ARM64: fix fmsub patterns which assumed accum operand was first

Confusingly, the NEON fmla instructions put the accumulator first but the
scalar versions put it at the end (like the fma lib function & LLVM's
intrinsic).

This should fix PR19345, assuming there's only one issue.

llvm-svn: 205758

llvm/lib/Target/ARM64/ARM64InstrInfo.td

@@ -2009,11 +2009,14 @@ defm FNMSUB : ThreeOperandFPData<1, 1, "fnmsub",
 
 // The following def pats catch the case where the LHS of an FMA is negated.
 // The TriOpFrag above catches the case where the middle operand is negated.
-def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Rd)),
-          (FMSUBSrrr FPR32:$Rd, FPR32:$Rn, FPR32:$Rm)>;
 
-def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Rd)),
+// N.b. FMSUB etc have the accumulator at the *end* of (outs), unlike
-          (FMSUBDrrr FPR64:$Rd, FPR64:$Rn, FPR64:$Rm)>;
+// the NEON variant.
+def : Pat<(f32 (fma (fneg FPR32:$Rn), FPR32:$Rm, FPR32:$Ra)),
+          (FMSUBSrrr FPR32:$Rn, FPR32:$Rm, FPR32:$Ra)>;
+
+def : Pat<(f64 (fma (fneg FPR64:$Rn), FPR64:$Rm, FPR64:$Ra)),
+          (FMSUBDrrr FPR64:$Rn, FPR64:$Rm, FPR64:$Ra)>;
 
 //===----------------------------------------------------------------------===//
 // Floating point comparison instructions.

llvm/test/CodeGen/ARM64/fmadd.ll

@@ -3,7 +3,7 @@
 define float @fma32(float %a, float %b, float %c) nounwind readnone ssp {
 entry:
 ; CHECK-LABEL: fma32:
-; CHECK: fmadd
+; CHECK: fmadd s0, s0, s1, s2
   %0 = tail call float @llvm.fma.f32(float %a, float %b, float %c)
   ret float %0
 }
@@ -11,7 +11,7 @@ entry:
 define float @fnma32(float %a, float %b, float %c) nounwind readnone ssp {
 entry:
 ; CHECK-LABEL: fnma32:
-; CHECK: fnmadd
+; CHECK: fnmadd s0, s0, s1, s2
   %0 = tail call float @llvm.fma.f32(float %a, float %b, float %c)
   %mul = fmul float %0, -1.000000e+00
   ret float %mul
@@ -20,7 +20,7 @@ entry:
 define float @fms32(float %a, float %b, float %c) nounwind readnone ssp {
 entry:
 ; CHECK-LABEL: fms32:
-; CHECK: fmsub
+; CHECK: fmsub s0, s0, s1, s2
   %mul = fmul float %b, -1.000000e+00
   %0 = tail call float @llvm.fma.f32(float %a, float %mul, float %c)
   ret float %0
@@ -29,7 +29,7 @@ entry:
 define float @fms32_com(float %a, float %b, float %c) nounwind readnone ssp {
 entry:
 ; CHECK-LABEL: fms32_com:
-; CHECK: fmsub
+; CHECK: fmsub s0, s1, s0, s2
   %mul = fmul float %b, -1.000000e+00
   %0 = tail call float @llvm.fma.f32(float %mul, float %a, float %c)
   ret float %0
@@ -38,7 +38,7 @@ entry:
 define float @fnms32(float %a, float %b, float %c) nounwind readnone ssp {
 entry:
 ; CHECK-LABEL: fnms32:
-; CHECK: fnmsub
+; CHECK: fnmsub s0, s0, s1, s2
   %mul = fmul float %c, -1.000000e+00
   %0 = tail call float @llvm.fma.f32(float %a, float %b, float %mul)
   ret float %0
@@ -46,7 +46,7 @@ entry:
 
 define double @fma64(double %a, double %b, double %c) nounwind readnone ssp {
 ; CHECK-LABEL: fma64:
-; CHECK: fmadd
+; CHECK: fmadd d0, d0, d1, d2
 entry:
   %0 = tail call double @llvm.fma.f64(double %a, double %b, double %c)
   ret double %0
@@ -54,7 +54,7 @@ entry:
 
 define double @fnma64(double %a, double %b, double %c) nounwind readnone ssp {
 ; CHECK-LABEL: fnma64:
-; CHECK: fnmadd
+; CHECK: fnmadd d0, d0, d1, d2
 entry:
   %0 = tail call double @llvm.fma.f64(double %a, double %b, double %c)
   %mul = fmul double %0, -1.000000e+00
@@ -63,7 +63,7 @@ entry:
 
 define double @fms64(double %a, double %b, double %c) nounwind readnone ssp {
 ; CHECK-LABEL: fms64:
-; CHECK: fmsub
+; CHECK: fmsub d0, d0, d1, d2
 entry:
   %mul = fmul double %b, -1.000000e+00
   %0 = tail call double @llvm.fma.f64(double %a, double %mul, double %c)
@@ -72,7 +72,7 @@ entry:
 
 define double @fms64_com(double %a, double %b, double %c) nounwind readnone ssp {
 ; CHECK-LABEL: fms64_com:
-; CHECK: fmsub
+; CHECK: fmsub d0, d1, d0, d2
 entry:
   %mul = fmul double %b, -1.000000e+00
   %0 = tail call double @llvm.fma.f64(double %mul, double %a, double %c)
@@ -81,7 +81,7 @@ entry:
 
 define double @fnms64(double %a, double %b, double %c) nounwind readnone ssp {
 ; CHECK-LABEL: fnms64:
-; CHECK: fnmsub
+; CHECK: fnmsub d0, d0, d1, d2
 entry:
   %mul = fmul double %c, -1.000000e+00
   %0 = tail call double @llvm.fma.f64(double %a, double %b, double %mul)