[llvm] [InstCombine] Try the flipped strictness of predicate in `foldICmpShlConstant` (PR #92773)

via llvm-commits llvm-commits at lists.llvm.org
Mon May 20 08:43:55 PDT 2024


llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-transforms

Author: Yingwei Zheng (dtcxzyw)

<details>
<summary>Changes</summary>

This patch extends the transform `(icmp pred iM (shl iM %v, N), C) -> (icmp pred i(M-N) (trunc %v iM to i(M-N)), (trunc (C>>N))` to handle icmps with the flipped strictness of predicate.

See the following case:
```
icmp ult i64 (shl X, 32), 8589934593 ->
icmp ule i64 (shl X, 32), 8589934592 ->
icmp ule i32 (trunc X, i32), 2 ->
icmp ult i32 (trunc X, i32), 3
```

Fixes the regression introduced by https://github.com/llvm/llvm-project/pull/86111#issuecomment-2098203152.

Alive2 proofs: https://alive2.llvm.org/ce/z/-sp5n3

`nuw` cannot be propagated as we always use `ashr` here. I don't see the value of fixing this (see the test `test_icmp_shl_nuw`).


---
Full diff: https://github.com/llvm/llvm-project/pull/92773.diff


2 Files Affected:

- (modified) llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp (+31-8) 
- (modified) llvm/test/Transforms/InstCombine/icmp.ll (+77) 


``````````diff
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 542a1c82b127a..f5952362f56bd 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -2414,14 +2414,37 @@ Instruction *InstCombinerImpl::foldICmpShlConstant(ICmpInst &Cmp,
   // free on the target. It has the additional benefit of comparing to a
   // smaller constant that may be more target-friendly.
   unsigned Amt = ShiftAmt->getLimitedValue(TypeBits - 1);
-  if (Shl->hasOneUse() && Amt != 0 && C.countr_zero() >= Amt &&
-      DL.isLegalInteger(TypeBits - Amt)) {
-    Type *TruncTy = IntegerType::get(Cmp.getContext(), TypeBits - Amt);
-    if (auto *ShVTy = dyn_cast<VectorType>(ShType))
-      TruncTy = VectorType::get(TruncTy, ShVTy->getElementCount());
-    Constant *NewC =
-        ConstantInt::get(TruncTy, C.ashr(*ShiftAmt).trunc(TypeBits - Amt));
-    return new ICmpInst(Pred, Builder.CreateTrunc(X, TruncTy), NewC);
+  if (Shl->hasOneUse() && Amt != 0 && DL.isLegalInteger(TypeBits - Amt)) {
+    auto FoldICmpShlToICmpTrunc = [&](ICmpInst::Predicate Pred,
+                                      const APInt &C) -> Instruction * {
+      if (C.countr_zero() < Amt)
+        return nullptr;
+      Type *TruncTy = ShType->getWithNewBitWidth(TypeBits - Amt);
+      Constant *NewC =
+          ConstantInt::get(TruncTy, C.ashr(*ShiftAmt).trunc(TypeBits - Amt));
+      return new ICmpInst(
+          Pred, Builder.CreateTrunc(X, TruncTy, "", Shl->hasNoSignedWrap()),
+          NewC);
+    };
+
+    if (Instruction *Res = FoldICmpShlToICmpTrunc(Pred, C))
+      return Res;
+
+    // Try the flipped strictness predicate.
+    // e.g.:
+    // icmp ult i64 (shl X, 32), 8589934593 ->
+    // icmp ule i64 (shl X, 32), 8589934592 ->
+    // icmp ule i32 (trunc X, i32), 2 ->
+    // icmp ult i32 (trunc X, i32), 3
+    if (auto FlippedStrictness =
+            InstCombiner::getFlippedStrictnessPredicateAndConstant(
+                Pred, ConstantInt::get(ShType->getContext(), C))) {
