mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-16 19:09:59 +00:00
chore(copyright) update library wide CMakeLists.txt copyright header template (#3313)
* chore(copyright) update library wide CMakeLists.txt files copyright header template
* Fix build
---------
Co-authored-by: Sami Remes <samremes@amd.com>
[ROCm/composable_kernel commit: 004784ef98]
This commit is contained in:
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_gemm_dl)
|
||||
|
||||
add_example_executable(example_gemm_dl_fp32 gemm_dl_fp32.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_gemm_bilinear_wmma_fp16 gemm_bilinear_wmma_fp16.cpp)
|
||||
add_example_executable(example_gemm_bilinear_wmma_int8 gemm_bilinear_wmma_int8.cpp)
|
||||
add_example_executable(example_gemm_bilinear_xdl_fp16 gemm_bilinear_xdl_fp16.cpp)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_gemm_bias_relu_xdl_fp16 gemm_bias_relu_xdl_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_gemm_add_add_fastgelu_xdl)
|
||||
add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
|
||||
add_example_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_convnd_fwd_xdl_fp32 convnd_fwd_xdl_fp32.cpp)
|
||||
add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
|
||||
add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_convnd_fwd_reduce_xdl)
|
||||
add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
|
||||
add_example_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_reduce_blockwise reduce_blockwise.cpp)
|
||||
add_example_executable(example_reduce_threadwise_multi_d reduce_threadwise_multi_d.cpp)
|
||||
add_example_executable(example_reduce_multiblock_atomic_add reduce_multiblock_atomic_add.cpp)
|
||||
|
||||
@@ -1,2 +1,5 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_pool2d_fwd_fp16 pool2d_fwd_fp16.cpp)
|
||||
add_example_executable(example_pool2d_fwd_fp32 pool2d_fwd_fp32.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
|
||||
add_example_executable(example_gemm_wmma_quantization_int8 gemm_wmma_quantization_int8.cpp)
|
||||
add_example_executable(example_gemm_xdl_bias_relu_quantization_int8 gemm_xdl_bias_relu_quantization_int8.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_grouped_gemm_xdl)
|
||||
add_example_executable(example_grouped_gemm_xdl_fp32 grouped_gemm_xdl_fp32.cpp)
|
||||
add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp32)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_gemm_reduce_xdl)
|
||||
add_custom_target(example_gemm_reduce_xdl_max)
|
||||
add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp)
|
||||
if(result EQUAL 0)
|
||||
target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_batched_gemm_reduce_xdl_fp16 batched_gemm_reduce_xdl_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_broadcast_add_2d_amn_bn broadcast_add_2d_amn_bn.cpp)
|
||||
add_example_executable(example_broadcast_add_3d_am_bmnk broadcast_add_3d_am_bmnk.cpp)
|
||||
add_example_executable(example_elementwise_add_1d elementwise_add_1d.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_grouped_conv_bwd_weight)
|
||||
add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
|
||||
add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_welford_fp16 gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp)
|
||||
add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_naive_fp16 gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp)
|
||||
add_example_executable(example_gemm_layernorm_xdl_naive_fp16 gemm_layernorm_xdl_naive_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_cgemm_xdl)
|
||||
|
||||
add_example_executable(example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_softmax_blockwise softmax_blockwise.cpp)
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_batched_gemm_xdl)
|
||||
|
||||
add_example_executable(example_batched_gemm_xdl_fp32 batched_gemm_xdl_fp32.cpp)
|
||||
|
||||
@@ -1,2 +1,5 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_gemm_bias_e_permute_g1m3n2k1_xdl_fp16 gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp)
|
||||
