[InstSimplify] Fix crash when optimizing minmax with bitcast constant vectors

[igorban-intel]

When simplifying min/max intrinsics with fixed-size vector constants,
InstructionSimplify attempts to optimize element-wise. However,
getAggregateElement() can return null for certain constant expressions
like bitcasts, leading to a null pointer dereference.

This patch adds a check to bail out of the optimization when
getAggregateElement() returns null, preventing the crash while
maintaining correct behavior for normal constant vectors.

Fixes crash with patterns like:
call <2 x half> @llvm.minnum.v2f16(<2 x half> %x,
<2 x half> bitcast (<1 x i32> to <2 x half>))

[github-actions]

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]

@llvm/pr-subscribers-llvm-transforms

@llvm/pr-subscribers-llvm-analysis

Author: Igor Gorban (igorban-intel)

Changes

When simplifying min/max intrinsics with fixed-size vector constants,
InstructionSimplify attempts to optimize element-wise. However,
getAggregateElement() can return null for certain constant expressions
like bitcasts, leading to a null pointer dereference.

This patch adds a check to bail out of the optimization when
getAggregateElement() returns null, preventing the crash while
maintaining correct behavior for normal constant vectors.

Fixes crash with patterns like:
call <2 x half> @llvm.minnum.v2f16(<2 x half> %x,
<2 x half> bitcast (<1 x i32> <i32 N> to <2 x half>))

---

Full diff: https://github.com/llvm/llvm-project/pull/168055.diff

2 Files Affected:

• (modified) llvm/lib/Analysis/InstructionSimplify.cpp (+4)
• (modified) llvm/test/Transforms/InstSimplify/fminmax-folds.ll (+70)

