[llvm] e1d2251 - [SLP]Fix minbitwidth analysis for gather nodes with icmp users.

[Alexey Bataev via llvm-commits]

Author: Alexey Bataev
Date: 2024-08-26T11:40:44-07:00
New Revision: e1d22512906e69846c8f6a2d29b30832b7c12b46

URL: https://github.com/llvm/llvm-project/commit/e1d22512906e69846c8f6a2d29b30832b7c12b46
DIFF: https://github.com/llvm/llvm-project/commit/e1d22512906e69846c8f6a2d29b30832b7c12b46.diff

LOG: [SLP]Fix minbitwidth analysis for gather nodes with icmp users.

If the node is not in MinBWs container and the user node is icmp node,
the compiler should not check the type size of the user instruction, it
is always 1 and is not good for actual bitwidth analysis.

Fixes https://github.com/llvm/llvm-project/issues/105988

Added: 
    

Modified: 
    llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
    llvm/test/Transforms/SLPVectorizer/X86/gather-with-cmp-user.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index def73e8d0c0db7..ed47ed661ab946 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -15975,6 +15975,10 @@ void BoUpSLP::computeMinimumValueSizes() {
                     auto It = MinBWs.find(TE);
                     if (It != MinBWs.end() && It->second.first > UserTESz)
                       return true;
+                    // The size of icmp is always 1 and should not be
+                    // considered.
+                    if (TE->getOpcode() == Instruction::ICmp)
+                      return true;
                     return DL->getTypeSizeInBits(U->getType()) > UserTESz;
                   }));
         })) {

diff  --git a/llvm/test/Transforms/SLPVectorizer/X86/gather-with-cmp-user.ll b/llvm/test/Transforms/SLPVectorizer/X86/gather-with-cmp-user.ll
index 10ca5a2700ebee..7e75970de34929 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/gather-with-cmp-user.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/gather-with-cmp-user.ll
@@ -7,15 +7,13 @@ define i1 @test(i32 %g, i16 %d) {
 ; CHECK-NEXT:  [[ENTRY:.*:]]
 ; CHECK-NEXT:    [[TMP0:%.*]] = and i16 [[D]], 1
 ; CHECK-NEXT:    [[XOR_I_I:%.*]] = xor i32 [[G]], 1
-; CHECK-NEXT:    [[TMP1:%.*]] = trunc i32 [[G]] to i8
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x i8> poison, i8 [[TMP1]], i32 0
-; CHECK-NEXT:    [[TMP3:%.*]] = trunc i32 [[XOR_I_I]] to i8
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x i8> [[TMP2]], i8 [[TMP3]], i32 1
+; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> poison, i32 [[G]], i32 0
+; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[XOR_I_I]], i32 1
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc <2 x i32> [[TMP9]] to <2 x i8>
 ; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <2 x i8> [[TMP4]], <2 x i8> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
 ; CHECK-NEXT:    [[TMP6:%.*]] = add <4 x i8> [[TMP5]], <i8 -9, i8 -9, i8 -1, i8 -1>
 ; CHECK-NEXT:    [[TMP7:%.*]] = icmp sgt <4 x i8> [[TMP6]], <i8 -3, i8 -3, i8 -3, i8 -3>
 ; CHECK-NEXT:    [[TMP8:%.*]] = zext <4 x i1> [[TMP7]] to <4 x i8>
-; CHECK-NEXT:    [[TMP9:%.*]] = sext <2 x i8> [[TMP4]] to <2 x i32>
 ; CHECK-NEXT:    [[TMP10:%.*]] = shufflevector <2 x i32> [[TMP9]], <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
 ; CHECK-NEXT:    [[TMP11:%.*]] = zext <4 x i8> [[TMP8]] to <4 x i32>
 ; CHECK-NEXT:    [[TMP12:%.*]] = icmp sgt <4 x i32> [[TMP10]], [[TMP11]]