[InstCombine] Remove the canonicalization of `trunc` to `i1` #84628

YanWQ-monad · 2024-03-09T13:39:25Z

Remove the canonicalization of trunc to i1 according to the suggestion of #83829 (comment)

llvm-project/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp

Lines 737 to 745 in a84e66a

    
           if (DestTy->isIntegerTy()) { 
        
             // Canonicalize trunc x to i1 -> icmp ne (and x, 1), 0 (scalar only). 
        
             // TODO: We canonicalize to more instructions here because we are probably 
        
             // lacking equivalent analysis for trunc relative to icmp. There may also 
        
             // be codegen concerns. If those trunc limitations were removed, we could 
        
             // remove this transform. 
        
             Value *And = Builder.CreateAnd(Src, ConstantInt::get(SrcTy, 1)); 
        
             return new ICmpInst(ICmpInst::ICMP_NE, And, Zero); 
        
           }

Alive2: https://alive2.llvm.org/ce/z/cacYVA

github-actions · 2024-03-09T13:39:44Z

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be
notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write
permissions for the repository. In which case you can instead tag reviewers by
name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review
by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate
is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

llvmbot · 2024-03-09T13:40:15Z

@llvm/pr-subscribers-llvm-transforms

Author: Monad (YanWQ-monad)

Changes

Remove the canonicalization of trunc to i1 according to the suggestion of #83829 (comment)

llvm-project/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp

Lines 737 to 745 in a84e66a

    
           if (DestTy->isIntegerTy()) { 
        
             // Canonicalize trunc x to i1 -> icmp ne (and x, 1), 0 (scalar only). 
        
             // TODO: We canonicalize to more instructions here because we are probably 
        
             // lacking equivalent analysis for trunc relative to icmp. There may also 
        
             // be codegen concerns. If those trunc limitations were removed, we could 
        
             // remove this transform. 
        
             Value *And = Builder.CreateAnd(Src, ConstantInt::get(SrcTy, 1)); 
        
             return new ICmpInst(ICmpInst::ICMP_NE, And, Zero); 
        
           }

Patch is 66.44 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/84628.diff

16 Files Affected:

(modified) llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp (-9)
(modified) llvm/test/Transforms/InstCombine/X86/x86-avx512-inseltpoison.ll (+56-84)
(modified) llvm/test/Transforms/InstCombine/X86/x86-avx512.ll (+56-84)
(modified) llvm/test/Transforms/InstCombine/apint-shl-trunc.ll (+4-6)
(modified) llvm/test/Transforms/InstCombine/cast.ll (+2-4)
(modified) llvm/test/Transforms/InstCombine/catchswitch-phi.ll (+9-10)
(modified) llvm/test/Transforms/InstCombine/icmp-mul-and.ll (+2-2)
(modified) llvm/test/Transforms/InstCombine/icmp-mul-zext.ll (+3-3)
(modified) llvm/test/Transforms/InstCombine/mul-masked-bits.ll (+4-6)
(modified) llvm/test/Transforms/InstCombine/mul.ll (+2-3)
(modified) llvm/test/Transforms/InstCombine/phi.ll (+4-5)
(modified) llvm/test/Transforms/InstCombine/ptr-int-cast.ll (+1-2)
(modified) llvm/test/Transforms/InstCombine/reduction-add-sext-zext-i1.ll (+1-2)
(modified) llvm/test/Transforms/InstCombine/reduction-xor-sext-zext-i1.ll (+7-11)
(modified) llvm/test/Transforms/InstCombine/select-ctlz-to-cttz.ll (+12-6)
(modified) llvm/test/Transforms/PhaseOrdering/X86/merge-functions.ll (+1-2)

