Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
magicwindyyd
mindspore
提交
c964916e
M
mindspore
项目概览
magicwindyyd
/
mindspore
与 Fork 源项目一致
Fork自
MindSpore / mindspore
通知
1
Star
1
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
0
列表
看板
标记
里程碑
合并请求
0
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
M
mindspore
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
0
Issue
0
列表
看板
标记
里程碑
合并请求
0
合并请求
0
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
提交
c964916e
编写于
5月 21, 2020
作者:
M
mindspore-ci-bot
提交者:
Gitee
5月 21, 2020
浏览文件
操作
浏览文件
下载
差异文件
!1254 change runtime stream type from uintptr_t to void*
Merge pull request !1254 from caifubi/change-uintprt_t-to-void-ptr
上级
ab6bc60d
5b963aef
变更
111
展开全部
隐藏空白更改
内联
并排
Showing
111 changed file
with
142 addition
and
155 deletion
+142
-155
mindspore/ccsrc/device/gpu/gpu_kernel_runtime.cc
mindspore/ccsrc/device/gpu/gpu_kernel_runtime.cc
+1
-1
mindspore/ccsrc/device/kernel_runtime.cc
mindspore/ccsrc/device/kernel_runtime.cc
+1
-2
mindspore/ccsrc/kernel/aicpu/aicpu_kernel_mod.cc
mindspore/ccsrc/kernel/aicpu/aicpu_kernel_mod.cc
+3
-4
mindspore/ccsrc/kernel/aicpu/aicpu_kernel_mod.h
mindspore/ccsrc/kernel/aicpu/aicpu_kernel_mod.h
+1
-1
mindspore/ccsrc/kernel/akg/gpu/akg_gpu_kernel_mod.cc
mindspore/ccsrc/kernel/akg/gpu/akg_gpu_kernel_mod.cc
+1
-1
mindspore/ccsrc/kernel/akg/gpu/akg_gpu_kernel_mod.h
mindspore/ccsrc/kernel/akg/gpu/akg_gpu_kernel_mod.h
+1
-1
mindspore/ccsrc/kernel/cpu/cpu_kernel.h
mindspore/ccsrc/kernel/cpu/cpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/arrays/argmax_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/arrays/argmax_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/arrays/array_reduce_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/arrays/array_reduce_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/arrays/concatv2_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/arrays/concatv2_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/arrays/gather_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/arrays/gather_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/arrays/one_hot_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/arrays/one_hot_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/arrays/select_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/arrays/select_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/arrays/slice_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/arrays/slice_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/arrays/slice_grad_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/arrays/slice_grad_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/arrays/transpose_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/arrays/transpose_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/arrays/unsorted_segment_sum_gpu_kernel.h
...ccsrc/kernel/gpu/arrays/unsorted_segment_sum_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/control/recv_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/control/recv_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/control/send_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/control/send_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/data/dataset_init_kernel.cc
mindspore/ccsrc/kernel/gpu/data/dataset_init_kernel.cc
+1
-1
mindspore/ccsrc/kernel/gpu/data/dataset_init_kernel.h
mindspore/ccsrc/kernel/gpu/data/dataset_init_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/data/dataset_iterator_kernel.cc
mindspore/ccsrc/kernel/gpu/data/dataset_iterator_kernel.cc
+1
-1
mindspore/ccsrc/kernel/gpu/data/dataset_iterator_kernel.h
mindspore/ccsrc/kernel/gpu/data/dataset_iterator_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/math/addn_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/math/addn_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/math/assign_add_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/math/assign_add_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/math/bias_add_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/math/bias_add_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/math/broadcast_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/math/broadcast_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/math/broadcast_grad_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/math/broadcast_grad_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/math/equalcount_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/math/equalcount_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/math/float_status_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/math/float_status_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/math/matmul_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/math/matmul_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/math/tensoradd_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/math/tensoradd_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/math/unary_op_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/math/unary_op_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nccl/nccl_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nccl/nccl_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/bias_add_grad_gpu_kenel.h
mindspore/ccsrc/kernel/gpu/nn/bias_add_grad_gpu_kenel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/conv2d_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/conv2d_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/conv2d_grad_filter_gpu_kernel.h
...spore/ccsrc/kernel/gpu/nn/conv2d_grad_filter_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/conv2d_grad_input_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/conv2d_grad_input_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/dropout_gpu_kernel.cc
mindspore/ccsrc/kernel/gpu/nn/dropout_gpu_kernel.cc
+1
-1
mindspore/ccsrc/kernel/gpu/nn/dropout_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/dropout_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/dropout_grad_kernel.cc
mindspore/ccsrc/kernel/gpu/nn/dropout_grad_kernel.cc
+1
-1
mindspore/ccsrc/kernel/gpu/nn/dropout_grad_kernel.h
mindspore/ccsrc/kernel/gpu/nn/dropout_grad_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/flatten_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/flatten_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/flatten_grad_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/flatten_grad_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/fused_batch_norm_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/fused_batch_norm_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/fused_batchnorm_grad_gpu_kernel.h
...ore/ccsrc/kernel/gpu/nn/fused_batchnorm_grad_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/gelu_grad_kernel.h
