[PATCH] D133649: [SLP] Add Preheader to CSE blocks after hoisting CSE-able instrs.

[Florian Hahn via Phabricator via llvm-commits]

fhahn created this revision.
fhahn added reviewers: RKSimon, spatel, ABataev.
Herald added subscribers: vporpo, hiraditya.
Herald added a project: All.
fhahn requested review of this revision.
Herald added a subscriber: pcwang-thead.
Herald added a project: LLVM.

Adding the pre-header to CSEBlocks ensures instructions are CSE'd even
after hoisting.

This was original discovered by @atrick a while ago.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D133649

Files:
  llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
  llvm/test/Transforms/SLPVectorizer/X86/cse.ll


Index: llvm/test/Transforms/SLPVectorizer/X86/cse.ll
===================================================================
--- llvm/test/Transforms/SLPVectorizer/X86/cse.ll
+++ llvm/test/Transforms/SLPVectorizer/X86/cse.ll
@@ -353,19 +353,17 @@
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <2 x i32> poison, i32 [[A:%.*]], i32 0
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> [[TMP0]], i32 [[A]], i32 1
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x i32> poison, i32 [[A]], i32 0
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x i32> [[TMP2]], i32 [[A]], i32 1
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
-; CHECK-NEXT:    [[TMP4:%.*]] = or <2 x i32> <i32 22, i32 22>, [[TMP1]]
+; CHECK-NEXT:    [[TMP2:%.*]] = or <2 x i32> <i32 22, i32 22>, [[TMP1]]
 ; CHECK-NEXT:    [[GEP_0:%.*]] = getelementptr inbounds i32, i32* [[DST:%.*]], i64 0
-; CHECK-NEXT:    [[TMP5:%.*]] = or <2 x i32> [[TMP4]], <i32 3, i32 3>
-; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32* [[GEP_0]] to <2 x i32>*
-; CHECK-NEXT:    store <2 x i32> [[TMP5]], <2 x i32>* [[TMP6]], align 4
-; CHECK-NEXT:    [[TMP7:%.*]] = or <2 x i32> [[TMP3]], <i32 3, i32 3>
+; CHECK-NEXT:    [[TMP3:%.*]] = or <2 x i32> [[TMP2]], <i32 3, i32 3>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i32* [[GEP_0]] to <2 x i32>*
+; CHECK-NEXT:    store <2 x i32> [[TMP3]], <2 x i32>* [[TMP4]], align 4
+; CHECK-NEXT:    [[TMP5:%.*]] = or <2 x i32> [[TMP1]], <i32 3, i32 3>
 ; CHECK-NEXT:    [[GEP_2:%.*]] = getelementptr inbounds i32, i32* [[DST]], i64 10
-; CHECK-NEXT:    [[TMP8:%.*]] = bitcast i32* [[GEP_2]] to <2 x i32>*
-; CHECK-NEXT:    store <2 x i32> [[TMP7]], <2 x i32>* [[TMP8]], align 4
+; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32* [[GEP_2]] to <2 x i32>*
+; CHECK-NEXT:    store <2 x i32> [[TMP5]], <2 x i32>* [[TMP6]], align 4
 ; CHECK-NEXT:    br i1 [[C:%.*]], label [[LOOP]], label [[EXIT:%.*]]
 ; CHECK:       exit:
 ; CHECK-NEXT:    ret void
Index: llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
===================================================================
--- llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -8962,6 +8962,7 @@
 
     // We can hoist this instruction. Move it to the pre-header.
     I->moveBefore(PreHeader->getTerminator());
+    CSEBlocks.insert(PreHeader);
   }
 
   // Make a list of all reachable blocks in our CSE queue.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D133649.459308.patch
Type: text/x-patch
Size: 2420 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20220910/a56d4018/attachment.bin>