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
index 45afa6363ae01f..2f34d825ceed58 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
@@ -734,15 +734,6 @@ Instruction *InstCombinerImpl::visitTrunc(TruncInst &Trunc) {
 
   if (DestWidth == 1) {
     Value *Zero = Constant::getNullValue(SrcTy);
-    if (DestTy->isIntegerTy()) {
-      // Canonicalize trunc x to i1 -> icmp ne (and x, 1), 0 (scalar only).
-      // TODO: We canonicalize to more instructions here because we are probably
-      // lacking equivalent analysis for trunc relative to icmp. There may also
-      // be codegen concerns. If those trunc limitations were removed, we could
-      // remove this transform.
-      Value *And = Builder.CreateAnd(Src, ConstantInt::get(SrcTy, 1));
-      return new ICmpInst(ICmpInst::ICMP_NE, And, Zero);
-    }
 
     // For vectors, we do not canonicalize all truncs to icmp, so optimize
     // patterns that would be covered within visitICmpInst.
diff --git a/llvm/test/Transforms/InstCombine/X86/x86-avx512-inseltpoison.ll b/llvm/test/Transforms/InstCombine/X86/x86-avx512-inseltpoison.ll
index 9b990480709c97..80d8e1b16ed28b 100644
--- a/llvm/test/Transforms/InstCombine/X86/x86-avx512-inseltpoison.ll
+++ b/llvm/test/Transforms/InstCombine/X86/x86-avx512-inseltpoison.ll
@@ -39,10 +39,9 @@ define <4 x float> @test_add_ss_mask(<4 x float> %a, <4 x float> %b, <4 x float>
 ; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> [[A:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = fadd float [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[TMP4:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i8 [[MASK:%.*]] to i1
 ; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[C:%.*]], i64 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], float [[TMP5]], float [[TMP3]]
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP4]], float [[TMP3]], float [[TMP5]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <4 x float> [[A]], float [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <4 x float> [[TMP7]]
 ;
@@ -117,10 +116,9 @@ define <2 x double> @test_add_sd_mask(<2 x double> %a, <2 x double> %b, <2 x dou
 ; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x double> [[A:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = fadd double [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[TMP4:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i8 [[MASK:%.*]] to i1
 ; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[C:%.*]], i64 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], double [[TMP5]], double [[TMP3]]
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP4]], double [[TMP3]], double [[TMP5]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x double> [[A]], double [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <2 x double> [[TMP7]]
 ;
@@ -191,10 +189,9 @@ define <4 x float> @test_sub_ss_mask(<4 x float> %a, <4 x float> %b, <4 x float>
 ; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> [[A:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = fsub float [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[TMP4:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i8 [[MASK:%.*]] to i1
 ; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[C:%.*]], i64 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], float [[TMP5]], float [[TMP3]]
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP4]], float [[TMP3]], float [[TMP5]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <4 x float> [[A]], float [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <4 x float> [[TMP7]]
 ;
@@ -269,10 +266,9 @@ define <2 x double> @test_sub_sd_mask(<2 x double> %a, <2 x double> %b, <2 x dou
 ; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x double> [[A:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = fsub double [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[TMP4:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i8 [[MASK:%.*]] to i1
 ; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[C:%.*]], i64 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], double [[TMP5]], double [[TMP3]]
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP4]], double [[TMP3]], double [[TMP5]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x double> [[A]], double [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <2 x double> [[TMP7]]
 ;
@@ -343,10 +339,9 @@ define <4 x float> @test_mul_ss_mask(<4 x float> %a, <4 x float> %b, <4 x float>
 ; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> [[A:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = fmul float [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[TMP4:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i8 [[MASK:%.*]] to i1
 ; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[C:%.*]], i64 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], float [[TMP5]], float [[TMP3]]
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP4]], float [[TMP3]], float [[TMP5]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <4 x float> [[A]], float [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <4 x float> [[TMP7]]
 ;
@@ -421,10 +416,9 @@ define <2 x double> @test_mul_sd_mask(<2 x double> %a, <2 x double> %b, <2 x dou
 ; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x double> [[A:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = fmul double [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[TMP4:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i8 [[MASK:%.*]] to i1
 ; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[C:%.*]], i64 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], double [[TMP5]], double [[TMP3]]
