uc *p2 = out + (s->img_y-1-j)*s->img_x*target; + for (i=0; i < (int) s->img_x*target; ++i) { + t = p1[i]; p1[i] = p2[i]; p2[i] = t; + } + } + } + + if (req_comp && req_comp != target) { + out = stbi__convert_format(out, target, req_comp, s->img_x, s->img_y); + if (out == NULL) return out; // stbi__convert_format frees input on failure + } + + *x = s->img_x; + *y = s->img_y; + if (comp) *comp = s->img_n; + return out; +} +#endif + +// Targa Truevision - TGA +// by Jonathan Dummer +#ifndef STBI_NO_TGA +// returns STBI_rgb or whatever, 0 on error +static int stbi__tga_get_comp(int bits_per_pixel, int is_grey, int* is_rgb16) +{ + // only RGB or RGBA (incl. 16bit) or grey allowed + if (is_rgb16) *is_rgb16 = 0; + switch(bits_per_pixel) { + case 8: return STBI_grey; + case 16: if(is_grey) return STBI_grey_alpha; + // fallthrough + case 15: if(is_rgb16) *is_rgb16 = 1; + return STBI_rgb; + case 24: // fallthrough + case 32: return bits_per_pixel/8; + default: return 0; + } +} + +static int stbi__tga_info(stbi__context *s, int *x, int *y, int *comp) +{ + int tga_w, tga_h, tga_comp, tga_image_type, tga_bits_per_pixel, tga_colormap_bpp; + int sz, tga_colormap_type; + stbi__get8(s); // discard Offset + tga_colormap_type = stbi__get8(s); // colormap type + if( tga_colormap_type > 1 ) { + stbi__rewind(s); + return 0; // only RGB or indexed allowed + } + tga_image_type = stbi__get8(s); // image type + if ( tga_colormap_type == 1 ) { // colormapped (paletted) image + if (tga_image_type != 1 && tga_image_type != 9) { + stbi__rewind(s); + return 0; + } + stbi__skip(s,4); // skip index of first colormap entry and number of entries + sz = stbi__get8(s); // check bits per palette color entry + if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) { + stbi__rewind(s); + return 0; + } + stbi__skip(s,4); // skip image x and y origin + tga_colormap_bpp = sz; + } else { // "normal" image w/o colormap - only RGB or grey allowed, +/- RLE + if ( (tga_image_type != 2) && (tga_image_type != 3) && (tga_image_type != 10) && (tga_image_type != 11) ) { + stbi__rewind(s); + return 0; // only RGB or grey allowed, +/- RLE + } + stbi__skip(s,9); // skip colormap specification and image x/y origin + tga_colormap_bpp = 0; + } + tga_w = stbi__get16le(s); + if( tga_w < 1 ) { + stbi__rewind(s); + return 0; // test width + } + tga_h = stbi__get16le(s); + if( tga_h < 1 ) { + stbi__rewind(s); + return 0; // test height + } + tga_bits_per_pixel = stbi__get8(s); // bits per pixel + stbi__get8(s); // ignore alpha bits + if (tga_colormap_bpp != 0) { + if((tga_bits_per_pixel != 8) && (tga_bits_per_pixel != 16)) { + // when using a colormap, tga_bits_per_pixel is the size of the indexes + // I don't think anything but 8 or 16bit indexes makes sense + stbi__rewind(s); + return 0; + } + tga_comp = stbi__tga_get_comp(tga_colormap_bpp, 0, NULL); + } else { + tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3) || (tga_image_type == 11), NULL); + } + if(!tga_comp) { + stbi__rewind(s); + return 0; + } + if (x) *x = tga_w; + if (y) *y = tga_h; + if (comp) *comp = tga_comp; + return 1; // seems to have passed everything +} + +static int stbi__tga_test(stbi__context *s) +{ + int res = 0; + int sz, tga_color_type; + stbi__get8(s); // discard Offset + tga_color_type = stbi__get8(s); // color type + if ( tga_color_type > 1 ) goto errorEnd; // only RGB or indexed allowed + sz = stbi__get8(s); // image type + if ( tga_color_type == 1 ) { // colormapped (paletted) image + if (sz != 1 && sz != 9) goto errorEnd; // colortype 1 demands image type 1 or 9 + stbi__skip(s,4); // skip index of first colormap entry and number of entries + sz = stbi__get8(s); // check bits per palette color entry + if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd; + stbi__skip(s,4); // skip image x and y origin + } else { // "normal" image w/o colormap + if ( (sz != 2) && (sz != 3) && (sz != 10) && (sz != 11) ) goto errorEnd; // only RGB or grey allowed, +/- RLE + stbi__skip(s,9); // skip colormap specification and image x/y origin + } + if ( stbi__get16le(s) < 1 ) goto errorEnd; // test width + if ( stbi__get16le(s) < 1 ) goto errorEnd; // test height + sz = stbi__get8(s); // bits per pixel + if ( (tga_color_type == 1) && (sz != 8) && (sz != 16) ) goto errorEnd; // for colormapped images, bpp is size of an index + if ( (sz != 8) && (sz != 15) && (sz != 16) && (sz != 24) && (sz != 32) ) goto errorEnd; + + res = 1; // if we got this far, everything's good and we can return 1 instead of 0 + +errorEnd: + stbi__rewind(s); + return res; +} + +// read 16bit value and convert to 24bit RGB +static void stbi__tga_read_rgb16(stbi__context *s, stbi_uc* out) +{ + stbi__uint16 px = (stbi__uint16)stbi__get16le(s); + stbi__uint16 fiveBitMask = 31; + // we have 3 channels with 5bits each + int r = (px >> 10) & fiveBitMask; + int g = (px >> 5) & fiveBitMask; + int b = px & fiveBitMask; + // Note that this saves the data in RGB(A) order, so it doesn't need to be swapped later + out[0] = (stbi_uc)((r * 255)/31); + out[1] = (stbi_uc)((g * 255)/31); + out[2] = (stbi_uc)((b * 255)/31); + + // some people claim that the most significant bit might be used for alpha + // (possibly if an alpha-bit is set in the "image descriptor byte") + // but that only made 16bit test images completely translucent.. + // so let's treat all 15 and 16bit TGAs as RGB with no alpha. +} + +static void *stbi__tga_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + // read in the TGA header stuff + int tga_offset = stbi__get8(s); + int tga_indexed = stbi__get8(s); + int tga_image_type = stbi__get8(s); + int tga_is_RLE = 0; + int tga_palette_start = stbi__get16le(s); + int tga_palette_len = stbi__get16le(s); + int tga_palette_bits = stbi__get8(s); + int tga_x_origin = stbi__get16le(s); + int tga_y_origin = stbi__get16le(s); + int tga_width = stbi__get16le(s); + int tga_height = stbi__get16le(s); + int tga_bits_per_pixel = stbi__get8(s); + int tga_comp, tga_rgb16=0; + int tga_inverted = stbi__get8(s); + // int tga_alpha_bits = tga_inverted & 15; // the 4 lowest bits - unused (useless?) + // image data + unsigned char *tga_data; + unsigned char *tga_palette = NULL; + int i, j; + unsigned char raw_data[4] = {0}; + int RLE_count = 0; + int RLE_repeating = 0; + int read_next_pixel = 1; + STBI_NOTUSED(ri); + STBI_NOTUSED(tga_x_origin); // @TODO + STBI_NOTUSED(tga_y_origin); // @TODO + + if (tga_height > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (tga_width > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + // do a tiny bit of precessing + if ( tga_image_type >= 8 ) + { + tga_image_type -= 8; + tga_is_RLE = 1; + } + tga_inverted = 1 - ((tga_inverted >> 5) & 1); + + // If I'm paletted, then I'll use the number of bits from the palette + if ( tga_indexed ) tga_comp = stbi__tga_get_comp(tga_palette_bits, 0, &tga_rgb16); + else tga_comp = stbi__tga_get_comp(tga_bits_per_pixel, (tga_image_type == 3), &tga_rgb16); + + if(!tga_comp) // shouldn't really happen, stbi__tga_test() should have ensured basic consistency + return stbi__errpuc("bad format", "Can't find out TGA pixelformat"); + + // tga info + *x = tga_width; + *y = tga_height; + if (comp) *comp = tga_comp; + + if (!stbi__mad3sizes_valid(tga_width, tga_height, tga_comp, 0)) + return stbi__errpuc("too large", "Corrupt TGA"); + + tga_data = (unsigned char*)stbi__malloc_mad3(tga_width, tga_height, tga_comp, 0); + if (!tga_data) return stbi__errpuc("outofmem", "Out of memory"); + + // skip to the data's starting position (offset usually = 0) + stbi__skip(s, tga_offset ); + + if ( !tga_indexed && !tga_is_RLE && !tga_rgb16 ) { + for (i=0; i < tga_height; ++i) { + int row = tga_inverted ? tga_height -i - 1 : i; + stbi_uc *tga_row = tga_data + row*tga_width*tga_comp; + stbi__getn(s, tga_row, tga_width * tga_comp); + } + } else { + // do I need to load a palette? + if ( tga_indexed) + { + if (tga_palette_len == 0) { /* you have to have at least one entry! */ + STBI_FREE(tga_data); + return stbi__errpuc("bad palette", "Corrupt TGA"); + } + + // any data to skip? (offset usually = 0) + stbi__skip(s, tga_palette_start ); + // load the palette + tga_palette = (unsigned char*)stbi__malloc_mad2(tga_palette_len, tga_comp, 0); + if (!tga_palette) { + STBI_FREE(tga_data); + return stbi__errpuc("outofmem", "Out of memory"); + } + if (tga_rgb16) { + stbi_uc *pal_entry = tga_palette; + STBI_ASSERT(tga_comp == STBI_rgb); + for (i=0; i < tga_palette_len; ++i) { + stbi__tga_read_rgb16(s, pal_entry); + pal_entry += tga_comp; + } + } else if (!stbi__getn(s, tga_palette, tga_palette_len * tga_comp)) { + STBI_FREE(tga_data); + STBI_FREE(tga_palette); + return stbi__errpuc("bad palette", "Corrupt TGA"); + } + } + // load the data + for (i=0; i < tga_width * tga_height; ++i) + { + // if I'm in RLE mode, do I need to get a RLE stbi__pngchunk? + if ( tga_is_RLE ) + { + if ( RLE_count == 0 ) + { + // yep, get the next byte as a RLE command + int RLE_cmd = stbi__get8(s); + RLE_count = 1 + (RLE_cmd & 127); + RLE_repeating = RLE_cmd >> 7; + read_next_pixel = 1; + } else if ( !RLE_repeating ) + { + read_next_pixel = 1; + } + } else + { + read_next_pixel = 1; + } + // OK, if I need to read a pixel, do it now + if ( read_next_pixel ) + { + // load however much data we did have + if ( tga_indexed ) + { + // read in index, then perform the lookup + int pal_idx = (tga_bits_per_pixel == 8) ? stbi__get8(s) : stbi__get16le(s); + if ( pal_idx >= tga_palette_len ) { + // invalid index + pal_idx = 0; + } + pal_idx *= tga_comp; + for (j = 0; j < tga_comp; ++j) { + raw_data[j] = tga_palette[pal_idx+j]; + } + } else if(tga_rgb16) { + STBI_ASSERT(tga_comp == STBI_rgb); + stbi__tga_read_rgb16(s, raw_data); + } else { + // read in the data raw + for (j = 0; j < tga_comp; ++j) { + raw_data[j] = stbi__get8(s); + } + } + // clear the reading flag for the next pixel + read_next_pixel = 0; + } // end of reading a pixel + + // copy data + for (j = 0; j < tga_comp; ++j) + tga_data[i*tga_comp+j] = raw_data[j]; + + // in case we're in RLE mode, keep counting down + --RLE_count; + } + // do I need to invert the image? + if ( tga_inverted ) + { + for (j = 0; j*2 < tga_height; ++j) + { + int index1 = j * tga_width * tga_comp; + int index2 = (tga_height - 1 - j) * tga_width * tga_comp; + for (i = tga_width * tga_comp; i > 0; --i) + { + unsigned char temp = tga_data[index1]; + tga_data[index1] = tga_data[index2]; + tga_data[index2] = temp; + ++index1; + ++index2; + } + } + } + // clear my palette, if I had one + if ( tga_palette != NULL ) + { + STBI_FREE( tga_palette ); + } + } + + // swap RGB - if the source data was RGB16, it already is in the right order + if (tga_comp >= 3 && !tga_rgb16) + { + unsigned char* tga_pixel = tga_data; + for (i=0; i < tga_width * tga_height; ++i) + { + unsigned char temp = tga_pixel[0]; + tga_pixel[0] = tga_pixel[2]; + tga_pixel[2] = temp; + tga_pixel += tga_comp; + } + } + + // convert to target component count + if (req_comp && req_comp != tga_comp) + tga_data = stbi__convert_format(tga_data, tga_comp, req_comp, tga_width, tga_height); + + // the things I do to get rid of an error message, and yet keep + // Microsoft's C compilers happy... [8^( + tga_palette_start = tga_palette_len = tga_palette_bits = + tga_x_origin = tga_y_origin = 0; + STBI_NOTUSED(tga_palette_start); + // OK, done + return tga_data; +} +#endif + +// ************************************************************************************************* +// Photoshop PSD loader -- PD by Thatcher Ulrich, integration by Nicolas Schulz, tweaked by STB + +#ifndef STBI_NO_PSD +static int stbi__psd_test(stbi__context *s) +{ + int r = (stbi__get32be(s) == 0x38425053); + stbi__rewind(s); + return r; +} + +static int stbi__psd_decode_rle(stbi__context *s, stbi_uc *p, int pixelCount) +{ + int count, nleft, len; + + count = 0; + while ((nleft = pixelCount - count) > 0) { + len = stbi__get8(s); + if (len == 128) { + // No-op. + } else if (len < 128) { + // Copy next len+1 bytes literally. + len++; + if (len > nleft) return 0; // corrupt data + count += len; + while (len) { + *p = stbi__get8(s); + p += 4; + len--; + } + } else if (len > 128) { + stbi_uc val; + // Next -len+1 bytes in the dest are replicated from next source byte. + // (Interpret len as a negative 8-bit int.) + len = 257 - len; + if (len > nleft) return 0; // corrupt data + val = stbi__get8(s); + count += len; + while (len) { + *p = val; + p += 4; + len--; + } + } + } + + return 1; +} + +static void *stbi__psd_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri, int bpc) +{ + int pixelCount; + int channelCount, compression; + int channel, i; + int bitdepth; + int w,h; + stbi_uc *out; + STBI_NOTUSED(ri); + + // Check identifier + if (stbi__get32be(s) != 0x38425053) // "8BPS" + return stbi__errpuc("not PSD", "Corrupt PSD image"); + + // Check file type version. + if (stbi__get16be(s) != 1) + return stbi__errpuc("wrong version", "Unsupported version of PSD image"); + + // Skip 6 reserved bytes. + stbi__skip(s, 6 ); + + // Read the number of channels (R, G, B, A, etc). + channelCount = stbi__get16be(s); + if (channelCount < 0 || channelCount > 16) + return stbi__errpuc("wrong channel count", "Unsupported number of channels in PSD image"); + + // Read the rows and columns of the image. + h = stbi__get32be(s); + w = stbi__get32be(s); + + if (h > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (w > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + // Make sure the depth is 8 bits. + bitdepth = stbi__get16be(s); + if (bitdepth != 8 && bitdepth != 16) + return stbi__errpuc("unsupported bit depth", "PSD bit depth is not 8 or 16 bit"); + + // Make sure the color mode is RGB. + // Valid options are: + // 0: Bitmap + // 1: Grayscale + // 2: Indexed color + // 3: RGB color + // 4: CMYK color + // 7: Multichannel + // 8: Duotone + // 9: Lab color + if (stbi__get16be(s) != 3) + return stbi__errpuc("wrong color format", "PSD is not in RGB color format"); + + // Skip the Mode Data. (It's the palette for indexed color; other info for other modes.) + stbi__skip(s,stbi__get32be(s) ); + + // Skip the image resources. (resolution, pen tool paths, etc) + stbi__skip(s, stbi__get32be(s) ); + + // Skip the reserved data. + stbi__skip(s, stbi__get32be(s) ); + + // Find out if the data is compressed. + // Known values: + // 0: no compression + // 1: RLE compressed + compression = stbi__get16be(s); + if (compression > 1) + return stbi__errpuc("bad compression", "PSD has an unknown compression format"); + + // Check size + if (!stbi__mad3sizes_valid(4, w, h, 0)) + return stbi__errpuc("too large", "Corrupt PSD"); + + // Create the destination image. + + if (!compression && bitdepth == 16 && bpc == 16) { + out = (stbi_uc *) stbi__malloc_mad3(8, w, h, 0); + ri->bits_per_channel = 16; + } else + out = (stbi_uc *) stbi__malloc(4 * w*h); + + if (!out) return stbi__errpuc("outofmem", "Out of memory"); + pixelCount = w*h; + + // Initialize the data to zero. + //memset( out, 0, pixelCount * 4 ); + + // Finally, the image data. + if (compression) { + // RLE as used by .PSD and .TIFF + // Loop until you get the number of unpacked bytes you are expecting: + // Read the next source byte into n. + // If n is between 0 and 127 inclusive, copy the next n+1 bytes literally. + // Else if n is between -127 and -1 inclusive, copy the next byte -n+1 times. + // Else if n is 128, noop. + // Endloop + + // The RLE-compressed data is preceded by a 2-byte data count for each row in the data, + // which we're going to just skip. + stbi__skip(s, h * channelCount * 2 ); + + // Read the RLE data by channel. + for (channel = 0; channel < 4; channel++) { + stbi_uc *p; + + p = out+channel; + if (channel >= channelCount) { + // Fill this channel with default data. + for (i = 0; i < pixelCount; i++, p += 4) + *p = (channel == 3 ? 255 : 0); + } else { + // Read the RLE data. + if (!stbi__psd_decode_rle(s, p, pixelCount)) { + STBI_FREE(out); + return stbi__errpuc("corrupt", "bad RLE data"); + } + } + } + + } else { + // We're at the raw image data. It's each channel in order (Red, Green, Blue, Alpha, ...) + // where each channel consists of an 8-bit (or 16-bit) value for each pixel in the image. + + // Read the data by channel. + for (channel = 0; channel < 4; channel++) { + if (channel >= channelCount) { + // Fill this channel with default data. + if (bitdepth == 16 && bpc == 16) { + stbi__uint16 *q = ((stbi__uint16 *) out) + channel; + stbi__uint16 val = channel == 3 ? 65535 : 0; + for (i = 0; i < pixelCount; i++, q += 4) + *q = val; + } else { + stbi_uc *p = out+channel; + stbi_uc val = channel == 3 ? 255 : 0; + for (i = 0; i < pixelCount; i++, p += 4) + *p = val; + } + } else { + if (ri->bits_per_channel == 16) { // output bpc + stbi__uint16 *q = ((stbi__uint16 *) out) + channel; + for (i = 0; i < pixelCount; i++, q += 4) + *q = (stbi__uint16) stbi__get16be(s); + } else { + stbi_uc *p = out+channel; + if (bitdepth == 16) { // input bpc + for (i = 0; i < pixelCount; i++, p += 4) + *p = (stbi_uc) (stbi__get16be(s) >> 8); + } else { + for (i = 0; i < pixelCount; i++, p += 4) + *p = stbi__get8(s); + } + } + } + } + } + + // remove weird white matte from PSD + if (channelCount >= 4) { + if (ri->bits_per_channel == 16) { + for (i=0; i < w*h; ++i) { + stbi__uint16 *pixel = (stbi__uint16 *) out + 4*i; + if (pixel[3] != 0 && pixel[3] != 65535) { + float a = pixel[3] / 65535.0f; + float ra = 1.0f / a; + float inv_a = 65535.0f * (1 - ra); + pixel[0] = (stbi__uint16) (pixel[0]*ra + inv_a); + pixel[1] = (stbi__uint16) (pixel[1]*ra + inv_a); + pixel[2] = (stbi__uint16) (pixel[2]*ra + inv_a); + } + } + } else { + for (i=0; i < w*h; ++i) { + unsigned char *pixel = out + 4*i; + if (pixel[3] != 0 && pixel[3] != 255) { + float a = pixel[3] / 255.0f; + float ra = 1.0f / a; + float inv_a = 255.0f * (1 - ra); + pixel[0] = (unsigned char) (pixel[0]*ra + inv_a); + pixel[1] = (unsigned char) (pixel[1]*ra + inv_a); + pixel[2] = (unsigned char) (pixel[2]*ra + inv_a); + } + } + } + } + + // convert to desired output format + if (req_comp && req_comp != 4) { + if (ri->bits_per_channel == 16) + out = (stbi_uc *) stbi__convert_format16((stbi__uint16 *) out, 4, req_comp, w, h); + else + out = stbi__convert_format(out, 4, req_comp, w, h); + if (out == NULL) return out; // stbi__convert_format frees input on failure + } + + if (comp) *comp = 4; + *y = h; + *x = w; + + return out; +} +#endif + +// ************************************************************************************************* +// Softimage PIC loader +// by Tom Seddon +// +// See http://softimage.wiki.softimage.com/index.php/INFO:_PIC_file_format +// See http://ozviz.wasp.uwa.edu.au/~pbourke/dataformats/softimagepic/ + +#ifndef STBI_NO_PIC +static int stbi__pic_is4(stbi__context *s,const char *str) +{ + int i; + for (i=0; i<4; ++i) + if (stbi__get8(s) != (stbi_uc)str[i]) + return 0; + + return 1; +} + +static int stbi__pic_test_core(stbi__context *s) +{ + int i; + + if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) + return 0; + + for(i=0;i<84;++i) + stbi__get8(s); + + if (!stbi__pic_is4(s,"PICT")) + return 0; + + return 1; +} + +typedef struct +{ + stbi_uc size,type,channel; +} stbi__pic_packet; + +static stbi_uc *stbi__readval(stbi__context *s, int channel, stbi_uc *dest) +{ + int mask=0x80, i; + + for (i=0; i<4; ++i, mask>>=1) { + if (channel & mask) { + if (stbi__at_eof(s)) return stbi__errpuc("bad file","PIC file too short"); + dest[i]=stbi__get8(s); + } + } + + return dest; +} + +static void stbi__copyval(int channel,stbi_uc *dest,const stbi_uc *src) +{ + int mask=0x80,i; + + for (i=0;i<4; ++i, mask>>=1) + if (channel&mask) + dest[i]=src[i]; +} + +static stbi_uc *stbi__pic_load_core(stbi__context *s,int width,int height,int *comp, stbi_uc *result) +{ + int act_comp=0,num_packets=0,y,chained; + stbi__pic_packet packets[10]; + + // this will (should...) cater for even some bizarre stuff like having data + // for the same channel in multiple packets. + do { + stbi__pic_packet *packet; + + if (num_packets==sizeof(packets)/sizeof(packets[0])) + return stbi__errpuc("bad format","too many packets"); + + packet = &packets[num_packets++]; + + chained = stbi__get8(s); + packet->size = stbi__get8(s); + packet->type = stbi__get8(s); + packet->channel = stbi__get8(s); + + act_comp |= packet->channel; + + if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (reading packets)"); + if (packet->size != 8) return stbi__errpuc("bad format","packet isn't 8bpp"); + } while (chained); + + *comp = (act_comp & 0x10 ? 