[FFmpeg-devel] [PATCH] checkasm/blockdsp: don't randomize the buffers for fill_block_tab

[FFmpeg-devel] [PATCH] checkasm/blockdsp: don't randomize the buffers for fill_block_tab

[James Almer]

It ignores and overwrites the previous values.
Fixes running the test under ubsan.

Signed-off-by: James Almer <jamrial@gmail.com>
---
 tests/checkasm/blockdsp.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/tests/checkasm/blockdsp.c b/tests/checkasm/blockdsp.c
index 19d69b8687..ab87fc8fa4 100644
--- a/tests/checkasm/blockdsp.c
+++ b/tests/checkasm/blockdsp.c
@@ -71,7 +71,8 @@ static void check_fill(BlockDSPContext *h){
                      ptrdiff_t line_size, int h);
         if (check_func(h->fill_block_tab[t], "blockdsp.%s", tests[t].name)) {
             uint8_t value = rnd();
-            randomize_buffers(tests[t].size);
+            memset(buf0, 0, sizeof(*buf0) * n * n);
+            memset(buf1, 0, sizeof(*buf0) * n * n);
             call_ref(buf0, value, n, n);
             call_new(buf1, value, n, n);
             if (memcmp(buf0, buf1, sizeof(*buf0) * n * n))
-- 
2.45.0

_______________________________________________
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".

Re: [FFmpeg-devel] [PATCH] checkasm/blockdsp: don't randomize the buffers for fill_block_tab

[Martin Storsjö]

On Mon, 6 May 2024, James Almer wrote:

> It ignores and overwrites the previous values.
> Fixes running the test under ubsan.
>
> Signed-off-by: James Almer <jamrial@gmail.com>
> ---
> tests/checkasm/blockdsp.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)

The change is probably correct, but what issue is ubsan complaining about? 
If this would just be a dead store of unused random values, that shouldn't 
be an ubsan issue in general, right?

// Martin

_______________________________________________
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".

Re: [FFmpeg-devel] [PATCH] checkasm/blockdsp: don't randomize the buffers for fill_block_tab

[Andreas Rheinhardt]

Martin Storsjö:
> On Mon, 6 May 2024, James Almer wrote:
> 
>> It ignores and overwrites the previous values.
>> Fixes running the test under ubsan.
>>
>> Signed-off-by: James Almer <jamrial@gmail.com>
>> ---
>> tests/checkasm/blockdsp.c | 3 ++-
>> 1 file changed, 2 insertions(+), 1 deletion(-)
> 
> The change is probably correct, but what issue is ubsan complaining
> about? If this would just be a dead store of unused random values, that
> shouldn't be an ubsan issue in general, right?
> 

UBSan complains about unaligned stores in randomize_buffers; which is
obvious given that i is incremented by 1, not by 2. I sent a patch that
fixes this without removing randomization:
https://ffmpeg.org/pipermail/ffmpeg-devel/2024-May/326945.html

(Ideally, this test should be extended to cover cases of stride != width.)

- Andreas

_______________________________________________
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".

Re: [FFmpeg-devel] [PATCH] checkasm/blockdsp: don't randomize the buffers for fill_block_tab

[Martin Storsjö]

On Tue, 7 May 2024, Andreas Rheinhardt wrote:

> Martin Storsjö:
>> On Mon, 6 May 2024, James Almer wrote:
>> 
>>> It ignores and overwrites the previous values.
>>> Fixes running the test under ubsan.
>>>
>>> Signed-off-by: James Almer <jamrial@gmail.com>
>>> ---
>>> tests/checkasm/blockdsp.c | 3 ++-
>>> 1 file changed, 2 insertions(+), 1 deletion(-)
>> 
>> The change is probably correct, but what issue is ubsan complaining
>> about? If this would just be a dead store of unused random values, that
>> shouldn't be an ubsan issue in general, right?
>> 
>
> UBSan complains about unaligned stores in randomize_buffers; which is
> obvious given that i is incremented by 1, not by 2. I sent a patch that
> fixes this without removing randomization:
> https://ffmpeg.org/pipermail/ffmpeg-devel/2024-May/326945.html

Thanks, that explains it. Those two patches LGTM.

// Martin
_______________________________________________
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".

