[PATCH] D125671: [InstCombine] Allow undef vectors when foldSelectToCopysign

[Chenbing.Zheng via Phabricator via llvm-commits]

This revision was landed with ongoing or failed builds.
This revision was automatically updated to reflect the committed changes.
Closed by commit rG51df77f36d8e: [InstCombine] Allow undef vectors when foldSelectToCopysign (authored by Chenbing.Zheng).

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D125671/new/

https://reviews.llvm.org/D125671

Files:
  llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
  llvm/test/Transforms/InstCombine/select.ll


Index: llvm/test/Transforms/InstCombine/select.ll
===================================================================
--- llvm/test/Transforms/InstCombine/select.ll
+++ llvm/test/Transforms/InstCombine/select.ll
@@ -1689,13 +1689,10 @@
   ret float %r
 }
 
-; TODO: Allow undefs when matching vectors.
-
 define <2 x float> @copysign_vec_undef(<2 x float> %x) {
 ; CHECK-LABEL: @copysign_vec_undef(
-; CHECK-NEXT:    [[I:%.*]] = bitcast <2 x float> [[X:%.*]] to <2 x i32>
-; CHECK-NEXT:    [[ISNEG:%.*]] = icmp slt <2 x i32> [[I]], zeroinitializer
-; CHECK-NEXT:    [[R:%.*]] = select nnan arcp <2 x i1> [[ISNEG]], <2 x float> <float 4.200000e+01, float undef>, <2 x float> <float -4.200000e+01, float -4.200000e+01>
+; CHECK-NEXT:    [[TMP1:%.*]] = fneg <2 x float> [[X:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = call <2 x float> @llvm.copysign.v2f32(<2 x float> <float 4.200000e+01, float 4.200000e+01>, <2 x float> [[TMP1]])
 ; CHECK-NEXT:    ret <2 x float> [[R]]
 ;
   %i = bitcast <2 x float> %x to <2 x i32>
@@ -1706,27 +1703,23 @@
 
 define <2 x float> @copysign_vec_undef1(<2 x float> %x) {
 ; CHECK-LABEL: @copysign_vec_undef1(
-; CHECK-NEXT:    [[I:%.*]] = bitcast <2 x float> [[X:%.*]] to <2 x i32>
-; CHECK-NEXT:    [[ISNEG:%.*]] = icmp slt <2 x i32> [[I]], zeroinitializer
-; CHECK-NEXT:    [[R:%.*]] = select nnan arcp <2 x i1> [[ISNEG]], <2 x float> <float 4.200000e+01, float 4.200000e+01>, <2 x float> <float undef, float -4.200000e+01>
+; CHECK-NEXT:    [[R:%.*]] = call <2 x float> @llvm.copysign.v2f32(<2 x float> <float 4.200000e+01, float 4.200000e+01>, <2 x float> [[X:%.*]])
 ; CHECK-NEXT:    ret <2 x float> [[R]]
 ;
   %i = bitcast <2 x float> %x to <2 x i32>
-  %isneg = icmp ugt <2 x i32> %i, <i32 2147483647, i32 2147483647>
+  %isneg = icmp ult <2 x i32> %i, <i32 2147483648, i32 2147483648>
   %r = select arcp nnan <2 x i1> %isneg, <2 x float> <float 42.0, float 42.0>, <2 x float> <float undef, float -42.0>
   ret <2 x float> %r
 }
 
 define <2 x float> @copysign_vec_undef3(<2 x float> %x) {
 ; CHECK-LABEL: @copysign_vec_undef3(
-; CHECK-NEXT:    [[I:%.*]] = bitcast <2 x float> [[X:%.*]] to <2 x i32>
-; CHECK-NEXT:    [[ISNEG:%.*]] = icmp slt <2 x i32> [[I]], zeroinitializer
-; CHECK-NEXT:    [[R:%.*]] = select nnan arcp <2 x i1> [[ISNEG]], <2 x float> <float 4.200000e+01, float undef>, <2 x float> <float -4.200000e+01, float undef>
+; CHECK-NEXT:    [[R:%.*]] = call <2 x float> @llvm.copysign.v2f32(<2 x float> <float 4.200000e+01, float 4.200000e+01>, <2 x float> [[X:%.*]])
 ; CHECK-NEXT:    ret <2 x float> [[R]]
 ;
   %i = bitcast <2 x float> %x to <2 x i32>
   %isneg = icmp ugt <2 x i32> %i, <i32 2147483647, i32 2147483647>
-  %r = select arcp nnan <2 x i1> %isneg, <2 x float> <float 42.0, float undef>, <2 x float> <float -42.0, float undef>
+  %r = select arcp nnan <2 x i1> %isneg, <2 x float> <float -42.0, float undef>, <2 x float> <float +42.0, float undef>
   ret <2 x float> %r
 }
 
Index: llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
===================================================================
--- llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -2276,7 +2276,8 @@
   // Match select ?, TC, FC where the constants are equal but negated.
   // TODO: Generalize to handle a negated variable operand?
   const APFloat *TC, *FC;
-  if (!match(TVal, m_APFloat(TC)) || !match(FVal, m_APFloat(FC)) ||
+  if (!match(TVal, m_APFloatAllowUndef(TC)) ||
+      !match(FVal, m_APFloatAllowUndef(FC)) ||
       !abs(*TC).bitwiseIsEqual(abs(*FC)))
     return nullptr;
 
@@ -2302,7 +2303,7 @@
 
   // Canonicalize the magnitude argument as the positive constant since we do
   // not care about its sign.
-  Value *MagArg = TC->isNegative() ? FVal : TVal;
+  Value *MagArg = ConstantFP::get(SelType, abs(*TC));
   Function *F = Intrinsic::getDeclaration(Sel.getModule(), Intrinsic::copysign,
                                           Sel.getType());
   return CallInst::Create(F, { MagArg, X });


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D125671.430560.patch
Type: text/x-patch
Size: 4021 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220519/ccacadc3/attachment.bin>