Fix png decoder
This commit is contained in:
@@ -34,6 +34,14 @@ struct PngContext {
|
||||
int dstHeight{0};
|
||||
int lastDstY{-1}; // Track last rendered destination Y to avoid duplicates
|
||||
|
||||
// tRNS support for grayscale / truecolor PNGs. PNGdec exposes a single transparent
|
||||
// sample value (grayscale) or RGB triplet (truecolor) via getTransparentColor() and
|
||||
// sets hasAlpha=1 on those pixel types when the chunk is present. Indexed PNGs use
|
||||
// the per-entry alpha array stored at palette[768..1023] instead.
|
||||
bool hasTrnsKey{false};
|
||||
uint8_t trnsGray{0};
|
||||
uint8_t trnsR{0}, trnsG{0}, trnsB{0};
|
||||
|
||||
PixelCache cache;
|
||||
bool caching{false};
|
||||
|
||||
@@ -175,48 +183,102 @@ constexpr size_t MIN_FREE_HEAP_FOR_PNG = PNG_DECODER_APPROX_SIZE + 16 * 1024; /
|
||||
// Required storage is therefore approximately: 2 * (pitch + 1) + alignment slack.
|
||||
// If PNG_MAX_BUFFERED_PIXELS is smaller than this requirement for a given image,
|
||||
// PNGdec can overrun its internal buffer before our draw callback executes.
|
||||
int bytesPerPixelFromType(int pixelType) {
|
||||
|
||||
// PNG row pitch in bytes — matches PNGdec's internal calculation in PNGParseInfo
|
||||
// (png.inl). bpp is the per-channel bit depth (1/2/4/8 for grayscale & indexed,
|
||||
// 8 elsewhere). For sub-byte depths the row is bit-packed: pitch < width.
|
||||
int pngPitchBytes(int srcWidth, int pixelType, int bpp) {
|
||||
switch (pixelType) {
|
||||
case PNG_PIXEL_TRUECOLOR:
|
||||
return 3;
|
||||
return ((3 * bpp) * srcWidth + 7) / 8;
|
||||
case PNG_PIXEL_GRAY_ALPHA:
|
||||
return 2;
|
||||
return ((2 * bpp) * srcWidth + 7) / 8;
|
||||
case PNG_PIXEL_TRUECOLOR_ALPHA:
|
||||
return 4;
|
||||
return ((4 * bpp) * srcWidth + 7) / 8;
|
||||
case PNG_PIXEL_GRAYSCALE:
|
||||
case PNG_PIXEL_INDEXED:
|
||||
default:
|
||||
return 1;
|
||||
return (srcWidth * bpp + 7) / 8;
|
||||
}
|
||||
}
|
||||
|
||||
int requiredPngInternalBufferBytes(int srcWidth, int pixelType) {
|
||||
int requiredPngInternalBufferBytes(int srcWidth, int pixelType, int bpp) {
|
||||
// +1 filter byte per scanline, *2 for current+previous lines, +32 for alignment margin.
|
||||
int pitch = srcWidth * bytesPerPixelFromType(pixelType);
|
||||
int pitch = pngPitchBytes(srcWidth, pixelType, bpp);
|
||||
return ((pitch + 1) * 2) + 32;
|
||||
}
|
||||
|
||||
// Extract the bpp-bit sample at index `x` from a packed scanline (MSB-first).
|
||||
// Supports bpp values 1, 2, 4, 8 — the only depths PNG allows for grayscale and indexed.
|
||||
inline uint8_t extractPackedSample(const uint8_t* pPixels, int x, int bpp) {
|
||||
if (bpp == 8) return pPixels[x];
|
||||
const int pixelsPerByte = 8 / bpp;
|
||||
const uint8_t mask = (uint8_t)((1 << bpp) - 1);
|
||||
const int byteIdx = x / pixelsPerByte;
|
||||
const int shift = (pixelsPerByte - 1 - (x % pixelsPerByte)) * bpp;
|
||||
return (pPixels[byteIdx] >> shift) & mask;
|
||||
}
|
||||
|
||||
// Convert entire source line to grayscale with alpha blending to white background.
