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

Rollback PR #39577: CUDNN v8 support 

PiperOrigin-RevId: 314219496
Change-Id: I3e36453044e5b73274e05d84f8d7f2cdb5b144ae
This commit is contained in:
Zhenyu Tan 2020-06-01 15:51:27 -07:00 committed by TensorFlower Gardener
parent 61a0c3bccd
commit 5cf536f78d
6 changed files with 18 additions and 3434 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

74
tensorflow/stream_executor/cuda/cuda_dnn.cc Normal file → Executable file
View File

@ -1278,11 +1278,7 @@ port::Status CheckAndFetchProjectionWeights(
cudnnRNNMode_t mode;
cudnnRNNAlgo_t algo;
cudnnDataType_t data_type;
#if CUDNN_VERSION >= 8000
RETURN_IF_CUDNN_ERROR(cudnnGetRNNDescriptor_v6(
#else
RETURN_IF_CUDNN_ERROR(cudnnGetRNNDescriptor(
#endif
/*handle=*/cudnn.handle(), /*rnnDesc=*/rnn_desc,
/*hiddenSize=*/&hidden_size_v,
/*numLayers=*/&num_layers_v,
@ -2428,28 +2424,6 @@ port::StatusOr<cudnnConvolutionFwdAlgo_t> GetCudnnConvolutionForwardAlgo(
const CudnnFilterDescriptor& filter, const CudnnConvolutionDescriptor& conv,
const CudnnTensorDescriptor& output_nd, bool specify_workspace_limit,
size_t memory_limit_bytes) {
#if CUDNN_VERSION >= 8000
const int num_requested_algos = 5;
int num_returned_algos = 0;
cudnnConvolutionFwdAlgoPerf_t perf_results[num_requested_algos];
RETURN_IF_CUDNN_ERROR(cudnnGetConvolutionForwardAlgorithm_v7(
cudnn.handle(), input_nd.handle(), filter.handle(), conv.handle(),
output_nd.handle(), num_requested_algos, &num_returned_algos,
perf_results));
size_t mem_limit = specify_workspace_limit ? memory_limit_bytes : 0ULL;
for (int r = 0; r < num_returned_algos; r++) {
if (perf_results[r].status == CUDNN_STATUS_SUCCESS &&
perf_results[r].algo != CUDNN_CONVOLUTION_FWD_ALGO_WINOGRAD_NONFUSED &&
perf_results[r].memory <= mem_limit) {
return perf_results[r].algo;
}
}
return port::Status(port::error::INTERNAL,
"cudnnGetConvolutionForwardAlgorithm_v7 returned "
"no suitable algorithms. This could be a cudnn bug.");
#else
cudnnConvolutionFwdPreference_t preference =
specify_workspace_limit ? CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT
: CUDNN_CONVOLUTION_FWD_NO_WORKSPACE;
@ -2458,7 +2432,6 @@ port::StatusOr<cudnnConvolutionFwdAlgo_t> GetCudnnConvolutionForwardAlgo(
cudnn.handle(), input_nd.handle(), filter.handle(), conv.handle(),
output_nd.handle(), preference, memory_limit_bytes, &algo_to_use));
return algo_to_use;
#endif
}
port::StatusOr<cudnnConvolutionBwdDataAlgo_t>
@ -2469,29 +2442,6 @@ GetCudnnConvolutionBackwardDataAlgo(const CudnnHandle& cudnn,
const CudnnTensorDescriptor& output_nd,
bool specify_workspace_limit,
size_t memory_limit_bytes) {
#if CUDNN_VERSION >= 8000
const int num_requested_algos = 5;
