[FFmpeg-devel] [RFC] x86 external assembler requirements

to prevent confusion. We could just change them to "nasm" and be done. We could provide compatability options. We could adopt Libav's generic "x86asm". James Darnley (1): configure: require NASM version 2.11 or newer for external x86 assembly configure | 17 +

Re: [FFmpeg-devel] [WIP][PATCH] Opus Piramid Vector Quantization Search in x86 SIMD asm

On 2017-06-09 13:41, Ivan Kalvachev wrote: > On 6/9/17, Michael Niedermayer wrote: >> seems this breaks build with mingw64, didnt investigate but it >> fails with these errors: >> >> libavcodec/libavcodec.a(opus_pvq_search.o):src/libavcodec/x86/opus_pvq_search.asm:(.text+0x2d): >> relocation trunc

[FFmpeg-devel] [WIP] [PATCH 0/5] sse2/avx functions for 8-bit simple_idct

still have a small optimisation to make and I need to use the correct coefficients. This will require a large change to the macros. I am sending this so that people can nitpick my changes. James Darnley (5): avcodec/x86: cleanup simple_idct10 avcodec/x86: add x86-64 8-bit simple_idct function

[FFmpeg-devel] [PATCH 3/5] more cleanup

--- libavcodec/x86/simple_idct10_template.asm | 16 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/libavcodec/x86/simple_idct10_template.asm b/libavcodec/x86/simple_idct10_template.asm index c0d1637ca2..3f398985a5 100644 --- a/libavcodec/x86/simple_idct10_template.

[FFmpeg-devel] [PATCH 1/5] avcodec/x86: cleanup simple_idct10

Use named arguments for the functions so we can remove a define. The stride/linesize argument is now ptrdiff_t type so we no longer need to sign extend the register. --- libavcodec/x86/proresdsp.asm | 2 +- libavcodec/x86/simple_idct10.asm | 8 ++-- libavcodec/x86/simple_i

[FFmpeg-devel] [PATCH 2/5] avcodec/x86: add x86-64 8-bit simple_idct function

Rounding contributed by Ronald S. Bultje --- libavcodec/tests/x86/dct.c | 2 ++ libavcodec/x86/idctdsp_init.c| 19 +++ libavcodec/x86/simple_idct.h | 3 +++ libavcodec/x86/simple_idct10.asm | 8 4 files changed, 32 insertions(+) diff --git a/libavcodec/te

[FFmpeg-devel] [PATCH 4/5] avcodec/x86: add x86-64 8-bit simple_idct put function

--- libavcodec/x86/idctdsp_init.c| 2 ++ libavcodec/x86/simple_idct.h | 3 +++ libavcodec/x86/simple_idct10.asm | 23 +++ 3 files changed, 28 insertions(+) diff --git a/libavcodec/x86/idctdsp_init.c b/libavcodec/x86/idctdsp_init.c index 4b2145e478..1826d01e0e 100644

[FFmpeg-devel] [PATCH 5/5] avcodec/x86: add x86-64 8-bit simple_idct add function

--- libavcodec/x86/idctdsp_init.c| 2 ++ libavcodec/x86/simple_idct.h | 3 ++ libavcodec/x86/simple_idct10.asm | 61 3 files changed, 66 insertions(+) diff --git a/libavcodec/x86/idctdsp_init.c b/libavcodec/x86/idctdsp_init.c index 1826d01e0e..ca

Re: [FFmpeg-devel] [WIP][PATCH] Opus Piramid Vector Quantization Search in x86 SIMD asm

On 2017-06-11 11:34, Ivan Kalvachev wrote: > On 6/10/17, James Darnley wrote: >> On 2017-06-09 13:41, Ivan Kalvachev wrote: >>> >>> const_*_edge is used on only one place is the code. >>> Would you check if this patch fixes the issue. >>> >>>

[FFmpeg-devel] [PATCH 0/6] sse2/avx functions for 8-bit simple idct

kindly give their opinion on the 2nd and 6th patches in particular I would greatly appreciate it. Performance gain decoding an MPEG2 HD sample over the old MMX: - Yorkfield: 210 to 224 fps - Haswell: 387 to 426 fps Would anyone like me to get some timer figures for the functions themselves? James

[FFmpeg-devel] [PATCH 1/6] avcodec/x86: cleanup simple_idct10

Use named arguments for the functions so we can remove a define. The stride/linesize argument is now ptrdiff_t type so we no longer need to sign extend the register. --- libavcodec/x86/proresdsp.asm | 2 +- libavcodec/x86/simple_idct10.asm | 8 ++-- libavcodec/x86/simple_i

[FFmpeg-devel] [PATCH 6/6] avcodec/x86: allow 8-bit simple_idct to use slightly different coefficients

This makes it exact to the old MMX one, as reported by libavcodec/tests/dct. --- libavcodec/x86/proresdsp.asm | 18 + libavcodec/x86/simple_idct10.asm | 33 ++- libavcodec/x86/simple_idct10_template.asm | 19 ++ 3 fi

[FFmpeg-devel] [PATCH 5/6] avcodec/x86: add x86-64 8-bit simple_idct add function

--- libavcodec/x86/idctdsp_init.c| 2 ++ libavcodec/x86/simple_idct.h | 3 ++ libavcodec/x86/simple_idct10.asm | 61 3 files changed, 66 insertions(+) diff --git a/libavcodec/x86/idctdsp_init.c b/libavcodec/x86/idctdsp_init.c index 1826d01e0e..9d