|
||||
// For indexed PNGs with tRNS chunk, alpha values are stored at palette[768] onwards.
|
||||
// `bpp` is the per-channel bit depth from PNGdec (1, 2, 4, or 8 for grayscale/indexed).
|
||||
// Sub-byte depths require unpacking the bit-packed scanline before lookup; treating each
|
||||
// byte as a single sample silently corrupts the output (PNGdec reports indexed-with-tRNS
|
||||
// PNGs at 1/2/4 bpp this way and the consequences are uninitialized palette reads).
|
||||
// For grayscale and truecolor PNGs, `ctx` carries any tRNS color-key set on the image —
|
||||
// matching pixels are composited to white instead of their decoded value.
|
||||
// Processing the whole line at once improves cache locality and reduces per-pixel overhead.
|
||||
void convertLineToGray(uint8_t* pPixels, uint8_t* grayLine, int width, int pixelType, uint8_t* palette, int hasAlpha) {
|
||||
void convertLineToGray(const PngContext& ctx, uint8_t* pPixels, uint8_t* grayLine, int width, int pixelType, int bpp,
|
||||
uint8_t* palette, int hasAlpha) {
|
||||
switch (pixelType) {
|
||||
case PNG_PIXEL_GRAYSCALE:
|
||||
memcpy(grayLine, pPixels, width);
|
||||
break;
|
||||
|
||||
case PNG_PIXEL_TRUECOLOR:
|
||||
for (int x = 0; x < width; x++) {
|
||||
uint8_t* p = &pPixels[x * 3];
|
||||
grayLine[x] = (uint8_t)((p[0] * 77 + p[1] * 150 + p[2] * 29) >> 8);
|
||||
case PNG_PIXEL_GRAYSCALE: {
|
||||
// tRNS for grayscale stores the transparent sample value at native bit depth.
|
||||
// Compare against the raw packed sample before scaling to 8-bit gray.
|
||||
const bool useKey = ctx.hasTrnsKey;
|
||||
const uint8_t key = ctx.trnsGray;
|
||||
if (bpp == 8) {
|
||||
if (useKey) {
|
||||
for (int x = 0; x < width; x++) {
|
||||
uint8_t sample = pPixels[x];
|
||||
grayLine[x] = (sample == key) ? 255 : sample;
|
||||
}
|
||||
} else {
|
||||
memcpy(grayLine, pPixels, width);
|
||||
}
|
||||
} else {
|
||||
const int maxVal = (1 << bpp) - 1;
|
||||
for (int x = 0; x < width; x++) {
|
||||
uint8_t sample = extractPackedSample(pPixels, x, bpp);
|
||||
if (useKey && sample == key) {
|
||||
grayLine[x] = 255;
|
||||
} else {
|
||||
grayLine[x] = (uint8_t)((sample * 255) / maxVal);
|
||||
}
|
||||
}
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
||||
case PNG_PIXEL_TRUECOLOR: {
|
||||
// tRNS for truecolor stores an 8-bit-per-channel RGB triplet to match exactly.
