* [FFmpeg-devel] [PATCH 2/4] tests/checkasm: add scene_sad checkasm test
2025-07-12 9:22 [FFmpeg-devel] [PATCH 1/4] avfilter/scene_sad: pass true depth to ff_scene_sad_get_fn() Niklas Haas
@ 2025-07-12 9:22 ` Niklas Haas
2025-07-12 9:22 ` [FFmpeg-devel] [PATCH 3/4] avfilter/x86/scene_sad: add AVX512 implementation Niklas Haas
2025-07-12 9:22 ` [FFmpeg-devel] [PATCH 4/4] avfilter/x86/scene_sad: add high bit depth AVX2/AVX512 version Niklas Haas
2 siblings, 0 replies; 4+ messages in thread
From: Niklas Haas @ 2025-07-12 9:22 UTC (permalink / raw)
To: ffmpeg-devel; +Cc: Niklas Haas
From: Niklas Haas <git@haasn.dev>
---
tests/checkasm/Makefile | 1 +
tests/checkasm/checkasm.c | 3 ++
tests/checkasm/checkasm.h | 1 +
tests/checkasm/scene_sad.c | 73 ++++++++++++++++++++++++++++++++++++++
4 files changed, 78 insertions(+)
create mode 100644 tests/checkasm/scene_sad.c
diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
index afd62d95ba..1938468bbd 100644
--- a/tests/checkasm/Makefile
+++ b/tests/checkasm/Makefile
@@ -52,6 +52,7 @@ AVCODECOBJS-$(CONFIG_VVC_DECODER) += vvc_alf.o vvc_mc.o vvc_sao.o
CHECKASMOBJS-$(CONFIG_AVCODEC) += $(AVCODECOBJS-yes)
# libavfilter tests
+AVFILTEROBJS-$(CONFIG_SCENE_SAD) += scene_sad.o
AVFILTEROBJS-$(CONFIG_AFIR_FILTER) += af_afir.o
AVFILTEROBJS-$(CONFIG_BLEND_FILTER) += vf_blend.o
AVFILTEROBJS-$(CONFIG_BWDIF_FILTER) += vf_bwdif.o
diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
index f4e3d4f433..66a8f8ff86 100644
--- a/tests/checkasm/checkasm.c
+++ b/tests/checkasm/checkasm.c
@@ -260,6 +260,9 @@ static const struct {
#endif
#endif
#if CONFIG_AVFILTER
+ #if CONFIG_SCENE_SAD
+ { "scene_sad", checkasm_check_scene_sad },
+ #endif
#if CONFIG_AFIR_FILTER
{ "af_afir", checkasm_check_afir },
#endif
diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h
index e829942d58..825e7ef52f 100644
--- a/tests/checkasm/checkasm.h
+++ b/tests/checkasm/checkasm.h
@@ -124,6 +124,7 @@ void checkasm_check_pixblockdsp(void);
void checkasm_check_sbrdsp(void);
void checkasm_check_rv34dsp(void);
void checkasm_check_rv40dsp(void);
+void checkasm_check_scene_sad(void);
void checkasm_check_svq1enc(void);
void checkasm_check_synth_filter(void);
void checkasm_check_sw_gbrp(void);
diff --git a/tests/checkasm/scene_sad.c b/tests/checkasm/scene_sad.c
new file mode 100644
index 0000000000..bc33c428c6
--- /dev/null
+++ b/tests/checkasm/scene_sad.c
@@ -0,0 +1,73 @@
+/*
+ * 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 "checkasm.h"
+#include "libavfilter/scene_sad.h"
+#include "libavutil/mem_internal.h"
+
+#define WIDTH 256
+#define HEIGHT 256
+#define STRIDE WIDTH
+
+#define randomize_buffers(type, buf, size, mask) \
+ do { \
+ int j; \
+ type *tmp_buf = (type *) buf; \
+ for (j = 0; j < size; j++) \
+ tmp_buf[j] = rnd() & mask; \
+ } while (0)
+
+static void check_scene_sad(int depth)
+{