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP4]], double [[TMP3]], double [[TMP5]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x double> [[A]], double [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <2 x double> [[TMP7]]
 ;
@@ -495,10 +489,9 @@ define <4 x float> @test_div_ss_mask(<4 x float> %a, <4 x float> %b, <4 x float>
 ; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> [[A:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = fdiv float [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[TMP4:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i8 [[MASK:%.*]] to i1
 ; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[C:%.*]], i64 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], float [[TMP5]], float [[TMP3]]
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP4]], float [[TMP3]], float [[TMP5]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <4 x float> [[A]], float [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <4 x float> [[TMP7]]
 ;
@@ -573,10 +566,9 @@ define <2 x double> @test_div_sd_mask(<2 x double> %a, <2 x double> %b, <2 x dou
 ; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x double> [[A:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = fdiv double [[TMP1]], [[TMP2]]
-; CHECK-NEXT:    [[TMP4:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP4]], 0
+; CHECK-NEXT:    [[TMP4:%.*]] = trunc i8 [[MASK:%.*]] to i1
 ; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[C:%.*]], i64 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], double [[TMP5]], double [[TMP3]]
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP4]], double [[TMP3]], double [[TMP5]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x double> [[A]], double [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <2 x double> [[TMP7]]
 ;
@@ -981,9 +973,8 @@ define <4 x float> @test_mask_vfmadd_ss(<4 x float> %a, <4 x float> %b, <4 x flo
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[C:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP4:%.*]] = call float @llvm.fma.f32(float [[TMP1]], float [[TMP2]], float [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP5]], 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], float [[TMP1]], float [[TMP4]]
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float [[TMP1]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <4 x float> [[A]], float [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <4 x float> [[TMP7]]
 ;
@@ -1011,9 +1002,8 @@ define float @test_mask_vfmadd_ss_0(<4 x float> %a, <4 x float> %b, <4 x float>
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[C:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP4:%.*]] = call float @llvm.fma.f32(float [[TMP1]], float [[TMP2]], float [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP5]], 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], float [[TMP1]], float [[TMP4]]
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float [[TMP1]]
 ; CHECK-NEXT:    ret float [[TMP6]]
 ;
   %1 = insertelement <4 x float> %a, float 1.000000e+00, i32 1
@@ -1060,9 +1050,8 @@ define <2 x double> @test_mask_vfmadd_sd(<2 x double> %a, <2 x double> %b, <2 x
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x double> [[C:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP4:%.*]] = call double @llvm.fma.f64(double [[TMP1]], double [[TMP2]], double [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP5]], 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], double [[TMP1]], double [[TMP4]]
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], double [[TMP4]], double [[TMP1]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x double> [[A]], double [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <2 x double> [[TMP7]]
 ;
@@ -1086,9 +1075,8 @@ define double @test_mask_vfmadd_sd_0(<2 x double> %a, <2 x double> %b, <2 x doub
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x double> [[C:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP4:%.*]] = call double @llvm.fma.f64(double [[TMP1]], double [[TMP2]], double [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP5]], 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], double [[TMP1]], double [[TMP4]]
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], double [[TMP4]], double [[TMP1]]
 ; CHECK-NEXT:    ret double [[TMP6]]
 ;
   %1 = insertelement <2 x double> %a, double 1.000000e+00, i32 1
@@ -1129,9 +1117,8 @@ define <4 x float> @test_maskz_vfmadd_ss(<4 x float> %a, <4 x float> %b, <4 x fl
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[C:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP4:%.*]] = call float @llvm.fma.f32(float [[TMP1]], float [[TMP2]], float [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP5]], 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], float 0.000000e+00, float [[TMP4]]
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float 0.000000e+00
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <4 x float> [[A]], float [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <4 x float> [[TMP7]]
 ;
@@ -1159,9 +1146,8 @@ define float @test_maskz_vfmadd_ss_0(<4 x float> %a, <4 x float> %b, <4 x float>
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[C:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP4:%.*]] = call float @llvm.fma.f32(float [[TMP1]], float [[TMP2]], float [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP5]], 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], float 0.000000e+00, float [[TMP4]]
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float 0.000000e+00
 ; CHECK-NEXT:    ret float [[TMP6]]
 ;
   %1 = insertelement <4 x float> %a, float 1.000000e+00, i32 1
@@ -1206,9 +1192,8 @@ define <2 x double> @test_maskz_vfmadd_sd(<2 x double> %a, <2 x double> %b, <2 x
