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
Commits
159e1354
"git@developer.sourcefind.cn:orangecat/ollama.git" did not exist on "abec7f06e5ec9e8dc5741893801d4548763d4c67"
Commit
159e1354
authored
Dec 04, 2023
by
Jun Liu
Browse files
Fix more places for workaround
parent
0f58a333
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
23 additions
and
20 deletions
+23
-20
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_scaleadd_ab.hpp
..._instance/gpu/grouped_convolution_forward_scaleadd_ab.hpp
+23
-20
No files found.
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_scaleadd_ab.hpp
View file @
159e1354
...
...
@@ -23,19 +23,20 @@ using ScaleAdd = ck::tensor_operation::element_wise::ScaleAdd;
#ifdef CK_ENABLE_BF16
// grouped conv3d forward multi AB scaleadd, NDHWGC/GKZYXC/NDHWGK
void
add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instances
(
std
::
vector
<
std
::
unique_ptr
<
DeviceGroupedConvFwdMultipleABD
<
3
,
NDHWGC
,
GKZYXC
,
ck
::
Tuple
<>
,
NDHWGK
,
ck
::
Tuple
<
BF16
,
BF16
>
,
ck
::
Tuple
<
BF16
,
BF16
>
,
ck
::
Tuple
<>
,
BF16
,
ScaleAdd
,
ScaleAdd
,
PassThrough
>>>&
instances
);
// TODO: Workaround for https://ontrack-internal.amd.com/browse/SWDEV-435347
// void add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
// std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
// NDHWGC,
// GKZYXC,
// ck::Tuple<>,
// NDHWGK,
// ck::Tuple<BF16, BF16>,
// ck::Tuple<BF16, BF16>,
// ck::Tuple<>,
// BF16,
// ScaleAdd,
// ScaleAdd,
// PassThrough>>>& instances);
#endif
#ifdef CK_ENABLE_FP16
...
...
@@ -151,13 +152,15 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
}
#endif
#ifdef CK_ENABLE_BF16
if
constexpr
(
is_same_v
<
InDataType
,
ck
::
Tuple
<
ck
::
bhalf_t
,
ck
::
bhalf_t
>>
&&
is_same_v
<
WeiDataType
,
ck
::
Tuple
<
ck
::
bhalf_t
,
ck
::
bhalf_t
>>
&&
is_same_v
<
OutDataType
,
ck
::
bhalf_t
>
&&
is_same_v
<
ComputeType
,
ck
::
bhalf_t
>
)
{
add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instances
(
op_ptrs
);
}
// TODO: Workaround for https://ontrack-internal.amd.com/browse/SWDEV-435347
// if constexpr(is_same_v<InDataType, ck::Tuple<ck::bhalf_t, ck::bhalf_t>> &&
// is_same_v<WeiDataType, ck::Tuple<ck::bhalf_t, ck::bhalf_t>> &&
// is_same_v<OutDataType, ck::bhalf_t> && is_same_v<ComputeType,
// ck::bhalf_t>)
// {
// add_device_grouped_conv3d_fwd_xdl_scaleadd_ab_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
// op_ptrs);
// }
#endif
#ifdef CK_ENABLE_INT8
if
constexpr
(
is_same_v
<
InDataType
,
ck
::
Tuple
<
int8_t
,
int8_t
>>
&&
...
...
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