Skip to content

Commit 3504d89

Browse files
zhangting2020zhangting2020
authored
Merge branch 'release/2.3' into release_2.3_cp
2 parents 68fdaf8 + b4adbe5 commit 3504d89

File tree

114 files changed

+3727
-1857
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

114 files changed

+3727
-1857
lines changed

‎cmake/external/xpu.cmake

Lines changed: 16 additions & 32 deletions
Original file line numberDiff line numberDiff line change
@@ -7,62 +7,46 @@ SET(XPU_PROJECT "extern_xpu")
77
SET(XPU_API_LIB_NAME "libxpuapi.so")
88
SET(XPU_RT_LIB_NAME "libxpurt.so")
99

10-
if(NOT DEFINED XPU_BASE_URL)
11-
SET(XPU_BASE_URL_WITHOUT_DATE "https://baidu-kunlun-product.cdn.bcebos.com/KL-SDK/klsdk-dev")
12-
SET(XPU_BASE_URL "${XPU_BASE_URL_WITHOUT_DATE}/20220411")
13-
else()
14-
SET(XPU_BASE_URL "${XPU_BASE_URL}")
15-
endif()
16-
17-
# ubuntu and centos: use output by XDNN API team
18-
if(NOT DEFINED XPU_XDNN_BASE_URL)
19-
SET(XPU_XDNN_BASE_URL_WITHOUT_DATE "https://klx-sdk-release-public.su.bcebos.com/xdnn/dev")
20-
SET(XPU_XDNN_BASE_URL "${XPU_XDNN_BASE_URL_WITHOUT_DATE}/20220412")
21-
else()
22-
SET(XPU_XDNN_BASE_URL "${XPU_XDNN_BASE_URL}")
23-
endif()
24-
2510
IF(WITH_AARCH64)
2611
SET(XPU_XRE_DIR_NAME "xre-kylin_aarch64")
27-
SET(XPU_XDNN_DIR_NAME "XDNN-kylin_aarch64")
12+
SET(XPU_XDNN_DIR_NAME "xdnn-kylin_aarch64")
2813
SET(XPU_XCCL_DIR_NAME "xccl-kylin_aarch64")
29-
SET(XPU_XDNN_URL "${XPU_XDNN_BASE_URL}/${XPU_XDNN_DIR_NAME}.tar.gz" CACHE STRING "" FORCE)
3014
ELSEIF(WITH_SUNWAY)
3115
SET(XPU_XRE_DIR_NAME "xre-deepin_sw6_64")
3216
SET(XPU_XDNN_DIR_NAME "xdnn-deepin_sw6_64")
3317
SET(XPU_XCCL_DIR_NAME "xccl-deepin_sw6_64")
34-
SET(XPU_XDNN_URL "${XPU_BASE_URL}/${XPU_XDNN_DIR_NAME}.tar.gz" CACHE STRING "" FORCE)
3518
ELSEIF(WITH_BDCENTOS)
3619
SET(XPU_XRE_DIR_NAME "xre-bdcentos_x86_64")
37-
SET(XPU_XDNN_DIR_NAME "XDNN-bdcentos_x86_64")
20+
SET(XPU_XDNN_DIR_NAME "xdnn-bdcentos_x86_64")
3821
SET(XPU_XCCL_DIR_NAME "xccl-bdcentos_x86_64")
39-
# ubuntu and centos: use output by XDNN API team
40-
SET(XPU_XDNN_URL "${XPU_XDNN_BASE_URL}/${XPU_XDNN_DIR_NAME}.tar.gz" CACHE STRING "" FORCE)
4122
ELSEIF(WITH_UBUNTU)
4223
SET(XPU_XRE_DIR_NAME "xre-ubuntu_x86_64")
43-
SET(XPU_XDNN_DIR_NAME "XDNN-ubuntu_x86_64")
24+
SET(XPU_XDNN_DIR_NAME "xdnn-ubuntu_x86_64")
4425
SET(XPU_XCCL_DIR_NAME "xccl-bdcentos_x86_64")
45-
# ubuntu and centos: use output by XDNN API team
46-
SET(XPU_XDNN_URL "${XPU_XDNN_BASE_URL}/${XPU_XDNN_DIR_NAME}.tar.gz" CACHE STRING "" FORCE)
4726
ELSEIF(WITH_CENTOS)
4827
SET(XPU_XRE_DIR_NAME "xre-centos7_x86_64")
49-
SET(XPU_XDNN_DIR_NAME "XDNN-bdcentos_x86_64")
28+
SET(XPU_XDNN_DIR_NAME "xdnn-centos7_x86_64")
5029
SET(XPU_XCCL_DIR_NAME "xccl-bdcentos_x86_64")
51-
# ubuntu and centos: use output by XDNN API team
52-
SET(XPU_XDNN_URL "${XPU_XDNN_BASE_URL}/${XPU_XDNN_DIR_NAME}.tar.gz" CACHE STRING "" FORCE)
53-
ELSE()
30+
31+
ELSE ()
5432
SET(XPU_XRE_DIR_NAME "xre-ubuntu_x86_64")
55-
SET(XPU_XDNN_DIR_NAME "XDNN-ubuntu_x86_64")
33+
SET(XPU_XDNN_DIR_NAME "xdnn-ubuntu_x86_64")
5634
SET(XPU_XCCL_DIR_NAME "xccl-bdcentos_x86_64")
57-
# default: use output by XDNN API team
58-
SET(XPU_XDNN_URL "${XPU_XDNN_BASE_URL}/${XPU_XDNN_DIR_NAME}.tar.gz" CACHE STRING "" FORCE)
5935
ENDIF()
6036