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x double> [[C:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP4:%.*]] = call double @llvm.fma.f64(double [[TMP1]], double [[TMP2]], double [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP5]], 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], double 0.000000e+00, double [[TMP4]]
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], double [[TMP4]], double 0.000000e+00
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x double> [[A]], double [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <2 x double> [[TMP7]]
 ;
@@ -1232,9 +1217,8 @@ define double @test_maskz_vfmadd_sd_0(<2 x double> %a, <2 x double> %b, <2 x dou
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x double> [[C:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP4:%.*]] = call double @llvm.fma.f64(double [[TMP1]], double [[TMP2]], double [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP5]], 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], double 0.000000e+00, double [[TMP4]]
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], double [[TMP4]], double 0.000000e+00
 ; CHECK-NEXT:    ret double [[TMP6]]
 ;
   %1 = insertelement <2 x double> %a, double 1.000000e+00, i32 1
@@ -1275,9 +1259,8 @@ define <4 x float> @test_mask3_vfmadd_ss(<4 x float> %a, <4 x float> %b, <4 x fl
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[C:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP4:%.*]] = call float @llvm.fma.f32(float [[TMP1]], float [[TMP2]], float [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP5]], 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], float [[TMP3]], float [[TMP4]]
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float [[TMP3]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <4 x float> [[C]], float [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <4 x float> [[TMP7]]
 ;
@@ -1305,9 +1288,8 @@ define float @test_mask3_vfmadd_ss_0(<4 x float> %a, <4 x float> %b, <4 x float>
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[C:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP4:%.*]] = call float @llvm.fma.f32(float [[TMP1]], float [[TMP2]], float [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP5]], 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], float [[TMP3]], float [[TMP4]]
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], float [[TMP4]], float [[TMP3]]
 ; CHECK-NEXT:    ret float [[TMP6]]
 ;
   %1 = insertelement <4 x float> %c, float 1.000000e+00, i32 1
@@ -1352,9 +1334,8 @@ define <2 x double> @test_mask3_vfmadd_sd(<2 x double> %a, <2 x double> %b, <2 x
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x double> [[C:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP4:%.*]] = call double @llvm.fma.f64(double [[TMP1]], double [[TMP2]], double [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP5]], 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], double [[TMP3]], double [[TMP4]]
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], double [[TMP4]], double [[TMP3]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x double> [[C]], double [[TMP6]], i64 0
 ; CHECK-NEXT:    ret <2 x double> [[TMP7]]
 ;
@@ -1378,9 +1359,8 @@ define double @test_mask3_vfmadd_sd_0(<2 x double> %a, <2 x double> %b, <2 x dou
 ; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[B:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x double> [[C:%.*]], i64 0
 ; CHECK-NEXT:    [[TMP4:%.*]] = call double @llvm.fma.f64(double [[TMP1]], double [[TMP2]], double [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP5]], 0
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[DOTNOT]], double [[TMP3]], double [[TMP4]]
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP5]], double [[TMP4]], double [[TMP3]]
 ; CHECK-NEXT:    ret double [[TMP6]]
 ;
   %1 = insertelement <2 x double> %c, double 1.000000e+00, i32 1
@@ -1423,9 +1403,8 @@ define <4 x float> @test_mask3_vfmsub_ss(<4 x float> %a, <4 x float> %b, <4 x fl
 ; CHECK-NEXT:    [[TMP4:%.*]] = fneg float [[TMP3]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = call float @llvm.fma.f32(float [[TMP1]], float [[TMP2]], float [[TMP4]])
 ; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[C]], i64 0
-; CHECK-NEXT:    [[TMP7:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP7]], 0
-; CHECK-NEXT:    [[TMP8:%.*]] = select i1 [[DOTNOT]], float [[TMP6]], float [[TMP5]]
+; CHECK-NEXT:    [[TMP7:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], float [[TMP5]], float [[TMP6]]
 ; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <4 x float> [[C]], float [[TMP8]], i64 0
 ; CHECK-NEXT:    ret <4 x float> [[TMP9]]
 ;
@@ -1457,9 +1436,8 @@ define float @test_mask3_vfmsub_ss_0(<4 x float> %a, <4 x float> %b, <4 x float>
 ; CHECK-NEXT:    [[TMP4:%.*]] = fneg float [[TMP3]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = call float @llvm.fma.f32(float [[TMP1]], float [[TMP2]], float [[TMP4]])
 ; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[C]], i64 0
-; CHECK-NEXT:    [[TMP7:%.*]] = and i8 [[MASK:%.*]], 1
-; CHECK-NEXT:    [[DOTNOT:%.*]] = icmp eq i8 [[TMP7]], 0
-; CHECK-NEXT:    [[TMP8:%.*]] = select i1 [[DOTNOT]], float [[TMP6]], float [[TMP5]]
+; CHECK-NEXT:    [[TMP7:%.*]] = trunc i8 [[MASK:%.*]] to i1
+; CHECK-NEXT:    [[TMP8:%.*]] = select i1 [[TMP7]], float [[TMP5]], float [[TMP6]]
 ; CHECK-NEXT:    ret float [[TMP8]]
 ;
   %1 = insertelement <4 x float> %c, float 1.000000e+00, i32 1
@@ -1532,9 +1510,8 @@ define <2 x double> @test_mask3_vfmsub_sd(<2 x double> %a, <2 x double> %b, <2 x
 ; CHECK-NEXT:    [[TMP4:%.*]] = fneg double [[TMP3]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = call double @llvm.fma.f64(double [[TMP1]], double [[TMP2]], double [[TMP4]])
 ; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x double> [[C]], i64 0
-; CHECK-NEXT:    [[TMP7:%.*]] = an...
[truncated]

