Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
Menu
Open sidebar
gaoqiong
composable_kernel_ROCM
Commits
7ebdd969
Commit
7ebdd969
authored
Aug 06, 2024
by
Jing Zhang
Browse files
add kbatch printout
parent
c3725a45
Changes
2
Hide whitespace changes
Inline
Side-by-side
Showing
2 changed files
with
160 additions
and
158 deletions
+160
-158
profiler/include/profiler/profile_gemm_multiply_multiply_impl.hpp
.../include/profiler/profile_gemm_multiply_multiply_impl.hpp
+8
-6
profiler/src/CMakeLists.txt
profiler/src/CMakeLists.txt
+152
-152
No files found.
profiler/include/profiler/profile_gemm_multiply_multiply_impl.hpp
View file @
7ebdd969
...
...
@@ -183,6 +183,7 @@ bool profile_gemm_multiply_multiply_impl(int do_verification,
float
best_ave_time
=
0
;
float
best_tflops
=
0
;
float
best_gb_per_sec
=
0
;
float
best_kbatch
=
0
;
// profile device GEMM instances
for
(
auto
&
op_ptr
:
op_ptrs
)
...
...
@@ -197,7 +198,6 @@ bool profile_gemm_multiply_multiply_impl(int do_verification,
for
(
std
::
size_t
i
=
0
;
i
<
kbatch_list
.
size
();
i
++
)
{
auto
kbatch_curr
=
kbatch_list
[
i
];
auto
argument_ptr
=
op_ptr
->
MakeArgumentPointer
(
...
...
@@ -213,7 +213,7 @@ bool profile_gemm_multiply_multiply_impl(int do_verification,
StrideB
,
std
::
array
<
ck
::
index_t
,
2
>
{
StrideD0
,
StrideD1
},
StrideE
,
KB
atch
,
kb
atch
_curr
,
a_element_op
,
b_element_op
,
c_element_op
);
...
...
@@ -269,7 +269,8 @@ bool profile_gemm_multiply_multiply_impl(int do_verification,
float
gb_per_sec
=
num_btype
/
1.E6
/
ave_time
;
std
::
cout
<<
"Perf: "
<<
std
::
setw
(
10
)
<<
ave_time
<<
" ms, "
<<
tflops
<<
" TFlops, "
<<
gb_per_sec
<<
" GB/s, "
<<
op_name
<<
std
::
endl
;
<<
" TFlops, "
<<
gb_per_sec
<<
" GB/s, "
<<
op_name
<<
", KBatch "
<<
kbatch_curr
<<
std
::
endl
;
#if defined CK_ENABLE_FP8
// set softer tolerances for fp8
...
...
@@ -296,6 +297,7 @@ bool profile_gemm_multiply_multiply_impl(int do_verification,
best_tflops
=
tflops
;
best_ave_time
=
ave_time
;
best_gb_per_sec
=
gb_per_sec
;
best_kbatch
=
kbatch_curr
;
}
}
else
...
...
@@ -342,9 +344,9 @@ bool profile_gemm_multiply_multiply_impl(int do_verification,
}
std
::
cout
<<
" M = "
<<
M
<<
" N = "
<<
N
<<
" K = "
<<
K
<<
" StrideA = "
<<
StrideA
<<
" StrideB = "
<<
StrideB
<<
" StrideE = "
<<
StrideE
<<
" : "
<<
best_ave_time
<<
"
ms,
"
<<
best_
tflops
<<
" TFlops,
"
<<
best_
gb_per_sec
<<
"
GB/
s, "
<<
best_op_name
<<
std
::
endl
;
<<
" StrideB = "
<<
StrideB
<<
" StrideE = "
<<
StrideE
<<
" : "
<<
"
KBatch =
"
<<
best_
kbatch
<<
" :
"
<<
best_
ave_time
<<
"
m
s, "
<<
best_tflops
<<
" TFlops, "
<<
best_gb_per_sec
<<
" GB/s, "
<<
best_op_name
<<
std
::
endl
;
return
pass
;
}
...
