[llvm-bugs] [Bug 45833] New: llvm::DAGTypeLegalizer::SplitVecOp_MSTORE crash
via llvm-bugs
llvm-bugs at lists.llvm.org
Thu May 7 12:52:15 PDT 2020
https://bugs.llvm.org/show_bug.cgi?id=45833
Bug ID: 45833
Summary: llvm::DAGTypeLegalizer::SplitVecOp_MSTORE crash
Product: new-bugs
Version: unspecified
Hardware: PC
OS: Linux
Status: NEW
Severity: enhancement
Priority: P
Component: new bugs
Assignee: unassignedbugs at nondot.org
Reporter: ajcbik at google.com
CC: htmldeveloper at gmail.com, llvm-bugs at lists.llvm.org
This bug is actually a continuation of the earlier bug
https://bugs.llvm.org/show_bug.cgi?id=45563
that exposed a problem with loads on unusual vector lengths.
As I suspected, a similar issue arises for stores.
Take the following MLIR input
func @transfer_write13_1d(%A : memref<?xf32>, %base: index) {
%f = constant 13.0 : f32
%v = splat %f : vector<13xf32>
vector.transfer_write %v, %A[%base]
{permutation_map = affine_map<(d0) -> (d0)>}
: vector<13xf32>, memref<?xf32>
return
}
and run it through the lowering to LLVM-IR path for execution, and it crashes
LLVM as follows
F0507 12:50:26.508680 34021 logging.cc:106] assert.h assertion failed at
third_party/llvm/llvm-project/llvm/include/llvm/CodeGen/ValueTypes.h:371 in
llvm::EVT llvm::EVT::getHalfNumVectorElementsVT(llvm::LLVMContext &) const:
!(EltCnt.Min & 1) && "Splitting vector, but not in half!"
*** Check failure stack trace: ***
@ 0x557cd9079f40 absl::logging_internal::LogMessage::DieIfFatal()
@ 0x557cd9079813 absl::logging_internal::LogMessage::SendToLog()
@ 0x557cd9078ade absl::logging_internal::LogMessage::Flush()
@ 0x557cd907ad98
absl::logging_internal::LogMessageFatal::~LogMessageFatal()
@ 0x557cd9076a40 __assert_fail
@ 0x557cd72a6807 llvm::EVT::getHalfNumVectorElementsVT()
@ 0x557cd729c1cb llvm::SelectionDAG::GetSplitDestVTs()
@ 0x557cd739c534 llvm::DAGTypeLegalizer::SplitVecOp_MSTORE()
@ 0x557cd7398366 llvm::DAGTypeLegalizer::SplitVectorOperand()
@ 0x557cd735cb0d llvm::DAGTypeLegalizer::run()
The fix is probably very similar.
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20200507/1f9e4903/attachment.html>
More information about the llvm-bugs
mailing list