4 : 3); // has alpha channel? + + for(y=0; ytype) { + default: + return stbi__errpuc("bad format","packet has bad compression type"); + + case 0: {//uncompressed + int x; + + for(x=0;xchannel,dest)) + return 0; + break; + } + + case 1://Pure RLE + { + int left=width, i; + + while (left>0) { + stbi_uc count,value[4]; + + count=stbi__get8(s); + if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pure read count)"); + + if (count > left) + count = (stbi_uc) left; + + if (!stbi__readval(s,packet->channel,value)) return 0; + + for(i=0; ichannel,dest,value); + left -= count; + } + } + break; + + case 2: {//Mixed RLE + int left=width; + while (left>0) { + int count = stbi__get8(s), i; + if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (mixed read count)"); + + if (count >= 128) { // Repeated + stbi_uc value[4]; + + if (count==128) + count = stbi__get16be(s); + else + count -= 127; + if (count > left) + return stbi__errpuc("bad file","scanline overrun"); + + if (!stbi__readval(s,packet->channel,value)) + return 0; + + for(i=0;ichannel,dest,value); + } else { // Raw + ++count; + if (count>left) return stbi__errpuc("bad file","scanline overrun"); + + for(i=0;ichannel,dest)) + return 0; + } + left-=count; + } + break; + } + } + } + } + + return result; +} + +static void *stbi__pic_load(stbi__context *s,int *px,int *py,int *comp,int req_comp, stbi__result_info *ri) +{ + stbi_uc *result; + int i, x,y, internal_comp; + STBI_NOTUSED(ri); + + if (!comp) comp = &internal_comp; + + for (i=0; i<92; ++i) + stbi__get8(s); + + x = stbi__get16be(s); + y = stbi__get16be(s); + + if (y > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (x > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + if (stbi__at_eof(s)) return stbi__errpuc("bad file","file too short (pic header)"); + if (!stbi__mad3sizes_valid(x, y, 4, 0)) return stbi__errpuc("too large", "PIC image too large to decode"); + + stbi__get32be(s); //skip `ratio' + stbi__get16be(s); //skip `fields' + stbi__get16be(s); //skip `pad' + + // intermediate buffer is RGBA + result = (stbi_uc *) stbi__malloc_mad3(x, y, 4, 0); + memset(result, 0xff, x*y*4); + + if (!stbi__pic_load_core(s,x,y,comp, result)) { + STBI_FREE(result); + result=0; + } + *px = x; + *py = y; + if (req_comp == 0) req_comp = *comp; + result=stbi__convert_format(result,4,req_comp,x,y); + + return result; +} + +static int stbi__pic_test(stbi__context *s) +{ + int r = stbi__pic_test_core(s); + stbi__rewind(s); + return r; +} +#endif + +// ************************************************************************************************* +// GIF loader -- public domain by Jean-Marc Lienher -- simplified/shrunk by stb + +#ifndef STBI_NO_GIF +typedef struct +{ + stbi__int16 prefix; + stbi_uc first; + stbi_uc suffix; +} stbi__gif_lzw; + +typedef struct +{ + int w,h; + stbi_uc *out; // output buffer (always 4 components) + stbi_uc *background; // The current "background" as far as a gif is concerned + stbi_uc *history; + int flags, bgindex, ratio, transparent, eflags; + stbi_uc pal[256][4]; + stbi_uc lpal[256][4]; + stbi__gif_lzw codes[8192]; + stbi_uc *color_table; + int parse, step; + int lflags; + int start_x, start_y; + int max_x, max_y; + int cur_x, cur_y; + int line_size; + int delay; +} stbi__gif; + +static int stbi__gif_test_raw(stbi__context *s) +{ + int sz; + if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') return 0; + sz = stbi__get8(s); + if (sz != '9' && sz != '7') return 0; + if (stbi__get8(s) != 'a') return 0; + return 1; +} + +static int stbi__gif_test(stbi__context *s) +{ + int r = stbi__gif_test_raw(s); + stbi__rewind(s); + return r; +} + +static void stbi__gif_parse_colortable(stbi__context *s, stbi_uc pal[256][4], int num_entries, int transp) +{ + int i; + for (i=0; i < num_entries; ++i) { + pal[i][2] = stbi__get8(s); + pal[i][1] = stbi__get8(s); + pal[i][0] = stbi__get8(s); + pal[i][3] = transp == i ? 0 : 255; + } +} + +static int stbi__gif_header(stbi__context *s, stbi__gif *g, int *comp, int is_info) +{ + stbi_uc version; + if (stbi__get8(s) != 'G' || stbi__get8(s) != 'I' || stbi__get8(s) != 'F' || stbi__get8(s) != '8') + return stbi__err("not GIF", "Corrupt GIF"); + + version = stbi__get8(s); + if (version != '7' && version != '9') return stbi__err("not GIF", "Corrupt GIF"); + if (stbi__get8(s) != 'a') return stbi__err("not GIF", "Corrupt GIF"); + + stbi__g_failure_reason = ""; + g->w = stbi__get16le(s); + g->h = stbi__get16le(s); + g->flags = stbi__get8(s); + g->bgindex = stbi__get8(s); + g->ratio = stbi__get8(s); + g->transparent = -1; + + if (g->w > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + if (g->h > STBI_MAX_DIMENSIONS) return stbi__err("too large","Very large image (corrupt?)"); + + if (comp != 0) *comp = 4; // can't actually tell whether it's 3 or 4 until we parse the comments + + if (is_info) return 1; + + if (g->flags & 0x80) + stbi__gif_parse_colortable(s,g->pal, 2 << (g->flags & 7), -1); + + return 1; +} + +static int stbi__gif_info_raw(stbi__context *s, int *x, int *y, int *comp) +{ + stbi__gif* g = (stbi__gif*) stbi__malloc(sizeof(stbi__gif)); + if (!stbi__gif_header(s, g, comp, 1)) { + STBI_FREE(g); + stbi__rewind( s ); + return 0; + } + if (x) *x = g->w; + if (y) *y = g->h; + STBI_FREE(g); + return 1; +} + +static void stbi__out_gif_code(stbi__gif *g, stbi__uint16 code) +{ + stbi_uc *p, *c; + int idx; + + // recurse to decode the prefixes, since the linked-list is backwards, + // and working backwards through an interleaved image would be nasty + if (g->codes[code].prefix >= 0) + stbi__out_gif_code(g, g->codes[code].prefix); + + if (g->cur_y >= g->max_y) return; + + idx = g->cur_x + g->cur_y; + p = &g->out[idx]; + g->history[idx / 4] = 1; + + c = &g->color_table[g->codes[code].suffix * 4]; + if (c[3] > 128) { // don't render transparent pixels; + p[0] = c[2]; + p[1] = c[1]; + p[2] = c[0]; + p[3] = c[3]; + } + g->cur_x += 4; + + if (g->cur_x >= g->max_x) { + g->cur_x = g->start_x; + g->cur_y += g->step; + + while (g->cur_y >= g->max_y && g->parse > 0) { + g->step = (1 << g->parse) * g->line_size; + g->cur_y = g->start_y + (g->step >> 1); + --g->parse; + } + } +} + +static stbi_uc *stbi__process_gif_raster(stbi__context *s, stbi__gif *g) +{ + stbi_uc lzw_cs; + stbi__int32 len, init_code; + stbi__uint32 first; + stbi__int32 codesize, codemask, avail, oldcode, bits, valid_bits, clear; + stbi__gif_lzw *p; + + lzw_cs = stbi__get8(s); + if (lzw_cs > 12) return NULL; + clear = 1 << lzw_cs; + first = 1; + codesize = lzw_cs + 1; + codemask = (1 << codesize) - 1; + bits = 0; + valid_bits = 0; + for (init_code = 0; init_code < clear; init_code++) { + g->codes[init_code].prefix = -1; + g->codes[init_code].first = (stbi_uc) init_code; + g->codes[init_code].suffix = (stbi_uc) init_code; + } + + // support no starting clear code + avail = clear+2; + oldcode = -1; + + len = 0; + for(;;) { + if (valid_bits < codesize) { + if (len == 0) { + len = stbi__get8(s); // start new block + if (len == 0) + return g->out; + } + --len; + bits |= (stbi__int32) stbi__get8(s) << valid_bits; + valid_bits += 8; + } else { + stbi__int32 code = bits & codemask; + bits >>= codesize; + valid_bits -= codesize; + // @OPTIMIZE: is there some way we can accelerate the non-clear path? + if (code == clear) { // clear code + codesize = lzw_cs + 1; + codemask = (1 << codesize) - 1; + avail = clear + 2; + oldcode = -1; + first = 0; + } else if (code == clear + 1) { // end of stream code + stbi__skip(s, len); + while ((len = stbi__get8(s)) > 0) + stbi__skip(s,len); + return g->out; + } else if (code <= avail) { + if (first) { + return stbi__errpuc("no clear code", "Corrupt GIF"); + } + + if (oldcode >= 0) { + p = &g->codes[avail++]; + if (avail > 8192) { + return stbi__errpuc("too many codes", "Corrupt GIF"); + } + + p->prefix = (stbi__int16) oldcode; + p->first = g->codes[oldcode].first; + p->suffix = (code == avail) ? p->first : g->codes[code].first; + } else if (code == avail) + return stbi__errpuc("illegal code in raster", "Corrupt GIF"); + + stbi__out_gif_code(g, (stbi__uint16) code); + + if ((avail & codemask) == 0 && avail <= 0x0FFF) { + codesize++; + codemask = (1 << codesize) - 1; + } + + oldcode = code; + } else { + return stbi__errpuc("illegal code in raster", "Corrupt GIF"); + } + } + } +} + +// this function is designed to support animated gifs, although stb_image doesn't support it +// two back is the image from two frames ago, used for a very specific disposal format +static stbi_uc *stbi__gif_load_next(stbi__context *s, stbi__gif *g, int *comp, int req_comp, stbi_uc *two_back) +{ + int dispose; + int first_frame; + int pi; + int pcount; + STBI_NOTUSED(req_comp); + + // on first frame, any non-written pixels get the background colour (non-transparent) + first_frame = 0; + if (g->out == 0) { + if (!stbi__gif_header(s, g, comp,0)) return 0; // stbi__g_failure_reason set by stbi__gif_header + if (!stbi__mad3sizes_valid(4, g->w, g->h, 0)) + return stbi__errpuc("too large", "GIF image is too large"); + pcount = g->w * g->h; + g->out = (stbi_uc *) stbi__malloc(4 * pcount); + g->background = (stbi_uc *) stbi__malloc(4 * pcount); + g->history = (stbi_uc *) stbi__malloc(pcount); + if (!g->out || !g->background || !g->history) + return stbi__errpuc("outofmem", "Out of memory"); + + // image is treated as "transparent" at the start - ie, nothing overwrites the current background; + // background colour is only used for pixels that are not rendered first frame, after that "background" + // color refers to the color that was there the previous frame. + memset(g->out, 0x00, 4 * pcount); + memset(g->background, 0x00, 4 * pcount); // state of the background (starts transparent) + memset(g->history, 0x00, pcount); // pixels that were affected previous frame + first_frame = 1; + } else { + // second frame - how do we dispose of the previous one? + dispose = (g->eflags & 0x1C) >> 2; + pcount = g->w * g->h; + + if ((dispose == 3) && (two_back == 0)) { + dispose = 2; // if I don't have an image to revert back to, default to the old background + } + + if (dispose == 3) { // use previous graphic + for (pi = 0; pi < pcount; ++pi) { + if (g->history[pi]) { + memcpy( &g->out[pi * 4], &two_back[pi * 4], 4 ); + } + } + } else if (dispose == 2) { + // restore what was changed last frame to background before that frame; + for (pi = 0; pi < pcount; ++pi) { + if (g->history[pi]) { + memcpy( &g->out[pi * 4], &g->background[pi * 4], 4 ); + } + } + } else { + // This is a non-disposal case eithe way, so just + // leave the pixels as is, and they will become the new background + // 1: do not dispose + // 0: not specified. + } + + // background is what out is after the undoing of the previou frame; + memcpy( g->background, g->out, 4 * g->w * g->h ); + } + + // clear my history; + memset( g->history, 0x00, g->w * g->h ); // pixels that were affected previous frame + + for (;;) { + int tag = stbi__get8(s); + switch (tag) { + case 0x2C: /* Image Descriptor */ + { + stbi__int32 x, y, w, h; + stbi_uc *o; + + x = stbi__get16le(s); + y = stbi__get16le(s); + w = stbi__get16le(s); + h = stbi__get16le(s); + if (((x + w) > (g->w)) || ((y + h) > (g->h))) + return stbi__errpuc("bad Image Descriptor", "Corrupt GIF"); + + g->line_size = g->w * 4; + g->start_x = x * 4; + g->start_y = y * g->line_size; + g->max_x = g->start_x + w * 4; + g->max_y = g->start_y + h * g->line_size; + g->cur_x = g->start_x; + g->cur_y = g->start_y; + + // if the width of the specified rectangle is 0, that means + // we may not see *any* pixels or the image is malformed; + // to make sure this is caught, move the current y down to + // max_y (which is what out_gif_code checks). + if (w == 0) + g->cur_y = g->max_y; + + g->lflags = stbi__get8(s); + + if (g->lflags & 0x40) { + g->step = 8 * g->line_size; // first interlaced spacing + g->parse = 3; + } else { + g->step = g->line_size; + g->parse = 0; + } + + if (g->lflags & 0x80) { + stbi__gif_parse_colortable(s,g->lpal, 2 << (g->lflags & 7), g->eflags & 0x01 ? g->transparent : -1); + g->color_table = (stbi_uc *) g->lpal; + } else if (g->flags & 0x80) { + g->color_table = (stbi_uc *) g->pal; + } else + return stbi__errpuc("missing color table", "Corrupt GIF"); + + o = stbi__process_gif_raster(s, g); + if (!o) return NULL; + + // if this was the first frame, + pcount = g->w * g->h; + if (first_frame && (g->bgindex > 0)) { + // if first frame, any pixel not drawn to gets the background color + for (pi = 0; pi < pcount; ++pi) { + if (g->history[pi] == 0) { + g->pal[g->bgindex][3] = 255; // just in case it was made transparent, undo that; It will be reset next frame if need be; + memcpy( &g->out[pi * 4], &g->pal[g->bgindex], 4 ); + } + } + } + + return o; + } + + case 0x21: // Comment Extension. + { + int len; + int ext = stbi__get8(s); + if (ext == 0xF9) { // Graphic Control Extension. + len = stbi__get8(s); + if (len == 4) { + g->eflags = stbi__get8(s); + g->delay = 10 * stbi__get16le(s); // delay - 1/100th of a second, saving as 1/1000ths. + + // unset old transparent + if (g->transparent >= 0) { + g->pal[g->transparent][3] = 255; + } + if (g->eflags & 0x01) { + g->transparent = stbi__get8(s); + if (g->transparent >= 0) { + g->pal[g->transparent][3] = 0; + } + } else { + // don't need transparent + stbi__skip(s, 1); + g->transparent = -1; + } + } else { + stbi__skip(s, len); + break; + } + } + while ((len = stbi__get8(s)) != 0) { + stbi__skip(s, len); + } + break; + } + + case 0x3B: // gif stream termination code + return (stbi_uc *) s; // using '1' causes warning on some compilers + + default: + return stbi__errpuc("unknown code", "Corrupt GIF"); + } + } +} + +static void *stbi__load_gif_main(stbi__context *s, int **delays, int *x, int *y, int *z, int *comp, int req_comp) +{ + if (stbi__gif_test(s)) { + int layers = 0; + stbi_uc *u = 0; + stbi_uc *out = 0; + stbi_uc *two_back = 0; + stbi__gif g; + int stride; + int out_size = 0; + int delays_size = 0; + memset(&g, 0, sizeof(g)); + if (delays) { + *delays = 0; + } + + do { + u = stbi__gif_load_next(s, &g, comp, req_comp, two_back); + if (u == (stbi_uc *) s) u = 0; // end of animated gif marker + + if (u) { + *x = g.w; + *y = g.h; + ++layers; + stride = g.w * g.h * 4; + + if (out) { + void *tmp = (stbi_uc*) STBI_REALLOC_SIZED( out, out_size, layers * stride ); + if (NULL == tmp) { + STBI_FREE(g.out); + STBI_FREE(g.history); + STBI_FREE(g.background); + return stbi__errpuc("outofmem", "Out of memory"); + } + else { + out = (stbi_uc*) tmp; + out_size = layers * stride; + } + + if (delays) { + *delays = (int*) STBI_REALLOC_SIZED( *delays, delays_size, sizeof(int) * layers ); + delays_size = layers * sizeof(int); + } + } else { + out = (stbi_uc*)stbi__malloc( layers * stride ); + out_size = layers * stride; + if (delays) { + *delays = (int*) stbi__malloc( layers * sizeof(int) ); + delays_size = layers * sizeof(int); + } + } + memcpy( out + ((layers - 1) * stride), u, stride ); + if (layers >= 2) { + two_back = out - 2 * stride; + } + + if (delays) { + (*delays)[layers - 1U] = g.delay; + } + } + } while (u != 0); + + // free temp buffer; + STBI_FREE(g.out); + STBI_FREE(g.history); + STBI_FREE(g.background); + + // do the final conversion after loading everything; + if (req_comp && req_comp != 4) + out = stbi__convert_format(out, 4, req_comp, layers * g.w, g.h); + + *z = layers; + return out; + } else { + return stbi__errpuc("not GIF", "Image was not as a gif type."); + } +} + +static void *stbi__gif_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + stbi_uc *u = 0; + stbi__gif g; + memset(&g, 0, sizeof(g)); + STBI_NOTUSED(ri); + + u = stbi__gif_load_next(s, &g, comp, req_comp, 0); + if (u == (stbi_uc *) s) u = 0; // end of animated gif marker + if (u) { + *x = g.w; + *y = g.h; + + // moved conversion to after successful load so that the same + // can be done for multiple frames. + if (req_comp && req_comp != 4) + u = stbi__convert_format(u, 4, req_comp, g.w, g.h); + } else if (g.out) { + // if there was an error and we allocated an image buffer, free it! + STBI_FREE(g.out); + } + + // free buffers needed for multiple frame loading; + STBI_FREE(g.history); + STBI_FREE(g.background); + + return u; +} + +static int stbi__gif_info(stbi__context *s, int *x, int *y, int *comp) +{ + return stbi__gif_info_raw(s,x,y,comp); +} +#endif + +// ************************************************************************************************* +// Radiance RGBE HDR loader +// originally by Nicolas Schulz +#ifndef STBI_NO_HDR +static int stbi__hdr_test_core(stbi__context *s, const char *signature) +{ + int i; + for (i=0; signature[i]; ++i) + if (stbi__get8(s) != signature[i]) + return 0; + stbi__rewind(s); + return 1; +} + +static int stbi__hdr_test(stbi__context* s) +{ + int r = stbi__hdr_test_core(s, "#?RADIANCE\n"); + stbi__rewind(s); + if(!r) { + r = stbi__hdr_test_core(s, "#?RGBE\n"); + stbi__rewind(s); + } + return r; +} + +#define STBI__HDR_BUFLEN 1024 +static char *stbi__hdr_gettoken(stbi__context *z, char *buffer) +{ + int len=0; + char c = '\0'; + + c = (char) stbi__get8(z); + + while (!stbi__at_eof(z) && c != '\n') { + buffer[len++] = c; + if (len == STBI__HDR_BUFLEN-1) { + // flush to end of line + while (!stbi__at_eof(z) && stbi__get8(z) != '\n') + ; + break; + } + c = (char) stbi__get8(z); + } + + buffer[len] = 0; + return buffer; +} + +static void stbi__hdr_convert(float *output, stbi_uc *input, int req_comp) +{ + if ( input[3] != 0 ) { + float f1; + // Exponent + f1 = (float) ldexp(1.0f, input[3] - (int)(128 + 8)); + if (req_comp <= 2) + output[0] = (input[0] + input[1] + input[2]) * f1 / 3; + else { + output[0] = input[0] * f1; + output[1] = input[1] * f1; + output[2] = input[2] * f1; + } + if (req_comp == 2) output[1] = 1; + if (req_comp == 4) output[3] = 1; + } else { + switch (req_comp) { + case 4: output[3] = 1; /* fallthrough */ + case 3: output[0] = output[1] = output[2] = 0; + break; + case 2: output[1] = 1; /* fallthrough */ + case 1: output[0] = 0; + break; + } + } +} + +static float *stbi__hdr_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + char buffer[STBI__HDR_BUFLEN]; + char *token; + int valid = 0; + int width, height; + stbi_uc *scanline; + float *hdr_data; + int len; + unsigned char count, value; + int i, j, k, c1,c2, z; + const char *headerToken; + STBI_NOTUSED(ri); + + // Check identifier + headerToken = stbi__hdr_gettoken(s,buffer); + if (strcmp(headerToken, "#?RADIANCE") != 0 && strcmp(headerToken, "#?RGBE") != 0) + return stbi__errpf("not HDR", "Corrupt HDR image"); + + // Parse header + for(;;) { + token = stbi__hdr_gettoken(s,buffer); + if (token[0] == 0) break; + if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1; + } + + if (!valid) return stbi__errpf("unsupported format", "Unsupported HDR format"); + + // Parse width and height + // can't use sscanf() if we're not using stdio! + token = stbi__hdr_gettoken(s,buffer); + if (strncmp(token, "-Y ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format"); + token += 3; + height = (int) strtol(token, &token, 10); + while (*token == ' ') ++token; + if (strncmp(token, "+X ", 3)) return stbi__errpf("unsupported data layout", "Unsupported HDR format"); + token += 3; + width = (int) strtol(token, NULL, 10); + + if (height > STBI_MAX_DIMENSIONS) return stbi__errpf("too large","Very large image (corrupt?)"); + if (width > STBI_MAX_DIMENSIONS) return stbi__errpf("too large","Very large image (corrupt?)"); + + *x = width; + *y = height; + + if (comp) *comp = 3; + if (req_comp == 0) req_comp = 3; + + if (!stbi__mad4sizes_valid(width, height, req_comp, sizeof(float), 0)) + return stbi__errpf("too large", "HDR image is too large"); + + // Read data + hdr_data = (float *) stbi__malloc_mad4(width, height, req_comp, sizeof(float), 0); + if (!hdr_data) + return stbi__errpf("outofmem", "Out of memory"); + + // Load image data + // image data is stored as some number of sca + if ( width < 8 || width >= 32768) { + // Read flat data + for (j=0; j < height; ++j) { + for (i=0; i < width; ++i) { + stbi_uc rgbe[4]; + main_decode_loop: + stbi__getn(s, rgbe, 4); + stbi__hdr_convert(hdr_data + j * width * req_comp + i * req_comp, rgbe, req_comp); + } + } + } else { + // Read RLE-encoded data + scanline = NULL; + + for (j = 0; j < height; ++j) { + c1 = stbi__get8(s); + c2 = stbi__get8(s); + len = stbi__get8(s); + if (c1 != 2 || c2 != 2 || (len & 0x80)) { + // not run-length encoded, so we have to actually use THIS data as a decoded + // pixel (note this can't be a valid pixel--one of RGB must be >= 128) + stbi_uc rgbe[4]; + rgbe[0] = (stbi_uc) c1; + rgbe[1] = (stbi_uc) c2; + rgbe[2] = (stbi_uc) len; + rgbe[3] = (stbi_uc) stbi__get8(s); + stbi__hdr_convert(hdr_data, rgbe, req_comp); + i = 1; + j = 0; + STBI_FREE(scanline); + goto main_decode_loop; // yes, this makes no sense + } + len <<= 8; + len |= stbi__get8(s); + if (len != width) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("invalid decoded scanline length", "corrupt HDR"); } + if (scanline == NULL) { + scanline = (stbi_uc *) stbi__malloc_mad2(width, 4, 0); + if (!scanline) { + STBI_FREE(hdr_data); + return stbi__errpf("outofmem", "Out of memory"); + } + } + + for (k = 0; k < 4; ++k) { + int nleft; + i = 0; + while ((nleft = width - i) > 0) { + count = stbi__get8(s); + if (count > 128) { + // Run + value = stbi__get8(s); + count -= 128; + if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); } + for (z = 0; z < count; ++z) + scanline[i++ * 4 + k] = value; + } else { + // Dump + if (count > nleft) { STBI_FREE(hdr_data); STBI_FREE(scanline); return stbi__errpf("corrupt", "bad RLE data in HDR"); } + for (z = 0; z < count; ++z) + scanline[i++ * 4 + k] = stbi__get8(s); + } + } + } + for (i=0; i < width; ++i) + stbi__hdr_convert(hdr_data+(j*width + i)*req_comp, scanline + i*4, req_comp); + } + if (scanline) + STBI_FREE(scanline); + } + + return hdr_data; +} + +static int stbi__hdr_info(stbi__context *s, int *x, int *y, int *comp) +{ + char buffer[STBI__HDR_BUFLEN]; + char *token; + int valid = 0; + int dummy; + + if (!x) x = &dummy; + if (!y) y = &dummy; + if (!comp) comp = &dummy; + + if (stbi__hdr_test(s) == 0) { + stbi__rewind( s ); + return 0; + } + + for(;;) { + token = stbi__hdr_gettoken(s,buffer); + if (token[0] == 0) break; + if (strcmp(token, "FORMAT=32-bit_rle_rgbe") == 0) valid = 1; + } + + if (!valid) { + stbi__rewind( s ); + return 0; + } + token = stbi__hdr_gettoken(s,buffer); + if (strncmp(token, "-Y ", 3)) { + stbi__rewind( s ); + return 0; + } + token += 3; + *y = (int) strtol(token, &token, 10); + while (*token == ' ') ++token; + if (strncmp(token, "+X ", 3)) { + stbi__rewind( s ); + return 0; + } + token += 3; + *x = (int) strtol(token, NULL, 10); + *comp = 3; + return 1; +} +#endif // STBI_NO_HDR + +#ifndef STBI_NO_BMP +static int stbi__bmp_info(stbi__context *s, int *x, int *y, int *comp) +{ + void *p; + stbi__bmp_data info; + + info.all_a = 255; + p = stbi__bmp_parse_header(s, &info); + stbi__rewind( s ); + if (p == NULL) + return 0; + if (x) *x = s->img_x; + if (y) *y = s->img_y; + if (comp) { + if (info.bpp == 24 && info.ma == 0xff000000) + *comp = 3; + else + *comp = info.ma ? 