[llvm] [RISCV] Disable fixed vectors in getOptimalMemOpType if minimum VLEN is 32. (PR #102974)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-risc-v

Author: Craig Topper (topperc)

<details>
<summary>Changes</summary>

This is needed to match #<!-- -->102405 which disabled fixed to scalable vector lowering for VLEN=32.

Fixes #<!-- -->102566 and #<!-- -->102568.

---
Full diff: https://github.com/llvm/llvm-project/pull/102974.diff


2 Files Affected:

- (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+5) 
- (added) llvm/test/CodeGen/RISCV/rvv/memcpy-crash-zvl32b.ll (+17) 


``````````diff
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 3beaa5198259d0..6fde09d89e4839 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -21024,6 +21024,11 @@ EVT RISCVTargetLowering::getOptimalMemOpType(const MemOp &Op,
     // which ends up using scalar sequences.
     return MVT::Other;
 
+  // If the minimum VLEN is less than RISCV::RVVBitsPerBlock we don't support
+  // fixed vectors.
+  if (MinVLenInBytes <= RISCV::RVVBitsPerBlock / 8)
+    return MVT::Other;
+
   // Prefer i8 for non-zero memset as it allows us to avoid materializing
   // a large scalar constant and instead use vmv.v.x/i to do the
   // broadcast.  For everything else, prefer ELenVT to minimize VL and thus
diff --git a/llvm/test/CodeGen/RISCV/rvv/memcpy-crash-zvl32b.ll b/llvm/test/CodeGen/RISCV/rvv/memcpy-crash-zvl32b.ll
new file mode 100644
index 00000000000000..e020fe1a0aa1ac
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/memcpy-crash-zvl32b.ll
@@ -0,0 +1,17 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -mtriple=riscv64 -mattr=+zve32x | FileCheck %s
+
+; Make sure we don't with VLEN=32.
+
+define void @c() {
+; CHECK-LABEL: c:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    lw a0, 0(zero)
+; CHECK-NEXT:    sw a0, 0(zero)
+; CHECK-NEXT:    ret
+entry:
+  call void @llvm.memcpy.p0.p0.i64(ptr null, ptr null, i64 4, i1 false)
+  ret void
+}
+
+declare void @llvm.memcpy.p0.p0.i64(ptr noalias nocapture writeonly, ptr noalias nocapture readonly, i64, i1 immarg) #1

``````````

</details>


https://github.com/llvm/llvm-project/pull/102974