Drop unit dims

iree-org · Aug 23, 2023 · 9cc4914 · 9cc4914
1 parent aec0744
commit 9cc4914
Showing 1 changed file with 16 additions and 0 deletions.
diff --git a/compiler/src/iree/compiler/Codegen/LLVMCPU/Passes.cpp b/compiler/src/iree/compiler/Codegen/LLVMCPU/Passes.cpp
@@ -507,8 +507,16 @@ void addMultiTilingExpertPassPipeline(
     nestedModulePM.addNestedPass<func::FuncOp>(createCSEPass());
   }
 
+  // Eliminate redundant transfer_read/write to avoid stack allocations.
+  nestedModulePM.addNestedPass<func::FuncOp>(
+      createOptimizeVectorTransferPass(/*flatten=*/false));
+
   addBufferizePasses(nestedModulePM);
 
+  // Perform memref-based transfer_read/write optimizations.
+  nestedModulePM.addNestedPass<func::FuncOp>(
+      createOptimizeVectorTransferPass(/*flatten=*/false));
+
   // Run IREE specific passes before vector lowering expert.
   nestedModulePM.addNestedPass<func::FuncOp>(
       createRemoveSingleIterationLoopPass());
@@ -655,8 +663,16 @@ void addCPUDataTilingPipeline(OpPassManager &passManager,
     nestedModulePM.addNestedPass<func::FuncOp>(createCSEPass());
   }
 
+  // Eliminate redundant transfer_read/write to avoid stack allocations.
+  nestedModulePM.addNestedPass<func::FuncOp>(
+      createOptimizeVectorTransferPass(/*flatten=*/false));
+
   addBufferizePasses(nestedModulePM);
 
+  // Perform memref-based transfer_read/write optimizations.
+  nestedModulePM.addNestedPass<func::FuncOp>(
+      createOptimizeVectorTransferPass(/*flatten=*/false));
+
   {
     LLVMCPUVectorLoweringPassOptions options;
     options.splitVectorTransfersTo = "linalg-copy";