4 : 3; + } + return 1; +} +#endif + +#ifndef STBI_NO_PSD +static int stbi__psd_info(stbi__context *s, int *x, int *y, int *comp) +{ + int channelCount, dummy, depth; + if (!x) x = &dummy; + if (!y) y = &dummy; + if (!comp) comp = &dummy; + if (stbi__get32be(s) != 0x38425053) { + stbi__rewind( s ); + return 0; + } + if (stbi__get16be(s) != 1) { + stbi__rewind( s ); + return 0; + } + stbi__skip(s, 6); + channelCount = stbi__get16be(s); + if (channelCount < 0 || channelCount > 16) { + stbi__rewind( s ); + return 0; + } + *y = stbi__get32be(s); + *x = stbi__get32be(s); + depth = stbi__get16be(s); + if (depth != 8 && depth != 16) { + stbi__rewind( s ); + return 0; + } + if (stbi__get16be(s) != 3) { + stbi__rewind( s ); + return 0; + } + *comp = 4; + return 1; +} + +static int stbi__psd_is16(stbi__context *s) +{ + int channelCount, depth; + if (stbi__get32be(s) != 0x38425053) { + stbi__rewind( s ); + return 0; + } + if (stbi__get16be(s) != 1) { + stbi__rewind( s ); + return 0; + } + stbi__skip(s, 6); + channelCount = stbi__get16be(s); + if (channelCount < 0 || channelCount > 16) { + stbi__rewind( s ); + return 0; + } + (void) stbi__get32be(s); + (void) stbi__get32be(s); + depth = stbi__get16be(s); + if (depth != 16) { + stbi__rewind( s ); + return 0; + } + return 1; +} +#endif + +#ifndef STBI_NO_PIC +static int stbi__pic_info(stbi__context *s, int *x, int *y, int *comp) +{ + int act_comp=0,num_packets=0,chained,dummy; + stbi__pic_packet packets[10]; + + if (!x) x = &dummy; + if (!y) y = &dummy; + if (!comp) comp = &dummy; + + if (!stbi__pic_is4(s,"\x53\x80\xF6\x34")) { + stbi__rewind(s); + return 0; + } + + stbi__skip(s, 88); + + *x = stbi__get16be(s); + *y = stbi__get16be(s); + if (stbi__at_eof(s)) { + stbi__rewind( s); + return 0; + } + if ( (*x) != 0 && (1 << 28) / (*x) < (*y)) { + stbi__rewind( s ); + return 0; + } + + stbi__skip(s, 8); + + do { + stbi__pic_packet *packet; + + if (num_packets==sizeof(packets)/sizeof(packets[0])) + return 0; + + packet = &packets[num_packets++]; + chained = stbi__get8(s); + packet->size = stbi__get8(s); + packet->type = stbi__get8(s); + packet->channel = stbi__get8(s); + act_comp |= packet->channel; + + if (stbi__at_eof(s)) { + stbi__rewind( s ); + return 0; + } + if (packet->size != 8) { + stbi__rewind( s ); + return 0; + } + } while (chained); + + *comp = (act_comp & 0x10 ? 4 : 3); + + return 1; +} +#endif + +// ************************************************************************************************* +// Portable Gray Map and Portable Pixel Map loader +// by Ken Miller +// +// PGM: http://netpbm.sourceforge.net/doc/pgm.html +// PPM: http://netpbm.sourceforge.net/doc/ppm.html +// +// Known limitations: +// Does not support comments in the header section +// Does not support ASCII image data (formats P2 and P3) +// Does not support 16-bit-per-channel + +#ifndef STBI_NO_PNM + +static int stbi__pnm_test(stbi__context *s) +{ + char p, t; + p = (char) stbi__get8(s); + t = (char) stbi__get8(s); + if (p != 'P' || (t != '5' && t != '6')) { + stbi__rewind( s ); + return 0; + } + return 1; +} + +static void *stbi__pnm_load(stbi__context *s, int *x, int *y, int *comp, int req_comp, stbi__result_info *ri) +{ + stbi_uc *out; + STBI_NOTUSED(ri); + + if (!stbi__pnm_info(s, (int *)&s->img_x, (int *)&s->img_y, (int *)&s->img_n)) + return 0; + + if (s->img_y > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + if (s->img_x > STBI_MAX_DIMENSIONS) return stbi__errpuc("too large","Very large image (corrupt?)"); + + *x = s->img_x; + *y = s->img_y; + if (comp) *comp = s->img_n; + + if (!stbi__mad3sizes_valid(s->img_n, s->img_x, s->img_y, 0)) + return stbi__errpuc("too large", "PNM too large"); + + out = (stbi_uc *) stbi__malloc_mad3(s->img_n, s->img_x, s->img_y, 0); + if (!out) return stbi__errpuc("outofmem", "Out of memory"); + stbi__getn(s, out, s->img_n * s->img_x * s->img_y); + + if (req_comp && req_comp != s->img_n) { + out = stbi__convert_format(out, s->img_n, req_comp, s->img_x, s->img_y); + if (out == NULL) return out; // stbi__convert_format frees input on failure + } + return out; +} + +static int stbi__pnm_isspace(char c) +{ + return c == ' ' || c == '\t' || c == '\n' || c == '\v' || c == '\f' || c == '\r'; +} + +static void stbi__pnm_skip_whitespace(stbi__context *s, char *c) +{ + for (;;) { + while (!stbi__at_eof(s) && stbi__pnm_isspace(*c)) + *c = (char) stbi__get8(s); + + if (stbi__at_eof(s) || *c != '#') + break; + + while (!stbi__at_eof(s) && *c != '\n' && *c != '\r' ) + *c = (char) stbi__get8(s); + } +} + +static int stbi__pnm_isdigit(char c) +{ + return c >= '0' && c <= '9'; +} + +static int stbi__pnm_getinteger(stbi__context *s, char *c) +{ + int value = 0; + + while (!stbi__at_eof(s) && stbi__pnm_isdigit(*c)) { + value = value*10 + (*c - '0'); + *c = (char) stbi__get8(s); + } + + return value; +} + +static int stbi__pnm_info(stbi__context *s, int *x, int *y, int *comp) +{ + int maxv, dummy; + char c, p, t; + + if (!x) x = &dummy; + if (!y) y = &dummy; + if (!comp) comp = &dummy; + + stbi__rewind(s); + + // Get identifier + p = (char) stbi__get8(s); + t = (char) stbi__get8(s); + if (p != 'P' || (t != '5' && t != '6')) { + stbi__rewind(s); + return 0; + } + + *comp = (t == '6') ? 3 : 1; // '5' is 1-component .pgm; '6' is 3-component .ppm + + c = (char) stbi__get8(s); + stbi__pnm_skip_whitespace(s, &c); + + *x = stbi__pnm_getinteger(s, &c); // read width + stbi__pnm_skip_whitespace(s, &c); + + *y = stbi__pnm_getinteger(s, &c); // read height + stbi__pnm_skip_whitespace(s, &c); + + maxv = stbi__pnm_getinteger(s, &c); // read max value + + if (maxv > 255) + return stbi__err("max value > 255", "PPM image not 8-bit"); + else + return 1; +} +#endif + +static int stbi__info_main(stbi__context *s, int *x, int *y, int *comp) +{ + #ifndef STBI_NO_JPEG + if (stbi__jpeg_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_PNG + if (stbi__png_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_GIF + if (stbi__gif_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_BMP + if (stbi__bmp_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_PSD + if (stbi__psd_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_PIC + if (stbi__pic_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_PNM + if (stbi__pnm_info(s, x, y, comp)) return 1; + #endif + + #ifndef STBI_NO_HDR + if (stbi__hdr_info(s, x, y, comp)) return 1; + #endif + + // test tga last because it's a crappy test! + #ifndef STBI_NO_TGA + if (stbi__tga_info(s, x, y, comp)) + return 1; + #endif + return stbi__err("unknown image type", "Image not of any known type, or corrupt"); +} + +static int stbi__is_16_main(stbi__context *s) +{ + #ifndef STBI_NO_PNG + if (stbi__png_is16(s)) return 1; + #endif + + #ifndef STBI_NO_PSD + if (stbi__psd_is16(s)) return 1; + #endif + + return 0; +} + +#ifndef STBI_NO_STDIO +STBIDEF int stbi_info(char const *filename, int *x, int *y, int *comp) +{ + FILE *f = stbi__fopen(filename, "rb"); + int result; + if (!f) return stbi__err("can't fopen", "Unable to open file"); + result = stbi_info_from_file(f, x, y, comp); + fclose(f); + return result; +} + +STBIDEF int stbi_info_from_file(FILE *f, int *x, int *y, int *comp) +{ + int r; + stbi__context s; + long pos = ftell(f); + stbi__start_file(&s, f); + r = stbi__info_main(&s,x,y,comp); + fseek(f,pos,SEEK_SET); + return r; +} + +STBIDEF int stbi_is_16_bit(char const *filename) +{ + FILE *f = stbi__fopen(filename, "rb"); + int result; + if (!f) return stbi__err("can't fopen", "Unable to open file"); + result = stbi_is_16_bit_from_file(f); + fclose(f); + return result; +} + +STBIDEF int stbi_is_16_bit_from_file(FILE *f) +{ + int r; + stbi__context s; + long pos = ftell(f); + stbi__start_file(&s, f); + r = stbi__is_16_main(&s); + fseek(f,pos,SEEK_SET); + return r; +} +#endif // !STBI_NO_STDIO + +STBIDEF int stbi_info_from_memory(stbi_uc const *buffer, int len, int *x, int *y, int *comp) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__info_main(&s,x,y,comp); +} + +STBIDEF int stbi_info_from_callbacks(stbi_io_callbacks const *c, void *user, int *x, int *y, int *comp) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user); + return stbi__info_main(&s,x,y,comp); +} + +STBIDEF int stbi_is_16_bit_from_memory(stbi_uc const *buffer, int len) +{ + stbi__context s; + stbi__start_mem(&s,buffer,len); + return stbi__is_16_main(&s); +} + +STBIDEF int stbi_is_16_bit_from_callbacks(stbi_io_callbacks const *c, void *user) +{ + stbi__context s; + stbi__start_callbacks(&s, (stbi_io_callbacks *) c, user); + return stbi__is_16_main(&s); +} + +#endif // STB_IMAGE_IMPLEMENTATION + +/* + revision history: + 2.20 (2019-02-07) support utf8 filenames in Windows; fix warnings and platform ifdefs + 2.19 (2018-02-11) fix warning + 2.18 (2018-01-30) fix warnings + 2.17 (2018-01-29) change sbti__shiftsigned to avoid clang -O2 bug + 1-bit BMP + *_is_16_bit api + avoid warnings + 2.16 (2017-07-23) all functions have 16-bit variants; + STBI_NO_STDIO works again; + compilation fixes; + fix rounding in unpremultiply; + optimize vertical flip; + disable raw_len validation; + documentation fixes + 2.15 (2017-03-18) fix png-1,2,4 bug; now all Imagenet JPGs decode; + warning fixes; disable run-time SSE detection on gcc; + uniform handling of optional "return" values; + thread-safe initialization of zlib tables + 2.14 (2017-03-03) remove deprecated STBI_JPEG_OLD; fixes for Imagenet JPGs + 2.13 (2016-11-29) add 16-bit API, only supported for PNG right now + 2.12 (2016-04-02) fix typo in 2.11 PSD fix that caused crashes + 2.11 (2016-04-02) allocate large structures on the stack + remove white matting for transparent PSD + fix reported channel count for PNG & BMP + re-enable SSE2 in non-gcc 64-bit + support RGB-formatted JPEG + read 16-bit PNGs (only as 8-bit) + 2.10 (2016-01-22) avoid warning introduced in 2.09 by STBI_REALLOC_SIZED + 2.09 (2016-01-16) allow comments in PNM files + 16-bit-per-pixel TGA (not bit-per-component) + info() for TGA could break due to .hdr handling + info() for BMP to shares code instead of sloppy parse + can use STBI_REALLOC_SIZED if allocator doesn't support realloc + code cleanup + 2.08 (2015-09-13) fix to 2.07 cleanup, reading RGB PSD as RGBA + 2.07 (2015-09-13) fix compiler warnings + partial animated GIF support + limited 16-bpc PSD support + #ifdef unused functions + bug with < 92 byte PIC,PNM,HDR,TGA + 2.06 (2015-04-19) fix bug where PSD returns wrong '*comp' value + 2.05 (2015-04-19) fix bug in progressive JPEG handling, fix warning + 2.04 (2015-04-15) try to re-enable SIMD on MinGW 64-bit + 2.03 (2015-04-12) extra corruption checking (mmozeiko) + stbi_set_flip_vertically_on_load (nguillemot) + fix NEON support; fix mingw support + 2.02 (2015-01-19) fix incorrect assert, fix warning + 2.01 (2015-01-17) fix various warnings; suppress SIMD on gcc 32-bit without -msse2 + 2.00b (2014-12-25) fix STBI_MALLOC in progressive JPEG + 2.00 (2014-12-25) optimize JPG, including x86 SSE2 & NEON SIMD (ryg) + progressive JPEG (stb) + PGM/PPM support (Ken Miller) + STBI_MALLOC,STBI_REALLOC,STBI_FREE + GIF bugfix -- seemingly never worked + STBI_NO_*, STBI_ONLY_* + 1.48 (2014-12-14) fix incorrectly-named assert() + 1.47 (2014-12-14) 1/2/4-bit PNG support, both direct and paletted (Omar Cornut & stb) + optimize PNG (ryg) + fix bug in interlaced PNG with user-specified channel count (stb) + 1.46 (2014-08-26) + fix broken tRNS chunk (colorkey-style transparency) in non-paletted PNG + 1.45 (2014-08-16) + fix MSVC-ARM internal compiler error by wrapping malloc + 1.44 (2014-08-07) + various warning fixes from Ronny Chevalier + 1.43 (2014-07-15) + fix MSVC-only compiler problem in code changed in 1.42 + 1.42 (2014-07-09) + don't define _CRT_SECURE_NO_WARNINGS (affects user code) + fixes to stbi__cleanup_jpeg path + added STBI_ASSERT to avoid requiring assert.h + 1.41 (2014-06-25) + fix search&replace from 1.36 that messed up comments/error messages + 1.40 (2014-06-22) + fix gcc struct-initialization warning + 1.39 (2014-06-15) + fix to TGA optimization when req_comp != number of components in TGA; + fix to GIF loading because BMP wasn't rewinding (whoops, no GIFs in my test suite) + add support for BMP version 5 (more ignored fields) + 1.38 (2014-06-06) + suppress MSVC warnings on integer casts truncating values + fix accidental rename of 'skip' field of I/O + 1.37 (2014-06-04) + remove duplicate typedef + 1.36 (2014-06-03) + convert to header file single-file library + if de-iphone isn't set, load iphone images color-swapped instead of returning NULL + 1.35 (2014-05-27) + various warnings + fix broken STBI_SIMD path + fix bug where stbi_load_from_file no longer left file pointer in correct place + fix broken non-easy path for 32-bit BMP (possibly never used) + TGA optimization by Arseny Kapoulkine + 1.34 (unknown) + use STBI_NOTUSED in stbi__resample_row_generic(), fix one more leak in tga failure case + 1.33 (2011-07-14) + make stbi_is_hdr work in STBI_NO_HDR (as specified), minor compiler-friendly improvements + 1.32 (2011-07-13) + support for "info" function for all supported filetypes (SpartanJ) + 1.31 (2011-06-20) + a few more leak fixes, bug in PNG handling (SpartanJ) + 1.30 (2011-06-11) + added ability to load files via callbacks to accomidate custom input streams (Ben Wenger) + removed deprecated format-specific test/load functions + removed support for installable file formats (stbi_loader) -- would have been broken for IO callbacks anyway + error cases in bmp and tga give messages and don't leak (Raymond Barbiero, grisha) + fix inefficiency in decoding 32-bit BMP (David Woo) + 1.29 (2010-08-16) + various warning fixes from Aurelien Pocheville + 1.28 (2010-08-01) + fix bug in GIF palette transparency (SpartanJ) + 1.27 (2010-08-01) + cast-to-stbi_uc to fix warnings + 1.26 (2010-07-24) + fix bug in file buffering for PNG reported by SpartanJ + 1.25 (2010-07-17) + refix trans_data warning (Won Chun) + 1.24 (2010-07-12) + perf improvements reading from files on platforms with lock-heavy fgetc() + minor perf improvements for jpeg + deprecated type-specific functions so we'll get feedback if they're needed + attempt to fix trans_data warning (Won Chun) + 1.23 fixed bug in iPhone support + 1.22 (2010-07-10) + removed image *writing* support + stbi_info support from Jetro Lauha + GIF support from Jean-Marc Lienher + iPhone PNG-extensions from James Brown + warning-fixes from Nicolas Schulz and Janez Zemva (i.stbi__err. Janez (U+017D)emva) + 1.21 fix use of 'stbi_uc' in header (reported by jon blow) + 1.20 added support for Softimage PIC, by Tom Seddon + 1.19 bug in interlaced PNG corruption check (found by ryg) + 1.18 (2008-08-02) + fix a threading bug (local mutable static) + 1.17 support interlaced PNG + 1.16 major bugfix - stbi__convert_format converted one too many pixels + 1.15 initialize some fields for thread safety + 1.14 fix threadsafe conversion bug + header-file-only version (#define STBI_HEADER_FILE_ONLY before including) + 1.13 threadsafe + 1.12 const qualifiers in the API + 1.11 Support installable IDCT, colorspace conversion routines + 1.10 Fixes for 64-bit (don't use "unsigned long") + optimized upsampling by Fabian "ryg" Giesen + 1.09 Fix format-conversion for PSD code (bad global variables!) + 1.08 Thatcher Ulrich's PSD code integrated by Nicolas Schulz + 1.07 attempt to fix C++ warning/errors again + 1.06 attempt to fix C++ warning/errors again + 1.05 fix TGA loading to return correct *comp and use good luminance calc + 1.04 default float alpha is 1, not 255; use 'void *' for stbi_image_free + 1.03 bugfixes to STBI_NO_STDIO, STBI_NO_HDR + 1.02 support for (subset of) HDR files, float interface for preferred access to them + 1.01 fix bug: possible bug in handling right-side up bmps... not sure + fix bug: the stbi__bmp_load() and stbi__tga_load() functions didn't work at all + 1.00 interface to zlib that skips zlib header + 0.99 correct handling of alpha in palette + 0.98 TGA loader by lonesock; dynamically add loaders (untested) + 0.97 jpeg errors on too large a file; also catch another malloc failure + 0.96 fix detection of invalid v value - particleman@mollyrocket forum + 0.95 during header scan, seek to markers in case of padding + 0.94 STBI_NO_STDIO to disable stdio usage; rename all #defines the same + 0.93 handle jpegtran output; verbose errors + 0.92 read 4,8,16,24,32-bit BMP files of several formats + 0.91 output 24-bit Windows 3.0 BMP files + 0.90 fix a few more warnings; bump version number to approach 1.0 + 0.61 bugfixes due to Marc LeBlanc, Christopher Lloyd + 0.60 fix compiling as c++ + 0.59 fix warnings: merge Dave Moore's -Wall fixes + 0.58 fix bug: zlib uncompressed mode len/nlen was wrong endian + 0.57 fix bug: jpg last huffman symbol before marker was >9 bits but less than 16 available + 0.56 fix bug: zlib uncompressed mode len vs. nlen + 0.55 fix bug: restart_interval not initialized to 0 + 0.54 allow NULL for 'int *comp' + 0.53 fix bug in png 3->4; speedup png decoding + 0.52 png handles req_comp=3,4 directly; minor cleanup; jpeg comments + 0.51 obey req_comp requests, 1-component jpegs return as 1-component, + on 'test' only check type, not whether we support this variant + 0.50 (2006-11-19) + first released version +*/ + + +/* +------------------------------------------------------------------------------ +This software is available under 2 licenses -- choose whichever you prefer. +------------------------------------------------------------------------------ +ALTERNATIVE A - MIT License +Copyright (c) 2017 Sean Barrett +Permission is hereby granted, free of charge, to any person obtaining a copy of +this software and associated documentation files (the "Software"), to deal in +the Software without restriction, including without limitation the rights to +use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies +of the Software, and to permit persons to whom the Software is furnished to do +so, subject to the following conditions: +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. +------------------------------------------------------------------------------ +ALTERNATIVE B - Public Domain (www.unlicense.org) +This is free and unencumbered software released into the public domain. +Anyone is free to copy, modify, publish, use, compile, sell, or distribute this +software, either in source code form or as a compiled binary, for any purpose, +commercial or non-commercial, and by any means. +In jurisdictions that recognize copyright laws, the author or authors of this +software dedicate any and all copyright interest in the software to the public +domain. We make this dedication for the benefit of the public at large and to +the detriment of our heirs and successors. We intend this dedication to be an +overt act of relinquishment in perpetuity of all present and future rights to +this software under copyright law. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN +ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. +------------------------------------------------------------------------------ +*/ diff --git a/kernels/knn_float4_topk.cl b/kernels/knn_float4_topk.cl new file mode 100644 index 0000000..d8660c0 --- /dev/null +++ b/kernels/knn_float4_topk.cl @@ -0,0 +1,191 @@ + +#ifndef DIM +#define DIM 128 +#endif + +#ifndef TOPK +#define TOPK 32 +#endif + +#ifndef QUERY_COUNT +#define QUERY_COUNT 1 +#endif + +/* + * Numero di work-item che collaborano per ogni query. + */ +#ifndef LANES_PER_QUERY +#define LANES_PER_QUERY 8 +#endif + +/* + * Ogni work-group contiene 8 work-item. + * + * Ogni work-item calcola un TOPK locale. + * Poi il work-item 0 fonde gli 8 TOPK locali. + */ + +__kernel void knn( + __global const float *db, + __global const float *queries, + __global uint *indices, + __global float *values, + uint n) +{ + uint lane = get_local_id(0); + uint q = get_group_id(0); + + if (q >= QUERY_COUNT) + return; + + /* + * TOPK locale per ogni work-item. + */ + float best_score[TOPK]; + uint best_index[TOPK]; + + for (uint k = 0; k < TOPK; ++k) { + best_score[k] = -3.402823466e+38f; + best_index[k] = 0xffffffff; + } + + __global const float *query = + &queries[q * DIM]; + + /* + * Ogni lane analizza una parte del database. + * + * Esempio con 8 lane: + * + * lane 0 -> 0, 8, 16, 24, ... + * lane 1 -> 1, 9, 17, 25, ... + * ... + * lane 7 -> 7, 15, 23, 31, ... + */ + for (uint i = lane; i < n; i += LANES_PER_QUERY) { + + __global const float *v = + &db[i * DIM]; + + float s = 0.0f; + + for (uint j = 0; j < DIM; ++j) + s += v[j] * query[j]; + + if (s <= best_score[TOPK - 1]) + continue; + + int pos = TOPK - 1; + + while (pos > 0 && + s > best_score[pos - 1]) { + --pos; + } + + for (int k = TOPK - 1; k > pos; --k) { + best_score[k] = + best_score[k - 1]; + + best_index[k] = + best_index[k - 1]; + } + + best_score[pos] = s; + best_index[pos] = i; + } + + /* + * TOPK locali nella local memory. + * + * 8 lane × 32 elementi. + */ + __local float local_scores[LANES_PER_QUERY][TOPK]; + __local uint local_indices[LANES_PER_QUERY][TOPK]; + + for (uint k = 0; k < TOPK; ++k) { + local_scores[lane][k] = best_score[k]; + local_indices[lane][k] = best_index[k]; + } + + /* + * Tutti i lane devono aver scritto + * la local memory prima della riduzione. + */ + barrier(CLK_LOCAL_MEM_FENCE); + + /* + * Solo lane 0 esegue il merge finale. + */ + if (lane == 0) { + + float final_score[TOPK]; + uint final_index[TOPK]; + + for (uint k = 0; k < TOPK; ++k) { + final_score[k] = + -3.402823466e+38f; + + final_index[k] = + 0xffffffff; + } + + /* + * Fonde gli 8 TOPK locali. + * + * Al massimo 8 × 32 = 256 candidati. + */ + for (uint l = 0; + l < LANES_PER_QUERY; + ++l) { + + for (uint k = 0; + k < TOPK; + ++k) { + + float s = + local_scores[l][k]; + + uint idx = + local_indices[l][k]; + + if (idx == 0xffffffff) + continue; + + if (s <= final_score[TOPK - 1]) + continue; + + int pos = TOPK - 1; + + while (pos > 0 && + s > final_score[pos - 1]) { + --pos; + } + + for (int p = TOPK - 1; + p > pos; + --p) { + + final_score[p] = + final_score[p - 1]; + + final_index[p] = + final_index[p - 1]; + } + + final_score[pos] = s; + final_index[pos] = idx; + } + } + + /* + * Risultato finale della query. + */ + for (uint k = 0; k < TOPK; ++k) { + indices[q * TOPK + k] = + final_index[k]; + + values[q * TOPK + k] = + final_score[k]; + } + } +} diff --git a/kernels/knn_float4_topk.cl.sha256 b/kernels/knn_float4_topk.cl.sha256 new file mode 100644 index 0000000..16dd8cf --- /dev/null +++ b/kernels/knn_float4_topk.cl.sha256 @@ -0,0 +1 @@ +6c4f502438aad3dd1600a00cb7988e0e35e56b2e8e4620d44ffcb0f3610a7226 knn_float4_topk.cl diff --git a/kernels/knn_reference.cl b/kernels/knn_reference.cl new file mode 100644 index 0000000..f7a0b43 --- /dev/null +++ b/kernels/knn_reference.cl @@ -0,0 +1,180 @@ + +#define DIM 128 +#define TOPK 32 + +/* + * Numero di work-item che collaborano per ogni query. + */ +#define LANES_PER_QUERY 8 + +/* + * Ogni work-group contiene 8 work-item. + * + * Ogni work-item calcola un TOPK locale. + * Poi il work-item 0 fonde gli 8 TOPK locali. + */ + +__kernel void knn( + __global const float *db, + __global const float *queries, + __global uint *indices, + __global float *values, + uint n) +{ + uint lane = get_local_id(0); + uint q = get_group_id(0); + + if (q >= 256) + return; + + /* + * TOPK locale per ogni work-item. + */ + float best_score[TOPK]; + uint best_index[TOPK]; + + for (uint k = 0; k < TOPK; ++k) { + best_score[k] = -3.402823466e+38f; + best_index[k] = 0xffffffff; + } + + __global const float *query = + &queries[q * DIM]; + + /* + * Ogni lane analizza una parte del database. + * + * Esempio con 8 lane: + * + * lane 0 -> 0, 8, 16, 24, ... + * lane 1 -> 1, 9, 17, 25, ... + * ... + * lane 7 -> 7, 15, 23, 31, ... + */ + for (uint i = lane; i < n; i += LANES_PER_QUERY) { + + __global const float *v = + &db[i * DIM]; + + float s = 0.0f; + + for (uint j = 0; j < DIM; ++j) + s += v[j] * query[j]; + + if (s <= best_score[TOPK - 1]) + continue; + + int pos = TOPK - 1; + + while (pos > 0 && + s > best_score[pos - 1]) { + --pos; + } + + for (int k = TOPK - 1; k > pos; --k) { + best_score[k] = + best_score[k - 1]; + + best_index[k] = + best_index[k - 1]; + } + + best_score[pos] = s; + best_index[pos] = i; + } + + /* + * TOPK locali nella local memory. + * + * 8 lane × 32 elementi. + */ + __local float local_scores[LANES_PER_QUERY][TOPK]; + __local uint local_indices[LANES_PER_QUERY][TOPK]; + + for (uint k = 0; k < TOPK; ++k) { + local_scores[lane][k] = best_score[k]; + local_indices[lane][k] = best_index[k]; + } + + /* + * Tutti i lane devono aver scritto + * la local memory prima della riduzione. + */ + barrier(CLK_LOCAL_MEM_FENCE); + + /* + * Solo lane 0 esegue il merge finale. + */ + if (lane == 0) { + + float final_score[TOPK]; + uint final_index[TOPK]; + + for (uint k = 0; k < TOPK; ++k) { + final_score[k] = + -3.402823466e+38f; + + final_index[k] = + 0xffffffff; + } + + /* + * Fonde gli 8 TOPK locali. + * + * Al massimo 8 × 32 = 256 candidati. + */ + for (uint l = 0; + l < LANES_PER_QUERY; + ++l) { + + for (uint k = 0; + k < TOPK; + ++k) { + + float s = + local_scores[l][k]; + + uint idx = + local_indices[l][k]; + + if (idx == 0xffffffff) + continue; + + if (s <= final_score[TOPK - 1]) + continue; + + int pos = TOPK - 1; + + while (pos > 0 && + s > final_score[pos - 1]) { + --pos; + } + + for (int p = TOPK - 1; + p > pos; + --p) { + + final_score[p] = + final_score[p - 1]; + + final_index[p] = + final_index[p - 1]; + } + + final_score[pos] = s; + final_index[pos] = idx; + } + } + + /* + * Risultato finale della query. + */ + for (uint k = 0; k < TOPK; ++k) { + indices[q * TOPK + k] = + final_index[k]; + + values[q * TOPK + k] = + final_score[k]; + } + } +} diff --git a/lib/README.md b/lib/README.md new file mode 100644 index 0000000..bcbac85 --- /dev/null +++ b/lib/README.md @@ -0,0 +1,17 @@ +# Runtime RKNN + +Copiare in questa directory la libreria runtime: + +```text +lib/librknnrt.so +``` + +La libreria deve essere compatibile con: + +- il dispositivo Rockchip utilizzato; +- la versione del driver RKNN; +- i modelli RKNN. + +Il runtime è escluso dal repository tramite `.gitignore`. + +Prima della distribuzione verificare le condizioni di licenza. diff --git a/models/README.md b/models/README.md new file mode 100644 index 0000000..2c291b5 --- /dev/null +++ b/models/README.md @@ -0,0 +1,25 @@ +# Modelli RKNN + +Copiare in questa directory i modelli richiesti: + +```text +models/SCRFD_500M_KPS_640.rknn +models/face_recognition_sface_2021dec.rknn +``` + +## SCRFD + +Il modello SCRFD rileva i volti e produce: + +- confidence score; +- bounding box; +- cinque landmark. + +## SFace + +Il modello SFace riceve un volto allineato a 112 x 112 pixel e produce un +embedding di 128 elementi. + +I modelli sono esclusi dal repository tramite `.gitignore`. + +Prima della distribuzione verificare le rispettive licenze. diff --git a/modules/hdbscan/README.md b/modules/hdbscan/README.md new file mode 100644 index 0000000..517e375 --- /dev/null +++ b/modules/hdbscan/README.md @@ -0,0 +1,48 @@ +# HDBSCAN + +Modulo futuro per completare la pipeline di clustering degli embedding +facciali. + +## Pipeline prevista + +```text +kNN graph + | + v +core distance + | + v +mutual reachability + | + v +minimum spanning tree + | + v +single linkage hierarchy + | + v +condensed tree + | + v +cluster stability + | + v +cluster assignment +``` + +## Output previsto + +Per ogni embedding: + +```text +face_index +cluster_id +membership_strength +noise_flag +``` + +Il risultato del clustering resterà distinto dall'identità semantica +assegnata dall'utente. + +Un cluster automatico non rappresenta necessariamente una persona +definitiva e più cluster potranno essere associati alla stessa persona. diff --git a/modules/knn-opencl/README.md b/modules/knn-opencl/README.md new file mode 100644 index 0000000..d2f741a --- /dev/null +++ b/modules/knn-opencl/README.md @@ -0,0 +1,43 @@ +# kNN OpenCL + +Modulo futuro per costruire il grafo dei vicini degli embedding SFace +tramite OpenCL. + +## Input previsto + +```text +embedding float32[N][128] +``` + +Gli embedding saranno già L2-normalizzati. + +Di conseguenza: + +```text +cosine_similarity(a,b) = dot(a,b) +``` + +## Output previsto + +```text +indices[N][32] +scores[N][32] +``` + +## Obiettivo iniziale + +```text +embedding float32[N][128] + | + v +brute-force dot product + | + v +TOPK=32 + | + +-- neighbor index + `-- similarity score +``` + +La prima implementazione privilegerà la correttezza e la validazione +CPU/GPU. diff --git a/modules/knn-opencl/reference/VALIDATION.md b/modules/knn-opencl/reference/VALIDATION.md new file mode 100644 index 0000000..90cbcee --- /dev/null +++ b/modules/knn-opencl/reference/VALIDATION.md @@ -0,0 +1,29 @@ +# Validazione kNN OpenCL Mali-G610 + +## Parametri + +- Database: 350000 embedding +- Query: 256 +- Dimensione: 128 +- TOPK: 32 +- Chunk: 50000 +- GPU: Mali-G610 r0p0 + +## Correttezza + +- Posizioni TOPK confrontate: 8192 +- Posizioni identiche CPU/OpenCL: 8192 +- Exact ratio: 100.000% + +## Prestazioni + +- CPU: 9663.226 ms +- OpenCL: 1373.797 ms +- Speedup: 7.034x +- Throughput: 65220680 vettori/s + +## Variante float4 min-replace + +- Kernel: 1244.576 ms +- Throughput: 71.992 M vettori/s +- Invalid indices: 0 diff --git a/modules/knn-opencl/reference/knn_benchmark.cpp b/modules/knn-opencl/reference/knn_benchmark.cpp new file mode 100644 index 0000000..69cd2e0 --- /dev/null +++ b/modules/knn-opencl/reference/knn_benchmark.cpp @@ -0,0 +1,968 @@ +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +static constexpr uint32_t DIM = 128; +static constexpr uint32_t TOPK = 32; +static constexpr uint32_t QUERIES = 256; + +// Numero massimo di vettori elaborati da un singolo kernel launch. +static constexpr size_t CHUNK_SIZE = 50000; + +static void check_cl(cl_int err, const char *what) { + if (err != CL_SUCCESS) { + std::cerr << "OpenCL error " << err << " at " << what << "\n"; + std::exit(1); + } +} + +static void normalize(float *v) { + double s = 0.0; + + for (uint32_t i = 0; i < DIM; ++i) + s += double(v[i]) * double(v[i]); + + float inv = 1.0f / std::sqrt(float(s)); + + for (uint32_t i = 0; i < DIM; ++i) + v[i] *= inv; +} + +static std::vector make_embeddings(size_t n) { + std::vector x(n * DIM); + + std::mt19937 rng(123456789); + std::normal_distribution d(0.0f, 1.0f); + + for (size_t i = 0; i < x.size(); ++i) + x[i] = d(rng); + + for (size_t i = 0; i < n; ++i) + normalize(&x[i * DIM]); + + return x; +} + +struct Neighbor { + float score; + uint32_t index; +}; + +static void cpu_knn_one( + const std::vector& db, + const float *query, + size_t n, + Neighbor *out) +{ + std::vector all(n); + + for (size_t i = 0; i < n; ++i) { + const float *v = &db[i * DIM]; + + float s = 0.0f; + + for (uint32_t j = 0; j < DIM; ++j) + s += v[j] * query[j]; + + all[i] = {s, static_cast(i)}; + } + + std::nth_element( + all.begin(), + all.begin() + TOPK, + all.end(), + [](const Neighbor& a, const Neighbor& b) { + return a.score > b.score; + }); + + std::sort( + all.begin(), + all.begin() + TOPK, + [](const Neighbor& a, const Neighbor& b) { + return a.score > b.score; + }); + + for (uint32_t k = 0; k < TOPK; ++k) + out[k] = all[k]; +} + +static double benchmark_cpu( + const std::vector& db, + const std::vector& queries, + std::vector& result, + size_t n) +{ + result.resize(QUERIES * TOPK); + + + auto t0 = std::chrono::steady_clock::now(); + + for (uint32_t q = 0; q < QUERIES; ++q) { + cpu_knn_one( + db, + &queries[q * DIM], + n, + &result[q * TOPK]); + } + + auto t1 = std::chrono::steady_clock::now(); + + return std::chrono::duration(t1 - t0).count(); +} + +static std::string cl_string(cl_device_id dev, cl_device_info info) { + size_t size = 0; + + check_cl( + clGetDeviceInfo(dev, info, 0, nullptr, &size), + "clGetDeviceInfo"); + + std::string s(size, '\0'); + + check_cl( + clGetDeviceInfo(dev, info, size, s.data(), nullptr), + "clGetDeviceInfo"); + + while (!s.empty() && s.back() == '\0') + s.pop_back(); + + return s; +} + +static cl_device_id find_mali() { + cl_uint np = 0; + + check_cl( + clGetPlatformIDs(0, nullptr, &np), + "clGetPlatformIDs"); + + std::vector platforms(np); + + check_cl( + clGetPlatformIDs(np, platforms.data(), nullptr), + "clGetPlatformIDs"); + + for (auto p : platforms) { + cl_uint nd = 0; + + if (clGetDeviceIDs( + p, + CL_DEVICE_TYPE_GPU, + 0, + nullptr, + &nd) != CL_SUCCESS) + continue; + + std::vector devices(nd); + + check_cl( + clGetDeviceIDs( + p, + CL_DEVICE_TYPE_GPU, + nd, + devices.data(), + nullptr), + "clGetDeviceIDs"); + + for (auto d : devices) { + std::string name = cl_string(d, CL_DEVICE_NAME); + std::string vendor = cl_string(d, CL_DEVICE_VENDOR); + + std::cout + << "OpenCL GPU: " + << vendor + << " / " + << name + << "\n"; + + if (name.find("Mali-G610") != std::string::npos) + return d; + } + } + + return nullptr; +} + +static void merge_topk( + const std::vector& chunk_indices, + const std::vector& chunk_values, + size_t chunk_offset, + std::vector& result) +{ + /* + * Per ogni query abbiamo TOPK candidati dal chunk corrente. + * + * Il risultato globale viene mantenuto in: + * + * result[q * TOPK ... q * TOPK + TOPK) + * + * Uniamo quindi al massimo 2*TOPK elementi per query. + */ + + for (uint32_t q = 0; q < QUERIES; ++q) { + + std::vector candidates; + candidates.reserve(TOPK * 2); + + Neighbor *dst = &result[q * TOPK]; + + for (uint32_t k = 0; k < TOPK; ++k) { + candidates.push_back(dst[k]); + } + + for (uint32_t k = 0; k < TOPK; ++k) { + uint32_t local_index = + chunk_indices[q * TOPK + k]; + + float score = + chunk_values[q * TOPK + k]; + + if (local_index == 0xffffffffu) + continue; + + candidates.push_back({ + score, + static_cast( + chunk_offset + local_index) + }); + } + + std::sort( + candidates.begin(), + candidates.end(), + [](const Neighbor& a, const Neighbor& b) { + return a.score > b.score; + }); + + for (uint32_t k = 0; k < TOPK; ++k) + dst[k] = candidates[k]; + } +} + +static double benchmark_opencl( + const std::vector& db, + const std::vector& queries, + std::vector& result, + size_t n) +{ + cl_int err; + + cl_device_id dev = find_mali(); + + if (!dev) { + std::cerr << "Mali-G610 not found\n"; + std::exit(1); + } + + cl_context ctx = + clCreateContext( + nullptr, + 1, + &dev, + nullptr, + nullptr, + &err); + + check_cl(err, "clCreateContext"); + + cl_command_queue queue = + clCreateCommandQueue(ctx, dev, 0, &err); + + check_cl(err, "clCreateCommandQueue"); + +const char *source = R"CLC( + +#define DIM 128 +#define TOPK 32 + +/* + * Numero di work-item che collaborano per ogni query. + */ +#define LANES_PER_QUERY 8 + +/* + * Ogni work-group contiene 8 work-item. + * + * Ogni work-item calcola un TOPK locale. + * Poi il work-item 0 fonde gli 8 TOPK locali. + */ + +__kernel void knn( + __global const float *db, + __global const float *queries, + __global uint *indices, + __global float *values, + uint n) +{ + uint lane = get_local_id(0); + uint q = get_group_id(0); + + if (q >= 256) + return; + + /* + * TOPK locale per ogni work-item. + */ + float best_score[TOPK]; + uint best_index[TOPK]; + + for (uint k = 0; k < TOPK; ++k) { + best_score[k] = -3.402823466e+38f; + best_index[k] = 0xffffffff; + } + + __global const float *query = + &queries[q * DIM]; + + /* + * Ogni lane analizza una parte del database. + * + * Esempio con 8 lane: + * + * lane 0 -> 0, 8, 16, 24, ... + * lane 1 -> 1, 9, 17, 25, ... + * ... + * lane 7 -> 7, 15, 23, 31, ... + */ + for (uint i = lane; i < n; i += LANES_PER_QUERY) { + + __global const float *v = + &db[i * DIM]; + + float s = 0.0f; + + for (uint j = 0; j < DIM; ++j) + s += v[j] * query[j]; + + if (s <= best_score[TOPK - 1]) + continue; + + int pos = TOPK - 1; + + while (pos > 0 && + s > best_score[pos - 1]) { + --pos; + } + + for (int k = TOPK - 1; k > pos; --k) { + best_score[k] = + best_score[k - 1]; + + best_index[k] = + best_index[k - 1]; + } + + best_score[pos] = s; + best_index[pos] = i; + } + + /* + * TOPK locali nella local memory. + * + * 8 lane × 32 elementi. + */ + __local float local_scores[LANES_PER_QUERY][TOPK]; + __local uint local_indices[LANES_PER_QUERY][TOPK]; + + for (uint k = 0; k < TOPK; ++k) { + local_scores[lane][k] = best_score[k]; + local_indices[lane][k] = best_index[k]; + } + + /* + * Tutti i lane devono aver scritto + * la local memory prima della riduzione. + */ + barrier(CLK_LOCAL_MEM_FENCE); + + /* + * Solo lane 0 esegue il merge finale. + */ + if (lane == 0) { + + float final_score[TOPK]; + uint final_index[TOPK]; + + for (uint k = 0; k < TOPK; ++k) { + final_score[k] = + -3.402823466e+38f; + + final_index[k] = + 0xffffffff; + } + + /* + * Fonde gli 8 TOPK locali. + * + * Al massimo 8 × 32 = 256 candidati. + */ + for (uint l = 0; + l < LANES_PER_QUERY; + ++l) { + + for (uint k = 0; + k < TOPK; + ++k) { + + float s = + local_scores[l][k]; + + uint idx = + local_indices[l][k]; + + if (idx == 0xffffffff) + continue; + + if (s <= final_score[TOPK - 1]) + continue; + + int pos = TOPK - 1; + + while (pos > 0 && + s > final_score[pos - 1]) { + --pos; + } + + for (int p = TOPK - 1; + p > pos; + --p) { + + final_score[p] = + final_score[p - 1]; + + final_index[p] = + final_index[p - 1]; + } + + final_score[pos] = s; + final_index[pos] = idx; + } + } + + /* + * Risultato finale della query. + */ + for (uint k = 0; k < TOPK; ++k) { + indices[q * TOPK + k] = + final_index[k]; + + values[q * TOPK + k] = + final_score[k]; + } + } +} + +)CLC"; + + cl_program program = + clCreateProgramWithSource( + ctx, + 1, + &source, + nullptr, + &err); + + check_cl(err, "clCreateProgramWithSource"); + + err = clBuildProgram( + program, + 1, + &dev, + "-cl-fast-relaxed-math", + nullptr, + nullptr); + + if (err != CL_SUCCESS) { + + size_t log_size = 0; + + clGetProgramBuildInfo( + program, + dev, + CL_PROGRAM_BUILD_LOG, + 0, + nullptr, + &log_size); + + std::string log(log_size, '\0'); + + clGetProgramBuildInfo( + program, + dev, + CL_PROGRAM_BUILD_LOG, + log_size, + log.data(), + nullptr); + + std::cerr << log << "\n"; + + check_cl(err, "clBuildProgram"); + } + + cl_kernel kernel = + clCreateKernel(program, "knn", &err); + + check_cl(err, "clCreateKernel"); + + /* + * Buffer GPU riutilizzato per ogni chunk. + */ + const size_t chunk_capacity = + std::min(CHUNK_SIZE, n); + + cl_mem db_buf = + clCreateBuffer( + ctx, + CL_MEM_READ_ONLY, + chunk_capacity * DIM * sizeof(float), + nullptr, + &err); + + check_cl(err, "db buffer"); + + cl_mem query_buf = + clCreateBuffer( + ctx, + CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, + queries.size() * sizeof(float), + const_cast(queries.data()), + &err); + + check_cl(err, "query buffer"); + + std::vector indices( + QUERIES * TOPK); + + std::vector values( + QUERIES * TOPK); + + cl_mem index_buf = + clCreateBuffer( + ctx, + CL_MEM_WRITE_ONLY, + indices.size() * sizeof(uint32_t), + nullptr, + &err); + + check_cl(err, "index buffer"); + + cl_mem value_buf = + clCreateBuffer( + ctx, + CL_MEM_WRITE_ONLY, + values.size() * sizeof(float), + nullptr, + &err); + + check_cl(err, "value buffer"); + + check_cl( + clSetKernelArg( + kernel, + 0, + sizeof(db_buf), + &db_buf), + "arg0"); + + check_cl( + clSetKernelArg( + kernel, + 1, + sizeof(query_buf), + &query_buf), + "arg1"); + + check_cl( + clSetKernelArg( + kernel, + 2, + sizeof(index_buf), + &index_buf), + "arg2"); + + check_cl( + clSetKernelArg( + kernel, + 3, + sizeof(value_buf), + &value_buf), + "arg3"); + + /* + * Risultato globale. + * + * Inizializziamo gli score al minimo possibile, + * così il primo chunk può essere mergiato + * senza una gestione speciale. + */ + result.resize(QUERIES * TOPK); + + for (size_t i = 0; i < result.size(); ++i) { + result[i] = { + -3.402823466e+38f, + 0xffffffffu + }; + } + + size_t num_chunks = + (n + CHUNK_SIZE - 1) / CHUNK_SIZE; + + std::cout + << "GPU CHUNK_SIZE : " + << CHUNK_SIZE + << "\n"; + + std::cout + << "GPU CHUNKS : " + << num_chunks + << "\n"; + + size_t global = QUERIES * 8; + size_t local = 8; + +double write_ms = 0.0; +double kernel_ms = 0.0; +double read_ms = 0.0; + + + auto t0 = + std::chrono::steady_clock::now(); + + for (size_t chunk_offset = 0; + chunk_offset < n; + chunk_offset += CHUNK_SIZE) { + + size_t chunk_n = + std::min( + CHUNK_SIZE, + n - chunk_offset); + + const float *chunk_ptr = + &db[chunk_offset * DIM]; + + /* + * Copia del chunk nel buffer GPU. + * + * La coda è in-order, quindi il write, + * kernel e read vengono eseguiti nell'ordine + * corretto. + */ +auto tw0 = std::chrono::steady_clock::now(); + +check_cl( + clEnqueueWriteBuffer( + queue, + db_buf, + CL_TRUE, + 0, + chunk_n * DIM * sizeof(float), + chunk_ptr, + 0, + nullptr, + nullptr), + "write db chunk"); + +auto tw1 = std::chrono::steady_clock::now(); + +write_ms += + std::chrono::duration( + tw1 - tw0).count(); + + uint32_t nn = + static_cast(chunk_n); + + check_cl( + clSetKernelArg( + kernel, + 4, + sizeof(nn), + &nn), + "arg4"); + +auto tk0 = std::chrono::steady_clock::now(); + +check_cl( + clEnqueueNDRangeKernel( + queue, + kernel, + 1, + nullptr, + &global, + &local, + 0, + nullptr, + nullptr), + "clEnqueueNDRangeKernel"); + +check_cl( + clFinish(queue), + "clFinish after kernel"); + +auto tk1 = std::chrono::steady_clock::now(); + +kernel_ms += + std::chrono::duration( + tk1 - tk0).count(); + +auto tr0 = std::chrono::steady_clock::now(); + +check_cl( + clEnqueueReadBuffer( + queue, + index_buf, + CL_TRUE, + 0, + indices.size() * sizeof(uint32_t), + indices.data(), + 0, + nullptr, + nullptr), + "read indices"); + +check_cl( + clEnqueueReadBuffer( + queue, + value_buf, + CL_TRUE, + 0, + values.size() * sizeof(float), + values.data(), + 0, + nullptr, + nullptr), + "read values"); + +auto tr1 = std::chrono::steady_clock::now(); + +read_ms += + std::chrono::duration( + tr1 - tr0).count(); + + /* + * Gli indici prodotti dal kernel sono locali + * al chunk. merge_topk() aggiunge chunk_offset + * per ottenere gli indici globali. + */ + merge_topk( + indices, + values, + chunk_offset, + result); + } + + auto t1 = + std::chrono::steady_clock::now(); + + clReleaseMemObject(value_buf); + clReleaseMemObject(index_buf); + clReleaseMemObject(query_buf); + clReleaseMemObject(db_buf); + clReleaseKernel(kernel); + clReleaseProgram(program); + clReleaseCommandQueue(queue); + clReleaseContext(ctx); + +std::cout + << "GPU WRITE TOTAL : " + << write_ms + << " ms\n"; + +std::cout + << "GPU KERNEL TOTAL: " + << kernel_ms + << " ms\n"; + +std::cout + << "GPU READ TOTAL : " + << read_ms + << " ms\n"; + + return std::chrono::duration( + t1 - t0).count(); +} + +int main(int argc, char **argv) { + + size_t n = 100000; + + if (argc > 1) + n = std::strtoull( + argv[1], + nullptr, + 10); + + std::cout + << "=== BATCH KNN BENCHMARK ===\n"; + + std::cout + << "DATABASE : " + << n + << "\n"; + + std::cout + << "QUERIES : " + << QUERIES + << "\n"; + + std::cout + << "DIM : " + << DIM + << "\n"; + + std::cout + << "TOPK : " + << TOPK + << "\n"; + + std::cout + << "SEED : 123456789\n\n"; + + std::cout + << "Generating database...