[llvm] r294011 - [NVPTX] Enable combineRepeatedFPDivisors for NVPTX.

[Justin Lebar via llvm-commits]

Author: jlebar
Date: Fri Feb  3 09:13:50 2017
New Revision: 294011

URL: http://llvm.org/viewvc/llvm-project?rev=294011&view=rev
Log:
[NVPTX] Enable combineRepeatedFPDivisors for NVPTX.

Reviewers: tra

Subscribers: jholewinski, llvm-commits

Differential Revision: https://reviews.llvm.org/D29477

Modified:
    llvm/trunk/lib/Target/NVPTX/NVPTXISelLowering.h
    llvm/trunk/test/CodeGen/NVPTX/fast-math.ll

Modified: llvm/trunk/lib/Target/NVPTX/NVPTXISelLowering.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/NVPTX/NVPTXISelLowering.h?rev=294011&r1=294010&r2=294011&view=diff
==============================================================================
--- llvm/trunk/lib/Target/NVPTX/NVPTXISelLowering.h (original)
+++ llvm/trunk/lib/Target/NVPTX/NVPTXISelLowering.h Fri Feb  3 09:13:50 2017
@@ -530,6 +530,8 @@ public:
                           int &ExtraSteps, bool &UseOneConst,
                           bool Reciprocal) const override;
 
+  unsigned combineRepeatedFPDivisors() const override { return 2; }
+
   bool allowFMA(MachineFunction &MF, CodeGenOpt::Level OptLevel) const;
   bool allowUnsafeFPMath(MachineFunction &MF) const;
 

Modified: llvm/trunk/test/CodeGen/NVPTX/fast-math.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/NVPTX/fast-math.ll?rev=294011&r1=294010&r2=294011&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/NVPTX/fast-math.ll (original)
+++ llvm/trunk/test/CodeGen/NVPTX/fast-math.ll Fri Feb  3 09:13:50 2017
@@ -117,5 +117,49 @@ define float @fcos_approx(float %a) #0 {
   ret float %r
 }
 
+; CHECK-LABEL: repeated_div_recip_allowed
+define float @repeated_div_recip_allowed(i1 %pred, float %a, float %b, float %divisor) {
+; CHECK: rcp.rn.f32
+; CHECK: mul.rn.f32
+; CHECK: mul.rn.f32
+  %x = fdiv arcp float %a, %divisor
+  %y = fdiv arcp float %b, %divisor
+  %z = select i1 %pred, float %x, float %y
+  ret float %z
+}
+
+; CHECK-LABEL: repeated_div_recip_allowed_ftz
+define float @repeated_div_recip_allowed_ftz(i1 %pred, float %a, float %b, float %divisor) #1 {
+; CHECK: rcp.rn.ftz.f32
+; CHECK: mul.rn.ftz.f32
+; CHECK: mul.rn.ftz.f32
+  %x = fdiv arcp float %a, %divisor
+  %y = fdiv arcp float %b, %divisor
+  %z = select i1 %pred, float %x, float %y
+  ret float %z
+}
+
+; CHECK-LABEL: repeated_div_fast
+define float @repeated_div_fast(i1 %pred, float %a, float %b, float %divisor) #0 {
+; CHECK: rcp.approx.f32
+; CHECK: mul.f32
+; CHECK: mul.f32
+  %x = fdiv float %a, %divisor
+  %y = fdiv float %b, %divisor
+  %z = select i1 %pred, float %x, float %y
+  ret float %z
+}
+
+; CHECK-LABEL: repeated_div_fast_ftz
+define float @repeated_div_fast_ftz(i1 %pred, float %a, float %b, float %divisor) #0 #1 {
+; CHECK: rcp.approx.ftz.f32
+; CHECK: mul.ftz.f32
+; CHECK: mul.ftz.f32
+  %x = fdiv float %a, %divisor
+  %y = fdiv float %b, %divisor
+  %z = select i1 %pred, float %x, float %y
+  ret float %z
+}
+
 attributes #0 = { "unsafe-fp-math" = "true" }
 attributes #1 = { "nvptx-f32ftz" = "true" }