Skip to content

L
libyuv
Switch branch/tag
  1. 02 Feb, 2018 2 commits
    • Frank Barchard's avatar
      I422ToUYVYRow_AVX2 use vpmovzxbd instead of vpermq · 5790a765
      Frank Barchard authored 
      I422ToUYVYRow_AVX2 optimized from 7 cycles per 32 pixels to 4.6 cycles.
Instead of 2 vpermq and vpunpcklbw:
vmovdqu    (%1),%%xmm2
vmovdqu    0x00(%1,%2,1),%%xmm3
vpermq     $0xd8,%%ymm2,%%ymm2
vpermq     $0xd8,%%ymm3,%%ymm3
vpunpcklbw %%ymm3,%%ymm2,%%ymm2

..use vpmovzxbd to expand the bytes to shorts, then vpslld and vpor
vpmovzxbd  (%1),%%ymm2
vpmovzxbd  0x00(%1,%2,1),%%ymm3
vpslld     $0x10,%%ymm3,%%ymm3
vpor       %%ymm3,%%ymm2,%%ymm2
which reduces the port 5 bottleneck by 1 cycle.

Bug: libyuv:556
Test: out/Release/libyuv_unittest --gtest_filter=*I42?To*UY*Opt

Change-Id: I53799e53cc6b090a1a695c839094c193be3eecaf
Reviewed-on: https://chromium-review.googlesource.com/899873
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
Reviewed-by: 's avatarrichard winterton <rrwinterton@gmail.com>
Reviewed-by: 's avatarCheng Wang <wangcheng@google.com>
      5790a765
    • Frank Barchard's avatar
      I422ToYUY2Row_AVX2 use vpmovzxbd instead of vpermq · 7ff53f32
      Frank Barchard authored 
      I422ToYUY2Row_AVX2 optimized from 7 cycles per 32 pixels to 6 cycles.
Instead of 2 vpermq and vpunpcklbw:
vmovdqu    (%1),%%xmm2
vmovdqu    0x00(%1,%2,1),%%xmm3
lea        0x10(%1),%1
vpermq     $0xd8,%%ymm2,%%ymm2
vpermq     $0xd8,%%ymm3,%%ymm3
vpunpcklbw %%ymm3,%%ymm2,%%ymm2

..use vpmovzxbd to expand the bytes to shorts, then vpslld and vpor
vpmovzxbd  (%1),%%ymm2
vpmovzxbd  0x00(%1,%2,1),%%ymm3
vpslld     $0x10,%%ymm3,%%ymm3
vpor       %%ymm3,%%ymm2,%%ymm2
which reduces the port 5 bottleneck by 1 cycle.

Bug: libyuv:556
Test: out/Release/libyuv_unittest --gtest_filter=*I42?To*UY*Opt

I422ToYUY2Row_AVX2 optimization

Improve performance of AVX2 code by avoiding vpermq

Bug: libyuv:556
Test: /usr/local/google/home/fbarchard/iaca-lin64/bin/iaca.sh -reduceout -arch BDW out/Release/obj/libyuv_internal/row_gcc.o
Change-Id: Ie36732da23ecea1ffcc6b297bacc962780b59ef1
Reviewed-on: https://chromium-review.googlesource.com/898067
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
Reviewed-by: 's avatarrichard winterton <rrwinterton@gmail.com>
      7ff53f32
  3. 01 Feb, 2018 1 commit
    • Frank Barchard's avatar
      I420ToYUY2_AVX2 port · 664c7356
      Frank Barchard authored 
      I420 and I422 To YUY2 and UYVY ported from SSE2 to AVX2.

