[InstCombine] Canonicalize icmp eq pow2 more thoroughly

We currently already canonicalize icmp eq (%x & Pow2), Pow2 to icmp ne (%x & Pow2), 0. This patch generalizes the fold based on known bits. In particular, this allows us to handle comparisons against !range !{i64 0, i64 2} loads, which addresses an optimization regression in Rust caused by 8df376d. Differential Revision: https://reviews.llvm.org/D146149
llvm · Mar 16, 2023 · 61d2f3a · 61d2f3a
1 parent ddbcd98
commit 61d2f3a
Show file tree

Hide file tree

Showing 6 changed files with 14 additions and 21 deletions.
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -3239,16 +3239,6 @@ Instruction *InstCombinerImpl::foldICmpBinOpEqualityWithConstant(
     }
     break;
   }
-  case Instruction::And: {
-    const APInt *BOC;
-    if (match(BOp1, m_APInt(BOC))) {
-      // If we have ((X & C) == C), turn it into ((X & C) != 0).
-      if (C == *BOC && C.isPowerOf2())
-        return new ICmpInst(isICMP_NE ? ICmpInst::ICMP_EQ : ICmpInst::ICMP_NE,
-                            BO, Constant::getNullValue(RHS->getType()));
-    }
-    break;
-  }
   case Instruction::UDiv:
     if (C.isZero()) {
       // (icmp eq/ne (udiv A, B), 0) -> (icmp ugt/ule i32 B, A)
@@ -5771,6 +5761,12 @@ Instruction *InstCombinerImpl::foldICmpUsingKnownBits(ICmpInst &I) {
         }
       }
     }
+
+    // Op0 eq C_Pow2 -> Op0 ne 0 if Op0 is known to be C_Pow2 or zero.
+    if (Op1Known.isConstant() && Op1Known.getConstant().isPowerOf2() &&
+        (Op0Known & Op1Known) == Op0Known)
+      return new ICmpInst(CmpInst::getInversePredicate(Pred), Op0,
+                          ConstantInt::getNullValue(Op1->getType()));
     break;
   }
   case ICmpInst::ICMP_ULT: {

diff --git a/llvm/test/Transforms/InstCombine/div.ll b/llvm/test/Transforms/InstCombine/div.ll
@@ -400,7 +400,7 @@ define i32 @test28(i32 %a) {
 
 define i32 @test29(i32 %a) {
 ; CHECK-LABEL: @test29(
-; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i32 [[A:%.*]], -1
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp ne i32 [[A:%.*]], 0
 ; CHECK-NEXT:    [[DIV:%.*]] = zext i1 [[TMP1]] to i32
 ; CHECK-NEXT:    ret i32 [[DIV]]
 ;

diff --git a/llvm/test/Transforms/InstCombine/icmp-range.ll b/llvm/test/Transforms/InstCombine/icmp-range.ll
@@ -1004,7 +1004,7 @@ define i1 @icmp_eq_bool_0(ptr %ptr) {
 define i1 @icmp_eq_bool_1(ptr %ptr) {
 ; CHECK-LABEL: @icmp_eq_bool_1(
 ; CHECK-NEXT:    [[VAL:%.*]] = load i64, ptr [[PTR:%.*]], align 8, !range [[RNG6]]
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i64 [[VAL]], 1
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ne i64 [[VAL]], 0
 ; CHECK-NEXT:    ret i1 [[CMP]]
 ;
   %val = load i64, ptr %ptr, align 8, !range !{i64 0, i64 2}
@@ -1026,7 +1026,7 @@ define i1 @icmp_ne_bool_0(ptr %ptr) {
 define i1 @icmp_ne_bool_1(ptr %ptr) {
 ; CHECK-LABEL: @icmp_ne_bool_1(
 ; CHECK-NEXT:    [[VAL:%.*]] = load i64, ptr [[PTR:%.*]], align 8, !range [[RNG6]]
-; CHECK-NEXT:    [[CMP:%.*]] = icmp ne i64 [[VAL]], 1
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i64 [[VAL]], 0
 ; CHECK-NEXT:    ret i1 [[CMP]]
 ;
   %val = load i64, ptr %ptr, align 8, !range !{i64 0, i64 2}

