Skip to content

[XPU] add bfloat16 supports for compare_kernel and add reduce_all_kernel #63602

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
QingshuChen merged 1 commit into from
Apr 18, 2024
Merged

[XPU] add bfloat16 supports for compare_kernel and add reduce_all_kernel #63602

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions paddle/phi/backends/xpu/xpu2_op_list.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -741,6 +741,7 @@ XPUOpMap& get_kl2_ops() {
{"reciprocal", XPUKernelSet({phi::DataType::FLOAT32})},
{"reciprocal_grad",
XPUKernelSet({phi::DataType::FLOAT32, phi::DataType::FLOAT16})},
{"reduce_all", XPUKernelSet({phi::DataType::BOOL})},
{"reduce_any", XPUKernelSet({phi::DataType::BOOL})},
{"reduce_max_grad", XPUKernelSet({phi::DataType::FLOAT32})},
{"reduce_max",
Expand Down
7 changes: 7 additions & 0 deletions paddle/phi/backends/xpu/xpu3_op_list.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -349,6 +349,7 @@ XPUOpMap& get_kl3_ops() {
XPUKernelSet({phi::DataType::INT64,
phi::DataType::INT32,
phi::DataType::FLOAT16,
phi::DataType::BFLOAT16,
phi::DataType::FLOAT32,
phi::DataType::BOOL})},
{"exp_grad", XPUKernelSet({phi::DataType::FLOAT32})},
Expand Down Expand Up @@ -517,11 +518,13 @@ XPUOpMap& get_kl3_ops() {
XPUKernelSet({phi::DataType::INT64,
phi::DataType::INT32,
phi::DataType::FLOAT16,
phi::DataType::BFLOAT16,
phi::DataType::FLOAT32})},
{"greater_than",
XPUKernelSet({phi::DataType::INT64,
phi::DataType::INT32,
phi::DataType::FLOAT16,
phi::DataType::BFLOAT16,
phi::DataType::FLOAT32})},
{"grid_sampler_grad", XPUKernelSet({phi::DataType::FLOAT32})},
{"group_norm_silu_xpu",
Expand Down Expand Up @@ -576,11 +579,13 @@ XPUOpMap& get_kl3_ops() {
XPUKernelSet({phi::DataType::INT64,
phi::DataType::INT32,
phi::DataType::FLOAT16,
phi::DataType::BFLOAT16,
phi::DataType::FLOAT32})},
{"less_than",
XPUKernelSet({phi::DataType::INT64,
phi::DataType::INT32,
phi::DataType::FLOAT16,
phi::DataType::BFLOAT16,
phi::DataType::FLOAT32})},
{"load", XPUKernelSet({phi::DataType::FLOAT32})},
{"load_combine",
Expand Down Expand Up @@ -669,6 +674,7 @@ XPUOpMap& get_kl3_ops() {
XPUKernelSet({phi::DataType::INT64,
phi::DataType::INT32,
phi::DataType::FLOAT16,
phi::DataType::BFLOAT16,
phi::DataType::FLOAT32})},
{"one_hot", XPUKernelSet({phi::DataType::INT32, phi::DataType::INT64})},
{"one_hot_v2",
Expand Down Expand Up @@ -716,6 +722,7 @@ XPUOpMap& get_kl3_ops() {
{"reciprocal", XPUKernelSet({phi::DataType::FLOAT32})},
{"reciprocal_grad",
XPUKernelSet({phi::DataType::FLOAT32, phi::DataType::FLOAT16})},
{"reduce_all", XPUKernelSet({phi::DataType::BOOL})},
{"reduce_any", XPUKernelSet({phi::DataType::BOOL})},
{"reduce_max_grad", XPUKernelSet({phi::DataType::FLOAT32})},
{"reduce_max",
Expand Down
4 changes: 4 additions & 0 deletions paddle/phi/kernels/reduce_all_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,3 +53,7 @@ PD_REGISTER_KERNEL(
#if defined(PADDLE_WITH_XPU_KP)
PD_REGISTER_KERNEL(all, KPS, ALL_LAYOUT, phi::AllKernel, bool) {}
#endif

#if defined(PADDLE_WITH_XPU)
PD_REGISTER_KERNEL(all, XPU, ALL_LAYOUT, phi::AllKernel, bool) {}
#endif
4 changes: 3 additions & 1 deletion paddle/phi/kernels/xpu/compare_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,7 +88,8 @@ PD_REGISTER_KERNEL(less_than,
int,
int64_t,
float,
phi::dtype::float16) {
phi::dtype::float16,
phi::dtype::bfloat16) {
kernel->OutputAt(0).SetDataType(phi::DataType::BOOL);
}