...
profiler/src/CMakeLists.txt
View file @
7ebdd969
# ckProfiler
set
(
PROFILER_SOURCES
profiler.cpp
profile_gemm.cpp
profile_reduce.cpp
profile_groupnorm_bwd_data.cpp
profile_groupnorm_fwd.cpp
profile_layernorm_bwd_data.cpp
profile_layernorm_bwd_gamma_beta.cpp
profile_groupnorm_bwd_gamma_beta.cpp
profile_layernorm_fwd.cpp
profile_max_pool3d_fwd.cpp
profile_avg_pool3d_bwd.cpp
profile_max_pool3d_bwd.cpp
profile_softmax.cpp
profile_batchnorm_fwd.cpp
profile_batchnorm_bwd.cpp
profile_batchnorm_infer.cpp
profile_conv_tensor_rearrange.cpp
profile_transpose.cpp
profile_permute_scale.cpp
#
profile_gemm.cpp
#
profile_reduce.cpp
#
profile_groupnorm_bwd_data.cpp
#
profile_groupnorm_fwd.cpp
#
profile_layernorm_bwd_data.cpp
#
profile_layernorm_bwd_gamma_beta.cpp
#
profile_groupnorm_bwd_gamma_beta.cpp
#
profile_layernorm_fwd.cpp
#
profile_max_pool3d_fwd.cpp
#
profile_avg_pool3d_bwd.cpp
#
profile_max_pool3d_bwd.cpp
#
profile_softmax.cpp
#
profile_batchnorm_fwd.cpp
#
profile_batchnorm_bwd.cpp
#
profile_batchnorm_infer.cpp
#
profile_conv_tensor_rearrange.cpp
#
profile_transpose.cpp
#
profile_permute_scale.cpp
)
if
(
GPU_TARGETS MATCHES
"gfx9"
)
if
(
DTYPES MATCHES
"fp32"
OR DTYPES MATCHES
"fp64"
OR NOT DEFINED DTYPES
)
list
(
APPEND PROFILER_SOURCES profile_contraction_bilinear.cpp
)
list
(
APPEND PROFILER_SOURCES profile_contraction_scale.cpp
)
endif
()
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
list
(
APPEND PROFILER_SOURCES profile_gemm_reduce.cpp
)
list
(
APPEND PROFILER_SOURCES profile_batched_gemm_gemm.cpp
)
list
(
APPEND PROFILER_SOURCES profile_batched_gemm_add_relu_gemm_add.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_add.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_add_add_fastgelu.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_add_fastgelu.cpp
)
list
(
APPEND PROFILER_SOURCES profile_grouped_gemm.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_streamk.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_fastgelu.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_add_relu.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_add_silu.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_add_relu_add_layernorm.cpp
)
list
(
APPEND PROFILER_SOURCES profile_grouped_gemm_fixed_nk.cpp
)
list
(
APPEND PROFILER_SOURCES profile_grouped_gemm_two_stage.cpp
)
list
(
APPEND PROFILER_SOURCES profile_grouped_gemm_fastgelu.cpp
)
list
(
APPEND PROFILER_SOURCES profile_grouped_gemm_tile_loop.cpp
)
list
(
APPEND PROFILER_SOURCES profile_grouped_gemm_multiply_tile_loop.cpp
)
endif
()
list
(
APPEND PROFILER_SOURCES profile_gemm_multiply_add.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_multiply_multiply.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_ab_scale.cpp
)
list
(
APPEND PROFILER_SOURCES profile_batched_gemm.cpp
)
list
(
APPEND PROFILER_SOURCES profile_batched_gemm_reduce.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_add_multiply.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_bias_add_reduce.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_splitk.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_universal.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_universal_reduce.cpp
)
list
(
APPEND PROFILER_SOURCES profile_gemm_universal_streamk.cpp
)
list
(
APPEND PROFILER_SOURCES profile_conv_fwd_bias_relu.cpp
)
list
(
APPEND PROFILER_SOURCES profile_conv_fwd_bias_relu_add.cpp
)
list
(
APPEND PROFILER_SOURCES profile_conv_bwd_data.cpp
)
list
(
APPEND PROFILER_SOURCES profile_conv_fwd.cpp
)
list
(
APPEND PROFILER_SOURCES profile_grouped_conv_fwd_outelementop.cpp
)
endif
()
if
(
