guard fneg with fmf sub flags

[mcberg2017]

Summary

This change uses fmf subflags to guard optimizations as well as unsafe. These changes originated from D46483.

[mcberg2017]

This review only contains one context, namely fneg. Please comment if needed, otherwise perhaps we can proceed.

[spatel]

lib/CodeGen/SelectionDAG/SelectionDAG.cpp

4052–4053

The FIXME comment is stale, and it should be removed.

The fneg does have FMF to propagate, but we're not doing that here, and that's a mistake visible in the test case.

test/CodeGen/PowerPC/fmf-propagation.ll

39–49

Please add this test as an NFC preliminary commit. That way, we won't have the constant pool noise in the unrelated tests. You might also just want to put this test at the bottom of the file to avoid that noise.

[mcberg2017]

Cleanup up code. Move test to bottom of lit test file, staged initial fneg_fsub test and checked in, with modifications here.

[spatel]

There's still an open question about why we're doing folds like this in getNode(), but I guess we'll answer that in another patch.

LGTM.

lib/CodeGen/SelectionDAG/SelectionDAG.cpp

case ISD::SCALAR_TO_VECTOR:
     if (OpOpcode == ISD::EXTRACT_VECTOR_ELT &&
         isa<ConstantSDNode>(Operand.getOperand(1)) &&
         Operand.getConstantOperandVal(1) == 0 &&
         Operand.getOperand(0).getValueType() == VT)
       return Operand.getOperand(0);
     break;
   case ISD::FNEG:
     // -(X-Y) -> (Y-X) is unsafe because when X==Y, -0.0 != +0.0
-    if (getTarget().Options.UnsafeFPMath && OpOpcode == ISD::FSUB)
-      // FIXME: FNEG has no fast-math-flags to propagate; use the FSUB's flags?
+    if ((getTarget().Options.UnsafeFPMath || Flags.hasNoSignedZeros()) &&
+        OpOpcode == ISD::FSUB)
       return getNode(ISD::FSUB, DL, VT, Operand.getOperand(1),
-                     Operand.getOperand(0), Operand.getNode()->getFlags());
+                     Operand.getOperand(0), Flags);
     if (OpOpcode == ISD::FNEG)  // --X -> X
       return Operand.getOperand(0);
     break;
   case ISD::FABS:
     if (OpOpcode == ISD::FNEG)  // abs(-X) -> abs(X)
       return getNode(ISD::FABS, DL, VT, Operand.getOperand(0));
     break;
   }

test/CodeGen/PowerPC/fmf-propagation.ll

 ; GLOBAL-NEXT:    xsmaddasp 3, 1, 2
 ; GLOBAL-NEXT:    fmr 1, 3
 ; GLOBAL-NEXT:    blr
   %mul = fmul float %x, %y
   %add = fadd contract float %mul, %z
   ret float %add
 }
 
 ; This shouldn't change anything - the intermediate fmul result is now also flagged.
 
 ; FMFDEBUG-LABEL: Optimized lowered selection DAG: %bb.0 'fmul_fadd_contract2:'
 ; FMFDEBUG:         fma {{t[0-9]+}}, {{t[0-9]+}}, {{t[0-9]+}}
 ; FMFDEBUG:       Type-legalized selection DAG: %bb.0 'fmul_fadd_contract2:'
 
 define float @fmul_fadd_contract2(float %x, float %y, float %z) {
 ; FMF-LABEL: fmul_fadd_contract2:
 ; FMF:       # %bb.0:
 ; FMF-NEXT:    xsmaddasp 3, 1, 2
 ; FMF-NEXT:    fmr 1, 3
 ; FMF-NEXT:    blr
 ;
 ; GLOBAL-LABEL: fmul_fadd_contract2:
 ; GLOBAL:       # %bb.0:
 ; GLOBAL-NEXT:    xsmaddasp 3, 1, 2
 ; GLOBAL-NEXT:    fmr 1, 3
 ; GLOBAL-NEXT:    blr
   %mul = fmul contract float %x, %y

 ; GLOBAL-NEXT:    blr
   %r = call afn double @log2(double %x)
   ret double %r
 }
 
 ; -(X - Y) --> (Y - X)
 
 ; FMFDEBUG-LABEL: Optimized lowered selection DAG: %bb.0 'fneg_fsub_nozeros_1:'
-; FMFDEBUG:         fsub {{t[0-9]+}}, {{t[0-9]+}}
+; FMFDEBUG:         fsub nsz {{t[0-9]+}}, {{t[0-9]+}}
 ; FMFDEBUG:       Type-legalized selection DAG: %bb.0 'fneg_fsub_nozeros_1:'
 
 ; GLOBALDEBUG-LABEL: Optimized lowered selection DAG: %bb.0 'fneg_fsub_nozeros_1:'
-; GLOBALDEBUG:         fsub {{t[0-9]+}}, {{t[0-9]+}}
+; GLOBALDEBUG:         fsub nsz {{t[0-9]+}}, {{t[0-9]+}}
 ; GLOBALDEBUG:       Type-legalized selection DAG: %bb.0 'fneg_fsub_nozeros_1:'
 
 define float @fneg_fsub_nozeros_1(float %x, float %y, float %z) {
 ; FMF-LABEL: fneg_fsub_nozeros_1:
 ; FMF:       # %bb.0:
-; FMF-NEXT:    xssubsp 0, 1, 2
-; FMF-NEXT:    xxlxor 13, 13, 13
-; FMF-NEXT:    xssubsp 1, 13, 0
+; FMF-NEXT:    xssubsp 1, 2, 1
 ; FMF-NEXT:    blr
 ;
 ; GLOBAL-LABEL: fneg_fsub_nozeros_1:
 ; GLOBAL:       # %bb.0:
-; GLOBAL-NEXT:    xssubsp 0, 1, 2
-; GLOBAL-NEXT:    xxlxor 13, 13, 13
-; GLOBAL-NEXT:    xssubsp 1, 13, 0
+; GLOBAL-NEXT:    xssubsp 1, 2, 1
 ; GLOBAL-NEXT:    blr
   %neg = fsub float %x, %y
-  %add = fsub float 0.0, %neg
+  %add = fsub nsz float 0.0, %neg
   ret float %add
 }
+