Expand All @@ -101,6 +102,7 @@ PD_REGISTER_KERNEL(less_than,
int64_t, \
float, \
phi::dtype::float16, \
phi::dtype::bfloat16, \
bool) { \
kernel->OutputAt(0).SetDataType(phi::DataType::BOOL); \
}
Expand Down
51 changes: 51 additions & 0 deletions paddle/phi/kernels/xpu/reduce_all_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,51 @@
// Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#include "paddle/phi/kernels/reduce_all_kernel.h"

#include "paddle/phi/backends/xpu/enforce_xpu.h"
#include "paddle/phi/backends/xpu/xpu_context.h"
#include "paddle/phi/core/kernel_registry.h"
#include "paddle/phi/kernels/xpu/reduce.h"

namespace phi {

template <typename T, typename Context>
void AllRawKernel(const Context& dev_ctx,
const DenseTensor& x,
const std::vector<int64_t>& dims,
bool keep_dim,
bool reduce_all,
DenseTensor* out) {
reduce_all = recompute_reduce_all(x, dims);
using XPUType = typename XPUTypeTrait<T>::Type;
auto f = [](xpu::Context* ctx,
const T* x,
T* y,
const std::vector<int>& xdims,
const std::vector<int>& reduce_dims) {
return xpu::reduce_all<XPUType>(ctx,
reinterpret_cast<const XPUType*>(x),
reinterpret_cast<XPUType*>(y),
xdims,
reduce_dims);
};

int r = XPUReduce<Context, T>(dev_ctx, x, dims, keep_dim, reduce_all, out, f);
PADDLE_ENFORCE_XDNN_SUCCESS(r, "reduce_all");
}

} // namespace phi

PD_REGISTER_KERNEL(all_raw, XPU, ALL_LAYOUT, phi::AllRawKernel, bool) {}
30 changes: 20 additions & 10 deletions test/xpu/test_reduce_all_op_xpu.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,16 +40,20 @@ def set_case(self):
self.op_type = 'reduce_all'
self.attrs = {
'use_xpu': True,
'reduce_all': True,
'keep_dim': True,
'reduce_all': False,
'keep_dim': False,
'dim': (3, 5, 4),
}
self.inputs = {
'X': np.random.randint(0, 2, (2, 5, 3, 2, 2, 3, 4, 2)).astype(
"bool"
)
}
self.outputs = {'Out': self.inputs['X'].all(axis=self.attrs['dim'])}
self.outputs = {
'Out': self.inputs['X'].all(
axis=self.attrs['dim'], keepdims=self.attrs['keep_dim']
)
}

def test_check_output(self):
self.check_output_with_place(self.place)
Expand All @@ -63,7 +67,7 @@ def set_case(self):
self.attrs = {
'use_xpu': True,
'reduce_all': True,
'keep_dim': True,
'keep_dim': False,
'dim': [1],
}
self.inputs = {
Expand All @@ -76,31 +80,37 @@ def set_case(self):
self.op_type = 'reduce_all'
self.attrs = {
'use_xpu': True,
'reduce_all': True,
'keep_dim': False,
'reduce_all': False,
'keep_dim': True,
'dim': (3, 6),
}
self.inputs = {
'X': np.random.randint(0, 2, (2, 5, 3, 2, 2, 3, 4, 2)).astype(
"bool"
)
}
self.outputs = {'Out': self.inputs['X'].all(axis=self.attrs['dim'])}
self.outputs = {
'Out': self.inputs['X'].all(
axis=self.attrs['dim'], keepdims=self.attrs['keep_dim']
)
}

class XPUTestReduceAllCase3(XPUTestReduceAllBase):
def set_case(self):
self.op_type = 'reduce_all'
self.attrs = {
'use_xpu': True,
'reduce_all': True,
'keep_dim': True,
'dim': [1]
# 'reduce_all': True,
'dim': [1],
}
self.inputs = {
'X': np.random.randint(0, 2, (5, 6, 10)).astype("bool")
}
self.outputs = {
'Out': np.expand_dims(self.inputs['X'].all(axis=1), axis=1)
'Out': self.inputs['X'].all(
axis=(0, 1, 2), keepdims=self.attrs['keep_dim']
)
}


Expand Down