add_example_executable(example_gemm_bias_e_permute_g1m2n3k1_xdl_fp16 gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_contraction)
|
||||
add_custom_target(example_contraction_scale)
|
||||
add_custom_target(example_contraction_bilinear)
|
||||
|
||||
@@ -1,2 +1,5 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_layernorm2d_fwd_fp16 layernorm2d_fwd_fp16.cpp)
|
||||
add_example_executable(example_layernorm2d_fwd_splitk_fp16 layernorm2d_fwd_splitk_fp16.cpp)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_grouped_gemm_bias_e_permute_xdl_fp16 grouped_gemm_bias_e_permute_xdl_fp16.cpp)
|
||||
|
||||
@@ -1,2 +1,5 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp)
|
||||
add_example_executable(example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_grouped_conv_fwd_multiple_d)
|
||||
add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
|
||||
add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
|
||||
add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
|
||||
add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_wmma_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_wmma_fp16.cpp)
|
||||
add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_wmma_fp16 batched_gemm_scale_softmax_gemm_permute_wmma_fp16.cpp)
|
||||
add_example_executable(example_self_attention_forward_wmma_fp16 self_attention_forward_wmma_fp16.cpp)
|
||||
|
||||
@@ -1,2 +1,5 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_dual_reduce_multiblock dual_reduce_multiblock.cpp)
|
||||
add_example_executable(example_dual_reduce_threadwise dual_reduce_threadwise.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_batchnorm_forward_training batchnorm_forward_training_nhwc.cpp)
|
||||
add_example_executable(example_batchnorm_forward_training_obsolete batchnorm_forward_training_nhwc_obsolete.cpp)
|
||||
add_example_executable(example_batchnorm_forward_inferring batchnorm_forward_inferring_nhwc.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_splitK_gemm_xdl)
|
||||
add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
|
||||
add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
add_example_executable(example_sparse_embedding3_forward_layernorm sparse_embedding3_forward_layernorm.cpp)
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_sparse_embedding3_forward_layernorm sparse_embedding3_forward_layernorm.cpp)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_batched_gemm_add_add_relu_gemm_add_xdl_fp16 batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_grouped_conv_bwd_data)
|
||||
|
||||
add_example_executable(example_grouped_conv_bwd_data_xdl_fp16 grouped_conv_bwd_data_xdl_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_permute)
|
||||
|
||||
add_example_executable(example_permute_1xHxW_fp16 permute_1xHxW_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_conv2d_fwd_xdl_perlayer_quantization_int8 conv2d_fwd_xdl_perlayer_quantization_int8.cpp)
|
||||
add_example_executable(example_conv2d_fwd_xdl_perchannel_quantization_int8 conv2d_fwd_xdl_perchannel_quantization_int8.cpp)
|
||||
add_example_executable(example_conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8 conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
|
||||
add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
|
||||
add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_groupnorm_fwd_sigmoid_mul_fp16 groupnorm_fwd_sigmoid_mul_fp16.cpp)
|
||||
add_example_executable(example_groupnorm_fwd_splitk_fp16 groupnorm_fwd_splitk_fp16.cpp)
|
||||
add_example_executable(example_groupnorm_fwd_swish_fp16 groupnorm_fwd_swish_fp16.cpp)
|
||||
|
||||
@@ -1,2 +1,5 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp16 splitk_gemm_bias_e_permute_xdl_fp16.cpp)
|
||||
add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp32 splitk_gemm_bias_e_permute_xdl_fp32.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_elementwise_permute_4D_fp16 elementwise_permute_4D_fp16.cpp)
|
||||
add_example_executable(example_elementwise_permute_4D_fp32_row elementwise_permute_4D_fp32_row.cpp)
|
||||