Re: [FFmpeg-devel] [PATCH] checkasm/blockdsp: don't randomize the buffers for fill_block_tab

[James Almer]

On 5/7/2024 7:49 AM, Andreas Rheinhardt wrote:
> Martin Storsjö:
>> On Mon, 6 May 2024, James Almer wrote:
>>
>>> It ignores and overwrites the previous values.
>>> Fixes running the test under ubsan.
>>>
>>> Signed-off-by: James Almer <jamrial@gmail.com>
>>> ---
>>> tests/checkasm/blockdsp.c | 3 ++-
>>> 1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> The change is probably correct, but what issue is ubsan complaining
>> about? If this would just be a dead store of unused random values, that
>> shouldn't be an ubsan issue in general, right?
>>
> 
> UBSan complains about unaligned stores in randomize_buffers; which is
> obvious given that i is incremented by 1, not by 2. I sent a patch that
> fixes this without removing randomization:

There's no reason to keep the randomization for this test. That's why i 
replaced it with a memset instead of fixing the unaligned writes.

> https://ffmpeg.org/pipermail/ffmpeg-devel/2024-May/326945.html
> 
> (Ideally, this test should be extended to cover cases of stride != width.)
> 
> - Andreas
> 
> _______________________________________________
> 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".
_______________________________________________
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".

[FFmpeg-devel] [PATCH 2/3] checkasm/blockdsp: use smallest allowed aligned buffers for fill_block_tab tests

[James Almer]

The requirement is either 8 or 16 bytes alignment, not 32.
This should help finding bugs in asm implementations.

Signed-off-by: James Almer <jamrial@gmail.com>
---
 tests/checkasm/blockdsp.c | 23 +++++++++--------------
 1 file changed, 9 insertions(+), 14 deletions(-)

diff --git a/tests/checkasm/blockdsp.c b/tests/checkasm/blockdsp.c
index ab87fc8fa4..f67a38d302 100644
--- a/tests/checkasm/blockdsp.c
+++ b/tests/checkasm/blockdsp.c
@@ -29,11 +29,6 @@
 #include "libavutil/intreadwrite.h"
 #include "libavutil/mem_internal.h"
 
-typedef struct {
-    const char *name;
-    int size;
-} test;
-
 #define randomize_buffers(size)             \
     do {                                    \
         int i;                              \
@@ -58,18 +53,18 @@ do {                                                                \
 } while (0)
 
 static void check_fill(BlockDSPContext *h){
-    const test tests[] = {
-        {"fill_block_tab[0]", 16},
-        {"fill_block_tab[1]", 8},
-    };
-    LOCAL_ALIGNED_32(uint8_t, buf0, [16 * 16]);
-    LOCAL_ALIGNED_32(uint8_t, buf1, [16 * 16]);
+    LOCAL_ALIGNED_16(uint8_t, buf0_16, [16 * 16]);
+    LOCAL_ALIGNED_16(uint8_t, buf1_16, [16 * 16]);
+    LOCAL_ALIGNED_8(uint8_t, buf0_8, [8 * 8]);
+    LOCAL_ALIGNED_8(uint8_t, buf1_8, [8 * 8]);
 
-    for (size_t t = 0; t < FF_ARRAY_ELEMS(tests); ++t) {
-        int n = tests[t].size;
+    for (int t = 0; t < 2; ++t) {
+        uint8_t *buf0 = t ? buf0_8 : buf0_16;
+        uint8_t *buf1 = t ? buf1_8 : buf1_16;
+        int n = 16 - 8 * t;
         declare_func(void, uint8_t *block, uint8_t value,
                      ptrdiff_t line_size, int h);
-        if (check_func(h->fill_block_tab[t], "blockdsp.%s", tests[t].name)) {
+        if (check_func(h->fill_block_tab[t], "blockdsp.fill_block_tab[%d]", t)) {
             uint8_t value = rnd();
             memset(buf0, 0, sizeof(*buf0) * n * n);
             memset(buf1, 0, sizeof(*buf0) * n * n);
-- 
2.45.0

_______________________________________________
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".

[FFmpeg-devel] [PATCH 3/3] x86/blockdsp: add sse2 and avx2 versions of fill_block_tab

[James Almer]

