AVX 2 SrcOver blits: color32, blitmask.

AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and unpack data that I think makes things clearer and more orthogonal, and it'll make it easier to try out a pmaddubsw lerp.  I have backported these changes to the SSE 4.1 code, and I hope that I can actually get a lot of this code templated for sharing between the two later.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should all be pixel-exact to the previous SSE 4.1 code.


GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1532613002

Committed: https://skia.googlesource.com/skia/+/5d2117015eb271e09faf4a7ddd89093c9d618a36

TODO:
- x86_64-cros-linux-gnu-g++ does not seem to support AVX2, despite accepting the -mavx2 command line option:
error: '__m256i' does not name a type (and many others)

- we don't appear to be passing /arch:avx2 correctly to MSVC, producing this warning-as-error:
warning C4752: found Intel(R) Advanced Vector Extensions; consider using /arch:AVX

CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot;client.skia.compile:Build-Ubuntu-GCC-x86_64-Release-CMake-Trybot,Build-Mac10.9-Clang-x86_64-Release-CMake-Trybot,Build-Ubuntu-GCC-x86_64-Debug-CrOS_Link-Trybot,Build-Win-MSVC-x86_64-Release-Trybot

[mtklein_C, 2015-12-16 17:02:24]

Description was changed from

==========
AVX 2 SrcOver blits: color32, blitmask.

Follow up to the SSE 4.1 CL.
==========

to

==========
AVX 2 SrcOver blits: color32, blitmask.

Follow up to the SSE 4.1 CL.
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot
==========

[mtklein, 2015-12-16 22:59:57]

The CQ bit was checked by mtklein@google.com to run a CQ dry run

[commit-bot: I haz the power, 2015-12-16 23:00:07]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/80001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/80001

[commit-bot: I haz the power, 2015-12-16 23:01:01]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-16 23:01:01]

Dry run: Try jobs failed on following builders:
  Build-Ubuntu-GCC-x86_64-Release-Trybot on client.skia.compile (JOB_FAILED,
http://build.chromium.org/p/client.skia.compile/builders/Build-Ubuntu-GCC-x86...)

[mtklein, 2015-12-17 16:11:55]

The CQ bit was checked by mtklein@google.com to run a CQ dry run

[commit-bot: I haz the power, 2015-12-17 16:11:59]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/100001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/100001

[mtklein, 2015-12-17 16:33:47]

Description was changed from

==========
AVX 2 SrcOver blits: color32, blitmask.

Follow up to the SSE 4.1 CL.
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot
==========

to

==========
AVX 2 SrcOver blits: color32, blitmask.

Follow up to the SSE 4.1 CL.
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

COMMIT=false
Depends on https://codereview.chromium.org/1535443003
==========

[mtklein, 2015-12-17 16:33:51]

The CQ bit was checked by mtklein@google.com to run a CQ dry run

[commit-bot: I haz the power, 2015-12-17 16:33:56]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/120001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/120001

[mtklein, 2015-12-17 16:38:14]

Description was changed from

==========
AVX 2 SrcOver blits: color32, blitmask.

Follow up to the SSE 4.1 CL.
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

COMMIT=false
Depends on https://codereview.chromium.org/1535443003
==========

to

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I plan to backport these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

COMMIT=false
Depends on https://codereview.chromium.org/1535443003
==========

[mtklein, 2015-12-17 16:46:12]

Description was changed from

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I plan to backport these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

COMMIT=false
Depends on https://codereview.chromium.org/1535443003
==========

to

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I plan to backport these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should be pixel-exact to the SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

COMMIT=false
Depends on https://codereview.chromium.org/1535443003
==========

[commit-bot: I haz the power, 2015-12-17 16:48:27]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-17 16:48:27]

Dry run: This issue passed the CQ dry run.