[FFmpeg-devel] [PATCH 2/6] avcodec/x86: modify simple_idct10 macros to add an action paramter

--- libavcodec/x86/proresdsp.asm | 2 +- libavcodec/x86/simple_idct10.asm | 8 +++ libavcodec/x86/simple_idct10_template.asm | 37 +-- 3 files changed, 25 insertions(+), 22 deletions(-) diff --git a/libavcodec/x86/proresdsp.asm b/libavcodec/

[FFmpeg-devel] [PATCH 4/6] avcodec/x86: add x86-64 8-bit simple_idct put function

--- libavcodec/x86/idctdsp_init.c| 2 ++ libavcodec/x86/simple_idct.h | 3 +++ libavcodec/x86/simple_idct10.asm | 23 +++ 3 files changed, 28 insertions(+) diff --git a/libavcodec/x86/idctdsp_init.c b/libavcodec/x86/idctdsp_init.c index 4b2145e478..1826d01e0e 100644

[FFmpeg-devel] [PATCH 3/6] avcodec/x86: add x86-64 8-bit simple_idct function

Rounding contributed by Ronald S. Bultje --- libavcodec/tests/x86/dct.c | 2 ++ libavcodec/x86/idctdsp_init.c| 19 +++ libavcodec/x86/simple_idct.h | 3 +++ libavcodec/x86/simple_idct10.asm | 8 4 files changed, 32 insertions(+) diff --git a/libavcodec/te

Re: [FFmpeg-devel] [PATCH 0/6] sse2/avx functions for 8-bit simple idct

On 2017-06-12 18:57, Michael Niedermayer wrote: > On Mon, Jun 12, 2017 at 03:36:06PM +0200, James Darnley wrote: >> Rounding contributed by Ronald S. Bultje >> --- >> libavcodec/tests/x86/dct.c | 2 ++ >> libavcodec/x86/idctdsp_init.c| 19 +++

Re: [FFmpeg-devel] [PATCH 0/6] sse2/avx functions for 8-bit simple idct

On 2017-06-13 00:18, James Darnley wrote: > On 2017-06-12 18:57, Michael Niedermayer wrote: >> ./ffplay ~/videos/matrixbench_mpeg2.mpg >> looks pretty bad > > If that would happen to be the FATE sample > mpeg2/matrixbench_mpeg2.lq1.mpg then I see that too. > >

[FFmpeg-devel] [PATCH 1/6] fate: add test of -idct simpleauto

--- tests/fate/video.mak | 3 +++ tests/ref/fate/idct-simpleauto | 27 +++ 2 files changed, 30 insertions(+) create mode 100644 tests/ref/fate/idct-simpleauto diff --git a/tests/fate/video.mak b/tests/fate/video.mak index d1d35335f2..455c1d3564 100644 --- a/tes

[FFmpeg-devel] [PATCH 0/6] [v2] sse2/avx functions for 8-bit simple idct

r platforms use their own functions for simpleauto. I might follow this with a patch to cleanup idctdsp_init.c James Darnley (6): fate: add test of -idct simpleauto avcodec/x86: cleanup simple_idct10 avcodec/x86: modify simple_idct10 macros to add an action paramter avcodec/x86: allow future

[FFmpeg-devel] [PATCH 5/6] avcodec/x86: allow future 8-bit simple idct to have "DC only hack"

Created by Ronald S. Bultje --- libavcodec/x86/simple_idct10_template.asm | 38 +++ 1 file changed, 38 insertions(+) diff --git a/libavcodec/x86/simple_idct10_template.asm b/libavcodec/x86/simple_idct10_template.asm index d8ea0bcc6b..51baf84c82 100644 --- a/libavcodec

[FFmpeg-devel] [PATCH 4/6] avcodec/x86: allow future 8-bit simple idct to use slightly different coefficients

--- libavcodec/x86/proresdsp.asm | 18 ++ libavcodec/x86/simple_idct10.asm | 29 + libavcodec/x86/simple_idct10_template.asm | 19 +++ 3 files changed, 50 insertions(+), 16 deletions(-) diff --git a/libavcodec/x86/p

[FFmpeg-devel] [PATCH 3/6] avcodec/x86: modify simple_idct10 macros to add an action paramter

--- libavcodec/x86/proresdsp.asm | 2 +- libavcodec/x86/simple_idct10.asm | 8 +++ libavcodec/x86/simple_idct10_template.asm | 37 +-- 3 files changed, 25 insertions(+), 22 deletions(-) diff --git a/libavcodec/x86/proresdsp.asm b/libavcodec/

[FFmpeg-devel] [PATCH 2/6] avcodec/x86: cleanup simple_idct10

Use named arguments for the functions so we can remove a define. The stride/linesize argument is now ptrdiff_t type so we no longer need to sign extend the register. --- libavcodec/x86/proresdsp.asm | 2 +- libavcodec/x86/simple_idct10.asm | 8 ++-- libavcodec/x86/simple_i

[FFmpeg-devel] [PATCH 6/6] avcodec/x86: add an 8-bit simple IDCT function based on the x86-64 high depth functions

Includes add/put functions Rounding contributed by Ronald S. Bultje --- libavcodec/tests/x86/dct.c | 2 + libavcodec/x86/idctdsp_init.c| 23 ++ libavcodec/x86/simple_idct.h | 9 libavcodec/x86/simple_idct10.asm | 94 4 files ch

[FFmpeg-devel] [PATCH] avcodec/x86: add an 8-bit simple IDCT function based on the x86-64 high depth functions

Includes add/put functions Rounding contributed by Ronald S. Bultje --- I must be stupid. I dropped the stack space change somewhere. libavcodec/tests/x86/dct.c | 2 + libavcodec/x86/idctdsp_init.c| 23 ++ libavcodec/x86/simple_idct.h | 9 libavcodec/x86/simple_idct

Re: [FFmpeg-devel] [PATCH] avcodec/x86: add an 8-bit simple IDCT function based on the x86-64 high depth functions

On 2017-06-16 03:58, Michael Niedermayer wrote: > On Thu, Jun 15, 2017 at 05:08:33PM +0200, James Darnley wrote: >> Includes add/put functions >> >> Rounding contributed by Ronald S. Bultje >> --- >> I must be stupid. I dropped the stack space change somewhere.