dtcxzyw · 2024-03-09T14:07:09Z

Failed Tests (1):
LLVM :: Transforms/InstCombine/phi.ll

Please update the test.

goldsteinn · 2024-03-09T15:59:48Z

This isn't a bad thought, but think the comment still applies, looking through the test changes there are some regressions that indicate we handle trunc i1 in all the cases we should.

I am supportive of this effort, but can you please also make patches to cleanup the regressions?

YanWQ-monad · 2024-03-09T17:40:58Z

May I ask a question about fixing this regression?

https://github.com/llvm/llvm-project/pull/84628/files#diff-8029a15d86a03b733e3d58345f03e5b45ec2c335e399cf6be07ac4247a67021cL249-R254

With the original canonicalization, trunc X to i1 would be canonicalized to Y = and X, 1 and icmp ne Y, 0, which enables some optimization like the folding of (icmp Pred1 V1, C1) | (icmp Pred2 V2, C2). But without the canonicalization, it failed to match trunc as icmp, which results in regression.

llvm-project/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp

Lines 3678 to 3682 in b4001e3

    
           ICmpInst *LHS = dyn_cast<ICmpInst>(Op0); 
        
           ICmpInst *RHS = dyn_cast<ICmpInst>(Op1); 
        
           if (LHS && RHS) 
        
             if (Value *Res = foldAndOrOfICmps(LHS, RHS, I, /* IsAnd */ false)) 
        
               return replaceInstUsesWith(I, Res);

The problem is that, the parameters for that series of functions are passed using ICmpInst. In order to reuse the code, it might be necessary to make changes to the signature and the body of the functions, to make them accept TruncInst.
I don't know whether it's elegant or acceptable, or there is a better way to handle it. Thanks.

goldsteinn · 2024-03-09T17:49:11Z

May I ask a question about fixing this regression?

