[llvm] [DAGCombine] Fix crash caused by illegal InterVT in ForwardStoreValueToDirectLoad (PR #181175)

via llvm-commits llvm-commits at lists.llvm.org
Thu Feb 12 08:33:36 PST 2026


https://github.com/Michael-Chen-NJU created https://github.com/llvm/llvm-project/pull/181175

This patch fixes an assertion failure in ForwardStoreValueToDirectLoad during DAGCombine.

The crash occurs when `STLF (Store-to-Load Forwarding)` creates an illegal intermediate bitcast type (e.g., `v128i1` when bridging a 128-bit store to a `<32 x i1>` load on X86). Since `v128i1` is not a legal mask type for the backend, it violates the expectations of the LegalizeDAG pass.

The fix adds a `TLI.isTypeLegal(InterVT)` check to ensure that the intermediate type used for the transformation is supported by the target.

Fixes #181130

>From 282956a57cbd16a7821d8c4bc6f81525a5d55aad Mon Sep 17 00:00:00 2001
From: Michael-Chen-NJU <2802328816 at qq.com>
Date: Fri, 13 Feb 2026 00:26:32 +0800
Subject: [PATCH] [DAGCombine] Fix crash caused by illegal InterVT in
 ForwardStoreValueToDirectLoad

---
 llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp |  3 ++-
 .../CodeGen/X86/stlf-v32i1-bitcast-crash.ll   | 21 +++++++++++++++++++
 2 files changed, 23 insertions(+), 1 deletion(-)
 create mode 100644 llvm/test/CodeGen/X86/stlf-v32i1-bitcast-crash.ll

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 19bcdeeefb143..e513a16ee0e27 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -20831,7 +20831,8 @@ SDValue DAGCombiner::ForwardStoreValueToDirectLoad(LoadSDNode *LD) {
 
         EVT InterVT = EVT::getVectorVT(*DAG.getContext(), EltVT,
                                        StMemSize.divideCoefficientBy(EltSize));
-        if (!TLI.isOperationLegalOrCustom(ISD::EXTRACT_SUBVECTOR, LDMemType))
+        if (!TLI.isOperationLegalOrCustom(ISD::EXTRACT_SUBVECTOR, LDMemType) ||
+            !TLI.isTypeLegal(InterVT))
           break;
 
         // In case of big-endian the offset is normalized to zero, denoting
diff --git a/llvm/test/CodeGen/X86/stlf-v32i1-bitcast-crash.ll b/llvm/test/CodeGen/X86/stlf-v32i1-bitcast-crash.ll
new file mode 100644
index 0000000000000..2abdc547dd479
--- /dev/null
+++ b/llvm/test/CodeGen/X86/stlf-v32i1-bitcast-crash.ll
@@ -0,0 +1,21 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc < %s -mtriple=x86_64-pc-windows-gnu -mcpu=tigerlake | FileCheck %s
+
+define <32 x i1> @v32i1_bitcast_crash(<4 x i8> %0) {
+; CHECK-LABEL: v32i1_bitcast_crash:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    subq $24, %rsp
+; CHECK-NEXT:    .seh_stackalloc 24
+; CHECK-NEXT:    .seh_endprologue
+; CHECK-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-NEXT:    vmovaps %xmm0, (%rsp)
+; CHECK-NEXT:    kmovd (%rsp), %k0
+; CHECK-NEXT:    vpmovm2b %k0, %ymm0
+; CHECK-NEXT:    .seh_startepilogue
+; CHECK-NEXT:    addq $24, %rsp
+; CHECK-NEXT:    .seh_endepilogue
+; CHECK-NEXT:    retq
+; CHECK-NEXT:    .seh_endproc
+  %2 = bitcast <4 x i8> %0 to <32 x i1>
+  ret <32 x i1> %2
+}



More information about the llvm-commits mailing list