diff options
Diffstat (limited to 'flang/test/Lower/CUDA/cuda-device-proc.cuf')
| -rw-r--r-- | flang/test/Lower/CUDA/cuda-device-proc.cuf | 12 |
1 files changed, 11 insertions, 1 deletions
diff --git a/flang/test/Lower/CUDA/cuda-device-proc.cuf b/flang/test/Lower/CUDA/cuda-device-proc.cuf index 83ee011..29c348c 100644 --- a/flang/test/Lower/CUDA/cuda-device-proc.cuf +++ b/flang/test/Lower/CUDA/cuda-device-proc.cuf @@ -438,7 +438,7 @@ end subroutine ! CHECK: nvvm.cp.async.bulk.commit.group ! CHECK: nvvm.cp.async.bulk.wait_group 0 -attributes(global) subroutine test_bulk_g2s(c, a, b, n) +attributes(global) subroutine test_bulk_g2s(a) real(8), device :: a(*) real(8), shared :: tmpa(1024) integer(8), shared :: barrier1 @@ -448,3 +448,13 @@ end subroutine ! CHECK-LABEL: func.func @_QPtest_bulk_g2s ! CHECK: nvvm.cp.async.bulk.shared.cluster.global %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : <7>, <1> + +attributes(global) subroutine test_bulk_s2g(a) + real(8), device :: a(*) + real(8), shared :: tmpa(1024) + integer(4) :: tx_count + call tma_bulk_s2g(tmpa, a(j), tx_count) +end subroutine + +! CHECK-LABEL: func.func @_QPtest_bulk_s2g +! CHECL: nvvm.cp.async.bulk.global.shared.cta %{{.*}}, %{{.*}}, %{{.*}} : <1>, <3> |