Was SSE2
I420ToYUY2_Opt (135 ms)
I420ToUYVY_Opt (148 ms)
I422ToYUY2_Opt (145 ms)
I422ToUYVY_Opt (142 ms)

Now AVX2
I420ToYUY2_Opt (133 ms)
I420ToUYVY_Opt (130 ms)
I422ToYUY2_Opt (127 ms)
I422ToUYVY_Opt (137 ms)

Bug: libyuv:556
Test: out/Release/libyuv_unittest --sandbox_unittests --gtest_filter=*I42?To*UY*Opt
Change-Id: Ic35f97cee02dc009fd98785589ba17c7cf50bb35
Reviewed-on: https://chromium-review.googlesource.com/892493
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
Reviewed-by: 's avatarrichard winterton <rrwinterton@gmail.com>
      664c7356
  5. 29 Jan, 2018 2 commits
  7. 27 Jan, 2018 1 commit
  9. 26 Jan, 2018 1 commit
  11. 25 Jan, 2018 1 commit
  13. 24 Jan, 2018 2 commits
  15. 23 Jan, 2018 4 commits
  17. 19 Jan, 2018 2 commits
    • Frank Barchard's avatar
      H010ToAR30 in 1 step with SSSE3 assembly · 09db0c4c
      Frank Barchard authored 
      Switch YUV conversion macro to output 16 bits per channel.
STOREAR30 macro to output AR30.

[ RUN      ] LibYUVConvertTest.TestH420ToARGB
uniques: B 220, G, 220, R 220
[       OK ] LibYUVConvertTest.TestH420ToARGB (0 ms)
[ RUN      ] LibYUVConvertTest.TestH010ToARGB
uniques: B 256, G, 256, R 256
[       OK ] LibYUVConvertTest.TestH010ToARGB (0 ms)
[ RUN      ] LibYUVConvertTest.TestH010ToAR30
uniques: B 883, G, 883, R 883
[       OK ] LibYUVConvertTest.TestH010ToAR30 (0 ms)

Bug: libyuv:751
Test: LibYUVConvertTest.H010ToAR30_Opt
Change-Id: I902b718e2c8b68ede69625ccafebc6519d5af70d
Reviewed-on: https://chromium-review.googlesource.com/869511Reviewed-by: 's avatarFrank Barchard <fbarchard@chromium.org>
Reviewed-by: 's avatarMiguel Casas <mcasas@chromium.org>
Reviewed-by: 's avatarrichard winterton <rrwinterton@gmail.com>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
      09db0c4c
    • Frank Barchard's avatar
      Add LibYUVConvertTest.TestH010ToAR30 unittest · 37f97210
      Frank Barchard authored 
      Tests accuracy of H010ToAR30 on grey scale ramp against float
and computes a histogram to detect number of unique shades for
each channel.

With 2 step intermediate using 8 bit RGB, the test shows 256
unique values.

[ RUN      ] LibYUVConvertTest.TestH420ToARGB
uniques: B 220, G, 220, R 220
[       OK ] LibYUVConvertTest.TestH420ToARGB (0 ms)
[ RUN      ] LibYUVConvertTest.TestH010ToARGB
uniques: B 256, G, 256, R 256
[       OK ] LibYUVConvertTest.TestH010ToARGB (0 ms)
[ RUN      ] LibYUVConvertTest.TestH010ToAR30
uniques: B 256, G, 256, R 256
[       OK ] LibYUVConvertTest.TestH010ToAR30 (0 ms)

Bug: libyuv:751
Test  LibYUVConvertTest.TestH010ToAR30 unittest

Change-Id: I6b1e1209247cb00b79b594127b02dae5217dc400
Reviewed-on: https://chromium-review.googlesource.com/875317Reviewed-by: 's avatarMiguel Casas <mcasas@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
      37f97210
  19. 16 Jan, 2018 1 commit
    • Frank Barchard's avatar
      Remove MEMOPREG x64 NaCL macros · ecab5430
      Frank Barchard authored 
      MEMOPREG macros are deprecated in row.h

Regular expressions to remove MEMOPREG macros:

MEMOPREG(movd, 0x00, [u_buf], [v_buf], 1, xmm1)                            \
MEMOPREG\((.*), (.*), (.*), (.*), (.*), (.*)\)
"\1    \2(%\3,%\4,\5),%%\6            \\n"

