1 // RUN: mlir-vulkan-runner %s \
2 // RUN: --shared-libs=%vulkan-runtime-wrappers,%mlir_runner_utils \
3 // RUN: --entry-point-result=void | FileCheck %s
8 spirv.target_env = #spirv.target_env<
9 #spirv.vce<v1.0, [Shader], [SPV_KHR_storage_buffer_storage_class]>, #spirv.resource_limits<>>
12 gpu.func @kernel_vector_shuffle(%arg0 : memref<2xi32>, %arg1 : memref<2xi32>, %arg2 : memref<4xi32>)
13 kernel attributes { spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [1, 1, 1]>} {
14 %idx0 = arith.constant 0 : index
15 %idx1 = arith.constant 1 : index
16 %idx2 = arith.constant 2 : index
17 %idx3 = arith.constant 3 : index
18 %idx4 = arith.constant 4 : index
20 %lhs = arith.constant dense<[0, 0]> : vector<2xi32>
21 %rhs = arith.constant dense<[0, 0]> : vector<2xi32>
23 %val0 = memref.load %arg0[%idx0] : memref<2xi32>
24 %val1 = memref.load %arg0[%idx1] : memref<2xi32>
25 %val2 = memref.load %arg1[%idx0] : memref<2xi32>
26 %val3 = memref.load %arg1[%idx1] : memref<2xi32>
28 %lhs0 = vector.insertelement %val0, %lhs[%idx0 : index] : vector<2xi32>
29 %lhs1 = vector.insertelement %val1, %lhs0[%idx1 : index] : vector<2xi32>
30 %rhs0 = vector.insertelement %val2, %rhs[%idx0 : index] : vector<2xi32>
31 %rhs1 = vector.insertelement %val3, %rhs0[%idx1 : index] : vector<2xi32>
33 %shuffle = vector.shuffle %lhs1, %rhs1[2, 1, 3, 3] : vector<2xi32>, vector<2xi32>
35 %res0 = vector.extractelement %shuffle[%idx0 : index] : vector<4xi32>
36 %res1 = vector.extractelement %shuffle[%idx1 : index] : vector<4xi32>
37 %res2 = vector.extractelement %shuffle[%idx2 : index] : vector<4xi32>
38 %res3 = vector.extractelement %shuffle[%idx3 : index] : vector<4xi32>
40 memref.store %res0, %arg2[%idx0]: memref<4xi32>
41 memref.store %res1, %arg2[%idx1]: memref<4xi32>
42 memref.store %res2, %arg2[%idx2]: memref<4xi32>
43 memref.store %res3, %arg2[%idx3]: memref<4xi32>
50 // Allocate 3 buffers.
51 %buf0 = memref.alloc() : memref<2xi32>
52 %buf1 = memref.alloc() : memref<2xi32>
53 %buf2 = memref.alloc() : memref<4xi32>
55 %idx0 = arith.constant 0 : index
56 %idx1 = arith.constant 1 : index
57 %idx4 = arith.constant 4 : index
59 // Initialize input buffer.
60 %buf0_vals = arith.constant dense<[0, 1]> : vector<2xi32>
61 %buf1_vals = arith.constant dense<[2, 3]> : vector<2xi32>
62 vector.store %buf0_vals, %buf0[%idx0] : memref<2xi32>, vector<2xi32>
63 vector.store %buf1_vals, %buf1[%idx0] : memref<2xi32>, vector<2xi32>
65 // Initialize output buffer.
66 %value0 = arith.constant 0 : i32
67 %buf3 = memref.cast %buf2 : memref<4xi32> to memref<?xi32>
68 call @fillResource1DInt(%buf3, %value0) : (memref<?xi32>, i32) -> ()
70 gpu.launch_func @kernels::@kernel_vector_shuffle
71 blocks in (%idx4, %idx1, %idx1) threads in (%idx1, %idx1, %idx1)
72 args(%buf0 : memref<2xi32>, %buf1 : memref<2xi32>, %buf2 : memref<4xi32>)
73 %buf4 = memref.cast %buf3 : memref<?xi32> to memref<*xi32>
74 call @printMemrefI32(%buf4) : (memref<*xi32>) -> ()
77 func.func private @fillResource1DInt(%0 : memref<?xi32>, %1 : i32)
78 func.func private @printMemrefI32(%ptr : memref<*xi32>)