[DAGCombiner] Handle extending EXTRACT_VECTOR_ELTs in calculateByteProvider #83963
Conversation
…ovider An EXTRACT_VECTOR_ELT can extend the element to the width of its result type, leaving the high bits undefined. Previously if we attempted to query the bytes in these high bits we would recurse and hit an assertion. This fixes it by bailing if the index is outside of the vector element size. I think the assertion Index < ByteWidth may still be incorrect, since ByteWidth is calculated from Op.getValueSizeInBits(). I believe this should be Op.getScalarValueSizeInBits() whenever VectorIndex is set since we're querying the element now, not the vector. But I couldn't think of a test case to trigger it. It can be addressed in a follow-up patch. Fixes llvm#83920
|
@llvm/pr-subscribers-llvm-selectiondag Author: Luke Lau (lukel97) ChangesAn EXTRACT_VECTOR_ELT can extend the element to the width of its result type, leaving the high bits undefined. Previously if we attempted to query the bytes in these high bits we would recurse and hit an assertion. This fixes it by bailing if the index is outside of the vector element size. I think the assertion Index < ByteWidth may still be incorrect, since ByteWidth is calculated from Op.getValueSizeInBits(). I believe this should be Op.getScalarValueSizeInBits() whenever VectorIndex is set since we're querying the element now, not the vector. But I couldn't think of a test case to trigger it. It can be addressed in a follow-up patch. Fixes #83920 Full diff: https://github.com/llvm/llvm-project/pull/83963.diff 2 Files Affected:
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 33ada3655dc731..f160dde25d26bf 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -8627,6 +8627,10 @@ calculateByteProvider(SDValue Op, unsigned Index, unsigned Depth,
if (NarrowBitWidth % 8 != 0)
return std::nullopt;
uint64_t NarrowByteWidth = NarrowBitWidth / 8;
+ // EXTRACT_VECTOR_ELT can extend the element type to the width of the return
+ // type, leaving the high bits undefined.
+ if (Index >= NarrowByteWidth)
+ return std::nullopt;
// Check to see if the position of the element in the vector corresponds
// with the byte we are trying to provide for. In the case of a vector of
diff --git a/llvm/test/CodeGen/RISCV/rvv/pr83920.ll b/llvm/test/CodeGen/RISCV/rvv/pr83920.ll
new file mode 100644
index 00000000000000..058417629c9c4e
--- /dev/null
+++ b/llvm/test/CodeGen/RISCV/rvv/pr83920.ll
@@ -0,0 +1,17 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; RUN: llc < %s -mtriple=riscv64 -mattr=+v -verify-machineinstrs | FileCheck %s
+
+define i8 @or_load_combine(ptr %p) {
+; CHECK-LABEL: or_load_combine:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: vsetivli zero, 2, e8, mf8, ta, ma
+; CHECK-NEXT: vle8.v v8, (a0)
+; CHECK-NEXT: vmv.x.s a0, v8
+; CHECK-NEXT: ori a0, a0, 1
+; CHECK-NEXT: ret
+entry:
+ %0 = load <2 x i8>, ptr %p
+ %1 = extractelement <2 x i8> %0, i64 0
+ %2 = or i8 %1, 1
+ ret i8 %2
+}
|
|
For reference, this was the failing assertion that I believe needs to be adjusted to handle the vector case: |
No, it doesn't seem to be handled in calculateByteProvider. |
An EXTRACT_VECTOR_ELT can extend the element to the width of its result type, leaving the high bits undefined. Previously if we attempted to query the bytes in these high bits we would recurse and hit an assertion. This fixes it by bailing if the index is outside of the vector element size.
I think the assertion Index < ByteWidth may still be incorrect, since ByteWidth is calculated from Op.getValueSizeInBits(). I believe this should be Op.getScalarValueSizeInBits() whenever VectorIndex is set since we're querying the element now, not the vector. But I couldn't think of a test case to trigger it. It can be addressed in a follow-up patch.
Fixes #83920