[llvm-branch-commits] [mlir] 4c372a3 - [mlir] Make GpuAsyncRegion pass depend on async dialect.

[Christian Sigg via llvm-branch-commits]

Author: Christian Sigg
Date: 2021-01-11T14:43:07+01:00
New Revision: 4c372a35cd892c0847da45540057447633a31d99

URL: https://github.com/llvm/llvm-project/commit/4c372a35cd892c0847da45540057447633a31d99
DIFF: https://github.com/llvm/llvm-project/commit/4c372a35cd892c0847da45540057447633a31d99.diff

LOG: [mlir] Make GpuAsyncRegion pass depend on async dialect.

Do not cache gpu.async.token type so that the pass can be created before the GPU dialect is registered.

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D94397

Added: 
    

Modified: 
    mlir/include/mlir/Dialect/GPU/Passes.td
    mlir/lib/Dialect/GPU/Transforms/AsyncRegionRewriter.cpp
    mlir/lib/Dialect/GPU/Transforms/PassDetail.h

Removed: 
    


################################################################################
diff  --git a/mlir/include/mlir/Dialect/GPU/Passes.td b/mlir/include/mlir/Dialect/GPU/Passes.td
index b9a83f7ef534..da1899c325fc 100644
--- a/mlir/include/mlir/Dialect/GPU/Passes.td
+++ b/mlir/include/mlir/Dialect/GPU/Passes.td
@@ -19,6 +19,7 @@ def GpuKernelOutlining : Pass<"gpu-kernel-outlining", "ModuleOp"> {
 def GpuAsyncRegionPass : FunctionPass<"gpu-async-region"> {
   let summary = "Make GPU ops async";
   let constructor = "mlir::createGpuAsyncRegionPass()";
+  let dependentDialects = ["async::AsyncDialect"];
 }
 
 #endif // MLIR_DIALECT_GPU_PASSES

diff  --git a/mlir/lib/Dialect/GPU/Transforms/AsyncRegionRewriter.cpp b/mlir/lib/Dialect/GPU/Transforms/AsyncRegionRewriter.cpp
index 4f0eefb05931..37e0b9373d42 100644
--- a/mlir/lib/Dialect/GPU/Transforms/AsyncRegionRewriter.cpp
+++ b/mlir/lib/Dialect/GPU/Transforms/AsyncRegionRewriter.cpp
@@ -78,6 +78,8 @@ struct GpuAsyncRegionPass::ThreadTokenCallback {
     if (op->getNumRegions() > 0)
       return op->emitOpError("regions are not supported");
 
+    auto tokenType = builder.getType<gpu::AsyncTokenType>();
+
     // If there is no current token, insert a `gpu.wait async` without
     // dependencies to create one.
     if (!currentToken)
@@ -108,7 +110,7 @@ struct GpuAsyncRegionPass::ThreadTokenCallback {
   }
 
   OpBuilder builder;
-  const Type tokenType = builder.getType<gpu::AsyncTokenType>();
+
   // The token that represents the current asynchronous dependency. It's valid
   // range starts with a `gpu.wait async` op, and ends with a `gpu.wait` op.
   // In between, each gpu::AsyncOpInterface depends on the current token and

diff  --git a/mlir/lib/Dialect/GPU/Transforms/PassDetail.h b/mlir/lib/Dialect/GPU/Transforms/PassDetail.h
index 2e6fb9217189..c83ac3035a40 100644
--- a/mlir/lib/Dialect/GPU/Transforms/PassDetail.h
+++ b/mlir/lib/Dialect/GPU/Transforms/PassDetail.h
@@ -9,6 +9,7 @@
 #ifndef DIALECT_GPU_TRANSFORMS_PASSDETAIL_H_
 #define DIALECT_GPU_TRANSFORMS_PASSDETAIL_H_
 
+#include "mlir/Dialect/Async/IR/Async.h"
 #include "mlir/Pass/Pass.h"
 
 namespace mlir {