[AArch64] Fix incorrect CSEL node created

[rogfer01]

Summary

Under -enable-unsafe-fp-math SELECT_CC lowering in AArch64
transforms floating point comparisons of the form "a == 0.0 ? 0.0 : x" to
"a == 0.0 ? a : x". But it incorrectly assumes that 'x' and 'a' have
the same type which can lead to a wrong CSEL node that fails later
due to nonsensical copies.

This issue was introduced in https://reviews.llvm.org/rL284531

[jmolloy]

Hi Roger,

This looks like a sensible fix - thanks for doing this!

If I understand right, the reason your testcase doesn't have a FileCheck invocation is because the previous behaviour was a crash. Could you please mention this as a comment in the testcase?

With that fixed, LGTM.

James

[rogfer01]

Mention in testcase that we are checking that we do not crash with the given input.

lib/Target/AArch64/AArch64ISelLowering.cpp

if (DAG.getTarget().Options.UnsafeFPMath) {
     // Transform "a == 0.0 ? 0.0 : x" to "a == 0.0 ? a : x" and
     // "a != 0.0 ? x : 0.0" to "a != 0.0 ? x : a" to avoid materializing 0.0.
     ConstantFPSDNode *RHSVal = dyn_cast<ConstantFPSDNode>(RHS);
     if (RHSVal && RHSVal->isZero()) {
       ConstantFPSDNode *CFVal = dyn_cast<ConstantFPSDNode>(FVal);
       ConstantFPSDNode *CTVal = dyn_cast<ConstantFPSDNode>(TVal);
 
       if ((CC == ISD::SETEQ || CC == ISD::SETOEQ || CC == ISD::SETUEQ) &&
-          CTVal && CTVal->isZero())
+          CTVal && CTVal->isZero() && TVal.getValueType() == LHS.getValueType())
         TVal = LHS;
       else if ((CC == ISD::SETNE || CC == ISD::SETONE || CC == ISD::SETUNE) &&
-               CFVal && CFVal->isZero())
+               CFVal && CFVal->isZero() &&
+               FVal.getValueType() == LHS.getValueType())
         FVal = LHS;
     }
   }
 
   // Emit first, and possibly only, CSEL.
   SDValue CC1Val = DAG.getConstant(CC1, dl, MVT::i32);
   SDValue CS1 = DAG.getNode(AArch64ISD::CSEL, dl, VT, TVal, FVal, CC1Val, Cmp);
 

test/CodeGen/AArch64/csel-zero-float.ll

+; RUN: llc -mtriple=aarch64-none-linux-gnu -enable-unsafe-fp-math < %s
+; There is no invocation to FileCheck as this
+; caused a crash in "Post-RA pseudo instruction expansion"
+
+define double @foo(float *%user, float %t17) {
+  %t16 = load float, float* %user, align 8
+  %conv = fpext float %t16 to double
+  %cmp26 = fcmp fast oeq float %t17, 0.000000e+00
+  %div = fdiv fast float %t16, %t17
+  %div.op = fmul fast float %div, 1.000000e+02
+  %t18 = fpext float %div.op to double
+  %conv31 = select i1 %cmp26, double 0.000000e+00, double %t18
+  ret double %conv31
+}
+