[webkit-changes] [WebKit/WebKit] 80778f: [JSC] Use frint32z if possible

[Yusuke Suzuki]

  Branch: refs/heads/main
  Home:   https://github.com/WebKit/WebKit
  Commit: 80778f6d90cfd5af2da37b8a47dd345dc4df5fee
      https://github.com/WebKit/WebKit/commit/80778f6d90cfd5af2da37b8a47dd345dc4df5fee
  Author: Yusuke Suzuki <ysuzuki at apple.com>
  Date:   2024-09-23 (Mon, 23 Sep 2024)

  Changed paths:
    A JSTests/microbenchmarks/hash-map-double.js
    M Source/JavaScriptCore/assembler/ARM64Assembler.h
    M Source/JavaScriptCore/assembler/MacroAssemblerARM64.cpp
    M Source/JavaScriptCore/assembler/MacroAssemblerARM64.h
    M Source/JavaScriptCore/dfg/DFGSpeculativeJIT.cpp
    M Source/JavaScriptCore/disassembler/ARM64/A64DOpcode.cpp
    M Source/JavaScriptCore/disassembler/ARM64/A64DOpcode.h
    M Source/JavaScriptCore/ftl/FTLLowerDFGToB3.cpp
    M Source/JavaScriptCore/ftl/FTLOutput.cpp
    M Source/JavaScriptCore/ftl/FTLOutput.h
    M Source/WTF/wtf/PlatformHave.h

  Log Message:
  -----------
  [JSC] Use frint32z if possible
https://bugs.webkit.org/show_bug.cgi?id=280094
rdar://136392239

Reviewed by Yijia Huang.

ARMv8.4 supports frint32z / frint32x / frint64z / frint64x and we can
convert double to int32 / int64 representing integers within FP
registers. This patch utilizes it so that it reduces traveling between
FP and GP.

                                ToT                     Patched

    hash-map-double       11.7468+-0.0558     ^     11.4932+-0.0750        ^ definitely 1.0221x faster

* JSTests/microbenchmarks/hash-map-double.js: Added.
(test):
* Source/JavaScriptCore/assembler/ARM64Assembler.h:
* Source/JavaScriptCore/assembler/MacroAssemblerARM64.cpp:
(JSC::MacroAssemblerARM64::collectCPUFeatures):
* Source/JavaScriptCore/assembler/MacroAssemblerARM64.h:
(JSC::MacroAssemblerARM64::roundTowardZeroInt32Double):
(JSC::MacroAssemblerARM64::roundTowardZeroInt32Float):
(JSC::MacroAssemblerARM64::roundTowardZeroInt64Double):
(JSC::MacroAssemblerARM64::roundTowardZeroInt64Float):
(JSC::MacroAssemblerARM64::branchConvertDoubleToInt32):
(JSC::MacroAssemblerARM64::supportsRoundFloatToIntegerFloat):
* Source/JavaScriptCore/dfg/DFGSpeculativeJIT.cpp:
* Source/JavaScriptCore/disassembler/ARM64/A64DOpcode.cpp:
(JSC::ARM64Disassembler::A64DOpcodeFloatingPointDataProcessing4Source::format):
* Source/JavaScriptCore/disassembler/ARM64/A64DOpcode.h:
(JSC::ARM64Disassembler::A64DOpcodeFloatingPointDataProcessing4Source::opName):
(JSC::ARM64Disassembler::A64DOpcodeFloatingPointDataProcessing4Source::opNum):
* Source/JavaScriptCore/ftl/FTLLowerDFGToB3.cpp:
(JSC::FTL::DFG::LowerDFGToB3::compileDoubleRep):
(JSC::FTL::DFG::LowerDFGToB3::tryDoubleToInt32AsDouble):
(JSC::FTL::DFG::LowerDFGToB3::tryDoubleToInt64AsDouble):
(JSC::FTL::DFG::LowerDFGToB3::compileArithPow):
(JSC::FTL::DFG::LowerDFGToB3::compileCompareStrictEq):
* Source/JavaScriptCore/ftl/FTLOutput.cpp:
(JSC::FTL::Output::doubleToInt32):
(JSC::FTL::Output::doubleToUInt32):
(JSC::FTL::Output::doubleToInt): Deleted.
(JSC::FTL::Output::doubleToUInt): Deleted.
* Source/JavaScriptCore/ftl/FTLOutput.h:
* Source/WTF/wtf/PlatformHave.h:

Canonical link: https://commits.webkit.org/284086@main



To unsubscribe from these emails, change your notification settings at https://github.com/WebKit/WebKit/settings/notifications