+      ICmpInst::Predicate NewPred = FlippedStrictness->first;
+      const APInt &NewC =
+          cast<ConstantInt>(FlippedStrictness->second)->getValue();
+      if (Instruction *Res = FoldICmpShlToICmpTrunc(NewPred, NewC))
+        return Res;
+    }
   }
 
   return nullptr;
diff --git a/llvm/test/Transforms/InstCombine/icmp.ll b/llvm/test/Transforms/InstCombine/icmp.ll
index 2d786c8f48833..319e13235ccd6 100644
--- a/llvm/test/Transforms/InstCombine/icmp.ll
+++ b/llvm/test/Transforms/InstCombine/icmp.ll
@@ -5198,3 +5198,80 @@ define i1 @icmp_freeze_sext(i16 %x, i16 %y) {
   %cmp2 = icmp uge i16 %ext.fr, %y
   ret i1 %cmp2
 }
+
+define i1 @test_icmp_shl(i64 %x) {
+; CHECK-LABEL: @test_icmp_shl(
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc i64 [[X:%.*]] to i32
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[TMP1]], 3
+; CHECK-NEXT:    ret i1 [[CMP]]
+;
+  %shl = shl i64 %x, 32
+  %cmp = icmp ult i64 %shl, 8589934593
+  ret i1 %cmp
+}
+
+define i1 @test_icmp_shl_multiuse(i64 %x) {
+; CHECK-LABEL: @test_icmp_shl_multiuse(
+; CHECK-NEXT:    [[SHL:%.*]] = shl i64 [[X:%.*]], 32
+; CHECK-NEXT:    call void @use_i64(i64 [[SHL]])
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i64 [[SHL]], 8589934593
+; CHECK-NEXT:    ret i1 [[CMP]]
+;
+  %shl = shl i64 %x, 32
+  call void @use_i64(i64 %shl)
+  %cmp = icmp ult i64 %shl, 8589934593
+  ret i1 %cmp
+}
+
+define i1 @test_icmp_shl_illegal_length(i64 %x) {
+; CHECK-LABEL: @test_icmp_shl_illegal_length(
+; CHECK-NEXT:    [[SHL:%.*]] = shl i64 [[X:%.*]], 31
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i64 [[SHL]], 8589934593
+; CHECK-NEXT:    ret i1 [[CMP]]
+;
+  %shl = shl i64 %x, 31
+  %cmp = icmp ult i64 %shl, 8589934593
+  ret i1 %cmp
+}
+
+define i1 @test_icmp_shl_invalid_rhsc(i64 %x) {
+; CHECK-LABEL: @test_icmp_shl_invalid_rhsc(
+; CHECK-NEXT:    [[SHL:%.*]] = shl i64 [[X:%.*]], 32
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i64 [[SHL]], 8589934595
+; CHECK-NEXT:    ret i1 [[CMP]]
+;
+  %shl = shl i64 %x, 32
+  %cmp = icmp ult i64 %shl, 8589934595
+  ret i1 %cmp
+}
+
+define i1 @test_icmp_shl_nuw(i64 %x) {
+; CHECK-LABEL: @test_icmp_shl_nuw(
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i64 [[X:%.*]], 3
+; CHECK-NEXT:    ret i1 [[CMP]]
+;
+  %shl = shl nuw i64 %x, 32
+  %cmp = icmp ult i64 %shl, 8589934593
+  ret i1 %cmp
+}
+
+define i1 @test_icmp_shl_nsw(i64 %x) {
+; CHECK-LABEL: @test_icmp_shl_nsw(
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i64 [[X:%.*]], 3
+; CHECK-NEXT:    ret i1 [[CMP]]
+;
+  %shl = shl nsw i64 %x, 32
+  %cmp = icmp ult i64 %shl, 8589934593
+  ret i1 %cmp
+}
+
+define <2 x i1> @test_icmp_shl_vec(<2 x i64> %x) {
+; CHECK-LABEL: @test_icmp_shl_vec(
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc <2 x i64> [[X:%.*]] to <2 x i32>
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult <2 x i32> [[TMP1]], <i32 3, i32 3>
+; CHECK-NEXT:    ret <2 x i1> [[CMP]]
+;
+  %shl = shl <2 x i64> %x, splat(i64 32)
+  %cmp = icmp ult <2 x i64> %shl, splat(i64 8589934593)
+  ret <2 x i1> %cmp
+}

``````````

</details>


https://github.com/llvm/llvm-project/pull/92773


More information about the llvm-commits mailing list