71 lines · plain
1// RUN: mlir-opt %s -test-spirv-cpu-runner-pipeline \2// RUN: | mlir-runner - -e main --entry-point-result=void --shared-libs=%mlir_runner_utils,%mlir_spirv_cpu_runtime --link-nested-modules \3// RUN: | FileCheck %s4 5// CHECK: [8, 8, 8, 8, 8, 8]6module attributes {7 gpu.container_module,8 spirv.target_env = #spirv.target_env<9 #spirv.vce<v1.0, [Shader], [SPV_KHR_variable_pointers]>,10 #spirv.resource_limits<11 max_compute_workgroup_invocations = 128,12 max_compute_workgroup_size = [128, 128, 64]>>13} {14 gpu.module @kernels {15 gpu.func @double(%arg0 : memref<6xi32>, %arg1 : memref<6xi32>)16 kernel attributes { spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [1, 1, 1]>} {17 %factor = arith.constant 2 : i3218 19 %i0 = arith.constant 0 : index20 %i1 = arith.constant 1 : index21 %i2 = arith.constant 2 : index22 %i3 = arith.constant 3 : index23 %i4 = arith.constant 4 : index24 %i5 = arith.constant 5 : index25 26 %x0 = memref.load %arg0[%i0] : memref<6xi32>27 %x1 = memref.load %arg0[%i1] : memref<6xi32>28 %x2 = memref.load %arg0[%i2] : memref<6xi32>29 %x3 = memref.load %arg0[%i3] : memref<6xi32>30 %x4 = memref.load %arg0[%i4] : memref<6xi32>31 %x5 = memref.load %arg0[%i5] : memref<6xi32>32 33 %y0 = arith.muli %x0, %factor : i3234 %y1 = arith.muli %x1, %factor : i3235 %y2 = arith.muli %x2, %factor : i3236 %y3 = arith.muli %x3, %factor : i3237 %y4 = arith.muli %x4, %factor : i3238 %y5 = arith.muli %x5, %factor : i3239 40 memref.store %y0, %arg1[%i0] : memref<6xi32>41 memref.store %y1, %arg1[%i1] : memref<6xi32>42 memref.store %y2, %arg1[%i2] : memref<6xi32>43 memref.store %y3, %arg1[%i3] : memref<6xi32>44 memref.store %y4, %arg1[%i4] : memref<6xi32>45 memref.store %y5, %arg1[%i5] : memref<6xi32>46 gpu.return47 }48 }49 func.func @main() {50 %input = memref.alloc() : memref<6xi32>51 %output = memref.alloc() : memref<6xi32>52 %four = arith.constant 4 : i3253 %zero = arith.constant 0 : i3254 %input_casted = memref.cast %input : memref<6xi32> to memref<?xi32>55 %output_casted = memref.cast %output : memref<6xi32> to memref<?xi32>56 call @fillI32Buffer(%input_casted, %four) : (memref<?xi32>, i32) -> ()57 call @fillI32Buffer(%output_casted, %zero) : (memref<?xi32>, i32) -> ()58 59 %one = arith.constant 1 : index60 gpu.launch_func @kernels::@double61 blocks in (%one, %one, %one) threads in (%one, %one, %one)62 args(%input : memref<6xi32>, %output : memref<6xi32>)63 %result = memref.cast %output : memref<6xi32> to memref<*xi32>64 call @printMemrefI32(%result) : (memref<*xi32>) -> ()65 return66 }67 68 func.func private @fillI32Buffer(%arg0 : memref<?xi32>, %arg1 : i32)69 func.func private @printMemrefI32(%ptr : memref<*xi32>)70}71