+ LOCAL_ALIGNED_32(uint8_t, src1, [WIDTH * HEIGHT * 2]);
+ LOCAL_ALIGNED_32(uint8_t, src2, [WIDTH * HEIGHT * 2]);
+ declare_func(void, const uint8_t *src1, ptrdiff_t stride1,
+ const uint8_t *src2, ptrdiff_t stride2,
+ ptrdiff_t width, ptrdiff_t height, uint64_t *sum);
+
+ const int width = WIDTH >> (depth > 8);
+ int mask = (1 << depth) - 1;
+ if (depth <= 8) {
+ randomize_buffers(uint8_t, src1, width * HEIGHT, mask);
+ randomize_buffers(uint8_t, src2, width * HEIGHT, mask);
+ } else if (depth <= 16) {
+ randomize_buffers(uint16_t, src1, width * HEIGHT, mask);
+ randomize_buffers(uint16_t, src2, width * HEIGHT, mask);
+ }
+
+ if (check_func(ff_scene_sad_get_fn(depth), "scene_sad%d", depth)) {
+ uint64_t sum1, sum2;
+ call_ref(src1, STRIDE, src2, STRIDE, width, HEIGHT, &sum1);
+ call_new(src1, STRIDE, src2, STRIDE, width, HEIGHT, &sum2);
+ if (sum1 != sum2) {
+ fprintf(stderr, "scene_sad%d: sum mismatch: %llu != %llu\n",
+ depth, (unsigned long long) sum1, (unsigned long long) sum2);
+ fail();
+ }
+ bench_new(src1, STRIDE, src2, STRIDE, WIDTH, HEIGHT, &sum2);
+ }
+}
+
+void checkasm_check_scene_sad(void)
+{
+ const int depths[] = { 8, 10, 12, 14, 15, 16 };
+ for (int i = 0; i < FF_ARRAY_ELEMS(depths); i++) {
+ check_scene_sad(depths[i]);
+ report("scene_sad%d", depths[i]);
+ }
+}
--
2.49.0
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 4+ messages in thread
* [FFmpeg-devel] [PATCH 3/4] avfilter/x86/scene_sad: add AVX512 implementation
2025-07-12 9:22 [FFmpeg-devel] [PATCH 1/4] avfilter/scene_sad: pass true depth to ff_scene_sad_get_fn() Niklas Haas
2025-07-12 9:22 ` [FFmpeg-devel] [PATCH 2/4] tests/checkasm: add scene_sad checkasm test Niklas Haas
@ 2025-07-12 9:22 ` Niklas Haas
2025-07-12 9:22 ` [FFmpeg-devel] [PATCH 4/4] avfilter/x86/scene_sad: add high bit depth AVX2/AVX512 version Niklas Haas
2 siblings, 0 replies; 4+ messages in thread
From: Niklas Haas @ 2025-07-12 9:22 UTC (permalink / raw)
To: ffmpeg-devel; +Cc: Niklas Haas
From: Niklas Haas <git@haasn.dev>
Trivial to add, but a lot faster (on my machine).
scene_sad8_c: 114476.4 ( 1.00x)
scene_sad8_sse2: 8644.3 (13.24x)
scene_sad8_avx2: 4520.1 (25.33x)
scene_sad8_avx512: 3153.0 (36.31x)
---
libavfilter/x86/scene_sad.asm | 7 +++++++
libavfilter/x86/scene_sad_init.c | 7 +++++++
2 files changed, 14 insertions(+)
diff --git a/libavfilter/x86/scene_sad.asm b/libavfilter/x86/scene_sad.asm
index bf7236b3a3..2cd9dddb5c 100644
--- a/libavfilter/x86/scene_sad.asm
+++ b/libavfilter/x86/scene_sad.asm
@@ -72,3 +72,10 @@ INIT_YMM avx2
SAD_FRAMES
%endif
+
+%if HAVE_AVX512_EXTERNAL
+
+INIT_ZMM avx512
+SAD_FRAMES
+
+%endif
diff --git a/libavfilter/x86/scene_sad_init.c b/libavfilter/x86/scene_sad_init.c
index 4a4c40195f..2d631b376a 100644
--- a/libavfilter/x86/scene_sad_init.c
+++ b/libavfilter/x86/scene_sad_init.c
