From 9c10ea18c56773ffbe8000b0c996ac3a2cb0eaf1 Mon Sep 17 00:00:00 2001 From: Robert Mader Date: Sun, 26 Jul 2026 12:17:23 +0200 Subject: [PATCH] RFC: ConvertToI420: Add WithSrcStride variant Some users - notably the libwebrtcs VideoCaptureImpl::IncomingFrame() - are currently limited by the fact that the source stride is derived from the format and buffer width, preventing it from supporting padded buffers. Padded buffers are useful if camera providers have hardware limitations regarding the stride, sometimes requiring alignments to e.g. 64/128/256 bytes. All the actual conversion implementations already support dedicated source strides, thus let's add a variant of ConvertToI420() - ConvertToI420WithSrcStride() - allowing users to pass through an explicit stride. Bug:538875176 Change-Id: I323729a315b26e96d7dbb93e68f04fbdd8120afe --- include/libyuv/convert.h | 41 +++++++++ source/convert_to_i420.cc | 185 +++++++++++++++++++++++++++----------- 2 files changed, 175 insertions(+), 51 deletions(-) diff --git a/include/libyuv/convert.h b/include/libyuv/convert.h index 4c4f8f1f9..5692aa93d 100644 --- a/include/libyuv/convert.h +++ b/include/libyuv/convert.h @@ -1164,6 +1164,47 @@ int ConvertToI420(const uint8_t* sample, enum RotationMode rotation, uint32_t fourcc); +// Convert camera sample to I420 with cropping, rotation and vertical flip. +// "src_size" is needed to parse MJPG. +// "dst_stride_y" number of bytes in a row of the dst_y plane. +// Normally this would be the same as dst_width, with recommended alignment +// to 16 bytes for better efficiency. +// If rotation of 90 or 270 is used, stride is affected. The caller should +// allocate the I420 buffer according to rotation. +// "dst_stride_u" number of bytes in a row of the dst_u plane. +// Normally this would be the same as (dst_width + 1) / 2, with +// recommended alignment to 16 bytes for better efficiency. +// If rotation of 90 or 270 is used, stride is affected. +// "crop_x" and "crop_y" are starting position for cropping. +// To center, crop_x = (src_width - dst_width) / 2 +// crop_y = (src_height - dst_height) / 2 +// "src_width" / "src_height" is size of src_frame in pixels. +// "src_height" can be negative indicating a vertically flipped image source. +// "crop_width" / "crop_height" is the size to crop the src to. +// Must be less than or equal to src_width/src_height +// Cropping parameters are pre-rotation. +// "rotation" can be 0, 90, 180 or 270. +// "fourcc" is a fourcc. ie 'I420', 'YUY2' +// Returns 0 for successful; -1 for invalid parameter. Non-zero for failure. +LIBYUV_API +int ConvertToI420WithSrcStride(const uint8_t* sample, + size_t sample_size, + uint8_t* dst_y, + int dst_stride_y, + uint8_t* dst_u, + int dst_stride_u, + uint8_t* dst_v, + int dst_stride_v, + int crop_x, + int crop_y, + int src_width, + int src_stride, + int src_height, + int crop_width, + int crop_height, + enum RotationMode rotation, + uint32_t fourcc); + #ifdef __cplusplus } // extern "C" } // namespace libyuv diff --git a/source/convert_to_i420.cc b/source/convert_to_i420.cc index baa4a9494..47f469809 100644 --- a/source/convert_to_i420.cc +++ b/source/convert_to_i420.cc @@ -44,6 +44,89 @@ int ConvertToI420(const uint8_t* sample, int crop_height, enum RotationMode rotation, uint32_t fourcc) { + uint32_t format = CanonicalFourCC(fourcc); + int src_stride; + + switch (format) { + case FOURCC_YUY2: + case FOURCC_UYVY: + case FOURCC_RGBP: + case FOURCC_RGBO: + case FOURCC_R444: + src_stride = src_width * 2; + break; + case FOURCC_24BG: + case FOURCC_RAW: + src_stride = src_width * 3; + break; + case FOURCC_ARGB: + case FOURCC_BGRA: + case FOURCC_ABGR: + case FOURCC_RGBA: + src_stride = src_width * 4; + break; + case FOURCC_I400: + case FOURCC_NV12: + case FOURCC_NV21: + case FOURCC_I420: + case FOURCC_YV12: + case FOURCC_I422: + case FOURCC_YV16: + case FOURCC_I444: + case FOURCC_YV24: + src_stride = src_width; + break; +#ifdef HAVE_JPEG + case FOURCC_MJPG: + src_stride = 0; + break; +#endif + default: + return -1; + } + + return ConvertToI420WithSrcStride(sample, + sample_size, + dst_y, + dst_stride_y, + dst_u, + dst_stride_u, + dst_v, + dst_stride_v, + crop_x, + crop_y, + src_width, + src_stride, + src_height, + crop_width, + crop_height, + rotation, + fourcc); +} + +// Convert camera sample to I420 with cropping, rotation and vertical flip. +// src_width is used for source stride computation +// src_height is used to compute location of planes, and indicate inversion +// sample_size is measured in bytes and is the size of the frame. +// With MJPEG it is the compressed size of the frame. +LIBYUV_API +int ConvertToI420WithSrcStride(const uint8_t* sample, + size_t sample_size, + uint8_t* dst_y, + int dst_stride_y, + uint8_t* dst_u, + int dst_stride_u, + uint8_t* dst_v, + int dst_stride_v, + int crop_x, + int crop_y, + int src_width, + int src_stride, + int src_height, + int crop_width, + int crop_height, + enum RotationMode rotation, + uint32_t fourcc) { if (src_height == INT_MIN || crop_height == INT_MIN) { return -1; } @@ -76,7 +159,7 @@ int ConvertToI420(const uint8_t* sample, uint8_t* rotate_buffer = NULL; const int inv_crop_height = (src_height < 0) ? -abs_crop_height : abs_crop_height; - int aligned_src_width = (src_width + 1) & ~1; + int aligned_src_stride = (src_stride + 1) & ~1; // One pass rotation is available for some formats. For the rest, convert // to I420 (with optional vertical flipping) into a temporary I420 buffer, @@ -109,8 +192,8 @@ int ConvertToI420(const uint8_t* sample, uint8_t* v = (crop_x & 1) ? dst_u : dst_v; int stride_u = (crop_x & 1) ? dst_stride_v : dst_stride_u; int stride_v = (crop_x & 1) ? dst_stride_u : dst_stride_v; - src = sample + ((ptrdiff_t)aligned_src_width * crop_y + crop_x) * 2; - r = YUY2ToI420(src, aligned_src_width * 2, dst_y, dst_stride_y, u, + src = sample + ((ptrdiff_t)aligned_src_stride * crop_y + crop_x); + r = YUY2ToI420(src, aligned_src_stride, dst_y, dst_stride_y, u, stride_u, v, stride_v, crop_width, inv_crop_height); break; } @@ -119,156 +202,156 @@ int ConvertToI420(const uint8_t* sample, uint8_t* v = (crop_x & 1) ? dst_u : dst_v; int stride_u = (crop_x & 1) ? dst_stride_v : dst_stride_u; int stride_v = (crop_x & 1) ? dst_stride_u : dst_stride_v; - src = sample + ((ptrdiff_t)aligned_src_width * crop_y + crop_x) * 2; - r = UYVYToI420(src, aligned_src_width * 2, dst_y, dst_stride_y, u, + src = sample + ((ptrdiff_t)aligned_src_stride * crop_y + crop_x); + r = UYVYToI420(src, aligned_src_stride, dst_y, dst_stride_y, u, stride_u, v, stride_v, crop_width, inv_crop_height); break; } case FOURCC_RGBP: - src = sample + ((ptrdiff_t)src_width * crop_y + crop_x) * 2; - r = RGB565ToI420(src, src_width * 2, dst_y, dst_stride_y, dst_u, + src = sample + ((ptrdiff_t)src_stride * crop_y + crop_x); + r = RGB565ToI420(src, src_stride, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height); break; case FOURCC_RGBO: - src = sample + ((ptrdiff_t)src_width * crop_y + crop_x) * 2; - r = ARGB1555ToI420(src, src_width * 2, dst_y, dst_stride_y, dst_u, + src = sample + ((ptrdiff_t)src_stride * crop_y + crop_x); + r = ARGB1555ToI420(src, src_stride, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height); break; case FOURCC_R444: - src = sample + ((ptrdiff_t)src_width * crop_y + crop_x) * 2; - r = ARGB4444ToI420(src, src_width * 