[mlir] Use the same pipeline tuning options as clang for execution engine

This change make sure that ExecutionEngine's pass pipeline is identical to one
used by clang. Previously, SLPVectorization was not enabled which caused
differences in code...

...generation.

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D137248

diff --git a/mlir/lib/ExecutionEngine/OptUtils.cpp b/mlir/lib/ExecutionEngine/OptUtils.cpp
index 403e54f..893a8a4 100644 (file)
--- a/mlir/lib/ExecutionEngine/OptUtils.cpp
+++ b/mlir/lib/ExecutionEngine/OptUtils.cpp
@@ -68,7 +68,13 @@ mlir::makeOptimizingTransformer(unsigned optLevel, unsigned sizeLevel,
     CGSCCAnalysisManager cgam;
     ModuleAnalysisManager mam;
 
-    PassBuilder pb(targetMachine);
+    PipelineTuningOptions tuningOptions;
+    tuningOptions.LoopUnrolling = true;
+    tuningOptions.LoopInterleaving = true;
+    tuningOptions.LoopVectorization = true;
+    tuningOptions.SLPVectorization = true;
+
+    PassBuilder pb(targetMachine, tuningOptions);
 
     pb.registerModuleAnalyses(mam);
     pb.registerCGSCCAnalyses(cgam);