add_example_executable(example_elementwise_permute_4D_fp16_row elementwise_permute_4D_fp16_row.cpp)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_elementwise_layernorm_blockwise elementwise_layernorm_blockwise.cpp)
|
||||
|
||||
@@ -1,2 +1,5 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_gemm_add_multiply_dl_fp16 gemm_add_multiply_dl_fp16.cpp)
|
||||
add_example_executable(example_gemm_add_multiply_xdl_fp16 gemm_add_multiply_xdl_fp16.cpp)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_gemm_bias_softmax_gemm_permute gemm_bias_softmax_gemm_permute_xdl.cpp)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_pool3d_fwd_fp16 pool3d_fwd_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_maxpool2d_bwd_bf16 maxpool2d_bwd_bf16.cpp)
|
||||
add_example_executable(example_maxpool2d_bwd_fp16 maxpool2d_bwd_fp16.cpp)
|
||||
add_example_executable(example_maxpool2d_bwd_fp32 maxpool2d_bwd_fp32.cpp)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_put_element_fp16 put_element_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_avgpool3d_bwd_bf16 avgpool3d_bwd_bf16.cpp)
|
||||
add_example_executable(example_avgpool3d_bwd_fp16 avgpool3d_bwd_fp16.cpp)
|
||||
add_example_executable(example_avgpool3d_bwd_fp32 avgpool3d_bwd_fp32.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_im2col_col2im)
|
||||
|
||||
add_example_executable(example_image_to_column_f32 image_to_column_f32.cpp)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_layernorm2d_bwd_fp32 layernorm2d_bwd_fp32.cpp)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_groupnorm_bwd_fp32 groupnorm_bwd_fp32.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_grouped_gemm_xdl_multi_abd)
|
||||
|
||||
add_example_executable(example_grouped_gemm_multi_abd_xdl_fixed_nk_bias_fp16 grouped_gemm_multi_abd_xdl_fixed_nk_bias_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_gemm_multi_ABD_wmma_fp16 gemm_multi_ABD_wmma_fp16.cpp)
|
||||
add_example_executable(example_gemm_multi_ABD_wmma_bias_fastgelu_bf16_i8 gemm_multi_ABD_wmma_bias_fastgelu_bf16_i8.cpp)
|
||||
add_example_executable(example_gemm_multi_ABD_wmma_multiply_bias_fastgelu_bf16_i8 gemm_multi_ABD_wmma_multiply_bias_fastgelu_bf16_i8.cpp)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_contraction_multi_ABD_xdl_fp16 contraction_multi_ABD_xdl_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_subdirectory(binary)
|
||||
add_subdirectory(convinvscale)
|
||||
add_subdirectory(convscale)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_convnd_activ_binary_xdl)
|
||||
# Bilinear residual
|
||||
add_example_executable(example_convnd_fwd_xdl_bilinear_residual_fp16 convnd_fwd_xdl_bilinear_residual_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
if (NOT GPU_TARGETS MATCHES "gfx11")
|
||||
add_custom_target(example_convnd_activ_xdl_convinvscale)
|
||||
add_example_executable(example_convnd_fwd_xdl_convinvscale_fp8 convnd_fwd_xdl_convinvscale_fp8.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
if (NOT GPU_TARGETS MATCHES "gfx11")
|
||||
add_custom_target(example_convnd_activ_xdl_convscale)
|
||||
add_example_executable(example_convnd_fwd_xdl_convscale_fp8 convnd_fwd_xdl_convscale_fp8.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
if (NOT GPU_TARGETS MATCHES "gfx11")
|
||||
add_custom_target(example_convnd_activ_xdl_convscale_add)
|
||||
add_example_executable(example_convnd_fwd_xdl_convscale_add_fp8 convnd_fwd_xdl_convscale_add_fp8.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
if (NOT GPU_TARGETS MATCHES "gfx11")
|
||||
add_custom_target(example_convnd_activ_xdl_convscale_reduce)
|
||||
add_example_executable(example_convnd_fwd_xdl_convscale_relu_amax_fp8 convnd_fwd_xdl_convscale_relu_amax_fp8.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
if (NOT GPU_TARGETS MATCHES "gfx11")
|
||||
add_custom_target(example_convnd_activ_xdl_convscale_relu)
|
||||
add_example_executable(example_convnd_fwd_xdl_convscale_relu_fp8 convnd_fwd_xdl_convscale_relu_fp8.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_convnd_activ_dynamic_unary_xdl)
|
||||
# Sigmoid
|
||||
add_example_executable(example_convnd_fwd_xdl_dynamic_sigmoid_fp16 convnd_fwd_xdl_dynamic_sigmoid_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_convnd_activ_multi_ab_xdl)
