* [FFmpeg-devel] [PATCH 1/6] swscale/yuv2rgb: fix conversion for widths not aligned to 8
@ 2024-06-16 22:28 Ramiro Polla
2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 2/6] swscale/yuv2rgb: add macros to simplify code generation Ramiro Polla
` (4 more replies)
0 siblings, 5 replies; 14+ messages in thread
From: Ramiro Polla @ 2024-06-16 22:28 UTC (permalink / raw)
To: ffmpeg-devel
The C code for some pixel formats (rgb555, rgb565, rgb444, and monob)
was not converting the last pixels on widths not aligned to 8.
NOTE: the last pixel for odd widths is still not converted for any of
the pixel formats in the C code for yuv2rgb except for monob.
---
libswscale/yuv2rgb.c | 101 +++++++++++++++++++++++++++++++++++++++----
1 file changed, 93 insertions(+), 8 deletions(-)
diff --git a/libswscale/yuv2rgb.c b/libswscale/yuv2rgb.c
index c1d6236f37..e641c765c7 100644
--- a/libswscale/yuv2rgb.c
+++ b/libswscale/yuv2rgb.c
@@ -172,10 +172,6 @@ const int *sws_getCoefficients(int colorspace)
return srcSliceH; \
}
-#define CLOSEYUV2RGBFUNC(dst_delta) \
- ENDYUV2RGBLINE(dst_delta, 0) \
- ENDYUV2RGBFUNC()
-
YUV2RGBFUNC(yuv2rgb_c_48, uint8_t, 0)
LOADCHROMA(0);
PUTRGB48(dst_1, py_1, 0);
@@ -432,7 +428,27 @@ YUV2RGBFUNC(yuv2rgb_c_16_ordered_dither, uint16_t, 0)
LOADCHROMA(3);
PUTRGB16(dst_2, py_2, 3, 6 + 8);
PUTRGB16(dst_1, py_1, 3, 6);
-CLOSEYUV2RGBFUNC(8)
+ENDYUV2RGBLINE(8, 0)
+ const uint8_t *d16 = ff_dither_2x2_8[y & 1];
+ const uint8_t *e16 = ff_dither_2x2_4[y & 1];
+ const uint8_t *f16 = ff_dither_2x2_8[(y & 1)^1];
+
+ LOADCHROMA(0);
+ PUTRGB16(dst_1, py_1, 0, 0);
+ PUTRGB16(dst_2, py_2, 0, 0 + 8);
+
+ LOADCHROMA(1);
+ PUTRGB16(dst_2, py_2, 1, 2 + 8);
+ PUTRGB16(dst_1, py_1, 1, 2);
+ENDYUV2RGBLINE(8, 1)
+ const uint8_t *d16 = ff_dither_2x2_8[y & 1];
+ const uint8_t *e16 = ff_dither_2x2_4[y & 1];
+ const uint8_t *f16 = ff_dither_2x2_8[(y & 1)^1];
+
+ LOADCHROMA(0);
+ PUTRGB16(dst_1, py_1, 0, 0);
+ PUTRGB16(dst_2, py_2, 0, 0 + 8);
+ENDYUV2RGBFUNC()
YUV2RGBFUNC(yuv2rgb_c_15_ordered_dither, uint16_t, 0)
const uint8_t *d16 = ff_dither_2x2_8[y & 1];
@@ -462,7 +478,25 @@ YUV2RGBFUNC(yuv2rgb_c_15_ordered_dither, uint16_t, 0)
LOADCHROMA(3);
PUTRGB15(dst_2, py_2, 3, 6 + 8);
PUTRGB15(dst_1, py_1, 3, 6);
-CLOSEYUV2RGBFUNC(8)
+ENDYUV2RGBLINE(8, 0)
+ const uint8_t *d16 = ff_dither_2x2_8[y & 1];
+ const uint8_t *e16 = ff_dither_2x2_8[(y & 1)^1];
+
+ LOADCHROMA(0);
+ PUTRGB15(dst_1, py_1, 0, 0);
+ PUTRGB15(dst_2, py_2, 0, 0 + 8);
+
+ LOADCHROMA(1);
+ PUTRGB15(dst_2, py_2, 1, 2 + 8);
+ PUTRGB15(dst_1, py_1, 1, 2);
+ENDYUV2RGBLINE(8, 1)
+ const uint8_t *d16 = ff_dither_2x2_8[y & 1];
+ const uint8_t *e16 = ff_dither_2x2_8[(y & 1)^1];
+
+ LOADCHROMA(0);
+ PUTRGB15(dst_1, py_1, 0, 0);
+ PUTRGB15(dst_2, py_2, 0, 0 + 8);
+ENDYUV2RGBFUNC()
// r, g, b, dst_1, dst_2
YUV2RGBFUNC(yuv2rgb_c_12_ordered_dither, uint16_t, 0)
@@ -493,7 +527,23 @@ YUV2RGBFUNC(yuv2rgb_c_12_ordered_dither, uint16_t, 0)
LOADCHROMA(3);
PUTRGB12(dst_2, py_2, 3, 6 + 8);
PUTRGB12(dst_1, py_1, 3, 6);
-CLOSEYUV2RGBFUNC(8)
+ENDYUV2RGBLINE(8, 0)
+ const uint8_t *d16 = ff_dither_4x4_16[y & 3];
+
+ LOADCHROMA(0);
+ PUTRGB12(dst_1, py_1, 0, 0);
+ PUTRGB12(dst_2, py_2, 0, 0 + 8);
+
+ LOADCHROMA(1);
+ PUTRGB12(dst_2, py_2, 1, 2 + 8);
+ PUTRGB12(dst_1, py_1, 1, 2);
+ENDYUV2RGBLINE(8, 1)
+ const uint8_t *d16 = ff_dither_4x4_16[y & 3];
+
+ LOADCHROMA(0);
+ PUTRGB12(dst_1, py_1, 0, 0);
+ PUTRGB12(dst_2, py_2, 0, 0 + 8);
+ENDYUV2RGBFUNC()
// r, g, b, dst_1, dst_2
YUV2RGBFUNC(yuv2rgb_c_8_ordered_dither, uint8_t, 0)
@@ -672,7 +722,42 @@ YUV2RGBFUNC(yuv2rgb_c_1_ordered_dither, uint8_t, 0)
dst_1[0] = out_1;
dst_2[0] = out_2;
-CLOSEYUV2RGBFUNC(1)
+
+ py_1 += 8;
+ py_2 += 8;
+ dst_1 += 1;
+ dst_2 += 1;
+ }
+ if (c->dstW & 7) {
+ int av_unused Y, U, V;
+ int pixels_left = c->dstW & 7;
+ const uint8_t *d128 = ff_dither_8x8_220[yd & 7];
+ char out_1 = 0, out_2 = 0;
+ g = c->table_gU[128 + YUVRGB_TABLE_HEADROOM] + c->table_gV[128 + YUVRGB_TABLE_HEADROOM];
+
+#define PUTRGB1_OR00(out, src, i, o) \
+ if (pixels_left) { \
+ PUTRGB1(out, src, i, o) \
+ pixels_left--; \
+ } else { \
+ out <<= 2; \
+ }
+
+ PUTRGB1_OR00(out_1, py_1, 0, 0);
+ PUTRGB1_OR00(out_2, py_2, 0, 0 + 8);
+
+ PUTRGB1_OR00(out_2, py_2, 1, 2 + 8);
+ PUTRGB1_OR00(out_1, py_1, 1, 2);
+
+ PUTRGB1_OR00(out_1, py_1, 2, 4);
+ PUTRGB1_OR00(out_2, py_2, 2, 4 + 8);
+
+ PUTRGB1_OR00(out_2, py_2, 3, 6 + 8);
+ PUTRGB1_OR00(out_1, py_1, 3, 6);
+
+ dst_1[0] = out_1;
+ dst_2[0] = out_2;
+ENDYUV2RGBFUNC()
SwsFunc ff_yuv2rgb_get_func_ptr(SwsContext *c)
{
--
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-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 14+ messages in thread* [FFmpeg-devel] [PATCH 2/6] swscale/yuv2rgb: add macros to simplify code generation 2024-06-16 22:28 [FFmpeg-devel] [PATCH 1/6] swscale/yuv2rgb: fix conversion for widths not aligned to 8 Ramiro Polla @ 2024-06-16 22:28 ` Ramiro Polla 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 3/6] swscale/yuv2rgb: fix yuv422p input in C code Ramiro Polla ` (3 subsequent siblings) 4 siblings, 0 replies; 14+ messages in thread From: Ramiro Polla @ 2024-06-16 22:28 UTC (permalink / raw) To: ffmpeg-devel --- libswscale/yuv2rgb.c | 574 +++++++++---------------------------------- 1 file changed, 113 insertions(+), 461 deletions(-) diff --git a/libswscale/yuv2rgb.c b/libswscale/yuv2rgb.c index e641c765c7..7386d3a2a2 100644 --- a/libswscale/yuv2rgb.c +++ b/libswscale/yuv2rgb.c @@ -72,13 +72,13 @@ const int *sws_getCoefficients(int colorspace) g = (void *)(c->table_gU[U+YUVRGB_TABLE_HEADROOM] + c->table_gV[V+YUVRGB_TABLE_HEADROOM]); \ b = (void *)c->table_bU[U+YUVRGB_TABLE_HEADROOM]; -#define PUTRGB(dst, src, i) \ +#define PUTRGB(dst, src, asrc, i, abase) \ Y = src[2 * i]; \ dst[2 * i] = r[Y] + g[Y] + b[Y]; \ Y = src[2 * i + 1]; \ dst[2 * i + 1] = r[Y] + g[Y] + b[Y]; -#define PUTRGB24(dst, src, i) \ +#define PUTRGB24(dst, src, asrc, i, abase) \ Y = src[2 * i]; \ dst[6 * i + 0] = r[Y]; \ dst[6 * i + 1] = g[Y]; \ @@ -88,7 +88,7 @@ const int *sws_getCoefficients(int colorspace) dst[6 * i + 4] = g[Y]; \ dst[6 * i + 5] = b[Y]; -#define PUTBGR24(dst, src, i) \ +#define PUTBGR24(dst, src, asrc, i, abase) \ Y = src[2 * i]; \ dst[6 * i + 0] = b[Y]; \ dst[6 * i + 1] = g[Y]; \ @@ -98,13 +98,13 @@ const int *sws_getCoefficients(int colorspace) dst[6 * i + 4] = g[Y]; \ dst[6 * i + 5] = r[Y]; -#define PUTRGBA(dst, ysrc, asrc, i, s) \ +#define PUTRGBA(dst, ysrc, asrc, i, abase) \ Y = ysrc[2 * i]; \ - dst[2 * i] = r[Y] + g[Y] + b[Y] + (asrc[2 * i] << s); \ + dst[2 * i] = r[Y] + g[Y] + b[Y] + (asrc[2 * i] << abase); \ Y = ysrc[2 * i + 1]; \ - dst[2 * i + 1] = r[Y] + g[Y] + b[Y] + (asrc[2 * i + 1] << s); + dst[2 * i + 1] = r[Y] + g[Y] + b[Y] + (asrc[2 * i + 1] << abase); -#define PUTRGB48(dst, src, i) \ +#define PUTRGB48(dst, src, asrc, i, abase) \ Y = src[ 2 * i]; \ dst[12 * i + 0] = dst[12 * i + 1] = r[Y]; \ dst[12 * i + 2] = dst[12 * i + 3] = g[Y]; \ @@ -114,7 +114,7 @@ const int *sws_getCoefficients(int colorspace) dst[12 * i + 8] = dst[12 * i + 9] = g[Y]; \ dst[12 * i + 10] = dst[12 * i + 11] = b[Y]; -#define PUTBGR48(dst, src, i) \ +#define PUTBGR48(dst, src, asrc, i, abase) \ Y = src[2 * i]; \ dst[12 * i + 0] = dst[12 * i + 1] = b[Y]; \ dst[12 * i + 2] = dst[12 * i + 3] = g[Y]; \ @@ -155,11 +155,15 @@ const int *sws_getCoefficients(int colorspace) while (h_size--) { \ int av_unused U, V, Y; \ -#define ENDYUV2RGBLINE(dst_delta, ss) \ +#define ENDYUV2RGBLINE(dst_delta, ss, alpha) \ pu += 4 >> ss; \ pv += 4 >> ss; \ py_1 += 8 >> ss; \ py_2 += 8 >> ss; \ + if (alpha) { \ + pa_1 += 8 >> ss; \ + pa_2 += 8 >> ss; \ + } \ dst_1 += dst_delta >> ss; \ dst_2 += dst_delta >> ss; \ } \ @@ -172,236 +176,77 @@ const int *sws_getCoefficients(int colorspace) return srcSliceH; \ } -YUV2RGBFUNC(yuv2rgb_c_48, uint8_t, 0) - LOADCHROMA(0); - PUTRGB48(dst_1, py_1, 0); - PUTRGB48(dst_2, py_2, 0); - - LOADCHROMA(1); - PUTRGB48(dst_2, py_2, 1); - PUTRGB48(dst_1, py_1, 1); - - LOADCHROMA(2); - PUTRGB48(dst_1, py_1, 2); - PUTRGB48(dst_2, py_2, 2); - - LOADCHROMA(3); - PUTRGB48(dst_2, py_2, 3); - PUTRGB48(dst_1, py_1, 3); -ENDYUV2RGBLINE(48, 0) - LOADCHROMA(0); - PUTRGB48(dst_1, py_1, 0); - PUTRGB48(dst_2, py_2, 0); - - LOADCHROMA(1); - PUTRGB48(dst_2, py_2, 1); - PUTRGB48(dst_1, py_1, 1); -ENDYUV2RGBLINE(48, 1) - LOADCHROMA(0); - PUTRGB48(dst_1, py_1, 0); - PUTRGB48(dst_2, py_2, 0); -ENDYUV2RGBFUNC() - -YUV2RGBFUNC(yuv2rgb_c_bgr48, uint8_t, 0) - LOADCHROMA(0); - PUTBGR48(dst_1, py_1, 0); - PUTBGR48(dst_2, py_2, 0); - - LOADCHROMA(1); - PUTBGR48(dst_2, py_2, 1); - PUTBGR48(dst_1, py_1, 1); - - LOADCHROMA(2); - PUTBGR48(dst_1, py_1, 2); - PUTBGR48(dst_2, py_2, 2); - - LOADCHROMA(3); - PUTBGR48(dst_2, py_2, 3); - PUTBGR48(dst_1, py_1, 3); -ENDYUV2RGBLINE(48, 0) - LOADCHROMA(0); - PUTBGR48(dst_1, py_1, 0); - PUTBGR48(dst_2, py_2, 0); - - LOADCHROMA(1); - PUTBGR48(dst_2, py_2, 1); - PUTBGR48(dst_1, py_1, 1); -ENDYUV2RGBLINE(48, 1) - LOADCHROMA(0); - PUTBGR48(dst_1, py_1, 0); - PUTBGR48(dst_2, py_2, 0); -ENDYUV2RGBFUNC() - -YUV2RGBFUNC(yuv2rgb_c_32, uint32_t, 0) - LOADCHROMA(0); - PUTRGB(dst_1, py_1, 0); - PUTRGB(dst_2, py_2, 0); - - LOADCHROMA(1); - PUTRGB(dst_2, py_2, 1); - PUTRGB(dst_1, py_1, 1); - - LOADCHROMA(2); - PUTRGB(dst_1, py_1, 2); - PUTRGB(dst_2, py_2, 2); - - LOADCHROMA(3); - PUTRGB(dst_2, py_2, 3); - PUTRGB(dst_1, py_1, 3); -ENDYUV2RGBLINE(8, 0) - LOADCHROMA(0); - PUTRGB(dst_1, py_1, 0); - PUTRGB(dst_2, py_2, 0); - - LOADCHROMA(1); - PUTRGB(dst_2, py_2, 1); - PUTRGB(dst_1, py_1, 1); -ENDYUV2RGBLINE(8, 1) - LOADCHROMA(0); - PUTRGB(dst_1, py_1, 0); - PUTRGB(dst_2, py_2, 0); -ENDYUV2RGBFUNC() - -#if HAVE_BIGENDIAN -YUV2RGBFUNC(yuva2argb_c, uint32_t, 1) -#else -YUV2RGBFUNC(yuva2rgba_c, uint32_t, 1) -#endif - LOADCHROMA(0); - PUTRGBA(dst_1, py_1, pa_1, 0, 24); - PUTRGBA(dst_2, py_2, pa_2, 0, 24); - - LOADCHROMA(1); - PUTRGBA(dst_2, py_2, pa_2, 1, 24); - PUTRGBA(dst_1, py_1, pa_1, 1, 24); - - LOADCHROMA(2); - PUTRGBA(dst_1, py_1, pa_1, 2, 24); - PUTRGBA(dst_2, py_2, pa_2, 2, 24); - - LOADCHROMA(3); - PUTRGBA(dst_2, py_2, pa_2, 3, 24); - PUTRGBA(dst_1, py_1, pa_1, 3, 24); - pa_1 += 8; - pa_2 += 8; -ENDYUV2RGBLINE(8, 0) - LOADCHROMA(0); - PUTRGBA(dst_1, py_1, pa_1, 0, 24); - PUTRGBA(dst_2, py_2, pa_2, 0, 24); - - LOADCHROMA(1); - PUTRGBA(dst_2, py_2, pa_2, 1, 24); - PUTRGBA(dst_1, py_1, pa_1, 1, 24); - pa_1 += 4; - pa_2 += 4; -ENDYUV2RGBLINE(8, 1) - LOADCHROMA(0); - PUTRGBA(dst_1, py_1, pa_1, 0, 24); - PUTRGBA(dst_2, py_2, pa_2, 0, 24); -ENDYUV2RGBFUNC() - -#if HAVE_BIGENDIAN -YUV2RGBFUNC(yuva2rgba_c, uint32_t, 1) -#else -YUV2RGBFUNC(yuva2argb_c, uint32_t, 1) -#endif - LOADCHROMA(0); - PUTRGBA(dst_1, py_1, pa_1, 0, 0); - PUTRGBA(dst_2, py_2, pa_2, 0, 0); - - LOADCHROMA(1); - PUTRGBA(dst_2, py_2, pa_2, 1, 0); - PUTRGBA(dst_1, py_1, pa_1, 1, 0); - - LOADCHROMA(2); - PUTRGBA(dst_1, py_1, pa_1, 2, 0); - PUTRGBA(dst_2, py_2, pa_2, 2, 0); - - LOADCHROMA(3); - PUTRGBA(dst_2, py_2, pa_2, 3, 0); - PUTRGBA(dst_1, py_1, pa_1, 