Signed-off-by: James Almer <jamrial@gmail.com>
---
 libavcodec/x86/blockdsp.asm    | 33 +++++++++++++++++++++++++++++++++
 libavcodec/x86/blockdsp_init.c | 13 +++++++++++++
 2 files changed, 46 insertions(+)

diff --git a/libavcodec/x86/blockdsp.asm b/libavcodec/x86/blockdsp.asm
index e380308d4a..cccc9a801a 100644
--- a/libavcodec/x86/blockdsp.asm
+++ b/libavcodec/x86/blockdsp.asm
@@ -80,3 +80,36 @@ INIT_XMM sse
 CLEAR_BLOCKS 1
 INIT_YMM avx
 CLEAR_BLOCKS 1
+
+;-----------------------------------------
+; void ff_fill_block_tab_%1(uint8_t *block, uint8_t value,
+;                           ptrdiff_t line_size, int h);
+;-----------------------------------------
+%macro FILL_BLOCK_TAB 2
+cglobal fill_block_tab_%1, 4, 5, 1, block, value, stride, h, stride3
+    lea stride3q, [strideq + strideq * 2]
+%if cpuflag(avx2)
+    movd m0, valued
+    vpbroadcastb m0, m0
+%else
+    SPLATB_REG m0, value, x
+%endif
+.loop:
+    mov%2 [blockq], m0
+    mov%2 [blockq + strideq], m0
+    mov%2 [blockq + strideq * 2], m0
+    mov%2 [blockq + stride3q], m0
+    lea blockq, [blockq + strideq * 4]
+    sub hd, 4
+    jg .loop
+    RET
+%endmacro
+
+INIT_XMM sse2
+FILL_BLOCK_TAB 8, q
+FILL_BLOCK_TAB 16, a
+%if HAVE_AVX2_EXTERNAL
+INIT_XMM avx2
+FILL_BLOCK_TAB 8, q
+FILL_BLOCK_TAB 16, a
+%endif
diff --git a/libavcodec/x86/blockdsp_init.c b/libavcodec/x86/blockdsp_init.c
index 996124114f..37f3bb6a84 100644
--- a/libavcodec/x86/blockdsp_init.c
+++ b/libavcodec/x86/blockdsp_init.c
@@ -29,6 +29,11 @@ void ff_clear_block_avx(int16_t *block);
 void ff_clear_blocks_sse(int16_t *blocks);
 void ff_clear_blocks_avx(int16_t *blocks);
 
+void ff_fill_block_tab_16_sse2(uint8_t *block, uint8_t value, ptrdiff_t line_size, int h);
+void ff_fill_block_tab_8_sse2(uint8_t *block, uint8_t value, ptrdiff_t line_size, int h);
+void ff_fill_block_tab_16_avx2(uint8_t *block, uint8_t value, ptrdiff_t line_size, int h);
+void ff_fill_block_tab_8_avx2(uint8_t *block, uint8_t value, ptrdiff_t line_size, int h);
+
 av_cold void ff_blockdsp_init_x86(BlockDSPContext *c)
 {
 #if HAVE_X86ASM
@@ -38,9 +43,17 @@ av_cold void ff_blockdsp_init_x86(BlockDSPContext *c)
         c->clear_block  = ff_clear_block_sse;
         c->clear_blocks = ff_clear_blocks_sse;
     }
+    if (EXTERNAL_SSE2(cpu_flags)) {
+        c->fill_block_tab[0] = ff_fill_block_tab_16_sse2;
+        c->fill_block_tab[1] = ff_fill_block_tab_8_sse2;
+    }
     if (EXTERNAL_AVX_FAST(cpu_flags)) {
         c->clear_block  = ff_clear_block_avx;
         c->clear_blocks = ff_clear_blocks_avx;
     }
+    if (EXTERNAL_AVX2(cpu_flags)) {
+        c->fill_block_tab[0] = ff_fill_block_tab_16_avx2;
+        c->fill_block_tab[1] = ff_fill_block_tab_8_avx2;
+    }
 #endif /* HAVE_X86ASM */
 }
-- 
2.45.0

_______________________________________________
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".

Re: [FFmpeg-devel] [PATCH 3/3] x86/blockdsp: add sse2 and avx2 versions of fill_block_tab

[Andreas Rheinhardt]