[mtklein, 2015-12-17 16:54:06]

The CQ bit was checked by mtklein@google.com to run a CQ dry run

[commit-bot: I haz the power, 2015-12-17 16:54:19]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/140001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/140001

[mtklein_C, 2015-12-17 16:55:51]

mtklein@chromium.org changed reviewers:
+ herb@google.com
- mtklein@google.com

[commit-bot: I haz the power, 2015-12-17 17:13:28]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-17 17:13:28]

Dry run: This issue passed the CQ dry run.

[mtklein, 2015-12-17 18:19:21]

Description was changed from

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I plan to backport these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should be pixel-exact to the SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

COMMIT=false
Depends on https://codereview.chromium.org/1535443003
==========

to

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I plan to backport these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should be pixel-exact to the SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot
==========

[mtklein, 2015-12-17 18:19:47]

Description was changed from

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I plan to backport these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should be pixel-exact to the SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot
==========

to

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I plan to backport these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should be pixel-exact to the SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot;client.skia.compile:Build-Ubuntu-GCC-x86_64-Release-CMake-Trybot,Build-Mac10.9-Clang-x86_64-Release-CMake-Trybot
==========

[mtklein, 2015-12-17 18:19:51]

The CQ bit was checked by mtklein@google.com to run a CQ dry run

[commit-bot: I haz the power, 2015-12-17 18:19:57]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/140001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/140001

[commit-bot: I haz the power, 2015-12-17 18:23:44]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-17 18:23:45]

Dry run: This issue passed the CQ dry run.

[herb_g, 2015-12-17 21:37:26]

https://codereview.chromium.org/1532613002/diff/140001/src/opts/SkOpts_avx2.cpp
File src/opts/SkOpts_avx2.cpp (right):

https://codereview.chromium.org/1532613002/diff/140001/src/opts/SkOpts_avx2.c...
src/opts/SkOpts_avx2.cpp:181: s_255_128 = div255_part1(_mm256_mullo_epi16(s,
_mm256_set1_epi16(255))),
Instead of _mm256_mullo_epi16, could we use _mm256_bslli_epi128? (s<<8) - s.
This should be two cycles instead of 5.

https://codereview.chromium.org/1532613002/diff/140001/src/opts/SkOpts_avx2.c...
src/opts/SkOpts_avx2.cpp:205: dst += dstRB / sizeof(*dst);
My paranoid lizard brain says that we should hoist the increments like : dstRB /
sizeof(*dst), just because my lizard brain does not trust the compiler.

[mtklein, 2015-12-18 14:36:06]

https://codereview.chromium.org/1532613002/diff/140001/src/opts/SkOpts_avx2.cpp
File src/opts/SkOpts_avx2.cpp (right):

https://codereview.chromium.org/1532613002/diff/140001/src/opts/SkOpts_avx2.c...
src/opts/SkOpts_avx2.cpp:181: s_255_128 = div255_part1(_mm256_mullo_epi16(s,
_mm256_set1_epi16(255))),
On 2015/12/17 at 21:37:26, herb_g wrote:
> Instead of _mm256_mullo_epi16, could we use _mm256_bslli_epi128? (s<<8) - s.
This should be two cycles instead of 5.

Done.  I admit I didn't pay much attention to things that asymptotically cost 0.

https://codereview.chromium.org/1532613002/diff/140001/src/opts/SkOpts_avx2.c...
src/opts/SkOpts_avx2.cpp:205: dst += dstRB / sizeof(*dst);
Nah, `ptr += k / sizeof(*ptr)` is as cheap as you can get for a pointer
increment in C++.  

Remember, all pointers are really byte pointers, so this compiles into a simple
increment by k, where something like
  int stride = k / sizeof(*ptr);
  ...
  ptr += stride;
is potentially more expensive, as it's got to remultiply stride by sizeof(*ptr)
to get it into bytes.  (Though, in practice I bet it optimizes to the same
code.)  It's also not anywhere near the critical path.

So, agreed, paranoid lizard brain.