mindspore/ccsrc/kernel/gpu/nn/gelu_grad_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/gelu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/gelu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/layer_norm_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/layer_norm_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/layer_norm_grad_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/layer_norm_grad_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/lstm_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/lstm_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/lstm_grad_data_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/lstm_grad_data_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/lstm_grad_weight_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/lstm_grad_weight_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/momentum_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/momentum_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/pooling_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/pooling_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/pooling_grad_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/pooling_grad_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/relu_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/relu_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/relu_grad_kernel.h
mindspore/ccsrc/kernel/gpu/nn/relu_grad_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/rmsprop_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/rmsprop_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/softmax_cross_entropy_with_logits_gpu_kernel.h
...nel/gpu/nn/softmax_cross_entropy_with_logits_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/softmax_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/softmax_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/softmax_grad_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/softmax_grad_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/sparse_softmax_cross_entropy_with_logits_gpu_kernel.h
.../nn/sparse_softmax_cross_entropy_with_logits_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/tanh_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/nn/tanh_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/nn/tanh_grad_kernel.h
mindspore/ccsrc/kernel/gpu/nn/tanh_grad_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/other/assign_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/other/assign_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/quant/batchnorm_fold2_gpu_kernel.h
...spore/ccsrc/kernel/gpu/quant/batchnorm_fold2_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/quant/batchnorm_fold2_grad_gpu_kernel.h
.../ccsrc/kernel/gpu/quant/batchnorm_fold2_grad_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/quant/batchnorm_fold_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/quant/batchnorm_fold_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/quant/batchnorm_fold_grad_gpu_kernel.h
...e/ccsrc/kernel/gpu/quant/batchnorm_fold_grad_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/quant/correction_mul_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/quant/correction_mul_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/quant/correction_mul_grad_gpu_kernel.h
...e/ccsrc/kernel/gpu/quant/correction_mul_grad_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/quant/fake_quant_gpu_kernel.cc
mindspore/ccsrc/kernel/gpu/quant/fake_quant_gpu_kernel.cc
+1
-1
mindspore/ccsrc/kernel/gpu/quant/fake_quant_gpu_kernel.h
mindspore/ccsrc/kernel/gpu/quant/fake_quant_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/quant/fake_quant_grad_gpu_kernel.cc
...pore/ccsrc/kernel/gpu/quant/fake_quant_grad_gpu_kernel.cc
+1
-1
mindspore/ccsrc/kernel/gpu/quant/fake_quant_grad_gpu_kernel.h
...spore/ccsrc/kernel/gpu/quant/fake_quant_grad_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/gpu/quant/fake_quant_per_channel_gpu_kernel.cc
...src/kernel/gpu/quant/fake_quant_per_channel_gpu_kernel.cc
+3
-3
mindspore/ccsrc/kernel/gpu/quant/fake_quant_per_channel_gpu_kernel.h
...csrc/kernel/gpu/quant/fake_quant_per_channel_gpu_kernel.h
+3
-3
mindspore/ccsrc/kernel/gpu/quant/fake_quant_per_channel_grad_gpu_kernel.cc
...ernel/gpu/quant/fake_quant_per_channel_grad_gpu_kernel.cc
+1
-1
mindspore/ccsrc/kernel/gpu/quant/fake_quant_per_channel_grad_gpu_kernel.h
...kernel/gpu/quant/fake_quant_per_channel_grad_gpu_kernel.h
+1
-1
mindspore/ccsrc/kernel/hccl/hcom_all_broadcast.cc
mindspore/ccsrc/kernel/hccl/hcom_all_broadcast.cc
+4
-4
mindspore/ccsrc/kernel/hccl/hcom_all_broadcast.h
mindspore/ccsrc/kernel/hccl/hcom_all_broadcast.h
+1
-1
mindspore/ccsrc/kernel/hccl/hcom_all_gather.cc
mindspore/ccsrc/kernel/hccl/hcom_all_gather.cc
+3
-4
mindspore/ccsrc/kernel/hccl/hcom_all_gather.h
mindspore/ccsrc/kernel/hccl/hcom_all_gather.h
+1
-1
mindspore/ccsrc/kernel/hccl/hcom_all_reduce.cc
mindspore/ccsrc/kernel/hccl/hcom_all_reduce.cc
+3
-4
mindspore/ccsrc/kernel/hccl/hcom_all_reduce.h
mindspore/ccsrc/kernel/hccl/hcom_all_reduce.h
+1
-1
mindspore/ccsrc/kernel/hccl/hcom_all_reduce_scatter.cc
mindspore/ccsrc/kernel/hccl/hcom_all_reduce_scatter.cc
+4
-4
mindspore/ccsrc/kernel/hccl/hcom_all_reduce_scatter.h
mindspore/ccsrc/kernel/hccl/hcom_all_reduce_scatter.h
+1
-1
mindspore/ccsrc/kernel/kernel.h
mindspore/ccsrc/kernel/kernel.h
+1
-1
mindspore/ccsrc/kernel/rts/assign.cc
mindspore/ccsrc/kernel/rts/assign.cc
+3
-5
mindspore/ccsrc/kernel/rts/assign.h
mindspore/ccsrc/kernel/rts/assign.h
+1
-1
mindspore/ccsrc/kernel/rts/label_goto.cc
mindspore/ccsrc/kernel/rts/label_goto.cc
+2
-2
mindspore/ccsrc/kernel/rts/label_goto.h
mindspore/ccsrc/kernel/rts/label_goto.h
+1
-1
mindspore/ccsrc/kernel/rts/label_set.cc
mindspore/ccsrc/kernel/rts/label_set.cc
+2
-2
mindspore/ccsrc/kernel/rts/label_set.h
mindspore/ccsrc/kernel/rts/label_set.h
+1
-1
mindspore/ccsrc/kernel/rts/label_switch.cc
mindspore/ccsrc/kernel/rts/label_switch.cc
+3
-2
mindspore/ccsrc/kernel/rts/label_switch.h
mindspore/ccsrc/kernel/rts/label_switch.h
+1
-1
mindspore/ccsrc/kernel/rts/memcpy_async.cc
mindspore/ccsrc/kernel/rts/memcpy_async.cc
+2
-4
mindspore/ccsrc/kernel/rts/memcpy_async.h
mindspore/ccsrc/kernel/rts/memcpy_async.h
+1
-1
mindspore/ccsrc/kernel/rts/profiling_kernel_mod.cc
mindspore/ccsrc/kernel/rts/profiling_kernel_mod.cc
+3
-4
mindspore/ccsrc/kernel/rts/profiling_kernel_mod.h
mindspore/ccsrc/kernel/rts/profiling_kernel_mod.h
+1
-1
mindspore/ccsrc/kernel/rts/recv.cc
mindspore/ccsrc/kernel/rts/recv.cc
+2
-3
mindspore/ccsrc/kernel/rts/recv.h
mindspore/ccsrc/kernel/rts/recv.h
+1
-1
mindspore/ccsrc/kernel/rts/send.cc
mindspore/ccsrc/kernel/rts/send.cc
+2
-3
mindspore/ccsrc/kernel/rts/send.h
mindspore/ccsrc/kernel/rts/send.h
+1
-1
mindspore/ccsrc/kernel/rts/stream_active.cc
mindspore/ccsrc/kernel/rts/stream_active.cc
+2
-3
mindspore/ccsrc/kernel/rts/stream_active.h
mindspore/ccsrc/kernel/rts/stream_active.h
+1
-1
mindspore/ccsrc/kernel/rts/stream_switch.cc
mindspore/ccsrc/kernel/rts/stream_switch.cc
+2
-3
mindspore/ccsrc/kernel/rts/stream_switch.h
mindspore/ccsrc/kernel/rts/stream_switch.h
+1
-1
mindspore/ccsrc/kernel/tbe/tbe_kernel_mod.cc
mindspore/ccsrc/kernel/tbe/tbe_kernel_mod.cc
+3
-4
mindspore/ccsrc/kernel/tbe/tbe_kernel_mod.h
mindspore/ccsrc/kernel/tbe/tbe_kernel_mod.h
+1
-1
未找到文件。
mindspore/ccsrc/device/gpu/gpu_kernel_runtime.cc
浏览文件 @
c964916e
...
...
@@ -199,7 +199,7 @@ bool GPUKernelRuntime::LaunchKernelDynamic(const session::KernelGraph *graph) {
AddressPtrList
kernel_workspaces
;
AddressPtrList
kernel_outputs
;
AllocKernelDynamicRes
(
*
kernel_mod
,
kernel
,
&
kernel_inputs
,
&
kernel_workspaces
,
&
kernel_outputs
);
if
(
!
kernel_mod
->
Launch
(
kernel_inputs
,
kernel_workspaces
,
kernel_outputs
,
reinterpret_cast
<
uintptr_t
>
(
stream_
)
))
{
if
(
!
kernel_mod
->
Launch
(
kernel_inputs
,
kernel_workspaces
,
kernel_outputs
,
stream_
))
{
MS_LOG
(
ERROR
)
<<
"Launch kernel failed."