\n"; + + auto db = + make_embeddings(n); + + std::cout + << "Generating queries...\n"; + + auto queries = + make_embeddings(QUERIES); + + std::vector cpu; + std::vector gpu; + + std::cout + << "\n=== CPU ===\n"; + + double cpu_ms = + benchmark_cpu( + db, + queries, + cpu, + n); + + std::cout + << std::fixed + << std::setprecision(3); + + std::cout + << "time_ms = " + << cpu_ms + << "\n"; + + std::cout + << "q0_top1 = " + << cpu[0].index + << "\n"; + + std::cout + << "\n=== OPENCL MALI ===\n"; + + double gpu_ms = + benchmark_opencl( + db, + queries, + gpu, + n); + + std::cout + << "time_ms = " + << gpu_ms + << "\n"; + + std::cout + << "q0_top1 = " + << gpu[0].index + << "\n"; + + size_t total = + QUERIES * TOPK; + + size_t same = 0; + + for (size_t i = 0; + i < total; + ++i) { + + if (cpu[i].index == gpu[i].index) + ++same; + } + + std::cout + << "exact_topk_positions = " + << same + << "/" + << total + << "\n"; + + std::cout + << "exact_ratio = " + << (double(same) / + double(total)) * 100.0 + << "%\n"; + +for (size_t i = 0; i < total; ++i) { + if (cpu[i].index != gpu[i].index) { + size_t q = i / TOPK; + size_t k = i % TOPK; + + std::cout + << "MISMATCH q=" << q + << " k=" << k + << " CPU index=" << cpu[i].index + << " score=" << cpu[i].score + << " GPU index=" << gpu[i].index + << " score=" << gpu[i].score + << "\n"; + } +} + + + + std::cout + << "\n=== SPEEDUP ===\n"; + + std::cout + << "CPU/OpenCL = " + << cpu_ms / gpu_ms + << "x\n"; + + double pairs = + double(n) * + double(QUERIES); + + double billion = + pairs * + double(DIM) / + 1e9; + + std::cout + << "\n=== WORK ===\n"; + + std::cout + << "similarity_ops = " + << billion + << " billion dot-product elements\n"; + + std::cout + << "\n=== THROUGHPUT ===\n"; + + std::cout + << "database_vectors_per_second = " + << (double(n) * + double(QUERIES)) + / (gpu_ms / 1000.0) + << "\n"; + + return 0; +} diff --git a/modules/knn-opencl/reference/knn_float4_minreplace_validate_benchmark.cpp b/modules/knn-opencl/reference/knn_float4_minreplace_validate_benchmark.cpp new file mode 100644 index 0000000..b0c1a75 --- /dev/null +++ b/modules/knn-opencl/reference/knn_float4_minreplace_validate_benchmark.cpp @@ -0,0 +1,972 @@ +#define CL_TARGET_OPENCL_VERSION 300 +#include + +#include +#include +#include +#include +#include +#include +#include +#include + +static constexpr int DIM = 128; +static constexpr int DIM4 = 32; +static constexpr int TOPK = 32; +static constexpr int QUERIES = 32; +static constexpr int CHUNK_SIZE = 5000; +static constexpr int LANES = 8; +static constexpr int TOTAL_DB = 12000; + +static void check(cl_int err, const char* what) { + if (err != CL_SUCCESS) { + std::cerr << what << " failed: " << err << "\n"; + std::exit(1); + } +} + +static const char* kernel_source = R"CLC( + +#define DIM4 32 +#define TOPK 32 +#define LANES 8 +#define QUERIES 32 + +inline float dot_f4( + __global const float4* db, + __global const float4* queries, + uint q, + uint db_index) +{ + uint db_base = db_index * DIM4; + uint q_base = q * DIM4; + + float4 acc0 = (float4)(0.0f); + float4 acc1 = (float4)(0.0f); + float4 acc2 = (float4)(0.0f); + float4 acc3 = (float4)(0.0f); + + for (uint d = 0; d < DIM4; d += 4) { + acc0 += db[db_base + d + 0] * queries[q_base + d + 0]; + acc1 += db[db_base + d + 1] * queries[q_base + d + 1]; + acc2 += db[db_base + d + 2] * queries[q_base + d + 2]; + acc3 += db[db_base + d + 3] * queries[q_base + d + 3]; + } + + float4 s = acc0 + acc1 + acc2 + acc3; + return s.x + s.y + s.z + s.w; +} + +__kernel void knn_minreplace( + __global const float4* db, + __global const float4* queries, + uint db_count, + __global float* local_scores_out, + __global uint* local_indices_out) +{ + uint gid = get_global_id(0); + uint lane = gid % LANES; + uint q = gid / LANES; + + if (q >= QUERIES) + return; + + float best[TOPK]; + uint best_idx[TOPK]; + + for (uint k = 0; k < TOPK; ++k) { + best[k] = -INFINITY; + best_idx[k] = 0; + } + + float min_score = -INFINITY; + uint min_pos = 0; + + for (uint i = lane; i < db_count; i += LANES) { + + float score = dot_f4(db, queries, q, i); + + if (score > min_score) { + + best[min_pos] = score; + best_idx[min_pos] = i; + + min_score = best[0]; + min_pos = 0; + + for (uint k = 1; k < TOPK; ++k) { + if (best[k] < min_score) { + min_score = best[k]; + min_pos = k; + } + } + } + } + + uint base = (q * LANES + lane) * TOPK; + + for (uint k = 0; k < TOPK; ++k) { + local_scores_out[base + k] = best[k]; + local_indices_out[base + k] = best_idx[k]; + } +} + +__kernel void merge_topk( + __global const float* local_scores, + __global const uint* local_indices, + __global float* final_scores, + __global uint* final_indices) +{ + uint q = get_global_id(0); + + if (q >= QUERIES) + return; + + float best[TOPK]; + uint best_idx[TOPK]; + + for (uint k = 0; k < TOPK; ++k) { + best[k] = -INFINITY; + best_idx[k] = 0; + } + + float min_score = -INFINITY; + uint min_pos = 0; + + for (uint lane = 0; lane < LANES; ++lane) { + + uint base = (q * LANES + lane) * TOPK; + + for (uint k = 0; k < TOPK; ++k) { + + float score = local_scores[base + k]; + + if (score > min_score) { + + best[min_pos] = score; + best_idx[min_pos] = local_indices[base + k]; + + min_score = best[0]; + min_pos = 0; + + for (uint j = 1; j < TOPK; ++j) { + if (best[j] < min_score) { + min_score = best[j]; + min_pos = j; + } + } + } + } + } + + // Sort final TOPK descending, to match the semantic output + // of the original benchmark. + for (uint i = 0; i < TOPK; ++i) { + uint max_pos = i; + + for (uint j = i + 1; j < TOPK; ++j) { + if (best[j] > best[max_pos]) + max_pos = j; + } + + if (max_pos != i) { + float ts = best[i]; + best[i] = best[max_pos]; + best[max_pos] = ts; + + uint ti = best_idx[i]; + best_idx[i] = best_idx[max_pos]; + best_idx[max_pos] = ti; + } + } + + uint out_base = q * TOPK; + + for (uint k = 0; k < TOPK; ++k) { + final_scores[out_base + k] = best[k]; + final_indices[out_base + k] = best_idx[k]; + } +} + +)CLC"; + +int main() { + cl_int err; + + cl_uint num_platforms = 0; + check( + clGetPlatformIDs(0, nullptr, &num_platforms), + "clGetPlatformIDs"); + + std::vector platforms(num_platforms); + + check( + clGetPlatformIDs( + num_platforms, + platforms.data(), + nullptr), + "clGetPlatformIDs(list)"); + + cl_platform_id platform = platforms[0]; + + cl_uint num_devices = 0; + + check( + clGetDeviceIDs( + platform, + CL_DEVICE_TYPE_GPU, + 0, + nullptr, + &num_devices), + "clGetDeviceIDs(count)"); + + std::vector devices(num_devices); + + check( + clGetDeviceIDs( + platform, + CL_DEVICE_TYPE_GPU, + num_devices, + devices.data(), + nullptr), + "clGetDeviceIDs(list)"); + + cl_device_id device = devices[0]; + + char device_name[256] = {}; + + clGetDeviceInfo( + device, + CL_DEVICE_NAME, + sizeof(device_name), + device_name, + nullptr); + + std::cout << "=== FULL KNN FLOAT4 + MIN-REPLACE ===\n"; + std::cout << "Device : " << device_name << "\n"; + std::cout << "DIM : " << DIM << "\n"; + std::cout << "TOPK : " << TOPK << "\n"; + std::cout << "QUERIES : " << QUERIES << "\n"; + std::cout << "TOTAL_DB : " << TOTAL_DB << "\n"; + std::cout << "CHUNK_SIZE : " << CHUNK_SIZE << "\n"; + std::cout << "LANES : " << LANES << "\n\n"; + + cl_context context = + clCreateContext( + nullptr, + 1, + &device, + nullptr, + nullptr, + &err); + + check(err, "clCreateContext"); + + cl_command_queue queue = + clCreateCommandQueue( + context, + device, + CL_QUEUE_PROFILING_ENABLE, + &err); + + check(err, "clCreateCommandQueue"); + + const size_t db_bytes = + static_cast(CHUNK_SIZE) * + DIM4 * + 4 * + sizeof(float); + + const size_t query_bytes = + static_cast(QUERIES) * + DIM4 * + 4 * + sizeof(float); + + const size_t lane_candidates = + static_cast(QUERIES) * + LANES * + TOPK; + + const size_t candidate_bytes = + lane_candidates * sizeof(float); + + const size_t index_bytes = + lane_candidates * sizeof(cl_uint); + + const size_t final_count = + static_cast(QUERIES) * TOPK; + + const size_t final_score_bytes = + final_count * sizeof(float); + + const size_t final_index_bytes = + final_count * sizeof(cl_uint); + + cl_mem db_buf = + clCreateBuffer( + context, + CL_MEM_READ_ONLY, + db_bytes, + nullptr, + &err); + + check(err, "clCreateBuffer(db)"); + + cl_mem query_buf = + clCreateBuffer( + context, + CL_MEM_READ_ONLY, + query_bytes, + nullptr, + &err); + + check(err, "clCreateBuffer(query)"); + + cl_mem local_scores_buf = + clCreateBuffer( + context, + CL_MEM_READ_WRITE, + candidate_bytes, + nullptr, + &err); + + check(err, "clCreateBuffer(local_scores)"); + + cl_mem local_indices_buf = + clCreateBuffer( + context, + CL_MEM_READ_WRITE, + index_bytes, + nullptr, + &err); + + check(err, "clCreateBuffer(local_indices)"); + + cl_mem final_scores_buf = + clCreateBuffer( + context, + CL_MEM_READ_WRITE, + final_score_bytes, + nullptr, + &err); + + check(err, "clCreateBuffer(final_scores)"); + + cl_mem final_indices_buf = + clCreateBuffer( + context, + CL_MEM_READ_WRITE, + final_index_bytes, + nullptr, + &err); + + check(err, "clCreateBuffer(final_indices)"); + + cl_program program = + clCreateProgramWithSource( + context, + 1, + &kernel_source, + nullptr, + &err); + + check(err, "clCreateProgramWithSource"); + + err = clBuildProgram( + program, + 1, + &device, + "-cl-fast-relaxed-math", + nullptr, + nullptr); + + if (err != CL_SUCCESS) { + + size_t log_size = 0; + + clGetProgramBuildInfo( + program, + device, + CL_PROGRAM_BUILD_LOG, + 0, + nullptr, + &log_size); + + std::vector log(log_size + 1); + + clGetProgramBuildInfo( + program, + device, + CL_PROGRAM_BUILD_LOG, + log_size, + log.data(), + nullptr); + + std::cerr << log.data() << "\n"; + + check(err, "clBuildProgram"); + } + + cl_kernel knn = + clCreateKernel( + program, + "knn_minreplace", + &err); + + check(err, "clCreateKernel(knn_minreplace)"); + + cl_kernel merge = + clCreateKernel( + program, + "merge_topk", + &err); + + check(err, "clCreateKernel(merge_topk)"); + + std::vector db( + static_cast(TOTAL_DB) * DIM); + + std::vector queries( + static_cast(QUERIES) * DIM); + + std::mt19937 rng_db(123456789); + std::mt19937 rng_q(987654321); + + std::normal_distribution normal(0.0f, 1.0f); + + for (int i = 0; i < TOTAL_DB; ++i) { + + float norm = 0.0f; + + for (int d = 0; d < DIM; ++d) { + + float x = normal(rng_db); + + db[ + static_cast(i) * DIM + d + ] = x; + + norm += x * x; + } + + norm = 1.0f / std::sqrt(norm); + + for (int d = 0; d < DIM; ++d) + db[ + static_cast(i) * DIM + d + ] *= norm; + } + + for (int q = 0; q < QUERIES; ++q) { + + float norm = 0.0f; + + for (int d = 0; d < DIM; ++d) { + + float x = normal(rng_q); + + queries[ + static_cast(q) * DIM + d + ] = x; + + norm += x * x; + } + + norm = 1.0f / std::sqrt(norm); + + for (int d = 0; d < DIM; ++d) + queries[ + static_cast(q) * DIM + d + ] *= norm; + } + + std::vector final_scores( + final_count); + + std::vector final_indices( + final_count); + + /* + * TOP-K globale persistente fra tutti i chunk. + * + * final_scores/final_indices contengono il risultato + * del chunk corrente. + * + * global_scores/global_indices contengono invece il + * TOP-K complessivo elaborato fino a quel momento. + */ + std::vector global_scores( + final_count, + -INFINITY); + + std::vector global_indices( + final_count, + 0); + + double total_knn_ms = 0.0; + double total_merge_ms = 0.0; + double total_global_merge_ms = 0.0; + + const int chunks = + (TOTAL_DB + CHUNK_SIZE - 1) / + CHUNK_SIZE; + + for (int chunk = 0; chunk < chunks; ++chunk) { + + const int begin = + chunk * CHUNK_SIZE; + + const int count = + std::min( + CHUNK_SIZE, + TOTAL_DB - begin); + + const size_t db_upload_bytes = + static_cast(count) * + DIM4 * + 4 * + sizeof(float); + + check( + clEnqueueWriteBuffer( + queue, + db_buf, + CL_TRUE, + 0, + db_upload_bytes, + db.data() + + static_cast(begin) * DIM, + 0, + nullptr, + nullptr), + "write db"); + + check( + clEnqueueWriteBuffer( + queue, + query_buf, + CL_TRUE, + 0, + query_bytes, + queries.data(), + 0, + nullptr, + nullptr), + "write queries"); + + check( + clSetKernelArg( + knn, + 0, + sizeof(cl_mem), + &db_buf), + "arg 0"); + + check( + clSetKernelArg( + knn, + 1, + sizeof(cl_mem), + &query_buf), + "arg 1"); + + cl_uint db_count = count; + + check( + clSetKernelArg( + knn, + 2, + sizeof(cl_uint), + &db_count), + "arg 2"); + + check( + clSetKernelArg( + knn, + 3, + sizeof(cl_mem), + &local_scores_buf), + "arg 3"); + + check( + clSetKernelArg( + knn, + 4, + sizeof(cl_mem), + &local_indices_buf), + "arg 4"); + + size_t global = + static_cast(QUERIES) * + LANES; + + size_t local = LANES; + + auto t0 = + std::chrono::steady_clock::now(); + + check( + clEnqueueNDRangeKernel( + queue, + knn, + 1, + nullptr, + &global, + &local, + 0, + nullptr, + nullptr), + "enqueue knn"); + + check(clFinish(queue), "finish knn"); + + auto t1 = + std::chrono::steady_clock::now(); + + total_knn_ms += + std::chrono::duration( + t1 - t0).count(); + + check( + clSetKernelArg( + merge, + 0, + sizeof(cl_mem), + &local_scores_buf), + "merge arg 0"); + + check( + clSetKernelArg( + merge, + 1, + sizeof(cl_mem), + &local_indices_buf), + "merge arg 1"); + + check( + clSetKernelArg( + merge, + 2, + sizeof(cl_mem), + &final_scores_buf), + "merge arg 2"); + + check( + clSetKernelArg( + merge, + 3, + sizeof(cl_mem), + &final_indices_buf), + "merge arg 3"); + + size_t merge_global = QUERIES; + + auto tm0 = + std::chrono::steady_clock::now(); + + check( + clEnqueueNDRangeKernel( + queue, + merge, + 1, + nullptr, + &merge_global, + nullptr, + 0, + nullptr, + nullptr), + "enqueue merge"); + + check(clFinish(queue), "finish merge"); + + auto tm1 = + std::chrono::steady_clock::now(); + + total_merge_ms += + std::chrono::duration( + tm1 - tm0).count(); + + check( + clEnqueueReadBuffer( + queue, + final_scores_buf, + CL_TRUE, + 0, + final_score_bytes, + final_scores.data(), + 0, + nullptr, + nullptr), + "read final scores"); + + check( + clEnqueueReadBuffer( + queue, + final_indices_buf, + CL_TRUE, + 0, + final_index_bytes, + final_indices.data(), + 0, + nullptr, + nullptr), + "read final indices"); + + /* + * Merge CPU fra: + * + * TOP-K globale accumulato finora + * TOP-K prodotto dal chunk corrente + * + * Gli indici restituiti dalla GPU sono locali al + * chunk. Li convertiamo quindi in indici globali + * aggiungendo begin. + */ + auto global_merge_start = + std::chrono::steady_clock::now(); + + for (int q = 0; q < QUERIES; ++q) { + + const size_t base = + static_cast(q) * TOPK; + + for (int k = 0; k < TOPK; ++k) { + + const float candidate_score = + final_scores[base + k]; + + const cl_uint candidate_index = + static_cast(begin) + + final_indices[base + k]; + + /* + * Trova il risultato peggiore nel TOP-K + * globale corrente. + */ + size_t worst_pos = base; + float worst_score = + global_scores[base]; + + for (int j = 1; j < TOPK; ++j) { + + const size_t pos = + base + + static_cast(j); + + if (global_scores[pos] < + worst_score) + { + worst_score = + global_scores[pos]; + + worst_pos = pos; + } + } + + /* + * Sostituzione del minimo. + */ + if (candidate_score > worst_score) { + + global_scores[worst_pos] = + candidate_score; + + global_indices[worst_pos] = + candidate_index; + } + } + } + + auto global_merge_end = + std::chrono::steady_clock::now(); + + total_global_merge_ms += + std::chrono::duration< + double, + std::milli>( + global_merge_end - + global_merge_start).count(); + } + + /* + * Ordina il TOP-K globale di ogni query in ordine + * decrescente di similarita. + */ + for (int q = 0; q < QUERIES; ++q) { + + const size_t base = + static_cast(q) * TOPK; + + for (int i = 0; i < TOPK; ++i) { + + int max_pos = i; + + for (int j = i + 1; + j < TOPK; + ++j) + { + if (global_scores[ + base + + static_cast(j) + ] > + global_scores[ + base + + static_cast(max_pos) + ]) + { + max_pos = j; + } + } + + if (max_pos != i) { + + std::swap( + global_scores[ + base + + static_cast(i) + ], + global_scores[ + base + + static_cast(max_pos) + ]); + + std::swap( + global_indices[ + base + + static_cast(i) + ], + global_indices[ + base + + static_cast(max_pos) + ]); + } + } + } + + /* + * Verifica intervallo indici globali. + */ + size_t invalid_indices = 0; + + for (cl_uint index : global_indices) { + if (index >= + static_cast(TOTAL_DB)) + { + ++invalid_indices; + } + } + + double checksum_scores = 0.0; + double checksum_indices = 0.0; + + for (size_t i = 0; i < final_count; ++i) { + checksum_scores += global_scores[i]; + checksum_indices += global_indices[i]; + } + + double vectors = + static_cast(TOTAL_DB) * + QUERIES; + + double vector_rate = + vectors / + total_knn_ms / + 1000.0; + + double gflops = + vector_rate * + DIM * + 2.0 / + 1000.0; + + double db_traffic = + vector_rate * + DIM * + sizeof(float) / + 1000.0; + + std::cout << "\n=== RESULT ===\n"; + + std::cout << "KNN kernel time : " + << std::fixed + << std::setprecision(3) + << total_knn_ms + << " ms\n"; + + std::cout << "Merge time : " + << std::fixed + << std::setprecision(3) + << total_merge_ms + << " ms\n"; + + std::cout << "Global CPU merge: " + << std::fixed + << std::setprecision(3) + << total_global_merge_ms + << " ms\n"; + + std::cout << "Total GPU time : " + << std::fixed + << std::setprecision(3) + << total_knn_ms + total_merge_ms + << " ms\n"; + + std::cout << "Search core time: " + << std::fixed + << std::setprecision(3) + << total_knn_ms + + total_merge_ms + + total_global_merge_ms + << " ms\n"; + + std::cout << "Invalid indices : " + << invalid_indices + << "\n"; + + std::cout << "Vector rate : " + << std::fixed + << std::setprecision(3) + << vector_rate + << " M vector/s\n"; + + std::cout << "Dot rate : " + << std::fixed + << std::setprecision(3) + << gflops + << " GFLOP/s\n"; + + std::cout << "DB traffic : " + << std::fixed + << std::setprecision(3) + << db_traffic + << " GB/s effective\n"; + + std::cout << "Score checksum : " + << std::fixed + << std::setprecision(3) + << checksum_scores + << "\n"; + + std::cout << "Index checksum : " + << std::fixed + << std::setprecision(3) + << checksum_indices + << "\n"; + + clReleaseKernel(knn); + clReleaseKernel(merge); + clReleaseProgram(program); + + clReleaseMemObject(db_buf); + clReleaseMemObject(query_buf); + clReleaseMemObject(local_scores_buf); + clReleaseMemObject(local_indices_buf); + clReleaseMemObject(final_scores_buf); + clReleaseMemObject(final_indices_buf); + + clReleaseCommandQueue(queue); + clReleaseContext(context); + + return 