Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

AVX512_changes #478

Open
wants to merge 13 commits into
base: master
Choose a base branch
from
Open

AVX512_changes #478

wants to merge 13 commits into from

Conversation

deeptiag1
Copy link

@deeptiag1 deeptiag1 commented Feb 19, 2020 •
edited by tianjunwork

Signed-off-by: deeptiag1 deepti.aggarwal@intel.com

AVX512_changes
2c2c7e4 
Signed-off-by: deeptiag1 <deepti.aggarwal@intel.comm>
@1480c1
Copy link
Member

1480c1 commented Feb 19, 2020

@deeptiag1, you may want to check your git config, it seems you might have misspelled your email, unless your email domain is intel.comm

@deeptiag1
AVX512_changes_P2
157cbe8 
Signed-off-by: deeptiag1 <deepti.aggarwal@intel.com>
@deeptiag1
AVX512_changes_P3
4093056 
Signed-off-by: deeptiag1 <deepti.aggarwal@intel.com>
@tianjunwork
Copy link
Contributor

tianjunwork commented Feb 19, 2020 •
edited

Hi @deeptiag1 , since AVX512 code doesn't execute when running encoder, our CI can't validate your code. Could you let us know what tests you did and the result? Thank you very much for your contribution.
Could you also give a better title to this PR which be more specific. Thank you.

@tianjunwork tianjunwork added the enhancement New feature or request label Feb 19, 2020
@deeptiag1
AVX512_changes_P4
b00574f 
Signed-off-by: deeptiag1 <deepti.aggarwal@intel.com>
@deeptiag1
Copy link
Author

Hi @tianjunwork, Enable the AVX512 flag in Ebdefinitions and then run the same test as one would run for any build. Let me know if all test pass.

@deeptiag1
AVX512_changes_P4
b827613 
Signed-off-by: deeptiag1 <deepti.aggarwal@intel.com>
@deeptiag1
AVX512_changes_P4
d14b735 
Signed-off-by: deeptiag1 <deepti.aggarwal@intel.com>
@deeptiag1
AVX512_changes_P4
0bc4093 
Signed-off-by: deeptiag1 <deepti.aggarwal@intel.com>
@deeptiag1
AVX512_changes_Revert
67ec8bf 
Signed-off-by: deeptiag1 <deepti.aggarwal@intel.com>
hassount
hassount reviewed Mar 4, 2020
View reviewed changes
Source/Lib/ASM_AVX2/EbTransforms_Intrinsic_AVX512.c Outdated
@@ -0,0 +1,195 @@
#include "EbTransforms_AVX2.h"
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

License header missing

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed

hassount
hassount reviewed Mar 4, 2020
View reviewed changes
Source/Lib/ASM_SSSE3/EbMcp_Intrinsic_AVX512.c Outdated
@@ -0,0 +1,457 @@
#include "EbMcp_SSSE3.h"
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

missing license header

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

the AVX512 code should be moved to the AVX2 project

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed

hassount
hassount reviewed Mar 4, 2020
View reviewed changes
Source/Lib/ASM_AVX2/EbTransforms_Intrinsic_AVX512.c Outdated
54, 67, -31, -73, 4, 78, 22, -82, 54, 67, -31, -73, 4, 78, 22, -82, -46, 85, 67, -88, -82, 90, 90, -90, -46, 85, 67, -88, -82, 90, 90, -90
};

extern void EbHevcTransform32_AVX512_INTRIN(EB_S16 *src, EB_U32 src_stride, EB_S16 *dst, EB_U32 dst_stride, EB_U32 shift)
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This would be more appropriate being moved to the AVX2 project to keep all AVX512 code in the same location.