37+
if(NOT DEFINED XPU_BASE_URL)
38+
SET(XPU_BASE_URL_WITHOUT_DATE "https://baidu-kunlun-product.cdn.bcebos.com/KL-SDK/klsdk-dev")
39+
SET(XPU_BASE_URL "${XPU_BASE_URL_WITHOUT_DATE}/20220411")
40+
else()
41+
SET(XPU_BASE_URL "${XPU_BASE_URL}")
42+
endif()
43+
6144
SET(XPU_XRE_URL "${XPU_BASE_URL}/${XPU_XRE_DIR_NAME}.tar.gz" CACHE STRING "" FORCE)
45+
SET(XPU_XDNN_URL "${XPU_BASE_URL}/${XPU_XDNN_DIR_NAME}.tar.gz" CACHE STRING "" FORCE)
6246
SET(XPU_XCCL_URL "${XPU_BASE_URL_WITHOUT_DATE}/20220411/${XPU_XCCL_DIR_NAME}.tar.gz" CACHE STRING "" FORCE)
6347
SET(XPU_PACK_DEPENCE_URL "https://baidu-kunlun-public.su.bcebos.com/paddle_depence/pack_paddle_depence.sh" CACHE STRING "" FORCE)
6448

65-
SET(SNAPPY_PREFIX_DIR "${THIRD_PARTY_PATH}/xpu")
49+
SET(SNAPPY_PREFIX_DIR "${THIRD_PARTY_PATH}/xpu")
6650
SET(XPU_DOWNLOAD_DIR "${SNAPPY_PREFIX_DIR}/src/${XPU_PROJECT}")
6751
SET(XPU_INSTALL_DIR "${THIRD_PARTY_PATH}/install/xpu")
6852
SET(XPU_INC_DIR "${THIRD_PARTY_PATH}/install/xpu/include")

‎paddle/fluid/distributed/collective/Common.cc

Lines changed: 6 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -17,7 +17,7 @@
1717
namespace paddle {
1818
namespace distributed {
1919

20-
std::vector<Place> GetPlaceList(const std::vector<Tensor>& tensors) {
20+
std::vector<Place> GetPlaceList(const std::vector<phi::DenseTensor>& tensors) {
2121
std::vector<Place> places;
2222
places.reserve(tensors.size());
2323
for (auto& tensor : tensors) {
@@ -40,15 +40,11 @@ std::string GetKeyFromPlaces(const std::vector<Place>& places) {
4040
return placeList;
4141
}
4242

