From: Niklas Haas <g...@haasn.dev> Signed-off-by: Niklas Haas <g...@haasn.dev> Sponsored-by: Sovereign Tech Fund --- tests/checkasm/sw_rgb.c | 308 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 308 insertions(+)
diff --git a/tests/checkasm/sw_rgb.c b/tests/checkasm/sw_rgb.c index 05370c1e41..82621b6adb 100644 --- a/tests/checkasm/sw_rgb.c +++ b/tests/checkasm/sw_rgb.c @@ -453,6 +453,305 @@ static void check_rgb_to_uv(SwsContext *sws) } } +static const int packed_rgb_fmts[] = { + AV_PIX_FMT_RGB24, + AV_PIX_FMT_BGR24, + AV_PIX_FMT_ARGB, + AV_PIX_FMT_RGBA, + AV_PIX_FMT_ABGR, + AV_PIX_FMT_BGRA, + AV_PIX_FMT_RGB48BE, + AV_PIX_FMT_RGB48LE, + AV_PIX_FMT_RGB565BE, + AV_PIX_FMT_RGB565LE, + AV_PIX_FMT_RGB555BE, + AV_PIX_FMT_RGB555LE, + AV_PIX_FMT_BGR565BE, + AV_PIX_FMT_BGR565LE, + AV_PIX_FMT_BGR555BE, + AV_PIX_FMT_BGR555LE, + AV_PIX_FMT_RGB444LE, + AV_PIX_FMT_RGB444BE, + AV_PIX_FMT_BGR444LE, + AV_PIX_FMT_BGR444BE, + AV_PIX_FMT_BGR48BE, + AV_PIX_FMT_BGR48LE, + AV_PIX_FMT_RGBA64BE, + AV_PIX_FMT_RGBA64LE, + AV_PIX_FMT_BGRA64BE, + AV_PIX_FMT_BGRA64LE, + AV_PIX_FMT_RGB8, + AV_PIX_FMT_BGR8, + AV_PIX_FMT_RGB4, + AV_PIX_FMT_BGR4, + AV_PIX_FMT_RGB4_BYTE, + AV_PIX_FMT_BGR4_BYTE, +}; + +#define INPUT_SIZE 512 + +static void check_yuv2packed1(void) +{ + SwsContext *sws; + SwsInternal *c; + + static const int alpha_values[] = {0, 2048, 4096}; + + declare_func_emms(AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT, + void, SwsInternal *c, const int16_t *lumSrc, + const int16_t *chrUSrc[2], const int16_t *chrVSrc[2], + const int16_t *alpSrc, uint8_t *dest, + int dstW, int uvalpha, int y); + + const int16_t *luma; + const int16_t *chru[2]; + const int16_t *chrv[2]; + const int16_t *alpha; + + LOCAL_ALIGNED_8(int32_t, src_y, [INPUT_SIZE]); + LOCAL_ALIGNED_8(int32_t, src_u, [INPUT_SIZE]); + LOCAL_ALIGNED_8(int32_t, src_v, [INPUT_SIZE]); + LOCAL_ALIGNED_8(int32_t, src_a, [INPUT_SIZE]); + + LOCAL_ALIGNED_8(uint8_t, dst0, [INPUT_SIZE * sizeof(int32_t[4])]); + LOCAL_ALIGNED_8(uint8_t, dst1, [INPUT_SIZE * sizeof(int32_t[4])]); + + randomize_buffers((uint8_t*)src_y, INPUT_SIZE * sizeof(int32_t)); + randomize_buffers((uint8_t*)src_u, INPUT_SIZE * sizeof(int32_t)); + randomize_buffers((uint8_t*)src_v, INPUT_SIZE * sizeof(int32_t)); + randomize_buffers((uint8_t*)src_a, INPUT_SIZE * sizeof(int32_t)); + + /* Limit to 14 bit input range */ + for (int i = 0; i < INPUT_SIZE; i++) { + src_y[i] &= 0x3FFF3FFF; + src_a[i] &= 0x3FFF3FFF; + src_u[i] &= 0x3FFF3FFF; + src_v[i] &= 0x3FFF3FFF; + src_u[i + INPUT_SIZE] &= 0x3FFF3FFF; + src_v[i + INPUT_SIZE] &= 0x3FFF3FFF; + } + + luma = (int16_t *)src_y; + alpha = (int16_t *)src_a; + for (int i = 0; i < 2; i++) { + chru[i] = (int16_t *)(src_u + i*INPUT_SIZE); + chrv[i] = (int16_t *)(src_v + i*INPUT_SIZE); + } + + for (int fmi = 0; fmi < FF_ARRAY_ELEMS(packed_rgb_fmts); fmi++) { + const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(packed_rgb_fmts[fmi]); + int line_size = INPUT_SIZE * desc->comp[0].step; + if (desc->flags & AV_PIX_FMT_FLAG_BITSTREAM) + line_size = AV_CEIL_RSHIFT(line_size, 3); + + sws = sws_getContext(MAX_LINE_SIZE, MAX_LINE_SIZE, AV_PIX_FMT_YUV420P, + MAX_LINE_SIZE, MAX_LINE_SIZE, packed_rgb_fmts[fmi], + SWS_ACCURATE_RND | SWS_BITEXACT, NULL, NULL, NULL); + if (!sws) + fail(); + + c = sws_internal(sws); + + for (int ai = 0; ai < FF_ARRAY_ELEMS(alpha_values); ai++) { + const int chr_alpha = alpha_values[ai]; + if (check_func(c->yuv2packed1, "yuv2%s_1_%d_%d", desc->name, chr_alpha, INPUT_SIZE)) { + memset(dst0, 0xFF, INPUT_SIZE * sizeof(int32_t[4])); + memset(dst1, 0xFF, INPUT_SIZE * sizeof(int32_t[4])); + + call_ref(c, luma, chru, chrv, alpha, dst0, INPUT_SIZE, chr_alpha, 0); + call_new(c, luma, chru, chrv, alpha, dst1, INPUT_SIZE, chr_alpha, 0); + + if (memcmp(dst0, dst1, line_size)) + fail(); + + bench_new(c, luma, chru, chrv, alpha, dst1, INPUT_SIZE, chr_alpha, 0); + } + } + } +} + +static void check_yuv2packed2(void) +{ + SwsContext *sws; + SwsInternal *c; + static const int alpha_values[] = {0, 2048, 4096}; + + declare_func_emms(AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT, + void, SwsInternal *c, const int16_t *lumSrc[2], + const int16_t *chrUSrc[2], const int16_t *chrVSrc[2], + const int16_t *alpSrc[2], uint8_t *dest, + int dstW, int yalpha, int uvalpha, int y); + + const int16_t *luma[2]; + const int16_t *chru[2]; + const int16_t *chrv[2]; + const int16_t *alpha[2]; + + LOCAL_ALIGNED_8(int32_t, src_y, [2 * INPUT_SIZE]); + LOCAL_ALIGNED_8(int32_t, src_u, [2 * INPUT_SIZE]); + LOCAL_ALIGNED_8(int32_t, src_v, [2 * INPUT_SIZE]); + LOCAL_ALIGNED_8(int32_t, src_a, [2 * INPUT_SIZE]); + + LOCAL_ALIGNED_8(uint8_t, dst0, [INPUT_SIZE * sizeof(int32_t[4])]); + LOCAL_ALIGNED_8(uint8_t, dst1, [INPUT_SIZE * sizeof(int32_t[4])]); + + randomize_buffers((uint8_t*)src_y, 2 * INPUT_SIZE * sizeof(int32_t)); + randomize_buffers((uint8_t*)src_u, 2 * INPUT_SIZE * sizeof(int32_t)); + randomize_buffers((uint8_t*)src_v, 2 * INPUT_SIZE * sizeof(int32_t)); + randomize_buffers((uint8_t*)src_a, 2 * INPUT_SIZE * sizeof(int32_t)); + + /* Limit to 14 bit input range */ + for (int i = 0; i < 2 * INPUT_SIZE; i++) { + src_y[i] &= 0x3FFF3FFF; + src_u[i] &= 0x3FFF3FFF; + src_v[i] &= 0x3FFF3FFF; + src_a[i] &= 0x3FFF3FFF; + } + + for (int i = 0; i < 2; i++) { + luma[i] = (int16_t *)(src_y + i*INPUT_SIZE); + chru[i] = (int16_t *)(src_u + i*INPUT_SIZE); + chrv[i] = (int16_t *)(src_v + i*INPUT_SIZE); + alpha[i] = (int16_t *)(src_a + i*INPUT_SIZE); + } + + for (int fmi = 0; fmi < FF_ARRAY_ELEMS(packed_rgb_fmts); fmi++) { + const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(packed_rgb_fmts[fmi]); + int line_size = INPUT_SIZE * desc->comp[0].step; + if (desc->flags & AV_PIX_FMT_FLAG_BITSTREAM) + line_size = AV_CEIL_RSHIFT(line_size, 3); + + sws = sws_getContext(MAX_LINE_SIZE, MAX_LINE_SIZE, AV_PIX_FMT_YUV420P, + MAX_LINE_SIZE, MAX_LINE_SIZE, packed_rgb_fmts[fmi], + SWS_ACCURATE_RND | SWS_BITEXACT, NULL, NULL, NULL); + if (!sws) + fail(); + + c = sws_internal(sws); + + for (int ai = 0; ai < FF_ARRAY_ELEMS(alpha_values); ai++) { + const int lum_alpha = alpha_values[ai]; + const int chr_alpha = alpha_values[ai]; + if (check_func(c->yuv2packed2, "yuv2%s_2_%d_%d", desc->name, lum_alpha, INPUT_SIZE)) { + memset(dst0, 0xFF, INPUT_SIZE * sizeof(int32_t[4])); + memset(dst1, 0xFF, INPUT_SIZE * sizeof(int32_t[4])); + + call_ref(c, luma, chru, chrv, alpha, dst0, INPUT_SIZE, lum_alpha, chr_alpha, 0); + call_new(c, luma, chru, chrv, alpha, dst1, INPUT_SIZE, lum_alpha, chr_alpha, 0); + + if (memcmp(dst0, dst1, line_size)) + fail(); + + bench_new(c, luma, chru, chrv, alpha, dst1, INPUT_SIZE, lum_alpha, chr_alpha, 0); + } + } + } +} + +static void check_yuv2packedX(void) +{ + SwsContext *sws; + SwsInternal *c; + +#define LARGEST_FILTER 16 + static const int filter_sizes[] = {2, 16}; + + declare_func_emms(AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT, + void, SwsInternal *c, const int16_t *lumFilter, + const int16_t **lumSrcx, int lumFilterSize, + const int16_t *chrFilter, const int16_t **chrUSrcx, + const int16_t **chrVSrcx, int chrFilterSize, + const int16_t **alpSrcx, uint8_t *dest, + int dstW, int y); + + const int16_t *luma[LARGEST_FILTER]; + const int16_t *chru[LARGEST_FILTER]; + const int16_t *chrv[LARGEST_FILTER]; + const int16_t *alpha[LARGEST_FILTER]; + + LOCAL_ALIGNED_8(int16_t, luma_filter, [LARGEST_FILTER]); + LOCAL_ALIGNED_8(int16_t, chr_filter, [LARGEST_FILTER]); + + LOCAL_ALIGNED_8(int32_t, src_y, [LARGEST_FILTER * INPUT_SIZE]); + LOCAL_ALIGNED_8(int32_t, src_u, [LARGEST_FILTER * INPUT_SIZE]); + LOCAL_ALIGNED_8(int32_t, src_v, [LARGEST_FILTER * INPUT_SIZE]); + LOCAL_ALIGNED_8(int32_t, src_a, [LARGEST_FILTER * INPUT_SIZE]); + + LOCAL_ALIGNED_8(uint8_t, dst0, [INPUT_SIZE * sizeof(int32_t[4])]); + LOCAL_ALIGNED_8(uint8_t, dst1, [INPUT_SIZE * sizeof(int32_t[4])]); + + randomize_buffers((uint8_t*)src_y, LARGEST_FILTER * INPUT_SIZE * sizeof(int32_t)); + randomize_buffers((uint8_t*)src_u, LARGEST_FILTER * INPUT_SIZE * sizeof(int32_t)); + randomize_buffers((uint8_t*)src_v, LARGEST_FILTER * INPUT_SIZE * sizeof(int32_t)); + randomize_buffers((uint8_t*)src_a, LARGEST_FILTER * INPUT_SIZE * sizeof(int32_t)); + + /* Limit to 14 bit input range */ + for (int i = 0; i < LARGEST_FILTER * INPUT_SIZE; i++) { + src_y[i] &= 0x3FFF3FFF; + src_u[i] &= 0x3FFF3FFF; + src_v[i] &= 0x3FFF3FFF; + src_a[i] &= 0x3FFF3FFF; + } + + for (int i = 0; i < LARGEST_FILTER; i++) { + luma[i] = (int16_t *)(src_y + i*INPUT_SIZE); + chru[i] = (int16_t *)(src_u + i*INPUT_SIZE); + chrv[i] = (int16_t *)(src_v + i*INPUT_SIZE); + alpha[i] = (int16_t *)(src_a + i*INPUT_SIZE); + } + + for (int fmi = 0; fmi < FF_ARRAY_ELEMS(packed_rgb_fmts); fmi++) { + const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(packed_rgb_fmts[fmi]); + int line_size = INPUT_SIZE * desc->comp[0].step; + if (desc->flags & AV_PIX_FMT_FLAG_BITSTREAM) + line_size = AV_CEIL_RSHIFT(line_size, 3); + + sws = sws_getContext(MAX_LINE_SIZE, MAX_LINE_SIZE, AV_PIX_FMT_YUV420P, + MAX_LINE_SIZE, MAX_LINE_SIZE, packed_rgb_fmts[fmi], + SWS_ACCURATE_RND | SWS_BITEXACT, NULL, NULL, NULL); + if (!sws) + fail(); + + c = sws_internal(sws); + + for (int fsi = 0; fsi < FF_ARRAY_ELEMS(filter_sizes); fsi++) { + const int luma_filter_size = filter_sizes[fsi]; + const int chr_filter_size = filter_sizes[fsi]; + + for (int i = 0; i < luma_filter_size; i++) + luma_filter[i] = -((1 << 12) / (luma_filter_size - 1)); + luma_filter[rnd() % luma_filter_size] = (1 << 13) - 1; + + for (int i = 0; i < chr_filter_size; i++) + chr_filter[i] = -((1 << 12) / (chr_filter_size - 1)); + chr_filter[rnd() % chr_filter_size] = (1 << 13) - 1; + + if (check_func(c->yuv2packedX, "yuv2%s_X_%d_%d", desc->name, luma_filter_size, INPUT_SIZE)) { + memset(dst0, 0xFF, INPUT_SIZE * sizeof(int32_t[4])); + memset(dst1, 0xFF, INPUT_SIZE * sizeof(int32_t[4])); + + call_ref(c, luma_filter, luma, luma_filter_size, + chr_filter, chru, chrv, chr_filter_size, + alpha, dst0, INPUT_SIZE, 0); + + call_new(c, luma_filter, luma, luma_filter_size, + chr_filter, chru, chrv, chr_filter_size, + alpha, dst1, INPUT_SIZE, 0); + + if (memcmp(dst0, dst1, line_size)) + fail(); + + bench_new(c, luma_filter, luma, luma_filter_size, + chr_filter, chru, chrv, chr_filter_size, + alpha, dst1, INPUT_SIZE, 0); + } + } + } +} + +#undef INPUT_SIZE +#undef LARGEST_FILTER + void checkasm_check_sw_rgb(void) { SwsContext *sws; @@ -499,4 +798,13 @@ void checkasm_check_sw_rgb(void) report("rgb24toyv12"); sws_freeContext(sws); + + check_yuv2packed1(); + report("yuv2packed1"); + + check_yuv2packed2(); + report("yuv2packed2"); + + check_yuv2packedX(); + report("yuv2packedX"); } -- 2.47.0 _______________________________________________ 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".