brintos

brintos / llvm-project-archived public Read only

0
0
Text · 2.2 KiB · 5c4b5ba Raw
47 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 52; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_100 -mattr=+ptx86 | FileCheck --check-prefixes=CHECK,CHECK-PTX64 %s3; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_100 -mattr=+ptx86 --nvptx-short-ptr | FileCheck --check-prefixes=CHECK,CHECK-PTX-SHARED32 %s4; RUN: %if ptxas-sm_100 && ptxas-isa-8.6 %{ llc < %s -mtriple=nvptx64 -mcpu=sm_100 -mattr=+ptx86 | %ptxas-verify -arch=sm_100 %}5; RUN: %if ptxas-sm_100 && ptxas-isa-8.6 %{ llc < %s -mtriple=nvptx64 -mcpu=sm_100 -mattr=+ptx86 --nvptx-short-ptr | %ptxas-verify -arch=sm_100 %}6 7declare void @llvm.nvvm.st.bulk(ptr, i64, i64)8define void @st_bulk(ptr %dest_addr, i64 %size) {9; CHECK-LABEL: st_bulk(10; CHECK:       {11; CHECK-NEXT:    .reg .b64 %rd<3>;12; CHECK-EMPTY:13; CHECK-NEXT:  // %bb.0:14; CHECK-NEXT:    ld.param.b64 %rd1, [st_bulk_param_0];15; CHECK-NEXT:    ld.param.b64 %rd2, [st_bulk_param_1];16; CHECK-NEXT:    st.bulk [%rd1], %rd2, 0;17; CHECK-NEXT:    ret;18  call void @llvm.nvvm.st.bulk(ptr %dest_addr, i64 %size, i64 0)19  ret void20}21 22declare void @llvm.nvvm.st.bulk.shared.cta(ptr addrspace(3), i64, i64)23define void @st_bulk_shared_cta(ptr addrspace(3) %dest_addr, i64 %size) {24; CHECK-PTX64-LABEL: st_bulk_shared_cta(25; CHECK-PTX64:       {26; CHECK-PTX64-NEXT:    .reg .b64 %rd<3>;27; CHECK-PTX64-EMPTY:28; CHECK-PTX64-NEXT:  // %bb.0:29; CHECK-PTX64-NEXT:    ld.param.b64 %rd1, [st_bulk_shared_cta_param_0];30; CHECK-PTX64-NEXT:    ld.param.b64 %rd2, [st_bulk_shared_cta_param_1];31; CHECK-PTX64-NEXT:    st.bulk.shared::cta [%rd1], %rd2, 0;32; CHECK-PTX64-NEXT:    ret;33;34; CHECK-PTX-SHARED32-LABEL: st_bulk_shared_cta(35; CHECK-PTX-SHARED32:       {36; CHECK-PTX-SHARED32-NEXT:    .reg .b32 %r<2>;37; CHECK-PTX-SHARED32-NEXT:    .reg .b64 %rd<2>;38; CHECK-PTX-SHARED32-EMPTY:39; CHECK-PTX-SHARED32-NEXT:  // %bb.0:40; CHECK-PTX-SHARED32-NEXT:    ld.param.b32 %r1, [st_bulk_shared_cta_param_0];41; CHECK-PTX-SHARED32-NEXT:    ld.param.b64 %rd1, [st_bulk_shared_cta_param_1];42; CHECK-PTX-SHARED32-NEXT:    st.bulk.shared::cta [%r1], %rd1, 0;43; CHECK-PTX-SHARED32-NEXT:    ret;44   call void @llvm.nvvm.st.bulk.shared.cta(ptr addrspace(3) %dest_addr, i64 %size, i64 0)45   ret void46}47