43-
static bool CheckTensorsInPlace(const std::vector<Tensor>& tensors,
44-
phi::AllocationType type) {
45-
return std::all_of(tensors.cbegin(), tensors.cend(), [&](const Tensor& t) {
46-
return t.place().GetType() == type;
47-
});
48-
}
49-
50-
bool CheckTensorsInCudaPlace(const std::vector<Tensor>& tensors) {
51-
return CheckTensorsInPlace(tensors, phi::AllocationType::GPU);
43+
bool CheckTensorsInCudaPlace(const std::vector<phi::DenseTensor>& tensors) {
44+
return std::all_of(tensors.cbegin(), tensors.cend(),
45+
[&](const phi::DenseTensor& t) {
46+
return platform::is_gpu_place(t.place());
47+
});
5248
}
5349

5450
} // namespace distributed

‎paddle/fluid/distributed/collective/Common.h

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -16,18 +16,18 @@
1616

1717
#include "paddle/fluid/platform/place.h"
1818
#include "paddle/phi/api/include/api.h"
19+
#include "paddle/phi/common/place.h"
20+
#include "paddle/phi/core/dense_tensor.h"
1921
namespace paddle {
2022
namespace distributed {
2123

22-
using Tensor = paddle::experimental::Tensor;
23-
2424
using Place = paddle::platform::Place;
2525
// Get the list of devices from list of tensors
26-
std::vector<Place> GetPlaceList(const std::vector<Tensor>& tensors);
26+
std::vector<Place> GetPlaceList(const std::vector<phi::DenseTensor>& tensors);
2727
// Get the deviceList String from the list of devices
2828
std::string GetKeyFromPlaces(const std::vector<Place>& places);
2929

30-
bool CheckTensorsInCudaPlace(const std::vector<Tensor>& tensors);
30+
bool CheckTensorsInCudaPlace(const std::vector<phi::DenseTensor>& tensors);
3131

3232
} // namespace distributed
3333
} // namespace paddle

‎paddle/fluid/distributed/collective/ProcessGroup.cc

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -17,7 +17,8 @@
1717
namespace paddle {
1818
namespace distributed {
1919

20-
ProcessGroup::Task::Task(int rank, const std::vector<Tensor>& inputTensors,
20+
ProcessGroup::Task::Task(int rank,
21+
const std::vector<phi::DenseTensor>& inputTensors,
2122
CommType comm_type)
2223
: rank_(rank), comm_type_(comm_type) {}
2324

‎paddle/fluid/distributed/collective/ProcessGroup.h

Lines changed: 17 additions & 20 deletions
Original file line numberDiff line numberDiff line change
@@ -54,7 +54,7 @@ class ProcessGroup {
5454
public:
5555
class Task {
5656
public:
57-
Task(int rank, const std::vector<Tensor>& inputTensors,
57+
Task(int rank, const std::vector<phi::DenseTensor>& inputTensors,
5858
CommType opType = CommType::UNKNOWN);
5959

6060
virtual ~Task();
@@ -79,68 +79,65 @@ class ProcessGroup {
7979
virtual const std::string GetBackendName() const = 0;
8080

8181
virtual std::shared_ptr<ProcessGroup::Task> AllReduce(
82-
std::vector<Tensor>& /* tensors */,
82+
std::vector<phi::DenseTensor>& /* input tensors */, // NOLINT
83+
std::vector<phi::DenseTensor>& /* output tensors */, // NOLINT
8384
const AllreduceOptions& = AllreduceOptions()) {
8485
PADDLE_THROW(platform::errors::InvalidArgument(
8586
"ProcessGroup%s does not support allreduce", GetBackendName()));
8687
}
8788

8889
virtual std::shared_ptr<ProcessGroup::Task> Broadcast(
89-
std::vector<Tensor>& /* tensors */,
90+
std::vector<phi::DenseTensor>& /* input tensors */, // NOLINT
91+
std::vector<phi::DenseTensor>& /* output tensors */, // NOLINT
9092
const BroadcastOptions& = BroadcastOptions()) {
9193
PADDLE_THROW(platform::errors::InvalidArgument(
9294
"ProcessGroup%s does not support broadcast", GetBackendName()));
9395
}
9496