GPU_TARGETS MATCHES
"gfx11"
OR GPU_TARGETS MATCHES
"gfx12"
OR GPU_TARGETS MATCHES
"gfx9"
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
list
(
APPEND PROFILER_SOURCES profile_gemm_bilinear.cpp
)
endif
()
list
(
APPEND PROFILER_SOURCES profile_grouped_conv_fwd.cpp
)
list
(
APPEND PROFILER_SOURCES profile_grouped_conv_bwd_data.cpp
)
list
(
APPEND PROFILER_SOURCES profile_grouped_conv_bwd_weight.cpp
)
endif
()
if
(
DL_KERNELS
)
list
(
APPEND PROFILER_SOURCES profile_batched_gemm_multi_d.cpp
)
list
(
APPEND PROFILER_SOURCES profile_grouped_conv_bwd_weight.cpp
)
endif
()
#
if(GPU_TARGETS MATCHES "gfx9")
#
if(DTYPES MATCHES "fp32" OR DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
#
list(APPEND PROFILER_SOURCES profile_contraction_bilinear.cpp)
#
list(APPEND PROFILER_SOURCES profile_contraction_scale.cpp)
#
endif()
#
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
#
list(APPEND PROFILER_SOURCES profile_gemm_reduce.cpp)
#
list(APPEND PROFILER_SOURCES profile_batched_gemm_gemm.cpp)
#
list(APPEND PROFILER_SOURCES profile_batched_gemm_add_relu_gemm_add.cpp)
#
list(APPEND PROFILER_SOURCES profile_gemm_add.cpp)
#
list(APPEND PROFILER_SOURCES profile_gemm_add_add_fastgelu.cpp)
#
list(APPEND PROFILER_SOURCES profile_gemm_add_fastgelu.cpp)
#
list(APPEND PROFILER_SOURCES profile_grouped_gemm.cpp)
#
list(APPEND PROFILER_SOURCES profile_gemm_streamk.cpp)
#
list(APPEND PROFILER_SOURCES profile_gemm_fastgelu.cpp)
#
list(APPEND PROFILER_SOURCES profile_gemm_add_relu.cpp)
#
list(APPEND PROFILER_SOURCES profile_gemm_add_silu.cpp)
#
list(APPEND PROFILER_SOURCES profile_gemm_add_relu_add_layernorm.cpp)
#
list(APPEND PROFILER_SOURCES profile_grouped_gemm_fixed_nk.cpp)
#
list(APPEND PROFILER_SOURCES profile_grouped_gemm_two_stage.cpp)
#
list(APPEND PROFILER_SOURCES profile_grouped_gemm_fastgelu.cpp)
#
list(APPEND PROFILER_SOURCES profile_grouped_gemm_tile_loop.cpp)
#
list(APPEND PROFILER_SOURCES profile_grouped_gemm_multiply_tile_loop.cpp)
#
endif()
#
list(APPEND PROFILER_SOURCES profile_gemm_multiply_add.cpp)
list
(
APPEND PROFILER_SOURCES profile_gemm_multiply_multiply.cpp
)
#
list(APPEND PROFILER_SOURCES profile_gemm_ab_scale.cpp)
#
list(APPEND PROFILER_SOURCES profile_batched_gemm.cpp)
#
list(APPEND PROFILER_SOURCES profile_batched_gemm_reduce.cpp)
#
list(APPEND PROFILER_SOURCES profile_gemm_add_multiply.cpp)
#
list(APPEND PROFILER_SOURCES profile_gemm_bias_add_reduce.cpp)
#
list(APPEND PROFILER_SOURCES profile_gemm_splitk.cpp)
list
(
APPEND PROFILER_SOURCES profile_gemm_universal.cpp
)
#
list(APPEND PROFILER_SOURCES profile_gemm_universal_reduce.cpp)
#
list(APPEND PROFILER_SOURCES profile_gemm_universal_streamk.cpp)
#
list(APPEND PROFILER_SOURCES profile_conv_fwd_bias_relu.cpp)
#
list(APPEND PROFILER_SOURCES profile_conv_fwd_bias_relu_add.cpp)
#
list(APPEND PROFILER_SOURCES profile_conv_bwd_data.cpp)
#
list(APPEND PROFILER_SOURCES profile_conv_fwd.cpp)
#
list(APPEND PROFILER_SOURCES profile_grouped_conv_fwd_outelementop.cpp)
#
#
endif()
#
#
if(GPU_TARGETS MATCHES "gfx11" OR GPU_TARGETS MATCHES "gfx12" OR GPU_TARGETS MATCHES "gfx9")
#
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
#
list(APPEND PROFILER_SOURCES profile_gemm_bilinear.cpp)
#
endif()
#
list(APPEND PROFILER_SOURCES profile_grouped_conv_fwd.cpp)
#
list(APPEND PROFILER_SOURCES profile_grouped_conv_bwd_data.cpp)
#
list(APPEND PROFILER_SOURCES profile_grouped_conv_bwd_weight.cpp)
#
endif()
#
#
if(DL_KERNELS)
#
list(APPEND PROFILER_SOURCES profile_batched_gemm_multi_d.cpp)
#
list(APPEND PROFILER_SOURCES profile_grouped_conv_bwd_weight.cpp)
#
endif()
set
(
PROFILER_EXECUTABLE ckProfiler
)
...