|
||||
const bool useKey = ctx.hasTrnsKey;
|
||||
const uint8_t kr = ctx.trnsR, kg = ctx.trnsG, kb = ctx.trnsB;
|
||||
for (int x = 0; x < width; x++) {
|
||||
uint8_t* p = &pPixels[x * 3];
|
||||
if (useKey && p[0] == kr && p[1] == kg && p[2] == kb) {
|
||||
grayLine[x] = 255;
|
||||
} else {
|
||||
grayLine[x] = (uint8_t)((p[0] * 77 + p[1] * 150 + p[2] * 29) >> 8);
|
||||
}
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
||||
case PNG_PIXEL_INDEXED:
|
||||
if (palette) {
|
||||
if (hasAlpha) {
|
||||
for (int x = 0; x < width; x++) {
|
||||
uint8_t idx = pPixels[x];
|
||||
uint8_t idx = extractPackedSample(pPixels, x, bpp);
|
||||
uint8_t* p = &palette[idx * 3];
|
||||
uint8_t gray = (uint8_t)((p[0] * 77 + p[1] * 150 + p[2] * 29) >> 8);
|
||||
uint8_t alpha = palette[768 + idx];
|
||||
@@ -224,12 +286,16 @@ void convertLineToGray(uint8_t* pPixels, uint8_t* grayLine, int width, int pixel
|
||||
}
|
||||
} else {
|
||||
for (int x = 0; x < width; x++) {
|
||||
uint8_t* p = &palette[pPixels[x] * 3];
|
||||
uint8_t idx = extractPackedSample(pPixels, x, bpp);
|
||||
uint8_t* p = &palette[idx * 3];
|
||||
grayLine[x] = (uint8_t)((p[0] * 77 + p[1] * 150 + p[2] * 29) >> 8);
|
||||
}
|
||||
}
|
||||
} else {
|
||||
memcpy(grayLine, pPixels, width);
|
||||
// Indexed PNG without palette is malformed (PNGdec should always populate it).
|
||||
// Fill white so downstream dithering produces a clean blank rather than treating
|
||||
// raw bit-packed data as gray values.
|
||||
memset(grayLine, 255, width);
|
||||
}
|
||||
break;
|
||||
|
||||
@@ -277,8 +343,8 @@ int pngDrawCallback(PNGDRAW* pDraw) {
|
||||
if (outY >= ctx->screenHeight) return 1;
|
||||
|
||||
// Convert entire source line to grayscale (improves cache locality)
|
||||
convertLineToGray(pDraw->pPixels, ctx->grayLineBuffer, srcWidth, pDraw->iPixelType, pDraw->pPalette,
|
||||
pDraw->iHasAlpha);
|
||||
convertLineToGray(*ctx, pDraw->pPixels, ctx->grayLineBuffer, srcWidth, pDraw->iPixelType, pDraw->iBpp,
|
||||
pDraw->pPalette, pDraw->iHasAlpha);
|
||||
|
||||
// Render scaled row using Bresenham-style integer stepping (no floating-point division)
|
||||
int dstWidth = ctx->dstWidth;
|
||||
@@ -307,10 +373,15 @@ int pngDrawCallback(PNGDRAW* pDraw) {
|
||||
|
||||
for (int dstX = 0; dstX < dstWidth; dstX++) {
|
||||
int outX = outXBase + dstX;
|
||||
if (outX < screenWidth) {
|
||||
uint8_t gray = ctx->grayLineBuffer[srcX];
|
||||
|
||||
uint8_t ditheredGray = ditherGray(*ctx, gray, dstX, outX, outY);
|
||||
// Always run dithering — error-diffusion ditherers carry per-column state across
|
||||
// rows, and skipping pixels that fall off the right edge would leave stale error
|
||||
// values that bleed into the next row's edge pixels. Only the framebuffer/cache
|
||||
// writes are guarded by the screen-bounds check.
|
||||
uint8_t gray = ctx->grayLineBuffer[srcX];
|
||||
uint8_t ditheredGray = ditherGray(*ctx, gray, dstX, outX, outY);
|
||||
|
||||
if (outX >= 0 && outX < screenWidth) {
|
||||
pw.writePixel(outX, ditheredGray);
|
||||
if (caching) cw.writePixel(outX, ditheredGray);
|
||||
}
|
||||
@@ -329,30 +400,49 @@ int pngDrawCallback(PNGDRAW* pDraw) {
|
||||
} // namespace
|
||||
|
||||
bool PngToFramebufferConverter::getDimensionsStatic(const std::string& imagePath, ImageDimensions& out) {
|
||||
size_t freeHeap = ESP.getFreeHeap();
|
||||
if (freeHeap < MIN_FREE_HEAP_FOR_PNG) {
|
||||
LOG_ERR("PNG", "Not enough heap for PNG decoder (%u free, need %u)", freeHeap, MIN_FREE_HEAP_FOR_PNG);
|
||||
// PNG file layout: 8-byte signature, then chunks. The IHDR chunk is mandatory and
|
||||
// must be the first chunk: 4 bytes length + "IHDR" + 13 bytes IHDR data + 4 bytes CRC.