0; +} diff --git a/modules/knn-opencl/reference/validation-minreplace-350k.txt b/modules/knn-opencl/reference/validation-minreplace-350k.txt new file mode 100644 index 0000000..bb7c76a --- /dev/null +++ b/modules/knn-opencl/reference/validation-minreplace-350k.txt @@ -0,0 +1,23 @@ +arm_release_ver: g24p0-00eac0, rk_so_ver: 10 +=== FULL KNN FLOAT4 + MIN-REPLACE === +Device : Mali-G610 r0p0 +DIM : 128 +TOPK : 32 +QUERIES : 256 +TOTAL_DB : 350000 +CHUNK_SIZE : 50000 +LANES : 8 + + +=== RESULT === +KNN kernel time : 1244.576 ms +Merge time : 8.022 ms +Global CPU merge: 14.039 ms +Total GPU time : 1252.599 ms +Search core time: 1266.638 ms +Invalid indices : 0 +Vector rate : 71.992 M vector/s +Dot rate : 18.430 GFLOP/s +DB traffic : 36.860 GB/s effective +Score checksum : 2804.191 +Index checksum : 1425305049.000 diff --git a/modules/knn-opencl/reference/validation-opencl-350k.txt b/modules/knn-opencl/reference/validation-opencl-350k.txt new file mode 100644 index 0000000..69e3df8 --- /dev/null +++ b/modules/knn-opencl/reference/validation-opencl-350k.txt @@ -0,0 +1,35 @@ +=== BATCH KNN BENCHMARK === +DATABASE : 350000 +QUERIES : 256 +DIM : 128 +TOPK : 32 +SEED : 123456789 + +Generating database... +Generating queries... + +=== CPU === +time_ms = 9663.226 +q0_top1 = 0 + +=== OPENCL MALI === +arm_release_ver: g24p0-00eac0, rk_so_ver: 10 +OpenCL GPU: ARM / Mali-G610 r0p0 +GPU CHUNK_SIZE : 50000 +GPU CHUNKS : 7 +GPU WRITE TOTAL : 48.242 ms +GPU KERNEL TOTAL: 1318.590 ms +GPU READ TOTAL : 0.712 ms +time_ms = 1373.797 +q0_top1 = 0 +exact_topk_positions = 8192/8192 +exact_ratio = 100.000% + +=== SPEEDUP === +CPU/OpenCL = 7.034x + +=== WORK === +similarity_ops = 11.469 billion dot-product elements + +=== THROUGHPUT === +database_vectors_per_second = 65220680.078 diff --git a/modules/mst/README.md b/modules/mst/README.md new file mode 100644 index 0000000..41e743f --- /dev/null +++ b/modules/mst/README.md @@ -0,0 +1,29 @@ +# Minimum Spanning Tree + +Modulo futuro per costruire il minimum spanning tree del grafo di mutual +reachability. + +## Input previsto + +Un grafo pesato: + +```text +source +destination +mutual_reachability_weight +``` + +## Output previsto + +Per un dataset con `N` vertici connessi, l'MST conterrà: + +```text +N - 1 archi +``` + +La prima implementazione privilegerà: + +- correttezza; +- determinismo; +- confronto con una versione CPU di riferimento; +- verificabilità dei risultati. diff --git a/modules/mutual-reachability/README.md b/modules/mutual-reachability/README.md new file mode 100644 index 0000000..9157c68 --- /dev/null +++ b/modules/mutual-reachability/README.md @@ -0,0 +1,37 @@ +# Mutual Reachability + +Modulo futuro per convertire il grafo kNN in un grafo pesato utilizzabile +da HDBSCAN. + +## Core distance + +Con `min_samples=32`, la core distance di un punto corrisponde alla +distanza dal suo trentaduesimo vicino. + +## Mutual reachability distance + +```text +mrd(i,j) = + max( + core_distance(i), + core_distance(j), + distance(i,j) + ) +``` + +## Input previsto + +```text +indices[N][32] +scores[N][32] +``` + +## Output previsto + +Un insieme di archi candidati: + +```text +source +destination +mutual_reachability_weight +``` diff --git a/src/face_engine.cc b/src/face_engine.cc new file mode 100644 index 0000000..a4a6a10 --- /dev/null +++ b/src/face_engine.cc @@ -0,0 +1,855 @@ +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define STB_IMAGE_IMPLEMENTATION +#include "stb_image.h" + +#include "rknn_api.h" +#include "face_engine.h" + +static constexpr int SCRFD_SIZE = 640; +static constexpr int SFACE_SIZE = 112; +static constexpr int EMBED_DIM = 128; + +static uint16_t float_to_fp16(float value) +{ + uint32_t bits; + std::memcpy(&bits, &value, sizeof(bits)); + + uint32_t sign = (bits >> 31) & 1; + int exp = ((bits >> 23) & 0xff) - 127; + uint32_t mant = bits & 0x7fffff; + + if (exp == 128) { + return (sign << 15) | 0x7c00; + } + + if (exp > 15) { + return (sign << 15) | 0x7c00; + } + + if (exp < -14) { + if (exp < -24) + return sign << 15; + + mant |= 0x800000; + int shift = -exp - 14; + uint16_t m = mant >> (shift + 13); + + return (sign << 15) | m; + } + + uint16_t h_exp = (uint16_t)(exp + 15); + uint16_t h_mant = (uint16_t)(mant >> 13); + + return (sign << 15) | (h_exp << 10) | h_mant; +} + +static float fp16_to_float(uint16_t h) +{ + uint32_t sign = (h >> 15) & 1; + uint32_t exp = (h >> 10) & 0x1f; + uint32_t mant = h & 0x3ff; + + uint32_t bits; + + if (exp == 0) { + if (mant == 0) { + bits = sign << 31; + } else { + float v = mant / 1024.0f; + v = std::ldexp(v, -14); + return sign ? -v : v; + } + } else if (exp == 31) { + bits = (sign << 31) | 0x7f800000 | (mant << 13); + } else { + uint32_t fexp = exp - 15 + 127; + bits = (sign << 31) | (fexp << 23) | (mant << 13); + } + + float v; + std::memcpy(&v, &bits, sizeof(v)); + return v; +} + +static float iou(const Face& a, const Face& b) +{ + float xx1 = std::max(a.x1, b.x1); + float yy1 = std::max(a.y1, b.y1); + float xx2 = std::min(a.x2, b.x2); + float yy2 = std::min(a.y2, b.y2); + + float w = std::max(0.0f, xx2 - xx1); + float h = std::max(0.0f, yy2 - yy1); + float inter = w * h; + + float area_a = std::max(0.0f, a.x2-a.x1) * + std::max(0.0f, a.y2-a.y1); + + float area_b = std::max(0.0f, b.x2-b.x1) * + std::max(0.0f, b.y2-b.y1); + + return inter / (area_a + area_b - inter + 1e-6f); +} + +static void nms(std::vector& faces, float threshold) +{ + std::sort(faces.begin(), faces.end(), + [](const Face& a, const Face& b) { + return a.score > b.score; + }); + + std::vector result; + + for (const auto& f : faces) { + bool keep = true; + + for (const auto& r : result) { + if (iou(f, r) > threshold) { + keep = false; + break; + } + } + + if (keep) + result.push_back(f); + } + + faces.swap(result); +} + +struct Image { + int w = 0; + int h = 0; + std::vector rgb; +}; + +static bool normalize_rotation( + int rotation_degrees, + int& normalized_rotation) +{ + int value = rotation_degrees % 360; + + if (value < 0) + value += 360; + + if ( + value != 0 + && value != 90 + && value != 180 + && value != 270 + ) { + return false; + } + + normalized_rotation = value; + return true; +} + +static Image rotate_image( + const Image& source, + int rotation_degrees) +{ + if (rotation_degrees == 0) + return source; + + Image destination; + + if ( + rotation_degrees == 90 + || rotation_degrees == 270 + ) { + destination.w = source.h; + destination.h = source.w; + } else { + destination.w = source.w; + destination.h = source.h; + } + + destination.rgb.resize( + static_cast(destination.w) + * static_cast(destination.h) + * 3); + + for (int source_y = 0; source_y < source.h; ++source_y) { + for (int source_x = 0; source_x < source.w; ++source_x) { + int destination_x = 0; + int destination_y = 0; + + switch (rotation_degrees) { + case 90: + destination_x = source.h - 1 - source_y; + destination_y = source_x; + break; + + case 180: + destination_x = source.w - 1 - source_x; + destination_y = source.h - 1 - source_y; + break; + + case 270: + destination_x = source_y; + destination_y = source.w - 1 - source_x; + break; + + default: + destination_x = source_x; + destination_y = source_y; + break; + } + + const std::size_t source_offset = + ( + static_cast(source_y) + * static_cast(source.w) + + static_cast(source_x) + ) * 3; + + const std::size_t destination_offset = + ( + static_cast(destination_y) + * static_cast(destination.w) + + static_cast(destination_x) + ) * 3; + + destination.rgb[destination_offset + 0] = + source.rgb[source_offset + 0]; + + destination.rgb[destination_offset + 1] = + source.rgb[source_offset + 1]; + + destination.rgb[destination_offset + 2] = + source.rgb[source_offset + 2]; + } + } + + return destination; +} + +static bool load_image(const std::string& path, Image& img) +{ + int c = 0; + + unsigned char* p = + stbi_load(path.c_str(), &img.w, &img.h, &c, 3); + + if (!p) { + std::cerr << "Errore caricamento: " << path << "\n"; + return false; + } + + img.rgb.assign(p, p + img.w * img.h * 3); + stbi_image_free(p); + + return true; +} + +static void resize_bilinear_rgb( + const Image& src, + std::vector& dst, + int dw, + int dh) +{ + dst.resize(dw * dh * 3); + + float sx = (float)src.w / dw; + float sy = (float)src.h / dh; + + for (int y = 0; y < dh; ++y) { + float fy = (y + 0.5f) * sy - 0.5f; + int y0 = (int)std::floor(fy); + float wy = fy - y0; + + if (y0 < 0) { + y0 = 0; + wy = 0; + } + + int y1 = std::min(y0 + 1, src.h - 1); + + for (int x = 0; x < dw; ++x) { + float fx = (x + 0.5f) * sx - 0.5f; + int x0 = (int)std::floor(fx); + float wx = fx - x0; + + if (x0 < 0) { + x0 = 0; + wx = 0; + } + + int x1 = std::min(x0 + 1, src.w - 1); + + for (int c = 0; c < 3; ++c) { + float p00 = src.rgb[(y0*src.w+x0)*3+c]; + float p01 = src.rgb[(y0*src.w+x1)*3+c]; + float p10 = src.rgb[(y1*src.w+x0)*3+c]; + float p11 = src.rgb[(y1*src.w+x1)*3+c]; + + float v = + p00 * (1-wx) * (1-wy) + + p01 * wx * (1-wy) + + p10 * (1-wx) * wy + + p11 * wx * wy; + + dst[(y*dw+x)*3+c] = + (uint8_t)std::clamp((int)std::round(v), 0, 255); + } + } + } +} + +class Scrfd { +public: + rknn_context ctx = 0; + + bool init(const char* model) + { + FILE* fp = fopen(model, "rb"); + if (!fp) { + perror(model); + return false; + } + + fseek(fp, 0, SEEK_END); + size_t size = ftell(fp); + fseek(fp, 0, SEEK_SET); + + std::vector data(size); + fread(data.data(), 1, size, fp); + fclose(fp); + + int ret = rknn_init(&ctx, data.data(), size, 0, nullptr); + if (ret != 0) { + std::cerr << "rknn_init SCRFD failed: " << ret << "\n"; + return false; + } + + return true; + } + + std::vector detect(const Image& image) + { + float scale = + std::min(640.0f / image.w, 640.0f / image.h); + + int rw = (int)std::round(image.w * scale); + int rh = (int)std::round(image.h * scale); + + std::vector resized; + resize_bilinear_rgb(image, resized, rw, rh); + + std::vector input( + SCRFD_SIZE * SCRFD_SIZE * 3); + + for (int y = 0; y < rh; ++y) { + for (int x = 0; x < rw; ++x) { + int dst = (y * SCRFD_SIZE + x) * 3; + int src = (y * rw + x) * 3; + + for (int c = 0; c < 3; ++c) { + float v = + ((float)resized[src+c] - 127.5f) / 128.0f; + + input[dst+c] = float_to_fp16(v); + } + } + } + + rknn_input in{}; + in.index = 0; + in.type = RKNN_TENSOR_FLOAT16; + in.fmt = RKNN_TENSOR_NHWC; + in.size = input.size() * sizeof(uint16_t); + in.buf = input.data(); + in.pass_through = 1; + + int ret = rknn_inputs_set(ctx, 1, &in); + if (ret != 0) { + std::cerr << "SCRFD inputs_set failed\n"; + return {}; + } + + ret = rknn_run(ctx, nullptr); + if (ret != 0) { + std::cerr << "SCRFD run failed\n"; + return {}; + } + + rknn_output outputs[9]{}; + + for (int i = 0; i < 9; ++i) { + outputs[i].want_float = 0; + } + + ret = rknn_outputs_get(ctx, 9, outputs, nullptr); + if (ret != 0) { + std::cerr << "SCRFD outputs_get failed\n"; + return {}; + } + + const int strides[3] = {8, 16, 32}; + const int counts[3] = {12800, 3200, 800}; + + std::vector faces; + + const float score_threshold = 0.50f; + + for (int level = 0; level < 3; ++level) { + const int stride = strides[level]; + const int count = counts[level]; + + const uint16_t* scores = + (const uint16_t*)outputs[level].buf; + + const uint16_t* bbox = + (const uint16_t*)outputs[3 + level].buf; + + const uint16_t* kps = + (const uint16_t*)outputs[6 + level].buf; + + int feat_w = SCRFD_SIZE / stride; + + for (int i = 0; i < count; ++i) { + float score = fp16_to_float(scores[i]); + + if (score < score_threshold) + continue; + + int anchor = i % 2; + int p = i / 2; + + int gx = p % feat_w; + int gy = p / feat_w; + + float cx = gx * stride; + float cy = gy * stride; + + /* + * AnchorGenerator con due anchor: + * centers duplicati sullo stesso punto. + */ + + (void)anchor; + + float l = fp16_to_float(bbox[i*4+0]) * stride; + float t = fp16_to_float(bbox[i*4+1]) * stride; + float r = fp16_to_float(bbox[i*4+2]) * stride; + float b = fp16_to_float(bbox[i*4+3]) * stride; + + Face f; + f.score = score; + + f.x1 = (cx - l) / scale; + f.y1 = (cy - t) / scale; + f.x2 = (cx + r) / scale; + f.y2 = (cy + b) / scale; + + for (int j = 0; j < 5; ++j) { + float x = + cx + fp16_to_float(kps[i*10+j*2]) + * stride; + + float y = + cy + fp16_to_float(kps[i*10+j*2+1]) + * stride; + + f.kps[j].x = x / scale; + f.kps[j].y = y / scale; + } + + f.x1 = std::clamp(f.x1, 0.0f, (float)image.w); + f.y1 = std::clamp(f.y1, 0.0f, (float)image.h); + f.x2 = std::clamp(f.x2, 0.0f, (float)image.w); + f.y2 = std::clamp(f.y2, 0.0f, (float)image.h); + + faces.push_back(f); + } + } + + rknn_outputs_release(ctx, 9, outputs); + + nms(faces, 0.45f); + + return faces; + } + + ~Scrfd() + { + if (ctx) + rknn_destroy(ctx); + } +}; + +static const Point sface_template[5] = { + {38.2946f, 51.6963f}, + {73.5318f, 51.5014f}, + {56.0252f, 71.7366f}, + {41.5493f, 92.3655f}, + {70.7299f, 92.2041f} +}; + +/* + * Trova la trasformazione similarity: + * + * x' = a*x - b*y + tx + * y' = b*x + a*y + ty + */ +static void similarity_transform( + const Point src[5], + float& a, + float& b, + float& tx, + float& ty) +{ + double sx = 0, sy = 0; + double dx = 0, dy = 0; + + for (int i = 0; i < 5; ++i) { + sx += src[i].x; + sy += src[i].y; + dx += sface_template[i].x; + dy += sface_template[i].y; + } + + sx /= 5; + sy /= 5; + dx /= 5; + dy /= 5; + + double num_a = 0; + double num_b = 0; + double den = 0; + + for (int i = 0; i < 5; ++i) { + double x = src[i].x - sx; + double y = src[i].y - sy; + + double X = sface_template[i].x - dx; + double Y = sface_template[i].y - dy; + + num_a += x*X + y*Y; + num_b += x*Y - y*X; + den += x*x + y*y; + } + + a = (float)(num_a / den); + b = (float)(num_b / den); + + tx = (float)(dx - a*sx + b*sy); + ty = (float)(dy - b*sx - a*sy); +} + +static std::vector align_face( + const Image& image, + const Face& face) +{ + float a, b, tx, ty; + + similarity_transform(face.kps, a, b, tx, ty); + + /* + * Inversa della similarity transform. + * + * src = A^-1 * (dst - t) + */ + float denom = a*a + b*b; + + std::vector output( + SFACE_SIZE * SFACE_SIZE * 3); + + for (int y = 0; y < SFACE_SIZE; ++y) { + for (int x = 0; x < SFACE_SIZE; ++x) { + + float X = x - tx; + float Y = y - ty; + + float sx = + (a*X + b*Y) / denom; + + float sy = + (-b*X + a*Y) / denom; + + int x0 = (int)std::floor(sx); + int y0 = (int)std::floor(sy); + + float wx = sx - x0; + float wy = sy - y0; + + uint8_t rgb[3] = {0,0,0}; + + if (x0 >= 0 && + y0 >= 0 && + x0 + 1 < image.w && + y0 + 1 < image.h) { + + int x1 = x0 + 1; + int y1 = y0 + 1; + + for (int c = 0; c < 3; ++c) { + float p00 = + image.rgb[(y0*image.w+x0)*3+c]; + + float p01 = + image.rgb[(y0*image.w+x1)*3+c]; + + float p10 = + image.rgb[(y1*image.w+x0)*3+c]; + + float p11 = + image.rgb[(y1*image.w+x1)*3+c]; + + float v = + p00*(1-wx)*(1-wy) + + p01*wx*(1-wy) + + p10*(1-wx)*wy + + p11*wx*wy; + + rgb[c] = + (uint8_t)std::clamp( + (int)std::round(v), 0, 255); + } + } + + int idx = (y*SFACE_SIZE+x)*3; + + /* + * SFace OpenCV: + * blobFromImage(..., swapRB=true, scalefactor=1) + * + * Noi partiamo già da RGB, quindi non facciamo + * nessun ulteriore swap e nessuna normalizzazione. + */ + output[idx+0] = float_to_fp16(rgb[0]); + output[idx+1] = float_to_fp16(rgb[1]); + output[idx+2] = float_to_fp16(rgb[2]); + } + } + + return output; +} + +class SFace { +public: + rknn_context ctx = 0; + + bool init(const char* model) + { + FILE* fp = fopen(model, "rb"); + if (!fp) { + perror(model); + return false; + } + + fseek(fp, 0, SEEK_END); + size_t size = ftell(fp); + fseek(fp, 0, SEEK_SET); + + std::vector data(size); + fread(data.data(), 1, size, fp); + fclose(fp); + + int ret = rknn_init(&ctx, data.data(), size, 0, nullptr); + + if (ret != 0) { + std::cerr << "rknn_init SFace failed: " + << ret << "\n"; + return false; + } + + return true; + } + + std::vector feature( + const std::vector& input) + { + rknn_input in{}; + + in.index = 0; + in.type = RKNN_TENSOR_FLOAT16; + in.fmt = RKNN_TENSOR_NHWC; + in.size = input.size() * sizeof(uint16_t); + in.buf = (void*)input.data(); + in.pass_through = 1; + + int ret = rknn_inputs_set(ctx, 1, &in); + + if (ret != 0) { + std::cerr << "SFace inputs_set failed\n"; + return {}; + } + + ret = rknn_run(ctx, nullptr); + + if (ret != 0) { + std::cerr << "SFace run failed\n"; + return {}; + } + + rknn_output output{}; + + output.want_float = 0; + + ret = rknn_outputs_get( + ctx, 1, &output, nullptr); + + if (ret != 0) { + std::cerr << "SFace outputs_get failed\n"; + return {}; + } + + const uint16_t* p = + (const uint16_t*)output.buf; + + std::vector emb(EMBED_DIM); + + for (int i = 0; i < EMBED_DIM; ++i) + emb[i] = fp16_to_float(p[i]); + + rknn_outputs_release(ctx, 1, &output); + + /* + * L2 normalization. + */ + double norm = 0; + + for (float v : emb) + norm += (double)v * v; + + norm = std::sqrt(norm); + + if (norm > 0) { + for (float& v : emb) + v /= (float)norm; + } + + return emb; + } + + ~SFace() + { + if (ctx) + rknn_destroy(ctx); + } +}; + +static std::vector process_image( + const Image& image, + Scrfd& detector, + SFace& recognizer) +{ + auto faces = detector.detect(image); + + for (auto& face : faces) { + auto aligned = align_face(image, face); + face.embedding = + recognizer.feature(aligned); + } + + return faces; +} + +struct FaceEngine::Impl { + Scrfd detector; + SFace recognizer; + bool initialized = false; +}; + +FaceEngine::FaceEngine() + : impl_(new Impl()) +{ +} + +FaceEngine::~FaceEngine() +{ + delete impl_; +} + +bool FaceEngine::init( + const std::string& scrfd_model_path, + const std::string& sface_model_path) +{ + if (impl_->initialized) + return true; + + if (!impl_->detector.init(scrfd_model_path.c_str())) + return false; + + if (!impl_->recognizer.init(sface_model_path.c_str())) + return false; + + impl_->initialized = true; + return true; +} + +bool FaceEngine::process_image( + const std::string& image_path, + std::vector& faces, + int& image_width, + int& image_height, + int rotation_degrees) +{ + faces.clear(); + image_width = 0; + image_height = 0; + + if (!impl_->initialized) { + std::cerr << "FaceEngine non inizializzato\n"; + return false; + } + + int normalized_rotation = 0; + + if (!normalize_rotation( + rotation_degrees, + normalized_rotation)) { + std::cerr + << "Rotazione non valida: " + << rotation_degrees + << ". Valori ammessi: 0, 90, 180, 270\n"; + + return false; + } + + Image source_image; + + if (!load_image(image_path, source_image)) + return false; + + Image oriented_image = rotate_image( + source_image, + normalized_rotation); + + image_width = oriented_image.w; + image_height = oriented_image.h; + + faces = ::process_image( + oriented_image, + impl_->detector, + impl_->recognizer); + + for (const Face& face : faces) { + if (face.embedding.size() != EMBED_DIM) { + std::cerr + << "Embedding non valido per " + << image_path + << ": attesi " + << EMBED_DIM + << " valori, trovati " + << face.embedding.size() + << "\n"; + + faces.clear(); + return false; + } + } + + return true; +} diff --git a/src/face_knn.cc b/src/face_knn.cc new file mode 100644 index 0000000..1b58ed8 --- /dev/null +++ b/src/face_knn.cc @@ -0,0 +1,781 @@ +#include "knn_opencl.