;
return
false
;
}
...
...
mindspore/ccsrc/device/kernel_runtime.cc
浏览文件 @
c964916e
...
...
@@ -664,8 +664,7 @@ bool KernelRuntime::LaunchKernelMod(const session::KernelGraph &graph) {
struct
timeval
start_time
,
end_time
;
(
void
)
gettimeofday
(
&
start_time
,
nullptr
);
#endif
auto
ret
=
kernel_mod
->
Launch
(
kernel_inputs
,
kernel_workspaces
,
kernel_outputs
,
reinterpret_cast
<
uintptr_t
>
(
stream_
));
auto
ret
=
kernel_mod
->
Launch
(
kernel_inputs
,
kernel_workspaces
,
kernel_outputs
,
stream_
);
if
(
!
ret
)
{
MS_LOG
(
ERROR
)
<<
"Launch kernel failed."
;
return
false
;
...
...
mindspore/ccsrc/kernel/aicpu/aicpu_kernel_mod.cc
浏览文件 @
c964916e
...
...
@@ -103,14 +103,13 @@ void AicpuOpKernelMod::CreateCpuKernelInfo(const std::vector<AddressPtr> &inputs
}
bool
AicpuOpKernelMod
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
if
(
stream_ptr
==
0
)
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
{
if
(
stream_ptr
==
nullptr
)
{
MS_LOG
(
ERROR
)
<<
"stream_ptr should not be nullptr."
;
return
false
;
}
CreateCpuKernelInfo
(
inputs
,
outputs
);
auto
*
stream
=
reinterpret_cast
<
rtStream_t
*>
(
stream_ptr
);
if
(
node_name_
==
"TopK"
)
{
node_name_
=
"TopKV2"
;
}
...
...
@@ -119,7 +118,7 @@ bool AicpuOpKernelMod::Launch(const std::vector<AddressPtr> &inputs, const std::
if
(
rtCpuKernelLaunch
(
reinterpret_cast
<
const
void
*>
(
node_so_
.
c_str
()),
reinterpret_cast
<
const
void
*>
(
node_name_
.
c_str
()),
1
,
reinterpret_cast
<
const
void
*>
(
args_
.
data
()),
static_cast
<
uint32_t
>
(
args_
.
length
()),
nullptr
,
stream
)
!=
RT_ERROR_NONE
)
{
stream
_ptr
)
!=
RT_ERROR_NONE
)
{
MS_LOG
(
ERROR
)
<<
"Aicpu op launch failed!"
;
return
false
;
...
...
mindspore/ccsrc/kernel/aicpu/aicpu_kernel_mod.h
浏览文件 @
c964916e
...
...
@@ -27,7 +27,7 @@ class AicpuOpKernelMod : public AscendKernelMod {
AicpuOpKernelMod
();
~
AicpuOpKernelMod
()
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
std
::
vector
<
TaskInfoPtr
>
GenTask
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uint32_t
stream_id
)
override
;
...
...
mindspore/ccsrc/kernel/akg/gpu/akg_gpu_kernel_mod.cc
浏览文件 @
c964916e
...
...
@@ -82,7 +82,7 @@ const std::vector<size_t> &GpuKernelMod::GetOutputSizeList() const { return outp
const
std
::
vector
<
size_t
>
&
GpuKernelMod
::
GetWorkspaceSizeList
()
const
{
return
workspace_size_list_
;
}
bool
GpuKernelMod
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
{
if
(
stream_ptr
==
0
)
{
MS_LOG
(
ERROR
)
<<
"stream_ptr should not be nullptr."
;
return
false
;
...
...
mindspore/ccsrc/kernel/akg/gpu/akg_gpu_kernel_mod.h
浏览文件 @
c964916e
...
...
@@ -64,7 +64,7 @@ class GpuKernelMod : public KernelMod {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
;
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
static
GpuKernelManagerPtr
kernelmanager_
;
...
...
mindspore/ccsrc/kernel/cpu/cpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -56,7 +56,7 @@ class CPUKernel : public kernel::KernelMod {
void
Init
(
const
CNodePtr
&
kernel_node
);
virtual
void
InitKernel
(
const
CNodePtr
&
kernel_node
)
=
0
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
/*stream_ptr*/
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
/*stream_ptr*/
)
override
{
return
Launch
(
inputs
,
workspace
,
outputs
);
};
virtual
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
...
...
mindspore/ccsrc/kernel/gpu/arrays/argmax_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -35,7 +35,7 @@ class ArgmaxGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
input
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
int
*
output
=
GetDeviceAddress
<
int
>
(
outputs
,
0
);
CalArgmax
(
input
,
SizeToInt
(
batch_size_
),
SizeToInt
(
channel_size_
),
axis_
,
output
,
...
...
mindspore/ccsrc/kernel/gpu/arrays/array_reduce_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -55,7 +55,7 @@ class ArrayReduceGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/arrays/concatv2_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -34,7 +34,7 @@ class ConcatV2GpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
if
(
inputs
.
size
()
==
2
)
{
T
*
input_0
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
input_1
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
...
...
mindspore/ccsrc/kernel/gpu/arrays/gather_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -35,7 +35,7 @@ class GatherGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
VARIABLE_NOT_USED
(
workspace
);
T
*
input_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
S
*
indices_addr
=
GetDeviceAddress
<
S
>
(
inputs
,
1
);
...
...
mindspore/ccsrc/kernel/gpu/arrays/one_hot_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -34,7 +34,7 @@ class OneHotGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
{
return
output_size_list_
;
}
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
VARIABLE_NOT_USED
(
workspace
);
const
S
*
indices
=
GetDeviceAddress
<
S
>
(
inputs
,
0
);
const
T
*
on_value
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
...
...
mindspore/ccsrc/kernel/gpu/arrays/select_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -34,7 +34,7 @@ class SelectGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
bool
*
input_cond
=
GetDeviceAddress
<
bool
>
(
inputs
,
0
);
T
*
input_x
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
T
*
input_y
=
GetDeviceAddress
<
T
>
(
inputs
,
2
);
...
...
mindspore/ccsrc/kernel/gpu/arrays/slice_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -34,7 +34,7 @@ class SliceGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
input
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
output
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
if
(
is_strided_slice_
)
{
...
...
mindspore/ccsrc/kernel/gpu/arrays/slice_grad_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -34,7 +34,7 @@ class SliceGradGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
dy
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
dx
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
FillDeviceArray
(
outputs
[
0
]
->
size
/
sizeof
(
T
),
dx
,
0.
f
,
reinterpret_cast
<
cudaStream_t
>
(
stream_ptr
));
...