[mtklein, 2015-12-18 14:36:12]

The CQ bit was checked by mtklein@google.com to run a CQ dry run

[commit-bot: I haz the power, 2015-12-18 14:36:20]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/160001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/160001

[commit-bot: I haz the power, 2015-12-18 14:50:49]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-18 14:50:52]

Dry run: This issue passed the CQ dry run.

[mtklein_C, 2015-12-30 19:54:12]

Description was changed from

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I plan to backport these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should be pixel-exact to the SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot;client.skia.compile:Build-Ubuntu-GCC-x86_64-Release-CMake-Trybot,Build-Mac10.9-Clang-x86_64-Release-CMake-Trybot
==========

to

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I have backported these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two later.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should be pixel-exact to the SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot;client.skia.compile:Build-Ubuntu-GCC-x86_64-Release-CMake-Trybot,Build-Mac10.9-Clang-x86_64-Release-CMake-Trybot
==========

[mtklein_C, 2015-12-30 19:54:12]

mtklein@chromium.org changed reviewers:
- mtklein@google.com

[mtklein_C, 2015-12-30 20:01:55]

Description was changed from

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I have backported these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two later.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should be pixel-exact to the SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot;client.skia.compile:Build-Ubuntu-GCC-x86_64-Release-CMake-Trybot,Build-Mac10.9-Clang-x86_64-Release-CMake-Trybot
==========

to

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I have backported these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two later.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should all be pixel-exact to the previous SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot,Test-Mac10.8-Clang-MacMini4.1-CPU-SSE4-x86_64-Release-Trybot;client.skia.compile:Build-Ubuntu-GCC-x86_64-Release-CMake-Trybot,Build-Mac10.9-Clang-x86_64-Release-CMake-Trybot
==========

[mtklein_C, 2015-12-30 20:02:01]

The CQ bit was checked by mtklein@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-12-30 20:02:06]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/200001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/200001

[mtklein_C, 2015-12-30 20:09:26]

The CQ bit was checked by mtklein@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-12-30 20:09:37]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/220001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/220001

[commit-bot: I haz the power, 2015-12-30 20:37:17]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-30 20:37:18]

Dry run: This issue passed the CQ dry run.

[mtklein, 2015-12-30 21:52:33]

The CQ bit was checked by mtklein@google.com to run a CQ dry run

[commit-bot: I haz the power, 2015-12-30 21:52:36]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/240001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/240001

[mtklein_C, 2015-12-30 22:26:08]

The CQ bit was checked by mtklein@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-12-30 22:26:14]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/260001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/260001

[commit-bot: I haz the power, 2015-12-30 22:28:20]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-30 22:28:21]

Dry run: Try jobs failed on following builders:
  Build-Win-MSVC-x86-Debug-Trybot on client.skia.compile (JOB_FAILED,
http://build.chromium.org/p/client.skia.compile/builders/Build-Win-MSVC-x86-D...)

[mtklein_C, 2015-12-30 22:32:11]

The CQ bit was checked by mtklein@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-12-30 22:32:20]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/320001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/320001

[commit-bot: I haz the power, 2015-12-30 22:34:21]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-30 22:34:22]

Dry run: Try jobs failed on following builders:
  Build-Win-MSVC-x86-Debug-Trybot on client.skia.compile (JOB_FAILED,
http://build.chromium.org/p/client.skia.compile/builders/Build-Win-MSVC-x86-D...)

[mtklein, 2015-12-31 01:39:22]

The CQ bit was checked by mtklein@google.com to run a CQ dry run

[commit-bot: I haz the power, 2015-12-31 01:39:37]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/340001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/340001

[mtklein, 2015-12-31 01:43:40]

The CQ bit was checked by mtklein@google.com to run a CQ dry run

[commit-bot: I haz the power, 2015-12-31 01:43:48]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/360001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/360001

[mtklein, 2015-12-31 01:55:58]

The CQ bit was checked by mtklein@google.com to run a CQ dry run

[commit-bot: I haz the power, 2015-12-31 01:56:08]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/380001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/380001

[commit-bot: I haz the power, 2015-12-31 02:10:41]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-31 02:10:41]

Dry run: This issue passed the CQ dry run.