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include + +namespace { + + +class ScopedStdoutToStderr { +public: + ScopedStdoutToStderr() + { + std::cout.flush(); + std::fflush(stdout); + + saved_stdout_ = + ::dup(STDOUT_FILENO); + + if (saved_stdout_ < 0) { + throw std::runtime_error( + "Impossibile salvare stdout" + ); + } + + if ( + ::dup2( + STDERR_FILENO, + STDOUT_FILENO + ) < 0 + ) { + ::close(saved_stdout_); + saved_stdout_ = -1; + + throw std::runtime_error( + "Impossibile deviare stdout" + ); + } + } + + ~ScopedStdoutToStderr() + { + if (saved_stdout_ < 0) { + return; + } + + std::cout.flush(); + std::fflush(stdout); + + ::dup2( + saved_stdout_, + STDOUT_FILENO + ); + + ::close(saved_stdout_); + } + + ScopedStdoutToStderr( + const ScopedStdoutToStderr& + ) = delete; + + ScopedStdoutToStderr& operator=( + const ScopedStdoutToStderr& + ) = delete; + +private: + int saved_stdout_ = -1; +}; + + +struct Arguments { + std::string queries_path; + + std::string database_path; + + std::string self_map_path; + + std::string backend = "cpu"; + + std::size_t query_count = 0; + + std::size_t database_count = 0; + + std::uint32_t dimension = 128; + + std::uint32_t top_k = 1; + + float threshold = 0.50f; +}; + + +void print_usage( + const char* program) +{ + std::cerr + << "Uso:\n" + << " " + << program + << " --queries FILE" + << " --database FILE" + << " --query-count N" + << " --database-count N" + << " --dimension N" + << " --top-k N" + << " --backend cpu" + << " [--threshold SCORE]" + << " [--self-map FILE]" + << "\n"; +} + + +std::string require_value( + int argc, + char** argv, + int& index) +{ + if (index + 1 >= argc) { + throw std::invalid_argument( + std::string( + "Valore mancante per " + ) + + argv[index] + ); + } + + index += 1; + + return argv[index]; +} + + +std::size_t parse_size( + const std::string& text, + const char* name) +{ + std::size_t consumed = 0; + + const auto value = + std::stoull( + text, + &consumed, + 10 + ); + + if ( + consumed != text.size() + || value == 0 + ) { + throw std::invalid_argument( + std::string(name) + + " non valido: " + + text + ); + } + + return static_cast( + value + ); +} + + +std::uint32_t parse_u32( + const std::string& text, + const char* name) +{ + const std::size_t value = + parse_size( + text, + name + ); + + if ( + value + > std::numeric_limits< + std::uint32_t + >::max() + ) { + throw std::invalid_argument( + std::string(name) + + " supera uint32" + ); + } + + return static_cast( + value + ); +} + + +float parse_float( + const std::string& text, + const char* name) +{ + std::size_t consumed = 0; + + const float value = + std::stof( + text, + &consumed + ); + + if (consumed != text.size()) { + throw std::invalid_argument( + std::string(name) + + " non valido: " + + text + ); + } + + return value; +} + + +Arguments parse_arguments( + int argc, + char** argv) +{ + Arguments arguments; + + for ( + int index = 1; + index < argc; + ++index + ) { + const std::string option = + argv[index]; + + if ( + option == "--help" + || option == "-h" + ) { + print_usage(argv[0]); + std::exit(0); + } + + if (option == "--queries") { + arguments.queries_path = + require_value( + argc, + argv, + index + ); + } else if ( + option == "--database" + ) { + arguments.database_path = + require_value( + argc, + argv, + index + ); + } else if ( + option == "--self-map" + ) { + arguments.self_map_path = + require_value( + argc, + argv, + index + ); + } else if ( + option == "--backend" + ) { + arguments.backend = + require_value( + argc, + argv, + index + ); + } else if ( + option == "--query-count" + ) { + arguments.query_count = + parse_size( + require_value( + argc, + argv, + index + ), + "query-count" + ); + } else if ( + option == "--database-count" + ) { + arguments.database_count = + parse_size( + require_value( + argc, + argv, + index + ), + "database-count" + ); + } else if ( + option == "--dimension" + ) { + arguments.dimension = + parse_u32( + require_value( + argc, + argv, + index + ), + "dimension" + ); + } else if ( + option == "--top-k" + ) { + arguments.top_k = + parse_u32( + require_value( + argc, + argv, + index + ), + "top-k" + ); + } else if ( + option == "--threshold" + ) { + arguments.threshold = + parse_float( + require_value( + argc, + argv, + index + ), + "threshold" + ); + } else { + throw std::invalid_argument( + "Opzione sconosciuta: " + + option + ); + } + } + + if ( + arguments.queries_path.empty() + || arguments.database_path.empty() + || arguments.query_count == 0 + || arguments.database_count == 0 + ) { + throw std::invalid_argument( + "Argomenti obbligatori mancanti" + ); + } + + if ( + arguments.backend != "cpu" + && arguments.backend != "opencl" + ) { + throw std::invalid_argument( + "Backend non supportato: " + + arguments.backend + ); + } + + return arguments; +} + + +std::vector read_float_file( + const std::string& path, + std::size_t expected_values) +{ + std::ifstream input( + path, + std::ios::binary + ); + + if (!input) { + throw std::runtime_error( + "Impossibile aprire: " + + path + ); + } + + input.seekg( + 0, + std::ios::end + ); + + const std::streamoff byte_count = + input.tellg(); + + input.seekg( + 0, + std::ios::beg + ); + + const auto expected_bytes = + static_cast( + expected_values + * sizeof(float) + ); + + if (byte_count != expected_bytes) { + throw std::runtime_error( + "Dimensione file incoerente: " + + path + + ", byte=" + + std::to_string(byte_count) + + ", attesi=" + + std::to_string( + expected_bytes + ) + ); + } + + std::vector values( + expected_values + ); + + input.read( + reinterpret_cast( + values.data() + ), + byte_count + ); + + if (!input) { + throw std::runtime_error( + "Lettura incompleta: " + + path + ); + } + + return values; +} + + +std::vector read_self_map( + const std::string& path, + std::size_t query_count) +{ + std::vector values( + query_count, + -1 + ); + + if (path.empty()) { + return values; + } + + std::ifstream input(path); + + if (!input) { + throw std::runtime_error( + "Impossibile aprire self-map: " + + path + ); + } + + std::size_t index = 0; + std::int64_t value = -1; + + while (input >> value) { + if (index >= query_count) { + throw std::runtime_error( + "Self-map contiene " + "troppe righe" + ); + } + + values[index] = value; + index += 1; + } + + if (index != query_count) { + throw std::runtime_error( + "Self-map contiene " + + std::to_string(index) + + " righe, attese " + + std::to_string( + query_count + ) + ); + } + + return values; +} + + +void write_json( + const face_knn::SearchOutput& output, + const std::vector& self_map, + float threshold) +{ + std::size_t recognized_count = 0; + + std::cout + << std::setprecision(9) + << "{\n" + << " \"backend\": \"" + << output.backend + << "\",\n" + << " \"device_name\": \"" + << output.device_name + << "\",\n" + << " \"query_count\": " + << output.query_count + << ",\n" + << " \"database_count\": " + << output.database_count + << ",\n" + << " \"dimension\": " + << output.dimension + << ",\n" + << " \"top_k\": " + << output.top_k + << ",\n" + << " \"threshold\": " + << threshold + << ",\n" + << " \"leave_one_out\": " + << ( + self_map.empty() + ? "false" + : "true" + ) + << ",\n" + << " \"elapsed_ms\": " + << output.elapsed_ms + << ",\n" + << " \"results\": [\n"; + + for ( + std::size_t query_index = 0; + query_index < output.query_count; + ++query_index + ) { + auto neighbors = + face_knn::query_neighbors( + output, + query_index + ); + + const std::int64_t excluded = + self_map[query_index]; + + if (excluded >= 0) { + neighbors = + face_knn::exclude_neighbor( + neighbors, + static_cast< + std::uint32_t + >(excluded) + ); + } + + if (neighbors.empty()) { + throw std::runtime_error( + "Nessun vicino disponibile " + "dopo leave-one-out" + ); + } + + const auto& best = + neighbors.front(); + + const bool recognized = + best.score >= threshold; + + if (recognized) { + recognized_count += 1; + } + + std::cout + << " {\n" + << " \"query_index\": " + << query_index + << ",\n" + << " \"excluded_index\": "; + + if (excluded < 0) { + std::cout << "null"; + } else { + std::cout << excluded; + } + + std::cout + << ",\n" + << " \"recognized\": " + << ( + recognized + ? "true" + : "false" + ) + << ",\n" + << " \"best_index\": " + << best.index + << ",\n" + << " \"best_score\": " + << best.score + << ",\n" + << " \"neighbors\": [\n"; + + for ( + std::size_t position = 0; + position < neighbors.size(); + ++position + ) { + const auto& neighbor = + neighbors[position]; + + std::cout + << " {" + << "\"index\": " + << neighbor.index + << ", \"score\": " + << neighbor.score + << "}"; + + if ( + position + 1 + < neighbors.size() + ) { + std::cout << ","; + } + + std::cout << "\n"; + } + + std::cout + << " ]\n" + << " }"; + + if ( + query_index + 1 + < output.query_count + ) { + std::cout << ","; + } + + std::cout << "\n"; + } + + const std::size_t below_count = + output.query_count + - recognized_count; + + const double coverage = + output.query_count > 0 + ? ( + static_cast( + recognized_count + ) + / static_cast( + output.query_count + ) + * 100.0 + ) + : 0.0; + + std::cout + << " ],\n" + << " \"recognized_count\": " + << recognized_count + << ",\n" + << " \"below_threshold_count\": " + << below_count + << ",\n" + << " \"coverage_percent\": " + << coverage + << "\n" + << "}\n"; +} + + +} // namespace + + +int main( + int argc, + char** argv) +{ + try { + const Arguments arguments = + parse_arguments( + argc, + argv + ); + + const std::size_t query_values = + arguments.query_count + * arguments.dimension; + + const std::size_t database_values = + arguments.database_count + * arguments.dimension; + + const auto queries = + read_float_file( + arguments.queries_path, + query_values + ); + + const auto database = + read_float_file( + arguments.database_path, + database_values + ); + + const auto self_map = + read_self_map( + arguments.self_map_path, + arguments.query_count + ); + + face_knn::SearchOptions options; + + options.dimension = + arguments.dimension; + + options.top_k = + arguments.top_k; + + options.kernel_path = + "kernels/knn_float4_topk.cl"; + + face_knn::SearchOutput output; + + if ( + arguments.backend == "opencl" + ) { + /* + * Il runtime Mali scrive la propria versione + * direttamente su stdout. Durante OpenCL + * deviamo stdout su stderr, affinché l'output + * JSON resti puro. + */ + ScopedStdoutToStderr redirect; + + output = + face_knn::search_opencl( + database, + queries, + arguments.database_count, + arguments.query_count, + options + ); + } else { + output = + face_knn::search_cpu( + database, + queries, + arguments.database_count, + arguments.query_count, + options + ); + } + + write_json( + output, + self_map, + arguments.threshold + ); + + return 0; + } catch (const std::exception& error) { + std::cerr + << "[face_knn] errore: " + << error.what() + << "\n"; + + print_usage(argv[0]); + + return 1; + } +} diff --git a/src/face_recognition.cc b/src/face_recognition.cc new file mode 100644 index 0000000..893c28c --- /dev/null +++ b/src/face_recognition.cc @@ -0,0 +1,144 @@ +#include +#include +#include +#include + +#include "face_engine.h" +#include "face_config.h" + +static constexpr float MATCH_THRESHOLD = 0.363f; + +static float cosine( + const std::vector& a, + const std::vector& b) +{ + if (a.size() != b.size()) + return 0.0f; + + float score = 0.0f; + + for (std::size_t i = 0; i < a.size(); ++i) + score += a[i] * b[i]; + + return score; +} + +static void print_faces( + const std::string& image_path, + const std::vector& faces) +{ + std::cout << "\n" << image_path << "\n"; + std::cout << "Faces: " << faces.size() << "\n"; + + for (std::size_t i = 0; i < faces.size(); ++i) { + const Face& face = faces[i]; + + std::cout + << "\nFace " << i + << " score=" << face.score + << "\n bbox=" + << face.x1 << " " + << face.y1 << " " + << face.x2 << " " + << face.y2 + << "\n"; + + std::cout << " landmarks:\n"; + + for (int j = 0; j < 5; ++j) { + std::cout + << " " + << j + << ": " + << face.kps[j].x + << " " + << face.kps[j].y + << "\n"; + } + } +} + +int main(int argc, char** argv) +{ + if (argc != 2 && argc != 3) { + std::cerr + << "Uso:\n" + << " " << argv[0] + << " image.jpg\n" + << " " << argv[0] + << " reference.jpg query.jpg\n"; + + return 1; + } + + const std::string root = + FACE_CLUSTERING_ROOT; + + const std::string scrfd_model = + root + "/models/SCRFD_500M_KPS_640.rknn"; + + const std::string sface_model = + root + "/models/face_recognition_sface_2021dec.rknn"; + + FaceEngine engine; + + if (!engine.init(scrfd_model, sface_model)) + return 1; + + std::vector faces1; + int image1_width = 0; + int image1_height = 0; + + if (!engine.process_image( + argv[1], + faces1, + image1_width, + image1_height)) { + return 1; + } + + print_faces(argv[1], faces1); + + if (argc == 2) + return 0; + + std::vector faces2; + int image2_width = 0; + int image2_height = 0; + + if (!engine.process_image( + argv[2], + faces2, + image2_width, + image2_height)) { + return 1; + } + + print_faces(argv[2], faces2); + + std::cout + << "\n========================================\n" + << "COSINE SIMILARITY\n" + << "========================================\n"; + + for (std::size_t i = 0; i < faces1.size(); ++i) { + for (std::size_t j = 0; j < faces2.size(); ++j) { + const float similarity = cosine( + faces1[i].embedding, + faces2[j].embedding); + + std::cout + << "ref[" << i + << "] vs query[" << j + << "] = " + << similarity; + + if (similarity >= MATCH_THRESHOLD) + std::cout << " MATCH"; + + std::cout << "\n"; + } + } + + return 0; +} diff --git a/src/face_scan.cc b/src/face_scan.cc new file mode 100644 index 0000000..08ddf04 --- /dev/null +++ b/src/face_scan.cc @@ -0,0 +1,368 @@ +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include + +#include "face_engine.h" + +static constexpr std::size_t EMBEDDING_SIZE = 128; + +static std::filesystem::path executable_path() +{ + std::vector buffer(4096); + + while (true) { + const ssize_t length = readlink( + "/proc/self/exe", + buffer.data(), + buffer.size() - 1); + + if (length < 0) { + throw std::runtime_error( + "Impossibile leggere /proc/self/exe"); + } + + const std::size_t path_length = + static_cast(length); + + if (path_length < buffer.size() - 1) { + buffer[path_length] = '\0'; + + return std::filesystem::path( + buffer.data()); + } + + buffer.resize(buffer.size() * 2); + } +} + +static std::filesystem::path application_root() +{ + /* + * Struttura prevista: + * + * root/ + * bin/face_scan + * lib/librknnrt.so + * models/.rknn + * + * executable_path(): + * root/bin/face_scan + * + * parent_path(): + * root/bin + * + * parent_path().parent_path(): + * root + */ + return executable_path() + .parent_path() + .parent_path(); +} + +static void write_json_string(const std::string& value) +{ + std::cout << '"'; + + for (unsigned char ch : value) { + switch (ch) { + case '"': + std::cout << "\\\""; + break; + + case '\\': + std::cout << "\\\\"; + break; + + case '\b': + std::cout << "\\b"; + break; + + case '\f': + std::cout << "\\f"; + break; + + case '\n': + std::cout << "\\n"; + break; + + case '\r': + std::cout << "\\r"; + break; + + case '\t': + std::cout << "\\t"; + break; + + default: + if (ch < 0x20) { + const char* hex = "0123456789abcdef"; + + std::cout + << "\\u00" + << hex[(ch >> 4) & 0x0f] + << hex[ch & 0x0f]; + } else { + std::cout << static_cast(ch); + } + } + } + + std::cout << '"'; +} + +static bool finite_face(const Face& face) +{ + if (!std::isfinite(face.score) || + !std::isfinite(face.x1) || + !std::isfinite(face.y1) || + !std::isfinite(face.x2) || + !std::isfinite(face.y2)) { + return false; + } + + for (const Point& point : face.kps) { + if (!std::isfinite(point.x) || + !std::isfinite(point.y)) { + return false; + } + } + + if (face.embedding.size() != EMBEDDING_SIZE) + return false; + + for (float value : face.embedding) { + if (!std::isfinite(value)) + return false; + } + + return true; +} + +static void write_face_json( + const Face& face, + std::size_t face_index) +{ + const float width = face.x2 - face.x1; + const float height = face.y2 - face.y1; + + std::cout << "{"; + + std::cout + << "\"face_index\":" + << face_index + << ","; + + std::cout + << "\"score\":" + << face.score + << ","; + + std::cout << "\"bbox\":{"; + + std::cout + << "\"x\":" + << face.x1 + << ","; + + std::cout + << "\"y\":" + << face.y1 + << ","; + + std::cout + << "\"width\":" + << width + << ","; + + std::cout + << "\"height\":" + << height; + + std::cout << "},"; + + std::cout << "\"landmarks\":["; + + for (int i = 0; i < 5; ++i) { + if (i != 0) + std::cout << ","; + + std::cout + << "[" + << face.kps[i].x + << "," + << face.kps[i].y + << "]"; + } + + std::cout << "],"; + + std::cout << "\"embedding\":["; + + for (std::size_t i = 0; + i < face.embedding.size(); + ++i) { + if (i != 0) + std::cout << ","; + + std::cout << face.embedding[i]; + } + + std::cout << "]"; + + std::cout << "}"; +} + +int main(int argc, char** argv) +{ + int rotation_degrees = 0; + const char* image_path = nullptr; + + if (argc == 2) { + image_path = argv[1]; + } else if ( + argc == 4 + && std::string(argv[1]) == "--rotation" + ) { + char* parse_end = nullptr; + + const long parsed_rotation = std::strtol( + argv[2], + &parse_end, + 10); + + if ( + parse_end == argv[2] + || *parse_end != '\0' + || parsed_rotation < -360000 + || parsed_rotation > 360000 + ) { + std::cerr + << "Rotazione non valida: " + << argv[2] + << "\n"; + + return 2; + } + + rotation_degrees = + static_cast(parsed_rotation); + + image_path = argv[3]; + } else { + std::cerr + << "Uso:\n" + << " " << argv[0] + << " image.jpg\n" + << " " << argv[0] + << " --rotation 90 image.jpg\n"; + + return 2; + } + + std::filesystem::path root; + + try { + root = application_root(); + } catch (const std::exception& error) { + std::cerr + << "Impossibile determinare la root applicativa: " + << error.what() + << "\n"; + + return 3; + } + + const std::string scrfd_model = + ( + root + / "models" + / "SCRFD_500M_KPS_640.rknn" + ).string(); + + const std::string sface_model = + ( + root + / "models" + / "face_recognition_sface_2021dec.rknn" + ).string(); + + FaceEngine engine; + + if (!engine.init(scrfd_model, sface_model)) { + std::cerr + << "Impossibile inizializzare FaceEngine\n"; + + return 3; + } + + std::vector faces; + int image_width = 0; + int image_height = 0; + + if (!engine.process_image( + image_path, + faces, + image_width, + image_height, + rotation_degrees)) { + std::cerr + << "Scansione fallita: " + << image_path + << "\n"; + + return 4; + } + + for (std::size_t i = 0; i < faces.size(); ++i) { + if (!finite_face(faces[i])) { + std::cerr + << "Risultato non valido per face_index=" + << i + << "\n"; + + return 5; + } + } + + std::cout << std::setprecision(9); + + std::cout << "{"; + + std::cout << "\"image\":{"; + + std::cout << "\"path\":"; + write_json_string(image_path); + + std::cout + << ",\"width\":" + << image_width + << ",\"height\":" + << image_height + << ",\"rotation\":" + << rotation_degrees; + + std::cout << "},"; + + std::cout << "\"faces\":["; + + for (std::size_t i = 0; i < faces.size(); ++i) { + if (i != 0) + std::cout << ","; + + write_face_json(faces[i], i); + } + + std::cout << "]"; + + std::cout << "}\n"; + + return 0; +} diff --git a/src/knn_opencl.cc b/src/knn_opencl.cc new file mode 100644 index 0000000..031a586 --- /dev/null +++ b/src/knn_opencl.cc @@ -0,0 +1,484 @@ +#include "knn_opencl.h" + +#include +#include +#include +#include +#include +#include +#include + +namespace face_knn { + +SearchOutput search_opencl_impl( + const std::vector& database, + const std::vector& queries, + std::size_t database_count, + std::size_t query_count, + const SearchOptions& options); + + +namespace { + +bool better_neighbor( + const Neighbor& left, + const Neighbor& right) +{ + if (left.score != right.score) { + return left.score > right.score; + } + + return left.index < right.index; +} + + +void validate_finite_values( + const std::vector& values, + const char* name) +{ + for ( + std::size_t index = 0; + index < values.size(); + ++index + ) { + if (!std::isfinite(values[index])) { + throw std::invalid_argument( + std::string(name) + + " contiene un valore non finito" + ); + } + } +} + + +std::uint32_t effective_top_k( + std::size_t database_count, + std::uint32_t requested_top_k) +{ + return std::min( + requested_top_k, + static_cast( + database_count + ) + ); +} + +} // namespace + + +void validate_input( + const std::vector& database, + const std::vector& queries, + std::size_t database_count, + std::size_t query_count, + const SearchOptions& options) +{ + if (options.dimension == 0) { + throw std::invalid_argument( + "dimension deve essere maggiore di zero" + ); + } + + if (options.top_k == 0) { + throw std::invalid_argument( + "top_k deve essere maggiore di zero" + ); + } + + if (options.lanes_per_query == 0) { + throw std::invalid_argument( + "lanes_per_query deve essere maggiore di zero" + ); + } + + if (options.chunk_size == 0) { + throw std::invalid_argument( + "chunk_size deve essere maggiore di zero" + ); + } + + if (database_count == 0) { + throw std::invalid_argument( + "database_count deve essere maggiore di zero" + ); + } + + if (query_count == 0) { + throw std::invalid_argument( + "query_count deve essere maggiore di zero" + ); + } + + if ( + database_count + > std::numeric_limits< + std::uint32_t + >::max() + ) { + throw std::invalid_argument( + "database_count supera uint32" + ); + } + + const std::size_t expected_database = + database_count + * static_cast( + options.dimension + ); + + const std::size_t expected_queries = + query_count + * static_cast( + options.dimension + ); + + if (database.size() != expected_database) { + throw std::invalid_argument( + "dimensione database incoerente" + ); + } + + if (queries.size() != expected_queries) { + throw std::invalid_argument( + "dimensione query incoerente" + ); + } + + validate_finite_values( + database, + "database" + ); + + validate_finite_values( + queries, + "queries" + ); +} + + +SearchOutput search_cpu( + const std::vector& database, + const std::vector& queries, + std::size_t database_count, + std::size_t query_count, + const SearchOptions& options) +{ + validate_input( + database, + queries, + database_count, + query_count, + options + ); + + const std::uint32_t result_top_k = + effective_top_k( + database_count, + options.top_k + ); + + SearchOutput output; + + output.query_count = query_count; + output.database_count = database_count; + output.dimension = options.dimension; + output.top_k = result_top_k; + output.backend = "cpu"; + output.device_name = "CPU"; + + output.neighbors.resize( + query_count + * static_cast( + result_top_k + ) + ); + + const auto started_at = + std::chrono::steady_clock::now(); + + for ( + std::size_t query_index = 0; + query_index < query_count; + ++query_index + ) { + const float* query = + &queries[ + query_index + * options.dimension + ]; + + std::vector candidates; + + candidates.reserve(database_count); + + for ( + std::size_t database_index = 0; + database_index < database_count; + ++database_index + ) { + const float* database_vector = + &database[ + database_index + * options.dimension + ]; + + float score = 0.0f; + + for ( + std::uint32_t component = 0; + component < options.dimension; + ++component + ) { + score += + database_vector[component] + * query[component]; + } + + candidates.push_back({ + score, + static_cast( + database_index + ), + }); + } + + if ( + result_top_k + < candidates.size() + ) { + std::nth_element( + candidates.begin(), + candidates.begin() + + result_top_k, + candidates.end(), + better_neighbor + ); + } + + std::sort( + candidates.begin(), + candidates.begin() + + result_top_k, + better_neighbor + ); + + const std::size_t output_offset = + query_index + * result_top_k; + + for ( + std::uint32_t position = 0; + position < result_top_k; + ++position + ) { + output.neighbors[ + output_offset + position + ] = candidates[position]; + } + } + + const auto completed_at = + std::chrono::steady_clock::now(); + + output.elapsed_ms = + std::chrono::duration< + double, + std::milli + >( + completed_at - started_at + ).count(); + + return output; +} + + +SearchOutput search_opencl( + const std::vector& database, + const std::vector& queries, + std::size_t database_count, + std::size_t query_count, + const SearchOptions& options) +{ + return search_opencl_impl( + database, + queries, + database_count, + query_count, + options + ); +} + + +std::vector query_neighbors( + const SearchOutput& output, + std::size_t query_index) +{ + if (query_index >= output.query_count) { + throw std::out_of_range( + "query_index fuori intervallo" + ); + } + + const std::size_t begin = + query_index + * output.top_k; + + const std::size_t end = + begin + + output.top_k; + + if (end > output.neighbors.size()) { + throw std::runtime_error( + "layout SearchOutput incoerente" + ); + } + + return { + output.neighbors.begin() + begin, + output.neighbors.begin() + end, + }; +} + + +std::vector exclude_neighbor( + const std::vector& neighbors, + std::uint32_t excluded_index) +{ + std::vector filtered; + + filtered.reserve(neighbors.size()); + + for (const Neighbor& neighbor : neighbors) { + if (neighbor.index == invalid_index) { + continue; + } + + if (neighbor.index == excluded_index) { + continue; + } + + filtered.push_back(neighbor); + } + + return filtered; +} + + +ValidationResult compare_outputs( + const SearchOutput& expected, + const SearchOutput& actual, + float score_tolerance, + float recognition_threshold) +{ + if (score_tolerance < 0.0f) { + throw std::invalid_argument( + "score_tolerance non valida" + ); + } + + if ( + expected.query_count + != actual.query_count + || expected.database_count + != actual.database_count + || expected.dimension + != actual.dimension + || expected.top_k + != actual.top_k + || expected.neighbors.size() + != actual.neighbors.size() + ) { + throw std::invalid_argument( + "Output kNN non confrontabili" + ); + } + + ValidationResult result; + + result.positions = + expected.neighbors.size(); + + for ( + std::size_t index = 0; + index < result.positions; + ++index + ) { + const Neighbor& expected_neighbor = + expected.neighbors[index]; + + const Neighbor& actual_neighbor = + actual.neighbors[index]; + + if ( + expected_neighbor.index + == actual_neighbor.index + ) { + result.exact_index_matches += 1; + } + + const float error = std::fabs( + expected_neighbor.score + - actual_neighbor.score + ); + + result.maximum_score_error = + std::max( + result.maximum_score_error, + error + ); + } + + for ( + std::size_t query_index = 0; + query_index < expected.query_count; + ++query_index + ) { + const auto expected_neighbors = + query_neighbors( + expected, + query_index + ); + + const auto actual_neighbors = + query_neighbors( + actual, + query_index + ); + + const bool expected_recognized = + !expected_neighbors.empty() + && expected_neighbors.front().score + >= recognition_threshold; + + const bool actual_recognized = + !actual_neighbors.empty() + && actual_neighbors.front().score + >= recognition_threshold; + + if ( + expected_recognized + == actual_recognized + ) { + result.decision_matches += 1; + } + } + + result.exact_indices = + result.exact_index_matches + == result.positions; + + result.decisions_equal = + result.decision_matches + == expected.query_count; + + if ( + result.maximum_score_error + > score_tolerance + ) { + result.decisions_equal = false; + } + + return result; +} + +} // namespace face_knn diff --git a/src/knn_opencl_backend.cc b/src/knn_opencl_backend.cc new file mode 100644 index 0000000..46a12c2 --- /dev/null +++ b/src/knn_opencl_backend.cc @@ -0,0 +1,944 @@ +#define CL_TARGET_OPENCL_VERSION 120 + +#include + +#include "knn_opencl.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +namespace face_knn { +namespace { + +void check_cl( + cl_int status, + const char* operation) +{ + if (status != CL_SUCCESS) { + throw std::runtime_error( + std::string(operation) + + ": OpenCL status " + + std::to_string(status) + ); + } +} + + +std::string device_string( + cl_device_id device, + cl_device_info property) +{ + size_t size = 0; + + check_cl( + clGetDeviceInfo( + device, + property, + 0, + nullptr, + &size + ), + "clGetDeviceInfo size" + ); + + std::string value( + size, + '\0' + ); + + check_cl( + clGetDeviceInfo( + device, + property, + size, + value.data(), + nullptr + ), + "clGetDeviceInfo value" + ); + + while ( + !value.empty() + && value.back() == '\0' + ) { + value.pop_back(); + } + + return value; +} + + +cl_device_id find_mali_device( + std::string& device_name) +{ + cl_uint platform_count = 0; + + check_cl( + clGetPlatformIDs( + 0, + nullptr, + &platform_count + ), + "clGetPlatformIDs count" + ); + + if (platform_count == 0) { + throw std::runtime_error( + "Nessuna piattaforma OpenCL" + ); + } + + std::vector platforms( + platform_count + ); + + check_cl( + clGetPlatformIDs( + platform_count, + platforms.data(), + nullptr + ), + "clGetPlatformIDs list" + ); + + cl_device_id fallback = nullptr; + std::string fallback_name; + + for ( + cl_platform_id platform + : platforms + ) { + cl_uint device_count = 0; + + const cl_int status = + clGetDeviceIDs( + platform, + CL_DEVICE_TYPE_GPU, + 0, + nullptr, + &device_count + ); + + if ( + status != CL_SUCCESS + || device_count == 0 + ) { + continue; + } + + std::vector devices( + device_count + ); + + check_cl( + clGetDeviceIDs( + platform, + CL_DEVICE_TYPE_GPU, + device_count, + devices.data(), + nullptr + ), + "clGetDeviceIDs list" + ); + + for ( + cl_device_id device + : devices + ) { + const std::string name = + device_string( + device, + CL_DEVICE_NAME + ); + + const std::string vendor = + device_string( + device, + CL_DEVICE_VENDOR + ); + + const std::string full_name = + vendor + " / " + name; + + if (fallback == nullptr) { + fallback = device; + fallback_name = full_name; + } + + if ( + name.find("Mali-G610") + != std::string::npos + ) { + device_name = full_name; + return device; + } + } + } + + if (fallback != nullptr) { + device_name = fallback_name; + + std::cerr + << "[face_knn] Mali-G610 non trovata; " + << "uso GPU OpenCL: " + << fallback_name + << "\n"; + + return fallback; + } + + throw std::runtime_error( + "Nessuna GPU OpenCL disponibile" + ); +} + + +std::string read_text_file( + const std::string& path) +{ + std::ifstream input(path); + + if (!input) { + throw std::runtime_error( + "Kernel OpenCL non leggibile: " + + path + ); + } + + const std::string content{ + std::istreambuf_iterator( + input + ), + std::istreambuf_iterator() + }; + + if (content.empty()) { + throw std::runtime_error( + "Kernel OpenCL vuoto: " + + path + ); + } + + return content; +} + + +bool better_neighbor( + const Neighbor& left, + const Neighbor& right) +{ + if (left.score != right.score) { + return left.score > right.score; + } + + return left.index < right.index; +} + + +Neighbor invalid_neighbor() +{ + return { + -std::numeric_limits::infinity(), + invalid_index, + }; +} + + +void merge_top_k( + const std::vector& chunk_indices, + const std::vector& chunk_scores, + std::size_t chunk_offset, + std::size_t query_count, + std::uint32_t top_k, + std::vector& global_result) +{ + for ( + std::size_t query_index = 0; + query_index < query_count; + ++query_index + ) { + const std::size_t result_offset = + query_index + * top_k; + + std::vector candidates; + + candidates.reserve( + static_cast( + top_k + ) * 2 + ); + + for ( + std::uint32_t position = 0; + position < top_k; + ++position + ) { + const Neighbor& current = + global_result[ + result_offset + + position + ]; + + if ( + current.index + != invalid_index + ) { + candidates.push_back( + current + ); + } + } + + for ( + std::uint32_t position = 0; + position < top_k; + ++position + ) { + const std::size_t chunk_position = + result_offset + + position; + + const std::uint32_t local_index = + chunk_indices[ + chunk_position + ]; + + if ( + local_index + == invalid_index + ) { + continue; + } + + candidates.push_back({ + chunk_scores[ + chunk_position + ], + static_cast( + chunk_offset + + local_index + ), + }); + } + + std::sort( + candidates.begin(), + candidates.end(), + better_neighbor + ); + + for ( + std::uint32_t position = 0; + position < top_k; + ++position + ) { + Neighbor value = + invalid_neighbor(); + + if ( + position + < candidates.size() + ) { + value = + candidates[position]; + } + + global_result[ + result_offset + + position + ] = value; + } + } +} + + +struct OpenCLResources { + cl_context context = nullptr; + + cl_command_queue queue = nullptr; + + cl_program program = nullptr; + + cl_kernel kernel = nullptr; + + cl_mem database_buffer = nullptr; + + cl_mem query_buffer = nullptr; + + cl_mem index_buffer = nullptr; + + cl_mem score_buffer = nullptr; + + ~OpenCLResources() + { + if (score_buffer != nullptr) { + clReleaseMemObject( + score_buffer + ); + } + + if (index_buffer != nullptr) { + clReleaseMemObject( + index_buffer + ); + } + + if (query_buffer != nullptr) { + clReleaseMemObject( + query_buffer + ); + } + + if (database_buffer != nullptr) { + clReleaseMemObject( + database_buffer + ); + } + + if (kernel != nullptr) { + clReleaseKernel(kernel); + } + + if (program != nullptr) { + clReleaseProgram(program); + } + + if (queue != nullptr) { + clReleaseCommandQueue(queue); + } + + if (context != nullptr) { + clReleaseContext(context); + } + } +}; + + +std::string build_log( + cl_program program, + cl_device_id device) +{ + size_t size = 0; + + clGetProgramBuildInfo( + program, + device, + CL_PROGRAM_BUILD_LOG, + 0, + nullptr, + &size + ); + + std::string value( + size, + '\0' + ); + + if (size > 0) { + clGetProgramBuildInfo( + program, + device, + CL_PROGRAM_BUILD_LOG, + size, + value.data(), + nullptr + ); + } + + while ( + !value.empty() + && value.back() == '\0' + ) { + value.pop_back(); + } + + return value; +} + +} // namespace + + +SearchOutput search_opencl_impl( + const std::vector& database, + const std::vector& queries, + std::size_t database_count, + std::size_t query_count, + const SearchOptions& options) +{ + validate_input( + database, + queries, + database_count, + query_count, + options + ); + + if ( + query_count + > std::numeric_limits< + std::uint32_t + >::max() + ) { + throw std::invalid_argument( + "query_count supera uint32" + ); + } + + const std::uint32_t top_k = + std::min( + options.top_k, + static_cast( + database_count + ) + ); + + const std::string kernel_path = + options.kernel_path.empty() + ? "kernels/knn_float4_topk.cl" + : options.kernel_path; + + const std::string kernel_source = + read_text_file( + kernel_path + ); + + std::string device_name; + + const cl_device_id device = + find_mali_device( + device_name + ); + + std::cerr + << "[face_knn] OpenCL GPU: " + << device_name + << "\n"; + + OpenCLResources resources; + + cl_int status = CL_SUCCESS; + + resources.context = + clCreateContext( + nullptr, + 1, + &device, + nullptr, + nullptr, + &status + ); + + check_cl( + status, + "clCreateContext" + ); + + resources.queue = + clCreateCommandQueue( + resources.context, + device, + 0, + &status + ); + + check_cl( + status, + "clCreateCommandQueue" + ); + + const char* source_pointer = + kernel_source.c_str(); + + const size_t source_size = + kernel_source.size(); + + resources.program = + clCreateProgramWithSource( + resources.context, + 1, + &source_pointer, + &source_size, + &status + ); + + check_cl( + status, + "clCreateProgramWithSource" + ); + + const std::string build_options = + "-cl-fast-relaxed-math" + " -D DIM=" + + std::to_string( + options.dimension + ) + + " -D TOPK=" + + std::to_string(top_k) + + " -D QUERY_COUNT=" + + std::to_string(query_count) + + " -D LANES_PER_QUERY=" + + std::to_string( + options.lanes_per_query + ); + + status = clBuildProgram( + resources.program, + 1, + &device, + build_options.c_str(), + nullptr, + nullptr + ); + + if (status != CL_SUCCESS) { + throw std::runtime_error( + "clBuildProgram fallito: " + + build_log( + resources.program, + device + ) + ); + } + + resources.kernel = + clCreateKernel( + resources.program, + "knn", + &status + ); + + check_cl( + status, + "clCreateKernel" + ); + + const std::size_t chunk_capacity = + std::min( + options.chunk_size, + database_count + ); + + resources.database_buffer = + clCreateBuffer( + resources.context, + CL_MEM_READ_ONLY, + chunk_capacity + * options.dimension + * sizeof(float), + nullptr, + &status + ); + + check_cl( + status, + "clCreateBuffer database" + ); + + resources.query_buffer = + clCreateBuffer( + resources.context, + CL_MEM_READ_ONLY + | CL_MEM_COPY_HOST_PTR, + queries.size() + * sizeof(float), + const_cast( + queries.data() + ), + &status + ); + + check_cl( + status, + "clCreateBuffer queries" + ); + + const std::size_t result_count = + query_count + * static_cast( + top_k + ); + + std::vector chunk_indices( + result_count, + invalid_index + ); + + std::vector chunk_scores( + result_count, + -std::numeric_limits::infinity() + ); + + resources.index_buffer = + clCreateBuffer( + resources.context, + CL_MEM_WRITE_ONLY, + result_count + * sizeof(std::uint32_t), + nullptr, + &status + ); + + check_cl( + status, + "clCreateBuffer indices" + ); + + resources.score_buffer = + clCreateBuffer( + resources.context, + CL_MEM_WRITE_ONLY, + result_count + * sizeof(float), + nullptr, + &status + ); + + check_cl( + status, + "clCreateBuffer scores" + ); + + check_cl( + clSetKernelArg( + resources.kernel, + 0, + sizeof( + resources.database_buffer + ), + &resources.database_buffer + ), + "clSetKernelArg database" + ); + + check_cl( + clSetKernelArg( + resources.kernel, + 1, + sizeof( + resources.query_buffer + ), + &resources.query_buffer + ), + "clSetKernelArg queries" + ); + + check_cl( + clSetKernelArg( + resources.kernel, + 2, + sizeof( + resources.index_buffer + ), + &resources.index_buffer + ), + "clSetKernelArg indices" + ); + + check_cl( + clSetKernelArg( + resources.kernel, + 3, + sizeof( + resources.score_buffer + ), + &resources.score_buffer + ), + "clSetKernelArg scores" + ); + + SearchOutput output; + + output.query_count = + query_count; + + output.database_count = + database_count; + + output.dimension = + options.dimension; + + output.top_k = + top_k; + + output.backend = + "opencl"; + + output.device_name = + device_name; + + output.neighbors.assign( + result_count, + invalid_neighbor() + ); + + const std::size_t global_size = + query_count + * options.lanes_per_query; + + const std::size_t local_size = + options.lanes_per_query; + + const std::size_t chunk_count = + ( + database_count + + options.chunk_size + - 1 + ) + / options.chunk_size; + + std::cerr + << "[face_knn] kernel_path=" + << kernel_path + << "\n" + << "[face_knn] build_options=" + << build_options + << "\n" + << "[face_knn] chunk_size=" + << chunk_capacity + << "\n" + << "[face_knn] chunks=" + << chunk_count + << "\n"; + + const auto started_at = + std::chrono::steady_clock::now(); + + for ( + std::size_t chunk_offset = 0; + chunk_offset < database_count; + chunk_offset += options.chunk_size + ) { + const std::size_t chunk_size = + std::min( + options.chunk_size, + database_count + - chunk_offset + ); + + const float* chunk_pointer = + &database[ + chunk_offset + * options.dimension + ]; + + check_cl( + clEnqueueWriteBuffer( + resources.queue, + resources.database_buffer, + CL_TRUE, + 0, + chunk_size + * options.dimension + * sizeof(float), + chunk_pointer, + 0, + nullptr, + nullptr + ), + "clEnqueueWriteBuffer database" + ); + + const std::uint32_t kernel_count = + static_cast( + chunk_size + ); + + check_cl( + clSetKernelArg( + resources.kernel, + 4, + sizeof(kernel_count), + &kernel_count + ), + "clSetKernelArg database count" + ); + + check_cl( + clEnqueueNDRangeKernel( + resources.queue, + resources.kernel, + 1, + nullptr, + &global_size, + &local_size, + 0, + nullptr, + nullptr + ), + "clEnqueueNDRangeKernel" + ); + + check_cl( + clFinish( + resources.queue + ), + "clFinish" + ); + + check_cl( + clEnqueueReadBuffer( + resources.queue, + resources.index_buffer, + CL_TRUE, + 0, + chunk_indices.size() + * sizeof(std::uint32_t), + chunk_indices.data(), + 0, + nullptr, + nullptr + ), + "clEnqueueReadBuffer indices" + ); + + check_cl( + clEnqueueReadBuffer( + resources.queue, + resources.score_buffer, + CL_TRUE, + 0, + chunk_scores.size() + * sizeof(float), + chunk_scores.data(), + 0, + nullptr, + nullptr + ), + "clEnqueueReadBuffer scores" + ); + + merge_top_k( + chunk_indices, + chunk_scores, + chunk_offset, + query_count, + top_k, + output.neighbors + ); + } + + const auto completed_at = + std::chrono::steady_clock::now(); + + output.elapsed_ms = + std::chrono::duration< + double, + std::milli + >( + completed_at - started_at + ).count(); + + return output; +} + + +} // namespace face_knn diff --git a/test/README.md b/test/README.md new file mode 100644 index 0000000..c4f8b69 --- /dev/null +++ b/test/README.md @@ -0,0 +1,24 @@ +# Immagini di test + +Inserire in questa directory immagini locali per verificare: + +- rilevamento dei volti; +- bounding box; +- landmark; +- generazione degli embedding; +- cosine similarity; +- output JSON. + +Esempio: + +```bash +./bin/face_recognition test/reference.jpg test/query.jpg +``` + +Oppure: + +```bash +./bin/face_scan test/image.jpg +``` + +Le immagini sono escluse dal repository tramite `.gitignore`. diff --git a/test/knn/test_knn_cpu.cc b/test/knn/test_knn_cpu.cc new file mode 100644 index 0000000..7176f22 --- /dev/null +++ b/test/knn/test_knn_cpu.cc @@ -0,0 +1,107 @@ +#include "knn_opencl.h" + +#include +#include +#include + +int main() +{ + const std::vector database = { + 1.0f, 0.0f, + 0.0f, 1.0f, + 0.70710678f, 0.70710678f, + }; + + const std::vector queries = { + 1.0f, 0.0f, + 0.0f, 1.0f, + }; + + face_knn::SearchOptions options; + + options.dimension = 2; + options.top_k = 2; + + const auto output = + face_knn::search_cpu( + database, + queries, + 3, + 2, + options + ); + + const auto first = + face_knn::query_neighbors( + output, + 0 + ); + + const auto second = + face_knn::query_neighbors( + output, + 1 + ); + + if ( + first.size() != 2 + || second.size() != 2 + ) { + std::cerr + << "Numero vicini errato\n"; + + return 1; + } + + if ( + first[0].index != 0 + || second[0].index != 1 + ) { + std::cerr + << "Top-1 errato\n"; + + return 1; + } + + if ( + std::fabs(first[0].score - 1.0f) + > 0.000001f + || std::fabs(second[0].score - 1.0f) + > 0.000001f + ) { + std::cerr + << "Score top-1 errato\n"; + + return 1; + } + + const auto filtered = + face_knn::exclude_neighbor( + first, + 0 + ); + + if ( + filtered.empty() + || filtered[0].index != 2 + ) { + std::cerr + << "Leave-one-out errato\n"; + + return 1; + } + + std::cout + << "OK: backend CPU dinamico\n" + << "query_count=" + << output.query_count + << "\n" + << "database_count=" + << output.database_count + << "\n" + << "top_k=" + << output.top_k + << "\n"; + + return 0; +} diff --git a/test/test.jpg b/test/test.jpg new file mode 100644 index 0000000..8e0c108 Binary files /dev/null and b/test/test.jpg differ diff --git a/test/test2.jpg b/test/test2.jpg new file mode 100644 index 0000000..544e0ad Binary files /dev/null and b/test/test2.jpg differ diff --git a/test/test3.jpg b/test/test3.jpg new file mode 100644 index 0000000..14921bb Binary files /dev/null and b/test/test3.jpg differ diff --git a/test/test3f.jpg b/test/test3f.jpg new file mode 100644 index 0000000..c124cda Binary files /dev/null and b/test/test3f.jpg differ diff --git a/test/test4.jpg b/test/test4.jpg new file mode 100644 index 0000000..3115472 Binary files /dev/null and b/test/test4.jpg differ diff --git a/test/test5.jpg b/test/test5.jpg new file mode 100644 index 0000000..0432520 Binary files /dev/null and b/test/test5.jpg differ diff --git a/test/test6.jpg b/test/test6.jpg new file mode 100644 index 0000000..eba3229 Binary files /dev/null and b/test/test6.jpg differ