[all-commits] [llvm/llvm-project] 0e1768: [AMDGPU] Speed up SIRegisterInfo::getReservedRegs ...

[Jay Foad via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 0e17684cf5ce3b99e8f732f53c3534990ad91fd5
      https://github.com/llvm/llvm-project/commit/0e17684cf5ce3b99e8f732f53c3534990ad91fd5
  Author: Jay Foad <jay.foad at amd.com>
  Date:   2024-01-30 (Tue, 30 Jan 2024)

  Changed paths:
    M llvm/include/llvm/CodeGen/TargetRegisterInfo.h
    M llvm/include/llvm/MC/MCRegisterInfo.h
    M llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp
    M llvm/unittests/CodeGen/MachineInstrTest.cpp
    M llvm/utils/TableGen/RegisterInfoEmitter.cpp

  Log Message:
  -----------
  [AMDGPU] Speed up SIRegisterInfo::getReservedRegs (#79844)

reserveRegisterTuples is slow because it uses MCRegAliasIterator and
hence ends up reserving the same aliased registers many times. This
patch changes getReservedRegs not to use it for reserving SGPRs, VGPRs
and AGPRs. Instead it iterates through base register classes, which
should come closer to reserving each register once only.

Overall this speeds up the time to run check-llvm-codegen-amdgpu in my
Release build from 18.4 seconds to 16.9 seconds (all timings +/- 0.2).