llvm.org GIT mirror llvm / 07bc0b6
[NVPTX] Add missing .v4 qualifier on vector store instruction git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@213276 91177308-0d34-0410-b5e6-96231b3b80d8 Justin Holewinski 6 years ago
2 changed file(s) with 13 addition(s) and 1 deletion(s). Raw diff Collapse all Expand all
19161916 def StoreParamV4I32 : NVPTXInst<(outs), (ins Int32Regs:$val, Int32Regs:$val2,
19171917 Int32Regs:$val3, Int32Regs:$val4,
19181918 i32imm:$a, i32imm:$b),
1919 "st.param.b32\t[param$a+$b], {{$val, $val2, $val3, $val4}};",
1919 "st.param.v4.b32\t[param$a+$b], {{$val, $val2, $val3, $val4}};",
19201920 []>;
19211921
19221922 def StoreParamV4I16 : NVPTXInst<(outs), (ins Int16Regs:$val, Int16Regs:$val2,
0 ; RUN: llc < %s -march=nvptx -mcpu=sm_20 | FileCheck %s
1
2 target triple = "nvptx-unknown-cuda"
3
4 declare void @bar(<4 x i32>)
5
6 ; CHECK-LABEL @foo
7 define void @foo(<4 x i32> %a) {
8 ; CHECK: st.param.v4.b32
9 tail call void @bar(<4 x i32> %a)
10 ret void
11 }