Re: [FFmpeg-devel] [PATCH] avcodec/x86: add an 8-bit simple IDCT function based on the x86-64 high depth functions

On 2017-06-16 12:48, Paul B Mahol wrote: > On 6/16/17, James Darnley wrote: >> On 2017-06-16 03:58, Michael Niedermayer wrote: >>> theres something wrong with this >>> it totally breaks this: >>> make -j12 ffmpeg && ./ffmpeg -ss 1 -i cache:matrixbench

[FFmpeg-devel] [PATCH 2/2] avcodec/x86/mpegenc: support transpose permuation type

--- libavcodec/x86/mpegvideoenc_template.c | 47 +- 1 file changed, 46 insertions(+), 1 deletion(-) diff --git a/libavcodec/x86/mpegvideoenc_template.c b/libavcodec/x86/mpegvideoenc_template.c index 3ce72e1367..1201be514b 100644 --- a/libavcodec/x86/mpegvideoenc_t

[FFmpeg-devel] [PATCH 1/2] avcodec/x86/mpegenc: check IDCT permutation type is a valid value

--- libavcodec/x86/mpegvideoenc_template.c | 8 +++- 1 file changed, 7 insertions(+), 1 deletion(-) diff --git a/libavcodec/x86/mpegvideoenc_template.c b/libavcodec/x86/mpegvideoenc_template.c index b2512744ca..3ce72e1367 100644 --- a/libavcodec/x86/mpegvideoenc_template.c +++ b/libavcodec/x

Re: [FFmpeg-devel] [PATCH 2/2] avcodec/x86/mpegenc: support transpose permuation type

On 2017-06-16 17:48, Michael Niedermayer wrote: > On Fri, Jun 16, 2017 at 03:53:28PM +0200, James Darnley wrote: >> --- >> libavcodec/x86/mpegvideoenc_template.c | 47 >> +- >> 1 file changed, 46 insertions(+), 1 deletion(-) >

Re: [FFmpeg-devel] [PATCH 1/6] fate: add test of -idct simpleauto

On 2017-06-16 20:31, Michael Niedermayer wrote: > On Thu, Jun 15, 2017 at 03:34:21PM +0200, James Darnley wrote: >> --- >> tests/fate/video.mak | 3 +++ >> tests/ref/fate/idct-simpleauto | 27 +++ >> 2 files changed, 30 insertions(+) &

[FFmpeg-devel] [PATCH 01/11] avcodec/x86/mpegenc: check IDCT permutation type is a valid value

--- libavcodec/x86/mpegvideoenc_template.c | 8 +++- 1 file changed, 7 insertions(+), 1 deletion(-) diff --git a/libavcodec/x86/mpegvideoenc_template.c b/libavcodec/x86/mpegvideoenc_template.c index b2512744ca..3ce72e1367 100644 --- a/libavcodec/x86/mpegvideoenc_template.c +++ b/libavcodec/x

[FFmpeg-devel] [PATCH 00/11] [v3] sse2/avx functions for 8-bit simple idct

since outdated and I added a comment to the "DC-only hack" above it in the contitional section. The other patches either fix or workaround bugs in other code. James Darnley (9): avcodec/x86/mpegenc: check IDCT permutation type is a valid value avcodec/x86/mpegenc: support transpose

[FFmpeg-devel] [PATCH 02/11] avcodec/x86/mpegenc: support transpose permuation type

--- libavcodec/x86/mpegvideoenc_template.c | 47 +- 1 file changed, 46 insertions(+), 1 deletion(-) diff --git a/libavcodec/x86/mpegvideoenc_template.c b/libavcodec/x86/mpegvideoenc_template.c index 3ce72e1367..1201be514b 100644 --- a/libavcodec/x86/mpegvideoenc_t

[FFmpeg-devel] [PATCH 03/11] avcodec/mdec: override IDCT choice before initing DSP structs

--- libavcodec/mdec.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/libavcodec/mdec.c b/libavcodec/mdec.c index 8e28aa04f0..97bfebbeb7 100644 --- a/libavcodec/mdec.c +++ b/libavcodec/mdec.c @@ -213,6 +213,9 @@ static av_cold int decode_init(AVCodecContext *avctx) {

[FFmpeg-devel] [PATCH 08/11] avcodec/x86: allow future 8-bit simple idct to use slightly different coefficients

--- libavcodec/x86/proresdsp.asm | 18 ++ libavcodec/x86/simple_idct10.asm | 29 + libavcodec/x86/simple_idct10_template.asm | 19 +++ 3 files changed, 50 insertions(+), 16 deletions(-) diff --git a/libavcodec/x86/p

[FFmpeg-devel] [PATCH 09/11] avcodec/x86: allow future 8-bit simple idct to have "DC only hack"

