projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: d17ea67) | patch
[AMDGPU] Use ds_read/write_b96/b128 when possible for SDag
authorMirko Brkusanin <Mirko.Brkusanin@amd.com>
Fri, 21 Aug 2020 09:51:06 +0000 (11:51 +0200)
committerMirko Brkusanin <Mirko.Brkusanin@amd.com>
Fri, 21 Aug 2020 10:26:31 +0000 (12:26 +0200)
commit0654ff703d4e99423133165db63083b831efb9b6
tree1117e532d719e97cfba0345b93abf7dc5881a20dtree | snapshot
parentd17ea67b92f6611a169dbd4a1399664078283648commit | diff
[AMDGPU] Use ds_read/write_b96/b128 when possible for SDag

Do not break down local loads and stores so ds_read/write_b96/b128 in
ISelLowering can be selected on subtargets that support them and if align
requirements allow them.

Differential Revision: https://reviews.llvm.org/D84403
15 files changed:
llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h diff | blob | history
llvm/lib/Target/AMDGPU/SIISelLowering.cpp diff | blob | history
llvm/test/CodeGen/AMDGPU/ds_read2.ll diff | blob | history
llvm/test/CodeGen/AMDGPU/ds_write2.ll diff | blob | history
llvm/test/CodeGen/AMDGPU/insert-subvector-unused-scratch.ll diff | blob | history
llvm/test/CodeGen/AMDGPU/lds-misaligned-bug.ll diff | blob | history
llvm/test/CodeGen/AMDGPU/load-local-f32.ll diff | blob | history
llvm/test/CodeGen/AMDGPU/load-local-i16.ll diff | blob | history
llvm/test/CodeGen/AMDGPU/load-local-i32.ll diff | blob | history
llvm/test/CodeGen/AMDGPU/load-local-i8.ll diff | blob | history
llvm/test/CodeGen/AMDGPU/load-local.128.ll [new file with mode: 0644] blob
llvm/test/CodeGen/AMDGPU/load-local.96.ll [new file with mode: 0644] blob
llvm/test/CodeGen/AMDGPU/store-local.128.ll [new file with mode: 0644] blob
llvm/test/CodeGen/AMDGPU/store-local.96.ll [new file with mode: 0644] blob
llvm/test/CodeGen/AMDGPU/store-local.ll diff | blob | history
Domain: System / Toolchain;