Use the 'arcp' fast-math-flag when combining repeated FP divisors

[oota-llvm.git] / test / CodeGen / X86 / fdiv-combine.ll

diff --git a/test/CodeGen/X86/fdiv-combine.ll b/test/CodeGen/X86/fdiv-combine.ll
index b65e9d01ab8ba7118f3603e87563853afb074194..d9d9ac401fb50ae594b45fc097f0858692583c8e 100644 (file)
--- a/test/CodeGen/X86/fdiv-combine.ll
+++ b/test/CodeGen/X86/fdiv-combine.ll
@@ -1,9 +1,11 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 | FileCheck %s
 
-; Anything more than one division using a single divisor operand
+; More than one 'arcp' division using a single divisor operand
 ; should be converted into a reciprocal and multiplication.
 
-define float @div1_arcp(float %x, float %y, float %z) #0 {
+; Don't do anything for just one division.
+
+define float @div1_arcp(float %x, float %y, float %z) {
 ; CHECK-LABEL: div1_arcp:
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    divss %xmm1, %xmm0
@@ -12,13 +14,15 @@ define float @div1_arcp(float %x, float %y, float %z) #0 {
   ret float %div1
 }
 
-define float @div2_arcp(float %x, float %y, float %z) #0 {
-; CHECK-LABEL: div2_arcp:
+; All math instructions are 'arcp', so optimize.
+
+define float @div2_arcp_all(float %x, float %y, float %z) {
+; CHECK-LABEL: div2_arcp_all:
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    movss {{.*#+}} xmm3 = mem[0],zero,zero,zero
 ; CHECK-NEXT:    divss %xmm2, %xmm3
-; CHECK-NEXT:    mulss %xmm1, %xmm0
 ; CHECK-NEXT:    mulss %xmm3, %xmm0
+; CHECK-NEXT:    mulss %xmm1, %xmm0
 ; CHECK-NEXT:    mulss %xmm3, %xmm0
 ; CHECK-NEXT:    retq
   %div1 = fdiv arcp float %x, %z
@@ -27,10 +31,57 @@ define float @div2_arcp(float %x, float %y, float %z) #0 {
   ret float %div2
 }
 
+; The first division is not 'arcp', so do not optimize.
+
+define float @div2_arcp_partial1(float %x, float %y, float %z) {
+; CHECK-LABEL: div2_arcp_partial1:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    divss %xmm2, %xmm0
+; CHECK-NEXT:    mulss %xmm1, %xmm0
+; CHECK-NEXT:    divss %xmm2, %xmm0
+; CHECK-NEXT:    retq
+  %div1 = fdiv float %x, %z
+  %mul = fmul arcp float %div1, %y
+  %div2 = fdiv arcp float %mul, %z
+  ret float %div2
+}
+
+; The second division is not 'arcp', so do not optimize.
+
+define float @div2_arcp_partial2(float %x, float %y, float %z) {
+; CHECK-LABEL: div2_arcp_partial2:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    divss %xmm2, %xmm0
+; CHECK-NEXT:    mulss %xmm1, %xmm0
+; CHECK-NEXT:    divss %xmm2, %xmm0
+; CHECK-NEXT:    retq
+  %div1 = fdiv arcp float %x, %z
+  %mul = fmul arcp float %div1, %y
+  %div2 = fdiv float %mul, %z
+  ret float %div2
+}
+
+; The multiply is not 'arcp', but that does not prevent optimizing the divisions.
+
+define float @div2_arcp_partial3(float %x, float %y, float %z) {
+; CHECK-LABEL: div2_arcp_partial3:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    movss {{.*#+}} xmm3 = mem[0],zero,zero,zero
+; CHECK-NEXT:    divss %xmm2, %xmm3
+; CHECK-NEXT:    mulss %xmm3, %xmm0
+; CHECK-NEXT:    mulss %xmm1, %xmm0
+; CHECK-NEXT:    mulss %xmm3, %xmm0
+; CHECK-NEXT:    retq
+  %div1 = fdiv arcp float %x, %z
+  %mul = fmul float %div1, %y
+  %div2 = fdiv arcp float %mul, %z
+  ret float %div2
+}
+
 ; If the reciprocal is already calculated, we should not
 ; generate an extra multiplication by 1.0. 
 
-define double @div3_arcp(double %x, double %y, double %z) #0 {
+define double @div3_arcp(double %x, double %y, double %z) {
 ; CHECK-LABEL: div3_arcp:
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    movsd{{.*#+}} xmm2 = mem[0],zero
@@ -44,7 +95,7 @@ define double @div3_arcp(double %x, double %y, double %z) #0 {
   ret double %ret
 }
 
-define void @PR24141() #0 {
+define void @PR24141() {
 ; CHECK-LABEL: PR24141:
 ; CHECK:       callq
 ; CHECK-NEXT:  divsd
@@ -57,11 +108,9 @@ while.body:
   %call = call { double, double } @g(double %x.0)
   %xv0 = extractvalue { double, double } %call, 0
   %xv1 = extractvalue { double, double } %call, 1
-  %div = fdiv double %xv0, %xv1
+  %div = fdiv arcp double %xv0, %xv1
   br label %while.body
 }
 
 declare { double, double } @g(double)
 
-; FIXME: If the backend understands 'arcp', then this attribute is unnecessary.
-attributes #0 = { "unsafe-fp-math"="true" }