3, 0); - pa_1 += 8; - pa_2 += 8; -ENDYUV2RGBLINE(8, 0) - LOADCHROMA(0); - PUTRGBA(dst_1, py_1, pa_1, 0, 0); - PUTRGBA(dst_2, py_2, pa_2, 0, 0); - - LOADCHROMA(1); - PUTRGBA(dst_2, py_2, pa_2, 1, 0); - PUTRGBA(dst_1, py_1, pa_1, 1, 0); - pa_1 += 4; - pa_2 += 4; -ENDYUV2RGBLINE(8, 1) - LOADCHROMA(0); - PUTRGBA(dst_1, py_1, pa_1, 0, 0); - PUTRGBA(dst_2, py_2, pa_2, 0, 0); -ENDYUV2RGBFUNC() - -YUV2RGBFUNC(yuv2rgb_c_24_rgb, uint8_t, 0) - LOADCHROMA(0); - PUTRGB24(dst_1, py_1, 0); - PUTRGB24(dst_2, py_2, 0); - - LOADCHROMA(1); - PUTRGB24(dst_2, py_2, 1); - PUTRGB24(dst_1, py_1, 1); - - LOADCHROMA(2); - PUTRGB24(dst_1, py_1, 2); - PUTRGB24(dst_2, py_2, 2); - - LOADCHROMA(3); - PUTRGB24(dst_2, py_2, 3); - PUTRGB24(dst_1, py_1, 3); -ENDYUV2RGBLINE(24, 0) - LOADCHROMA(0); - PUTRGB24(dst_1, py_1, 0); - PUTRGB24(dst_2, py_2, 0); - - LOADCHROMA(1); - PUTRGB24(dst_2, py_2, 1); - PUTRGB24(dst_1, py_1, 1); -ENDYUV2RGBLINE(24, 1) - LOADCHROMA(0); - PUTRGB24(dst_1, py_1, 0); - PUTRGB24(dst_2, py_2, 0); -ENDYUV2RGBFUNC() - -// only trivial mods from yuv2rgb_c_24_rgb -YUV2RGBFUNC(yuv2rgb_c_24_bgr, uint8_t, 0) - LOADCHROMA(0); - PUTBGR24(dst_1, py_1, 0); - PUTBGR24(dst_2, py_2, 0); - - LOADCHROMA(1); - PUTBGR24(dst_2, py_2, 1); - PUTBGR24(dst_1, py_1, 1); - - LOADCHROMA(2); - PUTBGR24(dst_1, py_1, 2); - PUTBGR24(dst_2, py_2, 2); - - LOADCHROMA(3); - PUTBGR24(dst_2, py_2, 3); - PUTBGR24(dst_1, py_1, 3); -ENDYUV2RGBLINE(24, 0) - LOADCHROMA(0); - PUTBGR24(dst_1, py_1, 0); - PUTBGR24(dst_2, py_2, 0); - - LOADCHROMA(1); - PUTBGR24(dst_2, py_2, 1); - PUTBGR24(dst_1, py_1, 1); -ENDYUV2RGBLINE(24, 1) - LOADCHROMA(0); - PUTBGR24(dst_1, py_1, 0); - PUTBGR24(dst_2, py_2, 0); -ENDYUV2RGBFUNC() - -YUV2RGBFUNC(yuv2rgb_c_16_ordered_dither, uint16_t, 0) - const uint8_t *d16 = ff_dither_2x2_8[y & 1]; - const uint8_t *e16 = ff_dither_2x2_4[y & 1]; +#define YUV420FUNC(func_name, dst_type, alpha, abase, PUTFUNC, dst_delta) \ + YUV2RGBFUNC(func_name, dst_type, alpha) \ + LOADCHROMA(0); \ + PUTFUNC(dst_1, py_1, pa_1, 0, abase); \ + PUTFUNC(dst_2, py_2, pa_2, 0, abase); \ + \ + LOADCHROMA(1); \ + PUTFUNC(dst_2, py_2, pa_2, 1, abase); \ + PUTFUNC(dst_1, py_1, pa_1, 1, abase); \ + \ + LOADCHROMA(2); \ + PUTFUNC(dst_1, py_1, pa_1, 2, abase); \ + PUTFUNC(dst_2, py_2, pa_2, 2, abase); \ + \ + LOADCHROMA(3); \ + PUTFUNC(dst_2, py_2, pa_2, 3, abase); \ + PUTFUNC(dst_1, py_1, pa_1, 3, abase); \ + ENDYUV2RGBLINE(dst_delta, 0, alpha) \ + LOADCHROMA(0); \ + PUTFUNC(dst_1, py_1, pa_1, 0, abase); \ + PUTFUNC(dst_2, py_2, pa_2, 0, abase); \ + \ + LOADCHROMA(1); \ + PUTFUNC(dst_2, py_2, pa_2, 1, abase); \ + PUTFUNC(dst_1, py_1, pa_1, 1, abase); \ + ENDYUV2RGBLINE(dst_delta, 1, alpha) \ + LOADCHROMA(0); \ + PUTFUNC(dst_1, py_1, pa_1, 0, abase); \ + PUTFUNC(dst_2, py_2, pa_2, 0, abase); \ + ENDYUV2RGBFUNC() + +#define YUV420FUNC_DITHER(func_name, dst_type, LOADDITHER, PUTFUNC, dst_delta) \ + YUV2RGBFUNC(func_name, dst_type, 0) \ + LOADDITHER \ + \ + LOADCHROMA(0); \ + PUTFUNC(dst_1, py_1, 0, 0); \ + PUTFUNC(dst_2, py_2, 0, 0 + 8); \ + \ + LOADCHROMA(1); \ + PUTFUNC(dst_2, py_2, 1, 2 + 8); \ + PUTFUNC(dst_1, py_1, 1, 2); \ + \ + LOADCHROMA(2); \ + PUTFUNC(dst_1, py_1, 2, 4); \ + PUTFUNC(dst_2, py_2, 2, 4 + 8); \ + \ + LOADCHROMA(3); \ + PUTFUNC(dst_2, py_2, 3, 6 + 8); \ + PUTFUNC(dst_1, py_1, 3, 6); \ + ENDYUV2RGBLINE(dst_delta, 0, 0) \ + LOADDITHER \ + \ + LOADCHROMA(0); \ + PUTFUNC(dst_1, py_1, 0, 0); \ + PUTFUNC(dst_2, py_2, 0, 0 + 8); \ + \ + LOADCHROMA(1); \ + PUTFUNC(dst_2, py_2, 1, 2 + 8); \ + PUTFUNC(dst_1, py_1, 1, 2); \ + ENDYUV2RGBLINE(dst_delta, 1, 0) \ + LOADDITHER \ + \ + LOADCHROMA(0); \ + PUTFUNC(dst_1, py_1, 0, 0); \ + PUTFUNC(dst_2, py_2, 0, 0 + 8); \ + ENDYUV2RGBFUNC() + +#define LOADDITHER16 \ + const uint8_t *d16 = ff_dither_2x2_8[y & 1]; \ + const uint8_t *e16 = ff_dither_2x2_4[y & 1]; \ const uint8_t *f16 = ff_dither_2x2_8[(y & 1)^1]; #define PUTRGB16(dst, src, i, o) \ @@ -413,45 +258,9 @@ YUV2RGBFUNC(yuv2rgb_c_16_ordered_dither, uint16_t, 0) dst[2 * i + 1] = r[Y + d16[1 + o]] + \ g[Y + e16[1 + o]] + \ b[Y + f16[1 + o]]; - LOADCHROMA(0); - PUTRGB16(dst_1, py_1, 0, 0); - PUTRGB16(dst_2, py_2, 0, 0 + 8); - - LOADCHROMA(1); - PUTRGB16(dst_2, py_2, 1, 2 + 8); - PUTRGB16(dst_1, py_1, 1, 2); - - LOADCHROMA(2); - PUTRGB16(dst_1, py_1, 2, 4); - PUTRGB16(dst_2, py_2, 2, 4 + 8); - - LOADCHROMA(3); - PUTRGB16(dst_2, py_2, 3, 6 + 8); - PUTRGB16(dst_1, py_1, 3, 6); -ENDYUV2RGBLINE(8, 0) - const uint8_t *d16 = ff_dither_2x2_8[y & 1]; - const uint8_t *e16 = ff_dither_2x2_4[y & 1]; - const uint8_t *f16 = ff_dither_2x2_8[(y & 1)^1]; - - LOADCHROMA(0); - PUTRGB16(dst_1, py_1, 0, 0); - PUTRGB16(dst_2, py_2, 0, 0 + 8); - - LOADCHROMA(1); - PUTRGB16(dst_2, py_2, 1, 2 + 8); - PUTRGB16(dst_1, py_1, 1, 2); -ENDYUV2RGBLINE(8, 1) - const uint8_t *d16 = ff_dither_2x2_8[y & 1]; - const uint8_t *e16 = ff_dither_2x2_4[y & 1]; - const uint8_t *f16 = ff_dither_2x2_8[(y & 1)^1]; - LOADCHROMA(0); - PUTRGB16(dst_1, py_1, 0, 0); - PUTRGB16(dst_2, py_2, 0, 0 + 8); -ENDYUV2RGBFUNC() - -YUV2RGBFUNC(yuv2rgb_c_15_ordered_dither, uint16_t, 0) - const uint8_t *d16 = ff_dither_2x2_8[y & 1]; +#define LOADDITHER15 \ + const uint8_t *d16 = ff_dither_2x2_8[y & 1]; \ const uint8_t *e16 = ff_dither_2x2_8[(y & 1)^1]; #define PUTRGB15(dst, src, i, o) \ @@ -463,43 +272,8 @@ YUV2RGBFUNC(yuv2rgb_c_15_ordered_dither, uint16_t, 0) dst[2 * i + 1] = r[Y + d16[1 + o]] + \ g[Y + d16[0 + o]] + \ b[Y + e16[1 + o]]; - LOADCHROMA(0); - PUTRGB15(dst_1, py_1, 0, 0); - PUTRGB15(dst_2, py_2, 0, 0 + 8); - - LOADCHROMA(1); - PUTRGB15(dst_2, py_2, 1, 2 + 8); - PUTRGB15(dst_1, py_1, 1, 2); - - LOADCHROMA(2); - PUTRGB15(dst_1, py_1, 2, 4); - PUTRGB15(dst_2, py_2, 2, 4 + 8); - - LOADCHROMA(3); - PUTRGB15(dst_2, py_2, 3, 6 + 8); - PUTRGB15(dst_1, py_1, 3, 6); -ENDYUV2RGBLINE(8, 0) - const uint8_t *d16 = ff_dither_2x2_8[y & 1]; - const uint8_t *e16 = ff_dither_2x2_8[(y & 1)^1]; - - LOADCHROMA(0); - PUTRGB15(dst_1, py_1, 0, 0); - PUTRGB15(dst_2, py_2, 0, 0 + 8); - - LOADCHROMA(1); - PUTRGB15(dst_2, py_2, 1, 2 + 8); - PUTRGB15(dst_1, py_1, 1, 2); -ENDYUV2RGBLINE(8, 1) - const uint8_t *d16 = ff_dither_2x2_8[y & 1]; - const uint8_t *e16 = ff_dither_2x2_8[(y & 1)^1]; - - LOADCHROMA(0); - PUTRGB15(dst_1, py_1, 0, 0); - PUTRGB15(dst_2, py_2, 0, 0 + 8); -ENDYUV2RGBFUNC() -// r, g, b, dst_1, dst_2 -YUV2RGBFUNC(yuv2rgb_c_12_ordered_dither, uint16_t, 0) +#define LOADDITHER12 \ const uint8_t *d16 = ff_dither_4x4_16[y & 3]; #define PUTRGB12(dst, src, i, o) \ @@ -512,42 +286,8 @@ YUV2RGBFUNC(yuv2rgb_c_12_ordered_dither, uint16_t, 0) g[Y + d16[1 + o]] + \ b[Y + d16[1 + o]]; - LOADCHROMA(0); - PUTRGB12(dst_1, py_1, 0, 0); - PUTRGB12(dst_2, py_2, 0, 0 + 8); - - LOADCHROMA(1); - PUTRGB12(dst_2, py_2, 1, 2 + 8); - PUTRGB12(dst_1, py_1, 1, 2); - - LOADCHROMA(2); - PUTRGB12(dst_1, py_1, 2, 4); - PUTRGB12(dst_2, py_2, 2, 4 + 8); - - LOADCHROMA(3); - PUTRGB12(dst_2, py_2, 3, 6 + 8); - PUTRGB12(dst_1, py_1, 3, 6); -ENDYUV2RGBLINE(8, 0) - const uint8_t *d16 = ff_dither_4x4_16[y & 3]; - - LOADCHROMA(0); - PUTRGB12(dst_1, py_1, 0, 0); - PUTRGB12(dst_2, py_2, 0, 0 + 8); - - LOADCHROMA(1); - PUTRGB12(dst_2, py_2, 1, 2 + 8); - PUTRGB12(dst_1, py_1, 1, 2); -ENDYUV2RGBLINE(8, 1) - const uint8_t *d16 = ff_dither_4x4_16[y & 3]; - - LOADCHROMA(0); - PUTRGB12(dst_1, py_1, 0, 0); - PUTRGB12(dst_2, py_2, 0, 0 + 8); -ENDYUV2RGBFUNC() - -// r, g, b, dst_1, dst_2 -YUV2RGBFUNC(yuv2rgb_c_8_ordered_dither, uint8_t, 0) - const uint8_t *d32 = ff_dither_8x8_32[yd & 7]; +#define LOADDITHER8 \ + const uint8_t *d32 = ff_dither_8x8_32[yd & 7]; \ const uint8_t *d64 = ff_dither_8x8_73[yd & 7]; #define PUTRGB8(dst, src, i, o) \ @@ -560,46 +300,9 @@ YUV2RGBFUNC(yuv2rgb_c_8_ordered_dither, uint8_t, 0) g[Y + d32[1 + o]] + \ b[Y + d64[1 + o]]; - LOADCHROMA(0); - PUTRGB8(dst_1, py_1, 0, 0); - PUTRGB8(dst_2, py_2, 0, 0 + 8); - - LOADCHROMA(1); - PUTRGB8(dst_2, py_2, 1, 2 + 8); - PUTRGB8(dst_1, py_1, 1, 2); - - LOADCHROMA(2); - PUTRGB8(dst_1, py_1, 2, 4); - PUTRGB8(dst_2, py_2, 2, 4 + 8); - - LOADCHROMA(3); - PUTRGB8(dst_2, py_2, 3, 6 + 8); - PUTRGB8(dst_1, py_1, 3, 6); - -ENDYUV2RGBLINE(8, 0) - const uint8_t *d32 = ff_dither_8x8_32[yd & 7]; - const uint8_t *d64 = ff_dither_8x8_73[yd & 7]; - LOADCHROMA(0); - PUTRGB8(dst_1, py_1, 0, 0); - PUTRGB8(dst_2, py_2, 0, 0 + 8); - - LOADCHROMA(1); - PUTRGB8(dst_2, py_2, 1, 2 + 8); - PUTRGB8(dst_1, py_1, 1, 2); - -ENDYUV2RGBLINE(8, 1) - const uint8_t *d32 = ff_dither_8x8_32[yd & 7]; - const uint8_t *d64 = ff_dither_8x8_73[yd & 7]; - LOADCHROMA(0); - PUTRGB8(dst_1, py_1, 0, 0); - PUTRGB8(dst_2, py_2, 0, 0 + 8); - -ENDYUV2RGBFUNC() - - -YUV2RGBFUNC(yuv2rgb_c_4_ordered_dither, uint8_t, 0) - const uint8_t * d64 = ff_dither_8x8_73[yd & 7]; - const uint8_t *d128 = ff_dither_8x8_220[yd & 7]; +#define LOADDITHER4D \ + const uint8_t * d64 = ff_dither_8x8_73[yd & 7]; \ + const uint8_t *d128 = ff_dither_8x8_220[yd & 7]; \ int acc; #define PUTRGB4D(dst, src, i, o) \ @@ -613,45 +316,8 @@ YUV2RGBFUNC(yuv2rgb_c_4_ordered_dither, uint8_t, 0) b[Y + d128[1 + o]]) << 4; \ dst[i] = acc; - LOADCHROMA(0); - PUTRGB4D(dst_1, py_1, 0, 0); - PUTRGB4D(dst_2, py_2, 0, 0 + 8); - - LOADCHROMA(1); - PUTRGB4D(dst_2, py_2, 1, 2 + 8); - PUTRGB4D(dst_1, py_1, 1, 2); - - LOADCHROMA(2); - PUTRGB4D(dst_1, py_1, 2, 4); - PUTRGB4D(dst_2, py_2, 2, 4 + 8); - - LOADCHROMA(3); - PUTRGB4D(dst_2, py_2, 3, 6 + 8); - PUTRGB4D(dst_1, py_1, 3, 6); - -ENDYUV2RGBLINE(4, 0) - const uint8_t * d64 = ff_dither_8x8_73[yd & 7]; - const uint8_t *d128 = ff_dither_8x8_220[yd & 7]; - int acc; - LOADCHROMA(0); - PUTRGB4D(dst_1, py_1, 0, 0); - PUTRGB4D(dst_2, py_2, 0, 0 + 8); - - LOADCHROMA(1); - PUTRGB4D(dst_2, py_2, 1, 2 + 8); - PUTRGB4D(dst_1, py_1, 1, 2); - -ENDYUV2RGBLINE(4, 1) - const uint8_t * d64 = ff_dither_8x8_73[yd & 7]; - const uint8_t *d128 = ff_dither_8x8_220[yd & 7]; - int acc; - LOADCHROMA(0); - PUTRGB4D(dst_1, py_1, 0, 0); - PUTRGB4D(dst_2, py_2, 0, 0 + 8); -ENDYUV2RGBFUNC() - -YUV2RGBFUNC(yuv2rgb_c_4b_ordered_dither, uint8_t, 0) - const uint8_t *d64 = ff_dither_8x8_73[yd & 7]; +#define LOADDITHER4DB \ + const uint8_t *d64 = ff_dither_8x8_73[yd & 7]; \ const uint8_t *d128 = ff_dither_8x8_220[yd & 7]; #define PUTRGB4DB(dst, src, i, o) \ @@ -664,39 +330,6 @@ YUV2RGBFUNC(yuv2rgb_c_4b_ordered_dither, uint8_t, 0) g[Y + d64[1 + o]] + \ b[Y + d128[1 + o]]; - LOADCHROMA(0); - PUTRGB4DB(dst_1, py_1, 0, 0); - PUTRGB4DB(dst_2, py_2, 0, 0 + 8); - - LOADCHROMA(1); - PUTRGB4DB(dst_2, py_2, 1, 2 + 8); - PUTRGB4DB(dst_1, py_1, 1, 2); - - LOADCHROMA(2); - PUTRGB4DB(dst_1, py_1, 2, 4); - PUTRGB4DB(dst_2, py_2, 2, 4 + 8); - - LOADCHROMA(3); - PUTRGB4DB(dst_2, py_2, 3, 6 + 8); - PUTRGB4DB(dst_1, py_1, 3, 6); -ENDYUV2RGBLINE(8, 0) - const uint8_t *d64 = ff_dither_8x8_73[yd & 7]; - const uint8_t *d128 = ff_dither_8x8_220[yd & 7]; - LOADCHROMA(0); - PUTRGB4DB(dst_1, py_1, 0, 0); - PUTRGB4DB(dst_2, py_2, 0, 0 + 8); - - LOADCHROMA(1); - PUTRGB4DB(dst_2, py_2, 1, 2 + 8); - PUTRGB4DB(dst_1, py_1, 1, 2); -ENDYUV2RGBLINE(8, 1) - const uint8_t *d64 = ff_dither_8x8_73[yd & 7]; - const uint8_t *d128 = ff_dither_8x8_220[yd & 7]; - LOADCHROMA(0); - PUTRGB4DB(dst_1, py_1, 0, 0); - PUTRGB4DB(dst_2, py_2, 0, 0 + 8); -ENDYUV2RGBFUNC() - YUV2RGBFUNC(yuv2rgb_c_1_ordered_dither, uint8_t, 0) const uint8_t *d128 = ff_dither_8x8_220[yd & 7]; char out_1 = 0, out_2 = 0; @@ -759,6 +392,25 @@ YUV2RGBFUNC(yuv2rgb_c_1_ordered_dither, uint8_t, 0) dst_2[0] = out_2; ENDYUV2RGBFUNC() +YUV420FUNC(yuv2rgb_c_48, uint8_t, 0, 0, PUTRGB48, 48) +YUV420FUNC(yuv2rgb_c_bgr48, uint8_t, 0, 0, PUTBGR48, 48) +YUV420FUNC(yuv2rgb_c_32, uint32_t, 0, 0, PUTRGB, 8) +#if HAVE_BIGENDIAN +YUV420FUNC(yuva2argb_c, uint32_t, 1, 24, PUTRGBA, 8) +YUV420FUNC(yuva2rgba_c, uint32_t, 1, 0, PUTRGBA, 8) +#else +YUV420FUNC(yuva2rgba_c, uint32_t, 1, 24, PUTRGBA, 8) +YUV420FUNC(yuva2argb_c, uint32_t, 1, 0, PUTRGBA, 8) +#endif +YUV420FUNC(yuv2rgb_c_24_rgb, uint8_t, 0, 0, PUTRGB24, 24) +YUV420FUNC(yuv2rgb_c_24_bgr, uint8_t, 0, 0, PUTBGR24, 24) +YUV420FUNC_DITHER(yuv2rgb_c_16_ordered_dither, uint16_t, LOADDITHER16, PUTRGB16, 8) +YUV420FUNC_DITHER(yuv2rgb_c_15_ordered_dither, uint16_t, LOADDITHER15, PUTRGB15, 8) +YUV420FUNC_DITHER(yuv2rgb_c_12_ordered_dither, uint16_t, LOADDITHER12, PUTRGB12, 8) +YUV420FUNC_DITHER(yuv2rgb_c_8_ordered_dither, uint8_t, LOADDITHER8, PUTRGB8, 8) +YUV420FUNC_DITHER(yuv2rgb_c_4_ordered_dither, uint8_t, LOADDITHER4D, PUTRGB4D, 4) +YUV420FUNC_DITHER(yuv2rgb_c_4b_ordered_dither, uint8_t, LOADDITHER4DB, PUTRGB4DB, 8) + SwsFunc ff_yuv2rgb_get_func_ptr(SwsContext *c) { SwsFunc t = NULL; -- 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-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 14+ messages in thread
* [FFmpeg-devel] [PATCH 3/6] swscale/yuv2rgb: fix yuv422p input in C code 2024-06-16 22:28 [FFmpeg-devel] [PATCH 1/6] swscale/yuv2rgb: fix conversion for widths not aligned to 8 Ramiro Polla 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 2/6] swscale/yuv2rgb: add macros to simplify code generation Ramiro Polla @ 2024-06-16 22:28 ` Ramiro Polla 2024-06-19 10:13 ` Michael Niedermayer 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 4/6] swscale/yuv2rgb: reindent after previous commit Ramiro Polla ` (2 subsequent siblings) 4 siblings, 1 reply; 14+ messages in thread From: Ramiro Polla @ 2024-06-16 22:28 UTC (permalink / raw) To: ffmpeg-devel The C code was silently ignoring the second chroma line on yuv422p input. --- libswscale/yuv2rgb.c | 228 +++++++++++++++++++++++++++++++++++++------ 1 file changed, 196 insertions(+), 32 deletions(-) diff --git a/libswscale/yuv2rgb.c b/libswscale/yuv2rgb.c index 7386d3a2a2..1ea87ac17a 100644 --- a/libswscale/yuv2rgb.c +++ b/libswscale/yuv2rgb.c @@ -65,7 +65,7 @@ const int *sws_getCoefficients(int colorspace) return ff_yuv2rgb_coeffs[colorspace]; } -#define LOADCHROMA(i) \ +#define LOADCHROMA(pu, pv, i) \ U = pu[i]; \ V = pv[i]; \ r = (void *)c->table_rV[V+YUVRGB_TABLE_HEADROOM]; \ @@ -124,17 +124,13 @@ const int *sws_getCoefficients(int colorspace) dst[12 * i + 8] = dst[12 * i + 9] = g[Y]; \ dst[12 * i + 10] = dst[12 * i + 11] = r[Y]; -#define YUV2RGBFUNC(func_name, dst_type, alpha) \ +#define YUV2RGBFUNC(func_name, dst_type, alpha, yuv422) \ static int func_name(SwsContext *c, const uint8_t *src[], \ int srcStride[], int srcSliceY, int srcSliceH, \ uint8_t *dst[], int dstStride[]) \ { \ int y; \ \ - if (!alpha && c->srcFormat == AV_PIX_FMT_YUV422P) { \ - srcStride[1] *= 2; \ - srcStride[2] *= 2; \ - } \ for (y = 0; y < srcSliceH; y += 2) { \ int yd = y + srcSliceY; \ dst_type *dst_1 = \ @@ -144,10 +140,15 @@ const int *sws_getCoefficients(int colorspace) dst_type av_unused *r, *g, *b; \ const uint8_t *py_1 = src[0] + y * srcStride[0]; \ const uint8_t *py_2 = py_1 + srcStride[0]; \ - const uint8_t av_unused *pu = src[1] + (y >> 1) * srcStride[1]; \ - const uint8_t av_unused *pv = src[2] + (y >> 1) * srcStride[2]; \ + const uint8_t av_unused *pu_1 = src[1] + (y >> !yuv422) * srcStride[1]; \ + const uint8_t av_unused *pv_1 = src[2] + (y >> !yuv422) * srcStride[2]; \ + const uint8_t av_unused *pu_2, *pv_2; \ const uint8_t av_unused *pa_1, *pa_2; \ unsigned int h_size = c->dstW >> 3; \ + if (yuv422) { \ + pu_2 = pu_1 + srcStride[1]; \ + pv_2 = pv_1 + srcStride[2]; \ + } \ if (alpha) { \ pa_1 = src[3] + y * srcStride[3]; \ pa_2 = pa_1 + srcStride[3]; \ @@ -155,9 +156,13 @@ const int *sws_getCoefficients(int colorspace) while (h_size--) { \ int av_unused U, V, Y; \ -#define ENDYUV2RGBLINE(dst_delta, ss, alpha) \ - pu += 4 >> ss; \ - pv += 4 >> ss; \ +#define ENDYUV2RGBLINE(dst_delta, ss, alpha, yuv422) \ + pu_1 += 4 >> ss; \ + pv_1 += 4 >> ss; \ + if (yuv422) { \ + pu_2 += 4 >> ss; \ + pv_2 += 4 >> ss; \ + } \ py_1 += 8 >> ss; \ py_2 += 8 >> ss; \ if (alpha) { \ @@ -177,73 +182,169 @@ const int *sws_getCoefficients(int colorspace) } #define YUV420FUNC(func_name, dst_type, alpha, abase, PUTFUNC, dst_delta) \ - YUV2RGBFUNC(func_name, dst_type, alpha) \ - LOADCHROMA(0); \ + YUV2RGBFUNC(func_name, dst_type, alpha, 0) \ + LOADCHROMA(pu_1, pv_1, 0); \ PUTFUNC(dst_1, py_1, pa_1, 0, abase); \ PUTFUNC(dst_2, py_2, pa_2, 0, abase); \ \ - LOADCHROMA(1); \ + LOADCHROMA(pu_1, pv_1, 1); \ PUTFUNC(dst_2, py_2, pa_2, 1, abase); \ PUTFUNC(dst_1, py_1, pa_1, 1, abase); \ \ - LOADCHROMA(2); \ + LOADCHROMA(pu_1, pv_1, 2); \ PUTFUNC(dst_1, py_1, pa_1, 2, abase); \ PUTFUNC(dst_2, py_2, pa_2, 2, abase); \ \ - LOADCHROMA(3); \ + LOADCHROMA(pu_1, pv_1, 3); \ PUTFUNC(dst_2, py_2, pa_2, 3, abase); \ PUTFUNC(dst_1, py_1, pa_1, 3, abase); \ - ENDYUV2RGBLINE(dst_delta, 0, alpha) \ - LOADCHROMA(0); \ + ENDYUV2RGBLINE(dst_delta, 0, alpha, 0) \ + LOADCHROMA(pu_1, pv_1, 0); \ PUTFUNC(dst_1, py_1, pa_1, 0, abase); \ PUTFUNC(dst_2, py_2, pa_2, 0, abase); \ \ - LOADCHROMA(1); \ + LOADCHROMA(pu_1, pv_1, 1); \ PUTFUNC(dst_2, py_2, pa_2, 1, abase); \ PUTFUNC(dst_1, py_1, pa_1, 1, abase); \ - ENDYUV2RGBLINE(dst_delta, 1, alpha) \ - LOADCHROMA(0); \ + ENDYUV2RGBLINE(dst_delta, 1, alpha, 0) \ + LOADCHROMA(pu_1, pv_1, 0); \ PUTFUNC(dst_1, py_1, pa_1, 0, abase); \ PUTFUNC(dst_2, py_2, pa_2, 0, abase); \ ENDYUV2RGBFUNC() +#define YUV422FUNC(func_name, dst_type, alpha, abase, PUTFUNC, dst_delta) \ + YUV2RGBFUNC(func_name, dst_type, alpha, 1) \ + LOADCHROMA(pu_1, pv_1, 0); \ + PUTFUNC(dst_1, py_1, pa_1, 0, abase); \ + \ + LOADCHROMA(pu_2, pv_2, 0); \ + PUTFUNC(dst_2, py_2, pa_2, 0, abase); \ + \ + LOADCHROMA(pu_2, pv_2, 1); \ + PUTFUNC(dst_2, py_2, pa_2, 1, abase); \ + \ + LOADCHROMA(pu_1, pv_1, 1); \ + PUTFUNC(dst_1, py_1, pa_1, 1, abase); \ + \ + LOADCHROMA(pu_1, pv_1, 2); \ + PUTFUNC(dst_1, py_1, pa_1, 2, abase); \ + \ + LOADCHROMA(pu_2, pv_2, 2); \ + PUTFUNC(dst_2, py_2, pa_2, 2, abase); \ + \ + LOADCHROMA(pu_2, pv_2, 3); \ + PUTFUNC(dst_2, py_2, pa_2, 3, abase); \ + \ + LOADCHROMA(pu_1, pv_1, 3); \ + PUTFUNC(dst_1, py_1, pa_1, 3, abase); \ + ENDYUV2RGBLINE(dst_delta, 0, alpha, 1) \ + LOADCHROMA(pu_1, pv_1, 0); \ + PUTFUNC(dst_1, py_1, pa_1, 0, abase); \ + \ + LOADCHROMA(pu_2, pv_2, 0); \ + PUTFUNC(dst_2, py_2, pa_2, 0, abase); \ + \ + LOADCHROMA(pu_2, pv_2, 1); \ + PUTFUNC(dst_2, py_2, pa_2, 1, abase); \ + \ + LOADCHROMA(pu_1, pv_1, 1); \ + PUTFUNC(dst_1, py_1, pa_1, 1, abase); \ + ENDYUV2RGBLINE(dst_delta, 1, alpha, 1) \ + LOADCHROMA(pu_1, pv_1, 0); \ + PUTFUNC(dst_1, py_1, pa_1, 0, abase); \ + \ + LOADCHROMA(pu_2, pv_2, 0); \ + PUTFUNC(dst_2, py_2, pa_2, 0, abase); \ + ENDYUV2RGBFUNC() + #define YUV420FUNC_DITHER(func_name, dst_type, LOADDITHER, PUTFUNC, dst_delta) \ - YUV2RGBFUNC(func_name, dst_type, 0) \ + YUV2RGBFUNC(func_name, dst_type, 0, 0) \ LOADDITHER \ \ - LOADCHROMA(0); \ + LOADCHROMA(pu_1, pv_1, 0); \ PUTFUNC(dst_1, py_1, 0, 0); \ PUTFUNC(dst_2, py_2, 0, 0 + 8); \ \ - LOADCHROMA(1); \ + LOADCHROMA(pu_1, pv_1, 1); \ PUTFUNC(dst_2, py_2, 1, 2 + 8); \ PUTFUNC(dst_1, py_1, 1, 2); \ \ - LOADCHROMA(2); \ + LOADCHROMA(pu_1, pv_1, 2); \ PUTFUNC(dst_1, py_1, 2, 4); \ PUTFUNC(dst_2, py_2, 2, 4 + 8); \ \ - LOADCHROMA(3); \ + LOADCHROMA(pu_1, pv_1, 3); \ PUTFUNC(dst_2, py_2, 3, 6 + 8); \ PUTFUNC(dst_1, py_1, 3, 6); \ - ENDYUV2RGBLINE(dst_delta, 0, 0) \ + ENDYUV2RGBLINE(dst_delta, 0, 0, 0) \ LOADDITHER \ \ - LOADCHROMA(0); \ + LOADCHROMA(pu_1, pv_1, 0); \ PUTFUNC(dst_1, py_1, 0, 0); \ PUTFUNC(dst_2, py_2, 0, 0 + 8); \ \ - LOADCHROMA(1); \ + LOADCHROMA(pu_1, pv_1, 1); \ PUTFUNC(dst_2, py_2, 1, 2 + 8); \ PUTFUNC(dst_1, py_1, 1, 2); \ - ENDYUV2RGBLINE(dst_delta, 1, 0) \ + ENDYUV2RGBLINE(dst_delta, 1, 0, 0) \ LOADDITHER \ \ - LOADCHROMA(0); \ + LOADCHROMA(pu_1, pv_1, 0); \ PUTFUNC(dst_1, py_1, 0, 0); \ PUTFUNC(dst_2, py_2, 0, 0 + 8); \ ENDYUV2RGBFUNC() +#define YUV422FUNC_DITHER(func_name, dst_type, LOADDITHER, PUTFUNC, dst_delta) \ + YUV2RGBFUNC(func_name, dst_type, 0, 1) \ + LOADDITHER \ + \ + LOADCHROMA(pu_1, pv_1, 0); \ + PUTFUNC(dst_1, py_1, 0, 0); \ + \ + LOADCHROMA(pu_2, pv_2, 0); \ + PUTFUNC(dst_2, py_2, 0, 0 + 8); \ + \ + LOADCHROMA(pu_2, pv_2, 1); \ + PUTFUNC(dst_2, py_2, 1, 2 + 8); \ + \ + LOADCHROMA(pu_1, pv_1, 1); \ + PUTFUNC(dst_1, py_1, 1, 2); \ + \ + LOADCHROMA(pu_1, pv_1, 2); \ + PUTFUNC(dst_1, py_1, 2, 4); \ + \ + LOADCHROMA(pu_2, pv_2, 2); \ + PUTFUNC(dst_2, py_2, 2, 4 + 8); \ + \ + LOADCHROMA(pu_2, pv_2, 3); \ + PUTFUNC(dst_2, py_2, 3, 6 + 8); \ + \ + LOADCHROMA(pu_1, pv_1, 3); \ + PUTFUNC(dst_1, py_1, 3, 6); \ + ENDYUV2RGBLINE(dst_delta, 0, 0, 1) \ + LOADDITHER \ + \ + LOADCHROMA(pu_1, pv_1, 0); \ + PUTFUNC(dst_1, py_1, 0, 0); \ + \ + LOADCHROMA(pu_2, pv_2, 0); \ + PUTFUNC(dst_2, py_2, 0, 0 + 8); \ + \ + LOADCHROMA(pu_2, pv_2, 1); \ + PUTFUNC(dst_2, py_2, 1, 2 + 8); \ + \ + LOADCHROMA(pu_1, pv_1, 1); \ + PUTFUNC(dst_1, py_1, 1, 2); \ + ENDYUV2RGBLINE(dst_delta, 1, 0, 1) \ + LOADDITHER \ + \ + LOADCHROMA(pu_1, pv_1, 0); \ + PUTFUNC(dst_1, py_1, 0, 0); \ + \ + LOADCHROMA(pu_2, pv_2, 0); \ + PUTFUNC(dst_2, py_2, 0, 0 + 8); \ + ENDYUV2RGBFUNC() + #define LOADDITHER16 \ const uint8_t *d16 = ff_dither_2x2_8[y & 1]; \ const uint8_t *e16 = ff_dither_2x2_4[y & 1]; \ @@ -330,7 +431,7 @@ const int *sws_getCoefficients(int colorspace) g[Y + d64[1 + o]] + \ b[Y + d128[1 + o]]; -YUV2RGBFUNC(yuv2rgb_c_1_ordered_dither, uint8_t, 0) +YUV2RGBFUNC(yuv2rgb_c_1_ordered_dither, uint8_t, 0, 0) const uint8_t *d128 = ff_dither_8x8_220[yd & 7]; char out_1 = 0, out_2 = 0; g = c->table_gU[128 + YUVRGB_TABLE_HEADROOM] + c->table_gV[128 + YUVRGB_TABLE_HEADROOM]; @@ -392,6 +493,7 @@ YUV2RGBFUNC(yuv2rgb_c_1_ordered_dither, uint8_t, 0) dst_2[0] = out_2; ENDYUV2RGBFUNC() +// YUV420 YUV420FUNC(yuv2rgb_c_48, uint8_t, 0, 0, PUTRGB48, 48) YUV420FUNC(yuv2rgb_c_bgr48, uint8_t, 0, 0, PUTBGR48, 48) YUV420FUNC(yuv2rgb_c_32, uint32_t, 0, 0, PUTRGB, 8) @@ -411,6 +513,26 @@ YUV420FUNC_DITHER(yuv2rgb_c_8_ordered_dither, uint8_t, LOADDITHER8, PUTRGB8, YUV420FUNC_DITHER(yuv2rgb_c_4_ordered_dither, uint8_t, LOADDITHER4D, PUTRGB4D, 4) YUV420FUNC_DITHER(yuv2rgb_c_4b_ordered_dither, uint8_t, LOADDITHER4DB, PUTRGB4DB, 8) +// YUV422 +YUV422FUNC(yuv422p_rgb48_c, uint8_t, 0, 0, PUTRGB48, 48) +YUV422FUNC(yuv422p_bgr48_c, uint8_t, 0, 0, PUTBGR48, 48) +YUV422FUNC(yuv422p_rgb32_c, uint32_t, 0, 0, PUTRGB, 8) +#if HAVE_BIGENDIAN +YUV422FUNC(yuva422p_argb_c, uint32_t, 1, 24, PUTRGBA, 8) +YUV422FUNC(yuva422p_rgba_c, uint32_t, 1, 0, PUTRGBA, 8) +#else +YUV422FUNC(yuva422p_rgba_c, uint32_t, 1, 24, PUTRGBA, 8) +YUV422FUNC(yuva422p_argb_c, uint32_t, 1, 0, PUTRGBA, 8) +#endif +YUV422FUNC(yuv422p_rgb24_c, uint8_t, 0, 0, PUTRGB24, 24) +YUV422FUNC(yuv422p_bgr24_c, uint8_t, 0, 0, PUTBGR24, 24) +YUV422FUNC_DITHER(yuv422p_bgr16, uint16_t, LOADDITHER16, PUTRGB16, 8) +YUV422FUNC_DITHER(yuv422p_bgr15, uint16_t, LOADDITHER15, PUTRGB15, 8) +YUV422FUNC_DITHER(yuv422p_bgr12, uint16_t, LOADDITHER12, PUTRGB12, 8) +YUV422FUNC_DITHER(yuv422p_bgr8, uint8_t, LOADDITHER8, PUTRGB8, 8) +YUV422FUNC_DITHER(yuv422p_bgr4, uint8_t, LOADDITHER4D, PUTRGB4D, 4) +YUV422FUNC_DITHER(yuv422p_bgr4_byte, uint8_t, LOADDITHER4DB, PUTRGB4DB, 8) + SwsFunc ff_yuv2rgb_get_func_ptr(SwsContext *c) { SwsFunc t = NULL; @@ -430,6 +552,47 @@ SwsFunc ff_yuv2rgb_get_func_ptr(SwsContext *c) "No accelerated colorspace conversion found from %s to %s.\n", av_get_pix_fmt_name(c->srcFormat), av_get_pix_fmt_name(c->dstFormat)); + if (c->srcFormat == AV_PIX_FMT_YUV422P) { + switch (c->dstFormat) { + case AV_PIX_FMT_BGR48BE: + case AV_PIX_FMT_BGR48LE: + return yuv422p_bgr48_c; + case AV_PIX_FMT_RGB48BE: + case AV_PIX_FMT_RGB48LE: + return yuv422p_rgb48_c; + case AV_PIX_FMT_ARGB: + case AV_PIX_FMT_ABGR: + if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat)) + return yuva422p_argb_c; + case AV_PIX_FMT_RGBA: + case AV_PIX_FMT_BGRA: + return (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat)) ? yuva422p_rgba_c : yuv422p_rgb32_c; + case AV_PIX_FMT_RGB24: + return yuv422p_rgb24_c; + case AV_PIX_FMT_BGR24: + return yuv422p_bgr24_c; + case AV_PIX_FMT_RGB565: + case AV_PIX_FMT_BGR565: + return yuv422p_bgr16; + case AV_PIX_FMT_RGB555: + case AV_PIX_FMT_BGR555: + return yuv422p_bgr15; + case AV_PIX_FMT_RGB444: + case AV_PIX_FMT_BGR444: + return yuv422p_bgr12; + case AV_PIX_FMT_RGB8: + case AV_PIX_FMT_BGR8: + return yuv422p_bgr8; + case AV_PIX_FMT_RGB4: + case AV_PIX_FMT_BGR4: + return yuv422p_bgr4; + case AV_PIX_FMT_RGB4_BYTE: + case AV_PIX_FMT_BGR4_BYTE: + return yuv422p_bgr4_byte; + case AV_PIX_FMT_MONOBLACK: + return yuv2rgb_c_1_ordered_dither; + } + } else { switch (c->dstFormat) { case AV_PIX_FMT_BGR48BE: case AV_PIX_FMT_BGR48LE: @@ -469,6 +632,7 @@ SwsFunc ff_yuv2rgb_get_func_ptr(SwsContext *c) case AV_PIX_FMT_MONOBLACK: return yuv2rgb_c_1_ordered_dither; } + } return NULL; } -- 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-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [FFmpeg-devel] [PATCH 3/6] swscale/yuv2rgb: fix yuv422p input in C code 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 3/6] swscale/yuv2rgb: fix yuv422p input in C code Ramiro Polla @ 2024-06-19 10:13 ` Michael Niedermayer 0 siblings, 0 replies; 14+ messages in thread From: Michael Niedermayer @ 2024-06-19 10:13 UTC (permalink / raw) To: FFmpeg development discussions and patches [-- Attachment #1.1: Type: text/plain, Size: 624 bytes --] On Mon, Jun 17, 2024 at 12:28:46AM +0200, Ramiro Polla wrote: > The C code was silently ignoring the second chroma line on yuv422p > input. > --- > libswscale/yuv2rgb.c | 228 +++++++++++++++++++++++++++++++++++++------ > 1 file changed, 196 insertions(+), 32 deletions(-) nice you found this. Thats an ugly bug thx [...] -- Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB If you fake or manipulate statistics in a paper in physics you will never get a job again. If you fake or manipulate statistics in a paper in medicin you will get a job for life at the pharma industry. [-- Attachment #1.2: signature.asc --] [-- Type: application/pgp-signature, Size: 195 bytes --] [-- Attachment #2: Type: text/plain, Size: 251 bytes --] _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 14+ messages in thread
* [FFmpeg-devel] [PATCH 4/6] swscale/yuv2rgb: reindent after previous commit 2024-06-16 22:28 [FFmpeg-devel] [PATCH 1/6] swscale/yuv2rgb: fix conversion for widths not aligned to 8 Ramiro Polla 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 2/6] swscale/yuv2rgb: add macros to simplify code generation Ramiro Polla 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 3/6] swscale/yuv2rgb: fix yuv422p input in C code Ramiro Polla @ 2024-06-16 22:28 ` Ramiro Polla 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 5/6] checkasm: add tests for yuv2rgb Ramiro Polla 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 6/6] swscale/yuv2rgb/x86: remove mmx/mmxext yuv2rgb functions Ramiro Polla 4 siblings, 0 replies; 14+ messages in thread From: Ramiro Polla @ 2024-06-16 22:28 UTC (permalink / raw) To: ffmpeg-devel --- libswscale/yuv2rgb.c | 78 ++++++++++++++++++++++---------------------- 1 file changed, 39 insertions(+), 39 deletions(-) diff --git a/libswscale/yuv2rgb.c b/libswscale/yuv2rgb.c index 1ea87ac17a..977eb3a7dd 100644 --- a/libswscale/yuv2rgb.c +++ b/libswscale/yuv2rgb.c @@ -593,45 +593,45 @@ SwsFunc ff_yuv2rgb_get_func_ptr(SwsContext *c) return yuv2rgb_c_1_ordered_dither; } } else { - switch (c->dstFormat) { - case AV_PIX_FMT_BGR48BE: - case AV_PIX_FMT_BGR48LE: - return yuv2rgb_c_bgr48; - case AV_PIX_FMT_RGB48BE: - case AV_PIX_FMT_RGB48LE: - return yuv2rgb_c_48; - case AV_PIX_FMT_ARGB: - case AV_PIX_FMT_ABGR: - if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat)) - return yuva2argb_c; - case AV_PIX_FMT_RGBA: - case AV_PIX_FMT_BGRA: - return (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat)) ? yuva2rgba_c : yuv2rgb_c_32; - case AV_PIX_FMT_RGB24: - return yuv2rgb_c_24_rgb; - case AV_PIX_FMT_BGR24: - return yuv2rgb_c_24_bgr; - case AV_PIX_FMT_RGB565: - case AV_PIX_FMT_BGR565: - return yuv2rgb_c_16_ordered_dither; - case AV_PIX_FMT_RGB555: - case AV_PIX_FMT_BGR555: - return yuv2rgb_c_15_ordered_dither; - case AV_PIX_FMT_RGB444: - case AV_PIX_FMT_BGR444: - return yuv2rgb_c_12_ordered_dither; - case AV_PIX_FMT_RGB8: - case AV_PIX_FMT_BGR8: - return yuv2rgb_c_8_ordered_dither; - case AV_PIX_FMT_RGB4: - case AV_PIX_FMT_BGR4: - return yuv2rgb_c_4_ordered_dither; - case AV_PIX_FMT_RGB4_BYTE: - case AV_PIX_FMT_BGR4_BYTE: - return yuv2rgb_c_4b_ordered_dither; - case AV_PIX_FMT_MONOBLACK: - return yuv2rgb_c_1_ordered_dither; - } + switch (c->dstFormat) { + case AV_PIX_FMT_BGR48BE: + case AV_PIX_FMT_BGR48LE: + return yuv2rgb_c_bgr48; + case AV_PIX_FMT_RGB48BE: + case AV_PIX_FMT_RGB48LE: + return yuv2rgb_c_48; + case AV_PIX_FMT_ARGB: + case AV_PIX_FMT_ABGR: + if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat)) + return yuva2argb_c; + case AV_PIX_FMT_RGBA: + case AV_PIX_FMT_BGRA: + return (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat)) ? yuva2rgba_c : yuv2rgb_c_32; + case AV_PIX_FMT_RGB24: + return yuv2rgb_c_24_rgb; + case AV_PIX_FMT_BGR24: + return yuv2rgb_c_24_bgr; + case AV_PIX_FMT_RGB565: + case AV_PIX_FMT_BGR565: + return yuv2rgb_c_16_ordered_dither; + case AV_PIX_FMT_RGB555: + case AV_PIX_FMT_BGR555: + return yuv2rgb_c_15_ordered_dither; + case AV_PIX_FMT_RGB444: + case AV_PIX_FMT_BGR444: + return yuv2rgb_c_12_ordered_dither; + case AV_PIX_FMT_RGB8: + case AV_PIX_FMT_BGR8: + return yuv2rgb_c_8_ordered_dither; + case AV_PIX_FMT_RGB4: + case AV_PIX_FMT_BGR4: + return yuv2rgb_c_4_ordered_dither; + case AV_PIX_FMT_RGB4_BYTE: + case AV_PIX_FMT_BGR4_BYTE: + return yuv2rgb_c_4b_ordered_dither; + case AV_PIX_FMT_MONOBLACK: + return yuv2rgb_c_1_ordered_dither; + } } return NULL; } -- 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-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 14+ messages in thread
* [FFmpeg-devel] [PATCH 5/6] checkasm: add tests for yuv2rgb 2024-06-16 22:28 [FFmpeg-devel] [PATCH 1/6] swscale/yuv2rgb: fix conversion for widths not aligned to 8 Ramiro Polla ` (2 preceding siblings ...) 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 4/6] swscale/yuv2rgb: reindent after previous commit Ramiro Polla @ 2024-06-16 22:28 ` Ramiro Polla 2024-06-20 14:59 ` Ramiro Polla 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 6/6] swscale/yuv2rgb/x86: remove mmx/mmxext yuv2rgb functions Ramiro Polla 4 siblings, 1 reply; 14+ messages in thread From: Ramiro Polla @ 2024-06-16 22:28 UTC (permalink / raw) To: ffmpeg-devel --- tests/checkasm/Makefile | 2 +- tests/checkasm/checkasm.c | 1 + tests/checkasm/checkasm.h | 1 + tests/checkasm/sw_yuv2rgb.c | 205 ++++++++++++++++++++++++++++++++++++ tests/fate/checkasm.mak | 1 + 5 files changed, 209 insertions(+), 1 deletion(-) create mode 100644 tests/checkasm/sw_yuv2rgb.c diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile index f20732b37a..3a7670e24b 100644 --- a/tests/checkasm/Makefile +++ b/tests/checkasm/Makefile @@ -63,7 +63,7 @@ AVFILTEROBJS-$(CONFIG_SOBEL_FILTER) += vf_convolution.o CHECKASMOBJS-$(CONFIG_AVFILTER) += $(AVFILTEROBJS-yes) # swscale tests -SWSCALEOBJS += sw_gbrp.o sw_range_convert.o sw_rgb.o sw_scale.o +SWSCALEOBJS += sw_gbrp.o sw_range_convert.o sw_rgb.o sw_scale.o sw_yuv2rgb.o CHECKASMOBJS-$(CONFIG_SWSCALE) += $(SWSCALEOBJS) diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c index 56232ab1e0..d9ac772a08 100644 --- a/tests/checkasm/checkasm.c +++ b/tests/checkasm/checkasm.c @@ -254,6 +254,7 @@ static const struct { { "sw_range_convert", checkasm_check_sw_range_convert }, { "sw_rgb", checkasm_check_sw_rgb }, { "sw_scale", checkasm_check_sw_scale }, + { "sw_yuv2rgb", checkasm_check_sw_yuv2rgb }, #endif #if CONFIG_AVUTIL { "fixed_dsp", checkasm_check_fixed_dsp }, diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h index e544007b67..4d5f3e387e 100644 --- a/tests/checkasm/checkasm.h +++ b/tests/checkasm/checkasm.h @@ -122,6 +122,7 @@ void checkasm_check_sw_gbrp(void); void checkasm_check_sw_range_convert(void); void checkasm_check_sw_rgb(void); void checkasm_check_sw_scale(void); +void checkasm_check_sw_yuv2rgb(void); void checkasm_check_takdsp(void); void checkasm_check_utvideodsp(void); void checkasm_check_v210dec(void); diff --git a/tests/checkasm/sw_yuv2rgb.c b/tests/checkasm/sw_yuv2rgb.c new file mode 100644 index 0000000000..fbe01a7788 --- /dev/null +++ b/tests/checkasm/sw_yuv2rgb.c @@ -0,0 +1,205 @@ +/* + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License along + * with FFmpeg; if not, write to the Free Software Foundation, Inc., + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. + */ + +#include <string.h> + +#include "libavutil/common.h" +#include "libavutil/intreadwrite.h" +#include "libavutil/mem_internal.h" +#include "libavutil/pixdesc.h" + +#include "libswscale/swscale.h" +#include "libswscale/swscale_internal.h" + +#include "checkasm.h" + +#define randomize_buffers(buf, size) \ + do { \ + for (int j = 0; j < size; j += 4) \ + AV_WN32(buf + j, rnd()); \ + } while (0) + +static const int