[PATCH] D46085: AMDGPU/SI: Don't promote alloca to vector for atomic load/store

Changpeng Fang via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Wed Apr 25 15:48:14 PDT 2018


cfang marked an inline comment as done.
cfang added inline comments.


================
Comment at: test/CodeGen/AMDGPU/vector-alloca-atomic.ll:2
+; RUN: opt -S -mtriple=amdgcn---amdgiz -amdgpu-promote-alloca -sroa -instcombine < %s | FileCheck -check-prefix=OPT %s
+target datalayout = "A5"
+
----------------
arsenm wrote:
> You shouldn't need this
If I remove the line "target datalayout = "A5"", I will get the following errors:

 chfang at shanghai:~/bug/llvm/build/bin$ ./opt -S -mtriple=amdgcn---amdgiz -amdgpu-promote-alloca -sroa -instcombine < ../../test/CodeGen/AMDGPU/vector-alloca-atomic.ll | ./FileCheck -check-prefix=OPT ../../test/CodeGen/AMDGPU/vector-alloca-atomic.ll
Allocation instruction pointer not in the stack address space!
  %alloca = alloca [3 x i32], addrspace(5)
Allocation instruction pointer not in the stack address space!
  %alloca = alloca [3 x i32], addrspace(5)
Allocation instruction pointer not in the stack address space!
  %alloca = alloca [3 x i32], addrspace(5)
./opt: -: error: input module is broken!
FileCheck error: '-' is empty.



https://reviews.llvm.org/D46085





More information about the llvm-commits mailing list