James Almer:
> Signed-off-by: James Almer <jamrial@gmail.com>
> ---
>  libavcodec/x86/blockdsp.asm    | 33 +++++++++++++++++++++++++++++++++
>  libavcodec/x86/blockdsp_init.c | 13 +++++++++++++
>  2 files changed, 46 insertions(+)
> 
> diff --git a/libavcodec/x86/blockdsp.asm b/libavcodec/x86/blockdsp.asm
> index e380308d4a..cccc9a801a 100644
> --- a/libavcodec/x86/blockdsp.asm
> +++ b/libavcodec/x86/blockdsp.asm
> @@ -80,3 +80,36 @@ INIT_XMM sse
>  CLEAR_BLOCKS 1
>  INIT_YMM avx
>  CLEAR_BLOCKS 1
> +
> +;-----------------------------------------
> +; void ff_fill_block_tab_%1(uint8_t *block, uint8_t value,
> +;                           ptrdiff_t line_size, int h);
> +;-----------------------------------------
> +%macro FILL_BLOCK_TAB 2
> +cglobal fill_block_tab_%1, 4, 5, 1, block, value, stride, h, stride3
> +    lea stride3q, [strideq + strideq * 2]
> +%if cpuflag(avx2)
> +    movd m0, valued
> +    vpbroadcastb m0, m0
> +%else
> +    SPLATB_REG m0, value, x
> +%endif
> +.loop:
> +    mov%2 [blockq], m0
> +    mov%2 [blockq + strideq], m0
> +    mov%2 [blockq + strideq * 2], m0
> +    mov%2 [blockq + stride3q], m0
> +    lea blockq, [blockq + strideq * 4]
> +    sub hd, 4
> +    jg .loop
> +    RET
> +%endmacro
> +
> +INIT_XMM sse2
> +FILL_BLOCK_TAB 8, q
> +FILL_BLOCK_TAB 16, a
> +%if HAVE_AVX2_EXTERNAL
> +INIT_XMM avx2
> +FILL_BLOCK_TAB 8, q
> +FILL_BLOCK_TAB 16, a
> +%endif
> diff --git a/libavcodec/x86/blockdsp_init.c b/libavcodec/x86/blockdsp_init.c
> index 996124114f..37f3bb6a84 100644
> --- a/libavcodec/x86/blockdsp_init.c
> +++ b/libavcodec/x86/blockdsp_init.c
> @@ -29,6 +29,11 @@ void ff_clear_block_avx(int16_t *block);
>  void ff_clear_blocks_sse(int16_t *blocks);
>  void ff_clear_blocks_avx(int16_t *blocks);
>  
> +void ff_fill_block_tab_16_sse2(uint8_t *block, uint8_t value, ptrdiff_t line_size, int h);
> +void ff_fill_block_tab_8_sse2(uint8_t *block, uint8_t value, ptrdiff_t line_size, int h);
> +void ff_fill_block_tab_16_avx2(uint8_t *block, uint8_t value, ptrdiff_t line_size, int h);
> +void ff_fill_block_tab_8_avx2(uint8_t *block, uint8_t value, ptrdiff_t line_size, int h);
> +
>  av_cold void ff_blockdsp_init_x86(BlockDSPContext *c)
>  {
>  #if HAVE_X86ASM
> @@ -38,9 +43,17 @@ av_cold void ff_blockdsp_init_x86(BlockDSPContext *c)
>          c->clear_block  = ff_clear_block_sse;
>          c->clear_blocks = ff_clear_blocks_sse;
>      }
> +    if (EXTERNAL_SSE2(cpu_flags)) {
> +        c->fill_block_tab[0] = ff_fill_block_tab_16_sse2;
> +        c->fill_block_tab[1] = ff_fill_block_tab_8_sse2;
> +    }
>      if (EXTERNAL_AVX_FAST(cpu_flags)) {
>          c->clear_block  = ff_clear_block_avx;
>          c->clear_blocks = ff_clear_blocks_avx;
>      }
> +    if (EXTERNAL_AVX2(cpu_flags)) {
> +        c->fill_block_tab[0] = ff_fill_block_tab_16_avx2;
> +        c->fill_block_tab[1] = ff_fill_block_tab_8_avx2;
> +    }
>  #endif /* HAVE_X86ASM */
>  }

Benchmarks?