diff --git a/llvm/lib/Analysis/InstructionSimplify.cpp b/llvm/lib/Analysis/InstructionSimplify.cpp
index 2a0a6a2d302b1..e5d2842a0ae7b 100644
--- a/llvm/lib/Analysis/InstructionSimplify.cpp
+++ b/llvm/lib/Analysis/InstructionSimplify.cpp
@@ -6985,6 +6985,10 @@ Value *llvm::simplifyBinaryIntrinsic(Intrinsic::ID IID, Type *ReturnType,
           // VectorShuffle instruction, which is not allowed in simplifyBinOp.
           OptResult = MinMaxOptResult::UseEither;
           for (unsigned i = 0; i != ElemCount.getFixedValue(); ++i) {
+            if (!C->getAggregateElement(i)) {
+              OptResult = MinMaxOptResult::CannotOptimize;
+              break;
+            }
             auto ElemResult = OptimizeConstMinMax(C->getAggregateElement(i),
                                                   IID, Call, &NewConst);
             if (ElemResult == MinMaxOptResult::CannotOptimize ||
diff --git a/llvm/test/Transforms/InstSimplify/fminmax-folds.ll b/llvm/test/Transforms/InstSimplify/fminmax-folds.ll
index 3a03f8627ab68..091e85920c0df 100644
--- a/llvm/test/Transforms/InstSimplify/fminmax-folds.ll
+++ b/llvm/test/Transforms/InstSimplify/fminmax-folds.ll
@@ -885,3 +885,73 @@ define void @minmax_minmax_xy_maxmin_yx(double %x, double %y, ptr %minnum_res, p
   store double %final_maximumnum, ptr %maximumnum_res
   ret void
 }
+
+;###############################################################
+;#            Constant Expression Vector Tests                 #
+;###############################################################
+; Test that minmax intrinsics with constant expression vectors don't crash
+; when getAggregateElement returns null for certain constant expressions.
+; These tests cover various scenarios where getAggregateElement() fails:
+; - Bitcast from mismatched vector element counts
+; - Bitcast from integer to float vectors
+; - Bitcast from i64 with different element boundaries
+
+; Test with bitcast from <1 x i32> to <2 x half> (element count mismatch)
+define <2 x half> @minmax_bitcast_v2f16_minnum(<2 x half> %x) {
+; CHECK-LABEL: @minmax_bitcast_v2f16_minnum(
+; CHECK-NEXT:    [[RESULT:%.*]] = call <2 x half> @llvm.minnum.v2f16(<2 x half> [[X:%.*]], <2 x half> bitcast (<1 x i32> splat (i32 1078530011) to <2 x half>))
+; CHECK-NEXT:    ret <2 x half> [[RESULT]]
+;
+  %result = call <2 x half> @llvm.minnum.v2f16(<2 x half> %x, <2 x half> bitcast (<1 x i32> <i32 1078530011> to <2 x half>))
+  ret <2 x half> %result
+}
+
+; Test with bitcast from <2 x i32> to <4 x half> (different element boundaries)
+define <4 x half> @minmax_bitcast_v4f16_maxnum(<4 x half> %x) {
+; CHECK-LABEL: @minmax_bitcast_v4f16_maxnum(
+; CHECK-NEXT:    [[RESULT:%.*]] = call <4 x half> @llvm.maxnum.v4f16(<4 x half> [[X:%.*]], <4 x half> bitcast (<2 x i32> <i32 1078530011, i32 1073741824> to <4 x half>))
+; CHECK-NEXT:    ret <4 x half> [[RESULT]]
+;
+  %result = call <4 x half> @llvm.maxnum.v4f16(<4 x half> %x, <4 x half> bitcast (<2 x i32> <i32 1078530011, i32 1073741824> to <4 x half>))
+  ret <4 x half> %result
+}
+
+; Test with bitcast from <1 x i64> to <2 x float> (scalar to vector bitcast)
+define <2 x float> @minmax_bitcast_v2f32_minimum(<2 x float> %x) {
+; CHECK-LABEL: @minmax_bitcast_v2f32_minimum(
+; CHECK-NEXT:    [[RESULT:%.*]] = call <2 x float> @llvm.minimum.v2f32(<2 x float> [[X:%.*]], <2 x float> bitcast (<1 x i64> splat (i64 4638564619268087808) to <2 x float>))
+; CHECK-NEXT:    ret <2 x float> [[RESULT]]
+;
+  %result = call <2 x float> @llvm.minimum.v2f32(<2 x float> %x, <2 x float> bitcast (<1 x i64> <i64 4638564619268087808> to <2 x float>))
+  ret <2 x float> %result
+}
+
+; Test with bitcast from <1 x double> to <4 x half> (type size mismatch)
+define <4 x half> @minmax_bitcast_v4f16_maximum(<4 x half> %x) {
+; CHECK-LABEL: @minmax_bitcast_v4f16_maximum(
+; CHECK-NEXT:    [[RESULT:%.*]] = call <4 x half> @llvm.maximum.v4f16(<4 x half> [[X:%.*]], <4 x half> bitcast (<1 x double> splat (double 0x400921FB54442D18) to <4 x half>))
+; CHECK-NEXT:    ret <4 x half> [[RESULT]]
+;
+  %result = call <4 x half> @llvm.maximum.v4f16(<4 x half> %x, <4 x half> bitcast (<1 x double> <double 0x400921FB54442D18> to <4 x half>))
+  ret <4 x half> %result
+}
+
+; Test with bitcast from <2 x i16> to <2 x half> (integer to float)
+define <2 x half> @minmax_bitcast_v2f16_minimumnum(<2 x half> %x) {
+; CHECK-LABEL: @minmax_bitcast_v2f16_minimumnum(
+; CHECK-NEXT:    [[RESULT:%.*]] = call <2 x half> @llvm.minimumnum.v2f16(<2 x half> [[X:%.*]], <2 x half> <half 0xH3F80, half 0xH3F00>)
+; CHECK-NEXT:    ret <2 x half> [[RESULT]]
+;
+  %result = call <2 x half> @llvm.minimumnum.v2f16(<2 x half> %x, <2 x half> bitcast (<2 x i16> <i16 16256, i16 16128> to <2 x half>))
+  ret <2 x half> %result
+}
+
+; Test with bitcast from <4 x i16> to <4 x half> (matching element count but getAggregateElement may fail)
+define <4 x half> @minmax_bitcast_v4f16_maximumnum(<4 x half> %x) {
+; CHECK-LABEL: @minmax_bitcast_v4f16_maximumnum(
+; CHECK-NEXT:    [[RESULT:%.*]] = call <4 x half> @llvm.maximumnum.v4f16(<4 x half> [[X:%.*]], <4 x half> <half 0xH3C00, half 0xH3F00, half 0xH3F80, half 0xH4000>)
+; CHECK-NEXT:    ret <4 x half> [[RESULT]]
+;
+  %result = call <4 x half> @llvm.maximumnum.v4f16(<4 x half> %x, <4 x half> bitcast (<4 x i16> <i16 15360, i16 16128, i16 16256, i16 16384> to <4 x half>))
+  ret <4 x half> %result
+}

[dtcxzyw]

LGTM. Nice catch!

[dtcxzyw]

Reuse the result in the following call. See also

llvm-project/llvm/include/llvm/IR/PatternMatch.h

Lines 388 to 390 in 3d41567

	Constant *Elt = C->getAggregateElement(i);
	if (!Elt)
	return false;

[igorban-intel]

Thanks, fixed.

[dtcxzyw]

@igorban-intel Do you have commit access?

[igorban-intel]

@igorban-intel Do you have commit access?

Do you mean the right to merge? Unfortunately, no.

[github-actions]

@igorban-intel 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!

[igorban-intel]

Thanks a lot!