m53297601
/
Paddle
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

fix cpplint error for the autmic max/min

Browse Source 
fix cpplint error for the autmic max/min
revert-27356-init_low_level_gloo
Zhong Hui 5 years ago committed by GitHub
parent ecfdfc9c58
commit a85592bcbf
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 20 additions and 21 deletions
Show Stats Download Patch File Download Diff File

17
paddle/fluid/operators/math/segment_pooling.cu
Unescape View File

@ -12,13 +12,12 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and See the License for the specific language governing permissions and
limitations under the License. */ limitations under the License. */
#include "paddle/fluid/operators/elementwise/elementwise_div_op.h" #include <algorithm>
#include "paddle/fluid/operators/gather.cu.h" #include "paddle/fluid/operators/gather.cu.h"
#include "paddle/fluid/operators/math/math_function.h" #include "paddle/fluid/operators/math/math_function.h"
#include "paddle/fluid/operators/math/segment_pooling.h" #include "paddle/fluid/operators/math/segment_pooling.h"
#include "paddle/fluid/platform/cuda_primitives.h" #include "paddle/fluid/platform/cuda_primitives.h"
#include "paddle/fluid/platform/gpu_launch_param_config.h" #include "paddle/fluid/platform/gpu_launch_param_config.h"
#include "paddle/fluid/platform/macros.h"
namespace paddle { namespace paddle {
namespace operators { namespace operators {
@ -100,7 +99,7 @@ __global__ void SegmentOpsKernel(const Index* segment_ids, const T* input,
CUDA_KERNEL_LOOP(stripe_index, h.total_stripe_count) { CUDA_KERNEL_LOOP(stripe_index, h.total_stripe_count) {
Index segment_offset, dim_index_base, actual_height; Index segment_offset, dim_index_base, actual_height;
Index inner_dim_size = h.inner_dim_size; Index inner_dim_size = h.inner_dim_size;
h.calculate(stripe_index, segment_offset, dim_index_base, actual_height); h.calculate(stripe_index, &segment_offset, &dim_index_base, &actual_height);
T minmax = pool.initial(); T minmax = pool.initial();
Index first_segment_id = segment_ids[dim_index_base]; Index first_segment_id = segment_ids[dim_index_base];
@ -154,7 +153,7 @@ __global__ void SegmentIndexGradKernel(const Index* segment_ids, const T* input,
T* in_grad, Helper h) { T* in_grad, Helper h) {
CUDA_KERNEL_LOOP(stripe_index, h.total_stripe_count) { CUDA_KERNEL_LOOP(stripe_index, h.total_stripe_count) {
Index segment_offset, dim_index_base, actual_height; Index segment_offset, dim_index_base, actual_height;
h.calculate(stripe_index, segment_offset, dim_index_base, actual_height); h.calculate(stripe_index, &segment_offset, &dim_index_base, &actual_height);
for (Index j = 0; j < actual_height; j++) { for (Index j = 0; j < actual_height; j++) {
Index current_segment_id = segment_ids[dim_index_base + j]; Index current_segment_id = segment_ids[dim_index_base + j];
@ -217,11 +216,11 @@ class ArrangeHelper {
total_stripe_count = inner_dim_size * input_outer_dim_num_stripe; total_stripe_count = inner_dim_size * input_outer_dim_num_stripe;
} }
DEVICE inline void calculate(T stripe_index, T& segment_offset, DEVICE inline void calculate(T stripe_index, T* segment_offset,
T& dim_index_base, T& actual_height) { T* dim_index_base, T* actual_height) {
segment_offset = stripe_index % inner_dim_size; *segment_offset = stripe_index % inner_dim_size;
dim_index_base = stripe_index / inner_dim_size * DimTileSize; *dim_index_base = stripe_index / inner_dim_size * DimTileSize;
actual_height = min(DimTileSize, input_length_size - dim_index_base); *actual_height = min(DimTileSize, input_length_size - *dim_index_base);
} }
}; };

24
paddle/fluid/platform/cuda_primitives.h
Unescape View File

@ -137,12 +137,12 @@ USE_CUDA_ATOMIC(Max, unsigned int);
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350 #if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
USE_CUDA_ATOMIC(Max, unsigned long long int); // NOLINT USE_CUDA_ATOMIC(Max, unsigned long long int); // NOLINT
#else #else
CUDA_ATOMIC_WRAPPER(Max, unsigned long long int) { CUDA_ATOMIC_WRAPPER(Max, unsigned long long int) { // NOLINT
if (*address >= val) { if (*address >= val) {
return; return;
} }
unsigned long long int old = *address, assumed; unsigned long long int old = *address, assumed; // NOLINT
do { do {
assumed = old; assumed = old;
@ -169,7 +169,7 @@ CUDA_ATOMIC_WRAPPER(Max, float) {
return; return;
} }
int *const address_as_i = (int *)address; int *const address_as_i = reinterpret_cast<int *>(address);
int old = *address_as_i, assumed; int old = *address_as_i, assumed;
do { do {
@ -187,9 +187,9 @@ CUDA_ATOMIC_WRAPPER(Max, double) {
return; return;
} }
unsigned long long int *const address_as_ull = unsigned long long int *const address_as_ull = // NOLINT
(unsigned long long int *)address; reinterpret_cast<unsigned long long int *>(address); // NOLINT
unsigned long long int old = *address_as_ull, assumed; unsigned long long int old = *address_as_ull, assumed; // NOLINT
do { do {
assumed = old; assumed = old;
@ -209,12 +209,12 @@ USE_CUDA_ATOMIC(Min, unsigned int);
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350 #if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
USE_CUDA_ATOMIC(Min, unsigned long long int); // NOLINT USE_CUDA_ATOMIC(Min, unsigned long long int); // NOLINT
#else #else
CUDA_ATOMIC_WRAPPER(Min, unsigned long long int) { CUDA_ATOMIC_WRAPPER(Min, unsigned long long int) { // NOLINT
if (*address <= val) { if (*address <= val) {
return; return;
} }
unsigned long long int old = *address, assumed; unsigned long long int old = *address, assumed; // NOLINT
do { do {
assumed = old; assumed = old;
@ -241,7 +241,7 @@ CUDA_ATOMIC_WRAPPER(Min, float) {
return; return;
} }
int *const address_as_i = (int *)address; int *const address_as_i = reinterpret_cast<int *>(address);
int old = *address_as_i, assumed; int old = *address_as_i, assumed;
do { do {
@ -259,9 +259,9 @@ CUDA_ATOMIC_WRAPPER(Min, double) {
return; return;
} }
unsigned long long int *const address_as_ull = unsigned long long int *const address_as_ull = // NOLINT
(unsigned long long int *)address; reinterpret_cast<unsigned long long int *>(address); // NOLINT
unsigned long long int old = *address_as_ull, assumed; unsigned long long int old = *address_as_ull, assumed; // NOLINT
do { do {
assumed = old; assumed = old;

Write Preview
Loading…
Cancel
Save