projects / oota-llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 5834764) | patch
AMDGPU: don't match vgpr loads for constant loads
authorMarek Olsak <marek.olsak@amd.com>
Mon, 27 Jul 2015 18:16:08 +0000 (18:16 +0000)
committerMarek Olsak <marek.olsak@amd.com>
Mon, 27 Jul 2015 18:16:08 +0000 (18:16 +0000)
commitdbd8d4f0567b33821e184231584aada79a15375b
tree9bde19ba2d12664cb35e44088b433a7a93e50762tree | snapshot
parent5834764c65865437796632982f99acc7f1da6566commit | diff
AMDGPU: don't match vgpr loads for constant loads

Author: Dave Airlie <airlied@redhat.com>

In order to implement indirect sampler loads, we don't
want to match on a VGPR load but an SGPR one for constants,
as we cannot feed VGPRs to the sampler only SGPRs.

this should be applicable for llvm 3.7 as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@243294 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Target/AMDGPU/SIInstructions.td diff | blob | history
test/CodeGen/AMDGPU/gv-const-addrspace.ll diff | blob | history
test/CodeGen/AMDGPU/smrd.ll diff | blob | history
C/C++ LLVM-based compilers that forbids OOTA behaviors