- Andreas

_______________________________________________
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".

Re: [FFmpeg-devel] [PATCH 2/3] checkasm/blockdsp: use smallest allowed aligned buffers for fill_block_tab tests

[Andreas Rheinhardt]

James Almer:
> The requirement is either 8 or 16 bytes alignment, not 32.
> This should help finding bugs in asm implementations.
> 
> Signed-off-by: James Almer <jamrial@gmail.com>
> ---
>  tests/checkasm/blockdsp.c | 23 +++++++++--------------
>  1 file changed, 9 insertions(+), 14 deletions(-)
> 
> diff --git a/tests/checkasm/blockdsp.c b/tests/checkasm/blockdsp.c
> index ab87fc8fa4..f67a38d302 100644
> --- a/tests/checkasm/blockdsp.c
> +++ b/tests/checkasm/blockdsp.c
> @@ -29,11 +29,6 @@
>  #include "libavutil/intreadwrite.h"
>  #include "libavutil/mem_internal.h"
>  
> -typedef struct {
> -    const char *name;
> -    int size;
> -} test;
> -
>  #define randomize_buffers(size)             \
>      do {                                    \
>          int i;                              \
> @@ -58,18 +53,18 @@ do {                                                                \
>  } while (0)
>  
>  static void check_fill(BlockDSPContext *h){
> -    const test tests[] = {
> -        {"fill_block_tab[0]", 16},
> -        {"fill_block_tab[1]", 8},
> -    };
> -    LOCAL_ALIGNED_32(uint8_t, buf0, [16 * 16]);
> -    LOCAL_ALIGNED_32(uint8_t, buf1, [16 * 16]);
> +    LOCAL_ALIGNED_16(uint8_t, buf0_16, [16 * 16]);
> +    LOCAL_ALIGNED_16(uint8_t, buf1_16, [16 * 16]);
> +    LOCAL_ALIGNED_8(uint8_t, buf0_8, [8 * 8]);
> +    LOCAL_ALIGNED_8(uint8_t, buf1_8, [8 * 8]);
>  
> -    for (size_t t = 0; t < FF_ARRAY_ELEMS(tests); ++t) {
> -        int n = tests[t].size;
> +    for (int t = 0; t < 2; ++t) {
> +        uint8_t *buf0 = t ? buf0_8 : buf0_16;
> +        uint8_t *buf1 = t ? buf1_8 : buf1_16;
> +        int n = 16 - 8 * t;
>          declare_func(void, uint8_t *block, uint8_t value,
>                       ptrdiff_t line_size, int h);
> -        if (check_func(h->fill_block_tab[t], "blockdsp.%s", tests[t].name)) {
> +        if (check_func(h->fill_block_tab[t], "blockdsp.fill_block_tab[%d]", t)) {
>              uint8_t value = rnd();
>              memset(buf0, 0, sizeof(*buf0) * n * n);
>              memset(buf1, 0, sizeof(*buf0) * n * n);

1. I wouldn't be surprised if the *_8 buffers were still 16 byte
aligned. You should probably force 8 byte alignment by using a 16
byte-aligned buffer with an offset of eight.
2. Can you also extend this test to actually test the case of stride !=
width? (And negative strides.)

- Andreas

_______________________________________________
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".

Re: [FFmpeg-devel] [PATCH 3/3] x86/blockdsp: add sse2 and avx2 versions of fill_block_tab

[James Almer]