...
mindspore/ccsrc/kernel/gpu/arrays/transpose_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -34,7 +34,7 @@ class TransposeGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
input
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
output
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
int
*
input_shape
=
GetDeviceAddress
<
int
>
(
workspace
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/arrays/unsorted_segment_sum_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -35,7 +35,7 @@ class UnsortedSegmentSumGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
input_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
S
*
indices_addr
=
GetDeviceAddress
<
S
>
(
inputs
,
1
);
T
*
output_addr
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/control/recv_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -33,7 +33,7 @@ class RecvGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
,
uintptr_t
)
override
{
void
*
)
override
{
CHECK_CUDA_RET_WITH_EXCEPT
(
cudaStreamWaitEvent
(
wait_stream_
,
wait_event_
,
0
),
"Waiting cuda event failed."
);
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/control/send_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -33,7 +33,7 @@ class SendGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
,
uintptr_t
)
override
{
void
*
)
override
{
CHECK_CUDA_RET_WITH_EXCEPT
(
cudaEventRecord
(
record_event_
,
record_stream_
),
"Recording cuda event failed."
);
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/data/dataset_init_kernel.cc
浏览文件 @
c964916e
...
...
@@ -53,7 +53,7 @@ bool DatasetInitKernel::Init(const CNodePtr &kernel_node) {
void
DatasetInitKernel
::
InitSizeLists
()
{
return
;
}
bool
DatasetInitKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
,
uintptr_t
)
{
const
std
::
vector
<
AddressPtr
>
&
,
void
*
)
{
void
*
addr
=
nullptr
;
size_t
len
=
total_bytes_
*
buffer_q_capacity_
;
...
...
mindspore/ccsrc/kernel/gpu/data/dataset_init_kernel.h
浏览文件 @
c964916e
...
...
@@ -33,7 +33,7 @@ class DatasetInitKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
;
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
bool
Init
(
const
CNodePtr
&
kernel_node
)
override
;
protected:
...
...
mindspore/ccsrc/kernel/gpu/data/dataset_iterator_kernel.cc
浏览文件 @
c964916e
...
...
@@ -64,7 +64,7 @@ bool DatasetIteratorKernel::Init(const CNodePtr &kernel_node) {
void
DatasetIteratorKernel
::
InitSizeLists
()
{
return
;
}
bool
DatasetIteratorKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
)
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
)
{
void
*
addr
=
nullptr
;
size_t
len
=
0
;
...
...
mindspore/ccsrc/kernel/gpu/data/dataset_iterator_kernel.h
浏览文件 @
c964916e
...
...
@@ -33,7 +33,7 @@ class DatasetIteratorKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
;
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
bool
Init
(
const
CNodePtr
&
kernel_node
)
override
;
protected:
...
...
mindspore/ccsrc/kernel/gpu/math/addn_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -43,7 +43,7 @@ class AddNGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/math/assign_add_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -35,7 +35,7 @@ class AssignAddGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/math/bias_add_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -42,7 +42,7 @@ class BiasAddGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
{
return
output_size_list_
;
}
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
VARIABLE_NOT_USED
(
workspace
);
VARIABLE_NOT_USED
(
stream_ptr
);
T
*
x_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/math/broadcast_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -39,7 +39,7 @@ class BroadcastOpGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
lhs
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
rhs
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
S
*
output
=
GetDeviceAddress
<
S
>
(
outputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/math/broadcast_grad_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -39,7 +39,7 @@ class BroadcastOpGradGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
x1
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
x2
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
T
*
dy
=
GetDeviceAddress
<
T
>
(
inputs
,
2
);
...
...
mindspore/ccsrc/kernel/gpu/math/equalcount_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -35,7 +35,7 @@ class EqualCountGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
VARIABLE_NOT_USED
(
workspace
);
T
*
input1
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
input2
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
...
...
mindspore/ccsrc/kernel/gpu/math/float_status_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -40,7 +40,7 @@ class FloatStatusGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
input
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
switch
(
kernel_name_
)
{
...
...
mindspore/ccsrc/kernel/gpu/math/matmul_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -48,7 +48,7 @@ class MatMulGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
VARIABLE_NOT_USED
(
workspace
);
VARIABLE_NOT_USED
(
stream_ptr
);
auto
input1_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/math/tensoradd_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -43,7 +43,7 @@ class TensorAddGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
)
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
)
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/math/unary_op_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -62,7 +62,7 @@ class UnaryOpGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
VARIABLE_NOT_USED
(
workspace
);
T
*
input_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
output_addr
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/nccl/nccl_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -60,7 +60,7 @@ class NcclGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
{
return
output_size_list_
;
}
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
input_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
output_addr
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/nn/bias_add_grad_gpu_kenel.h
浏览文件 @
c964916e
...
...
@@ -45,7 +45,7 @@ class BiasAddGradGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
{
return
output_size_list_
;
}
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
dy_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
db_addr
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
T
*
indices_addr
=
GetDeviceAddress
<
T
>
(
workspace
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/nn/conv2d_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -60,7 +60,7 @@ class Conv2dGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/nn/conv2d_grad_filter_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -61,7 +61,7 @@ class ConvGradFilterGpuBkwKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/nn/conv2d_grad_input_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -61,7 +61,7 @@ class ConvGradInputGpuBkwKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/nn/dropout_gpu_kernel.cc
浏览文件 @
c964916e
...
...
@@ -76,7 +76,7 @@ void DropoutGpuFwdKernel::InitSizeLists() {
}
bool
DropoutGpuFwdKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/nn/dropout_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -37,7 +37,7 @@ class DropoutGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
bool
Init
(
const
CNodePtr
&
kernel_node
)
override
;
...
...
mindspore/ccsrc/kernel/gpu/nn/dropout_grad_kernel.cc
浏览文件 @
c964916e
...
...
@@ -75,7 +75,7 @@ void DropoutGradGpuFwdKernel::InitSizeLists() {
}
bool
DropoutGradGpuFwdKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/nn/dropout_grad_kernel.h
浏览文件 @
c964916e
...
...
@@ -32,7 +32,7 @@ class DropoutGradGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
;
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
bool
Init
(
const
CNodePtr
&
kernel_node
)
override
;
protected:
...
...
mindspore/ccsrc/kernel/gpu/nn/flatten_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -35,7 +35,7 @@ class FlattenGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
input
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
output
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
cudaError_t
ret
=
...
...
mindspore/ccsrc/kernel/gpu/nn/flatten_grad_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -35,7 +35,7 @@ class FlattenGardGpuBkwKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
VARIABLE_NOT_USED
(
workspace
);
T
*
input
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
output
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/nn/fused_batch_norm_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -49,7 +49,7 @@ class FusedBatchNormGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
VARIABLE_NOT_USED
(
workspace
);
VARIABLE_NOT_USED
(
stream_ptr
);
if
(
is_null_input_
)
{
...
...
mindspore/ccsrc/kernel/gpu/nn/fused_batchnorm_grad_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -47,7 +47,7 @@ class FusedBatchNormGradGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
{
return
output_size_list_
;
}
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
VARIABLE_NOT_USED
(
workspace
);
VARIABLE_NOT_USED
(
stream_ptr
);
if
(
is_null_input_
)
{
...
