1// RUN: mlir-opt -allow-unregistered-dialect %s | FileCheck %s
2// Verify the printed output can be parsed.
3// RUN: mlir-opt -allow-unregistered-dialect %s | mlir-opt -allow-unregistered-dialect | FileCheck %s
4// Verify the generic form can be parsed.
5// RUN: mlir-opt -allow-unregistered-dialect -mlir-print-op-generic %s | mlir-opt -allow-unregistered-dialect | FileCheck %s
6
7// CHECK-LABEL: func @raw_buffer_load_f32_from_rank_1
8func.func @raw_buffer_load_f32_from_rank_1(%src : memref<128xf32>, %offset : i32, %idx0 : i32) -> f32 {
9  // CHECK: amdgpu.raw_buffer_load {boundsCheck = true, indexOffset = 1 : i32} %{{.*}}[{{.*}}] sgprOffset %{{.*}} : memref<128xf32>, i32 -> f32
10  %0 = amdgpu.raw_buffer_load {boundsCheck = true, indexOffset = 1 : i32} %src[%idx0] sgprOffset %offset : memref<128xf32>, i32 -> f32
11  func.return %0 : f32
12}
13
14// CHECK-LABEL: func @raw_buffer_load_f32_from_rank_4
15func.func @raw_buffer_load_f32_from_rank_4(%src : memref<128x64x32x16xf32>, %offset : i32, %idx0 : i32, %idx1 : i32, %idx2 : i32, %idx3 : i32) -> f32 {
16  // CHECK: amdgpu.raw_buffer_load {boundsCheck = true, indexOffset = 1 : i32} %{{.*}}[%{{.*}}, %{{.*}}, %{{.*}}] sgprOffset %{{.*}} : memref<128x64x32x16xf32>, i32, i32, i32, i32 -> f32
17  %0 = amdgpu.raw_buffer_load {boundsCheck = true, indexOffset = 1 : i32} %src[%idx0, %idx1, %idx2, %idx3] sgprOffset %offset : memref<128x64x32x16xf32>, i32, i32, i32, i32 -> f32
18  func.return %0 : f32
19}
20
21// CHECK-LABEL: func @raw_buffer_load_4xf32_from_rank_4
22func.func @raw_buffer_load_4xf32_from_rank_4(%src : memref<128x64x32x16xf32>, %offset : i32, %idx0 : i32, %idx1 : i32, %idx2 : i32, %idx3 : i32) -> vector<4xf32> {
23  // CHECK: amdgpu.raw_buffer_load {boundsCheck = true, indexOffset = 1 : i32} %{{.*}}[%{{.*}}, %{{.*}}, %{{.*}}] sgprOffset %{{.*}} : memref<128x64x32x16xf32>, i32, i32, i32, i32 -> vector<4xf32>
24  %0 = amdgpu.raw_buffer_load {boundsCheck = true, indexOffset = 1 : i32} %src[%idx0, %idx1, %idx2, %idx3] sgprOffset %offset : memref<128x64x32x16xf32>, i32, i32, i32, i32 -> vector<4xf32>
25  func.return %0 : vector<4xf32>
26}
27
28// CHECK-LABEL: func @raw_buffer_store_f32_to_rank_1
29func.func @raw_buffer_store_f32_to_rank_1(%value : f32, %dst : memref<128xf32>, %offset : i32, %idx0 : i32) {
30  // CHECK: amdgpu.raw_buffer_store {boundsCheck = true, indexOffset = 1 : i32} %{{.*}} -> %{{.*}}[{{.*}}] sgprOffset %{{.*}} : f32 -> memref<128xf32>, i32
31  amdgpu.raw_buffer_store {boundsCheck = true, indexOffset = 1 : i32} %value -> %dst[%idx0] sgprOffset %offset : f32 -> memref<128xf32>, i32
32  func.return