https://github.com/llvm/llvm-project/pull/84628/files#diff-8029a15d86a03b733e3d58345f03e5b45ec2c335e399cf6be07ac4247a67021cL249-R254

With the original canonicalization, trunc X to i1 would be canonicalized to Y = and X, 1 and icmp ne Y, 0, which enables some optimization like the folding of (icmp Pred1 V1, C1) | (icmp Pred2 V2, C2). But without the canonicalization, it failed to match trunc as icmp, which results in regression.

llvm-project/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp

Lines 3678 to 3682 in b4001e3

ICmpInst *LHS = dyn_cast<ICmpInst>(Op0);

ICmpInst *RHS = dyn_cast<ICmpInst>(Op1);

if (LHS && RHS)

if (Value *Res = foldAndOrOfICmps(LHS, RHS, I, /* IsAnd */ false))

return replaceInstUsesWith(I, Res);

The problem is that, the parameters for that series of functions are passed using ICmpInst. In order to reuse the code, it might be necessary to make changes to the signature and the body of the functions, to make them accept TruncInst. I don't know whether it's elegant or acceptable, or there is a better way to handle it. Thanks.

It is a limitation that these folds take completed ICmpInst instead of decomposed components. And I would fully support an NFC patch to change that.

Another approach might be to create a pattern match m_TruncLike and match the (icmp ne (and X, 1), 0) in it. Then modify the places you see use us matching m_Trunc to to use m_TruncLike.

YanWQ-monad · 2024-03-10T17:25:45Z

After examining the code, I found that those folds are not the key problem.
In fact, the removal of canonicalization affects two folds in that regression:

((C1 << X) & C2) != 0   ->  X < (Log2(C2+C1) - Log2(C1))           (foldICmpUsingKnownBits)
((C1 << X) >> C2) & C3  ->  X == (cttz(C3)+C2-cttz(C1)) ? C3 : 0   (visitAnd)

Making those folds inside foldAndOrOfICmps accept trunc seems provide no help to these two folds, unless a more thorough decomposition is performed. So I just copied two folds to visitTrunc.

llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp

goldsteinn · 2024-03-10T19:44:10Z

After examining the code, I found that those folds are not the key problem. In fact, the removal of canonicalization affects two folds in that regression:
((C1 << X) & C2) != 0   ->  X < (Log2(C2+C1) - Log2(C1))           (foldICmpUsingKnownBits)
((C1 << X) >> C2) & C3  ->  X == (cttz(C3)+C2-cttz(C1)) ? C3 : 0   (visitAnd)
Making those folds inside foldAndOrOfICmps accept trunc seems provide no help to these two folds, unless a more thorough decomposition is performed. So I just copied two folds to visitTrunc.

Seems reasonable, not sure if this canonicalization is relied on elsewhere thought.

YanWQ-monad · 2024-03-21T13:56:39Z

ping? Please let me know if there are further changes needed, thanks.

llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp

goldsteinn · 2024-03-21T16:17:26Z

Im generally fine with this change.
@dtcxzyw can you test this?

dtcxzyw

LGTM.

llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp

dtcxzyw · 2024-03-21T19:53:56Z

llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp

    Value *X;
+    const APInt *C1;
+    Constant *C2;
+    if (match(Src, m_OneUse(m_LShr(m_Shl(m_Power2(C1), m_Value(X)),


Please provide the alive2 proof.

Sure, it's https://alive2.llvm.org/ce/z/6Frpdc.

Could actually be any right shift: https://alive2.llvm.org/ce/z/vmPhjM

(see m_Shr)

Alive2: https://alive2.llvm.org/ce/z/6Frpdc

YanWQ-monad · 2024-03-29T11:46:56Z

ping?

github-actions · 2024-03-29T13:47:56Z

@YanWQ-monad Congratulations on having your first Pull Request (PR) merged into the LLVM Project!

Your changes will be combined with recent changes from other authors, then tested
by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR.