...
mindspore/ccsrc/kernel/gpu/nn/gelu_grad_kernel.h
浏览文件 @
c964916e
...
...
@@ -35,7 +35,7 @@ class GeLUGpuGradKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
dy_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
x_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
T
*
dx_addr
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/nn/gelu_kernel.h
浏览文件 @
c964916e
...
...
@@ -35,7 +35,7 @@ class GeluGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
input_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
output_addr
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/nn/layer_norm_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -35,7 +35,7 @@ class LayerNormGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
auto
x
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
auto
gamma
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
auto
beta
=
GetDeviceAddress
<
T
>
(
inputs
,
2
);
...
...
mindspore/ccsrc/kernel/gpu/nn/layer_norm_grad_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -35,7 +35,7 @@ class LayerNormGradGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
auto
x
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
auto
dy
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
auto
var
=
GetDeviceAddress
<
T
>
(
inputs
,
2
);
...
...
mindspore/ccsrc/kernel/gpu/nn/lstm_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -59,7 +59,7 @@ class LstmGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
VARIABLE_NOT_USED
(
stream_ptr
);
auto
x_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
auto
hx_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
...
...
mindspore/ccsrc/kernel/gpu/nn/lstm_grad_data_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -61,7 +61,7 @@ class LstmGradDataGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
{
return
output_size_list_
;
}
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
VARIABLE_NOT_USED
(
stream_ptr
);
auto
y_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
auto
dy_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
...
...
mindspore/ccsrc/kernel/gpu/nn/lstm_grad_weight_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -54,7 +54,7 @@ class LstmGradWeightGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
{
return
output_size_list_
;
}
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
VARIABLE_NOT_USED
(
stream_ptr
);
auto
x_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
auto
hx_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
...
...
mindspore/ccsrc/kernel/gpu/nn/momentum_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -34,7 +34,7 @@ class MomentumGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
,
uintptr_t
stream_ptr
)
override
{
void
*
stream_ptr
)
override
{
T
*
variable
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
accumulation
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
T
*
learning_rate
=
GetDeviceAddress
<
T
>
(
inputs
,
2
);
...
...
mindspore/ccsrc/kernel/gpu/nn/pooling_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -59,7 +59,7 @@ class PoolingGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
{
return
output_size_list_
;
}
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/nn/pooling_grad_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -61,7 +61,7 @@ class PoolingGradGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
{
return
output_size_list_
;
}
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/nn/relu_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -43,7 +43,7 @@ class ReLUGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/nn/relu_grad_kernel.h
浏览文件 @
c964916e
...
...
@@ -41,7 +41,7 @@ class ReluGradGpuFwdKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/nn/rmsprop_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -35,7 +35,7 @@ class RMSPropGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream
)
override
{
if
(
!
use_center_
)
{
T
*
variable
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
mean_square
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
...
...
mindspore/ccsrc/kernel/gpu/nn/softmax_cross_entropy_with_logits_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -52,7 +52,7 @@ class SoftmaxCrossEntropyWithLogitsGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
{
return
output_size_list_
;
}
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/nn/softmax_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -52,7 +52,7 @@ class SoftmaxGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/nn/softmax_grad_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -51,7 +51,7 @@ class SoftmaxGradGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/nn/sparse_softmax_cross_entropy_with_logits_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -52,7 +52,7 @@ class SparseSoftmaxCrossEntropyWithLogitsGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
{
return
output_size_list_
;
}
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/nn/tanh_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -37,7 +37,7 @@ class TanhGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
auto
x_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
auto
y_addr
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/nn/tanh_grad_kernel.h
浏览文件 @
c964916e
...
...
@@ -37,7 +37,7 @@ class TanhGradKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
auto
y_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
auto
dy_addr
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
auto
dx_addr
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/other/assign_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -33,7 +33,7 @@ class AssignGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
T
*
var
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
T
*
value
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
T
*
output
=
GetDeviceAddress
<
T
>
(
outputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/quant/batchnorm_fold2_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -45,7 +45,7 @@ class BatchNormFold2GpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/quant/batchnorm_fold2_grad_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -45,7 +45,7 @@ class BatchNormFold2GradGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
if
(
is_null_input_
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/gpu/quant/batchnorm_fold_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -53,7 +53,7 @@ class BatchNormFoldGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
(
void
)
workspace
;
auto
x
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
auto
mean
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
...
...
mindspore/ccsrc/kernel/gpu/quant/batchnorm_fold_grad_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -47,7 +47,7 @@ class BatchNormFoldGradGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
(
void
)
workspace
;
// 'd_batch_mean', 'd_batch_std', 'x', 'batch_mean', 'batch_std', 'current_step'
T
*
d_batch_mean
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/quant/correction_mul_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -34,7 +34,7 @@ class CorrectionMulGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
{
return
output_size_list_
;
}
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
auto
*
weight
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
auto
*
gamma
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
auto
*
running_std
=
GetDeviceAddress
<
T
>
(
inputs
,
2
);
...
...
mindspore/ccsrc/kernel/gpu/quant/correction_mul_grad_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -35,7 +35,7 @@ class CorrectionMulGradGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
{
return
workspace_size_list_
;
}
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
{
auto
*
d_out
=
GetDeviceAddress
<
T
>
(
inputs
,
0
);
auto
*
weight
=
GetDeviceAddress
<
T
>
(
inputs
,
1
);
auto
*
gamma
=
GetDeviceAddress
<
T
>
(
inputs
,
2
);
...
...
mindspore/ccsrc/kernel/gpu/quant/fake_quant_gpu_kernel.cc
浏览文件 @
c964916e
...
...
@@ -114,7 +114,7 @@ void FakeQuantGpuKernel::InitSizeLists() {
}
bool
FakeQuantGpuKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
{
float
*
output
=
GetDeviceAddress
<
float
>
(
outputs
,
0
);
float
*
input
=
GetDeviceAddress
<
float
>
(
inputs
,
0
);
float
*
input_min
=
GetDeviceAddress
<
float
>
(
inputs
,
1
);
...
...
mindspore/ccsrc/kernel/gpu/quant/fake_quant_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -32,7 +32,7 @@ class FakeQuantGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
;
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
bool
Init
(
const
CNodePtr
&
kernel
)
override
;
protected:
...
...
mindspore/ccsrc/kernel/gpu/quant/fake_quant_grad_gpu_kernel.cc
浏览文件 @
c964916e
...
...
@@ -92,7 +92,7 @@ void FakeQuantGradGpuKernel::InitSizeLists() {
}
bool
FakeQuantGradGpuKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
{
float
*
output
=
GetDeviceAddress
<
float
>
(
outputs
,
0
);
float
*
gradient
=
GetDeviceAddress
<
float
>
(
inputs
,
0
);
float
*
input
=
GetDeviceAddress
<
float
>
(
inputs
,
1
);
...
...
mindspore/ccsrc/kernel/gpu/quant/fake_quant_grad_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -32,7 +32,7 @@ class FakeQuantGradGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
;
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
bool
Init
(
const
CNodePtr
&
kernel_node
)
override
;
protected:
...