[mtklein_C, 2015-12-31 02:28:24]

The CQ bit was checked by mtklein@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-12-31 02:28:32]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/420001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/420001

[commit-bot: I haz the power, 2015-12-31 02:43:37]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-31 02:43:38]

Dry run: This issue passed the CQ dry run.

[mtklein_C, 2015-12-31 07:27:43]

The CQ bit was checked by mtklein@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2015-12-31 07:27:47]

Dry run: CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/360002
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/360002

[commit-bot: I haz the power, 2015-12-31 07:42:13]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-12-31 07:42:14]

Dry run: This issue passed the CQ dry run.

[herb_g, 2016-01-22 18:58:25]

lgtm

[mtklein, 2016-01-22 20:00:06]

The CQ bit was checked by mtklein@google.com

[commit-bot: I haz the power, 2016-01-22 20:00:14]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/360002
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/360002

[commit-bot: I haz the power, 2016-01-22 20:06:03]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2016-01-22 20:06:04]

Try jobs failed on following builders:
  Test-Mac10.8-Clang-MacMini4.1-CPU-SSE4-x86_64-Release-Trybot on client.skia
(JOB_FAILED,
http://build.chromium.org/p/client.skia/builders/Test-Mac10.8-Clang-MacMini4....)

[mtklein, 2016-01-22 20:08:20]

The CQ bit was checked by mtklein@google.com

[commit-bot: I haz the power, 2016-01-22 20:08:35]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/360002
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/360002

[commit-bot: I haz the power, 2016-01-22 20:11:52]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2016-01-22 20:11:53]

Try jobs failed on following builders:
  Test-Mac10.8-Clang-MacMini4.1-CPU-SSE4-x86_64-Release-Trybot on client.skia
(JOB_FAILED,
http://build.chromium.org/p/client.skia/builders/Test-Mac10.8-Clang-MacMini4....)

[mtklein, 2016-01-25 14:08:17]

The CQ bit was checked by mtklein@google.com

[commit-bot: I haz the power, 2016-01-25 14:08:25]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/360002
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/360002

[commit-bot: I haz the power, 2016-01-25 14:09:47]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2016-01-25 14:09:50]

Try jobs failed on following builders:
  Build-Mac10.8-Clang-x86_64-Release-Trybot on client.skia.compile (JOB_FAILED,
http://build.chromium.org/p/client.skia.compile/builders/Build-Mac10.8-Clang-...)

[mtklein, 2016-01-25 15:14:35]

The CQ bit was checked by mtklein@google.com

[mtklein, 2016-01-25 15:14:35]

The patchset sent to the CQ was uploaded after l-g-t-m from herb@google.com
Link to the patchset: https://codereview.chromium.org/1532613002/#ps450001
(title: "i'd have sworn other_cflags used to work")

[commit-bot: I haz the power, 2016-01-25 15:14:40]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/450001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/450001

[commit-bot: I haz the power, 2016-01-25 15:23:59]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2016-01-25 15:24:00]

Try jobs failed on following builders:
  Test-Mac10.8-Clang-MacMini4.1-CPU-SSE4-x86_64-Release-Trybot on client.skia
(JOB_FAILED,
http://build.chromium.org/p/client.skia/builders/Test-Mac10.8-Clang-MacMini4....)