MEMOPREG(movdqu,0x00,1,4,1,xmm2)
MEMOPREG\((.*),(.*),(.*),(.*),(.*),(.*)\)
"\1    \2(%\3,%\4,\5),%%\6            \\n"

MEMOPREG(movdqu,0x00,1,4,1,xmm2)
MEMOPREG\((.*),(.*),(.*),(.*),(.*),(.*)\)(.*)(//.*)
"\1    \2(%\3,%\4,\5),%%\6           \\n"

TBR=braveyao@chromium.org

Bug: libyuv:702
Test: try bots pass
Change-Id: If8743abd9af2e8c549d0c7d3d49733a9b0f0ca86
Reviewed-on: https://chromium-review.googlesource.com/865964Reviewed-by: 's avatarFrank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
      ecab5430
  21. 13 Jan, 2018 1 commit
    • Frank Barchard's avatar
      Remove MEMOPMEM x64 NaCL macros · b33e0f97
      Frank Barchard authored 
      MEMOPMEM macros are deprecated in row.h

Usage examples
    MEMOPMEM(vmovdqu,ymm0,0x00,0,1,1)          //  vmovdqu %%ymm0,(%0,%1)
    MEMOPMEM(movdqu,xmm2,0x00,1,0,1)

Regular expressions to remove MEMACCESS macros:

MEMOPMEM\((.*),(.*),(.*),(.*),(.*),(.*)\)(.*)(//.*)
"\1    %%\2,\3(%\4,%\5,\6)\7 \\n"

MEMOPMEM\((.*),(.*),(.*),(.*),(.*),(.*)\)
"\1    %%\2,\3(%\4,%\5,\6)            \\n"

TBR=braveyao@chromium.org
Bug: libyuv:702
Test: try bots pass
Change-Id: Id8c6963d544d16e39bb6a9a0536babfb7f554b3a
Reviewed-on: https://chromium-review.googlesource.com/865934Reviewed-by: 's avatarFrank Barchard <fbarchard@chromium.org>
      b33e0f97
  23. 12 Jan, 2018 4 commits
    • Frank Barchard's avatar
      Remove VMEMOPREG x64 NaCL macros · a875ed17
      Frank Barchard authored 
      VMEMOPREG macros are deprecated in row.h

Usage examples
    VMEMOPREG(vpavgb,0x00,0,4,1,ymm0,ymm0)     // vpavgb (%0,%4,1),%%ymm0,%%ymm0
    VMEMOPREG(vpavgb,0x20,0,4,1,ymm1,ymm1)

Regular expressions to remove MEMACCESS macros:

VMEMOPREG\((.*),(.*),(.*),(.*),(.*),(.*),(.*)\)(.*)(//.*)
"\1    \2(%\3,%\4,\5),%%\6,%%\7      \\n"

VMEMOPREG\((.*),(.*),(.*),(.*),(.*),(.*),(.*)\)
"\1    \2(%\3,%\4,\5),%%\6,%%\7            \\n"

TBR=braveyao@chromium.org

Bug: libyuv:702
Test: try bots pass
Change-Id: I472446606f7fd568fdf33aaacc22d5ed78673dab
Reviewed-on: https://chromium-review.googlesource.com/865640Reviewed-by: 's avatarFrank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
      a875ed17
    • Frank Barchard's avatar
      Remove VEXTOPMEM x64 NaCL macros · 030042a2
      Frank Barchard authored 
      VEXTOPMEM macros are deprecated in row.h

Usage examples
    VEXTOPMEM(vextractf128,1,ymm0,0x0,1,2,1) // vextractf128 $1,%%ymm0,(%1,%2,1)

Regular expressions to remove MEMACCESS macros:

VEXTOPMEM\((.*),(.*),(.*),(.*),(.*),(.*),(.*)\)(.*//.*)
"\1 $\2,%\3,\4(%\5,%\6,\7)        \\n"

