[webkit-changes] [WebKit/WebKit] 67d43c: Make MacroAssemblerARMv7 scratch regs available to...

[aoikonomopoulos]

  Branch: refs/heads/main
  Home:   https://github.com/WebKit/WebKit
  Commit: 67d43c36bbf9f2111a98316b5bd4a7ea1a539761
      https://github.com/WebKit/WebKit/commit/67d43c36bbf9f2111a98316b5bd4a7ea1a539761
  Author: Angelos Oikonomopoulos <angelos at igalia.com>
  Date:   2022-12-22 (Thu, 22 Dec 2022)

  Changed paths:
    M Source/JavaScriptCore/assembler/MacroAssemblerARMv7.h
    M Source/JavaScriptCore/b3/B3Common.cpp
    M Source/JavaScriptCore/b3/B3LowerToAir.cpp
    M Source/JavaScriptCore/b3/B3MemoryValueInlines.h
    M Source/JavaScriptCore/b3/air/AirAllocateRegistersAndStackAndGenerateCode.cpp
    M Source/JavaScriptCore/b3/air/AirArg.h
    M Source/JavaScriptCore/b3/air/AirCode.cpp
    M Source/JavaScriptCore/b3/air/AirHelpers.h
    M Source/JavaScriptCore/b3/air/AirLowerStackArgs.cpp
    M Source/JavaScriptCore/b3/air/opcode_generator.rb
    M Source/JavaScriptCore/wasm/WasmAirIRGenerator32_64.cpp
    M Source/JavaScriptCore/wasm/WasmAirIRGenerator64.cpp
    M Source/JavaScriptCore/wasm/WasmAirIRGeneratorBase.h

  Log Message:
  -----------
  Make MacroAssemblerARMv7 scratch regs available to Air
https://bugs.webkit.org/show_bug.cgi?id=249648

Reviewed by Justin Michaud.

Make addressTempRegister/dataTempRegister available for use in Air. We
still need to reserve one of the two as the extendedOffsetAddrRegister,
so this results in one extra usable register in Air.

Legal load offsets on ARM differ between words and doubles, so we need
to thread the Air opcode all the way to isValidAddrForm.

We also make calls to MacroAssemblerARMv7::short_move (which make use of
the cachedAddressTempRegister) conditional on m_allowScratchRegister, so
that we don't accidentally try to use a cached value.

The trickiest part is in the use of the extendedOffsetAddrRegister in
Air. The register allocator in
AirAllocateRegistersAndStackAndGenerateCode.cpp assumes that it can
flush a register value without needing any extra registers. This is not
the case on ARM when the offset relative to the callFrameRegister is too
large. It used to be the case before this patch, because
MacroAssemblerARMv7 was privately reserving a register.

However, the extendedOffsetAddrRegister may have already been allocated
by lowerStackArgs, so we can't rely on it being available when we later
need to spill a value.

Instead, we move the lowering of the extended offset to
GenerateAndAllocateRegisters::generate, which becomes the sole user of
extendedOffsetAddrRegister.

Note that this is not a pure win on ARMv7. We lose the reuse of
previously computed addresess in MacroAssemblerARMv7, which results in
larger code size overall. We can hopefully claim those back by
implementing the same caching logic generically in Air. This is a wash
on the total score for JetStream2 but the corresponding code size is 18%
larger.

* Source/JavaScriptCore/assembler/MacroAssemblerARMv7.h:
(JSC::MacroAssemblerARMv7::store32):
(JSC::MacroAssemblerARMv7::cachedRegisterGetValue):
(JSC::MacroAssemblerARMv7::cachedRegisterSetValue):
(JSC::MacroAssemblerARMv7::short_move):
(JSC::MacroAssemblerARMv7::move):
(JSC::MacroAssemblerARMv7::setupArmAddress):
(JSC::MacroAssemblerARMv7::absoluteAddressWithinShortOffset):
(JSC::MacroAssemblerARMv7::moveFixedWidthEncoding):
* Source/JavaScriptCore/b3/B3Common.cpp:
(JSC::B3::extendedOffsetAddrRegister):
* Source/JavaScriptCore/b3/B3LowerToAir.cpp:
* Source/JavaScriptCore/b3/B3MemoryValueInlines.h:
(JSC::B3::MemoryValue::isLegalOffsetImpl const):
* Source/JavaScriptCore/b3/air/AirAllocateRegistersAndStackAndGenerateCode.cpp:
(JSC::B3::Air::callFrameAddr):
(JSC::B3::Air::GenerateAndAllocateRegisters::flush):
(JSC::B3::Air::GenerateAndAllocateRegisters::alloc):
(JSC::B3::Air::GenerateAndAllocateRegisters::generate):
* Source/JavaScriptCore/b3/air/AirArg.h:
(JSC::B3::Air::Arg::isValidAddrForm):
(JSC::B3::Air::Arg::isValidForm const):
* Source/JavaScriptCore/b3/air/AirCode.cpp:
(JSC::B3::Air::Code::Code):
* Source/JavaScriptCore/b3/air/AirHelpers.h:
(JSC::B3::Air::moveForType):
* Source/JavaScriptCore/b3/air/AirLowerStackArgs.cpp:
(JSC::B3::Air::lowerStackArgs):
* Source/JavaScriptCore/b3/air/opcode_generator.rb:
* Source/JavaScriptCore/wasm/WasmAirIRGenerator32_64.cpp:
(JSC::Wasm::AirIRGenerator32::emitZeroInitialize):
(JSC::Wasm::AirIRGenerator32::emitLoad):
(JSC::Wasm::AirIRGenerator32::emitStore):
(JSC::Wasm::isFPLoadOp):
(JSC::Wasm::AirIRGenerator32::emitLoadOp):
(JSC::Wasm::isFPStoreOp):
(JSC::Wasm::AirIRGenerator32::emitStoreOp):
* Source/JavaScriptCore/wasm/WasmAirIRGenerator64.cpp:
(JSC::Wasm::AirIRGenerator64::materializeAddrArg):
* Source/JavaScriptCore/wasm/WasmAirIRGeneratorBase.h:
(JSC::Wasm::AirIRGeneratorBase::emitPatchpoint):
(JSC::Wasm::ExpressionType>::restoreWebAssemblyGlobalState):
(JSC::Wasm::ExpressionType>::addCurrentMemory):
(JSC::Wasm::ExpressionType>::getGlobal):
(JSC::Wasm::ExpressionType>::setGlobal):
(JSC::Wasm::ExpressionType>::fixupPointerPlusOffsetForAtomicOps):
(JSC::Wasm::ExpressionType>::addCallIndirect):

Canonical link: https://commits.webkit.org/258279@main