|
||||
# ScaleAdd on A and B
|
||||
add_example_executable(example_conv_fwd_xdl_scaleadd_ab_fp16 conv_fwd_xdl_scaleadd_ab_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_convnd_activ_unary_xdl)
|
||||
# Sigmoid
|
||||
add_example_executable(example_convnd_fwd_xdl_sigmoid_fp16 convnd_fwd_xdl_sigmoid_fp16.cpp)
|
||||
|
||||
@@ -1,2 +1,5 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_layernorm4d_fwd_fp16 layernorm4d_fwd_fp16.cpp)
|
||||
add_example_executable(example_layernorm4d_fwd_splitk_fp16 layernorm4d_fwd_splitk_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_fpAintB_gemm_wmma)
|
||||
add_example_executable(example_fp16int8_gemm_wmma fp16int8_gemm_wmma.cpp)
|
||||
add_example_dependencies(example_fpAintB_gemm_wmma example_fp16int8_gemm_wmma)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_gemm_multiply_multiply_xdl_fp8 gemm_multiply_multiply_xdl_fp8.cpp)
|
||||
add_example_executable(example_gemm_multiply_multiply_xdl_fp8_ab_scale gemm_multiply_multiply_xdl_fp8_ab_scale.cpp)
|
||||
add_example_executable(example_gemm_multiply_multiply_xdl_fp8_blockscale_bpreshuffle gemm_multiply_multiply_xdl_fp8_blockscale_bpreshuffle.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_example_executable(example_complex_contraction_bilinear_xdl_fp32 complex_contraction_bilinear_xdl_fp32.cpp)
|
||||
add_example_executable(example_complex_contraction_bilinear_xdl_fp64 complex_contraction_bilinear_xdl_fp64.cpp)
|
||||
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_gemm_mx)
|
||||
|
||||
add_example_executable(example_gemm_mx_fp8 gemm_mx_fp8.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_gemm_add_xdl)
|
||||
|
||||
add_example_executable(example_gemm_add_xdl_fp16 gemm_add_xdl_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_custom_target(example_gemm_add_relu_xdl)
|
||||
|
||||
add_example_executable(example_gemm_add_relu_xdl_fp16 gemm_add_relu_xdl_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
include_directories(BEFORE
|
||||
${PROJECT_SOURCE_DIR}/include
|
||||
${PROJECT_SOURCE_DIR}/library/include
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
set(INST_TARGETS ${SUPPORTED_GPU_TARGETS})
|
||||
# Currently only gfx9 and gfx12 archs are supported by FMHA
|
||||
list(FILTER INST_TARGETS INCLUDE REGEX "gfx9|gfx12")
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
set(LAYERNORM2D_FWD_KNOWN_APIS "fwd;bwd")
|
||||
set(LAYERNORM2D_FWD_ENABLE_APIS "fwd" CACHE STRING
|
||||
"semicolon-separated list of APIs to generate (${LAYERNORM2D_FWD_KNOWN_APIS}) & link, or \"all\".")
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
if(GPU_TARGETS MATCHES "gfx94|gfx95|gfx90a")
|
||||
add_executable(tile_example_gemm_basic gemm_basic.cpp)
|
||||
add_executable(tile_example_gemm_universal universal_gemm.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
# not using add_example_executable() to add this target, since we don't want this to have
|
||||
# to be included in "make all/install/check"
|
||||
add_executable(tile_example_img2col image_to_column.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
set(EXAMPLE_REDUCE "tile_example_reduce")
|
||||
# not using add_example_executable() to add this target, since we don't want this to have
|
||||
# to be included in "make all/install/check"
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
# not using add_example_executable() to add this target, since we don't want this to have
|
||||
# to be included in "make all/install/check"
|
||||
add_executable(tile_example_permute permute.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_executable(tile_example_topk_softmax topk_softmax.cpp topk_softmax_api.cpp)
|
||||
target_include_directories(tile_example_topk_softmax PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/)
|
||||
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
set(RMSNORM2D_FWD_KNOWN_APIS "fwd;bwd")
|
||||
set(RMSNORM2D_FWD_ENABLE_APIS "fwd" CACHE STRING
|
||||
"semicolon-separated list of APIs to generate (${RMSNORM2D_FWD_KNOWN_APIS}) & link, or \"all\".")
