Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
Showing
1 changed file
with
81 additions
and
0 deletions.
There are no files selected for viewing
81 changes: 81 additions & 0 deletions
81
mlir/test/mlir-miopen-driver/conv2d_harness_cyxk_chwn_khwn.mlir
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,81 @@ | ||
| // RUN: mlir-miopen-driver -p --host %s | FileCheck %s --check-prefix=HARNESS | ||
| // RUN: mlir-miopen-driver -pc --host %s | FileCheck %s --check-prefix=LOWERING | ||
| // RUN: mlir-miopen-driver -p -fil_layout=cyxk -in_layout=chwn -out_layout=khwn -c --host %s | mlir-rocm-runner --shared-libs=%rocm_wrapper_library_dir/librocm-runtime-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext --entry-point-result=void | FileCheck %s --check-prefix=E2E | ||
|
|
||
| func @conv2d(%filter : memref<8x3x3x128xf32>, %input : memref<8x32x32x128xf32>, %output : memref<128x30x30x128xf32>) { | ||
| // Convolution host-side logic would be populated here. | ||
| return | ||
| } | ||
| // HARNESS: module | ||
| // HARNESS: func @conv2d([[FILTER_MEMREF:%.*]]: memref<8x3x3x128xf32>, [[INPUT_MEMREF:%.*]]: memref<8x32x32x128xf32>, [[OUTPUT_MEMREF:%.*]]: memref<128x30x30x128xf32>) | ||
| // LOWERING: module | ||
| // LOWERING: func @conv2d([[FILTER_MEMREF:%.*]]: memref<8x3x3x128xf32>, [[INPUT_MEMREF:%.*]]: memref<8x32x32x128xf32>, [[OUTPUT_MEMREF:%.*]]: memref<128x30x30x128xf32>) | ||
| // LOWERING: "gpu.launch_func"(%{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, [[FILTER_MEMREF]], [[INPUT_MEMREF]], [[OUTPUT_MEMREF]]) {kernel = @miopen_kernel_module::@miopen_conv2d_kcyx_nchw_nkhw} : (index, index, index, index, index, index, memref<8x3x3x128xf32>, memref<8x32x32x128xf32>, memref<128x30x30x128xf32>) -> () | ||
|
|
||
| func @main() { | ||
| // allocate CPU memory. | ||
| %0 = alloc() : memref<8x3x3x128xf32> | ||
| %1 = alloc() : memref<8x32x32x128xf32> | ||
| %2 = alloc() : memref<128x30x30x128xf32> | ||
|
|
||
| %3 = memref_cast %0 : memref<8x3x3x128xf32> to memref<?x?x?x?xf32> | ||
| %4 = memref_cast %1 : memref<8x32x32x128xf32> to memref<?x?x?x?xf32> | ||
| %5 = memref_cast %2 : memref<128x30x30x128xf32> to memref<?x?x?x?xf32> | ||
|
|
||
| // populate initial values. | ||
| %cst = constant 1.0 : f32 | ||
| %cst0 = constant 0.0 : f32 | ||
| call @mcpuMemset4DFloat(%3, %cst) : (memref<?x?x?x?xf32>, f32) -> () | ||
| call @mcpuMemset4DFloat(%4, %cst) : (memref<?x?x?x?xf32>, f32) -> () | ||
| call @mcpuMemset4DFloat(%5, %cst0) : (memref<?x?x?x?xf32>, f32) -> () | ||
|
|
||
| // allocate GPU memory. | ||
| %6 = call @mgpuMemAlloc4DFloat(%3) : (memref<?x?x?x?xf32>) -> (memref<?x?x?x?xf32>) | ||
| %7 = call @mgpuMemAlloc4DFloat(%4) : (memref<?x?x?x?xf32>) -> (memref<?x?x?x?xf32>) | ||
| %8 = call @mgpuMemAlloc4DFloat(%5) : (memref<?x?x?x?xf32>) -> (memref<?x?x?x?xf32>) | ||
|
|
||
| // copy direction constants. | ||
| %cst_h2d = constant 1 : i32 | ||
| %cst_d2h = constant 2 : i32 | ||
|
|
||
| // transfer data CPU -> GPU. | ||
| call @mgpuMemCopy4DFloat(%3, %6, %cst_h2d) : (memref<?x?x?x?xf32>, memref<?x?x?x?xf32>, i32) -> () | ||
| call @mgpuMemCopy4DFloat(%4, %7, %cst_h2d) : (memref<?x?x?x?xf32>, memref<?x?x?x?xf32>, i32) -> () | ||
| call @mgpuMemCopy4DFloat(%5, %8, %cst_h2d) : (memref<?x?x?x?xf32>, memref<?x?x?x?xf32>, i32) -> () | ||
|
|
||
| // launch kernel. | ||
| %filter = memref_cast %6 : memref<?x?x?x?xf32> to memref<8x3x3x128xf32> | ||
| %input = memref_cast %7 : memref<?x?x?x?xf32> to memref<8x32x32x128xf32> | ||
| %output = memref_cast %8 : memref<?x?x?x?xf32> to memref<128x30x30x128xf32> | ||
| call @conv2d(%filter, %input, %output) : (memref<8x3x3x128xf32>, memref<8x32x32x128xf32>, memref<128x30x30x128xf32>) -> () | ||
|
|
||
| // transfer data GPU -> CPU. | ||
| call @mgpuMemCopy4DFloat(%8, %5, %cst_d2h) : (memref<?x?x?x?xf32>, memref<?x?x?x?xf32>, i32) -> () | ||
|
|
||
| // verify result. | ||
| // TBD. Add more verifying logic. | ||
| %9 = memref_cast %5 : memref<?x?x?x?xf32> to memref<*xf32> | ||
| call @print_memref_f32(%9) : (memref<*xf32>) -> () | ||
|
|
||
| // dellocate GPU memory. | ||
| call @mgpuMemDealloc4DFloat(%6) : (memref<?x?x?x?xf32>) -> () | ||
| call @mgpuMemDealloc4DFloat(%7) : (memref<?x?x?x?xf32>) -> () | ||
| call @mgpuMemDealloc4DFloat(%8) : (memref<?x?x?x?xf32>) -> () | ||
|
|
||
| // deallocate CPU memory. | ||
| dealloc %0 : memref<8x3x3x128xf32> | ||
| dealloc %1 : memref<8x32x32x128xf32> | ||
| dealloc %2 : memref<128x30x30x128xf32> | ||
|
|
||
| return | ||
| } | ||
|
|
||
| func @mcpuMemset4DFloat(%ptr : memref<?x?x?x?xf32>, %value: f32) -> () | ||
| func @mgpuMemAlloc4DFloat(%ptr : memref<?x?x?x?xf32>) -> (memref<?x?x?x?xf32>) | ||
| func @mgpuMemDealloc4DFloat(%ptr : memref<?x?x?x?xf32>) -> () | ||
| func @mgpuMemCopy4DFloat(%src : memref<?x?x?x?xf32>, %dst : memref<?x?x?x?xf32>, %dir : i32) -> () | ||
| func @print_memref_f32(%ptr : memref<*xf32>) | ||
| // LOWERING: gpu.module @miopen_kernel_module | ||
| // LOWERING: gpu.func @miopen_conv2d_kcyx_nchw_nkhw | ||
| // TBD. Add more verifying logic. | ||
| // E2E: Unranked Memref base@ = 0x{{.*}} rank = 4 offset = 0 sizes = [128, 30, 30, 128] strides = [115200, 3840, 128, 1] data = |