...
@@ -89,83 +89,83 @@ if(NOT WIN32 AND ${hip_VERSION_FLAT} GREATER 600241132)
endif
()
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE utility getopt::getopt
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_normalization_fwd_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_normalization_bwd_data_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_normalization_bwd_gamma_beta_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_softmax_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_reduce_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_batchnorm_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_pool3d_fwd_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_avg_pool3d_bwd_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_max_pool_bwd_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_image_to_column_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_column_to_image_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_transpose_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_permute_scale_instance
)
if
(
GPU_TARGETS MATCHES
"gfx9"
)
if
(
DTYPES MATCHES
"fp32"
OR DTYPES MATCHES
"fp64"
OR NOT DEFINED DTYPES
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_contraction_bilinear_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_contraction_scale_instance
)
endif
()
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_add_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_add_add_fastgelu_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_fastgelu_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_batched_gemm_gemm_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_batched_gemm_add_relu_gemm_add_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_gemm_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_streamk_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_add_fastgelu_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_add_relu_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_add_silu_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_add_relu_add_layernorm_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_gemm_fixed_nk_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_gemm_fastgelu_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_gemm_tile_loop_instance
)
endif
()
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_batched_gemm_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_batched_gemm_reduce_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_multiply_add_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_multiply_multiply_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_ab_scale_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_splitk_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_universal_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_universal_reduce_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_universal_streamk_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_add_multiply_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_reduce_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_bias_add_reduce_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_conv2d_fwd_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_conv2d_fwd_bias_relu_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_conv2d_fwd_bias_relu_add_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_conv1d_fwd_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_conv1d_bwd_data_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_conv3d_bwd_data_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_conv2d_bwd_data_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_conv1d_bwd_weight_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_conv2d_bwd_weight_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_conv3d_fwd_convscale_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_conv3d_fwd_convinvscale_instance
)
endif
()
if
(
GPU_TARGETS MATCHES
"gfx9"
OR GPU_TARGETS MATCHES
"gfx11"
OR GPU_TARGETS MATCHES
"gfx12"
)
if
(
DTYPES MATCHES
"fp16"
OR NOT DEFINED DTYPES
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_bilinear_instance
)
endif
()
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_conv3d_fwd_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_conv2d_bwd_data_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_conv3d_bwd_data_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_conv2d_fwd_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_conv3d_bwd_weight_instance
)
endif
()
if
(
DL_KERNELS
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_batched_gemm_multi_d_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_conv1d_bwd_weight_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_conv2d_bwd_weight_instance
)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_grouped_conv3d_bwd_weight_instance
)
endif
()
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_fwd_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_bwd_data_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_bwd_gamma_beta_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_softmax_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_reduce_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool3d_fwd_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_avg_pool3d_bwd_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_max_pool_bwd_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_image_to_column_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_column_to_image_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_transpose_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_permute_scale_instance)
#
#
if(GPU_TARGETS MATCHES "gfx9")
#
if(DTYPES MATCHES "fp32" OR DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
#
endif()
#
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_add_fastgelu_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_fastgelu_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_gemm_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_add_relu_gemm_add_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_streamk_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_fastgelu_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_silu_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_add_layernorm_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fixed_nk_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgelu_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_tile_loop_instance)
#
endif()
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_reduce_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_multiply_add_instance)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_multiply_multiply_instance
)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_ab_scale_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_splitk_instance)
target_link_libraries
(
${
PROFILER_EXECUTABLE
}
PRIVATE device_gemm_universal_instance
)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_universal_reduce_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_universal_streamk_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_multiply_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_reduce_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bias_add_reduce_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_bias_relu_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_bias_relu_add_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_fwd_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv1d_bwd_data_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv3d_bwd_data_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_bwd_data_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_bwd_weight_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_weight_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_fwd_convscale_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_fwd_convinvscale_instance)
#
endif()
#
#
if(GPU_TARGETS MATCHES "gfx9" OR GPU_TARGETS MATCHES "gfx11" OR GPU_TARGETS MATCHES "gfx12")
#
if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bilinear_instance)
#
endif()
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_fwd_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_fwd_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_weight_instance)
#
endif()
#
#
if(DL_KERNELS)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_bwd_weight_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_weight_instance)
#
target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_weight_instance)
#
endif()
rocm_install
(
TARGETS
${
PROFILER_EXECUTABLE
}
COMPONENT profiler
)
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment
