44 lines · plain
1// RUN: mlir-opt %s --pass-pipeline="builtin.module(gpu.module(convert-to-llvm{dynamic=true}))" | FileCheck %s2// RUN: mlir-opt %s --pass-pipeline="builtin.module(gpu.module(convert-to-llvm{dynamic=true allow-pattern-rollback=0}))" | FileCheck %s3 4// CHECK-LABEL: gpu.module @nvvm_module5gpu.module @nvvm_module [#nvvm.target] {6 // CHECK-LABEL: llvm.func @kernel_0()7 func.func @kernel_0() -> index {8 // CHECK: = nvvm.read.ptx.sreg.tid.x : i329 // CHECK: = llvm.sext %{{.*}} : i32 to i6410 %tIdX = gpu.thread_id x11 // CHECK: = nvvm.read.ptx.sreg.laneid range <i32, 0, 32> : i3212 // CHECK: = llvm.sext %{{.*}} : i32 to i6413 %laneId = gpu.lane_id14 %sum = index.add %tIdX, %laneId15 func.return %sum : index16 }17 18// CHECK-LABEL: llvm.func @kernel_119// CHECK: (%{{.*}}: !llvm.ptr<1>, %arg1: !llvm.ptr<1>, %arg2: i64)20// CHECK: attributes {gpu.kernel, gpu.known_block_size = array<i32: 128, 1, 1>, nvvm.kernel, nvvm.maxntid = array<i32: 128, 1, 1>} 21 gpu.func @kernel_1(%arg0 : memref<f32, #gpu.address_space<global>>) kernel attributes {known_block_size = array<i32: 128, 1, 1>} {22 gpu.return23 }24}25 26// CHECK-LABEL: gpu.module @nvvm_module_227gpu.module @nvvm_module_2 {28 // CHECK-LABEL: llvm.func @kernel_0()29 func.func @kernel_0() -> index {30 // CHECK: = gpu.thread_id x31 %tIdX = gpu.thread_id x32 // CHECK: = gpu.lane_id33 %laneId = gpu.lane_id34 %sum = index.add %tIdX, %laneId35 func.return %sum : index36 }37 38// CHECK-LABEL: gpu.func @kernel_139// CHECK: (%{{.*}}: memref<f32, #gpu.address_space<global>>) kernel attributes {known_block_size = array<i32: 128, 1, 1>}40 gpu.func @kernel_1(%arg0 : memref<f32, #gpu.address_space<global>>) kernel attributes {known_block_size = array<i32: 128, 1, 1>} {41 gpu.return42 }43}44