Skip to content

Remove cuDNN compute capablity check as its inaccurate #2960

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
umar456 merged 2 commits into from
Jul 3, 2020
Merged

Remove cuDNN compute capablity check as its inaccurate #2960

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
181f907
Remove cuDNN compute capablity check as its inaccurate
umar456 Jul 3, 2020
b2de525
Fix cuDNN runtime version checks and warnings
umar456 Jul 3, 2020
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
26 changes: 11 additions & 15 deletions src/backend/cuda/cudnnModule.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,6 +51,12 @@ auto cudnnVersionComponents(size_t version) {
return make_tuple(major, minor, patch);
}

auto cudaRuntimeVersionComponents(size_t version) {
auto major = version / 1000;
auto minor = (version - (major * 1000)) / 10;
return make_tuple(major, minor);
}

cudnnModule::cudnnModule()
: module({"cudnn"}, {"", "64_7", "64_8", "64_6", "64_5", "64_4"}, {""},
cudnnVersions.size(), cudnnVersions.data()) {
Expand Down Expand Up @@ -83,8 +89,7 @@ cudnnModule::cudnnModule()
major, minor);
}

std::tie(rtmajor, rtminor, std::ignore) =
cudnnVersionComponents(cudnn_rtversion);
std::tie(rtmajor, rtminor) = cudaRuntimeVersionComponents(cudnn_rtversion);

AF_TRACE("cuDNN Version: {}.{}.{} cuDNN CUDA Runtime: {}.{}", major, minor,
patch, rtmajor, rtminor);
Expand All @@ -101,15 +106,16 @@ cudnnModule::cudnnModule()

int afcuda_runtime = 0;
cudaRuntimeGetVersion(&afcuda_runtime);
if (afcuda_runtime != static_cast<int>(cudnn_version)) {
if (afcuda_runtime != static_cast<int>(cudnn_rtversion)) {
getLogger()->warn(
"WARNING: ArrayFire CUDA Runtime({}) and cuDNN CUDA "
"Runtime({}.{}) do not match. For maximum compatibility, make sure "
"Runtime({}) do not match. For maximum compatibility, make sure "
"the two versions match.(Ignoring check)",
// NOTE: the int version formats from CUDA and cuDNN are different
// so we are using int_version_to_string for the ArrayFire CUDA
// runtime
int_version_to_string(afcuda_runtime), rtmajor, rtminor);
int_version_to_string(afcuda_runtime),
int_version_to_string(cudnn_rtversion));
}

MODULE_FUNCTION_INIT(cudnnConvolutionBackwardData);
Expand All @@ -135,16 +141,6 @@ cudnnModule::cudnnModule()
MODULE_FUNCTION_INIT(cudnnSetStream);
MODULE_FUNCTION_INIT(cudnnSetTensor4dDescriptor);

// Check to see if the cuDNN runtime is compatible with the current device
cudaDeviceProp prop = getDeviceProp(getActiveDeviceId());
if (!checkDeviceWithRuntime(cudnn_rtversion, {prop.major, prop.minor})) {
string error_message = fmt::format(
"Error: cuDNN CUDA Runtime({}.{}) does not support the "
"current device's compute capability(sm_{}{}).",
rtmajor, rtminor, prop.major, prop.minor);
AF_ERROR(error_message, AF_ERR_RUNTIME);
}

if (!module.symbolsLoaded()) {
string error_message =
"Error loading cuDNN symbols. ArrayFire was unable to load some "
Expand Down