Created by Ronald S. Bultje --- libavcodec/x86/simple_idct10_template.asm | 38 +++ 1 file changed, 38 insertions(+) diff --git a/libavcodec/x86/simple_idct10_template.asm b/libavcodec/x86/simple_idct10_template.asm index d8ea0bcc6b..51baf84c82 100644 --- a/libavcodec

[FFmpeg-devel] [PATCH 06/11] avcodec/x86: cleanup simple_idct10

Use named arguments for the functions so we can remove a define. The stride/linesize argument is now ptrdiff_t type so we no longer need to sign extend the register. --- libavcodec/x86/proresdsp.asm | 2 +- libavcodec/x86/simple_idct10.asm | 8 ++-- libavcodec/x86/simple_i

[FFmpeg-devel] [PATCH 10/11] avcodec/x86: add an 8-bit simple IDCT function based on the x86-64 high depth functions

Includes add/put functions Rounding contributed by Ronald S. Bultje --- libavcodec/tests/x86/dct.c| 2 + libavcodec/x86/idctdsp_init.c | 23 libavcodec/x86/simple_idct.h | 9 +++ libavcodec/x86/simple_idct10.asm | 92 +++

[FFmpeg-devel] [PATCH 07/11] avcodec/x86: modify simple_idct10 macros to add an action paramter

--- libavcodec/x86/proresdsp.asm | 2 +- libavcodec/x86/simple_idct10.asm | 8 +++ libavcodec/x86/simple_idct10_template.asm | 37 +-- 3 files changed, 25 insertions(+), 22 deletions(-) diff --git a/libavcodec/x86/proresdsp.asm b/libavcodec/

[FFmpeg-devel] [PATCH 05/11] avcodec/mpegenc: do not use unquantize shortcuts for wmv1

From: "Ronald S. Bultje" Commit message by James Darnley The shortcut is based on end-of-block positions. This leads to some coefficients not being unquantized. This is the symptom of the bug. A possible candidate for the real bug is the scan table used here in unquantize does not

[FFmpeg-devel] [PATCH 04/11] avcodec/mdec: allow/use permuted IDCTs

From: "Ronald S. Bultje" --- libavcodec/mdec.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/libavcodec/mdec.c b/libavcodec/mdec.c index 97bfebbeb7..1e1c8f4c55 100644 --- a/libavcodec/mdec.c +++ b/libavcodec/mdec.c @@ -94,7 +94,7 @@ static inline int mdec_decode_block

[FFmpeg-devel] [PATCH 11/11] avcodec/x86: use new x86-64 functions for -idct simple

They now match according to FATE, barring any further bugs with untested parts --- libavcodec/x86/idctdsp_init.c | 6 -- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/libavcodec/x86/idctdsp_init.c b/libavcodec/x86/idctdsp_init.c index 9da60d1a1e..162560d411 100644 --- a/libavco

Re: [FFmpeg-devel] [PATCH 05/11] avcodec/mpegenc: do not use unquantize shortcuts for wmv1

On 2017-06-20 00:08, Michael Niedermayer wrote: > On Mon, Jun 19, 2017 at 05:10:58PM +0200, James Darnley wrote: >> From: "Ronald S. Bultje" >> >> Commit message by James Darnley >> >> The shortcut is based on end-of-block positions. This leads to

Re: [FFmpeg-devel] [PATCH 02/11] avcodec/x86/mpegenc: support transpose permuation type

On 2017-06-19 18:30, Michael Niedermayer wrote: > On Mon, Jun 19, 2017 at 05:10:55PM +0200, James Darnley wrote: >> --- >> libavcodec/x86/mpegvideoenc_template.c | 47 >> +- >> 1 file changed, 46 insertions(+), 1 deletion(-) > &

Re: [FFmpeg-devel] [PATCH 06/11] avcodec/x86: cleanup simple_idct10

On 2017-06-19 20:30, Ronald S. Bultje wrote: > Hi, > > On Mon, Jun 19, 2017 at 11:10 AM, James Darnley wrote: > >> Use named arguments for the functions so we can remove a define. The >> stride/linesize argument is now ptrdiff_t type so we no longer need to >

Re: [FFmpeg-devel] [PATCH 07/11] avcodec/x86: modify simple_idct10 macros to add an action paramter

On 2017-06-19 20:31, Ronald S. Bultje wrote: > Hi, > > On Mon, Jun 19, 2017 at 11:11 AM, James Darnley wrote: > >> --- >> libavcodec/x86/proresdsp.asm | 2 +- >> libavcodec/x86/simple_idct10.asm | 8 +++ >> libavcodec

Re: [FFmpeg-devel] [PATCH 09/11] avcodec/x86: allow future 8-bit simple idct to have "DC only hack"

On 2017-06-20 13:56, Ronald S. Bultje wrote: > Hi, > > On Mon, Jun 19, 2017 at 11:11 AM, James Darnley wrote: > >> Created by Ronald S. Bultje >> --- >> libavcodec/x86/simple_idct10_template.asm | 38 >> +++ >> 1 file ch

Re: [FFmpeg-devel] [PATCH] mdec: use correctly permutated quant matrix for dequantization.

On 2017-06-20 14:47, Ronald S. Bultje wrote: > This allows using non-simple (e.g. simplemmx) IDCT implementations. > The result is not bitexact (which is why the fate test continues to > use -idct simple), but the PSNR between C/MMX goes from ~35 to ~90. > --- > libavcodec/mdec.c | 14 ++--

Re: [FFmpeg-devel] [PATCH] mdec: use correctly permutated quant matrix for dequantization.