33}
34
35// CHECK-LABEL: func @raw_buffer_store_f32_to_rank_4
36func.func @raw_buffer_store_f32_to_rank_4(%value : f32, %dst : memref<128x64x32x16xf32>, %offset : i32, %idx0 : i32, %idx1 : i32, %idx2 : i32, %idx3 : i32) {
37  // CHECK: amdgpu.raw_buffer_store {boundsCheck = true, indexOffset = 1 : i32} %{{.*}} -> %{{.*}}[%{{.*}}, %{{.*}}, %{{.*}}] sgprOffset %{{.*}} : f32 -> memref<128x64x32x16xf32>, i32, i32, i32, i32
38  amdgpu.raw_buffer_store {boundsCheck = true, indexOffset = 1 : i32} %value -> %dst[%idx0, %idx1, %idx2, %idx3] sgprOffset %offset : f32 -> memref<128x64x32x16xf32>, i32, i32, i32, i32
39  func.return
40}
41
42// CHECK-LABEL: func @raw_buffer_store_4xf32_to_rank_4
43func.func @raw_buffer_store_4xf32_to_rank_4(%value : vector<4xf32>, %dst : memref<128x64x32x16xf32>, %offset : i32, %idx0 : i32, %idx1 : i32, %idx2 : i32, %idx3 : i32) {
44  // CHECK: amdgpu.raw_buffer_store {boundsCheck = true, indexOffset = 1 : i32} %{{.*}} -> %{{.*}}[%{{.*}}, %{{.*}}, %{{.*}}] sgprOffset %{{.*}} : vector<4xf32> -> memref<128x64x32x16xf32>, i32, i32, i32, i32
45  amdgpu.raw_buffer_store {boundsCheck = true, indexOffset = 1 : i32} %value -> %dst[%idx0, %idx1, %idx2, %idx3] sgprOffset %offset : vector<4xf32> -> memref<128x64x32x16xf32>, i32, i32, i32, i32
46  func.return
47}
48
49// CHECK-LABEL: func @raw_buffer_atomic_fadd_f32_to_rank_1
50func.func @raw_buffer_atomic_fadd_f32_to_rank_1(%value : f32, %dst : memref<128xf32>, %offset : i32, %idx0 : i32) {
51  // CHECK: amdgpu.raw_buffer_atomic_fadd {boundsCheck = true, indexOffset = 1 : i32} %{{.*}} -> %{{.*}}[{{.*}}] sgprOffset %{{.*}} : f32 -> memref<128xf32>, i32
52  amdgpu.raw_buffer_atomic_fadd {boundsCheck = true, indexOffset = 1 : i32} %value -> %dst[%idx0] sgprOffset %offset : f32 -> memref<128xf32>, i32
53  func.return
54}
55
56// CHECK-LABEL: func @raw_buffer_atomic_fadd_f32_to_rank_4
57func.func @raw_buffer_atomic_fadd_f32_to_rank_4(%value : f32, %dst : memref<128x64x32x16xf32>, %offset : i32, %idx0 : i32, %idx1 : i32, %idx2 : i32, %idx3 : i32) {
58  // CHECK: amdgpu.raw_buffer_atomic_fadd {boundsCheck = true, indexOffset = 1 : i32} %{{.*}} -> %{{.*}}[%{{.*}}, %{{.*}}, %{{.*}}] sgprOffset %{{.*}} : f32 -> memref<128x64x32x16xf32>, i32, i32, i32, i32
59  amdgpu.raw_buffer_atomic_fadd {boundsCheck = true, indexOffset = 1 : i32} %value -> %dst[%idx0, %idx1, %idx2, %idx3] sgprOffset %offset : f32 -> memref<128x64x32x16xf32>, i32, i32, i32, i32
60  func.return
61}
62
63// CHECK-LABEL: func @lds_barrier
64func.func @lds_barrier() {
65  // CHECK: amdgpu.lds_barrier
66  amdgpu.lds_barrier
67  func.return
68}
69