|
||||
// Width and height live at bytes 16..23 (big-endian uint32s) of the file.
|
||||
// Reading those bytes directly avoids allocating PNGdec's ~42 KB working buffers
|
||||
// — important on the C3 where dimension queries can run while decode buffers from
|
||||
// a previous image are still pinned. We deliberately don't validate the CRC here;
|
||||
// a corrupt IHDR will surface during the actual decode.
|
||||
FsFile f;
|
||||
if (!Storage.openFileForRead("PNG", imagePath, f)) {
|
||||
LOG_ERR("PNG", "Failed to open file for dimensions: %s", imagePath.c_str());
|
||||
return false;
|
||||
}
|
||||
|
||||
std::unique_ptr<PNG> png(new (std::nothrow) PNG());
|
||||
if (!png) {
|
||||
LOG_ERR("PNG", "Failed to allocate PNG decoder for dimensions");
|
||||
uint8_t hdr[24];
|
||||
int n = f.read(hdr, sizeof(hdr));
|
||||
f.close();
|
||||
if (n < (int)sizeof(hdr)) {
|
||||
LOG_ERR("PNG", "Short read on PNG header: %s", imagePath.c_str());
|
||||
return false;
|
||||
}
|
||||
|
||||
int rc = png->open(imagePath.c_str(), pngOpenWithHandle, pngCloseWithHandle, pngReadWithHandle, pngSeekWithHandle,
|
||||
nullptr);
|
||||
|
||||
if (rc != 0) {
|
||||
LOG_ERR("PNG", "Failed to open PNG for dimensions: %d", rc);
|
||||
// Validate PNG signature: 89 50 4E 47 0D 0A 1A 0A
|
||||
static constexpr uint8_t kPngSig[8] = {0x89, 0x50, 0x4E, 0x47, 0x0D, 0x0A, 0x1A, 0x0A};
|
||||
if (memcmp(hdr, kPngSig, 8) != 0) {
|
||||
LOG_ERR("PNG", "Not a PNG file: %s", imagePath.c_str());
|
||||
return false;
|
||||
}
|
||||
// Chunk type at hdr[12..15] must be "IHDR"
|
||||
if (hdr[12] != 'I' || hdr[13] != 'H' || hdr[14] != 'D' || hdr[15] != 'R') {
|
||||
LOG_ERR("PNG", "First chunk not IHDR: %s", imagePath.c_str());
|
||||
return false;
|
||||
}
|
||||
|
||||
out.width = png->getWidth();
|
||||
out.height = png->getHeight();
|
||||
uint32_t width = ((uint32_t)hdr[16] << 24) | ((uint32_t)hdr[17] << 16) | ((uint32_t)hdr[18] << 8) | (uint32_t)hdr[19];
|
||||
uint32_t height =
|
||||
((uint32_t)hdr[20] << 24) | ((uint32_t)hdr[21] << 16) | ((uint32_t)hdr[22] << 8) | (uint32_t)hdr[23];
|
||||
if (width == 0 || height == 0 || width > 0x7FFF || height > 0x7FFF) {
|
||||
LOG_ERR("PNG", "Implausible PNG dimensions %ux%u: %s", width, height, imagePath.c_str());
|
||||
return false;
|
||||
}
|
||||
|
||||
png->close();
|
||||
out.width = (int16_t)width;
|
||||
out.height = (int16_t)height;
|
||||
return true;
|
||||
}
|
||||
|
||||
@@ -412,11 +502,15 @@ bool PngToFramebufferConverter::decodeToFramebuffer(const std::string& imagePath
|
||||
}
|
||||
ctx.lastDstY = -1; // Reset row tracking
|
||||
|
||||
LOG_DBG("PNG", "PNG %dx%d -> %dx%d (scale %.2f), bpp: %d", ctx.srcWidth, ctx.srcHeight, ctx.dstWidth, ctx.dstHeight,
|
||||
ctx.scale, png->getBpp());
|
||||
|
||||
// PNGdec's getBpp() actually returns per-channel bit depth (1/2/4/8), not bits-per-pixel.
