[Codegen] Do not consider parallel regions in bufferization analysis (#17757)
When there is a buffer used inside of an `scf.forall` op that is defined
outside of the `scf.forall`, bufferization will unconditionally
bufferize out of place by default in order to avoid race conditions.
However, handling parallel accesses to a buffer should generally be the
responsibility of the source program, and if there is a race condition,
then it should be handled outside of bufferization. This PR disables the
parallel region check in IREE to simplify the bufferization analysis and
enable more buffer reuse.
It is possible that this PR could cause race conditions if data races
are not handled properly, and we are relying too much on bufferization
to be conservative. Turning this option off could be a good early step
in diagnosing data races on GPU.
---------
Signed-off-by: Max Dawkins <max.dawkins@gmail.com>
diff --git a/compiler/src/iree/compiler/Codegen/Common/IREEComprehensiveBufferizePass.cpp b/compiler/src/iree/compiler/Codegen/Common/IREEComprehensiveBufferizePass.cpp
index 97a2821..9b46b84 100644
--- a/compiler/src/iree/compiler/Codegen/Common/IREEComprehensiveBufferizePass.cpp
+++ b/compiler/src/iree/compiler/Codegen/Common/IREEComprehensiveBufferizePass.cpp
@@ -206,6 +206,11 @@
options.printConflicts = printConflicts;
options.allocationFn = allocationFn;
options.memCpyFn = memCpyFn;
+ // Turning off checkParallelRegions assumes that we are not relying too much
+ // on bufferization being conservative. If we are, then this could cause race
+ // conditions. Turning this option off could be a good step in diagnosing
+ // data races on GPU.
+ options.checkParallelRegions = false;
if (failed(runIREEOneShotBufferize(funcOp, options))) {
return signalPassFailure();
