-
Notifications
You must be signed in to change notification settings - Fork 64
Conversation
Testing patch series with bf9935b as its head.
|
@rakuco , could you please look at the coding style check in bots? I didn't experience it locally. And it says "src/runtime.h" not found is wired. Since this patch doesn't remove runtime.h. It should be there. Thanks in advance. |
and Int32x4Array data types, implement the SIMD operations on the Float32x4, Float64x2 and Int32x4 with C++ runtime functions and implement Float32x4Array, Float64x2Array and Int32x4Array element loading and storing. Conflicts: include/v8.h src/arm/assembler-arm.h src/hydrogen-instructions.cc src/ia32/assembler-ia32.h src/ia32/lithium-codegen-ia32.cc src/types.cc src/types.h src/x64/assembler-x64.h src/x64/lithium-codegen-x64.cc src/x64/lithium-x64.cc src/x64/lithium-x64.h
These objects contains a "value" pointer pointing to a corresponding FixedTypedArray which is the real data storage.
Testing patch series with 3c70675 as its head.
|
Update the PR to implement the new SIMD128 JSObject layout. |
Again, the coding style failure is wired. All three ports passed tests locally. |
The implementation is for ia32 and x64 ports. Optimize Float32x4Array, Float64x2Array and Int32x4Array element loading and storing for ia32 and x64 port by: 1). Allocating SIMD128 registers 2). Emitting SIMD movups instruction 3). Adding deoptimization support for SIMD128 registers or stack slots Inline all SIMD operations and emit SIMD instructions for ia32 and x64 port Conflicts: src/hydrogen-instructions.cc src/hydrogen-instructions.h src/ia32/assembler-ia32.h src/ia32/deoptimizer-ia32.cc src/ia32/lithium-codegen-ia32.cc src/ia32/lithium-codegen-ia32.h src/ia32/macro-assembler-ia32.cc src/x64/assembler-x64.h src/x64/lithium-codegen-x64.cc src/x64/lithium-x64.h src/hydrogen-instructions.cc src/hydrogen-instructions.h src/hydrogen.cc src/ia32/lithium-ia32.cc src/x64/lithium-x64.cc test/cctest/test-disasm-ia32.cc
- Fix the runtime function test auto generation. - Fix the coding style issues. TODO(ningxin): squash into patch 1
Implement the simd128-tag inlining for the new JSObject-based SIMD128 types' layout. TODO(ningxin): squash into patch-2
Ping @fenghaitao. I'll try to fix the style checker. |
I think I've fixed the style checker -- instead of running the presubmit tests as the "check coding style" step, I'm now running them properly as actual tests. Please submit your patch again, I think there were some failures already present in the tree without your patch that were caught by those new tests. |
Testing patch series with 97783aa as its head.
|
Testing patch series with d66d271 as its head.
|
The x64 failure passes on my local test. I am going to trigger the bots once more. |
Testing patch series with af33544 as its head.
|
Bots give different results without any code change... |
This is something I couldn't understand so far either -- when we moved to M37's V8 I had to remove the build directory in build.crosswalk-project.org (not the tryslaves here) to get Check to stop crashing in x64. I did the same thing today with the tryslaves and Check still fails. Any insights here would be appreciated. |
I didn't meet any x64 Check failure issues on my local build for M37. @fenghaitao @weilianglin , do you have any ideas? |
Is it related to VM configuration? I remember previously we have one issue only in VM but not in real machines. |
Testing patch series with 0fe7e99 as its head.
|
I've added a commit to implement the inlining of simd128-tag. It improve the performance of all benchmarks especially for With this, there are no significant performance regressions comparing to M36. (Actually, it has some performance regression since the new implementation employs more complex object layout.) |
The bots finally decided to cooperate :-) I just need @fenghaitao or anyone else with any V8 expertise to lgtm this to merge. |
LGTM |
Thanks for the review! |
This is essentially a cherry-pick that was applied before plus the removal of a test that is now failing because it depends on more involved changes. The test case checks comparison of different undetectable JSReceivers, which is not relevant in practice, as there's only one of these at most, which is document.all. Merged 55b4df7 Merged d00da47 Merged 4da2e3d Merged c1507e1 [runtime] Unify comparison operator runtime entries. [turbofan] Don't use the CompareIC in JSGenericLowering. PPC: [runtime] Unify comparison operator runtime entries. PPC: [turbofan] Don't use the CompareIC in JSGenericLowering. R=yangguo@chromium.org BUG=chromium:590832,v8:4788,chromium:606181 LOG=N NOTRY=true NOPRESUBMIT=true Cr-Commit-Position: refs/branch-heads/5.0@{#13} Cr-Branched-From: ad16e6c-refs/heads/5.0.71@{#1} Cr-Branched-From: bd9df50-refs/heads/master@{#34215} Review URL: https://codereview.chromium.org/1925463003 Cr-Commit-Position: refs/branch-heads/5.0@{#44} Cr-Branched-From: ad16e6c-refs/heads/5.0.71@{#1} Cr-Branched-From: bd9df50-refs/heads/master@{#34215}
Merged f98caf4 Various species micro-optimizations BUG=chromium:609739 LOG=N R=adamk@chromium.org, adamk Review URL: https://codereview.chromium.org/1977943002 . Cr-Commit-Position: refs/branch-heads/5.1@{#44} Cr-Branched-From: 167dc63-refs/heads/5.1.281@{#1} Cr-Branched-From: 03953f5-refs/heads/master@{#35282}
Merged 46428e4 Make it possible to create a v8::Function directly w/o a prototype BUG=chromium:625823 LOG=N TBR=hablich@chromium.org Review URL: https://codereview.chromium.org/2133683002 . Cr-Commit-Position: refs/branch-heads/5.2@{#44} Cr-Branched-From: 2cd36d6-refs/heads/5.2.361@{#1} Cr-Branched-From: 3fef34e-refs/heads/master@{#36332}
Revision: 6cf621e BUG=chromium:628542 LOG=N NOTRY=true NOPRESUBMIT=true NOTREECHECKS=true R=hablich@chromium.org Review-Url: https://codereview.chromium.org/2248213003 Cr-Commit-Position: refs/branch-heads/5.3@{#44} Cr-Branched-From: 820a23a-refs/heads/5.3.332@{#2} Cr-Branched-From: 37538cb-refs/heads/master@{#37308}
Cr-Commit-Position: refs/branch-heads/5.9@{crosswalk-project#44} Cr-Branched-From: fe9bb7e-refs/heads/5.9.211@{crosswalk-project#1} Cr-Branched-From: 70ad237-refs/heads/master@{#44591}
Cr-Commit-Position: refs/branch-heads/5.8@{crosswalk-project#44} Cr-Branched-From: eda659c-refs/heads/5.8.283@{crosswalk-project#1} Cr-Branched-From: 4310cd0-refs/heads/master@{#43429}
Implement the SIMD types as JSObject.