[llvm] 6578356 - [TableGen] Ignore inaccessible memory when checking pattern flags (#90061)
via llvm-commits
llvm-commits at lists.llvm.org
Fri Apr 26 02:28:56 PDT 2024
Author: Jay Foad
Date: 2024-04-26T10:28:52+01:00
New Revision: 6578356a4e3e6acd7983c74feab43ac96925894c
URL: https://github.com/llvm/llvm-project/commit/6578356a4e3e6acd7983c74feab43ac96925894c
DIFF: https://github.com/llvm/llvm-project/commit/6578356a4e3e6acd7983c74feab43ac96925894c.diff
LOG: [TableGen] Ignore inaccessible memory when checking pattern flags (#90061)
In the AMDGPU backend we have some cases where we'd like to mark an
intrinsic as IntrInaccessibleMemOnly to model dependencies, but the
corresponding MachineInstrs use uses/defs of a special physical register
to express the same thing. In this case TableGen would complain:
Pattern doesn't match mayLoad/mayStore = 0
but the error is not useful.
Added:
Modified:
llvm/utils/TableGen/Common/CodeGenDAGPatterns.cpp
Removed:
################################################################################
diff --git a/llvm/utils/TableGen/Common/CodeGenDAGPatterns.cpp b/llvm/utils/TableGen/Common/CodeGenDAGPatterns.cpp
index 88d353e89a4614..e0e31739e26262 100644
--- a/llvm/utils/TableGen/Common/CodeGenDAGPatterns.cpp
+++ b/llvm/utils/TableGen/Common/CodeGenDAGPatterns.cpp
@@ -3616,7 +3616,15 @@ class InstAnalyzer {
hasChain = true;
if (const CodeGenIntrinsic *IntInfo = N.getIntrinsicInfo(CDP)) {
- ModRefInfo MR = IntInfo->ME.getModRef();
+ // Ignore reads/writes to inaccessible memory. These should not imply
+ // mayLoad/mayStore on the instruction because they are often used to
+ // model dependencies that Machine IR expresses as uses/defs of a
+ // special physical register.
+ ModRefInfo MR = ModRefInfo::NoModRef;
+ for (MemoryEffects::Location Loc : MemoryEffects::locations()) {
+ if (Loc != MemoryEffects::Location::InaccessibleMem)
+ MR |= IntInfo->ME.getModRef();
+ }
// If this is an intrinsic, analyze it.
if (isRefSet(MR))
mayLoad = true; // These may load memory.
More information about the llvm-commits
mailing list