[PATCH] D68826: AMDGPU: Fix redundant setting of m0 for atomic load/store
Matt Arsenault via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Thu Oct 10 12:47:02 PDT 2019
arsenm created this revision.
arsenm added a reviewer: rampitec.
Herald added subscribers: jfb, t-tye, tpr, dstuttard, yaxunl, nhaehnle, wdng, jvesely, kzhuravl.
Atomic load/store would have their setting of m0 handled twice, which
happened to be optimized out later.
https://reviews.llvm.org/D68826
Files:
lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
Index: lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -715,12 +715,17 @@
return; // Already selected.
}
- if (isa<AtomicSDNode>(N) ||
+ // isa<MemSDNode> almost works but is slightly too permisssive for some DS
+ // intrinsics.
+ if (Opc == ISD::LOAD || Opc == ISD::STORE || isa<AtomicSDNode>(N) ||
(Opc == AMDGPUISD::ATOMIC_INC || Opc == AMDGPUISD::ATOMIC_DEC ||
Opc == ISD::ATOMIC_LOAD_FADD ||
Opc == AMDGPUISD::ATOMIC_LOAD_FMIN ||
- Opc == AMDGPUISD::ATOMIC_LOAD_FMAX))
+ Opc == AMDGPUISD::ATOMIC_LOAD_FMAX)) {
N = glueCopyToM0LDSInit(N);
+ SelectCode(N);
+ return;
+ }
switch (Opc) {
default:
@@ -817,14 +822,6 @@
ReplaceNode(N, buildSMovImm64(DL, Imm, N->getValueType(0)));
return;
}
- case ISD::LOAD:
- case ISD::STORE:
- case ISD::ATOMIC_LOAD:
- case ISD::ATOMIC_STORE: {
- N = glueCopyToM0LDSInit(N);
- break;
- }
-
case AMDGPUISD::BFE_I32:
case AMDGPUISD::BFE_U32: {
// There is a scalar version available, but unlike the vector version which
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D68826.224450.patch
Type: text/x-patch
Size: 1213 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20191010/99444956/attachment.bin>
More information about the llvm-commits
mailing list