Please check whether problems have been caused by your change specifically, as
the builds can include changes from many authors. It is not uncommon for your
change to be included in a build that fails due to someone else's changes, or
infrastructure issues.

How to do this, and the rest of the post-merge process, is covered in detail here.

If your change does cause a problem, it may be reverted, or you can revert it yourself.
This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again.

If you don't get any reports, no action is required from you. Your changes are working as expected, well done!

With the addition of llvm#84628, truncs to i1 are being emitted as conditions to branch instructions. This caused significant regressions in cases which were previously improved by loop unswitch. Adding truncs to i1 restore the previous performance seen.

With the addition of #84628, truncs to i1 are being emitted as conditions to branch instructions. This caused significant regressions in cases which were previously improved by loop unswitch. Adding truncs to i1 restore the previous performance seen.

This patch attempts to help the SelectOpt pass detect select groups made up of conditions and not(conditions). Usually these are canonicalized in instcombine to remove the not and invert the true/false values, but this will not happen for Loginal operations, which can be beneficial to convert if they are part of a larger select group. The handling for not's are mostly handled in the SelectLike, which can be marked as Inverted in order to reverse the TrueValue and FalseValue. This helps fix a regression in fortran minloc constructs, after llvm#84628 helped simplify a loop with branches into a loop with selects.

This patch attempts to help the SelectOpt pass detect select groups made up of conditions and not(conditions). Usually these are canonicalized in instcombine to remove the not and invert the true/false values, but this will not happen for Loginal operations, which can be beneficial to convert if they are part of a larger select group. The handling for not's are mostly handled in the SelectLike, which can be marked as Inverted in order to reverse the TrueValue and FalseValue. This helps fix a regression in fortran minloc constructs, after #84628 helped simplify a loop with branches into a loop with selects.

…lvm#84628)" This reverts commit 56b3222.

Remove the canonicalization of trunc to i1

59e41d9

YanWQ-monad requested a review from nikic as a code owner March 9, 2024 13:39

llvmbot added the llvm:transforms label Mar 9, 2024

dtcxzyw requested a review from goldsteinn March 9, 2024 13:52

YanWQ-monad added 2 commits March 10, 2024 02:05

Rename the variables to avoid conflict

55ace09

Fix regressions

64fc00c

goldsteinn requested a review from dtcxzyw March 10, 2024 19:37

goldsteinn reviewed Mar 10, 2024

View reviewed changes

llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp Outdated Show resolved Hide resolved

goldsteinn reviewed Mar 10, 2024

View reviewed changes

llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp Outdated Show resolved Hide resolved

YanWQ-monad added 2 commits March 11, 2024 04:14

Revise the code

6b3ae69

Change variable name

fcaafc3

goldsteinn reviewed Mar 21, 2024

View reviewed changes

llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp Outdated Show resolved Hide resolved

Use m_Power2

ad99d1f

dtcxzyw mentioned this pull request Mar 21, 2024

Test PR84628 plctlab/llvm-ci#1117

Closed

dtcxzyw approved these changes Mar 21, 2024

View reviewed changes

YanWQ-monad added 3 commits March 22, 2024 04:08

Remove old comment

4015d83

Remove unnecessary C2 < BitWidth

6594005

Alive2: https://alive2.llvm.org/ce/z/6Frpdc

Match any right shift

f905604

dtcxzyw approved these changes Mar 22, 2024

View reviewed changes

dtcxzyw merged commit 56b3222 into llvm:main Mar 29, 2024
4 checks passed

MDevereau mentioned this pull request Apr 23, 2024

[LoopUnswitch] Allow i1 truncs in loop unswitch #89738

Merged

davemgreen mentioned this pull request May 17, 2024

[SelectOpt] Add handling for not conditions. #92517

Merged

cyyself mentioned this pull request Jul 24, 2024

[InstCombine] Missed optimization after #84628 #100293

Open

cyyself added a commit to cyyself/llvm-project that referenced this pull request Jul 24, 2024