2, dst_y, dst_stride_y, dst_u, + src = sample + ((ptrdiff_t)src_stride * crop_y + crop_x); + r = ARGB4444ToI420(src, src_stride, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height); break; case FOURCC_24BG: - src = sample + ((ptrdiff_t)src_width * crop_y + crop_x) * 3; - r = RGB24ToI420(src, src_width * 3, dst_y, dst_stride_y, dst_u, + src = sample + ((ptrdiff_t)src_stride * crop_y + crop_x); + r = RGB24ToI420(src, src_stride, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height); break; case FOURCC_RAW: - src = sample + ((ptrdiff_t)src_width * crop_y + crop_x) * 3; - r = RAWToI420(src, src_width * 3, dst_y, dst_stride_y, dst_u, + src = sample + ((ptrdiff_t)src_stride * crop_y + crop_x); + r = RAWToI420(src, src_stride, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height); break; case FOURCC_ARGB: - src = sample + ((ptrdiff_t)src_width * crop_y + crop_x) * 4; - r = ARGBToI420(src, src_width * 4, dst_y, dst_stride_y, dst_u, + src = sample + ((ptrdiff_t)src_stride * crop_y + crop_x); + r = ARGBToI420(src, src_stride, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height); break; case FOURCC_BGRA: - src = sample + ((ptrdiff_t)src_width * crop_y + crop_x) * 4; - r = BGRAToI420(src, src_width * 4, dst_y, dst_stride_y, dst_u, + src = sample + ((ptrdiff_t)src_stride * crop_y + crop_x); + r = BGRAToI420(src, src_stride, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height); break; case FOURCC_ABGR: - src = sample + ((ptrdiff_t)src_width * crop_y + crop_x) * 4; - r = ABGRToI420(src, src_width * 4, dst_y, dst_stride_y, dst_u, + src = sample + ((ptrdiff_t)src_stride * crop_y + crop_x); + r = ABGRToI420(src, src_stride, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height); break; case FOURCC_RGBA: - src = sample + ((ptrdiff_t)src_width * crop_y + crop_x) * 4; - r = RGBAToI420(src, src_width * 4, dst_y, dst_stride_y, dst_u, + src = sample + ((ptrdiff_t)src_stride * crop_y + crop_x); + r = RGBAToI420(src, src_stride, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height); break; // TODO(fbarchard): Add AR30 and AB30 case FOURCC_I400: - src = sample + (ptrdiff_t)src_width * crop_y + crop_x; - r = I400ToI420(src, src_width, dst_y, dst_stride_y, dst_u, dst_stride_u, + src = sample + (ptrdiff_t)src_stride * crop_y + crop_x; + r = I400ToI420(src, src_stride, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height); break; // Biplanar formats case FOURCC_NV12: - src = sample + ((ptrdiff_t)src_width * crop_y + crop_x); - src_uv = sample + ((ptrdiff_t)src_width * abs_src_height) + - ((ptrdiff_t)(crop_y / 2) * aligned_src_width) + + src = sample + ((ptrdiff_t)src_stride * crop_y + crop_x); + src_uv = sample + ((ptrdiff_t)src_stride * abs_src_height) + + ((ptrdiff_t)(crop_y / 2) * aligned_src_stride) + ((crop_x / 2) * 2); - r = NV12ToI420Rotate(src, src_width, src_uv, aligned_src_width, dst_y, + r = NV12ToI420Rotate(src, src_stride, src_uv, aligned_src_stride, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height, rotation); break; case FOURCC_NV21: - src = sample + ((ptrdiff_t)src_width * crop_y + crop_x); - src_uv = sample + ((ptrdiff_t)src_width * abs_src_height) + - ((ptrdiff_t)(crop_y / 2) * aligned_src_width) + + src = sample + ((ptrdiff_t)src_stride * crop_y + crop_x); + src_uv = sample + ((ptrdiff_t)src_stride * abs_src_height) + + ((ptrdiff_t)(crop_y / 2) * aligned_src_stride) + ((crop_x / 2) * 2); // Call NV12 but with dst_u and dst_v parameters swapped. - r = NV12ToI420Rotate(src, src_width, src_uv, aligned_src_width, dst_y, + r = NV12ToI420Rotate(src, src_stride, src_uv, aligned_src_stride, dst_y, dst_stride_y, dst_v, dst_stride_v, dst_u, dst_stride_u, crop_width, inv_crop_height, rotation); break; // Triplanar formats case FOURCC_I420: case FOURCC_YV12: { - const uint8_t* src_y = sample + ((ptrdiff_t)src_width * crop_y + crop_x); + const uint8_t* src_y = sample + ((ptrdiff_t)src_stride * crop_y + crop_x); const uint8_t* src_u; const uint8_t* src_v; int halfwidth = (src_width + 1) / 2; int halfheight = (abs_src_height + 1) / 2; if (format == FOURCC_YV12) { - src_v = sample + (ptrdiff_t)src_width * abs_src_height + + src_v = sample + (ptrdiff_t)src_stride * abs_src_height + (ptrdiff_t)halfwidth * (crop_y / 2) + (crop_x / 2); - src_u = sample + (ptrdiff_t)src_width * abs_src_height + + src_u = sample + (ptrdiff_t)src_stride * abs_src_height + halfwidth * ((ptrdiff_t)halfheight + (crop_y / 2)) + (crop_x / 2); } else { - src_u = sample + (ptrdiff_t)src_width * abs_src_height + + src_u = sample + (ptrdiff_t)src_stride * abs_src_height + (ptrdiff_t)halfwidth * (crop_y / 2) + (crop_x / 2); - src_v = sample + (ptrdiff_t)src_width * abs_src_height + + src_v = sample + (ptrdiff_t)src_stride * abs_src_height + halfwidth * ((ptrdiff_t)halfheight + (crop_y / 2)) + (crop_x / 2); } - r = I420Rotate(src_y, src_width, src_u, halfwidth, src_v, halfwidth, + r = I420Rotate(src_y, src_stride, src_u, halfwidth, src_v, halfwidth, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height, rotation); break; } case FOURCC_I422: case FOURCC_YV16: { - const uint8_t* src_y = sample + (ptrdiff_t)src_width * crop_y + crop_x; + const uint8_t* src_y = sample + (ptrdiff_t)src_stride * crop_y + crop_x; const uint8_t* src_u; const uint8_t* src_v; int halfwidth = (src_width + 1) / 2; if (format == FOURCC_YV16) { - src_v = sample + (ptrdiff_t)src_width * abs_src_height + + src_v = sample + (ptrdiff_t)src_stride * abs_src_height + (ptrdiff_t)halfwidth * crop_y + (crop_x / 2); - src_u = sample + (ptrdiff_t)src_width * abs_src_height + + src_u = sample + (ptrdiff_t)src_stride * abs_src_height + halfwidth * ((ptrdiff_t)abs_src_height + crop_y) + (crop_x / 2); } else { - src_u = sample + (ptrdiff_t)src_width * abs_src_height + + src_u = sample + (ptrdiff_t)src_stride * abs_src_height + (ptrdiff_t)halfwidth * crop_y + (crop_x / 2); - src_v = sample + (ptrdiff_t)src_width * abs_src_height + + src_v = sample + (ptrdiff_t)src_stride * abs_src_height + halfwidth * ((ptrdiff_t)abs_src_height + crop_y) + (crop_x / 2); } - r = I422ToI420(src_y, src_width, src_u, halfwidth, src_v, halfwidth, + r = I422ToI420(src_y, src_stride, src_u, halfwidth, src_v, halfwidth, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height); break; } case FOURCC_I444: case FOURCC_YV24: { - const uint8_t* src_y = sample + (ptrdiff_t)src_width * crop_y + crop_x; + const uint8_t* src_y = sample + (ptrdiff_t)src_stride * crop_y + crop_x; const uint8_t* src_u; const uint8_t* src_v; if (format == FOURCC_YV24) { src_v = - sample + src_width * ((ptrdiff_t)abs_src_height + crop_y) + crop_x; - src_u = sample + src_width * ((ptrdiff_t)abs_src_height * 2 + crop_y) + + sample + src_stride * ((ptrdiff_t)abs_src_height + crop_y) + crop_x; + src_u = sample + src_stride * ((ptrdiff_t)abs_src_height * 2 + crop_y) + crop_x; } else { src_u = - sample + src_width * ((ptrdiff_t)abs_src_height + crop_y) + crop_x; - src_v = sample + src_width * ((ptrdiff_t)abs_src_height * 2 + crop_y) + + sample + src_stride * ((ptrdiff_t)abs_src_height + crop_y) + crop_x; + src_v = sample + src_stride * ((ptrdiff_t)abs_src_height * 2 + crop_y) + crop_x; } - r = I444ToI420(src_y, src_width, src_u, src_width, src_v, src_width, + r = I444ToI420(src_y, src_stride, src_u, src_stride, src_v, src_stride, dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v, crop_width, inv_crop_height); break;