...
mindspore/ccsrc/kernel/gpu/quant/fake_quant_per_channel_gpu_kernel.cc
浏览文件 @
c964916e
...
...
@@ -118,7 +118,7 @@ void FakeQuantPerChannelGpuKernel::InitSizeLists() {
void
FakeQuantPerChannelGpuKernel
::
CalFakeQuantizeForTraining
(
float
*
input
,
float
*
output
,
float
*
input_min
,
float
*
input_max
,
float
*
d_nudge_min
,
float
*
d_nudge_max
,
float
*
d_scale
,
uintptr_t
stream_ptr
)
{
float
*
d_scale
,
void
*
stream_ptr
)
{
// calculate the input min and max according by the parameter ema and ema_decay.
CalMinMaxPerChannel
(
input
,
input_min
,
input_max
,
input_size_
/
sizeof
(
float
),
channel_out_
,
ema_decay_
,
ema_
,
reinterpret_cast
<
cudaStream_t
>
(
stream_ptr
));
...
...
@@ -139,7 +139,7 @@ void FakeQuantPerChannelGpuKernel::CalFakeQuantizeForTraining(float *input, floa
void
FakeQuantPerChannelGpuKernel
::
CalFakeQuantizeForInfer
(
float
*
input
,
float
*
output
,
float
*
input_min
,
float
*
input_max
,
float
*
d_nudge_min
,
float
*
d_nudge_max
,
float
*
d_scale
,
uintptr_t
stream_ptr
)
{
float
*
d_scale
,
void
*
stream_ptr
)
{
// real launch
CalNudgePerChannel
(
input_min
,
input_max
,
quant_min_
,
quant_max_
,
d_nudge_min
,
d_nudge_max
,
d_scale
,
channel_out_
,
reinterpret_cast
<
cudaStream_t
>
(
stream_ptr
));
...
...
@@ -149,7 +149,7 @@ void FakeQuantPerChannelGpuKernel::CalFakeQuantizeForInfer(float *input, float *
bool
FakeQuantPerChannelGpuKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
{
(
void
)
workspace
;
float
*
output
=
GetDeviceAddress
<
float
>
(
outputs
,
0
);
float
*
input
=
GetDeviceAddress
<
float
>
(
inputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/quant/fake_quant_per_channel_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -32,7 +32,7 @@ class FakeQuantPerChannelGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
;
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
bool
Init
(
const
CNodePtr
&
kernel
)
override
;
protected:
...
...
@@ -40,9 +40,9 @@ class FakeQuantPerChannelGpuKernel : public GpuKernel {
private:
void
CalFakeQuantizeForTraining
(
float
*
input
,
float
*
output
,
float
*
input_min
,
float
*
input_max
,
float
*
d_nudge_min
,
float
*
d_nudge_max
,
float
*
d_scale
,
uintptr_t
stream_ptr
);
float
*
d_nudge_max
,
float
*
d_scale
,
void
*
stream_ptr
);
void
CalFakeQuantizeForInfer
(
float
*
input
,
float
*
output
,
float
*
input_min
,
float
*
input_max
,
float
*
d_nudge_min
,
float
*
d_nudge_max
,
float
*
d_scale
,
uintptr_t
stream_ptr
);
float
*
d_nudge_max
,
float
*
d_scale
,
void
*
stream_ptr
);
size_t
input_size_
;
size_t
min_size_
;
...
...
mindspore/ccsrc/kernel/gpu/quant/fake_quant_per_channel_grad_gpu_kernel.cc
浏览文件 @
c964916e
...
...
@@ -104,7 +104,7 @@ void FakeQuantPerChannelGradGpuKernel::InitSizeLists() {
bool
FakeQuantPerChannelGradGpuKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
{
(
void
)
workspace
;
float
*
output
=
GetDeviceAddress
<
float
>
(
outputs
,
0
);
float
*
gradient
=
GetDeviceAddress
<
float
>
(
inputs
,
0
);
...
...
mindspore/ccsrc/kernel/gpu/quant/fake_quant_per_channel_grad_gpu_kernel.h
浏览文件 @
c964916e
...
...
@@ -32,7 +32,7 @@ class FakeQuantPerChannelGradGpuKernel : public GpuKernel {
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
override
;
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
bool
Init
(
const
CNodePtr
&
kernel_node
)
override
;
protected:
...
...
mindspore/ccsrc/kernel/hccl/hcom_all_broadcast.cc
浏览文件 @
c964916e
...
...
@@ -24,17 +24,17 @@
namespace
mindspore
{
namespace
kernel
{
bool
HcomAllBroadCastKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
bool
HcomAllBroadCastKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
/*workspace*/
,
const
std
::
vector
<
AddressPtr
>
&
/*outputs*/
,
void
*
stream_ptr
)
{
auto
context_ptr
=
MsContext
::
GetInstance
();
MS_EXCEPTION_IF_NULL
(
context_ptr
);
if
(
context_ptr
->
enable_task_sink
())
{
return
true
;
}
const
char
*
tag
=
"Hccl-BroadCast"
;
auto
stream
=
reinterpret_cast
<
rtStream_t
>
(
stream_ptr
);
hcclResult_t
ret
=
hcom_broadcast
(
tag
,
inputs
[
0
]
->
addr
,
hccl_count_
,
hccl_data_type_list_
[
0
],
root_id_
,
nullptr
,
stream
);
hcom_broadcast
(
tag
,
inputs
[
0
]
->
addr
,
hccl_count_
,
hccl_data_type_list_
[
0
],
root_id_
,
nullptr
,
stream
_ptr
);
if
(
ret
!=
HCCL_SUCCESS
)
{
MS_LOG
(
ERROR
)
<<
"HcomBroadcastOp : hcom_broadcast fail, return: "
<<
static_cast
<
int
>
(
ret
);
return
false
;
...
...
mindspore/ccsrc/kernel/hccl/hcom_all_broadcast.h
浏览文件 @
c964916e
...
...
@@ -31,7 +31,7 @@ class HcomAllBroadCastKernel : public HcclKernel {
/* Inherit from kernelmod */
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
private:
};
...
...
mindspore/ccsrc/kernel/hccl/hcom_all_gather.cc
浏览文件 @
c964916e
...
...
@@ -24,17 +24,16 @@
namespace
mindspore
{
namespace
kernel
{
bool
HcomAllGatherKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
bool
HcomAllGatherKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
/*workspace*/
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
{
auto
context_ptr
=
MsContext
::
GetInstance
();
MS_EXCEPTION_IF_NULL
(
context_ptr
);
if
(
context_ptr
->
enable_task_sink
())
{
return
true
;
}
const
char
*
tag
=
"Hccl-AllGather"
;
auto
stream
=
reinterpret_cast
<
rtStream_t
>
(
stream_ptr
);
hcclResult_t
ret
=
hcom_all_gather
(
tag
,
inputs
[
0
]
->
addr
,
outputs
[
0
]
->
addr
,
hccl_count_
,
hccl_data_type_list_
[
0
],
nullptr
,
stream
);
hcom_all_gather
(
tag
,
inputs
[
0
]
->
addr
,
outputs
[
0
]
->
addr
,
hccl_count_
,
hccl_data_type_list_
[
0
],
nullptr
,
stream
_ptr
);
if
(
ret
!=
HCCL_SUCCESS
)
{
MS_LOG
(
ERROR
)
<<
"HcomAllGatherKernelOp : hcom_all_gather fail, return: "
<<
static_cast
<
int
>
(
ret
);
return
false
;
...
