[llvm] 501dcab - [RISCV] Limit VLEN in getOptimalMemOpType to prevent creating invalid MVTs. (#139116)

[via llvm-commits]

Author: Craig Topper
Date: 2025-05-08T16:27:36-07:00
New Revision: 501dcab68ea978d8a3dbc46a8030b9227c4720de

URL: https://github.com/llvm/llvm-project/commit/501dcab68ea978d8a3dbc46a8030b9227c4720de
DIFF: https://github.com/llvm/llvm-project/commit/501dcab68ea978d8a3dbc46a8030b9227c4720de.diff

LOG: [RISCV] Limit VLEN in getOptimalMemOpType to prevent creating invalid MVTs. (#139116)

We only guarantee that types that are 1024 bytes or smaller exist in the
MVT enum.

Fixes #139075.

Added: 
    llvm/test/CodeGen/RISCV/rvv/pr139075.ll

Modified: 
    llvm/lib/Target/RISCV/RISCVISelLowering.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index c726fa74b922e..8403b51483323 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -23380,7 +23380,10 @@ EVT RISCVTargetLowering::getOptimalMemOpType(const MemOp &Op,
   // combining will typically form larger LMUL operations from the LMUL1
   // operations emitted here, and that's okay because combining isn't
   // introducing new memory operations; it's just merging existing ones.
-  const unsigned MinVLenInBytes = Subtarget.getRealMinVLen()/8;
+  // NOTE: We limit to 1024 bytes to avoid creating an invalid MVT.
+  const unsigned MinVLenInBytes =
+      std::min(Subtarget.getRealMinVLen() / 8, 1024U);
+
   if (Op.size() < MinVLenInBytes)
     // TODO: Figure out short memops.  For the moment, do the default thing
     // which ends up using scalar sequences.

diff  --git a/llvm/test/CodeGen/RISCV/rvv/pr139075.ll b/llvm/test/CodeGen/RISCV/rvv/pr139075.ll
new file mode 100644
index 0000000000000..33e8e13a21588
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/pr139075.ll
@@ -0,0 +1,18 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -mtriple=riscv64 -mattr=+v,+zvl16384b | FileCheck %s
+
+define void @a(ptr %0, ptr %1) {
+; CHECK-LABEL: a:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    li a2, 1024
+; CHECK-NEXT:    vsetvli zero, a2, e8, mf2, ta, ma
+; CHECK-NEXT:    vle8.v v8, (a1)
+; CHECK-NEXT:    vse8.v v8, (a0)
+; CHECK-NEXT:    addi a1, a1, 1024
+; CHECK-NEXT:    vle8.v v8, (a1)
+; CHECK-NEXT:    addi a0, a0, 1024
+; CHECK-NEXT:    vse8.v v8, (a0)
+; CHECK-NEXT:    ret
+  call void @llvm.memcpy.p0.p0.i64(ptr align 1 %0, ptr align 4 %1, i64 2048, i1 false)
+  ret void
+}