[FFmpeg-devel] [PATCH] libavcodec/h264pred: Remove pred8x8_horizontal_8_mmxext

[FFmpeg-devel] [PATCH] libavcodec/h264pred: Remove pred8x8_horizontal_8_mmxext

[Kieran Kunhya]

[-- Attachment #1: Type: text/plain, Size: 203 bytes --]

$subj

Old:
pred8x8_horizontal_8_c: 6.8
pred8x8_horizontal_8_mmxext: 8.6
pred8x8_horizontal_8_ssse3: 4.8

New:
pred8x8_horizontal_8_c: 9.2
pred8x8_horizontal_8_sse2: 12.2
pred8x8_horizontal_8_ssse3: 4.9

[-- Attachment #2: 0001-libavcodec-h264pred-Remove-pred8x8_horizontal_8_mmxe.patch --]
[-- Type: application/octet-stream, Size: 2982 bytes --]

[-- 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".

Re: [FFmpeg-devel] [PATCH] libavcodec/h264pred: Remove pred8x8_horizontal_8_mmxext

[Andreas Rheinhardt]

Kieran Kunhya:
> $subj
> 
> Old:
> pred8x8_horizontal_8_c: 6.8
> pred8x8_horizontal_8_mmxext: 8.6
> pred8x8_horizontal_8_ssse3: 4.8
> 
> New:
> pred8x8_horizontal_8_c: 9.2
> pred8x8_horizontal_8_sse2: 12.2
> pred8x8_horizontal_8_ssse3: 4.9
> 

You do realize that the SSE2 version is worse than the mmxext version?
In fact, worse than the C version. Given that both the mmxext and sse2
versions are worse than C, there is no point in having the mmxext one at
all and no need for an SSE2 replacement.

- 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] libavcodec/h264pred: Remove pred8x8_horizontal_8_mmxext

[James Almer]

On 3/2/2024 6:20 PM, Andreas Rheinhardt wrote:
> Kieran Kunhya:
>> $subj
>>
>> Old:
>> pred8x8_horizontal_8_c: 6.8
>> pred8x8_horizontal_8_mmxext: 8.6
>> pred8x8_horizontal_8_ssse3: 4.8
>>
>> New:
>> pred8x8_horizontal_8_c: 9.2
>> pred8x8_horizontal_8_sse2: 12.2
>> pred8x8_horizontal_8_ssse3: 4.9
>>
> 
> You do realize that the SSE2 version is worse than the mmxext version?
> In fact, worse than the C version. Given that both the mmxext and sse2
> versions are worse than C, there is no point in having the mmxext one at
> all and no need for an SSE2 replacement.

The SSSE3 version needs to be updated to use xmm registers either way.

> 
> - 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".

Re: [FFmpeg-devel] [PATCH] libavcodec/h264pred: Remove pred8x8_horizontal_8_mmxext

[Kieran Kunhya]

On Sat, 2 Mar 2024 at 21:18, Andreas Rheinhardt <
andreas.rheinhardt@outlook.com> wrote:

> Kieran Kunhya:
> > $subj
> >
> > Old:
> > pred8x8_horizontal_8_c: 6.8
> > pred8x8_horizontal_8_mmxext: 8.6
> > pred8x8_horizontal_8_ssse3: 4.8
> >
> > New:
> > pred8x8_horizontal_8_c: 9.2
> > pred8x8_horizontal_8_sse2: 12.2
> > pred8x8_horizontal_8_ssse3: 4.9
> >
>
> You do realize that the SSE2 version is worse than the mmxext version?
> In fact, worse than the C version. Given that both the mmxext and sse2
> versions are worse than C, there is no point in having the mmxext one at
> all and no need for an SSE2 replacement.
>
> - Andreas
>

On my machine, others may have different speeds.

Kieran
_______________________________________________
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] libavcodec/h264pred: Remove pred8x8_horizontal_8_mmxext

[Henrik Gramner via ffmpeg-devel]

On Sat, Mar 2, 2024 at 10:13 PM Kieran Kunhya <kierank@obe.tv> wrote:
>      SPLATB_LOAD m0, r0+r1*0-1, m2
>      SPLATB_LOAD m1, r0+r1*1-1, m2

This adds an extra unnecessary shuffle in the SSE2 code as it splats
to a full register. The easiest way of fixing it would probably be to
unroll the macro and manually get rid of it.

Although on x86-64 it might be faster to do a 1->8 byte splat using a
GPR multiply with 0x0101010101010101.
_______________________________________________
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".