Skip to content

Add functionality test suite for ov op translation #401

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
wants to merge 4 commits into
base: main
Choose a base branch
from
Draft

Add functionality test suite for ov op translation #401

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
041557b
add functionality test
zhczhong Oct 30, 2024
48090c4
Merge branch 'main' of https://github.com/intel/graph-compiler into z…
zhczhong Nov 11, 2024
0cb9c87
update
zhczhong Nov 11, 2024
8c2592b
Merge branch 'main' into zhicong/add_functionality_test
kurapov-peter Nov 26, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
57 changes: 57 additions & 0 deletions test/mlir/test/gc/Integration/op/binary.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,57 @@
// RUN: gc-opt %s --gc-gpu-pipeline -split-input-file | FileCheck %s

// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>} {
func.func @multiply(%arg0: memref<1024x1024xf16>, %arg1: memref<1024x1024xf16>, %arg2: memref<1024x1024xf16>, %arg3: memref<1024x1024xf16>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<1024x1024xf16>
%1 = bufferization.to_tensor %arg1 restrict : memref<1024x1024xf16>
%2 = bufferization.to_tensor %arg2 restrict : memref<1024x1024xf16>
%3 = tensor.empty() : tensor<1024x1024xf16>
%4 = linalg.mul ins(%0, %1 : tensor<1024x1024xf16>, tensor<1024x1024xf16>) outs(%3 : tensor<1024x1024xf16>) -> tensor<1024x1024xf16>
bufferization.materialize_in_destination %4 in restrict writable %arg3 : (tensor<1024x1024xf16>, memref<1024x1024xf16>) -> ()
return
}
}

// -----

// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>} {
func.func @add(%arg0: memref<1024x1024xf16>, %arg1: memref<1024x1024xf16>, %arg2: memref<1024x1024xf16>, %arg3: memref<1024x1024xf16>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<1024x1024xf16>
%1 = bufferization.to_tensor %arg1 restrict : memref<1024x1024xf16>
%2 = bufferization.to_tensor %arg2 restrict : memref<1024x1024xf16>
%3 = tensor.empty() : tensor<1024x1024xf16>
%4 = linalg.add ins(%0, %1 : tensor<1024x1024xf16>, tensor<1024x1024xf16>) outs(%3 : tensor<1024x1024xf16>) -> tensor<1024x1024xf16>
bufferization.materialize_in_destination %4 in restrict writable %arg3 : (tensor<1024x1024xf16>, memref<1024x1024xf16>) -> ()
return
}
}

// -----
// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>} {
func.func @subtract(%arg0: memref<1024x1024xf16>, %arg1: memref<1024x1024xf16>, %arg2: memref<1024x1024xf16>, %arg3: memref<1024x1024xf16>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<1024x1024xf16>
%1 = bufferization.to_tensor %arg1 restrict : memref<1024x1024xf16>
%2 = bufferization.to_tensor %arg2 restrict : memref<1024x1024xf16>
%3 = tensor.empty() : tensor<1024x1024xf16>
%4 = linalg.sub ins(%0, %1 : tensor<1024x1024xf16>, tensor<1024x1024xf16>) outs(%3 : tensor<1024x1024xf16>) -> tensor<1024x1024xf16>
bufferization.materialize_in_destination %4 in restrict writable %arg3 : (tensor<1024x1024xf16>, memref<1024x1024xf16>) -> ()
return
}
}

// -----
// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>} {
func.func @divide(%arg0: memref<1024x1024xf16>, %arg1: memref<1024x1024xf16>, %arg2: memref<1024x1024xf16>, %arg3: memref<1024x1024xf16>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<1024x1024xf16>
%1 = bufferization.to_tensor %arg1 restrict : memref<1024x1024xf16>
%2 = bufferization.to_tensor %arg2 restrict : memref<1024x1024xf16>
%3 = tensor.empty() : tensor<1024x1024xf16>
%4 = linalg.div ins(%0, %1 : tensor<1024x1024xf16>, tensor<1024x1024xf16>) outs(%3 : tensor<1024x1024xf16>) -> tensor<1024x1024xf16>
bufferization.materialize_in_destination %4 in restrict writable %arg3 : (tensor<1024x1024xf16>, memref<1024x1024xf16>) -> ()
return
}
}
79 changes: 79 additions & 0 deletions test/mlir/test/gc/Integration/op/matmul.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,79 @@
// RUN: gc-opt %s --gc-gpu-pipeline -split-input-file | FileCheck %s

// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>} {
func.func @matmul_f16(%arg0: memref<4096x4096xf16>, %arg1: memref<4096x4096xf16>, %arg2: memref<4096x4096xf16>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<4096x4096xf16>
%1 = bufferization.to_tensor %arg1 restrict : memref<4096x4096xf16>
%2 = tensor.empty() : tensor<4096x4096xf16>
%cst = arith.constant 0.000000e+00 : f16
%3 = linalg.fill ins(%cst : f16) outs(%2 : tensor<4096x4096xf16>) -> tensor<4096x4096xf16>
%4 = linalg.matmul_transpose_b ins(%0, %1 : tensor<4096x4096xf16>, tensor<4096x4096xf16>) outs(%3 : tensor<4096x4096xf16>) -> tensor<4096x4096xf16>
bufferization.materialize_in_destination %4 in restrict writable %arg2 : (tensor<4096x4096xf16>, memref<4096x4096xf16>) -> ()
return
}
}

// -----
// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>} {
func.func @corner_shape_matmul_f16(%arg0: memref<521x521xf16>, %arg1: memref<521x521xf16>, %arg2: memref<521x521xf16>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<521x521xf16>
%1 = bufferization.to_tensor %arg1 restrict : memref<521x521xf16>
%2 = tensor.empty() : tensor<521x521xf16>
%cst = arith.constant 0.000000e+00 : f16
%3 = linalg.fill ins(%cst : f16) outs(%2 : tensor<521x521xf16>) -> tensor<521x521xf16>
%4 = linalg.matmul_transpose_b ins(%0, %1 : tensor<521x521xf16>, tensor<521x521xf16>) outs(%3 : tensor<521x521xf16>) -> tensor<521x521xf16>
bufferization.materialize_in_destination %4 in restrict writable %arg2 : (tensor<521x521xf16>, memref<521x521xf16>) -> ()
return
}
}

// -----
// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>}{
func.func @dynamic_matmul_f16(%arg0: memref<?x?xf16>, %arg1: memref<1024x1024xf16>, %arg2: memref<?x1024xf16>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<?x?xf16>
%c0 = arith.constant 0 : index
%dim = tensor.dim %0, %c0 : tensor<?x?xf16>
%c1 = arith.constant 1 : index
%dim_0 = tensor.dim %0, %c1 : tensor<?x?xf16>
%1 = bufferization.to_tensor %arg1 restrict : memref<1024x1024xf16>
%2 = tensor.empty(%dim) : tensor<?x1024xf16>
%cst = arith.constant 0.000000e+00 : f16
%3 = linalg.fill ins(%cst : f16) outs(%2 : tensor<?x1024xf16>) -> tensor<?x1024xf16>
%4 = linalg.matmul_transpose_b ins(%0, %1 : tensor<?x?xf16>, tensor<1024x1024xf16>) outs(%3 : tensor<?x1024xf16>) -> tensor<?x1024xf16>
bufferization.materialize_in_destination %4 in restrict writable %arg2 : (tensor<?x1024xf16>, memref<?x1024xf16>) -> ()
return
}
}

// -----
// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>} {
func.func @matmul_bf16(%arg0: memref<4096x4096xbf16>, %arg1: memref<4096x4096xbf16>, %arg2: memref<4096x4096xbf16>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<4096x4096xbf16>
%1 = bufferization.to_tensor %arg1 restrict : memref<4096x4096xbf16>
%2 = tensor.empty() : tensor<4096x4096xbf16>
%cst = arith.constant 0.000000e+00 : bf16
%3 = linalg.fill ins(%cst : bf16) outs(%2 : tensor<4096x4096xbf16>) -> tensor<4096x4096xbf16>
%4 = linalg.matmul_transpose_b ins(%0, %1 : tensor<4096x4096xbf16>, tensor<4096x4096xbf16>) outs(%3 : tensor<4096x4096xbf16>) -> tensor<4096x4096xbf16>
bufferization.materialize_in_destination %4 in restrict writable %arg2 : (tensor<4096x4096xbf16>, memref<4096x4096xbf16>) -> ()
return
}
}

// -----
// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>} {
func.func @matmul_f32(%arg0: memref<4096x4096xf32>, %arg1: memref<4096x4096xf32>, %arg2: memref<4096x4096xf32>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<4096x4096xf32>
%1 = bufferization.to_tensor %arg1 restrict : memref<4096x4096xf32>
%2 = tensor.empty() : tensor<4096x4096xf32>
%cst = arith.constant 0.000000e+00 : f32
%3 = linalg.fill ins(%cst : f32) outs(%2 : tensor<4096x4096xf32>) -> tensor<4096x4096xf32>
%4 = linalg.matmul_transpose_b ins(%0, %1 : tensor<4096x4096xf32>, tensor<4096x4096xf32>) outs(%3 : tensor<4096x4096xf32>) -> tensor<4096x4096xf32>
bufferization.materialize_in_destination %4 in restrict writable %arg2 : (tensor<4096x4096xf32>, memref<4096x4096xf32>) -> ()
return
}
}
76 changes: 76 additions & 0 deletions test/mlir/test/gc/Integration/op/relu.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,76 @@
// RUN: gc-opt %s --gc-gpu-pipeline -split-input-file | FileCheck %s