dst_fmts[] = { +// AV_PIX_FMT_BGR48BE, +// AV_PIX_FMT_BGR48LE, +// AV_PIX_FMT_RGB48BE, +// AV_PIX_FMT_RGB48LE, + AV_PIX_FMT_ARGB, + AV_PIX_FMT_ABGR, + AV_PIX_FMT_RGBA, + AV_PIX_FMT_BGRA, + AV_PIX_FMT_RGB24, + AV_PIX_FMT_BGR24, + AV_PIX_FMT_RGB565, + AV_PIX_FMT_BGR565, + AV_PIX_FMT_RGB555, + AV_PIX_FMT_BGR555, +// AV_PIX_FMT_RGB444, +// AV_PIX_FMT_BGR444, +// 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, +// AV_PIX_FMT_MONOBLACK, +}; + +static int cmp_off_by_n(const uint8_t *ref, const uint8_t *test, size_t n, int accuracy) +{ + for (size_t i = 0; i < n; i++) { + if (abs(ref[i] - test[i]) > accuracy) + return 1; + } + return 0; +} + +static int cmp_555_by_n(const uint8_t *ref, const uint8_t *test, size_t n, int accuracy) +{ + const uint16_t *ref16 = (const uint16_t *) ref; + const uint16_t *test16 = (const uint16_t *) test; + for (size_t i = 0; i < n; i++) { + if (abs(( ref16[i] & 0x1f) - ( test16[i] & 0x1f)) > accuracy) + return 1; + if (abs(((ref16[i] >> 5) & 0x1f) - ((test16[i] >> 5) & 0x1f)) > accuracy) + return 1; + if (abs(((ref16[i] >> 10) & 0x1f) - ((test16[i] >> 10) & 0x1f)) > accuracy) + return 1; + } + return 0; +} + +static int cmp_565_by_n(const uint8_t *ref, const uint8_t *test, size_t n, int accuracy) +{ + const uint16_t *ref16 = (const uint16_t *) ref; + const uint16_t *test16 = (const uint16_t *) test; + for (size_t i = 0; i < n; i++) { + if (abs(( ref16[i] & 0x1f) - ( test16[i] & 0x1f)) > accuracy) + return 1; + if (abs(((ref16[i] >> 5) & 0x3f) - ((test16[i] >> 5) & 0x3f)) > accuracy) + return 1; + if (abs(((ref16[i] >> 11) & 0x1f) - ((test16[i] >> 11) & 0x1f)) > accuracy) + return 1; + } + return 0; +} + +static void check_yuv2rgb(int src_pix_fmt) +{ + const AVPixFmtDescriptor *src_desc = av_pix_fmt_desc_get(src_pix_fmt); +#define MAX_LINE_SIZE 1920 + static const int input_sizes[] = {8, 128, 1080, MAX_LINE_SIZE}; + + declare_func_emms(AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT, + int, SwsContext *c, const uint8_t *src[], + int srcStride[], int srcSliceY, int srcSliceH, + uint8_t *dst[], int dstStride[]); + + LOCAL_ALIGNED_8(uint8_t, src_y, [MAX_LINE_SIZE * 2]); + LOCAL_ALIGNED_8(uint8_t, src_u, [MAX_LINE_SIZE]); + LOCAL_ALIGNED_8(uint8_t, src_v, [MAX_LINE_SIZE]); + LOCAL_ALIGNED_8(uint8_t, src_a, [MAX_LINE_SIZE * 2]); + const uint8_t *src[4] = { src_y, src_u, src_v, src_a }; + + LOCAL_ALIGNED_8(uint8_t, dst0_, [2 * MAX_LINE_SIZE * 6]); + uint8_t *dst0[4] = { dst0_ }; + uint8_t *lines0[2] = { dst0_, dst0_ + MAX_LINE_SIZE * 6 }; + + LOCAL_ALIGNED_8(uint8_t, dst1_, [2 * MAX_LINE_SIZE * 6]); + uint8_t *dst1[4] = { dst1_ }; + uint8_t *lines1[2] = { dst1_, dst1_ + MAX_LINE_SIZE * 6 }; + + randomize_buffers(src_y, MAX_LINE_SIZE * 2); + randomize_buffers(src_u, MAX_LINE_SIZE); + randomize_buffers(src_v, MAX_LINE_SIZE); + randomize_buffers(src_a, MAX_LINE_SIZE * 2); + + for (int dfi = 0; dfi < FF_ARRAY_ELEMS(dst_fmts); dfi++) { + int dst_pix_fmt = dst_fmts[dfi]; + const AVPixFmtDescriptor *dst_desc = av_pix_fmt_desc_get(dst_pix_fmt); + int sample_size = av_get_padded_bits_per_pixel(dst_desc) >> 3; + for (int isi = 0; isi < FF_ARRAY_ELEMS(input_sizes); isi++) { + struct SwsContext *ctx; + int width = input_sizes[isi]; + int srcSliceY = 0; + int srcSliceH = 2; + int srcStride[4] = { + width, + width >> src_desc->log2_chroma_w, + width >> src_desc->log2_chroma_w, + width, + }; + int dstStride[4] = { MAX_LINE_SIZE * 6 }; + + ctx = sws_getContext(width, srcSliceH, src_pix_fmt, + width, srcSliceH, dst_pix_fmt, + 0, NULL, NULL, NULL); + if (!ctx) + fail(); + + if (check_func(ctx->convert_unscaled, "%s_%s_%d", src_desc->name, dst_desc->name, width)) { + memset(dst0_, 0xFF, 2 * MAX_LINE_SIZE * 6); + memset(dst1_, 0xFF, 2 * MAX_LINE_SIZE * 6); + + call_ref(ctx, src, srcStride, srcSliceY, + srcSliceH, dst0, dstStride); + call_new(ctx, src, srcStride, srcSliceY, + srcSliceH, dst1, dstStride); + + if (dst_pix_fmt == AV_PIX_FMT_ARGB || + dst_pix_fmt == AV_PIX_FMT_ABGR || + dst_pix_fmt == AV_PIX_FMT_RGBA || + dst_pix_fmt == AV_PIX_FMT_BGRA || + dst_pix_fmt == AV_PIX_FMT_RGB24 || + dst_pix_fmt == AV_PIX_FMT_BGR24) { + if (cmp_off_by_n(lines0[0], lines1[0], width * sample_size, 3) || + cmp_off_by_n(lines0[1], lines1[1], width * sample_size, 3)) + fail(); + } else if (dst_pix_fmt == AV_PIX_FMT_RGB565 || + dst_pix_fmt == AV_PIX_FMT_BGR565) { + if (cmp_565_by_n(lines0[0], lines1[0], width, 2) || + cmp_565_by_n(lines0[1], lines1[1], width, 2)) + fail(); + } else if (dst_pix_fmt == AV_PIX_FMT_RGB555 || + dst_pix_fmt == AV_PIX_FMT_BGR555) { + if (cmp_555_by_n(lines0[0], lines1[0], width, 2) || + cmp_555_by_n(lines0[1], lines1[1], width, 2)) + fail(); + } else { + fail(); + } + + bench_new(ctx, src, srcStride, srcSliceY, + srcSliceH, dst0, dstStride); + } + sws_freeContext(ctx); + } + } +} + +#undef MAX_LINE_SIZE + +void checkasm_check_sw_yuv2rgb(void) +{ + check_yuv2rgb(AV_PIX_FMT_YUV420P); + report("yuv420p"); + check_yuv2rgb(AV_PIX_FMT_YUV422P); + report("yuv422p"); + check_yuv2rgb(AV_PIX_FMT_YUVA420P); + report("yuva420p"); +} diff --git a/tests/fate/checkasm.mak b/tests/fate/checkasm.mak index 0ed2ea5be6..49832b09bf 100644 --- a/tests/fate/checkasm.mak +++ b/tests/fate/checkasm.mak @@ -44,6 +44,7 @@ FATE_CHECKASM = fate-checkasm-aacencdsp \ fate-checkasm-sw_range_convert \ fate-checkasm-sw_rgb \ fate-checkasm-sw_scale \ + fate-checkasm-sw_yuv2rgb \ fate-checkasm-takdsp \ fate-checkasm-utvideodsp \ fate-checkasm-v210dec \ -- 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-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [FFmpeg-devel] [PATCH 5/6] checkasm: add tests for yuv2rgb 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 5/6] checkasm: add tests for yuv2rgb Ramiro Polla @ 2024-06-20 14:59 ` Ramiro Polla 2024-06-25 21:19 ` Ramiro Polla 0 siblings, 1 reply; 14+ messages in thread From: Ramiro Polla @ 2024-06-20 14:59 UTC (permalink / raw) To: ffmpeg-devel On Mon, Jun 17, 2024 at 12:28 AM Ramiro Polla <ramiro.polla@gmail.com> wrote: > > --- > tests/checkasm/Makefile | 2 +- > tests/checkasm/checkasm.c | 1 + > tests/checkasm/checkasm.h | 1 + > tests/checkasm/sw_yuv2rgb.c | 205 ++++++++++++++++++++++++++++++++++++ > tests/fate/checkasm.mak | 1 + > 5 files changed, 209 insertions(+), 1 deletion(-) > create mode 100644 tests/checkasm/sw_yuv2rgb.c > > diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile > index f20732b37a..3a7670e24b 100644 > --- a/tests/checkasm/Makefile > +++ b/tests/checkasm/Makefile > @@ -63,7 +63,7 @@ AVFILTEROBJS-$(CONFIG_SOBEL_FILTER) += vf_convolution.o > CHECKASMOBJS-$(CONFIG_AVFILTER) += $(AVFILTEROBJS-yes) > > # swscale tests > -SWSCALEOBJS += sw_gbrp.o sw_range_convert.o sw_rgb.o sw_scale.o > +SWSCALEOBJS += sw_gbrp.o sw_range_convert.o sw_rgb.o sw_scale.o sw_yuv2rgb.o > > CHECKASMOBJS-$(CONFIG_SWSCALE) += $(SWSCALEOBJS) > > diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c > index 56232ab1e0..d9ac772a08 100644 > --- a/tests/checkasm/checkasm.c > +++ b/tests/checkasm/checkasm.c > @@ -254,6 +254,7 @@ static const struct { > { "sw_range_convert", checkasm_check_sw_range_convert }, > { "sw_rgb", checkasm_check_sw_rgb }, > { "sw_scale", checkasm_check_sw_scale }, > + { "sw_yuv2rgb", checkasm_check_sw_yuv2rgb }, > #endif > #if CONFIG_AVUTIL > { "fixed_dsp", checkasm_check_fixed_dsp }, > diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h > index e544007b67..4d5f3e387e 100644 > --- a/tests/checkasm/checkasm.h > +++ b/tests/checkasm/checkasm.h > @@ -122,6 +122,7 @@ void checkasm_check_sw_gbrp(void); > void checkasm_check_sw_range_convert(void); > void checkasm_check_sw_rgb(void); > void checkasm_check_sw_scale(void); > +void checkasm_check_sw_yuv2rgb(void); > void checkasm_check_takdsp(void); > void checkasm_check_utvideodsp(void); > void checkasm_check_v210dec(void); > diff --git a/tests/checkasm/sw_yuv2rgb.c b/tests/checkasm/sw_yuv2rgb.c > new file mode 100644 > index 0000000000..fbe01a7788 > --- /dev/null > +++ b/tests/checkasm/sw_yuv2rgb.c > @@ -0,0 +1,205 @@ > +/* > + * This file is part of FFmpeg. > + * > + * FFmpeg is free software; you can redistribute it and/or modify > + * it under the terms of the GNU General Public License as published by > + * the Free Software Foundation; either version 2 of the License, or > + * (at your option) any later version. > + * > + * FFmpeg is distributed in the hope that it will be useful, > + * but WITHOUT ANY WARRANTY; without even the implied warranty of > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the > + * GNU General Public License for more details. > + * > + * You should have received a copy of the GNU General Public License along > + * with FFmpeg; if not, write to the Free Software Foundation, Inc., > + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. > + */ > + > +#include <string.h> > + > +#include "libavutil/common.h" > +#include "libavutil/intreadwrite.h" > +#include "libavutil/mem_internal.h" > +#include "libavutil/pixdesc.h" > + > +#include "libswscale/swscale.h" > +#include "libswscale/swscale_internal.h" > + > +#include "checkasm.h" > + > +#define randomize_buffers(buf, size) \ > + do { \ > + for (int j = 0; j < size; j += 4) \ > + AV_WN32(buf + j, rnd()); \ > + } while (0) > + > +static const int dst_fmts[] = { > +// AV_PIX_FMT_BGR48BE, > +// AV_PIX_FMT_BGR48LE, > +// AV_PIX_FMT_RGB48BE, > +// AV_PIX_FMT_RGB48LE, > + AV_PIX_FMT_ARGB, > + AV_PIX_FMT_ABGR, > + AV_PIX_FMT_RGBA, > + AV_PIX_FMT_BGRA, > + AV_PIX_FMT_RGB24, > + AV_PIX_FMT_BGR24, > + AV_PIX_FMT_RGB565, > + AV_PIX_FMT_BGR565, > + AV_PIX_FMT_RGB555, > + AV_PIX_FMT_BGR555, > +// AV_PIX_FMT_RGB444, > +// AV_PIX_FMT_BGR444, > +// 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, > +// AV_PIX_FMT_MONOBLACK, > +}; > + > +static int cmp_off_by_n(const uint8_t *ref, const uint8_t *test, size_t n, int accuracy) > +{ > + for (size_t i = 0; i < n; i++) { > + if (abs(ref[i] - test[i]) > accuracy) > + return 1; > + } > + return 0; > +} > + > +static int cmp_555_by_n(const uint8_t *ref, const uint8_t *test, size_t n, int accuracy) > +{ > + const uint16_t *ref16 = (const uint16_t *) ref; > + const uint16_t *test16 = (const uint16_t *) test; > + for (size_t i = 0; i < n; i++) { > + if (abs(( ref16[i] & 0x1f) - ( test16[i] & 0x1f)) > accuracy) > + return 1; > + if (abs(((ref16[i] >> 5) & 0x1f) - ((test16[i] >> 5) & 0x1f)) > accuracy) > + return 1; > + if (abs(((ref16[i] >> 10) & 0x1f) - ((test16[i] >> 10) & 0x1f)) > accuracy) > + return 1; > + } > + return 0; > +} > + > +static int cmp_565_by_n(const uint8_t *ref, const uint8_t *test, size_t n, int accuracy) > +{ > + const uint16_t *ref16 = (const uint16_t *) ref; > + const uint16_t *test16 = (const uint16_t *) test; > + for (size_t i = 0; i < n; i++) { > + if (abs(( ref16[i] & 0x1f) - ( test16[i] & 0x1f)) > accuracy) > + return 1; > + if (abs(((ref16[i] >> 5) & 0x3f) - ((test16[i] >> 5) & 0x3f)) > accuracy) > + return 1; > + if (abs(((ref16[i] >> 11) & 0x1f) - ((test16[i] >> 11) & 0x1f)) > accuracy) > + return 1; > + } > + return 0; > +} > + > +static void