On 2017-06-20 18:16, Ronald S. Bultje wrote: > On Tue, Jun 20, 2017 at 12:04 PM, James Darnley wrote: >>> @@ -231,6 +230,13 @@ static av_cold int decode_init(AVCodecContext >> *avctx) >>> avctx->pix_fmt = AV_PIX_FMT_YUVJ420P; >>>

Re: [FFmpeg-devel] [PATCH 08/11] avcodec/x86: allow future 8-bit simple idct to use slightly different coefficients

On 2017-06-20 13:55, Ronald S. Bultje wrote: > Hi, > > On Mon, Jun 19, 2017 at 11:11 AM, James Darnley wrote: > >> --- >> libavcodec/x86/proresdsp.asm | 18 ++ >> libavcodec/x86/simple_idct10.asm | 29 >>

Re: [FFmpeg-devel] [PATCH 10/11] avcodec/x86: add an 8-bit simple IDCT function based on the x86-64 high depth functions

On 2017-06-19 17:11, James Darnley wrote: > diff --git a/libavcodec/x86/simple_idct10_template.asm > b/libavcodec/x86/simple_idct10_template.asm > index 51baf84c82..02fd445ec0 100644 > --- a/libavcodec/x86/simple_idct10_template.asm > +++ b/libavcodec/x86/simple_idct10_template.

Re: [FFmpeg-devel] [PATCH 1/2] avcodec/mpegvideo: Use intra_scantable in dct_unquantize_h263_intra_c()

On 2017-06-20 00:37, Michael Niedermayer wrote: > Signed-off-by: Michael Niedermayer > --- > libavcodec/mpegvideo.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c > index 63a30b93ce..e29558b3a2 100644 > --- a/libavcodec/mp

Re: [FFmpeg-devel] [PATCH 6/8] x86inc: reduce difference to x264 upstream

On 2017-10-31 04:30, Michael Niedermayer wrote: > On Mon, Oct 30, 2017 at 02:08:33PM +0100, James Darnley wrote: >> These changes were commited to x264 in b568a256 "Experimental nasm >> support" >> --- >> libavutil/x86/x86inc.asm | 16 ++-- &g

Re: [FFmpeg-devel] [PATCH 6/8] x86inc: reduce difference to x264 upstream

On 2017-11-06 21:15, James Almer wrote: > On 11/6/2017 4:56 PM, James Darnley wrote: >> Line 733 is the align command to align the start of the function. I >> can't see why it fails here but not on any other function in that file >> or any other file. >> >> A

[FFmpeg-devel] [PATCH 00/11] AVX-512 support (v.2)

hat acutally use ZMM registers. Henrik Gramner (1): x86inc: AVX-512 support James Darnley (10): configure: test whether x86 assembler supports AVX-512 avutil: add AVX-512 flags avutil: detect when AVX-512 is available avutil: add alignment needed for AVX-512 avcodec: add stride alignmen

[FFmpeg-devel] [PATCH 01/11] configure: test whether x86 assembler supports AVX-512

--- configure | 5 + 1 file changed, 5 insertions(+) diff --git a/configure b/configure index f396abda5b..146a87324c 100755 --- a/configure +++ b/configure @@ -406,6 +406,7 @@ Optimization options (experts only): --disable-fma3 disable FMA3 optimizations --disable-fma4

[FFmpeg-devel] [PATCH 04/11] avutil: add alignment needed for AVX-512

--- This patch gained the alignmnet increase in mem.c libavutil/mem.c | 2 +- libavutil/x86/cpu.c | 2 ++ 2 files changed, 3 insertions(+), 1 deletion(-) diff --git a/libavutil/mem.c b/libavutil/mem.c index 6ad409daf4..cdf539306f 100644 --- a/libavutil/mem.c +++ b/libavutil/mem.c @@ -61,7 +6

[FFmpeg-devel] [PATCH 11/11] avcodec/lossless_videodsp: add AVX-512 version of add_bytes

--- libavcodec/x86/lossless_videodsp.asm| 5 + libavcodec/x86/lossless_videodsp_init.c | 5 + 2 files changed, 10 insertions(+) diff --git a/libavcodec/x86/lossless_videodsp.asm b/libavcodec/x86/lossless_videodsp.asm index ba4d4f0153..5649348f86 100644 --- a/libavcodec/x86/lossless_v

[FFmpeg-devel] [PATCH 07/11] checkasm: support for AVX-512 functions

--- tests/checkasm/checkasm.c | 1 + 1 file changed, 1 insertion(+) diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c index b8b0e32dbd..9fb1438bdb 100644 --- a/tests/checkasm/checkasm.c +++ b/tests/checkasm/checkasm.c @@ -192,6 +192,7 @@ static const struct { { "FMA3", "

[FFmpeg-devel] [PATCH 10/11] avcodec/blockdsp: add AVX-512 version of clear_block(s)

From: James Darnley Also adjust alignment requirements where nessecary. --- Whether this patch is committed or not the change to 4xm.c should be picked to master because the alignment is wrong for the AVX version of this function. I assume it hasn't been noticed yet because it manages to

[FFmpeg-devel] [PATCH 06/11] x86inc: AVX-512 support

From: Henrik Gramner AVX-512 consists of a plethora of different extensions, but in order to keep things a bit more manageable we group together the following extensions under a single baseline cpu flag which should cover SKL-X and future CPUs: * AVX-512 Foundation (F) * AVX-512 Conflict Detect