@deeptiag1
changes_incorpoarted
d92a7c4 
Signed-off-by: deeptiag1 <deepti.aggarwal@intel.com>
@intelmark
Copy link
Contributor

intelmark commented Mar 27, 2020 •
edited

From @hassount 3/17/2020:

From the results below, it seems that we need to disable the optimization for the LumaInterpolationFilterOneDOutRawHorizontal_AVX512 kernel and make sure that the documentation contains a clear reference to the gcc version requirement when vnni is on, then I don’t see any blockers to this PR moving forward assuming functional testing is passing._

The current PR shows:

#ifdef VNNI_SUPPORT
#ifndef NON_AVX512_SUPPORT
#define LumaInterpolationFilterOneDOutRawHorizontalLumaInterpolationFilterOneDOutRawHorizontal_AVX512
#else
#define LumaInterpolationFilterOneDOutRawHorizontalLumaInterpolationFilterOneDOutRawHorizontal_SSSE3
#endif

But isn't VNNI a subset of AVX512 - Are all combinations of the VNNI_SUPPORT and AVX512_SUPPORT defines possible?

Also there doesn't seem to be a mention of the gcc requirements for compiling the VNNI specific code anywhere in the PR

@deeptiag1
avx512-changes_v3
d983a3b 
Signed-off-by: deeptiag1 <deepti.aggarwal@intel.com>
@deeptiag1
avx512-changes_v3
06ec13d 
Signed-off-by: deeptiag1 <deepti.aggarwal@intel.com>
intelmark
intelmark reviewed Apr 1, 2020
View reviewed changes
README.md Outdated
@@ -69,6 +69,7 @@ In order to run the highest resolution supported by the encoder, at least 64GB o
- Download the yasm exe from the following [link](http://www.tortall.net/projects/yasm/releases/yasm-1.3.0-win64.exe)
- Rename yasm-1.3.0-win64.exe to yasm.exe
- Copy yasm.exe into a location that is in the PATH environment variable
- Vnni requires gcc version greater then 9.2.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

  1. greater than
  2. On my server it seems to be working with version: gcc version 9.2.1

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed

@deeptiag1
avx512-changes_v3
0f5ce17 
Signed-off-by: deeptiag1 <deepti.aggarwal@intel.com>
@deeptiag1
avx512-changes_v3
a23033f 
Signed-off-by: deeptiag1 <deepti.aggarwal@intel.com>
@deeptiag1
Copy link
Author

#define vnni in EbDefinitions enable vnni code committed in this PR.

Perf Data;
Encoder mode :0
reference_avx512 : branch vnni_changes_1 @ d92a7c4
vnni_avx512: reference_avx512 + VNNI_SUPPORT + LumaInterpolationFilterOneDOutRawHorizontal_SSSE3
speed
reference_avx512 | 0.95fps
vnni_avx512 | 0.96fps
Speed gain: 1%

Encoder mode :0
reference_avx2 : branch vnni_changes_1 @ d92a7c4 + NON_AVX512_SUPPORT
vnni_avx2: reference_avx2 + VNNI_SUPPORT + LumaInterpolationFilterOneDOutRawHorizontal_SSSE3
speed
reference_avx2 | 0.96fps
vnni_avx2 | 0.97fps
Speed gain: 1%

@intelmark
Copy link
Contributor

@deeptiag1 Regarding the performance data - great to see the speed up, but what size video and is there a command line that was used?

inteltiger
inteltiger reviewed Aug 6, 2021
View reviewed changes
Source/Lib/ASM_SSSE3/EbMcp_Intrinsic_AVX512VNNI.c
};

#ifndef NON_AVX512_SUPPORT
void LumaInterpolationFilterOneDOutRawHorizontal_AVX512(
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why no code is calling LumaInterpolationFilterOneDOutRawHorizontal_AVX512 ?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@hassount hassount hassount left review comments

@intelmark intelmark intelmark left review comments

@inteltiger inteltiger inteltiger left review comments

Assignees
No one assigned
Labels
enhancement New feature or request
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

None yet
6 participants
@deeptiag1 @1480c1 @tianjunwork @intelmark @hassount @inteltiger