Change IR generation for return (in the simple case) to avoid doing silly

load/store nonsense in the epilog.  For example, for:

int foo(int X) {
  int A[100];
  return A[X];
}

we used to generate:

  %arrayidx = getelementptr inbounds [100 x i32]* %A, i32 0, i64 %idxprom ; <i32*> [#uses=1]
  %tmp1 = load i32* %arrayidx                     ; <i32> [#uses=1]
  store i32 %tmp1, i32* %retval
  %0 = load i32* %retval                          ; <i32> [#uses=1]
  ret i32 %0
}

which codegen'd to this code:

_foo:                                   ## @foo
## BB#0:                                ## %entry
	subq	$408, %rsp              ## imm = 0x198
	movl	%edi, 400(%rsp)
	movl	400(%rsp), %edi
	movslq	%edi, %rax
	movl	(%rsp,%rax,4), %edi
	movl	%edi, 404(%rsp)
	movl	404(%rsp), %eax
	addq	$408, %rsp              ## imm = 0x198
	ret

Now we generate:

  %arrayidx = getelementptr inbounds [100 x i32]* %A, i32 0, i64 %idxprom ; <i32*> [#uses=1]
  %tmp1 = load i32* %arrayidx                     ; <i32> [#uses=1]
  ret i32 %tmp1
}

and:

_foo:                                   ## @foo
## BB#0:                                ## %entry
	subq	$408, %rsp              ## imm = 0x198
	movl	%edi, 404(%rsp)
	movl	404(%rsp), %edi
	movslq	%edi, %rax
	movl	(%rsp,%rax,4), %eax
	addq	$408, %rsp              ## imm = 0x198
	ret

This actually does matter, cutting out 2000 lines of IR from CGStmt.ll 
for example.

Another interesting effect is that altivec.h functions which are dead
now get dce'd by the inliner.  Hence all the changes to 
builtins-ppc-altivec.c to ensure the calls aren't dead.

llvm-svn: 106970

1 files changed, 37 insertions, 18 deletions

diff --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp
index 283f3610af1..051ef086990 100644
--- a/clang/lib/CodeGen/CGCall.cpp
+++ b/clang/lib/CodeGen/CGCall.cpp
@@ -373,18 +373,18 @@ static llvm::Value *CreateCoercedLoad(llvm::Value *SrcPtr,
     // FIXME: Use better alignment / avoid requiring aligned load.
     Load->setAlignment(1);
     return Load;
-  } else {
-    // Otherwise do coercion through memory. This is stupid, but
-    // simple.
-    llvm::Value *Tmp = CGF.CreateTempAlloca(Ty);
-    llvm::Value *Casted =
-      CGF.Builder.CreateBitCast(Tmp, llvm::PointerType::getUnqual(SrcTy));
-    llvm::StoreInst *Store =
-      CGF.Builder.CreateStore(CGF.Builder.CreateLoad(SrcPtr), Casted);
-    // FIXME: Use better alignment / avoid requiring aligned store.
-    Store->setAlignment(1);
-    return CGF.Builder.CreateLoad(Tmp);
   }
+  
+  // Otherwise do coercion through memory. This is stupid, but
+  // simple.
+  llvm::Value *Tmp = CGF.CreateTempAlloca(Ty);
+  llvm::Value *Casted =
+    CGF.Builder.CreateBitCast(Tmp, llvm::PointerType::getUnqual(SrcTy));
+  llvm::StoreInst *Store =
+    CGF.Builder.CreateStore(CGF.Builder.CreateLoad(SrcPtr), Casted);
+  // FIXME: Use better alignment / avoid requiring aligned store.
+  Store->setAlignment(1);
+  return CGF.Builder.CreateLoad(Tmp);
 }
 
 /// CreateCoercedStore - Create a store to \arg DstPtr from \arg Src,
@@ -798,8 +798,7 @@ void CodeGenFunction::EmitFunctionProlog(const CGFunctionInfo &FI,
   assert(AI == Fn->arg_end() && "Argument mismatch!");
 }
 
-void CodeGenFunction::EmitFunctionEpilog(const CGFunctionInfo &FI,
-                                         llvm::Value *ReturnValue) {
+void CodeGenFunction::EmitFunctionEpilog(const CGFunctionInfo &FI) {
   // Functions with no result always return void.
   if (ReturnValue == 0) {
     Builder.CreateRetVoid();
@@ -824,12 +823,32 @@ void CodeGenFunction::EmitFunctionEpilog(const CGFunctionInfo &FI,
     break;
 
   case ABIArgInfo::Extend:
-  case ABIArgInfo::Direct:
-    // The internal return value temp always will have
-    // pointer-to-return-type type.
-    RV = Builder.CreateLoad(ReturnValue);
+  case ABIArgInfo::Direct: {
+    // The internal return value temp always will have pointer-to-return-type
+    // type, just do a load.
+      
+    // If the instruction right before the insertion point is a store to the
+    // return value, we can elide the load, zap the store, and usually zap the
+    // alloca.
+    llvm::BasicBlock *InsertBB = Builder.GetInsertBlock();
+    llvm::StoreInst *SI = 0;
+    if (InsertBB->empty() || 
+        !(SI = dyn_cast<llvm::StoreInst>(&InsertBB->back())) ||
+        SI->getPointerOperand() != ReturnValue || SI->isVolatile()) {
+      RV = Builder.CreateLoad(ReturnValue);
+    } else {
+      // Get the stored value and nuke the now-dead store.
+      RV = SI->getValueOperand();
+      SI->eraseFromParent();
+      
+      // If that was the only use of the return value, nuke it as well now.
+      if (ReturnValue->use_empty() && isa<llvm::AllocaInst>(ReturnValue)) {
+        cast<llvm::AllocaInst>(ReturnValue)->eraseFromParent();
+        ReturnValue = 0;
+      }
+    }
     break;
-
+  }
   case ABIArgInfo::Ignore:
     break;