[DAGCombiner] rename function attribute for disabling ftrunc transform

[spatel]

Summary

This is the matching name change for the Clang patch at D46236.

[spatel]

The only thing up for debate here was the spelling of the attribute, and D46236 was reviewed and accepted with this version of the string, so I'm going to mark this part as accepted too. That way, Phab won't be confused.

We want to commit both of these in close proximity to limit any breakage from mismatch in the attribute name that clang is creating and LLVM is expecting.

llvm/trunk/docs/ReleaseNotes.rst

* A new tool named :doc:`llvm-mca <CommandGuide/llvm-mca>` has been added.
   machine code for a specific CPU.
 
 * The optimization flag to merge constants (-fmerge-all-constants) is no longer
   applied by default.
 
 * Optimization of floating-point casts is improved. This may cause surprising
   results for code that is relying on the undefined behavior of overflowing 
   casts. The optimization can be disabled by specifying a function attribute:
-  "fp-cast-overflow-workaround"="true". This attribute may be created by the
-  clang option :option:`-ffp-cast-overflow-workaround`.
+  "strict-float-cast-overflow"="false". This attribute may be created by the
+  clang option :option:`-fno-strict-float-cast-overflow`.
   Code sanitizers can be used to detect affected patterns. The option for
   detecting this problem alone is "-fsanitize=float-cast-overflow":
 
 .. code-block:: c
 
     int main() {
       float x = 4294967296.0f;
       x = (float)((int)x);

llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

 }
 
 static SDValue foldFPToIntToFP(SDNode *N, SelectionDAG &DAG,
                                const TargetLowering &TLI) {
   // This optimization is guarded by a function attribute because it may produce
   // unexpected results. Ie, programs may be relying on the platform-specific
   // undefined behavior when the float-to-int conversion overflows.
   const Function &F = DAG.getMachineFunction().getFunction();
-  Attribute CastWorkaround = F.getFnAttribute("fp-cast-overflow-workaround");
-  if (CastWorkaround.getValueAsString().equals("true"))
+  Attribute StrictOverflow = F.getFnAttribute("strict-float-cast-overflow");
+  if (StrictOverflow.getValueAsString().equals("false"))
     return SDValue();
 
   // We only do this if the target has legal ftrunc. Otherwise, we'd likely be
   // replacing casts with a libcall.
   EVT VT = N->getValueType(0);
   if (!TLI.isOperationLegal(ISD::FTRUNC, VT))
     return SDValue();
 

llvm/trunk/test/CodeGen/X86/ftrunc.ll

 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vroundpd $11, %ymm0, %ymm0
 ; AVX1-NEXT:    retq
   %i = fptosi <4 x double> %x to <4 x i64>
   %r = sitofp <4 x i64> %i to <4 x double>
   ret <4 x double> %r
 }
 
-; The attribute name is subject to change, but the fold may be
-; guarded to allow existing code to continue working based on its
-; assumptions of float->int overflow.
+; The fold may be guarded to allow existing code to continue 
+; working based on its assumptions of float->int overflow.
 
 define float @trunc_unsigned_f32_disable_via_attr(float %x) #1 {
 ; SSE2-LABEL: trunc_unsigned_f32_disable_via_attr:
 ; SSE2:       # %bb.0:
 ; SSE2-NEXT:    cvttss2si %xmm0, %rax
 ; SSE2-NEXT:    movl %eax, %eax
 ; SSE2-NEXT:    xorps %xmm0, %xmm0
 ; SSE2-NEXT:    cvtsi2ssq %rax, %xmm0

 ; AVX1-NEXT:    vcvttsd2si %xmm0, %rax
 ; AVX1-NEXT:    vcvtsi2sdq %rax, %xmm1, %xmm0
 ; AVX1-NEXT:    retq
   %i = fptosi double %x to i64
   %r = sitofp i64 %i to double
   ret double %r
 }
 
-attributes #1 = { nounwind "fp-cast-overflow-workaround"="true" }
+attributes #1 = { nounwind "strict-float-cast-overflow"="false" }