On 5/7/2024 12:10 PM, Andreas Rheinhardt wrote:
> James Almer:
>> Signed-off-by: James Almer <jamrial@gmail.com>
>> ---
>>   libavcodec/x86/blockdsp.asm    | 33 +++++++++++++++++++++++++++++++++
>>   libavcodec/x86/blockdsp_init.c | 13 +++++++++++++
>>   2 files changed, 46 insertions(+)
>>
>> diff --git a/libavcodec/x86/blockdsp.asm b/libavcodec/x86/blockdsp.asm
>> index e380308d4a..cccc9a801a 100644
>> --- a/libavcodec/x86/blockdsp.asm
>> +++ b/libavcodec/x86/blockdsp.asm
>> @@ -80,3 +80,36 @@ INIT_XMM sse
>>   CLEAR_BLOCKS 1
>>   INIT_YMM avx
>>   CLEAR_BLOCKS 1
>> +
>> +;-----------------------------------------
>> +; void ff_fill_block_tab_%1(uint8_t *block, uint8_t value,
>> +;                           ptrdiff_t line_size, int h);
>> +;-----------------------------------------
>> +%macro FILL_BLOCK_TAB 2
>> +cglobal fill_block_tab_%1, 4, 5, 1, block, value, stride, h, stride3
>> +    lea stride3q, [strideq + strideq * 2]
>> +%if cpuflag(avx2)
>> +    movd m0, valued
>> +    vpbroadcastb m0, m0
>> +%else
>> +    SPLATB_REG m0, value, x
>> +%endif
>> +.loop:
>> +    mov%2 [blockq], m0
>> +    mov%2 [blockq + strideq], m0
>> +    mov%2 [blockq + strideq * 2], m0
>> +    mov%2 [blockq + stride3q], m0
>> +    lea blockq, [blockq + strideq * 4]
>> +    sub hd, 4
>> +    jg .loop
>> +    RET
>> +%endmacro
>> +
>> +INIT_XMM sse2
>> +FILL_BLOCK_TAB 8, q
>> +FILL_BLOCK_TAB 16, a
>> +%if HAVE_AVX2_EXTERNAL
>> +INIT_XMM avx2
>> +FILL_BLOCK_TAB 8, q
>> +FILL_BLOCK_TAB 16, a
>> +%endif
>> diff --git a/libavcodec/x86/blockdsp_init.c b/libavcodec/x86/blockdsp_init.c
>> index 996124114f..37f3bb6a84 100644
>> --- a/libavcodec/x86/blockdsp_init.c
>> +++ b/libavcodec/x86/blockdsp_init.c
>> @@ -29,6 +29,11 @@ void ff_clear_block_avx(int16_t *block);
>>   void ff_clear_blocks_sse(int16_t *blocks);
>>   void ff_clear_blocks_avx(int16_t *blocks);
>>   
>> +void ff_fill_block_tab_16_sse2(uint8_t *block, uint8_t value, ptrdiff_t line_size, int h);
>> +void ff_fill_block_tab_8_sse2(uint8_t *block, uint8_t value, ptrdiff_t line_size, int h);
>> +void ff_fill_block_tab_16_avx2(uint8_t *block, uint8_t value, ptrdiff_t line_size, int h);
>> +void ff_fill_block_tab_8_avx2(uint8_t *block, uint8_t value, ptrdiff_t line_size, int h);
>> +
>>   av_cold void ff_blockdsp_init_x86(BlockDSPContext *c)
>>   {
>>   #if HAVE_X86ASM
>> @@ -38,9 +43,17 @@ av_cold void ff_blockdsp_init_x86(BlockDSPContext *c)
>>           c->clear_block  = ff_clear_block_sse;
>>           c->clear_blocks = ff_clear_blocks_sse;
>>       }
>> +    if (EXTERNAL_SSE2(cpu_flags)) {
>> +        c->fill_block_tab[0] = ff_fill_block_tab_16_sse2;
>> +        c->fill_block_tab[1] = ff_fill_block_tab_8_sse2;
>> +    }
>>       if (EXTERNAL_AVX_FAST(cpu_flags)) {
>>           c->clear_block  = ff_clear_block_avx;
>>           c->clear_blocks = ff_clear_blocks_avx;
>>       }
>> +    if (EXTERNAL_AVX2(cpu_flags)) {
>> +        c->fill_block_tab[0] = ff_fill_block_tab_16_avx2;
>> +        c->fill_block_tab[1] = ff_fill_block_tab_8_avx2;
>> +    }
>>   #endif /* HAVE_X86ASM */
>>   }
> 
> Benchmarks?

blockdsp.fill_block_tab[0]_c: 13.4
blockdsp.fill_block_tab[0]_sse2: 8.4
blockdsp.fill_block_tab[0]_avx2: 7.4
blockdsp.fill_block_tab[1]_c: 7.9
blockdsp.fill_block_tab[1]_sse2: 4.5
blockdsp.fill_block_tab[1]_avx2: 4.0

On an Alder Lake using --cpu-alderlake. If i let gcc compile without cpu 
specific optimizations, the C version is even slower.
_______________________________________________
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".

Re: [FFmpeg-devel] [PATCH 2/3] checkasm/blockdsp: use smallest allowed aligned buffers for fill_block_tab tests

[James Almer]

On 5/7/2024 12:14 PM, Andreas Rheinhardt wrote:
> James Almer:
>> The requirement is either 8 or 16 bytes alignment, not 32.
>> This should help finding bugs in asm implementations.
>>
>> Signed-off-by: James Almer <jamrial@gmail.com>
>> ---
>>   tests/checkasm/blockdsp.c | 23 +++++++++--------------
>>   1 file changed, 9 insertions(+), 14 deletions(-)
>>
>> diff --git a/tests/checkasm/blockdsp.c b/tests/checkasm/blockdsp.c
>> index ab87fc8fa4..f67a38d302 100644
>> --- a/tests/checkasm/blockdsp.c
>> +++ b/tests/checkasm/blockdsp.c
>> @@ -29,11 +29,6 @@
>>   #include "libavutil/intreadwrite.h"
>>   #include "libavutil/mem_internal.h"
>>   
>> -typedef struct {
>> -    const char *name;
>> -    int size;
>> -} test;
>> -
>>   #define randomize_buffers(size)             \
>>       do {                                    \
>>           int i;                              \
>> @@ -58,18 +53,18 @@ do {                                                                \
>>   } while (0)
>>   
>>   static void check_fill(BlockDSPContext *h){
>> -    const test tests[] = {
>> -        {"fill_block_tab[0]", 16},
>> -        {"fill_block_tab[1]", 8},
>> -    };
>> -    LOCAL_ALIGNED_32(uint8_t, buf0, [16 * 16]);
>> -    LOCAL_ALIGNED_32(uint8_t, buf1, [16 * 16]);
>> +    LOCAL_ALIGNED_16(uint8_t, buf0_16, [16 * 16]);
>> +    LOCAL_ALIGNED_16(uint8_t, buf1_16, [16 * 16]);
>> +    LOCAL_ALIGNED_8(uint8_t, buf0_8, [8 * 8]);
>> +    LOCAL_ALIGNED_8(uint8_t, buf1_8, [8 * 8]);
>>   
>> -    for (size_t t = 0; t < FF_ARRAY_ELEMS(tests); ++t) {
>> -        int n = tests[t].size;
>> +    for (int t = 0; t < 2; ++t) {
>> +        uint8_t *buf0 = t ? buf0_8 : buf0_16;
>> +        uint8_t *buf1 = t ? buf1_8 : buf1_16;
>> +        int n = 16 - 8 * t;
>>           declare_func(void, uint8_t *block, uint8_t value,
>>                        ptrdiff_t line_size, int h);
>> -        if (check_func(h->fill_block_tab[t], "blockdsp.%s", tests[t].name)) {
>> +        if (check_func(h->fill_block_tab[t], "blockdsp.fill_block_tab[%d]", t)) {
>>               uint8_t value = rnd();
>>               memset(buf0, 0, sizeof(*buf0) * n * n);
>>               memset(buf1, 0, sizeof(*buf0) * n * n);
> 
> 1. I wouldn't be surprised if the *_8 buffers were still 16 byte
> aligned. You should probably force 8 byte alignment by using a 16
> byte-aligned buffer with an offset of eight.

Amended the following locally:

> diff --git a/tests/checkasm/blockdsp.c b/tests/checkasm/blockdsp.c
> index 8c1f8281d2..5f4d46b8fa 100644
> --- a/tests/checkasm/blockdsp.c
> +++ b/tests/checkasm/blockdsp.c
> @@ -55,12 +55,10 @@ do {                                                                \
>  static void check_fill(BlockDSPContext *h){
>      LOCAL_ALIGNED_16(uint8_t, buf0_16, [16 * 16]);
>      LOCAL_ALIGNED_16(uint8_t, buf1_16, [16 * 16]);
> -    LOCAL_ALIGNED_8(uint8_t, buf0_8, [8 * 8]);
> -    LOCAL_ALIGNED_8(uint8_t, buf1_8, [8 * 8]);
> 
>      for (int t = 0; t < 2; ++t) {
> -        uint8_t *buf0 = t ? buf0_8 : buf0_16;
> -        uint8_t *buf1 = t ? buf1_8 : buf1_16;
> +        uint8_t *buf0 = buf0_16 + t * /* force 8 byte alignment */ 8;
> +        uint8_t *buf1 = buf1_16 + t * /* force 8 byte alignment */ 8;
>          int n = 16 - 8 * t;
>          declare_func(void, uint8_t *block, uint8_t value,
>                       ptrdiff_t line_size, int h);

> 2. Can you also extend this test to actually test the case of stride !=
> width? (And negative strides.)

Maybe later.
_______________________________________________
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".