Allow speculating llvm.sqrt, fma and fmuladd

This doesn't set errno, so this should be OK. Also update the documentation to explicitly state that errno are not set. llvm-svn: 200501
llvm · Jan 31, 2014 · ee364ee · ee364ee
1 parent f33d8b9
commit ee364ee
Show file tree

Hide file tree

Showing 3 changed files with 67 additions and 2 deletions.
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
@@ -7490,7 +7490,7 @@ Semantics:
 """"""""""
 
 This function returns the same values as the libm ``fma`` functions
-would.
+would, and does not set errno.
 
 '``llvm.fabs.*``' Intrinsic
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
@@ -8298,7 +8298,8 @@ is equivalent to the expression a \* b + c, except that rounding will
 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:
 """""""""

diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
@@ -2036,6 +2036,12 @@ bool llvm::isSafeToSpeculativelyExecute(const Value *V,
        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?

diff --git a/llvm/test/Transforms/SimplifyCFG/speculate-math.ll b/llvm/test/Transforms/SimplifyCFG/speculate-math.ll
@@ -0,0 +1,58 @@
+; 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: select
+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: select
+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: select
+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
+}
+