[llvm] [InstCombine] Fold `(trunc nuw A to i1) == (trunc nuw B to i1)` to `A == B` (PR #133368)

[via llvm-commits]

https://github.com/veera-sivarajan created https://github.com/llvm/llvm-project/pull/133368

Fixes #133344

Proof: https://alive2.llvm.org/ce/z/X3Uh23 

InstCombine couldn't optimize `i1`  because `canonicalizeICmpBool()` was transforming the comparison into bitwise operations before `foldICmpTruncWithTruncOrExt()` was called.

This PR solves the ordering issue by placing `foldICmpTruncWithTruncOrExt()` before `canonicalizeICmpBool()`.

I believe this will not cause any regressions since all tests are passing.


>From 87fb724da0e6a24713687abf9e87c512a10d3b14 Mon Sep 17 00:00:00 2001
From: Veera <sveera.2001 at gmail.com>
Date: Fri, 28 Mar 2025 04:28:24 +0000
Subject: [PATCH 1/2] Add Test

---
 .../Transforms/InstCombine/icmp-of-trunc-ext.ll     | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll b/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll
index 2c2de5dbf09f6..4ff9f8339be1a 100644
--- a/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll
@@ -409,6 +409,19 @@ define i1 @trunc_equality_either(i16 %x, i16 %y) {
   ret i1 %c
 }
 
+define i1 @trunc_equality_bool(i8 %a, i8 %b) {
+; CHECK-LABEL: @trunc_equality_bool(
+; CHECK-NEXT:    [[TMP1:%.*]] = xor i8 [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT:    [[TMP2:%.*]] = trunc i8 [[TMP1]] to i1
+; CHECK-NEXT:    [[EQ:%.*]] = xor i1 [[TMP2]], true
+; CHECK-NEXT:    ret i1 [[EQ]]
+;
+  %at = trunc nuw i8 %a to i1
+  %bt = trunc nuw i8 %b to i1
+  %eq = icmp eq i1 %at, %bt
+  ret i1 %eq
+}
+
 define i1 @trunc_unsigned_nuw_zext(i32 %x, i8 %y) {
 ; CHECK-LABEL: @trunc_unsigned_nuw_zext(
 ; CHECK-NEXT:    [[TMP1:%.*]] = zext i8 [[Y:%.*]] to i32

>From ed615d224cc4726d1f685c63bf728ca13262f03f Mon Sep 17 00:00:00 2001
From: Veera <sveera.2001 at gmail.com>
Date: Fri, 28 Mar 2025 04:42:27 +0000
Subject: [PATCH 2/2] [InstCombine] Fold `(trunc nuw A to i1) == (trunc nuw B
 to i1)` to `A == B`

---
 llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp | 6 +++---
 llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll   | 4 +---
 2 files changed, 4 insertions(+), 6 deletions(-)

diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 76020d2b1dbf4..e75b4026d5424 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -7451,6 +7451,9 @@ Instruction *InstCombinerImpl::visitICmpInst(ICmpInst &I) {
     }
   }
 
+  if (Instruction *Res = foldICmpTruncWithTruncOrExt(I, Q))
+    return Res;
+
   if (Op0->getType()->isIntOrIntVectorTy(1))
     if (Instruction *Res = canonicalizeICmpBool(I, Builder))
       return Res;
@@ -7473,9 +7476,6 @@ Instruction *InstCombinerImpl::visitICmpInst(ICmpInst &I) {
   if (Instruction *Res = foldICmpUsingKnownBits(I))
     return Res;
 
-  if (Instruction *Res = foldICmpTruncWithTruncOrExt(I, Q))
-    return Res;
-
   // Test if the ICmpInst instruction is used exclusively by a select as
   // part of a minimum or maximum operation. If so, refrain from doing
   // any other folding. This helps out other analyses which understand
diff --git a/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll b/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll
index 4ff9f8339be1a..aeb70a636682f 100644
--- a/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll
+++ b/llvm/test/Transforms/InstCombine/icmp-of-trunc-ext.ll
@@ -411,9 +411,7 @@ define i1 @trunc_equality_either(i16 %x, i16 %y) {
 
 define i1 @trunc_equality_bool(i8 %a, i8 %b) {
 ; CHECK-LABEL: @trunc_equality_bool(
-; CHECK-NEXT:    [[TMP1:%.*]] = xor i8 [[A:%.*]], [[B:%.*]]
-; CHECK-NEXT:    [[TMP2:%.*]] = trunc i8 [[TMP1]] to i1
-; CHECK-NEXT:    [[EQ:%.*]] = xor i1 [[TMP2]], true
+; CHECK-NEXT:    [[EQ:%.*]] = icmp eq i8 [[A:%.*]], [[B:%.*]]
 ; CHECK-NEXT:    ret i1 [[EQ]]
 ;
   %at = trunc nuw i8 %a to i1