95-
virtual void Broadcast(const phi::DenseTensor* in, phi::DenseTensor* out) {
96-
PADDLE_THROW(platform::errors::Fatal(
97-
"ProcessGroup%s does not support broadcast for static mode runtime",
98-
GetBackendName()));
99-
}
100-
10197
virtual std::shared_ptr<ProcessGroup::Task> Barrier(
10298
const BarrierOptions& = BarrierOptions()) {
10399
PADDLE_THROW(platform::errors::InvalidArgument(
104100
"ProcessGroup%s does not support barrier", GetBackendName()));
105101
}
106102

107103
virtual std::shared_ptr<ProcessGroup::Task> Send(
108-
std::vector<Tensor>& tensors /* tensors */, int dst_rank) { // NOLINT
104+
std::vector<phi::DenseTensor>&, int) { // NOLINT
109105
PADDLE_THROW(platform::errors::InvalidArgument(
110106
"ProcessGroup%s does not support send", GetBackendName()));
111107
}
112108

113109
virtual std::shared_ptr<ProcessGroup::Task> Recv(
114-
std::vector<Tensor>& tensors /* tensors */, int src_rank) { // NOLINT
110+
std::vector<phi::DenseTensor>& tensors, int) { // NOLINT
115111
PADDLE_THROW(platform::errors::InvalidArgument(
116112
"ProcessGroup%s does not support receive", GetBackendName()));
117113
}
118114

119115
virtual std::shared_ptr<ProcessGroup::Task> AllGather(
120-
std::vector<Tensor>& in_tensors /* tensors */, // NOLINT
121-
std::vector<Tensor>& out_tensors /* tensors */) { // NOLINT
116+
std::vector<phi::DenseTensor>&, // NOLINT
117+
std::vector<phi::DenseTensor>&) { // NOLINT
122118
PADDLE_THROW(platform::errors::InvalidArgument(
123119
"ProcessGroup%s does not support AllGather", GetBackendName()));
124120
}
125121

126122
virtual std::shared_ptr<ProcessGroup::Task> AllToAll(
127-
std::vector<Tensor>& in /* tensors */, // NOLINT
128-
std::vector<Tensor>& out /* tensors */) { // NOLINT
123+
std::vector<phi::DenseTensor>&, // NOLINT
124+
std::vector<phi::DenseTensor>&) { // NOLINT
129125
PADDLE_THROW(platform::errors::InvalidArgument(
130126
"ProcessGroup%s does not support AllToAll", GetBackendName()));
131127
}
132128

133129
virtual std::shared_ptr<ProcessGroup::Task> Reduce(
134-
std::vector<Tensor>& tensors /* tensors */, // NOLINT
135-
const ReduceOptions& opts) { // NOLINT
130+
std::vector<phi::DenseTensor>&, // NOLINT
131+
std::vector<phi::DenseTensor>&, // NOLINT
132+
const ReduceOptions& opts) {
136133
PADDLE_THROW(platform::errors::InvalidArgument(
137134
"ProcessGroup%s does not support Reduce", GetBackendName()));
138135
}
139136

140137
virtual std::shared_ptr<ProcessGroup::Task> Scatter(
141-
std::vector<Tensor>& in_tensors /* tensors */, // NOLINT
142-
std::vector<Tensor>& out_tensors /* tensors */, // NOLINT
143-
const ScatterOptions&) { // NOLINT
138+
std::vector<phi::DenseTensor>&, // NOLINT
139+
std::vector<phi::DenseTensor>&, // NOLINT
140+
const ScatterOptions&) { // NOLINT
144141
PADDLE_THROW(platform::errors::InvalidArgument(
145142
"ProcessGroup%s does not support Scatter", GetBackendName()));
146143
}

0 commit comments

Comments
 (0)