From: Benjamin Kramer <benny.kra@googlemail.com>
Date: Fri, 24 Dec 2010 21:17:12 +0000 (+0000)
Subject: MemCpyOpt: Turn memcpys from a constant into a memset if possible.
X-Git-Url: http://plrg.eecs.uci.edu/git/?a=commitdiff_plain;h=a112087e4298ca8ec1bc8aef8a2b272e49faa7ac;p=oota-llvm.git

MemCpyOpt: Turn memcpys from a constant into a memset if possible.

This allows us to compile "int cst[] = {-1, -1, -1};" into
  movl  $-1, 16(%rsp)
  movq  $-1, 8(%rsp)
instead of
  movl  _cst+8(%rip), %eax
  movl  %eax, 16(%rsp)
  movq  _cst(%rip), %rax
  movq  %rax, 8(%rsp)


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122548 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
index 6dcb49f9358..c6381205c89 100644
--- a/lib/Transforms/Scalar/MemCpyOptimizer.cpp
+++ b/lib/Transforms/Scalar/MemCpyOptimizer.cpp
@@ -14,6 +14,7 @@
 
 #define DEBUG_TYPE "memcpyopt"
 #include "llvm/Transforms/Scalar.h"
+#include "llvm/GlobalVariable.h"
 #include "llvm/IntrinsicInst.h"
 #include "llvm/Instructions.h"
 #include "llvm/ADT/SmallVector.h"
@@ -31,6 +32,7 @@ using namespace llvm;
 STATISTIC(NumMemCpyInstr, "Number of memcpy instructions deleted");
 STATISTIC(NumMemSetInfer, "Number of memsets inferred");
 STATISTIC(NumMoveToCpy,   "Number of memmoves converted to memcpy");
+STATISTIC(NumCpyToSet,    "Number of memcpys converted to memset");
 
 /// isBytewiseValue - If the specified value can be set by repeating the same
 /// byte in memory, return the i8 value that it is represented with.  This is
@@ -38,6 +40,13 @@ STATISTIC(NumMoveToCpy,   "Number of memmoves converted to memcpy");
 /// i16 0xF0F0, double 0.0 etc.  If the value can't be handled with a repeated
 /// byte store (e.g. i16 0x1234), return null.
 static Value *isBytewiseValue(Value *V) {
+  // Look through constant globals.
+  if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V)) {
+    if (GV->mayBeOverridden() || !GV->isConstant() || !GV->hasInitializer())
+      return 0;
+    V = GV->getInitializer();
+  }
+
   // All byte-wide stores are splatable, even of arbitrary variables.
   if (V->getType()->isIntegerTy(8)) return V;
   
@@ -73,7 +82,24 @@ static Value *isBytewiseValue(Value *V) {
       return ConstantInt::get(V->getContext(), Val);
     }
   }
-  
+
+  // A ConstantArray is splatable if all its members are equal and also
+  // splatable.
+  if (ConstantArray *CA = dyn_cast<ConstantArray>(V)) {
+    if (CA->getNumOperands() == 0)
+      return 0;
+
+    Value *Val = isBytewiseValue(CA->getOperand(0));
+    if (!Val)
+      return 0;
+
+    for (unsigned I = 1, E = CA->getNumOperands(); I != E; ++I)
+      if (CA->getOperand(I-1) != CA->getOperand(I))
+        return 0;
+
+    return Val;
+  }
+
   // Conceptually, we could handle things like:
   //   %a = zext i8 %X to i16
   //   %b = shl i16 %a, 8
@@ -765,8 +791,24 @@ bool MemCpyOpt::processMemCpy(MemCpyInst *M) {
     M->eraseFromParent();
     return false;
   }
-  
-  
+
+  // If copying from a constant, try to turn the memcpy into a memset.
+  if (Value *ByteVal = isBytewiseValue(M->getSource())) {
+    Value *Ops[] = {
+      M->getRawDest(), ByteVal,               // Start, value
+      CopySize,                               // Size
+      M->getAlignmentCst(),                   // Alignment
+      ConstantInt::getFalse(M->getContext()), // volatile
+    };
+    const Type *Tys[] = { Ops[0]->getType(), Ops[2]->getType() };
+    Module *Mod = M->getParent()->getParent()->getParent();
+    Function *MemSetF = Intrinsic::getDeclaration(Mod, Intrinsic::memset, Tys, 2);
+    CallInst::Create(MemSetF, Ops, Ops+5, "", M);
+    M->eraseFromParent();
+    ++NumCpyToSet;
+    return true;
+  }
+
   // The are two possible optimizations we can do for memcpy:
   //   a) memcpy-memcpy xform which exposes redundance for DSE.
   //   b) call-memcpy xform for return slot optimization.
diff --git a/test/Transforms/MemCpyOpt/memcpy-to-memset.ll b/test/Transforms/MemCpyOpt/memcpy-to-memset.ll
new file mode 100644
index 00000000000..b18d176f003
--- /dev/null
+++ b/test/Transforms/MemCpyOpt/memcpy-to-memset.ll
@@ -0,0 +1,19 @@
+; RUN: opt -memcpyopt -S < %s | FileCheck %s
+
+@cst = internal constant [3 x i32] [i32 -1, i32 -1, i32 -1], align 4
+
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i32, i1) nounwind
+declare void @foo(i32*) nounwind
+
+define void @test1() nounwind {
+  %arr = alloca [3 x i32], align 4
+  %arr_i8 = bitcast [3 x i32]* %arr to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %arr_i8, i8* bitcast ([3 x i32]* @cst to i8*), i64 12, i32 4, i1 false)
+  %arraydecay = getelementptr inbounds [3 x i32]* %arr, i64 0, i64 0
+  call void @foo(i32* %arraydecay) nounwind
+  ret void
+; CHECK: @test1
+; CHECK: call void @llvm.memset
+; CHECK-NOT: call void @llvm.memcpy
+; CHECK: ret void
+}