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
yangql
composable_kernel-1
Commits
a68b16a5
"composable_kernel/include/config.hpp" did not exist on "ac62d13ecdf1d959e62582a6bce017d6e4f90692"
Commit
a68b16a5
authored
Jun 06, 2019
by
Chao Liu
Browse files
refactor
parent
0a386c46
Changes
2
Show whitespace changes
Inline
Side-by-side
Showing
2 changed files
with
6 additions
and
27 deletions
+6
-27
driver/driver.hip.cpp
driver/driver.hip.cpp
+2
-2
src/include/blockwise_generic_tensor_slice_op.hip.hpp
src/include/blockwise_generic_tensor_slice_op.hip.hpp
+4
-25
No files found.
driver/driver.hip.cpp
View file @
a68b16a5
...
...
@@ -646,9 +646,9 @@ int main(int argc, char* argv[])
device_convolution_implicit_gemm_v1_nchw_cyxk_nkhw
#elif 0
device_convolution_implicit_gemm_v2_chwn_cyxk_khwn
#elif 1
device_convolution_implicit_gemm_v3_nchw_cyxk_nkhw
#elif 0
device_convolution_implicit_gemm_v3_nchw_cyxk_nkhw
#elif 1
device_convolution_implicit_gemm_v4_nchw_kcyx_nkhw
#endif
(
in_nchw_desc
,
in_nchw
,
wei_kcyx_desc
,
wei_kcyx
,
out_nkhw_desc
,
out_nkhw_device
,
nrepeat
);
...
...
src/include/blockwise_generic_tensor_slice_op.hip.hpp
View file @
a68b16a5
...
...
@@ -203,7 +203,7 @@ struct BlockwiseGenericTensorSliceCopy_v1
make_ConstantTensorDescriptor_packed
(
thread_sub_tensor_lengths
*
repeat_lengths
);
static_ford
<
decltype
(
repeat_lengths
)
>
{}([
&
](
auto
repeat_multi_id_
)
{
#if
1
#if
0
constexpr auto repeat_multi_id = sequence2array(decltype(repeat_multi_id_){});
const auto src_thread_data_multi_id_begin = repeat_multi_id * data_per_cluster_per_dims;
...
...
@@ -215,7 +215,7 @@ struct BlockwiseGenericTensorSliceCopy_v1
const index_t clipboard_offset =
thread_tensor_desc.GetOffsetFromMultiIndex(clipboard_data_multi_id_begin);
#else
#else
// HIP compiler performs better with these codes
constexpr
auto
repeat_multi_id
=
decltype
(
repeat_multi_id_
){};
constexpr
auto
src_thread_data_multi_id_begin
=
...
...
@@ -256,7 +256,7 @@ struct BlockwiseGenericTensorSliceCopy_v1
make_ConstantTensorDescriptor_packed
(
thread_sub_tensor_lengths
*
repeat_lengths
);
static_ford
<
decltype
(
repeat_lengths
)
>
{}([
&
](
auto
repeat_multi_id_
)
{
#if
1
#if
0
constexpr auto repeat_multi_id = sequence2array(decltype(repeat_multi_id_){});
const auto clipboard_data_multi_id_begin = repeat_multi_id * thread_sub_tensor_lengths;
...
...
@@ -267,7 +267,7 @@ struct BlockwiseGenericTensorSliceCopy_v1
thread_tensor_desc.GetOffsetFromMultiIndex(clipboard_data_multi_id_begin);
const index_t dst_offset = DstDesc{}.GetOffsetFromMultiIndex(dst_data_multi_id_begin);
#else
#else
// HIP compiler performs better with these codes
constexpr
auto
repeat_multi_id
=
decltype
(
repeat_multi_id_
){};
constexpr
auto
clipboard_data_multi_id_begin
=
...
...
@@ -338,27 +338,6 @@ struct BlockwiseGenericTensorSliceCopy_v1
src_partial_original_desc
.
UpdateMultiIndexGivenStepSizeOf1dIndex
(
old_src_partial_original_multi_id
,
StepSize
,
direction
);
#if 0
{
if(debug_flag && get_block_1d_id() == 0)
{
printf("id %5u %5u: "
"old_src_partial_original_multi_id %u %u %u, "
"new_src_partial_original_multi_id %u %u %u, "
"mThreadSrcOffset %u, mThreadDstOffset %u \n",
get_block_1d_id(),
get_thread_local_1d_id(),
old_src_partial_original_multi_id[0],
old_src_partial_original_multi_id[1],
old_src_partial_original_multi_id[2],
new_src_partial_original_multi_id[0],
new_src_partial_original_multi_id[1],
new_src_partial_original_multi_id[2]
);
}
}
#endif
// update "mThreadSrcOriginalMultiId"
static_for
<
0
,
decltype
(
src_partial_original_dims
)
::
GetSize
(),
1
>
{}([
&
](
auto
I_
)
{
constexpr
auto
I
=
decltype
(
I_
){};
...
...
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