[FFmpeg-devel] [PATCH 03/11] avutil: detect when AVX-512 is available

--- I've changed this patch slightly because I discovered that it would cause an illegal instruction exception on much older processors (probably all without AVX). I was running xgetbv() almost uncontitionally. Now it is a little more like what is the in x264 patch. libavutil/x86/cpu.c | 12 +++

[FFmpeg-devel] [PATCH 09/11] avcodec/blockdsp: roll-up x86asm preprocessor loop

From: James Darnley --- libavcodec/x86/blockdsp.asm | 11 --- 1 file changed, 4 insertions(+), 7 deletions(-) diff --git a/libavcodec/x86/blockdsp.asm b/libavcodec/x86/blockdsp.asm index 9d203df8f5..9d0e8a3242 100644 --- a/libavcodec/x86/blockdsp.asm +++ b/libavcodec/x86/blockdsp.asm

[FFmpeg-devel] [PATCH 02/11] avutil: add AVX-512 flags

--- libavutil/cpu.c | 6 +- libavutil/cpu.h | 1 + libavutil/tests/cpu.c | 1 + libavutil/x86/cpu.h | 2 ++ 4 files changed, 9 insertions(+), 1 deletion(-) diff --git a/libavutil/cpu.c b/libavutil/cpu.c index c8401b8258..6548cc3042 100644 --- a/libavutil/cpu.c +++ b/libavutil/cp

[FFmpeg-devel] [PATCH 08/11] avcodec/v210enc: add AVX-512 10-bit line pack function

--- libavcodec/x86/v210enc.asm| 5 + libavcodec/x86/v210enc_init.c | 7 +++ 2 files changed, 12 insertions(+) diff --git a/libavcodec/x86/v210enc.asm b/libavcodec/x86/v210enc.asm index 965f2bea3c..5068af27f8 100644 --- a/libavcodec/x86/v210enc.asm +++ b/libavcodec/x86/v210enc.asm @@ -

[FFmpeg-devel] [PATCH 05/11] avcodec: add stride alignment needed for AVX-512

--- configure | 2 ++ libavcodec/internal.h | 4 +++- 2 files changed, 5 insertions(+), 1 deletion(-) diff --git a/configure b/configure index 146a87324c..fce8030d91 100755 --- a/configure +++ b/configure @@ -1886,6 +1886,7 @@ ARCH_FEATURES=" local_aligned simd_align_16

Re: [FFmpeg-devel] [PATCH 10/11] avcodec/blockdsp: add AVX-512 version of clear_block(s)

On 2017-11-09 20:35, Martin Vignali wrote: > 2017-11-09 12:58 GMT+01:00 James Darnley : > >> From: James Darnley >> >> Also adjust alignment requirements where nessecary. >> --- >> Whether this patch is committed or not the change to 4xm.c should be >>

Re: [FFmpeg-devel] [PATCH 11/11] avcodec/lossless_videodsp: add AVX-512 version of add_bytes

On 2017-11-09 20:43, Martin Vignali wrote: > 2017-11-09 20:37 GMT+01:00 Martin Vignali : >> lgtm >> >> Can you post your checkasm benchmark result for this ? Yep > $ ./tests/checkasm/checkasm --bench --test=llviddsp > benchmarking with native FFmpeg timers > nop: 26.0 > checkasm: using random see

Re: [FFmpeg-devel] [PATCH 08/11] avcodec/v210enc: add AVX-512 10-bit line pack function

On 2017-11-09 20:42, Martin Vignali wrote: > I doesn't want to block this patch, but > like you say (in your previous version), that this version is not faster, > i'm not sure, it's interesting to apply it. > You already made "real" avx512 version for other funcs, in order to check > the rest of yo

Re: [FFmpeg-devel] [PATCH 05/11] avcodec: add stride alignment needed for AVX-512

On 2017-11-10 02:38, James Almer wrote: > On 11/9/2017 8:58 AM, James Darnley wrote: >> --- >> configure | 2 ++ >> libavcodec/internal.h | 4 +++- >> 2 files changed, 5 insertions(+), 1 deletion(-) >> >> diff --git a/configure b/configure

Re: [FFmpeg-devel] [PATCH 08/11] avcodec/v210enc: add AVX-512 10-bit line pack function

On 2017-11-10 14:32, James Darnley wrote: > I mentioned previously that using ZMM registers will cause the CPU to > reduce its frequency. > > Gramner said on IRC that a user should spend 20-30% of time in > AVX-512/ZMM code for it to be a net gain in speed. > From ffmpeg-devel

Re: [FFmpeg-devel] [PATCH] avfilter/vf_threshold: add x86 SIMD

On 2017-11-12 21:15, Rostislav Pehlivanov wrote: > On 12 November 2017 at 19:15, Paul B Mahol wrote: > +movam7, [pb_128] >> +addinq, wq >> +add thresholdq, wq >> +add minq, wq >> +add maxq, wq >> +add outq, wq >> +neg wq >> +.ne

Re: [FFmpeg-devel] [PATCH 08/11] avcodec/v210enc: add AVX-512 10-bit line pack function

On 2017-11-10 22:13, James Darnley wrote: > The IRC log should appear at the link below. >> https://lists.ffmpeg.org/pipermail/ffmpeg-devel-irc/2017-November/004651.html Of course when I try to predict what number an email will get based on the past few it ends up being out of order. T

[FFmpeg-devel] [PATCH] configure: add audio_frame_queue dependency for aptx codec

--- configure | 2 ++ 1 file changed, 2 insertions(+) diff --git a/configure b/configure index 8b7b7e164b..48761934be 100755 --- a/configure +++ b/configure @@ -2439,6 +2439,8 @@ amv_encoder_select="aandcttables jpegtables mpegvideoenc" ape_decoder_select="bswapdsp llauddsp" apng_decoder_select

Re: [FFmpeg-devel] [PATCH 03/11] avutil: detect when AVX-512 is available

On 2017-11-10 03:11, James Almer wrote: > On 11/9/2017 8:58 AM, James Darnley wrote: >> @@ -154,6 +155,13 @@ int ff_get_cpu_flags_x86(void) >> if (ebx & 0x0100) >> rval |= AV_CPU_FLAG_BMI2; >> } >> +#if HAVE_AVX512 /*

[FFmpeg-devel] [PATCH 5/8] lavc/x86/flac_dsp_gpl: cosmetic whitespace alignment

--- libavcodec/x86/flac_dsp_gpl.asm | 40 1 file changed, 20 insertions(+), 20 deletions(-) diff --git a/libavcodec/x86/flac_dsp_gpl.asm b/libavcodec/x86/flac_dsp_gpl.asm index 4d212ed212..952fc8b86b 100644 --- a/libavcodec/x86/flac_dsp_gpl.asm +++ b/libav

[FFmpeg-devel] [PATCH 0/8] left-overs of an ancient patch set for the flac encoder

benchmarking I originally did a little less useful because both types of the lpc coder are used for both sample depths (16 and 24). That does make the 32-bit version more useful though because it gets used with 16-bit samples when the intermediates overflow 32 bits. James Darnley (8): avcodec/flac

[FFmpeg-devel] [PATCH 4/8] avcodec/flac: partially unroll loop in flac_enc_lpc_32

Now does 6 samples per iteration, up from 2. From 1.6 to 2.1 times faster again. 2.5 to 3.9 times faster overall. Runtime is reduced by a further 4 to 17%. Reduced by 9 to 65% overall. Same conditions as previously. --- libavcodec/x86/flac_dsp_gpl.asm | 30 +- 1 fil

[FFmpeg-devel] [PATCH 2/8] avcodec/flac: add AVX2 version of the 16-bit LPC encoder

When compared to the SSE4 version, runtime is reduced by 0.5 to 20%. After a bug fix log, long ago in e609cfd697 the 16-bit lpc encoder is used so little that the runtime reduction is no longer correct. The function itself is around 2 times faster. (As one might expect for doing twice as many sam

[FFmpeg-devel] [PATCH 8/8] checkasm: add tests for flacenc lpc coder

--- tests/checkasm/flacdsp.c | 72 1 file changed, 72 insertions(+) diff --git a/tests/checkasm/flacdsp.c b/tests/checkasm/flacdsp.c index dccb54d672..08e5e264ea 100644 --- a/tests/checkasm/flacdsp.c +++ b/tests/checkasm/flacdsp.c @@ -20,13 +20,16

[FFmpeg-devel] [PATCH 3/8] avcodec/flac: add SSE4.2 version of the 32-bit lpc encoder

From 1.3 to 2.5 times faster. Runtime reduced by 4 to 58%. As with the 16-bit version the speed-up generally increases with compression_level. Also like the 16-bit version, it is not used with levels less than 3. After this bug fix in long, long ago in e609cfd697 this 32-bit lpc encoder is heav

[FFmpeg-devel] [PATCH 1/8] avcodec/flac: document limitations of the LPC encoder

State that the maximum value of order is 32. This limit is used in both C and x86 assebly code. --- libavcodec/flacdsp.h | 8 1 file changed, 8 insertions(+) diff --git a/libavcodec/flacdsp.h b/libavcodec/flacdsp.h index 7bb0dd0e9a..90fd3f04b5 100644 --- a/libavcodec/flacdsp.h +++ b/lib

[FFmpeg-devel] [PATCH 6/8] lavc/x86/flac_dsp_gpl: partially unroll 32-bit LPC encoder

Around 1.1 times faster and reduces runtime by up to 6%. --- libavcodec/x86/flac_dsp_gpl.asm | 91 - 1 file changed, 72 insertions(+), 19 deletions(-) diff --git a/libavcodec/x86/flac_dsp_gpl.asm b/libavcodec/x86/flac_dsp_gpl.asm index 952fc8b86b..91989ce56

[FFmpeg-devel] [PATCH 7/8] lavc/flacenc: add AVX2 version of the 32-bit LPC encoder

When compared to the SSE4.2 version runtime, is reduced by 1 to 26%. The function itself is around 2 times faster. --- libavcodec/x86/flac_dsp_gpl.asm | 56 +++-- libavcodec/x86/flacdsp_init.c | 5 +++- 2 files changed, 47 insertions(+), 14 deletions(-) dif

Re: [FFmpeg-devel] [PATCH 7/8] lavc/flacenc: add AVX2 version of the 32-bit LPC encoder

On 2017-11-27 00:13, Rostislav Pehlivanov wrote: > On 26 November 2017 at 22:51, James Darnley wrote: >> @@ -123,7 +123,10 @@ RET >> %endmacro >> >> %macro PMINSQ 3 >> -pcmpgtq %3, %2, %1 >> +mova%3, %2 >> +; We cannot use the

Re: [FFmpeg-devel] [PATCH 6/8] lavc/x86/flac_dsp_gpl: partially unroll 32-bit LPC encoder

On 2017-11-27 00:17, Rostislav Pehlivanov wrote: > On 26 November 2017 at 22:51, James Darnley wrote: >> @@ -152,13 +152,13 @@ RET >> %macro FUNCTION_BODY_32 0 >> >> %if ARCH_X86_64 >> -cglobal flac_enc_lpc_32, 5, 7, 8, mmsize, res, smp, len, order, coefs &

Re: [FFmpeg-devel] avutil/x86util : add macro for 128 bits constant load

On 2017-11-27 20:19, Martin Vignali wrote: > +%macro VBROADCASTI128 2 ; dst xmm/ymm, src : 128bits val > +%if mmsize == 32 > +vbroadcasti128 %1, %2 > +%else > +mova %1, %2 > +%endif > +%endmacro If the condition was made "mmsize > 16" would this work correctly for zmm registers?

Re: [FFmpeg-devel] [PATCH 7/8] lavc/flacenc: add AVX2 version of the 32-bit LPC encoder

On 2017-11-27 17:50, Henrik Gramner wrote: > On Sun, Nov 26, 2017 at 11:51 PM, James Darnley > wrote: >> -pd_0_int_min: times 2 dd 0, -2147483648 >> -pq_int_min: times 2 dq -2147483648 >> -pq_int_max: times 2 dq 2147483647 >> +pd_0_int_min: times 4 dd

Re: [FFmpeg-devel] avfilter/x86/vf_threshold : add SSE4 and AVX2 for threshold 16

On 2017-12-03 19:30, Martin Vignali wrote: > libavfilter/x86/vf_threshold.asm| 19 ++- > libavfilter/x86/vf_threshold_init.c | 34 -- > 2 files changed, 34 insertions(+), 19 deletions(-) > > diff --git a/libavfilter/x86/vf_threshold.asm > b/lib

[FFmpeg-devel] [PATCH 6/7] x86inc: AVX-512 support

From: Henrik Gramner AVX-512 consists of a plethora of different extensions, but in order to keep things a bit more manageable we group together the following extensions under a single baseline cpu flag which should cover SKL-X and future CPUs: * AVX-512 Foundation (F) * AVX-512 Conflict Detect

[FFmpeg-devel] [PATCH 3/7] avutil: detect when AVX-512 is available

--- libavutil/x86/cpu.c | 12 ++-- 1 file changed, 10 insertions(+), 2 deletions(-) diff --git a/libavutil/x86/cpu.c b/libavutil/x86/cpu.c index f33088c8c7..696f47b3bf 100644 --- a/libavutil/x86/cpu.c +++ b/libavutil/x86/cpu.c @@ -97,6 +97,7 @@ int ff_get_cpu_flags_x86(void) int max_

[FFmpeg-devel] [PATCH 5/7] avcodec: add stride alignment needed for AVX-512

--- configure | 2 ++ libavcodec/internal.h | 4 +++- 2 files changed, 5 insertions(+), 1 deletion(-) diff --git a/configure b/configure index 07fb825f91..d3187d71ed 100755 --- a/configure +++ b/configure @@ -1892,6 +1892,7 @@ ARCH_FEATURES=" local_aligned simd_align_16

[FFmpeg-devel] [PATCH 1/7] configure: test whether x86 assembler supports AVX-512

--- configure | 5 + 1 file changed, 5 insertions(+) diff --git a/configure b/configure index d09eec4155..07fb825f91 100755 --- a/configure +++ b/configure @@ -411,6 +411,7 @@ Optimization options (experts only): --disable-fma3 disable FMA3 optimizations --disable-fma4

[FFmpeg-devel] [PATCH 7/7] checkasm: support for AVX-512 functions

--- tests/checkasm/checkasm.c | 1 + 1 file changed, 1 insertion(+) diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c index 45a70aa87f..ff0ca5b68d 100644 --- a/tests/checkasm/checkasm.c +++ b/tests/checkasm/checkasm.c @@ -204,6 +204,7 @@ static const struct { { "FMA3", "

[FFmpeg-devel] [PATCH 0/7] AVX-512 support (v.3)

I have addressed all the comments raised in the previous threads. While some patches were okayed last time I am still sending them as part of these to give everyone a final change to see them again and to object if they wish. Henrik Gramner (1): x86inc: AVX-512 support James Darnley (6

[FFmpeg-devel] [PATCH 2/7] avutil: add AVX-512 flags

--- Changelog | 1 + doc/APIchanges| 3 +++ libavutil/cpu.c | 6 +- libavutil/cpu.h | 1 + libavutil/tests/cpu.c | 1 + libavutil/version.h | 2 +- libavutil/x86/cpu.h | 2 ++ 7 files changed, 14 insertions(+), 2 deletions(-) diff --git a/Changelog b/Change

[FFmpeg-devel] [PATCH 4/7] avutil: add alignment needed for AVX-512

--- libavutil/mem.c | 2 +- libavutil/x86/cpu.c | 2 ++ 2 files changed, 3 insertions(+), 1 deletion(-) diff --git a/libavutil/mem.c b/libavutil/mem.c index 6ad409daf4..79e8b597f1 100644 --- a/libavutil/mem.c +++ b/libavutil/mem.c @@ -61,7 +61,7 @@ void free(void *ptr); #include "mem_inte

Re: [FFmpeg-devel] [PATCH 0/7] AVX-512 support (v.3)

On 2017-12-21 15:06, Carl Eugen Hoyos wrote: > 2017-12-21 14:40 GMT+01:00 James Darnley : >> I have addressed all the comments raised in the previous threads. >> While some patches were okayed last time I am still sending them >> as part of these to give everyone a final cha

<    1   2   3   4   5   6   >