This commit also reduces the number of times ff_sws_init_scale() gets called (only once per bit depth), and the number of times randomize_buffers() gets called (only if the function must be checked).
Benchmarks are only performed on bit depths 8 and 16 (since they are different functions, and not only different constants). --- tests/checkasm/sw_range_convert.c | 94 +++++++++++++++++++++---------- 1 file changed, 63 insertions(+), 31 deletions(-) diff --git a/tests/checkasm/sw_range_convert.c b/tests/checkasm/sw_range_convert.c index e97388d14a..e3e5096729 100644 --- a/tests/checkasm/sw_range_convert.c +++ b/tests/checkasm/sw_range_convert.c @@ -28,6 +28,33 @@ #include "checkasm.h" +static const enum AVPixelFormat pixel_formats[] = { + AV_PIX_FMT_YUV444P, + AV_PIX_FMT_YUV444P9, + AV_PIX_FMT_YUV444P10, + AV_PIX_FMT_YUV444P12, + AV_PIX_FMT_YUV444P14, + AV_PIX_FMT_YUV444P16, +}; + +static void randomize_buffers(int16_t *buf0, int16_t *buf1, int bit_depth, int width) +{ + int32_t *buf0_32 = (int32_t *) buf0; + int32_t *buf1_32 = (int32_t *) buf1; + int mask = (1 << bit_depth) - 1; + int src_shift = bit_depth <= 14 ? 15 - bit_depth : 19 - bit_depth; + for (int i = 0; i < width; i++) { + int32_t r = rnd() & mask; + if (bit_depth == 16) { + buf0_32[i] = r << src_shift; + buf1_32[i] = r << src_shift; + } else { + buf0[i] = r << src_shift; + buf1[i] = r << src_shift; + } + } +} + static void check_lumConvertRange(int from) { const char *func_str = from ? "lumRangeFromJpeg" : "lumRangeToJpeg"; @@ -35,8 +62,8 @@ static void check_lumConvertRange(int from) static const int input_sizes[] = {8, LARGEST_INPUT_SIZE}; struct SwsContext *ctx; - LOCAL_ALIGNED_32(int16_t, dst0, [LARGEST_INPUT_SIZE]); - LOCAL_ALIGNED_32(int16_t, dst1, [LARGEST_INPUT_SIZE]); + LOCAL_ALIGNED_32(int16_t, dst0, [LARGEST_INPUT_SIZE * 2]); + LOCAL_ALIGNED_32(int16_t, dst1, [LARGEST_INPUT_SIZE * 2]); declare_func(void, int16_t *dst, int width); @@ -44,28 +71,31 @@ static void check_lumConvertRange(int from) if (sws_init_context(ctx, NULL, NULL) < 0) fail(); - ctx->srcFormat = AV_PIX_FMT_YUV444P; - ctx->dstFormat = AV_PIX_FMT_YUV444P; ctx->srcRange = from; ctx->dstRange = !from; + for (int pfi = 0; pfi < FF_ARRAY_ELEMS(pixel_formats); pfi++) { + enum AVPixelFormat pix_fmt = pixel_formats[pfi]; + const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt); + int bit_depth = desc->comp[0].depth; + int sample_size = bit_depth == 16 ? sizeof(int32_t) : sizeof(int16_t); + ctx->srcFormat = pix_fmt; + ctx->dstFormat = pix_fmt; + ctx->dstBpc = bit_depth; + ff_sws_init_scale(ctx); for (int dstWi = 0; dstWi < FF_ARRAY_ELEMS(input_sizes); dstWi++) { int width = input_sizes[dstWi]; - for (int i = 0; i < width; i++) { - uint8_t r = rnd(); - dst0[i] = (int16_t) r << 7; - dst1[i] = (int16_t) r << 7; - } - ff_sws_init_scale(ctx); - if (check_func(ctx->lumConvertRange, "%s_%d", func_str, width)) { + if (check_func(ctx->lumConvertRange, "%s%d_%d", func_str, bit_depth, width)) { + randomize_buffers(dst0, dst1, bit_depth, width); call_ref(dst0, width); call_new(dst1, width); - if (memcmp(dst0, dst1, width * sizeof(int16_t))) + if (memcmp(dst0, dst1, width * sample_size)) fail(); - if (width == LARGEST_INPUT_SIZE) + if (width == LARGEST_INPUT_SIZE && (bit_depth == 8 || bit_depth == 16)) bench_new(dst1, width); } } + } sws_freeContext(ctx); } @@ -78,10 +108,10 @@ static void check_chrConvertRange(int from) static const int input_sizes[] = {8, LARGEST_INPUT_SIZE}; struct SwsContext *ctx; - LOCAL_ALIGNED_32(int16_t, dstU0, [LARGEST_INPUT_SIZE]); - LOCAL_ALIGNED_32(int16_t, dstV0, [LARGEST_INPUT_SIZE]); - LOCAL_ALIGNED_32(int16_t, dstU1, [LARGEST_INPUT_SIZE]); - LOCAL_ALIGNED_32(int16_t, dstV1, [LARGEST_INPUT_SIZE]); + LOCAL_ALIGNED_32(int16_t, dstU0, [LARGEST_INPUT_SIZE * 2]); + LOCAL_ALIGNED_32(int16_t, dstV0, [LARGEST_INPUT_SIZE * 2]); + LOCAL_ALIGNED_32(int16_t, dstU1, [LARGEST_INPUT_SIZE * 2]); + LOCAL_ALIGNED_32(int16_t, dstV1, [LARGEST_INPUT_SIZE * 2]); declare_func(void, int16_t *dstU, int16_t *dstV, int width); @@ -89,31 +119,33 @@ static void check_chrConvertRange(int from) if (sws_init_context(ctx, NULL, NULL) < 0) fail(); - ctx->srcFormat = AV_PIX_FMT_YUV444P; - ctx->dstFormat = AV_PIX_FMT_YUV444P; ctx->srcRange = from; ctx->dstRange = !from; + for (int pfi = 0; pfi < FF_ARRAY_ELEMS(pixel_formats); pfi++) { + enum AVPixelFormat pix_fmt = pixel_formats[pfi]; + const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt); + int bit_depth = desc->comp[0].depth; + int sample_size = bit_depth == 16 ? sizeof(int32_t) : sizeof(int16_t); + ctx->srcFormat = pix_fmt; + ctx->dstFormat = pix_fmt; + ctx->dstBpc = bit_depth; + ff_sws_init_scale(ctx); for (int dstWi = 0; dstWi < FF_ARRAY_ELEMS(input_sizes); dstWi++) { int width = input_sizes[dstWi]; - for (int i = 0; i < width; i++) { - uint8_t r = rnd(); - dstU0[i] = (int16_t) r << 7; - dstV0[i] = (int16_t) r << 7; - dstU1[i] = (int16_t) r << 7; - dstV1[i] = (int16_t) r << 7; - } - ff_sws_init_scale(ctx); - if (check_func(ctx->chrConvertRange, "%s_%d", func_str, width)) { + if (check_func(ctx->chrConvertRange, "%s%d_%d", func_str, bit_depth, width)) { + randomize_buffers(dstU0, dstU1, bit_depth, width); + randomize_buffers(dstV0, dstV1, bit_depth, width); call_ref(dstU0, dstV0, width); call_new(dstU1, dstV1, width); - if (memcmp(dstU0, dstU1, width * sizeof(int16_t)) || - memcmp(dstV0, dstV1, width * sizeof(int16_t))) + if (memcmp(dstU0, dstU1, width * sample_size) || + memcmp(dstV0, dstV1, width * sample_size)) fail(); - if (width == LARGEST_INPUT_SIZE) + if (width == LARGEST_INPUT_SIZE && (bit_depth == 8 || bit_depth == 16)) bench_new(dstU1, dstV1, width); } } + } sws_freeContext(ctx); } -- 2.30.2 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".