...
mindspore/ccsrc/kernel/hccl/hcom_all_gather.h
浏览文件 @
c964916e
...
...
@@ -31,7 +31,7 @@ class HcomAllGatherKernel : public HcclKernel {
/* Inherit from kernelmod */
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
private:
};
...
...
mindspore/ccsrc/kernel/hccl/hcom_all_reduce.cc
浏览文件 @
c964916e
...
...
@@ -24,17 +24,16 @@
namespace
mindspore
{
namespace
kernel
{
bool
HcomAllReduceKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
bool
HcomAllReduceKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
/*workspace*/
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
{
auto
context_ptr
=
MsContext
::
GetInstance
();
MS_EXCEPTION_IF_NULL
(
context_ptr
);
if
(
context_ptr
->
enable_task_sink
())
{
return
true
;
}
const
char
*
tag
=
"Hccl-AllReduce"
;
auto
stream
=
reinterpret_cast
<
rtStream_t
>
(
stream_ptr
);
hcclResult_t
ret
=
hcom_all_reduce
(
tag
,
inputs
[
0
]
->
addr
,
outputs
[
0
]
->
addr
,
hccl_count_
,
hccl_data_type_list_
[
0
],
op_type_
,
nullptr
,
stream
);
op_type_
,
nullptr
,
stream
_ptr
);
if
(
ret
!=
HCCL_SUCCESS
)
{
MS_LOG
(
ERROR
)
<<
"HcomAllReduceKernelOp : hcom_all_reduce fail, return: "
<<
static_cast
<
int
>
(
ret
);
return
false
;
...
...
mindspore/ccsrc/kernel/hccl/hcom_all_reduce.h
浏览文件 @
c964916e
...
...
@@ -30,7 +30,7 @@ class HcomAllReduceKernel : public HcclKernel {
/* Inherit from kernelmod */
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
private:
};
...
...
mindspore/ccsrc/kernel/hccl/hcom_all_reduce_scatter.cc
浏览文件 @
c964916e
...
...
@@ -24,17 +24,17 @@
namespace
mindspore
{
namespace
kernel
{
bool
HcomAllReduceScatterKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
bool
HcomAllReduceScatterKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
/*workspace*/
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
{
auto
context_ptr
=
MsContext
::
GetInstance
();
MS_EXCEPTION_IF_NULL
(
context_ptr
);
if
(
context_ptr
->
enable_task_sink
())
{
return
true
;
}
const
char
*
tag
=
"Hccl-ReduceScatter"
;
auto
stream
=
reinterpret_cast
<
rtStream_t
>
(
stream_ptr
);
hcclResult_t
ret
=
hcom_reduce_scatter
(
tag
,
inputs
[
0
]
->
addr
,
outputs
[
0
]
->
addr
,
hccl_count_
,
hccl_data_type_list_
[
0
],
op_type_
,
nullptr
,
stream
);
op_type_
,
nullptr
,
stream
_ptr
);
if
(
ret
!=
HCCL_SUCCESS
)
{
MS_LOG
(
ERROR
)
<<
"HcomReduceScatterOp : hcom_reduce_scatter fail, return: "
<<
static_cast
<
int
>
(
ret
);
return
false
;
...
...
mindspore/ccsrc/kernel/hccl/hcom_all_reduce_scatter.h
浏览文件 @
c964916e
...
...
@@ -31,7 +31,7 @@ class HcomAllReduceScatterKernel : public HcclKernel {
/* Inherit from kernelmod */
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
private:
};
...
...
mindspore/ccsrc/kernel/kernel.h
浏览文件 @
c964916e
...
...
@@ -124,7 +124,7 @@ class KernelMod {
virtual
const
std
::
vector
<
size_t
>
&
GetOutputSizeList
()
const
=
0
;
virtual
const
std
::
vector
<
size_t
>
&
GetWorkspaceSizeList
()
const
=
0
;
virtual
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
=
0
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
=
0
;
virtual
std
::
vector
<
size_t
>
GenParameters
()
{
return
{};
}
virtual
~
KernelMod
()
=
default
;
...
...
mindspore/ccsrc/kernel/rts/assign.cc
浏览文件 @
c964916e
...
...
@@ -30,10 +30,8 @@ AssignKernel::AssignKernel() {}
AssignKernel
::~
AssignKernel
()
{}
bool
AssignKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
auto
stream
=
reinterpret_cast
<
rtStream_t
>
(
stream_ptr
);
bool
AssignKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
/*workspace*/
,
const
std
::
vector
<
AddressPtr
>
&
/*outputs*/
,
void
*
stream_ptr
)
{
if
(
inputs
.
size
()
!=
2
)
{
MS_LOG
(
ERROR
)
<<
"inputs size is not two"
;
return
false
;
...
...
@@ -44,7 +42,7 @@ bool AssignKernel::Launch(const std::vector<AddressPtr> &inputs, const std::vect
return
true
;
}
rtError_t
status
=
rtMemcpyAsync
(
inputs
[
0
]
->
addr
,
inputs
[
0
]
->
size
,
inputs
[
1
]
->
addr
,
inputs
[
1
]
->
size
,
RT_MEMCPY_DEVICE_TO_DEVICE
,
stream
);
RT_MEMCPY_DEVICE_TO_DEVICE
,
stream
_ptr
);
if
(
status
!=
RT_ERROR_NONE
)
{
MS_LOG
(
ERROR
)
<<
"Assign op rtMemcpyAsync failed!"
;
return
false
;
...
...
mindspore/ccsrc/kernel/rts/assign.h
浏览文件 @
c964916e
...
...
@@ -29,7 +29,7 @@ class AssignKernel : public RtKernel {
~
AssignKernel
()
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
std
::
vector
<
TaskInfoPtr
>
GenTask
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uint32_t
stream_id
)
override
;
};
...
...
mindspore/ccsrc/kernel/rts/label_goto.cc
浏览文件 @
c964916e
...
...
@@ -45,8 +45,8 @@ bool LabelGotoKernel::Init(const AnfNodePtr &anf_node) {
return
true
;
}
bool
LabelGotoKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
bool
LabelGotoKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
/*inputs*/
,
const
std
::
vector
<
AddressPtr
>
&
/*workspace*/
,
const
std
::
vector
<
AddressPtr
>
&
/*outputs*/
,
void
*
/*stream_ptr*/
)
{
MS_LOG
(
INFO
)
<<
"LabelGotoKernel launch"
;
return
true
;
}
...
...
mindspore/ccsrc/kernel/rts/label_goto.h
浏览文件 @
c964916e
...