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
set(TILE_ADD_RMSNORM2D_RDQUANT_FWD "tile_add_rmsnorm2d_rdquant_fwd")
|
||||
# not using add_example_executable() to add this target, since we don't want this to have
|
||||
# to be included in "make all/install/check"
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
function (add_smoothquant_example TARGET_NAME MAIN_SRC)
|
||||
message(DEBUG "adding ${TARGET_NAME}")
|
||||
# not using add_example_executable() to add target, since we don't want this to have
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_executable(tile_example_moe_sorting moe_sorting.cpp moe_sorting_api.cpp)
|
||||
target_include_directories(tile_example_moe_sorting PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/)
|
||||
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
function (add_moe_smoothquant_example TARGET_NAME MAIN_SRC)
|
||||
message(DEBUG "adding ${TARGET_NAME}")
|
||||
# not using add_example_executable() to add target, since we don't want this to have
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
if(GPU_TARGETS MATCHES "gfx94|gfx95")
|
||||
set(TILE_EXAMPLE_FUSED_MOE "tile_example_fused_moe")
|
||||
message(DEBUG "adding ${TILE_EXAMPLE_FUSED_MOE}")
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_executable(tile_example_batched_gemm batched_gemm.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
if(GPU_TARGETS MATCHES "gfx94|gfx95|gfx90a")
|
||||
add_executable(tile_example_grouped_gemm grouped_gemm.cpp)
|
||||
add_executable(tile_example_quant_grouped_gemm quant_grouped_gemm.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
set(SUPPORTED_GPUS gfx908 gfx90a gfx942 gfx950)
|
||||
|
||||
set(has_supported_gpu FALSE)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_executable(tile_example_gemm_multi_d_fp16 gemm_multi_d_fp16.cpp)
|
||||
set(EXAMPLE_GEMM_COMPILE_OPTIONS)
|
||||
if(CK_USE_OCP_FP8)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
if(GPU_TARGETS MATCHES "gfx94|gfx95|gfx90a")
|
||||
set(EXAMPLE_CONV_COMPILE_OPTIONS)
|
||||
list(APPEND EXAMPLE_CONV_COMPILE_OPTIONS -mllvm -enable-noalias-to-md-conversion=0)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
# Elementwise example targets 2D inputs
|
||||
set(TARGET_NAME_2D_INPUT tile_example_elementwise)
|
||||
add_executable(${TARGET_NAME_2D_INPUT} elementwise_example.cpp)
|
||||
|
||||
@@ -1 +1,4 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_executable(tile_example_gemm_multi_abd_fp16 gemm_multi_abd_fp16.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
if(GPU_TARGETS MATCHES "gfx94|gfx95|gfx90a")
|
||||
set(TARGET_NAME tile_example_batched_transpose)
|
||||
add_executable(${TARGET_NAME} batched_transpose_example.cpp batched_transpose_api.cpp)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
set(EXAMPLE_POOL_3D "tile_example_pool3d")
|
||||
message(DEBUG "adding example ${EXAMPLE_POOL_3D}")
|
||||
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
set(EXAMPLE_GEMM_COMPILE_OPTIONS)
|
||||
if(CK_USE_OCP_FP8)
|
||||
list(APPEND EXAMPLE_GEMM_COMPILE_OPTIONS -DCK_TILE_USE_OCP_FP8)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
if(GPU_TARGETS MATCHES "gfx9")
|
||||
add_executable(tile_example_streamk_gemm_basic streamk_gemm_basic.cpp)
|
||||
set(EXAMPLE_GEMM_COMPILE_OPTIONS)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
add_executable(tile_example_batched_contraction batched_contraction.cpp)
|
||||
set(EXAMPLE_CONTRACTION_COMPILE_OPTIONS)
|
||||
if(CK_USE_OCP_FP8)
|
||||
|
||||
@@ -1,3 +1,6 @@
|
||||
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
# SPDX-License-Identifier: MIT
|
||||
|
||||
include_directories(AFTER
|
||||
${CMAKE_CURRENT_LIST_DIR}
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user