// -----
// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>} {
func.func @relu_f16(%arg0: memref<1024x1024xf16>, %arg1: memref<1024x1024xf16>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<1024x1024xf16>
%1 = tensor.empty() : tensor<1024x1024xf16>
%cst = arith.constant 0.000000e+00 : f16
%2 = linalg.fill ins(%cst : f16) outs(%1 : tensor<1024x1024xf16>) -> tensor<1024x1024xf16>
%3 = linalg.max ins(%0, %2 : tensor<1024x1024xf16>, tensor<1024x1024xf16>) outs(%1 : tensor<1024x1024xf16>) -> tensor<1024x1024xf16>
bufferization.materialize_in_destination %3 in restrict writable %arg1 : (tensor<1024x1024xf16>, memref<1024x1024xf16>) -> ()
return
}
}

// -----
// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>} {
func.func @dynamic_relu(%arg0: memref<?x?xf16>, %arg1: memref<?x?xf16>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<?x?xf16>
%c0 = arith.constant 0 : index
%dim = tensor.dim %0, %c0 : tensor<?x?xf16>
%c1 = arith.constant 1 : index
%dim_0 = tensor.dim %0, %c1 : tensor<?x?xf16>
%1 = tensor.empty(%dim, %dim_0) : tensor<?x?xf16>
%cst = arith.constant 0.000000e+00 : f16
%2 = linalg.fill ins(%cst : f16) outs(%1 : tensor<?x?xf16>) -> tensor<?x?xf16>
%3 = linalg.max ins(%0, %2 : tensor<?x?xf16>, tensor<?x?xf16>) outs(%1 : tensor<?x?xf16>) -> tensor<?x?xf16>
bufferization.materialize_in_destination %3 in restrict writable %arg1 : (tensor<?x?xf16>, memref<?x?xf16>) -> ()
return
}
}

// -----
// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>} {
func.func @relu_bf16(%arg0: memref<1024x1024xbf16>, %arg1: memref<1024x1024xbf16>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<1024x1024xbf16>
%1 = tensor.empty() : tensor<1024x1024xbf16>
%cst = arith.constant 0.000000e+00 : bf16
%2 = linalg.fill ins(%cst : bf16) outs(%1 : tensor<1024x1024xbf16>) -> tensor<1024x1024xbf16>
%3 = linalg.max ins(%0, %2 : tensor<1024x1024xbf16>, tensor<1024x1024xbf16>) outs(%1 : tensor<1024x1024xbf16>) -> tensor<1024x1024xbf16>
bufferization.materialize_in_destination %3 in restrict writable %arg1 : (tensor<1024x1024xbf16>, memref<1024x1024xbf16>) -> ()
return
}
}

// -----
// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>} {
func.func @relu_f32(%arg0: memref<1024x1024xf32>, %arg1: memref<1024x1024xf32>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<1024x1024xf32>
%1 = tensor.empty() : tensor<1024x1024xf32>
%cst = arith.constant 0.000000e+00 : f32
%2 = linalg.fill ins(%cst : f32) outs(%1 : tensor<1024x1024xf32>) -> tensor<1024x1024xf32>
%3 = linalg.max ins(%0, %2 : tensor<1024x1024xf32>, tensor<1024x1024xf32>) outs(%1 : tensor<1024x1024xf32>) -> tensor<1024x1024xf32>
bufferization.materialize_in_destination %3 in restrict writable %arg1 : (tensor<1024x1024xf32>, memref<1024x1024xf32>) -> ()
return
}
}

// -----
// CHECK-LABEL: llvm
module @fragment_name attributes {"#dlti.sys_spec" = #dlti.target_system_spec<"GPU" : #dlti.target_device_spec<#dlti.dl_entry<"num_exec_units", 448 : i32>, #dlti.dl_entry<"num_exec_units_per_slice", 32 : i32>, #dlti.dl_entry<"num_threads_per_eu", 8 : i32>, #dlti.dl_entry<"L1_cache_size_in_bytes", 67108864 : i32>, #dlti.dl_entry<"max_vector_op_width", 256 : i32>, #dlti.dl_entry<"max_work_group_size", 1024 : i32>>>} {
func.func @relu_f32_corner_shape(%arg0: memref<1061x1061xf32>, %arg1: memref<1061x1061xf32>) {
%0 = bufferization.to_tensor %arg0 restrict : memref<1061x1061xf32>
%1 = tensor.empty() : tensor<1061x1061xf32>
%cst = arith.constant 0.000000e+00 : f32
%2 = linalg.fill ins(%cst : f32) outs(%1 : tensor<1061x1061xf32>) -> tensor<1061x1061xf32>
%3 = linalg.max ins(%0, %2 : tensor<1061x1061xf32>, tensor<1061x1061xf32>) outs(%1 : tensor<1061x1061xf32>) -> tensor<1061x1061xf32>
bufferization.materialize_in_destination %3 in restrict writable %arg1 : (tensor<1061x1061xf32>, memref<1061x1061xf32>) -> ()
return
}
}
Loading