...
@@ -32,7 +32,7 @@ class LabelGotoKernel : public RtKernel {
bool
Init
(
const
AnfNodePtr
&
anf_node
)
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
std
::
vector
<
TaskInfoPtr
>
GenTask
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uint32_t
stream_id
)
override
;
...
...
mindspore/ccsrc/kernel/rts/label_set.cc
浏览文件 @
c964916e
...
...
@@ -45,8 +45,8 @@ bool LabelSetKernel::Init(const AnfNodePtr &anf_node) {
return
true
;
}
bool
LabelSetKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
bool
LabelSetKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
/*inputs*/
,
const
std
::
vector
<
AddressPtr
>
&
/*workspace*/
,
const
std
::
vector
<
AddressPtr
>
&
/*outputs*/
,
void
*
/*stream_ptr*/
)
{
MS_LOG
(
INFO
)
<<
"LabelSetKernel launch"
;
return
true
;
}
...
...
mindspore/ccsrc/kernel/rts/label_set.h
浏览文件 @
c964916e
...
...
@@ -32,7 +32,7 @@ class LabelSetKernel : public RtKernel {
bool
Init
(
const
AnfNodePtr
&
anf_node
)
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
std
::
vector
<
TaskInfoPtr
>
GenTask
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uint32_t
stream_id
)
override
;
...
...
mindspore/ccsrc/kernel/rts/label_switch.cc
浏览文件 @
c964916e
...
...
@@ -53,8 +53,9 @@ bool LabelSwitchKernel::Init(const AnfNodePtr &anf_node) {
return
true
;
}
bool
LabelSwitchKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
bool
LabelSwitchKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
/*inputs*/
,
const
std
::
vector
<
AddressPtr
>
&
/*workspace*/
,
const
std
::
vector
<
AddressPtr
>
&
/*outputs*/
,
void
*
/*stream_ptr*/
)
{
MS_LOG
(
INFO
)
<<
"LabelSwitchKernel launch"
;
return
true
;
}
...
...
mindspore/ccsrc/kernel/rts/label_switch.h
浏览文件 @
c964916e
...
...
@@ -32,7 +32,7 @@ class LabelSwitchKernel : public RtKernel {
bool
Init
(
const
AnfNodePtr
&
anf_node
)
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
std
::
vector
<
TaskInfoPtr
>
GenTask
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uint32_t
stream_id
)
override
;
...
...
mindspore/ccsrc/kernel/rts/memcpy_async.cc
浏览文件 @
c964916e
...
...
@@ -34,9 +34,7 @@ MemCpyAsyncKernel::MemCpyAsyncKernel() {}
MemCpyAsyncKernel
::~
MemCpyAsyncKernel
()
{}
bool
MemCpyAsyncKernel
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
/*workspace*/
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
auto
stream
=
reinterpret_cast
<
rtStream_t
>
(
stream_ptr
);
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
{
if
(
inputs
.
size
()
!=
1
)
{
MS_LOG
(
ERROR
)
<<
"inputs size is not one"
;
return
false
;
...
...
@@ -51,7 +49,7 @@ bool MemCpyAsyncKernel::Launch(const std::vector<AddressPtr> &inputs, const std:
return
true
;
}
rtError_t
status
=
rtMemcpyAsync
(
outputs
[
0
]
->
addr
,
outputs
[
0
]
->
size
,
inputs
[
0
]
->
addr
,
inputs
[
0
]
->
size
,
RT_MEMCPY_DEVICE_TO_DEVICE
,
stream
);
RT_MEMCPY_DEVICE_TO_DEVICE
,
stream
_ptr
);
if
(
status
!=
RT_ERROR_NONE
)
{
MS_LOG
(
ERROR
)
<<
"MemCpyAsync op rtMemcpyAsync failed!"
;
return
false
;
...
...
mindspore/ccsrc/kernel/rts/memcpy_async.h
浏览文件 @
c964916e
...
...
@@ -31,7 +31,7 @@ class MemCpyAsyncKernel : public RtKernel {
bool
Init
(
const
AnfNodePtr
&
anf_node
)
override
;
bool
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
override
;
const
std
::
vector
<
AddressPtr
>
&
outputs
,
void
*
stream_ptr
)
override
;
std
::
vector
<
TaskInfoPtr
>
GenTask
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uint32_t
stream_id
)
override
;
...
...
mindspore/ccsrc/kernel/rts/profiling_kernel_mod.cc
浏览文件 @
c964916e
...
...
@@ -50,10 +50,9 @@ bool ProfilingKernelMod::Init(const AnfNodePtr &anf_node) {
return
true
;
}
bool
ProfilingKernelMod
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
inputs
,
const
std
::
vector
<
AddressPtr
>
&
workspace
,
const
std
::
vector
<
AddressPtr
>
&
outputs
,
uintptr_t
stream_ptr
)
{
MS_LOG
(
INFO
)
<<
"gen task inputs size:"
<<
inputs
.
size
()
<<
", workspace size:"
<<
workspace
.
size
()
<<
", outputs size:"
<<
outputs
.
size
()
<<
", stream_ptr:"
<<
stream_ptr
;
bool
ProfilingKernelMod
::
Launch
(
const
std
::
vector
<
AddressPtr
>
&
/*inputs*/
,
const
std
::
vector
<
AddressPtr
>
&
/*workspace*/
,
const
std
::
vector
<
AddressPtr
>
&
/*outputs*/
,
void
*
/*stream_ptr*/
)
{
return
true
;
}
...
...
mindspore/ccsrc/kernel/rts/profiling_kernel_mod.h
浏览文件 @
c964916e
此差异已折叠。
点击以展开。
mindspore/ccsrc/kernel/rts/recv.cc
浏览文件 @
c964916e
此差异已折叠。
点击以展开。
mindspore/ccsrc/kernel/rts/recv.h
浏览文件 @
c964916e
此差异已折叠。
点击以展开。
mindspore/ccsrc/kernel/rts/send.cc
浏览文件 @
c964916e
此差异已折叠。
点击以展开。
mindspore/ccsrc/kernel/rts/send.h
浏览文件 @
c964916e
此差异已折叠。
点击以展开。
mindspore/ccsrc/kernel/rts/stream_active.cc
浏览文件 @
c964916e
此差异已折叠。
点击以展开。
mindspore/ccsrc/kernel/rts/stream_active.h
浏览文件 @
c964916e
此差异已折叠。
点击以展开。
mindspore/ccsrc/kernel/rts/stream_switch.cc
浏览文件 @
c964916e
此差异已折叠。
点击以展开。
mindspore/ccsrc/kernel/rts/stream_switch.h
浏览文件 @
c964916e
此差异已折叠。
点击以展开。
mindspore/ccsrc/kernel/tbe/tbe_kernel_mod.cc
浏览文件 @
c964916e
此差异已折叠。
点击以展开。
mindspore/ccsrc/kernel/tbe/tbe_kernel_mod.h
浏览文件 @
c964916e
此差异已折叠。
点击以展开。
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录