check_yuv2rgb(int src_pix_fmt) > +{ > + const AVPixFmtDescriptor *src_desc = av_pix_fmt_desc_get(src_pix_fmt); > +#define MAX_LINE_SIZE 1920 > + static const int input_sizes[] = {8, 128, 1080, MAX_LINE_SIZE}; > + > + declare_func_emms(AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT, > + int, SwsContext *c, const uint8_t *src[], > + int srcStride[], int srcSliceY, int srcSliceH, > + uint8_t *dst[], int dstStride[]); > + > + LOCAL_ALIGNED_8(uint8_t, src_y, [MAX_LINE_SIZE * 2]); > + LOCAL_ALIGNED_8(uint8_t, src_u, [MAX_LINE_SIZE]); > + LOCAL_ALIGNED_8(uint8_t, src_v, [MAX_LINE_SIZE]); > + LOCAL_ALIGNED_8(uint8_t, src_a, [MAX_LINE_SIZE * 2]); > + const uint8_t *src[4] = { src_y, src_u, src_v, src_a }; > + > + LOCAL_ALIGNED_8(uint8_t, dst0_, [2 * MAX_LINE_SIZE * 6]); > + uint8_t *dst0[4] = { dst0_ }; > + uint8_t *lines0[2] = { dst0_, dst0_ + MAX_LINE_SIZE * 6 }; > + > + LOCAL_ALIGNED_8(uint8_t, dst1_, [2 * MAX_LINE_SIZE * 6]); > + uint8_t *dst1[4] = { dst1_ }; > + uint8_t *lines1[2] = { dst1_, dst1_ + MAX_LINE_SIZE * 6 }; > + > + randomize_buffers(src_y, MAX_LINE_SIZE * 2); > + randomize_buffers(src_u, MAX_LINE_SIZE); > + randomize_buffers(src_v, MAX_LINE_SIZE); > + randomize_buffers(src_a, MAX_LINE_SIZE * 2); > + > + for (int dfi = 0; dfi < FF_ARRAY_ELEMS(dst_fmts); dfi++) { > + int dst_pix_fmt = dst_fmts[dfi]; > + const AVPixFmtDescriptor *dst_desc = av_pix_fmt_desc_get(dst_pix_fmt); > + int sample_size = av_get_padded_bits_per_pixel(dst_desc) >> 3; > + for (int isi = 0; isi < FF_ARRAY_ELEMS(input_sizes); isi++) { > + struct SwsContext *ctx; > + int width = input_sizes[isi]; > + int srcSliceY = 0; > + int srcSliceH = 2; > + int srcStride[4] = { > + width, > + width >> src_desc->log2_chroma_w, > + width >> src_desc->log2_chroma_w, > + width, > + }; > + int dstStride[4] = { MAX_LINE_SIZE * 6 }; > + > + ctx = sws_getContext(width, srcSliceH, src_pix_fmt, > + width, srcSliceH, dst_pix_fmt, > + 0, NULL, NULL, NULL); checkasm gets a little spammy with this commit by printing "No accelerated colorspace conversion found from %s to %s.\n" from libswscale/yuv2rgb.c multiple times. Any suggestions? Perhaps downgrading that message to AV_LOG_VERBOSE, or raising loglevel around the call to sws_getContext()? > + if (!ctx) > + fail(); > + > + if (check_func(ctx->convert_unscaled, "%s_%s_%d", src_desc->name, dst_desc->name, width)) { > + memset(dst0_, 0xFF, 2 * MAX_LINE_SIZE * 6); > + memset(dst1_, 0xFF, 2 * MAX_LINE_SIZE * 6); > + > + call_ref(ctx, src, srcStride, srcSliceY, > + srcSliceH, dst0, dstStride); > + call_new(ctx, src, srcStride, srcSliceY, > + srcSliceH, dst1, dstStride); > + > + if (dst_pix_fmt == AV_PIX_FMT_ARGB || > + dst_pix_fmt == AV_PIX_FMT_ABGR || > + dst_pix_fmt == AV_PIX_FMT_RGBA || > + dst_pix_fmt == AV_PIX_FMT_BGRA || > + dst_pix_fmt == AV_PIX_FMT_RGB24 || > + dst_pix_fmt == AV_PIX_FMT_BGR24) { > + if (cmp_off_by_n(lines0[0], lines1[0], width * sample_size, 3) || > + cmp_off_by_n(lines0[1], lines1[1], width * sample_size, 3)) > + fail(); > + } else if (dst_pix_fmt == AV_PIX_FMT_RGB565 || > + dst_pix_fmt == AV_PIX_FMT_BGR565) { > + if (cmp_565_by_n(lines0[0], lines1[0], width, 2) || > + cmp_565_by_n(lines0[1], lines1[1], width, 2)) > + fail(); > + } else if (dst_pix_fmt == AV_PIX_FMT_RGB555 || > + dst_pix_fmt == AV_PIX_FMT_BGR555) { > + if (cmp_555_by_n(lines0[0], lines1[0], width, 2) || > + cmp_555_by_n(lines0[1], lines1[1], width, 2)) > + fail(); > + } else { > + fail(); > + } > + > + bench_new(ctx, src, srcStride, srcSliceY, > + srcSliceH, dst0, dstStride); > + } > + sws_freeContext(ctx); > + } > + } > +} > + > +#undef MAX_LINE_SIZE > + > +void checkasm_check_sw_yuv2rgb(void) > +{ > + check_yuv2rgb(AV_PIX_FMT_YUV420P); > + report("yuv420p"); > + check_yuv2rgb(AV_PIX_FMT_YUV422P); > + report("yuv422p"); > + check_yuv2rgb(AV_PIX_FMT_YUVA420P); > + report("yuva420p"); > +} > diff --git a/tests/fate/checkasm.mak b/tests/fate/checkasm.mak > index 0ed2ea5be6..49832b09bf 100644 > --- a/tests/fate/checkasm.mak > +++ b/tests/fate/checkasm.mak > @@ -44,6 +44,7 @@ FATE_CHECKASM = fate-checkasm-aacencdsp \ > fate-checkasm-sw_range_convert \ > fate-checkasm-sw_rgb \ > fate-checkasm-sw_scale \ > + fate-checkasm-sw_yuv2rgb \ > fate-checkasm-takdsp \ > fate-checkasm-utvideodsp \ > fate-checkasm-v210dec \ > -- > 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-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [FFmpeg-devel] [PATCH 5/6] checkasm: add tests for yuv2rgb 2024-06-20 14:59 ` Ramiro Polla @ 2024-06-25 21:19 ` Ramiro Polla 2024-06-27 14:02 ` Ramiro Polla 0 siblings, 1 reply; 14+ messages in thread From: Ramiro Polla @ 2024-06-25 21:19 UTC (permalink / raw) To: ffmpeg-devel [-- Attachment #1: Type: text/plain, Size: 456 bytes --] On Thu, Jun 20, 2024 at 4:59 PM Ramiro Polla <ramiro.polla@gmail.com> wrote: > checkasm gets a little spammy with this commit by printing "No > accelerated colorspace conversion found from %s to %s.\n" from > libswscale/yuv2rgb.c multiple times. Any suggestions? Perhaps > downgrading that message to AV_LOG_VERBOSE, or raising loglevel around > the call to sws_getContext()? New patch attached that overrides the log level to prevent spamming. [-- Attachment #2: 0001-checkasm-add-tests-for-yuv2rgb.patch --] [-- Type: text/x-patch, Size: 11009 bytes --] From c0359653f11a5e6ae39445134c2e73cb211c4a65 Mon Sep 17 00:00:00 2001 From: Ramiro Polla <ramiro.polla@gmail.com> Date: Sun, 16 Jun 2024 22:40:13 +0200 Subject: [PATCH] checkasm: add tests for yuv2rgb --- tests/checkasm/Makefile | 2 +- tests/checkasm/checkasm.c | 1 + tests/checkasm/checkasm.h | 1 + tests/checkasm/sw_yuv2rgb.c | 211 ++++++++++++++++++++++++++++++++++++ tests/fate/checkasm.mak | 1 + 5 files changed, 215 insertions(+), 1 deletion(-) create mode 100644 tests/checkasm/sw_yuv2rgb.c diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile index f20732b37a..3a7670e24b 100644 --- a/tests/checkasm/Makefile +++ b/tests/checkasm/Makefile @@ -63,7 +63,7 @@ AVFILTEROBJS-$(CONFIG_SOBEL_FILTER) += vf_convolution.o CHECKASMOBJS-$(CONFIG_AVFILTER) += $(AVFILTEROBJS-yes) # swscale tests -SWSCALEOBJS += sw_gbrp.o sw_range_convert.o sw_rgb.o sw_scale.o +SWSCALEOBJS += sw_gbrp.o sw_range_convert.o sw_rgb.o sw_scale.o sw_yuv2rgb.o CHECKASMOBJS-$(CONFIG_SWSCALE) += $(SWSCALEOBJS) diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c index b88b128e4c..de0024099a 100644 --- a/tests/checkasm/checkasm.c +++ b/tests/checkasm/checkasm.c @@ -254,6 +254,7 @@ static const struct { { "sw_range_convert", checkasm_check_sw_range_convert }, { "sw_rgb", checkasm_check_sw_rgb }, { "sw_scale", checkasm_check_sw_scale }, + { "sw_yuv2rgb", checkasm_check_sw_yuv2rgb }, #endif #if CONFIG_AVUTIL { "fixed_dsp", checkasm_check_fixed_dsp }, diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h index e544007b67..4d5f3e387e 100644 --- a/tests/checkasm/checkasm.h +++ b/tests/checkasm/checkasm.h @@ -122,6 +122,7 @@ void checkasm_check_sw_gbrp(void); void checkasm_check_sw_range_convert(void); void checkasm_check_sw_rgb(void); void checkasm_check_sw_scale(void); +void checkasm_check_sw_yuv2rgb(void); void checkasm_check_takdsp(void); void checkasm_check_utvideodsp(void); void checkasm_check_v210dec(void); diff --git a/tests/checkasm/sw_yuv2rgb.c b/tests/checkasm/sw_yuv2rgb.c new file mode 100644 index 0000000000..02ed9a74d5 --- /dev/null +++ b/tests/checkasm/sw_yuv2rgb.c @@ -0,0 +1,211 @@ +/* + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License along + * with FFmpeg; if not, write to the Free Software Foundation, Inc., + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. + */ + +#include <string.h> + +#include "libavutil/common.h" +#include "libavutil/intreadwrite.h" +#include "libavutil/mem_internal.h" +#include "libavutil/pixdesc.h" + +#include "libswscale/swscale.h" +#include "libswscale/swscale_internal.h" + +#include "checkasm.h" + +#define randomize_buffers(buf, size) \ + do { \ + for (int j = 0; j < size; j += 4) \ + AV_WN32(buf + j, rnd()); \ + } while (0) + +static const int dst_fmts[] = { +// AV_PIX_FMT_BGR48BE, +// AV_PIX_FMT_BGR48LE, +// AV_PIX_FMT_RGB48BE, +// AV_PIX_FMT_RGB48LE, + AV_PIX_FMT_ARGB, + AV_PIX_FMT_ABGR, + AV_PIX_FMT_RGBA, + AV_PIX_FMT_BGRA, + AV_PIX_FMT_RGB24, + AV_PIX_FMT_BGR24, + AV_PIX_FMT_RGB565, + AV_PIX_FMT_BGR565, + AV_PIX_FMT_RGB555, + AV_PIX_FMT_BGR555, +// AV_PIX_FMT_RGB444, +// AV_PIX_FMT_BGR444, +// 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, +// AV_PIX_FMT_MONOBLACK, +}; + +static int cmp_off_by_n(const uint8_t *ref, const uint8_t *test, size_t n, int accuracy) +{ + for (size_t i = 0; i < n; i++) { + if (abs(ref[i] - test[i]) > accuracy) + return 1; + } + return 0; +} + +static int cmp_555_by_n(const uint8_t *ref, const uint8_t *test, size_t n, int accuracy) +{ + const uint16_t *ref16 = (const uint16_t *) ref; + const uint16_t *test16 = (const uint16_t *) test; + for (size_t i = 0; i < n; i++) { + if (abs(( ref16[i] & 0x1f) - ( test16[i] & 0x1f)) > accuracy) + return 1; + if (abs(((ref16[i] >> 5) & 0x1f) - ((test16[i] >> 5) & 0x1f)) > accuracy) + return 1; + if (abs(((ref16[i] >> 10) & 0x1f) - ((test16[i] >> 10) & 0x1f)) > accuracy) + return 1; + } + return 0; +} + +static int cmp_565_by_n(const uint8_t *ref, const uint8_t *test, size_t n, int accuracy) +{ + const uint16_t *ref16 = (const uint16_t *) ref; + const uint16_t *test16 = (const uint16_t *) test; + for (size_t i = 0; i < n; i++) { + if (abs(( ref16[i] & 0x1f) - ( test16[i] & 0x1f)) > accuracy) + return 1; + if (abs(((ref16[i] >> 5) & 0x3f) - ((test16[i] >> 5) & 0x3f)) > accuracy) + return 1; + if (abs(((ref16[i] >> 11) & 0x1f) - ((test16[i] >> 11) & 0x1f)) > accuracy) + return 1; + } + return 0; +} + +static void check_yuv2rgb(int src_pix_fmt) +{ + const AVPixFmtDescriptor *src_desc = av_pix_fmt_desc_get(src_pix_fmt); +#define MAX_LINE_SIZE 1920 + static const int input_sizes[] = {8, 128, 1080, MAX_LINE_SIZE}; + + declare_func_emms(AV_CPU_FLAG_MMX | AV_CPU_FLAG_MMXEXT, + int, SwsContext *c, const uint8_t *src[], + int srcStride[], int srcSliceY, int srcSliceH, + uint8_t *dst[], int dstStride[]); + + LOCAL_ALIGNED_8(uint8_t, src_y, [MAX_LINE_SIZE * 2]); + LOCAL_ALIGNED_8(uint8_t, src_u, [MAX_LINE_SIZE]); + LOCAL_ALIGNED_8(uint8_t, src_v, [MAX_LINE_SIZE]); + LOCAL_ALIGNED_8(uint8_t, src_a, [MAX_LINE_SIZE * 2]); + const uint8_t *src[4] = { src_y, src_u, src_v, src_a }; + + LOCAL_ALIGNED_8(uint8_t, dst0_, [2 * MAX_LINE_SIZE * 6]); + uint8_t *dst0[4] = { dst0_ }; + uint8_t *lines0[2] = { dst0_, dst0_ + MAX_LINE_SIZE * 6 }; + + LOCAL_ALIGNED_8(uint8_t, dst1_, [2 * MAX_LINE_SIZE * 6]); + uint8_t *dst1[4] = { dst1_ }; + uint8_t *lines1[2] = { dst1_, dst1_ + MAX_LINE_SIZE * 6 }; + + randomize_buffers(src_y, MAX_LINE_SIZE * 2); + randomize_buffers(src_u, MAX_LINE_SIZE); + randomize_buffers(src_v, MAX_LINE_SIZE); + randomize_buffers(src_a, MAX_LINE_SIZE * 2); + + for (int dfi = 0; dfi < FF_ARRAY_ELEMS(dst_fmts); dfi++) { + int dst_pix_fmt = dst_fmts[dfi]; + const AVPixFmtDescriptor *dst_desc = av_pix_fmt_desc_get(dst_pix_fmt); + int sample_size = av_get_padded_bits_per_pixel(dst_desc) >> 3; + for (int isi = 0; isi < FF_ARRAY_ELEMS(input_sizes); isi++) { + struct SwsContext *ctx; + int log_level; + int width = input_sizes[isi]; + int srcSliceY = 0; + int srcSliceH = 2; + int srcStride[4] = { + width, + width >> src_desc->log2_chroma_w, + width >> src_desc->log2_chroma_w, + width, + }; + int dstStride[4] = { MAX_LINE_SIZE * 6 }; + + // override log level to prevent spamming of the message + // "No accelerated colorspace conversion found from %s to %s" + log_level = av_log_get_level(); + av_log_set_level(AV_LOG_ERROR); + ctx = sws_getContext(width, srcSliceH, src_pix_fmt, + width, srcSliceH, dst_pix_fmt, + 0, NULL, NULL, NULL); + av_log_set_level(log_level); + if (!ctx) + fail(); + + if (check_func(ctx->convert_unscaled, "%s_%s_%d", src_desc->name, dst_desc->name, width)) { + memset(dst0_, 0xFF, 2 * MAX_LINE_SIZE * 6); + memset(dst1_, 0xFF, 2 * MAX_LINE_SIZE * 6); + + call_ref(ctx, src, srcStride, srcSliceY, + srcSliceH, dst0, dstStride); + call_new(ctx, src, srcStride, srcSliceY, + srcSliceH, dst1, dstStride); + + if (dst_pix_fmt == AV_PIX_FMT_ARGB || + dst_pix_fmt == AV_PIX_FMT_ABGR || + dst_pix_fmt == AV_PIX_FMT_RGBA || + dst_pix_fmt == AV_PIX_FMT_BGRA || + dst_pix_fmt == AV_PIX_FMT_RGB24 || + dst_pix_fmt == AV_PIX_FMT_BGR24) { + if (cmp_off_by_n(lines0[0], lines1[0], width * sample_size, 3) || + cmp_off_by_n(lines0[1], lines1[1], width * sample_size, 3)) + fail(); + } else if (dst_pix_fmt == AV_PIX_FMT_RGB565 || + dst_pix_fmt == AV_PIX_FMT_BGR565) { + if (cmp_565_by_n(lines0[0], lines1[0], width, 2) || + cmp_565_by_n(lines0[1], lines1[1], width, 2)) + fail(); + } else if (dst_pix_fmt == AV_PIX_FMT_RGB555 || + dst_pix_fmt == AV_PIX_FMT_BGR555) { + if (cmp_555_by_n(lines0[0], lines1[0], width, 2) || + cmp_555_by_n(lines0[1], lines1[1], width, 2)) + fail(); + } else { + fail(); + } + + bench_new(ctx, src, srcStride, srcSliceY, + srcSliceH, dst0, dstStride); + } + sws_freeContext(ctx); + } + } +} + +#undef MAX_LINE_SIZE + +void checkasm_check_sw_yuv2rgb(void) +{ + check_yuv2rgb(AV_PIX_FMT_YUV420P); + report("yuv420p"); + check_yuv2rgb(AV_PIX_FMT_YUV422P); + report("yuv422p"); + check_yuv2rgb(AV_PIX_FMT_YUVA420P); + report("yuva420p"); +} diff --git a/tests/fate/checkasm.mak b/tests/fate/checkasm.mak index 0ed2ea5be6..49832b09bf 100644 --- a/tests/fate/checkasm.mak +++ b/tests/fate/checkasm.mak @@ -44,6 +44,7 @@ FATE_CHECKASM = fate-checkasm-aacencdsp \ fate-checkasm-sw_range_convert \ fate-checkasm-sw_rgb \ fate-checkasm-sw_scale \ + fate-checkasm-sw_yuv2rgb \ fate-checkasm-takdsp \ fate-checkasm-utvideodsp \ fate-checkasm-v210dec \ -- 2.30.2 [-- Attachment #3: Type: text/plain, Size: 251 bytes --] _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [FFmpeg-devel] [PATCH 5/6] checkasm: add tests for yuv2rgb 2024-06-25 21:19 ` Ramiro Polla @ 2024-06-27 14:02 ` Ramiro Polla 2024-06-30 0:33 ` Sean McGovern 0 siblings, 1 reply; 14+ messages in thread From: Ramiro Polla @ 2024-06-27 14:02 UTC (permalink / raw) To: ffmpeg-devel On Tue, Jun 25, 2024 at 11:19 PM Ramiro Polla <ramiro.polla@gmail.com> wrote: > On Thu, Jun 20, 2024 at 4:59 PM Ramiro Polla <ramiro.polla@gmail.com> wrote: > > checkasm gets a little spammy with this commit by printing "No > > accelerated colorspace conversion found from %s to %s.\n" from > > libswscale/yuv2rgb.c multiple times. Any suggestions? Perhaps > > downgrading that message to AV_LOG_VERBOSE, or raising loglevel around > > the call to sws_getContext()? > > New patch attached that overrides the log level to prevent spamming. I'll apply this tomorrow if there are no comments. _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [FFmpeg-devel] [PATCH 5/6] checkasm: add tests for yuv2rgb 2024-06-27 14:02 ` Ramiro Polla @ 2024-06-30 0:33 ` Sean McGovern 0 siblings, 0 replies; 14+ messages in thread From: Sean McGovern @ 2024-06-30 0:33 UTC (permalink / raw) To: FFmpeg development discussions and patches Hi Ramiro, On Thu, Jun 27, 2024, 10:37 Ramiro Polla <ramiro.polla@gmail.com> wrote: > On Tue, Jun 25, 2024 at 11:19 PM Ramiro Polla <ramiro.polla@gmail.com> > wrote: > > On Thu, Jun 20, 2024 at 4:59 PM Ramiro Polla <ramiro.polla@gmail.com> > wrote: > > > checkasm gets a little spammy with this commit by printing "No > > > accelerated colorspace conversion found from %s to %s.\n" from > > > libswscale/yuv2rgb.c multiple times. Any suggestions? Perhaps > > > downgrading that message to AV_LOG_VERBOSE, or raising loglevel around > > > the call to sws_getContext()? > > > > New patch attached that overrides the log level to prevent spamming. > > I'll apply this tomorrow if there are no comments. > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email > ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". > It looks like this exposed an issue in the AltiVec acceleration in libswscale. libswscale/ppc/yuv2rgb.c looks a bit too complex for me to investigate. I'd be curious to see if this fails on a 64-bit POWER machine. I'll try to test it later. -- Sean McGovern > _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 14+ messages in thread
* [FFmpeg-devel] [PATCH 6/6] swscale/yuv2rgb/x86: remove mmx/mmxext yuv2rgb functions 2024-06-16 22:28 [FFmpeg-devel] [PATCH 1/6] swscale/yuv2rgb: fix conversion for widths not aligned to 8 Ramiro Polla ` (3 preceding siblings ...) 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 5/6] checkasm: add tests for yuv2rgb Ramiro Polla @ 2024-06-16 22:28 ` Ramiro Polla 2024-06-16 23:15 ` James Almer 4 siblings, 1 reply; 14+ messages in thread From: Ramiro Polla @ 2024-06-16 22:28 UTC (permalink / raw) To: ffmpeg-devel These functions are either slower or barely faster than the C LUT yuv2rgb code. --- libswscale/x86/yuv2rgb.c | 51 ----------------- libswscale/x86/yuv2rgb_template.c | 4 -- libswscale/x86/yuv_2_rgb.asm | 93 +------------------------------ 3 files changed, 3 insertions(+), 145 deletions(-) diff --git a/libswscale/x86/yuv2rgb.c b/libswscale/x86/yuv2rgb.c index 6754062245..41dfa80f33 100644 --- a/libswscale/x86/yuv2rgb.c +++ b/libswscale/x86/yuv2rgb.c @@ -41,25 +41,8 @@ #define DITHER1XBPP // only for MMX -//MMX versions -#if HAVE_MMX -#undef RENAME -#define COMPILE_TEMPLATE_MMX -#define RENAME(a) a ## _mmx -#include "yuv2rgb_template.c" -#undef COMPILE_TEMPLATE_MMX -#endif /* HAVE_MMX */ - -// MMXEXT versions -#undef RENAME -#define COMPILE_TEMPLATE_MMXEXT -#define RENAME(a) a ## _mmxext -#include "yuv2rgb_template.c" -#undef COMPILE_TEMPLATE_MMXEXT - //SSSE3 versions #undef RENAME -#define COMPILE_TEMPLATE_SSSE3 #define RENAME(a) a ## _ssse3 #include "yuv2rgb_template.c" @@ -99,40 +82,6 @@ av_cold SwsFunc ff_yuv2rgb_init_x86(SwsContext *c) } } - if (EXTERNAL_MMXEXT(cpu_flags)) { - switch (c->dstFormat) { - case AV_PIX_FMT_RGB24: - return yuv420_rgb24_mmxext; - case AV_PIX_FMT_BGR24: - return yuv420_bgr24_mmxext; - } - } - - if (EXTERNAL_MMX(cpu_flags)) { - switch (c->dstFormat) { - case AV_PIX_FMT_RGB32: - if (c->srcFormat == AV_PIX_FMT_YUVA420P) { -#if CONFIG_SWSCALE_ALPHA - return yuva420_rgb32_mmx; -#endif - break; - } else - return yuv420_rgb32_mmx; - case AV_PIX_FMT_BGR32: - if (c->srcFormat == AV_PIX_FMT_YUVA420P) { -#if CONFIG_SWSCALE_ALPHA - return yuva420_bgr32_mmx; -#endif - break; - } else - return yuv420_bgr32_mmx; - case AV_PIX_FMT_RGB565: - return yuv420_rgb16_mmx; - case AV_PIX_FMT_RGB555: - return yuv420_rgb15_mmx; - } - } - #endif /* HAVE_X86ASM */ return NULL; } diff --git a/libswscale/x86/yuv2rgb_template.c b/libswscale/x86/yuv2rgb_template.c index 596943bb73..a4741e6873 100644 --- a/libswscale/x86/yuv2rgb_template.c +++ b/libswscale/x86/yuv2rgb_template.c @@ -47,7 +47,6 @@ extern void RENAME(ff_yuv_420_bgr24)(x86_reg index, uint8_t *image, const uint8_ const uint8_t *pv_index, const uint64_t *pointer_c_dither, const uint8_t *py_2index); -#ifndef COMPILE_TEMPLATE_MMXEXT extern void RENAME(ff_yuv_420_rgb15)(x86_reg index, uint8_t *image, const uint8_t *pu_index, const uint8_t *pv_index, const uint64_t *pointer_c_dither, const uint8_t *py_2index); @@ -163,9 +162,7 @@ static inline int RENAME(yuva420_bgr32)(SwsContext *c, const uint8_t *src[], } return srcSliceH; } -#endif -#if !defined(COMPILE_TEMPLATE_MMX) static inline int RENAME(yuv420_rgb24)(SwsContext *c, const uint8_t *src[], int srcStride[], int srcSliceY, int srcSliceH, @@ -193,4 +190,3 @@ static inline int RENAME(yuv420_bgr24)(SwsContext *c, const uint8_t *src[], } return srcSliceH; } -#endif diff --git a/libswscale/x86/yuv_2_rgb.asm b/libswscale/x86/yuv_2_rgb.asm index a1f9134e08..b67ab162d2 100644 --- a/libswscale/x86/yuv_2_rgb.asm +++ b/libswscale/x86/yuv_2_rgb.asm @@ -38,12 +38,6 @@ pb_e0: times 16 db 224 pb_03: times 16 db 3 pb_07: times 16 db 7 -mask_1101: dw -1, -1, 0, -1 -mask_0010: dw 0, 0, -1, 0 -mask_0110: dw 0, -1, -1, 0 -mask_1001: dw -1, 0, 0, -1 -mask_0100: dw 0, -1, 0, 0 - SECTION .text ;----------------------------------------------------------------------------- @@ -55,14 +49,6 @@ SECTION .text ; ;----------------------------------------------------------------------------- -%macro MOV_H2L 1 -%if mmsize == 8 - psrlq %1, 32 -%else ; mmsize == 16 - psrldq %1, 8 -%endif -%endmacro - %macro yuv2rgb_fn 3 %if %3 == 32 @@ -91,18 +77,6 @@ SECTION .text %define m_blue m1 %endif -%if mmsize == 8 -%define time_num 1 -%define reg_num 8 -%define y_offset [pointer_c_ditherq + 8 * 8] -%define u_offset [pointer_c_ditherq + 9 * 8] -%define v_offset [pointer_c_ditherq + 10 * 8] -%define ug_coff [pointer_c_ditherq + 7 * 8] -%define vg_coff [pointer_c_ditherq + 6 * 8] -%define y_coff [pointer_c_ditherq + 3 * 8] -%define ub_coff [pointer_c_ditherq + 5 * 8] -%define vr_coff [pointer_c_ditherq + 4 * 8] -%elif mmsize == 16 %define time_num 2 %if ARCH_X86_32 %define reg_num 8 @@ -125,13 +99,11 @@ SECTION .text %define ub_coff m14 %define vr_coff m15 %endif ; ARCH_X86_32/64 -%endif ; coeff define mmsize == 8/16 cglobal %1_420_%2%3, GPR_num, GPR_num, reg_num, parameters %if ARCH_X86_64 movsxd indexq, indexd -%if mmsize == 16 VBROADCASTSD y_offset, [pointer_c_ditherq + 8 * 8] VBROADCASTSD u_offset, [pointer_c_ditherq + 9 * 8] VBROADCASTSD v_offset, [pointer_c_ditherq + 10 * 8] @@ -141,7 +113,6 @@ cglobal %1_420_%2%3, GPR_num, GPR_num, reg_num, parameters VBROADCASTSD ub_coff, [pointer_c_ditherq + 5 * 8] VBROADCASTSD vr_coff, [pointer_c_ditherq + 4 * 8] %endif -%endif .loop0: movu m_y, [py_2indexq + 2 * indexq] movh m_u, [pu_indexq + indexq] @@ -157,7 +128,7 @@ cglobal %1_420_%2%3, GPR_num, GPR_num, reg_num, parameters psllw m1, 3 psllw m6, 3 psllw m7, 3 -%if (ARCH_X86_32 && mmsize == 16) +%if ARCH_X86_32 VBROADCASTSD m2, mu_offset VBROADCASTSD m3, mv_offset VBROADCASTSD m4, my_offset @@ -176,7 +147,7 @@ cglobal %1_420_%2%3, GPR_num, GPR_num, reg_num, parameters pmulhw m0, m5 VBROADCASTSD m4, mvr_coff