int num_returned_algos = 0;
cudnnConvolutionBwdDataAlgoPerf_t perf_results[num_requested_algos];
RETURN_IF_CUDNN_ERROR(cudnnGetConvolutionBackwardDataAlgorithm_v7(
cudnn.handle(), filter.handle(), output_nd.handle(), conv.handle(),
input_nd.handle(), num_requested_algos, &num_returned_algos,
perf_results));
size_t mem_limit = specify_workspace_limit ? memory_limit_bytes : 0ULL;
for (int r = 0; r < num_returned_algos; r++) {
if (perf_results[r].status == CUDNN_STATUS_SUCCESS &&
perf_results[r].algo !=
CUDNN_CONVOLUTION_BWD_DATA_ALGO_WINOGRAD_NONFUSED &&
perf_results[r].memory <= mem_limit) {
return perf_results[r].algo;
}
}
return port::Status(port::error::INTERNAL,
"cudnnGetConvolutionBackwardDataAlgorithm_v7 returned "
"no suitable algorithms. This could be a cudnn bug.");
#else
cudnnConvolutionBwdDataPreference_t preference =
specify_workspace_limit
? CUDNN_CONVOLUTION_BWD_DATA_SPECIFY_WORKSPACE_LIMIT
@ -2501,7 +2451,6 @@ GetCudnnConvolutionBackwardDataAlgo(const CudnnHandle& cudnn,
cudnn.handle(), filter.handle(), output_nd.handle(), conv.handle(),
input_nd.handle(), preference, memory_limit_bytes, &algo_to_use));
return algo_to_use;
#endif
}
port::StatusOr<cudnnConvolutionBwdFilterAlgo_t>
@ -2512,28 +2461,6 @@ GetCudnnConvolutionBackwardFilterAlgo(const CudnnHandle& cudnn,
const CudnnTensorDescriptor& output_nd,
bool specify_workspace_limit,
size_t memory_limit_bytes) {
#if CUDNN_VERSION >= 8000
const int num_requested_algos = 5;
int num_returned_algos = 0;
cudnnConvolutionBwdFilterAlgoPerf_t perf_results[num_requested_algos];
RETURN_IF_CUDNN_ERROR(cudnnGetConvolutionBackwardFilterAlgorithm_v7(
cudnn.handle(), input_nd.handle(), output_nd.handle(), conv.handle(),
filter.handle(), num_requested_algos, &num_returned_algos, perf_results));
size_t mem_limit = specify_workspace_limit ? memory_limit_bytes : 0ULL;
for (int r = 0; r < num_returned_algos; r++) {
if (perf_results[r].status == CUDNN_STATUS_SUCCESS &&
perf_results[r].algo !=
CUDNN_CONVOLUTION_BWD_FILTER_ALGO_WINOGRAD_NONFUSED &&
perf_results[r].memory <= mem_limit) {
return perf_results[r].algo;
}
}
return port::Status(port::error::INTERNAL,
"cudnnGetConvolutionBackwardFilterAlgorithm_v7 returned "
"no suitable algorithms. This could be a cudnn bug.");
#else
cudnnConvolutionBwdFilterPreference_t preference =
specify_workspace_limit
? CUDNN_CONVOLUTION_BWD_FILTER_SPECIFY_WORKSPACE_LIMIT
@ -2543,7 +2470,6 @@ GetCudnnConvolutionBackwardFilterAlgo(const CudnnHandle& cudnn,
cudnn.handle(), input_nd.handle(), output_nd.handle(), conv.handle(),
filter.handle(), preference, memory_limit_bytes, &algo_to_use));
return algo_to_use;
#endif
}
port::StatusOr<DeviceMemory<uint8>> AllocateCudnnConvolutionForwardWorkspace(

