[CodeGenPrepare] Remove load-based heuristic

Summary:
Both the hardware and LLVM have changed since 2012.
Now, load-based heuristic don't show big differences any more on OoO cores.

There is no notable regressons and improvements on spec2000/2006. (Cortex-A57, Core i5).

Reviewers: spatel, zansari
    
Differential Revision: http://reviews.llvm.org/D16836

llvm-svn: 261809

llvm/lib/CodeGen/CodeGenPrepare.cpp

@@ -4477,17 +4477,6 @@ static bool isFormingBranchFromSelectProfitable(const TargetTransformInfo *TTI,
   if (!Cmp || !Cmp->hasOneUse())
     return false;
 
-  Value *CmpOp0 = Cmp->getOperand(0);
-  Value *CmpOp1 = Cmp->getOperand(1);
-
-  // Emit "cmov on compare with a memory operand" as a branch to avoid stalls
-  // on a load from memory. But if the load is used more than once, do not
-  // change the select to a branch because the load is probably needed
-  // regardless of whether the branch is taken or not.
-  if ((isa<LoadInst>(CmpOp0) && CmpOp0->hasOneUse()) ||
-      (isa<LoadInst>(CmpOp1) && CmpOp1->hasOneUse()))
-    return true;
-
   // If either operand of the select is expensive and only needed on one side
   // of the select, we should form a branch.
   if (sinkSelectOperand(TTI, SI->getTrueValue()) ||

llvm/test/CodeGen/AArch64/a57-csel.ll

@@ -1,8 +1,9 @@
 ; RUN: llc -mtriple=aarch64-none-linux-gnu < %s -mcpu=cortex-a57 -aarch64-enable-early-ifcvt=false | FileCheck %s
 
-; Check that the select is expanded into a branch sequence.
+; Check that the select isn't expanded into a branch sequence
+; when the icmp's first operand %x0 is from load.
 define i64 @f(i64 %a, i64 %b, i64* %c, i64 %d, i64 %e) {
-  ; CHECK: cbz
+  ; CHECK: csel
   %x0 = load i64, i64* %c
   %x1 = icmp eq i64 %x0, 0
   %x2 = select i1 %x1, i64 %a, i64 %b

llvm/test/CodeGen/X86/cmov-into-branch.ll

@@ -1,6 +1,6 @@
 ; RUN: llc -march=x86-64 -mcpu=core2 < %s | FileCheck %s
 
-; cmp with single-use load, should not form cmov.
+; cmp with single-use load, should not form branch.
 define i32 @test1(double %a, double* nocapture %b, i32 %x, i32 %y)  {
   %load = load double, double* %b, align 8
   %cmp = fcmp olt double %load, %a
@@ -8,9 +8,7 @@ define i32 @test1(double %a, double* nocapture %b, i32 %x, i32 %y)  {
   ret i32 %cond
 ; CHECK-LABEL: test1:
 ; CHECK: ucomisd
-; CHECK-NOT: cmov
-; CHECK: j
-; CHECK-NOT: cmov
+; CHECK: cmovbel
 }
 
 ; Sanity check: no load.
@@ -23,19 +21,6 @@ define i32 @test2(double %a, double %b, i32 %x, i32 %y)  {
 ; CHECK: cmov
 }
 
-; Multiple uses of %a, should not form cmov.
-define i32 @test3(i32 %a, i32* nocapture %b, i32 %x)  {
-  %load = load i32, i32* %b, align 4
-  %cmp = icmp ult i32 %load, %a
-  %cond = select i1 %cmp, i32 %a, i32 %x
-  ret i32 %cond
-; CHECK-LABEL: test3:
-; CHECK: cmpl
-; CHECK-NOT: cmov
-; CHECK: j
-; CHECK-NOT: cmov
-}
-
 ; Multiple uses of the load.
 define i32 @test4(i32 %a, i32* nocapture %b, i32 %x, i32 %y)  {
   %load = load i32, i32* %b, align 4

llvm/test/Transforms/CodeGenPrepare/X86/select.ll

@@ -2,8 +2,7 @@
 
 target triple = "x86_64-unknown-unknown"
 
-; Nothing to sink here, but this gets converted to a branch to
-; avoid stalling an out-of-order CPU on a predictable branch.
+; Nothing to sink and convert here.
 
 define i32 @no_sink(double %a, double* %b, i32 %x, i32 %y)  {
 entry:
@@ -15,11 +14,7 @@ entry:
 ; CHECK-LABEL: @no_sink(
 ; CHECK:    %load = load double, double* %b, align 8
 ; CHECK:    %cmp = fcmp olt double %load, %a
-; CHECK:    br i1 %cmp, label %select.end, label %select.false
-; CHECK:  select.false:
-; CHECK:    br label %select.end
-; CHECK:  select.end:
-; CHECK:    %sel = phi i32 [ %x, %entry ], [ %y, %select.false ] 
+; CHECK:    %sel = select i1 %cmp, i32 %x, i32 %y
 ; CHECK:    ret i32 %sel
 }