Revert "[InstCombine] Remove the canonicalization of trunc to i1 (l…

e2de1ad

…lvm#84628)" This reverts commit 56b3222.

cyyself added a commit to cyyself/llvm-project that referenced this pull request Aug 14, 2024

Revert "[InstCombine] Remove the canonicalization of trunc to i1 (l…

66e03ed

…lvm#84628)" This reverts commit 56b3222.

nikic mentioned this pull request Oct 2, 2024

[InstCombine] Regression in and of masked comparison fold #110919

Closed

This was referenced Dec 2, 2024

[ValueTracking] Handle assume(trunc x to i1) in ComputeKnownBits #118406

Merged

[InstCombine] Fold (trunc X to i1) & !iszero(X & Pow2)) -> (X & (1 | Pow2)) == (1 | Pow2) #119196

Closed

dianqk mentioned this pull request Feb 25, 2025

Add Three Codegen Tests rust-lang/rust#134626

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[InstCombine] Remove the canonicalization of `trunc` to `i1` #84628

[InstCombine] Remove the canonicalization of `trunc` to `i1` #84628

YanWQ-monad commented Mar 9, 2024 •

edited by dtcxzyw

Loading

github-actions bot commented Mar 9, 2024

llvmbot commented Mar 9, 2024

dtcxzyw commented Mar 9, 2024

goldsteinn commented Mar 9, 2024

YanWQ-monad commented Mar 9, 2024

goldsteinn commented Mar 9, 2024

YanWQ-monad commented Mar 10, 2024

goldsteinn commented Mar 10, 2024

YanWQ-monad commented Mar 21, 2024 •

edited

Loading

goldsteinn commented Mar 21, 2024

dtcxzyw left a comment

dtcxzyw Mar 21, 2024

YanWQ-monad Mar 21, 2024

goldsteinn Mar 21, 2024

goldsteinn Mar 21, 2024

YanWQ-monad commented Mar 29, 2024

github-actions bot commented Mar 29, 2024

	if (DestTy->isIntegerTy()) {
	// Canonicalize trunc x to i1 -> icmp ne (and x, 1), 0 (scalar only).
	// TODO: We canonicalize to more instructions here because we are probably
	// lacking equivalent analysis for trunc relative to icmp. There may also
	// be codegen concerns. If those trunc limitations were removed, we could
	// remove this transform.
	Value *And = Builder.CreateAnd(Src, ConstantInt::get(SrcTy, 1));
	return new ICmpInst(ICmpInst::ICMP_NE, And, Zero);
	}

[InstCombine] Remove the canonicalization of trunc to i1 #84628

[InstCombine] Remove the canonicalization of trunc to i1 #84628

Conversation

YanWQ-monad commented Mar 9, 2024 • edited by dtcxzyw Loading

github-actions bot commented Mar 9, 2024

llvmbot commented Mar 9, 2024

dtcxzyw commented Mar 9, 2024

goldsteinn commented Mar 9, 2024

YanWQ-monad commented Mar 9, 2024

goldsteinn commented Mar 9, 2024

YanWQ-monad commented Mar 10, 2024

goldsteinn commented Mar 10, 2024

YanWQ-monad commented Mar 21, 2024 • edited Loading

goldsteinn commented Mar 21, 2024

dtcxzyw left a comment

Choose a reason for hiding this comment

dtcxzyw Mar 21, 2024

Choose a reason for hiding this comment

YanWQ-monad Mar 21, 2024

Choose a reason for hiding this comment

goldsteinn Mar 21, 2024

Choose a reason for hiding this comment

goldsteinn Mar 21, 2024

Choose a reason for hiding this comment

YanWQ-monad commented Mar 29, 2024

github-actions bot commented Mar 29, 2024

[InstCombine] Remove the canonicalization of `trunc` to `i1` #84628

[InstCombine] Remove the canonicalization of `trunc` to `i1` #84628

YanWQ-monad commented Mar 9, 2024 •

edited by dtcxzyw

Loading

YanWQ-monad commented Mar 21, 2024 •

edited

Loading