@@ -41,6 +41,9 @@ SCENE_SAD_FUNC(scene_sad_sse2, ff_scene_sad_sse2, 16)
#if HAVE_AVX2_EXTERNAL
SCENE_SAD_FUNC(scene_sad_avx2, ff_scene_sad_avx2, 32)
#endif
+#if HAVE_AVX512_EXTERNAL
+SCENE_SAD_FUNC(scene_sad_avx512, ff_scene_sad_avx512, 64)
+#endif
#endif
ff_scene_sad_fn ff_scene_sad_get_fn_x86(int depth)
@@ -48,6 +51,10 @@ ff_scene_sad_fn ff_scene_sad_get_fn_x86(int depth)
#if HAVE_X86ASM
int cpu_flags = av_get_cpu_flags();
if (depth <= 8) {
+#if HAVE_AVX512_EXTERNAL
+ if (EXTERNAL_AVX512(cpu_flags))
+ return scene_sad_avx512;
+#endif
#if HAVE_AVX2_EXTERNAL
if (EXTERNAL_AVX2_FAST(cpu_flags))
return scene_sad_avx2;
--
2.49.0
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 4+ messages in thread
* [FFmpeg-devel] [PATCH 4/4] avfilter/x86/scene_sad: add high bit depth AVX2/AVX512 version
2025-07-12 9:22 [FFmpeg-devel] [PATCH 1/4] avfilter/scene_sad: pass true depth to ff_scene_sad_get_fn() Niklas Haas
2025-07-12 9:22 ` [FFmpeg-devel] [PATCH 2/4] tests/checkasm: add scene_sad checkasm test Niklas Haas
2025-07-12 9:22 ` [FFmpeg-devel] [PATCH 3/4] avfilter/x86/scene_sad: add AVX512 implementation Niklas Haas
@ 2025-07-12 9:22 ` Niklas Haas
2 siblings, 0 replies; 4+ messages in thread
From: Niklas Haas @ 2025-07-12 9:22 UTC (permalink / raw)
To: ffmpeg-devel; +Cc: Niklas Haas
From: Niklas Haas <git@haasn.dev>
Since psadbw only exists for 8-bits, we have to emulate it for 16-bit
inputs. The simplest sequence is to use a normal subtraction, which is safe
as long as the inputs do not exceed 32767 - so limit this implementation
to 15-bit inputs and below.
For 16-bit inputs, we could in theory instead use a pminw / pmaxw to ensure
the resulting difference does not overflow, but this is slower, and also
breaks the subsequent use of pmaddwd, so I opted to skip 16-bit SIMD for
now.
scene_sad10_c: 114175.6 ( 1.00x)
scene_sad10_avx2: 9617.7 (11.87x)
scene_sad10_avx512: 5208.8 (21.92x)
scene_sad12_c: 114537.8 ( 1.00x)
scene_sad12_avx2: 9614.0 (11.91x)
scene_sad12_avx512: 5186.3 (22.08x)
scene_sad14_c: 114113.9 ( 1.00x)
scene_sad14_avx2: 9612.9 (11.87x)
scene_sad14_avx512: 5186.0 (22.00x)
scene_sad15_c: 114108.9 ( 1.00x)
scene_sad15_avx2: 9612.3 (11.87x)
scene_sad15_avx512: 5186.4 (22.00x)
scene_sad16_c: 114136.0 ( 1.00x)
---
libavfilter/x86/scene_sad.asm | 44 ++++++++++++++++++++++++--------
libavfilter/x86/scene_sad_init.c | 33 +++++++++++++++++++++---
2 files changed, 64 insertions(+), 13 deletions(-)
diff --git a/libavfilter/x86/scene_sad.asm b/libavfilter/x86/scene_sad.asm
index 2cd9dddb5c..181c290893 100644
--- a/libavfilter/x86/scene_sad.asm
+++ b/libavfilter/x86/scene_sad.asm
@@ -24,25 +24,47 @@
%include "libavutil/x86/x86util.asm"
+SECTION_RODATA
+
+pw_1: times 32 dw 1
+
SECTION .text
-%macro SAD_INIT 0
-cglobal scene_sad, 6, 7, 2, src1, stride1, src2, stride2, width, end, x
+%macro SAD_INIT 1 ; depth
+cglobal scene_sad%1, 6, 7, 3, src1, stride1, src2, stride2, width, end, x
add src1q, widthq
add src2q, widthq
neg widthq
pxor m1, m1
%endmacro
+%macro PSADQ 4 ; depth, dst, [src2], tmp
+%if %1 == 8
+ psadbw %2, %3
+%else
+ psubw %2, %3
+ pabsw %2, %2
+ pmaddwd %2, [pw_1]
+ %if mmsize == 64
+ vextracti32x8 ymm%4, %2, 1
+ paddd ymm%2, ymm%4
+ pmovzxdq %2, ymm%2
+ %else
+ vextracti128 xmm%4, %2, 1
+ paddd xmm%2, xmm%4
+ pmovzxdq %2, xmm%2
+ %endif
+%endif
+%endmacro
-%macro SAD_LOOP 0
+%macro SAD_LOOP 1 ; depth
.nextrow:
mov xq, widthq
.loop:
movu m0, [src1q + xq]
- psadbw m0, [src2q + xq]
+ PSADQ %1, m0, [src2q + xq], m2
paddq m1, m0
add xq, mmsize
jl .loop
@@ -57,25 +79,27 @@ RET
%endmacro
-%macro SAD_FRAMES 0
- SAD_INIT
- SAD_LOOP
+%macro SAD_FRAMES 1 ; depth
+ SAD_INIT %1
+ SAD_LOOP %1
%endmacro
INIT_XMM sse2
-SAD_FRAMES
+SAD_FRAMES 8
%if HAVE_AVX2_EXTERNAL
INIT_YMM avx2
-SAD_FRAMES
+SAD_FRAMES 8
+SAD_FRAMES 16
%endif
%if HAVE_AVX512_EXTERNAL
INIT_ZMM avx512
-SAD_FRAMES
+SAD_FRAMES 8
+SAD_FRAMES 16
%endif
diff --git a/libavfilter/x86/scene_sad_init.c b/libavfilter/x86/scene_sad_init.c
index 2d631b376a..9863839b4e 100644
--- a/libavfilter/x86/scene_sad_init.c
+++ b/libavfilter/x86/scene_sad_init.c
@@ -36,13 +36,31 @@ static void FUNC_NAME(SCENE_SAD_PARAMS) { \
*sum += sad[0]; \
}
+#define SCENE_SAD16_FUNC(FUNC_NAME, ASM_FUNC_NAME, MMSIZE) \
+void ASM_FUNC_NAME(SCENE_SAD_PARAMS); \
+ \
+static void FUNC_NAME(SCENE_SAD_PARAMS) { \
+ uint64_t sad[MMSIZE / 8] = {0}; \
+ ptrdiff_t bytes = (width << 1) & ~(MMSIZE - 1); \
+ *sum = 0; \
+ ASM_FUNC_NAME(src1, stride1, src2, stride2, bytes, height, sad); \
+ for (int i = 0; i < MMSIZE / 8; i++) \
+ *sum += sad[i]; \
+ ff_scene_sad16_c(src1 + bytes, stride1, \
+ src2 + bytes, stride2, \
+ width - (bytes >> 1), height, sad); \
+ *sum += sad[0]; \
+}
+
#if HAVE_X86ASM
-SCENE_SAD_FUNC(scene_sad_sse2, ff_scene_sad_sse2, 16)
+SCENE_SAD_FUNC(scene_sad_sse2, ff_scene_sad8_sse2, 16)
#if HAVE_AVX2_EXTERNAL
-SCENE_SAD_FUNC(scene_sad_avx2, ff_scene_sad_avx2, 32)
+SCENE_SAD_FUNC(scene_sad_avx2, ff_scene_sad8_avx2, 32)
+SCENE_SAD16_FUNC(scene_sad16_avx2, ff_scene_sad16_avx2, 32)
#endif
#if HAVE_AVX512_EXTERNAL
-SCENE_SAD_FUNC(scene_sad_avx512, ff_scene_sad_avx512, 64)
+SCENE_SAD_FUNC(scene_sad_avx512, ff_scene_sad8_avx512, 64)
+SCENE_SAD16_FUNC(scene_sad16_avx512, ff_scene_sad16_avx512, 64)
#endif
#endif
@@ -61,6 +79,15 @@ ff_scene_sad_fn ff_scene_sad_get_fn_x86(int depth)
#endif
if (EXTERNAL_SSE2(cpu_flags))
return scene_sad_sse2;
+ } else if (depth < 16) { /* this routine is only safe up to 15 bits */
+#if HAVE_AVX512_EXTERNAL
+ if (EXTERNAL_AVX512(cpu_flags))
+ return scene_sad16_avx512;
+#endif
+#if HAVE_AVX2_EXTERNAL
+ if (EXTERNAL_AVX2_FAST(cpu_flags))
+ return scene_sad16_avx2;
+#endif
}
#endif
return NULL;
--
2.49.0
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 4+ messages in thread