3316
tensorflow/stream_executor/cuda/cudnn_8_0.inc
View File

File diff suppressed because it is too large Load Diff

12
tensorflow/stream_executor/cuda/cudnn_stub.cc
View File

@ -51,17 +51,15 @@ cudnnStatus_t GetSymbolNotFoundError() { return CUDNN_STATUS_INTERNAL_ERROR; }
#error cuDNN version earlier than 6 is not supported.
#elif CUDNN_MAJOR < 7
#include "tensorflow/stream_executor/cuda/cudnn_6_0.inc"
#elif CUDNN_MAJOR == 7 && CUDNN_MINOR < 1
#elif CUDNN_MINOR < 1
#include "tensorflow/stream_executor/cuda/cudnn_7_0.inc"
// 2 instead of 3: see https://github.com/tensorflow/tensorflow/issues/32350
#elif CUDNN_MAJOR == 7 && CUDNN_MINOR < 2
#elif CUDNN_MINOR < 2
#include "tensorflow/stream_executor/cuda/cudnn_7_1.inc"
#elif CUDNN_MAJOR == 7 && CUDNN_MINOR < 4
#elif CUDNN_MINOR < 4
#include "tensorflow/stream_executor/cuda/cudnn_7_3.inc"
#elif CUDNN_MAJOR == 7 && CUDNN_MINOR < 6
#elif CUDNN_MINOR < 6
#include "tensorflow/stream_executor/cuda/cudnn_7_4.inc"
#elif CUDNN_MAJOR == 7
#include "tensorflow/stream_executor/cuda/cudnn_7_6.inc"
#else
#include "tensorflow/stream_executor/cuda/cudnn_8_0.inc"
#include "tensorflow/stream_executor/cuda/cudnn_7_6.inc"
#endif

24
third_party/gpus/cuda_configure.bzl vendored
View File

@ -1069,31 +1069,11 @@ def _create_local_cuda_repository(repository_ctx):
],
))
if [int(x) for x in cuda_config.cudnn_version.split(".")] < [8, 0]:
cudnn_headers = ["cudnn.h"]
else:
cudnn_headers = [
"cudnn_adv_infer.h",
"cudnn_adv_train.h",
"cudnn_cnn_infer.h",
"cudnn_cnn_train.h",
"cudnn_ops_infer.h",
"cudnn_ops_train.h",
"cudnn.h",
"cudnn_version.h",
]
cudnn_srcs = []
cudnn_outs = []
for header in cudnn_headers:
cudnn_srcs.append(cudnn_header_dir + "/" + header)
cudnn_outs.append("cudnn/include/" + header)
copy_rules.append(make_copy_files_rule(
repository_ctx,
name = "cudnn-include",
srcs = cudnn_srcs,
outs = cudnn_outs,
srcs = [cudnn_header_dir + "/cudnn.h"],
outs = ["cudnn/include/cudnn.h"],
))
# Set up BUILD file for cuda/

24
third_party/gpus/find_cuda_config.py vendored
View File

@ -219,20 +219,17 @@ def _find_library(base_paths, library_name, required_version):
return _find_file(base_paths, _library_paths(), filepattern)
def _find_versioned_file(base_paths, relative_paths, filepatterns,
def _find_versioned_file(base_paths, relative_paths, filepattern,
required_version, get_version):
"""Returns first valid path to a file that matches the requested version."""
if type(filepatterns) not in [list, tuple]:
filepatterns = [filepatterns]
for path in _cartesian_product(base_paths, relative_paths):
for filepattern in filepatterns:
for file in glob.glob(os.path.join(path, filepattern)):
actual_version = get_version(file)
if _matches_version(actual_version, required_version):
return file, actual_version
for file in glob.glob(os.path.join(path, filepattern)):
actual_version = get_version(file)
if _matches_version(actual_version, required_version):
return file, actual_version
raise _not_found_error(
base_paths, relative_paths,
", ".join(filepatterns) + " matching version '%s'" % required_version)
filepattern + " matching version '%s'" % required_version)
def _find_header(base_paths, header_name, required_version, get_version):
@ -429,13 +426,12 @@ def _find_cufft_config(base_paths, required_version, cuda_version):
def _find_cudnn_config(base_paths, required_version):
def get_header_version(path):
version = [
version = (
_get_header_version(path, name)
for name in ("CUDNN_MAJOR", "CUDNN_MINOR", "CUDNN_PATCHLEVEL")]
return ".".join(version) if version[0] else None
for name in ("CUDNN_MAJOR", "CUDNN_MINOR", "CUDNN_PATCHLEVEL"))
return ".".join(version)
header_path, header_version = _find_header(base_paths,
("cudnn.h", "cudnn_version.h"),
header_path, header_version = _find_header(base_paths, "cudnn.h",
required_version,
get_header_version)
cudnn_version = header_version.split(".")[0]

2
third_party/gpus/find_cuda_config.py.gz.base64 vendored
View File

File diff suppressed because one or more lines are too long