Allow speculating llvm.sqrt, fma and fmuladd

[arsenm]

Summary

This should be OK since the sqrt intrinsic isn't supposed to be able to set errno, and the standard fma function doesn't anyway. The comment mentions others could be, but the documentation for these is very unclear. sqrt specifically says it doesn't set errno, but the others don't, though it probably shouldn't be different.

[arsenm]

Explicitly state in the documentation that llvm.fma and llvm.fmuladd do not set errno.

[arsenm]

Ping

[hfinkel]

LGTM.

As you point out, we should fix up the definitions of the other intrinsics w.r.t. errno, but that's a separate issue. (I had a patch to do this: http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20131111/194571.html -- but it seems that unfortunately I forgot about it -- perhaps you can comment on this at some point).

[arsenm]

r200501

[arsenm]

r200501

docs/LangRef.rst

 """"""""""
 
 This function returns the same values as the libm ``fma`` functions
-would.
+would, and does not set errno.
 
 '``llvm.fabs.*``' Intrinsic
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^

 not be performed between the multiplication and addition steps if the
 code generator fuses the operations. Fusion is not guaranteed, even if
 the target platform supports it. If a fused multiply-add is required the
-corresponding llvm.fma.\* intrinsic function should be used instead.
+corresponding llvm.fma.\* intrinsic function should be used
+instead. This never sets errno, just as '``llvm.fma.*``'.
 
 Examples:
 """""""""

lib/Analysis/ValueTracking.cpp

        case Intrinsic::umul_with_overflow:
        case Intrinsic::usub_with_overflow:
          return true;
+       // Sqrt should be OK, since the llvm sqrt intrinsic isn't defined to set
+       // errno like libm sqrt would.
+       case Intrinsic::sqrt:
+       case Intrinsic::fma:
+       case Intrinsic::fmuladd:
+         return true;
        // TODO: some fp intrinsics are marked as having the same error handling
        // as libm. They're safe to speculate when they won't error.
        // TODO: are convert_{from,to}_fp16 safe?

test/Transforms/SimplifyCFG/speculate-math.ll

+; RUN: opt -S -simplifycfg -phi-node-folding-threshold=2 < %s | FileCheck %s
+
+declare float @llvm.sqrt.f32(float) nounwind readonly
+declare float @llvm.fma.f32(float, float, float) nounwind readonly
+declare float @llvm.fmuladd.f32(float, float, float) nounwind readonly
+
+; CHECK-LABEL: @sqrt_test(
+; CHECK-NOT: call
+define void @sqrt_test(float addrspace(1)* noalias nocapture %out, float %a) nounwind {
+entry:
+  %cmp.i = fcmp olt float %a, 0.000000e+00
+  br i1 %cmp.i, label %test_sqrt.exit, label %cond.else.i
+
+cond.else.i:                                      ; preds = %entry
+  %0 = tail call float @llvm.sqrt.f32(float %a) nounwind readnone
+  br label %test_sqrt.exit
+
+test_sqrt.exit:                                   ; preds = %cond.else.i, %entry
+  %cond.i = phi float [ %0, %cond.else.i ], [ 0x7FF8000000000000, %entry ]
+  store float %cond.i, float addrspace(1)* %out, align 4
+  ret void
+}
+
+
+; CHECK-LABEL: @fma_test(
+; CHECK-NOT: call
+define void @fma_test(float addrspace(1)* noalias nocapture %out, float %a, float %b, float %c) nounwind {
+entry:
+  %cmp.i = fcmp olt float %a, 0.000000e+00
+  br i1 %cmp.i, label %test_fma.exit, label %cond.else.i
+
+cond.else.i:                                      ; preds = %entry
+  %0 = tail call float @llvm.fma.f32(float %a, float %b, float %c) nounwind readnone
+  br label %test_fma.exit
+
+test_fma.exit:                                   ; preds = %cond.else.i, %entry
+  %cond.i = phi float [ %0, %cond.else.i ], [ 0x7FF8000000000000, %entry ]
+  store float %cond.i, float addrspace(1)* %out, align 4
+  ret void
+}
+
+; CHECK-LABEL: @fmuladd_test(
+; CHECK-NOT: call
+define void @fmuladd_test(float addrspace(1)* noalias nocapture %out, float %a, float %b, float %c) nounwind {
+entry:
+  %cmp.i = fcmp olt float %a, 0.000000e+00
+  br i1 %cmp.i, label %test_fmuladd.exit, label %cond.else.i
+
+cond.else.i:                                      ; preds = %entry
+  %0 = tail call float @llvm.fmuladd.f32(float %a, float %b, float %c) nounwind readnone
+  br label %test_fmuladd.exit
+
+test_fmuladd.exit:                                   ; preds = %cond.else.i, %entry
+  %cond.i = phi float [ %0, %cond.else.i ], [ 0x7FF8000000000000, %entry ]
+  store float %cond.i, float addrspace(1)* %out, align 4
+  ret void
+}
+