diff --git a/llvm/test/Transforms/InstCombine/icmp-shl-nsw.ll b/llvm/test/Transforms/InstCombine/icmp-shl-nsw.ll
@@ -162,7 +162,7 @@ define i1 @icmp_sgt8(i8 %x) {
 
 define i1 @icmp_sgt9(i8 %x) {
 ; CHECK-LABEL: @icmp_sgt9(
-; CHECK-NEXT:    [[CMP:%.*]] = icmp ne i8 [[X:%.*]], -1
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i8 [[X:%.*]], 0
 ; CHECK-NEXT:    ret i1 [[CMP]]
 ;
   %shl = shl nsw i8 %x, 7
@@ -303,7 +303,7 @@ define i1 @icmp_sle8(i8 %x) {
 
 define i1 @icmp_sle9(i8 %x) {
 ; CHECK-LABEL: @icmp_sle9(
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i8 [[X:%.*]], -1
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ne i8 [[X:%.*]], 0
 ; CHECK-NEXT:    ret i1 [[CMP]]
 ;
   %shl = shl nsw i8 %x, 7

diff --git a/llvm/test/Transforms/InstCombine/icmp-shr-lt-gt.ll b/llvm/test/Transforms/InstCombine/icmp-shr-lt-gt.ll
@@ -2767,7 +2767,7 @@ define i1 @lshrult_03_00_exact(i4 %x) {
 
 define i1 @lshrult_03_01_exact(i4 %x) {
 ; CHECK-LABEL: @lshrult_03_01_exact(
-; CHECK-NEXT:    [[C:%.*]] = icmp ne i4 [[X:%.*]], -8
+; CHECK-NEXT:    [[C:%.*]] = icmp eq i4 [[X:%.*]], 0
 ; CHECK-NEXT:    ret i1 [[C]]
 ;
   %s = lshr exact i4 %x, 3

diff --git a/llvm/test/Transforms/InstCombine/zext.ll b/llvm/test/Transforms/InstCombine/zext.ll
@@ -718,9 +718,7 @@ define i64 @zext_icmp_eq_bool_0(ptr %ptr) {
 define i64 @zext_icmp_eq_bool_1(ptr %ptr) {
 ; CHECK-LABEL: @zext_icmp_eq_bool_1(
 ; CHECK-NEXT:    [[VAL:%.*]] = load i64, ptr [[PTR:%.*]], align 8, !range [[RNG0]]
-; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i64 [[VAL]], 1
-; CHECK-NEXT:    [[LEN:%.*]] = zext i1 [[CMP]] to i64
-; CHECK-NEXT:    ret i64 [[LEN]]
+; CHECK-NEXT:    ret i64 [[VAL]]
 ;
   %val = load i64, ptr %ptr, align 8, !range !{i64 0, i64 2}
   %cmp = icmp eq i64 %val, 1
@@ -742,8 +740,7 @@ define i64 @zext_icmp_ne_bool_0(ptr %ptr) {
 define i64 @zext_icmp_ne_bool_1(ptr %ptr) {
 ; CHECK-LABEL: @zext_icmp_ne_bool_1(
 ; CHECK-NEXT:    [[VAL:%.*]] = load i64, ptr [[PTR:%.*]], align 8, !range [[RNG0]]
-; CHECK-NEXT:    [[CMP:%.*]] = icmp ne i64 [[VAL]], 1
-; CHECK-NEXT:    [[LEN:%.*]] = zext i1 [[CMP]] to i64
+; CHECK-NEXT:    [[LEN:%.*]] = xor i64 [[VAL]], 1
 ; CHECK-NEXT:    ret i64 [[LEN]]
 ;
   %val = load i64, ptr %ptr, align 8, !range !{i64 0, i64 2}