brintos

brintos / llvm-project-archived public Read only

0
0
Text · 1.0 KiB · 3dceefb Raw
29 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --function foo --extra_scrub --default-march nvptx64 --filter-out ".*//.*" --filter-out "[\{\}\(\)]" --version 52 3; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_50 | FileCheck %s4; RUN: %if ptxas %{ llc < %s -mtriple=nvptx64 -mcpu=sm_50 | %ptxas-verify %}5 6target triple = "nvptx-nvidia-cuda"7 8define ptx_kernel void @foo(ptr noalias readonly %ptr, ptr noalias %retval) {9; CHECK-LABEL: foo(10; CHECK:    .reg .b32 %r<4>;11; CHECK:    .reg .b64 %rd<5>;12; CHECK-EMPTY:13; CHECK:    ld.param.b64 %rd1, [foo_param_0];14; CHECK:    cvta.to.global.u64 %rd2, %rd1;15; CHECK:    ld.param.b64 %rd3, [foo_param_1];16; CHECK:    cvta.to.global.u64 %rd4, %rd3;17; CHECK:    ld.global.nc.b8 %r1, [%rd2];18; CHECK:    add.s32 %r2, %r1, 1;19; CHECK:    and.b32 %r3, %r2, 1;20; CHECK:    st.global.b32 [%rd4], %r3;21; CHECK:    ret;22  %ld = load i1, ptr %ptr, align 123  %zext = zext i1 %ld to i3224  %add = add i32 %zext, 125  %and = and i32 %add, 126  store i32 %and, ptr %retval27  ret void28}29