[PATCH] D55182: InstCombine: Scalarize single use icmp/fcmp

[Matt Arsenault via Phabricator via llvm-commits]

arsenm updated this revision to Diff 176918.
arsenm added a comment.

Added more tests in r348456, use match


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D55182/new/

https://reviews.llvm.org/D55182

Files:
  lib/Transforms/InstCombine/InstCombineVectorOps.cpp
  test/Transforms/InstCombine/scalarization.ll


Index: test/Transforms/InstCombine/scalarization.ll
===================================================================
--- test/Transforms/InstCombine/scalarization.ll
+++ test/Transforms/InstCombine/scalarization.ll
@@ -137,10 +137,10 @@
 
 define i1 @cheap_to_extract_icmp(<4 x i32> %x, <4 x i1> %y) {
 ; CHECK-LABEL: @cheap_to_extract_icmp(
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq <4 x i32> [[X:%.*]], zeroinitializer
-; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x i1> [[CMP]], i32 2
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x i1> [[Y:%.*]], i32 2
-; CHECK-NEXT:    [[R:%.*]] = and i1 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x i32> [[X:%.*]], i32 2
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x i1> [[Y:%.*]], i32 2
+; CHECK-NEXT:    [[R:%.*]] = and i1 [[TMP2]], [[TMP3]]
 ; CHECK-NEXT:    ret i1 [[R]]
 ;
   %cmp = icmp eq <4 x i32> %x, zeroinitializer
@@ -151,10 +151,10 @@
 
 define i1 @cheap_to_extract_fcmp(<4 x float> %x, <4 x i1> %y) {
 ; CHECK-LABEL: @cheap_to_extract_fcmp(
-; CHECK-NEXT:    [[CMP:%.*]] = fcmp oeq <4 x float> [[X:%.*]], zeroinitializer
-; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x i1> [[CMP]], i32 2
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x i1> [[Y:%.*]], i32 2
-; CHECK-NEXT:    [[R:%.*]] = and i1 [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> [[X:%.*]], i32 2
+; CHECK-NEXT:    [[TMP2:%.*]] = fcmp oeq float [[TMP1]], 0.000000e+00
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x i1> [[Y:%.*]], i32 2
+; CHECK-NEXT:    [[R:%.*]] = and i1 [[TMP2]], [[TMP3]]
 ; CHECK-NEXT:    ret i1 [[R]]
 ;
   %cmp = fcmp oeq <4 x float> %x, zeroinitializer
@@ -165,8 +165,8 @@
 
 define i1 @extractelt_vector_icmp_constrhs(<2 x i32> %arg) {
 ; CHECK-LABEL: @extractelt_vector_icmp_constrhs(
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq <2 x i32> [[ARG:%.*]], zeroinitializer
-; CHECK-NEXT:    [[EXT:%.*]] = extractelement <2 x i1> [[CMP]], i32 0
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x i32> [[ARG:%.*]], i32 0
+; CHECK-NEXT:    [[EXT:%.*]] = icmp eq i32 [[TMP1]], 0
 ; CHECK-NEXT:    ret i1 [[EXT]]
 ;
   %cmp = icmp eq <2 x i32> %arg, zeroinitializer
@@ -176,8 +176,8 @@
 
 define i1 @extractelt_vector_fcmp_constrhs(<2 x float> %arg) {
 ; CHECK-LABEL: @extractelt_vector_fcmp_constrhs(
-; CHECK-NEXT:    [[CMP:%.*]] = fcmp oeq <2 x float> [[ARG:%.*]], zeroinitializer
-; CHECK-NEXT:    [[EXT:%.*]] = extractelement <2 x i1> [[CMP]], i32 0
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x float> [[ARG:%.*]], i32 0
+; CHECK-NEXT:    [[EXT:%.*]] = fcmp oeq float [[TMP1]], 0.000000e+00
 ; CHECK-NEXT:    ret i1 [[EXT]]
 ;
   %cmp = fcmp oeq <2 x float> %arg, zeroinitializer
@@ -187,8 +187,8 @@
 
 define i1 @extractelt_vector_icmp_constrhs_dynidx(<2 x i32> %arg, i32 %idx) {
 ; CHECK-LABEL: @extractelt_vector_icmp_constrhs_dynidx(
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq <2 x i32> [[ARG:%.*]], zeroinitializer
-; CHECK-NEXT:    [[EXT:%.*]] = extractelement <2 x i1> [[CMP]], i32 [[IDX:%.*]]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x i32> [[ARG:%.*]], i32 [[IDX:%.*]]
+; CHECK-NEXT:    [[EXT:%.*]] = icmp eq i32 [[TMP1]], 0
 ; CHECK-NEXT:    ret i1 [[EXT]]
 ;
   %cmp = icmp eq <2 x i32> %arg, zeroinitializer
@@ -198,8 +198,8 @@
 
 define i1 @extractelt_vector_fcmp_constrhs_dynidx(<2 x float> %arg, i32 %idx) {
 ; CHECK-LABEL: @extractelt_vector_fcmp_constrhs_dynidx(
-; CHECK-NEXT:    [[CMP:%.*]] = fcmp oeq <2 x float> [[ARG:%.*]], zeroinitializer
-; CHECK-NEXT:    [[EXT:%.*]] = extractelement <2 x i1> [[CMP]], i32 [[IDX:%.*]]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x float> [[ARG:%.*]], i32 [[IDX:%.*]]
+; CHECK-NEXT:    [[EXT:%.*]] = fcmp oeq float [[TMP1]], 0.000000e+00
 ; CHECK-NEXT:    ret i1 [[EXT]]
 ;
   %cmp = fcmp oeq <2 x float> %arg, zeroinitializer
Index: lib/Transforms/InstCombine/InstCombineVectorOps.cpp
===================================================================
--- lib/Transforms/InstCombine/InstCombineVectorOps.cpp
+++ lib/Transforms/InstCombine/InstCombineVectorOps.cpp
@@ -310,6 +310,16 @@
     return BinaryOperator::CreateWithCopiedFlags(BO->getOpcode(), E0, E1, BO);
   }
 
+  Value *X, *Y;
+  CmpInst::Predicate Pred;
+  if (match(SrcVec, m_Cmp(Pred, m_Value(X), m_Value(Y))) &&
+      cheapToScalarize(SrcVec, IndexC)) {
+    // extelt (cmp X, Y), Index --> cmp (extelt X, Index), (extelt Y, Index)
+    Value *E0 = Builder.CreateExtractElement(X, Index);
+    Value *E1 = Builder.CreateExtractElement(Y, Index);
+    return CmpInst::Create(cast<CmpInst>(SrcVec)->getOpcode(), Pred, E0, E1);
+  }
+
   if (auto *I = dyn_cast<Instruction>(SrcVec)) {
     if (auto *IE = dyn_cast<InsertElementInst>(I)) {
       // Extracting the inserted element?


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D55182.176918.patch
Type: text/x-patch
Size: 4811 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20181206/9050b5e2/attachment.bin>