[PATCH] D84403: [AMDGPU] Use ds_read/write_b96/b128 when possible for SDag
Matt Arsenault via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Mon Aug 10 12:14:12 PDT 2020
arsenm added inline comments.
================
Comment at: llvm/lib/Target/AMDGPU/SIISelLowering.cpp:7954
+ // Use ds_read_b128 or ds_read_b96 when possible.
+ if (Subtarget->getGeneration() >= AMDGPUSubtarget::SEA_ISLANDS &&
+ MemVT.getStoreSize() <= (Subtarget->useDS128() ? 16 : 12) &&
----------------
I don't want to spread generation checks anywhere outside of AMDGPUSubtarget, so this should check some other feature or helper inside the subtarget
================
Comment at: llvm/lib/Target/AMDGPU/SIISelLowering.cpp:8432
+ // Use ds_write_b128 or ds_write_b96 when possible.
+ if (Subtarget->getGeneration() >= AMDGPUSubtarget::SEA_ISLANDS &&
+ VT.getStoreSize() <= (Subtarget->useDS128() ? 16 : 12) &&
----------------
Ditto
================
Comment at: llvm/lib/Target/AMDGPU/SIISelLowering.cpp:8435
+ allowsMisalignedMemoryAccessesImpl(VT.getSizeInBits(), AS,
+ Align(Store->getAlignment()))) {
+ return SDValue();
----------------
I assume there's a Store->getAlign() to avoid the Align construction?
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D84403/new/
https://reviews.llvm.org/D84403
More information about the llvm-commits
mailing list