Bug: libyuv:702
Test: try bots pass
Change-Id: I177edf9813128408e74816672dd25abb03a5e1ca
Reviewed-on: https://chromium-review.googlesource.com/865283Reviewed-by: 's avatarFrank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
      030042a2
    • Frank Barchard's avatar
      Remove MEMACCESS x64 NaCL macros · 5088f001
      Frank Barchard authored 
      MEMACCESS macros are deprecated in row.h

Usage examples
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"

Regular expressions to remove MEMACCESS macros:

" MEMACCESS2\((.*),(.*)\) "(.*)\\n"
\1(%\2)\3              \\n"

" MEMACCESS\((.*)\) "(.*)\\n"
(%\1)\2            \\n"

Bug: libyuv:702
Test: try bots pass
Change-Id: I42f62d5dede8ef2ea643e78c204371a7659d25e6
Reviewed-on: https://chromium-review.googlesource.com/862803Reviewed-by: 's avatarFrank Barchard <fbarchard@chromium.org>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
      5088f001
    • Frank Barchard's avatar
      Remove MEMOPARG x64 NaCL macros · e3797d17
      Frank Barchard authored 
      MEMOPARG macros are deprecated in row.h

  #opcode " " #offset "(%" #base ",%" #index "," #scale "),%" #arg "\n"

Usage examples
    MEMOPARG(movzwl,0x00,1,3,1,k2)             //  movzwl  (%1,%3,1),%k2

Regular expression to remove MEMACCESS macro:

MEMOPARG\((.*),(.*),(.*),(.*),(.*),(.*)\)(.*//.*)
"\1    \2(%\3,%\4,\5),%\6                \\n"

Bug: libyuv:702
Test: try bots pass
Change-Id: I4a5ad2abf5017e651576f4c8c784be1c8dbf5a83
Reviewed-on: https://chromium-review.googlesource.com/863108Reviewed-by: 's avatarFrank Barchard <fbarchard@chromium.org>
      e3797d17
  25. 10 Jan, 2018 4 commits
  27. 09 Jan, 2018 1 commit
  29. 08 Jan, 2018 2 commits
  31. 07 Jan, 2018 1 commit
    • Frank Barchard's avatar
      H010ToAR30 optimized to 2 step conversion · 9d2cd6a3
      Frank Barchard authored 
      Previously H010ToAR30 was done in a 3 step conversion:
H010ToH420, H420ToARGB, ARGBToAR30.
This CL merges the first 2 steps into H010ToARGB, to
improve performance.
Caveat - only 10 bit YUV is supported at this time.
Previously the low level code supported different numbers
of bits - 9, 10, 12 or 16.

Was 3 step conversion:
LibYUVConvertTest.H010ToAR30_Any (1263 ms)
LibYUVConvertTest.H010ToAR30_Unaligned (951 ms)
LibYUVConvertTest.H010ToAR30_Invert (913 ms)
LibYUVConvertTest.H010ToAR30_Opt (901 ms)

Now 2 step conversion:
LibYUVConvertTest.H010ToAR30_Any (853 ms)
LibYUVConvertTest.H010ToAR30_Unaligned (811 ms)
LibYUVConvertTest.H010ToAR30_Invert (781 ms)
LibYUVConvertTest.H010ToAR30_Opt (755 ms)

Bug: libyuv:751
Test: LibYUVConvertTest.H010ToAR30_Opt
Change-Id: Ica7574040401cd57145a4827acdf3c0e58346a2a
Reviewed-on: https://chromium-review.googlesource.com/853288Reviewed-by: 's avatarFrank Barchard <fbarchard@chromium.org>
Reviewed-by: 's avatarMiguel Casas <mcasas@chromium.org>
      9d2cd6a3
  33. 05 Jan, 2018 2 commits
  35. 04 Jan, 2018 1 commit
  37. 03 Jan, 2018 1 commit
  39. 02 Jan, 2018 4 commits
  41. 28 Dec, 2017 1 commit
  43. 21 Dec, 2017 1 commit