Invoke simplifycfg and sroa before instcombine.

danielcdh · danielcdh · commit bd3ed3c55bfc · 2016-06-23T20:13:10.000Z
Summary: InstCombine needs to be performed after simplifycfg and sroa, otherwise it may make bad optimization decisions. Reviewers: davidxl, wmi, dnovillo Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D21568 llvm-svn: 273606
diff --git a/clang/lib/CodeGen/BackendUtil.cpp b/clang/lib/CodeGen/BackendUtil.cpp
@@ -178,8 +178,14 @@ static void addAddDiscriminatorsPass(const PassManagerBuilder &Builder,
   PM.add(createAddDiscriminatorsPass());
 }
 
-static void addInstructionCombiningPass(const PassManagerBuilder &Builder,
-                                        legacy::PassManagerBase &PM) {
+static void addCleanupPassesForSampleProfiler(
+    const PassManagerBuilder &Builder, legacy::PassManagerBase &PM) {
+  // instcombine is needed before sample profile annotation because it converts
+  // certain function calls to be inlinable. simplifycfg and sroa are needed
+  // before instcombine for necessary preparation. E.g. load store is eliminated
+  // properly so that instcombine will not introduce unecessary liverange.
+  PM.add(createCFGSimplificationPass());
+  PM.add(createSROAPass());
   PM.add(createInstructionCombiningPass());
 }
 
@@ -492,7 +498,7 @@ void EmitAssemblyHelper::CreatePasses(ModuleSummaryIndex *ModuleSummary) {
     MPM->add(createPruneEHPass());
     MPM->add(createSampleProfileLoaderPass(CodeGenOpts.SampleProfileFile));
     PMBuilder.addExtension(PassManagerBuilder::EP_EarlyAsPossible,
-                           addInstructionCombiningPass);
+                           addCleanupPassesForSampleProfiler);
   }
 
   PMBuilder.populateFunctionPassManager(*FPM);
diff --git a/clang/test/CodeGen/pgo-sample-preparation.c b/clang/test/CodeGen/pgo-sample-preparation.c
@@ -0,0 +1,16 @@
+// Test if PGO sample use preparation passes are executed correctly.
+//
+// Ensure that instcombine is executed after simplifycfg and sroa so that
+// "a < 255" will not be converted to a * 256 < 255 * 256.
+// RUN: %clang_cc1 -O2 -fprofile-sample-use=%S/Inputs/pgo-sample.prof %s -emit-llvm -o - 2>&1 | FileCheck %s
+
+void bar(int);
+void foo(int x, int y, int z) {
+  int m;
+  for (m = 0; m < x ; m++) {
+    int a = (((y >> 8) & 0xff) * z) / 256;
+    bar(a < 255 ? a : 255);
+  }
+}
+
+// CHECK-NOT: icmp slt i32 %mul, 65280
diff --git a/clang/test/CodeGen/pgo-sample.c b/clang/test/CodeGen/pgo-sample.c
@@ -2,5 +2,8 @@
 //
 // Ensure Pass PGOInstrumentationGenPass is invoked.
 // RUN: %clang_cc1 -O2 -fprofile-sample-use=%S/Inputs/pgo-sample.prof %s -mllvm -debug-pass=Structure -emit-llvm -o - 2>&1 | FileCheck %s
+// CHECK: Simplify the CFG
+// CHECK: SROA
+// CHECK: Combine redundant instructions
 // CHECK: Remove unused exception handling info
 // CHECK: Sample profile pass
