71 lines · plain
1// RUN: %clang_cc1 "-triple" "nvptx64-nvidia-cuda" "-target-feature" "+ptx80" "-target-cpu" "sm_90" -emit-llvm -fcuda-is-device -o - %s | FileCheck %s2 3// CHECK: define{{.*}} void @_Z6kernelPlPvj(4__attribute__((global)) void kernel(long *out, void *ptr, unsigned u) {5 int i = 0;6 // CHECK: call i1 @llvm.nvvm.isspacep.shared.cluster7 out[i++] = __nvvm_isspacep_shared_cluster(ptr);8 9 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.clusterid.x()10 out[i++] = __nvvm_read_ptx_sreg_clusterid_x();11 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.clusterid.y()12 out[i++] = __nvvm_read_ptx_sreg_clusterid_y();13 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.clusterid.z()14 out[i++] = __nvvm_read_ptx_sreg_clusterid_z();15 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.clusterid.w()16 out[i++] = __nvvm_read_ptx_sreg_clusterid_w();17 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.nclusterid.x()18 out[i++] = __nvvm_read_ptx_sreg_nclusterid_x();19 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.nclusterid.y()20 out[i++] = __nvvm_read_ptx_sreg_nclusterid_y();21 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.nclusterid.z()22 out[i++] = __nvvm_read_ptx_sreg_nclusterid_z();23 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.nclusterid.w()24 out[i++] = __nvvm_read_ptx_sreg_nclusterid_w();25 26 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.x()27 out[i++] = __nvvm_read_ptx_sreg_cluster_ctaid_x();28 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.y()29 out[i++] = __nvvm_read_ptx_sreg_cluster_ctaid_y();30 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.z()31 out[i++] = __nvvm_read_ptx_sreg_cluster_ctaid_z();32 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctaid.w()33 out[i++] = __nvvm_read_ptx_sreg_cluster_ctaid_w();34 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.x()35 out[i++] = __nvvm_read_ptx_sreg_cluster_nctaid_x();36 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.y()37 out[i++] = __nvvm_read_ptx_sreg_cluster_nctaid_y();38 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.z()39 out[i++] = __nvvm_read_ptx_sreg_cluster_nctaid_z();40 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctaid.w()41 out[i++] = __nvvm_read_ptx_sreg_cluster_nctaid_w();42 43 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.cluster.ctarank()44 out[i++] = __nvvm_read_ptx_sreg_cluster_ctarank();45 // CHECK: call i32 @llvm.nvvm.read.ptx.sreg.cluster.nctarank()46 out[i++] = __nvvm_read_ptx_sreg_cluster_nctarank();47 // CHECK: call i1 @llvm.nvvm.is_explicit_cluster()48 out[i++] = __nvvm_is_explicit_cluster();49 50 auto * sptr = (__attribute__((address_space(3))) void *)ptr;51 // CHECK: call ptr @llvm.nvvm.mapa(ptr %{{.*}}, i32 %{{.*}})52 out[i++] = (long) __nvvm_mapa(ptr, u);53 // CHECK: call ptr addrspace(7) @llvm.nvvm.mapa.shared.cluster(ptr addrspace(3) %{{.*}}, i32 %{{.*}})54 out[i++] = (long) __nvvm_mapa_shared_cluster(sptr, u);55 // CHECK: call i32 @llvm.nvvm.getctarank(ptr {{.*}})56 out[i++] = __nvvm_getctarank(ptr);57 // CHECK: call i32 @llvm.nvvm.getctarank.shared.cluster(ptr addrspace(3) {{.*}})58 out[i++] = __nvvm_getctarank_shared_cluster(sptr);59 60 // CHECK: call void @llvm.nvvm.barrier.cluster.arrive()61 __nvvm_barrier_cluster_arrive();62 // CHECK: call void @llvm.nvvm.barrier.cluster.arrive.relaxed()63 __nvvm_barrier_cluster_arrive_relaxed();64 // CHECK: call void @llvm.nvvm.barrier.cluster.wait()65 __nvvm_barrier_cluster_wait();66 // CHECK: call void @llvm.nvvm.fence.sc.cluster()67 __nvvm_fence_sc_cluster();68 69 // CHECK: ret void70}71