experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Automated g4 rollback of changelist 172041133

Browse Source 
PiperOrigin-RevId: 172129075
This commit is contained in:
A. Unique TensorFlower 2017-10-13 12:09:58 -07:00 committed by TensorFlower Gardener
parent 6a04ffd447
commit 26cc81e405
3 changed files with 5 additions and 81 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
tensorflow/core/util/cuda_kernel_helper.h
View File

@ -154,15 +154,11 @@ struct CudaLaunchConfig {
// Calculate the Cuda launch config we should use for a kernel launch.
// This is assuming the kernel is quite simple and will largely be
// memory-limited.
// REQUIRES: work_element_count > 0.
inline CudaLaunchConfig GetCudaLaunchConfig(int work_element_count,
const GPUDevice& d) {
CHECK_GT(work_element_count, 0);
CudaLaunchConfig config;
// in case of invalid input, return the default value config, which has all -1
if (work_element_count <= 0) {
return config;
}
const int virtual_thread_count = work_element_count;
const int physical_thread_count = std::min(
d.getNumCudaMultiProcessors() * d.maxCudaThreadsPerMultiProcessor(),
@ -180,17 +176,14 @@ inline CudaLaunchConfig GetCudaLaunchConfig(int work_element_count,
// Calculate the Cuda launch config we should use for a kernel launch. This
// variant takes the resource limits of func into account to maximize occupancy.
// REQUIRES: work_element_count > 0.
template <typename DeviceFunc>
inline CudaLaunchConfig GetCudaLaunchConfig(int work_element_count,
const GPUDevice& d, DeviceFunc func,
size_t dynamic_shared_memory_size,
int block_size_limit) {
CHECK_GT(work_element_count, 0);
CudaLaunchConfig config;
if (work_element_count <= 0) {
return config;
}
int block_count = 0;
int thread_per_block = 0;

69
tensorflow/core/util/cuda_kernel_helper_test.cu.cc
View File

@ -111,28 +111,6 @@ class CudaLaunchConfigTest : public ::testing::Test {
TEST_F(CudaLaunchConfigTest, GetCudaLaunchConfig) {
CudaLaunchConfig cfg;
// test invalid inputs
CudaLaunchConfig default_value;
cfg = GetCudaLaunchConfig(0, d);
EXPECT_EQ(default_value.virtual_thread_count, cfg.virtual_thread_count);
EXPECT_EQ(default_value.block_count, cfg.block_count);
EXPECT_EQ(default_value.thread_per_block, cfg.thread_per_block);
cfg = GetCudaLaunchConfig(-1, d);
EXPECT_EQ(default_value.virtual_thread_count, cfg.virtual_thread_count);
EXPECT_EQ(default_value.block_count, cfg.block_count);
EXPECT_EQ(default_value.thread_per_block, cfg.thread_per_block);
cfg = GetCudaLaunchConfig(0, d, Count1D, 0, 0);
EXPECT_EQ(default_value.virtual_thread_count, cfg.virtual_thread_count);
EXPECT_EQ(default_value.block_count, cfg.block_count);
EXPECT_EQ(default_value.thread_per_block, cfg.thread_per_block);
cfg = GetCudaLaunchConfig(-1, d, Count1D, 0, 0);
EXPECT_EQ(default_value.virtual_thread_count, cfg.virtual_thread_count);
EXPECT_EQ(default_value.block_count, cfg.block_count);
EXPECT_EQ(default_value.thread_per_block, cfg.thread_per_block);
// test valid inputs
#define TEST_LAUNCH_PARAMETER(work_element_count) \
cfg = GetCudaLaunchConfig(bufsize, d); \
@ -184,34 +162,6 @@ TEST_F(CudaLaunchConfigTest, GetCuda2DLaunchConfig) {
Cuda2DLaunchConfig cfg;
CudaLaunchConfig cfg1d;
// test invalid inputs
Cuda2DLaunchConfig default_value;
cfg = GetCuda2DLaunchConfig(1, 0, d);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda2DLaunchConfig(1, -1, d);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda2DLaunchConfig(-1, 1, d);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda2DLaunchConfig(-1, 1, d);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda2DLaunchConfig(0, -1, d);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda2DLaunchConfig(0, 0, d);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda2DLaunchConfig(1, 0, d, Count2D, 0, 0);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda2DLaunchConfig(1, -1, d, Count2D, 0, 0);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda2DLaunchConfig(-1, 1, d, Count2D, 0, 0);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda2DLaunchConfig(-1, 1, d, Count2D, 0, 0);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda2DLaunchConfig(0, -1, d, Count2D, 0, 0);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda2DLaunchConfig(0, 0, d, Count2D, 0, 0);
EXPECT_EQ(default_value, cfg);
// test valid inputs
#define TEST_LAUNCH_PARAMETER(dimx, dimy) \
cfg1d = GetCudaLaunchConfig(bufsize, d); \
@ -252,25 +202,6 @@ TEST_F(CudaLaunchConfigTest, GetCuda3DLaunchConfig) {
Cuda3DLaunchConfig cfg;
CudaLaunchConfig cfg1d;
// test invalid inputs
Cuda3DLaunchConfig default_value;
cfg = GetCuda3DLaunchConfig(0, 1, 1, d, Count3D, 0, 0);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda3DLaunchConfig(-1, 1, 1, d, Count3D, 0, 0);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda3DLaunchConfig(1, 0, 1, d, Count3D, 0, 0);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda3DLaunchConfig(1, -1, 1, d, Count3D, 0, 0);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda3DLaunchConfig(1, 1, 0, d, Count3D, 0, 0);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda3DLaunchConfig(1, 1, -1, d, Count3D, 0, 0);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda3DLaunchConfig(0, 0, 0, d, Count3D, 0, 0);
EXPECT_EQ(default_value, cfg);
cfg = GetCuda3DLaunchConfig(-1, -1, -1, d, Count3D, 0, 0);
EXPECT_EQ(default_value, cfg);
// test valid inputs
#define TEST_LAUNCH_PARAMETER(dimx, dimy, dimz) \
cfg1d = GetCudaLaunchConfig(bufsize, d, SetOutbufZero, 0, 0); \

2
tensorflow/tensorflow.bzl
View File

@ -647,7 +647,7 @@ def tf_cuda_only_cc_test(name,
clean_dep("//tensorflow:darwin"): 1,
"//conditions:default": 0,
}),
tags=tags)
tags=tags + tf_cuda_tests_tags())
# Create a cc_test for each of the tensorflow tests listed in "tests"
def tf_cc_tests(srcs,