[SOL] Expand memcmp earlier (#172)

Author: LucasSte

llvm/lib/Target/SBF/SBFTargetMachine.cpp

@@ -10,19 +10,21 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "MCTargetDesc/SBFMCAsmInfo.h"
 #include "SBF.h"
+#include "SBFFunctionInfo.h"
 #include "SBFTargetMachine.h"
 #include "SBFTargetTransformInfo.h"
-#include "SBFFunctionInfo.h"
-#include "MCTargetDesc/SBFMCAsmInfo.h"
 #include "TargetInfo/SBFTargetInfo.h"
+#include "llvm/CodeGen/ExpandMemCmp.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
 #include "llvm/CodeGen/TargetPassConfig.h"
 #include "llvm/IR/PassManager.h"
 #include "llvm/MC/TargetRegistry.h"
 #include "llvm/Passes/PassBuilder.h"
 #include "llvm/Target/TargetOptions.h"
+#include "llvm/Transforms/InstCombine/InstCombine.h"
 #include "llvm/Transforms/Scalar.h"
 #include "llvm/Transforms/Scalar/SimplifyCFG.h"
 #include "llvm/Transforms/Utils/SimplifyCFGOptions.h"
@@ -117,6 +119,8 @@ void SBFTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
       });
   PB.registerPeepholeEPCallback([=](FunctionPassManager &FPM,
                                     OptimizationLevel Level) {
+    FPM.addPass(ExpandMemCmpPass(this));
+    FPM.addPass(InstCombinePass());
     FPM.addPass(SimplifyCFGPass(
         SimplifyCFGOptions().hoistCommonInsts(true).convertSwitchToLookupTable(
             true)));

llvm/lib/Target/SBF/SBFTargetTransformInfo.h

@@ -87,6 +87,7 @@ class SBFTTIImpl : public BasicTTIImplBase<SBFTTIImpl> {
     TTI::MemCmpExpansionOptions Options;
     Options.LoadSizes = {8, 4, 2, 1};
     Options.MaxNumLoads = TLI->getMaxExpandSizeMemcmp(OptSize);
+    Options.AllowOverlappingLoads = true;
     return Options;
   }
 

llvm/test/CodeGen/SBF/expand_memcmp.ll

@@ -0,0 +1,36 @@
+; RUN: opt -O2 -S < %s | FileCheck %s
+
+target datalayout = "e-m:e-p:64:64-i64:64-n32:64-S128"
+target triple = "sbf"
+
+; Function Attrs: mustprogress nofree nounwind willreturn memory(argmem: read)
+declare i32 @memcmp(ptr nocapture, ptr nocapture, i64) local_unnamed_addr
+
+; The memcmp is expanded with an overlaping load
+define i1 @yes_expand(ptr %a, ptr %b) {
+entry:
+    %res = call i32 @memcmp(ptr %a, ptr %b, i64 15)
+    %is_zero = icmp eq i32 %res, 0
+    ret i1 %is_zero
+
+; CHECK: entry:
+; CHECK: %0 = load i64, ptr %a, align 1
+; CHECK: %1 = load i64, ptr %b, align 1
+; CHECK: %.not = icmp eq i64 %0, %1
+; CHECK: br i1 %.not, label %loadbb1, label %res_block
+
+; CHECK: res_block:
+; CHECK: br label %endblock
+
+; CHECK: loadbb1:
+; CHECK: %2 = getelementptr i8, ptr %a, i64 7
+; CHECK: %3 = getelementptr i8, ptr %b, i64 7
+; CHECK: %4 = load i64, ptr %2, align 1
+; CHECK: %5 = load i64, ptr %3, align 1
+; CHECK: %.not2 = icmp eq i64 %4, %5
+; CHECK: br i1 %.not2, label %endblock, label %res_block
+
+; CHECK: endblock:
+; CHECK: %is_zero = phi i1 [ true, %loadbb1 ], [ false, %res_block ]
+; CHECK: ret i1 %is_zero
+}