pmulhw m1, m4 -%else ; ARCH_X86_64 || mmsize == 8 +%else ; ARCH_X86_64 psubsw m0, u_offset ; U = U - 128 psubsw m1, v_offset ; V = V - 128 psubw m6, y_offset @@ -207,49 +178,10 @@ cglobal %1_420_%2%3, GPR_num, GPR_num, reg_num, parameters packuswb m2, m7 ; G0 G2 G4 G6 ... G1 G3 G5 G7 ... mova m3, m_red mova m6, m_blue - MOV_H2L m_red + psrldq m_red, 8 punpcklbw m3, m2 ; R0 G0 R2 G2 R4 G4 R6 G6 R8 G8 ... punpcklbw m6, m_red ; B0 R1 B2 R3 B4 R5 B6 R7 B8 R9 ... - mova m5, m3 punpckhbw m2, m_blue ; G1 B1 G3 B3 G5 B5 G7 B7 G9 B9 ... -%if mmsize == 8 - punpcklwd m3 ,m6 ; R0 G0 B0 R1 R2 G2 B2 R3 - punpckhwd m5, m6 ; R4 G4 B4 R5 R6 G6 B6 R7 -%if cpuflag(mmxext) - pshufw m1, m2, 0xc6 - pshufw m6, m3, 0x84 - pshufw m7, m5, 0x38 - pand m6, [mask_1101] ; R0 G0 B0 R1 -- -- R2 G2 - movq m0, m1 - pand m7, [mask_0110] ; -- -- R6 G6 B6 R7 -- -- - movq m2, m1 - pand m1, [mask_0100] ; -- -- G3 B3 -- -- -- -- - psrlq m3, 48 ; B2 R3 -- -- -- -- -- -- - pand m0, [mask_0010] ; -- -- -- -- G1 B1 -- -- - psllq m5, 32 ; -- -- -- -- R4 G4 B4 R5 - pand m2, [mask_1001] ; G5 B5 -- -- -- -- G7 B7 - por m1, m3 - por m0, m6 - por m1, m5 - por m2, m7 - movntq [imageq], m0 - movntq [imageq + 8], m1 - movntq [imageq + 16], m2 -%else ; cpuflag(mmx) - movd [imageq], m3 ; R0 G0 R2 G2 - movd [imageq + 4], m2 ; G1 B1 - psrlq m3, 32 - psrlq m2, 16 - movd [imageq + 6], m3 ; R2 G2 B2 R3 - movd [imageq + 10], m2 ; G3 B3 - psrlq m2, 16 - movd [imageq + 12], m5 ; R4 G4 B4 R5 - movd [imageq + 16], m2 ; G5 B5 - psrlq m5, 32 - movd [imageq + 20], m2 ; -- -- G7 B7 - movd [imageq + 18], m5 ; R6 G6 B6 R7 -%endif ; cpuflag -%else ; mmsize == 16 pshufb m3, [rgb24_shuf1] ; r0 g0 r6 g6 r12 g12 r2 g2 r8 g8 r14 g14 r4 g4 r10 g10 pshufb m6, [rgb24_shuf2] ; b10 r11 b0 r1 b6 r7 b12 r13 b2 r3 b8 r9 b14 r15 b4 r5 pshufb m2, [rgb24_shuf3] ; g5 b5 g11 b11 g1 b1 g7 b7 g13 b13 g3 b3 g9 b9 g15 b15 @@ -274,7 +206,6 @@ cglobal %1_420_%2%3, GPR_num, GPR_num, reg_num, parameters movu [imageq], m0 movu [imageq + 16], m1 movu [imageq + 32], m2 -%endif ; mmsize = 16 %else ; PACK RGB15/16/32 packuswb m0, m1 packuswb m3, m5 @@ -309,18 +240,12 @@ cglobal %1_420_%2%3, GPR_num, GPR_num, reg_num, parameters movu [imageq + 24 * time_num], m_alpha %else ; PACK RGB15/16 %define depth 2 -%if cpuflag(ssse3) %define red_dither m3 %define green_dither m4 %define blue_dither m5 VBROADCASTSD red_dither, [pointer_c_ditherq + 0 * 8] VBROADCASTSD green_dither, [pointer_c_ditherq + 1 * 8] VBROADCASTSD blue_dither, [pointer_c_ditherq + 2 * 8] -%else ; cpuflag(mmx/mmxext) -%define blue_dither [pointer_c_ditherq + 2 * 8] -%define green_dither [pointer_c_ditherq + 1 * 8] -%define red_dither [pointer_c_ditherq + 0 * 8] -%endif %if %3 == 15 %define gmask pb_03 %define isRGB15 1 @@ -358,18 +283,6 @@ RET %endmacro -INIT_MMX mmx -yuv2rgb_fn yuv, rgb, 32 -yuv2rgb_fn yuv, bgr, 32 -yuv2rgb_fn yuva, rgb, 32 -yuv2rgb_fn yuva, bgr, 32 -yuv2rgb_fn yuv, rgb, 15 -yuv2rgb_fn yuv, rgb, 16 - -INIT_MMX mmxext -yuv2rgb_fn yuv, rgb, 24 -yuv2rgb_fn yuv, bgr, 24 - INIT_XMM ssse3 yuv2rgb_fn yuv, rgb, 24 yuv2rgb_fn yuv, bgr, 24 -- 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-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [FFmpeg-devel] [PATCH 6/6] swscale/yuv2rgb/x86: remove mmx/mmxext yuv2rgb functions 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 6/6] swscale/yuv2rgb/x86: remove mmx/mmxext yuv2rgb functions Ramiro Polla @ 2024-06-16 23:15 ` James Almer 2024-06-17 12:46 ` Ramiro Polla 0 siblings, 1 reply; 14+ messages in thread From: James Almer @ 2024-06-16 23:15 UTC (permalink / raw) To: ffmpeg-devel On 6/16/2024 7:28 PM, Ramiro Polla wrote: > These functions are either slower or barely faster than the C LUT > yuv2rgb code. > --- > libswscale/x86/yuv2rgb.c | 51 ----------------- > libswscale/x86/yuv2rgb_template.c | 4 -- > libswscale/x86/yuv_2_rgb.asm | 93 +------------------------------ > 3 files changed, 3 insertions(+), 145 deletions(-) > > diff --git a/libswscale/x86/yuv2rgb.c b/libswscale/x86/yuv2rgb.c > index 6754062245..41dfa80f33 100644 > --- a/libswscale/x86/yuv2rgb.c > +++ b/libswscale/x86/yuv2rgb.c > @@ -41,25 +41,8 @@ > > #define DITHER1XBPP // only for MMX Shouldn't this be removed too? > > -//MMX versions > -#if HAVE_MMX > -#undef RENAME > -#define COMPILE_TEMPLATE_MMX > -#define RENAME(a) a ## _mmx > -#include "yuv2rgb_template.c" > -#undef COMPILE_TEMPLATE_MMX > -#endif /* HAVE_MMX */ > - > -// MMXEXT versions > -#undef RENAME > -#define COMPILE_TEMPLATE_MMXEXT > -#define RENAME(a) a ## _mmxext > -#include "yuv2rgb_template.c" > -#undef COMPILE_TEMPLATE_MMXEXT > - > //SSSE3 versions > #undef RENAME > -#define COMPILE_TEMPLATE_SSSE3 > #define RENAME(a) a ## _ssse3 > #include "yuv2rgb_template.c" You could write a seventh patch that moves the template stuff back to this file, now that SSSE3 is the only version. See commit 8b62fb231a78. _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [FFmpeg-devel] [PATCH 6/6] swscale/yuv2rgb/x86: remove mmx/mmxext yuv2rgb functions 2024-06-16 23:15 ` James Almer @ 2024-06-17 12:46 ` Ramiro Polla 2024-06-20 14:50 ` Ramiro Polla 0 siblings, 1 reply; 14+ messages in thread From: Ramiro Polla @ 2024-06-17 12:46 UTC (permalink / raw) To: FFmpeg development discussions and patches On Mon, Jun 17, 2024 at 1:16 AM James Almer <jamrial@gmail.com> wrote: > On 6/16/2024 7:28 PM, Ramiro Polla wrote: > > These functions are either slower or barely faster than the C LUT > > yuv2rgb code. > > --- > > libswscale/x86/yuv2rgb.c | 51 ----------------- > > libswscale/x86/yuv2rgb_template.c | 4 -- > > libswscale/x86/yuv_2_rgb.asm | 93 +------------------------------ > > 3 files changed, 3 insertions(+), 145 deletions(-) > > > > diff --git a/libswscale/x86/yuv2rgb.c b/libswscale/x86/yuv2rgb.c > > index 6754062245..41dfa80f33 100644 > > --- a/libswscale/x86/yuv2rgb.c > > +++ b/libswscale/x86/yuv2rgb.c > > @@ -41,25 +41,8 @@ > > > > #define DITHER1XBPP // only for MMX > > Shouldn't this be removed too? I think this #define can already be removed from everywhere. It seems to be unconditionally set in swscale_internal.h (I haven't tracked down since when this is the case). > > -//MMX versions > > -#if HAVE_MMX > > -#undef RENAME > > -#define COMPILE_TEMPLATE_MMX > > -#define RENAME(a) a ## _mmx > > -#include "yuv2rgb_template.c" > > -#undef COMPILE_TEMPLATE_MMX > > -#endif /* HAVE_MMX */ > > - > > -// MMXEXT versions > > -#undef RENAME > > -#define COMPILE_TEMPLATE_MMXEXT > > -#define RENAME(a) a ## _mmxext > > -#include "yuv2rgb_template.c" > > -#undef COMPILE_TEMPLATE_MMXEXT > > - > > //SSSE3 versions > > #undef RENAME > > -#define COMPILE_TEMPLATE_SSSE3 > > #define RENAME(a) a ## _ssse3 > > #include "yuv2rgb_template.c" > > You could write a seventh patch that moves the template stuff back to > this file, now that SSSE3 is the only version. See commit 8b62fb231a78. Will do in the next version of this patchset. _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [FFmpeg-devel] [PATCH 6/6] swscale/yuv2rgb/x86: remove mmx/mmxext yuv2rgb functions 2024-06-17 12:46 ` Ramiro Polla @ 2024-06-20 14:50 ` Ramiro Polla 0 siblings, 0 replies; 14+ messages in thread From: Ramiro Polla @ 2024-06-20 14:50 UTC (permalink / raw) To: FFmpeg development discussions and patches On Mon, Jun 17, 2024 at 2:46 PM Ramiro Polla <ramiro.polla@gmail.com> wrote: > On Mon, Jun 17, 2024 at 1:16 AM James Almer <jamrial@gmail.com> wrote: > > On 6/16/2024 7:28 PM, Ramiro Polla wrote: > > > These functions are either slower or barely faster than the C LUT > > > yuv2rgb code. > > > --- > > > libswscale/x86/yuv2rgb.c | 51 ----------------- > > > libswscale/x86/yuv2rgb_template.c | 4 -- > > > libswscale/x86/yuv_2_rgb.asm | 93 +------------------------------ > > > 3 files changed, 3 insertions(+), 145 deletions(-) > > > > > > diff --git a/libswscale/x86/yuv2rgb.c b/libswscale/x86/yuv2rgb.c > > > index 6754062245..41dfa80f33 100644 > > > --- a/libswscale/x86/yuv2rgb.c > > > +++ b/libswscale/x86/yuv2rgb.c > > > @@ -41,25 +41,8 @@ > > > > > > #define DITHER1XBPP // only for MMX > > > > Shouldn't this be removed too? > > I think this #define can already be removed from everywhere. It seems > to be unconditionally set in swscale_internal.h (I haven't tracked > down since when this is the case). > > > > -//MMX versions > > > -#if HAVE_MMX > > > -#undef RENAME > > > -#define COMPILE_TEMPLATE_MMX > > > -#define RENAME(a) a ## _mmx > > > -#include "yuv2rgb_template.c" > > > -#undef COMPILE_TEMPLATE_MMX > > > -#endif /* HAVE_MMX */ > > > - > > > -// MMXEXT versions > > > -#undef RENAME > > > -#define COMPILE_TEMPLATE_MMXEXT > > > -#define RENAME(a) a ## _mmxext > > > -#include "yuv2rgb_template.c" > > > -#undef COMPILE_TEMPLATE_MMXEXT > > > - > > > //SSSE3 versions > > > #undef RENAME > > > -#define COMPILE_TEMPLATE_SSSE3 > > > #define RENAME(a) a ## _ssse3 > > > #include "yuv2rgb_template.c" > > > > You could write a seventh patch that moves the template stuff back to > > this file, now that SSSE3 is the only version. See commit 8b62fb231a78. > > Will do in the next version of this patchset. I'll apply this patchset if there are no more comments, before submitting more patches to deal with DITHER1XBPP and detemplatizing. _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe". ^ permalink raw reply [flat|nested] 14+ messages in thread
end of thread, other threads:[~2024-06-30 0:33 UTC | newest] Thread overview: 14+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2024-06-16 22:28 [FFmpeg-devel] [PATCH 1/6] swscale/yuv2rgb: fix conversion for widths not aligned to 8 Ramiro Polla 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 2/6] swscale/yuv2rgb: add macros to simplify code generation Ramiro Polla 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 3/6] swscale/yuv2rgb: fix yuv422p input in C code Ramiro Polla 2024-06-19 10:13 ` Michael Niedermayer 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 4/6] swscale/yuv2rgb: reindent after previous commit Ramiro Polla 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 5/6] checkasm: add tests for yuv2rgb Ramiro Polla 2024-06-20 14:59 ` Ramiro Polla 2024-06-25 21:19 ` Ramiro Polla 2024-06-27 14:02 ` Ramiro Polla 2024-06-30 0:33 ` Sean McGovern 2024-06-16 22:28 ` [FFmpeg-devel] [PATCH 6/6] swscale/yuv2rgb/x86: remove mmx/mmxext yuv2rgb functions Ramiro Polla 2024-06-16 23:15 ` James Almer 2024-06-17 12:46 ` Ramiro Polla 2024-06-20 14:50 ` Ramiro Polla
Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel This inbox may be cloned and mirrored by anyone: git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git # If you have public-inbox 1.1+ installed, you may # initialize and index your mirror using the following commands: public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \ ffmpegdev@gitmailbox.com public-inbox-index ffmpegdev Example config snippet for mirrors. AGPL code for this site: git clone https://public-inbox.org/public-inbox.git