[mtklein, 2016-01-25 16:36:36]

Description was changed from

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I have backported these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two later.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should all be pixel-exact to the previous SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot,Test-Mac10.8-Clang-MacMini4.1-CPU-SSE4-x86_64-Release-Trybot;client.skia.compile:Build-Ubuntu-GCC-x86_64-Release-CMake-Trybot,Build-Mac10.9-Clang-x86_64-Release-CMake-Trybot
==========

to

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I have backported these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two later.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should all be pixel-exact to the previous SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot;client.skia.compile:Build-Ubuntu-GCC-x86_64-Release-CMake-Trybot,Build-Mac10.9-Clang-x86_64-Release-CMake-Trybot
==========

[mtklein, 2016-01-25 16:36:43]

The CQ bit was checked by mtklein@google.com

[commit-bot: I haz the power, 2016-01-25 16:36:55]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1532613002/450001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1532613002/450001

[commit-bot: I haz the power, 2016-01-25 16:37:33]

Description was changed from

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I have backported these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two later.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should all be pixel-exact to the previous SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot;client.skia.compile:Build-Ubuntu-GCC-x86_64-Release-CMake-Trybot,Build-Mac10.9-Clang-x86_64-Release-CMake-Trybot
==========

to

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I have backported these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two later.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should all be pixel-exact to the previous SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot;client.skia.compile:Build-Ubuntu-GCC-x86_64-Release-CMake-Trybot,Build-Mac10.9-Clang-x86_64-Release-CMake-Trybot

Committed:
https://skia.googlesource.com/skia/+/5d2117015eb271e09faf4a7ddd89093c9d618a36
==========

[commit-bot: I haz the power, 2016-01-25 16:37:34]

Committed patchset #24 (id:450001) as
https://skia.googlesource.com/skia/+/5d2117015eb271e09faf4a7ddd89093c9d618a36

[msarett, 2016-01-25 16:54:35]

A revert of this CL (patchset #24 id:450001) has been created in
https://codereview.chromium.org/1632713002/ by msarett@google.com.

The reason for reverting is: Bot failures.

[mtklein, 2016-01-25 17:07:17]

Description was changed from

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I have backported these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two later.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should all be pixel-exact to the previous SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot;client.skia.compile:Build-Ubuntu-GCC-x86_64-Release-CMake-Trybot,Build-Mac10.9-Clang-x86_64-Release-CMake-Trybot

Committed:
https://skia.googlesource.com/skia/+/5d2117015eb271e09faf4a7ddd89093c9d618a36
==========

to

==========
AVX 2 SrcOver blits: color32, blitmask.

As a follow up to the SSE 4.1 CL, this should look pretty familiar.

I've made some organizational changes around how we load, store, pack, and
unpack data that I think makes things clearer and more orthogonal, and it'll
make it easier to try out a pmaddubsw lerp.  I have backported these changes to
the SSE 4.1 code, and I hope that I can actually get a lot of this code
templated for sharing between the two later.

Perf changes (relative to SSE 4.1):
Xfermode_SrcOver:      1650 -> 1180  (0.71x)  // large opaque blit
Xfermode_SrcOver_aa:   1794 -> 1653  (0.92x)  // large opaque + small
transparent
text_16_AA_{FF,BK,WT}: 1.72 -> 1.59  (0.92x)  // small opaque blit
text_16_AA_88:         1.83 -> 1.77  (0.97x)  // small transparent blit

This should be a big throughout win, and a small latency win.
This should all be pixel-exact to the previous SSE 4.1 code.


GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...

Committed:
https://skia.googlesource.com/skia/+/5d2117015eb271e09faf4a7ddd89093c9d618a36

TODO:
- x86_64-cros-linux-gnu-g++ does not seem to support AVX2, despite accepting the
-mavx2 command line option:
error: '__m256i' does not name a type (and many others)

- we don't appear to be passing /arch:avx2 correctly to MSVC, producing this
warning-as-error:
warning C4752: found Intel(R) Advanced Vector Extensions; consider using
/arch:AVX

CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot;client.skia.compile:Build-Ubuntu-GCC-x86_64-Release-CMake-Trybot,Build-Mac10.9-Clang-x86_64-Release-CMake-Trybot,Build-Ubuntu-GCC-x86_64-Debug-CrOS_Link-Trybot,Build-Win-MSVC-x86_64-Release-Trybot
==========