|
||||
// Capture both type and depth here so we can size buffers and validate up front.
|
||||
const int pixelType = png->getPixelType();
|
||||
const int requiredInternal = requiredPngInternalBufferBytes(ctx.srcWidth, pixelType);
|
||||
const int channelDepth = png->getBpp();
|
||||
|
||||
LOG_DBG("PNG", "PNG %dx%d -> %dx%d (scale %.2f), pixelType=%d channelDepth=%d", ctx.srcWidth, ctx.srcHeight,
|
||||
ctx.dstWidth, ctx.dstHeight, ctx.scale, pixelType, channelDepth);
|
||||
|
||||
const int requiredInternal = requiredPngInternalBufferBytes(ctx.srcWidth, pixelType, channelDepth);
|
||||
if (requiredInternal > PNG_MAX_BUFFERED_PIXELS) {
|
||||
LOG_ERR("PNG",
|
||||
"PNG row buffer too small: need %d bytes for width=%d type=%d, configured PNG_MAX_BUFFERED_PIXELS=%d",
|
||||
@@ -426,8 +520,31 @@ bool PngToFramebufferConverter::decodeToFramebuffer(const std::string& imagePath
|
||||
return false;
|
||||
}
|
||||
|
||||
if (png->getBpp() != 8) {
|
||||
warnUnsupportedFeature("bit depth (" + std::to_string(png->getBpp()) + "bpp)", imagePath);
|
||||
// Validate per-channel bit depth for the pixel type. Grayscale/indexed allow 1/2/4/8
|
||||
// (all unpacked by convertLineToGray); the alpha and truecolor variants only allow 8
|
||||
// in practice — PNGdec rejects 16-bit at open() — so anything else here is a surprise.
|
||||
if (channelDepth != 8 && pixelType != PNG_PIXEL_GRAYSCALE && pixelType != PNG_PIXEL_INDEXED) {
|
||||
warnUnsupportedFeature("bit depth (" + std::to_string(channelDepth) + " bits/channel)", imagePath);
|
||||
}
|
||||
|
||||
// Capture tRNS color-key for grayscale / truecolor PNGs. PNGdec sets iHasAlpha=1 on
|
||||
// those types when a tRNS chunk is present and stores the transparent value via
|
||||
// getTransparentColor(). Indexed PNGs use the per-entry alpha array at palette[768..]
|
||||
// and don't need the color-key path.
|
||||
if (png->hasAlpha() && pixelType != PNG_PIXEL_INDEXED &&
|
||||
(pixelType == PNG_PIXEL_GRAYSCALE || pixelType == PNG_PIXEL_TRUECOLOR)) {
|
||||
uint32_t trns = png->getTransparentColor();
|
||||
ctx.hasTrnsKey = true;
|
||||
if (pixelType == PNG_PIXEL_GRAYSCALE) {
|
||||
// PNGdec stores the lower byte of the 2-byte tRNS sample. For 1/2/4 bpp grayscale
|
||||
// this matches the bit-packed sample value we extract during line conversion.
|
||||
ctx.trnsGray = (uint8_t)(trns & 0xFF);
|
||||
} else {
|
||||
// Truecolor: R<<16 | G<<8 | B (each component is the lower byte of a 2-byte sample).
|
||||
ctx.trnsR = (uint8_t)((trns >> 16) & 0xFF);
|
||||
ctx.trnsG = (uint8_t)((trns >> 8) & 0xFF);
|
||||
ctx.trnsB = (uint8_t)(trns & 0xFF);
|
||||
}
|
||||
}
|
||||
|
||||
// Allocate grayscale line buffer on demand (~3.2 KB) - freed after decode
|
||||
|
||||
Reference in New Issue
Block a user