p04896132
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

!1175 LayerNormGrad fix & codex 

Merge pull request !1175 from chenweifeng/layernorm
This commit is contained in:
mindspore-ci-bot 2020-05-15 14:55:36 +08:00 committed by Gitee
parent 42ba885e58 1991a89f40
commit dbac31e787
8 changed files with 12 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
mindspore/ccsrc/kernel/gpu/arrays/unsorted_segment_sum_gpu_kernel.cc
View File

@ -37,6 +37,5 @@ MS_REG_GPU_KERNEL_TWO(
UnsortedSegmentSum,
KernelAttr().AddInputAttr(kNumberTypeInt32).AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt32),
UnsortedSegmentSumGpuKernel, int, int64_t)
} // namespace kernel
} // namespace mindspore

2
mindspore/ccsrc/kernel/gpu/data/dataset_utils.cc
View File

@ -18,7 +18,6 @@
namespace mindspore {
namespace kernel {
size_t UnitSizeInBytes(const mindspore::TypeId &t) {
size_t bytes = 0;
switch (t) {
@ -65,6 +64,5 @@ int ElementNums(const std::vector<int> &shape) {
return nums;
}
} // namespace kernel
} // namespace mindspore

1
mindspore/ccsrc/kernel/gpu/data/dataset_utils.h
View File

@ -21,7 +21,6 @@
#include "ir/dtype/type.h"
namespace mindspore {
namespace kernel {
size_t UnitSizeInBytes(const mindspore::TypeId &t);
int ElementNums(const std::vector<int> &shape);
} // namespace kernel

13
mindspore/ccsrc/kernel/gpu/math/broadcast_gpu_kernel.h
View File

@ -27,7 +27,6 @@
#include "kernel/gpu/kernel_constants.h"
namespace mindspore {
namespace kernel {
template <typename T, typename S>
class BroadcastOpGpuKernel : public GpuKernel {
public:
@ -70,14 +69,14 @@ class BroadcastOpGpuKernel : public GpuKernel {
output_num_ *= shape3[i];
}
int offset = shape3.size() - shape1.size();
for (size_t i = 0; i < shape1.size(); i++) {
lhs_shape_[i + offset] = shape1[i];
input1_num_ *= shape1[i];
for (size_t j = 0; j < shape1.size(); j++) {
lhs_shape_[j + offset] = shape1[j];
input1_num_ *= shape1[j];
}
offset = shape3.size() - shape2.size();
for (size_t i = 0; i < shape2.size(); i++) {
rhs_shape_[i + offset] = shape2[i];
input2_num_ *= shape2[i];
for (size_t k = 0; k < shape2.size(); k++) {
rhs_shape_[k + offset] = shape2[k];
input2_num_ *= shape2[k];
}
InitSizeLists();

2
mindspore/ccsrc/kernel/gpu/math/broadcast_grad_gpu_kernel.h
View File

@ -1,4 +1,3 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
@ -28,7 +27,6 @@
#include "kernel/gpu/kernel_constants.h"
namespace mindspore {
namespace kernel {
template <typename T>
class BroadcastOpGradGpuKernel : public GpuKernel {
public:

4
mindspore/ccsrc/kernel/gpu/nn/layer_norm_grad_gpu_kernel.h
View File

@ -36,8 +36,8 @@ class LayerNormGradGpuKernel : public GpuKernel {
bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &,
const std::vector<AddressPtr> &outputs, uintptr_t stream_ptr) override {
auto dy = GetDeviceAddress<T>(inputs, 0);
auto x = GetDeviceAddress<T>(inputs, 1);
auto x = GetDeviceAddress<T>(inputs, 0);
auto dy = GetDeviceAddress<T>(inputs, 1);
auto var = GetDeviceAddress<T>(inputs, 2);
auto mean = GetDeviceAddress<T>(inputs, 3);
auto gamma = GetDeviceAddress<T>(inputs, 4);

1
mindspore/ccsrc/kernel/gpu/nn/rmsprop_gpu_kernel.cc
View File

@ -44,6 +44,5 @@ MS_REG_GPU_KERNEL_ONE(ApplyCenteredRMSProp,
.AddInputAttr(kNumberTypeFloat32)
.AddOutputAttr(kNumberTypeFloat32),
RMSPropGpuKernel, float)
} // namespace kernel
} // namespace mindspore

8
tests/st/ops/gpu/test_layer_norm_grad_op.py
View File

@ -80,7 +80,7 @@ def test_layernormgrad0():
gamma_ms = Tensor(gamma_np)
net = LayerNormGradNet(begin_norm_axis, begin_params_axis)
dx_ms, dg_ms, db_ms = net(dy_ms, x_ms, var_ms, mean_ms, gamma_ms)
dx_ms, dg_ms, db_ms = net(x_ms, dy_ms, var_ms, mean_ms, gamma_ms)
assert np.allclose(dx_ms.asnumpy(), dx_np, rtol=1e-6, atol=1e-6)
assert np.allclose(dg_ms.asnumpy(), dg_np, rtol=1e-6, atol=1e-3)
@ -107,7 +107,7 @@ def test_layernormgrad1():
gamma_ms = Tensor(gamma_np)
net = LayerNormGradNet(begin_norm_axis, begin_params_axis)
dx_ms, dg_ms, db_ms = net(dy_ms, x_ms, var_ms, mean_ms, gamma_ms)
dx_ms, dg_ms, db_ms = net(x_ms, dy_ms, var_ms, mean_ms, gamma_ms)
assert np.allclose(dx_ms.asnumpy(), dx_np, rtol=1e-6, atol=1e-6)
assert np.allclose(dg_ms.asnumpy(), dg_np, rtol=1e-6, atol=1e-3)
@ -133,8 +133,8 @@ def test_layernormgrad2():
gamma_ms = Tensor(gamma_np)
net = LayerNormGradNet(begin_norm_axis, begin_params_axis)
dx_ms, dg_ms, db_ms = net(dy_ms, x_ms, var_ms, mean_ms, gamma_ms)
dx_ms, dg_ms, db_ms = net(x_ms, dy_ms, var_ms, mean_ms, gamma_ms)
assert np.allclose(dx_ms.asnumpy(), dx_np, rtol=1e-6, atol=1e-6)
assert np.allclose(dg_ms.asnumpy(), dg_np, rtol=1e-6, atol=1e-3)
assert np.allclose(db_ms.asnumpy(), db_np, rtol=1e-6, atol=1e-3)
assert np.allclose(db_ms.asnumpy(), db_np, rtol=1e-6, atol=1e-3)