Revert "We can mask load and store with just AVX."
authorBrian Osman <brianosman@google.com>
Mon, 22 May 2017 19:25:36 +0000 (19:25 +0000)
committerSkia Commit-Bot <skia-commit-bot@chromium.org>
Mon, 22 May 2017 19:30:46 +0000 (19:30 +0000)
commit74fc593209c28efdf57127219322d944dfdf7c10
tree8c1dd3bd1f3d70f8f3a2fd9a4a5937e8415739df
parent459889967899be7258edcd8e0567f63bb3ae33df
Revert "We can mask load and store with just AVX."

This reverts commit 139e463dc6f965fdaed854efcb20c6cafbb6dbdc.

Reason for revert: Crashes on Valgrind bots.

Original change's description:
> We can mask load and store with just AVX.
>
> Previously we were using AVX2 instructions to generate the masks,
> and AVX2 instructions for the mask load and stores themselves.
>
> AVX came with float mask loads and stores, which will work perfectly
> fine.  I don't really get what the point of the 32-bit int loads and
> stores are in AVX2, beyond maybe syntax sugar?
>
> Change-Id: I81fa55fb09daea4f5546f8c9ebbc886015edce51
> Reviewed-on: https://skia-review.googlesource.com/17452
> Reviewed-by: Herb Derby <herb@google.com>
> Commit-Queue: Ravi Mistry <rmistry@google.com>
>

TBR=mtklein@chromium.org,rmistry@google.com,herb@google.com,reed@google.com
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true

Change-Id: I3a48f006c20475f6334ff94998281f381c696c93
Reviewed-on: https://skia-review.googlesource.com/17524
Reviewed-by: Brian Osman <brianosman@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
src/jumper/SkJumper_generated.S
src/jumper/SkJumper_generated_win.S
src/jumper/SkJumper_stages.cpp