SWDEV-304151 - Reset file descriptor after Module Load

File descriptors opened during hipModuleLoad are not closed
until hipModuleUnLoad. In large models, this exceeds the ulimit
for no. of open file descriptors causing out of resource issues.
Hence, close the file descriptors once the required data is read
from the file, and not wait until unload.

Change-Id: Ic50d424e5ca388715b1e6309a0de72a480321582


[ROCm/clr commit: 3b157b3d5f]
Esse commit está contido em:
Satyanvesh Dittakavi
2021-10-26 13:53:22 +00:00
commit 6ef56f09f1
3 arquivos alterados com 22 adições e 14 exclusões
@@ -509,6 +509,9 @@ hipError_t DynCO::loadCodeObject(const char* fname, const void* image) {
// Define Global functions
IHIP_RETURN_ONFAIL(populateDynGlobalFuncs());
// Resets the file info such as desc, size and name
// as they are not necessary once the code object is loaded
fb_info_->ResetFileInfo();
return hipSuccess;
}
+1 -14
Ver Arquivo
@@ -29,20 +29,7 @@ FatBinaryInfo::~FatBinaryInfo() {
delete fbd;
}
if (fdesc_ > 0) {
if (fsize_ && !amd::Os::MemoryUnmapFile(image_, fsize_)) {
guarantee(false, "Cannot unmap file");
}
if (!amd::Os::CloseFileHandle(fdesc_)) {
guarantee(false, "Cannot close file");
}
}
fname_ = std::string();
fdesc_ = amd::Os::FDescInit();
fsize_ = 0;
image_ = nullptr;
uri_ = std::string();
ResetFileInfo();
}
hipError_t FatBinaryInfo::ExtractFatBinary(const std::vector<hip::Device*>& devices) {
+18
Ver Arquivo
@@ -67,6 +67,24 @@ public:
return hipSuccess;
}
// Resets the file information
void ResetFileInfo() {
if (fdesc_ > 0) {
if (fsize_ && !amd::Os::MemoryUnmapFile(image_, fsize_)) {
guarantee(false, "Cannot unmap file");
}
if (!amd::Os::CloseFileHandle(fdesc_)) {
guarantee(false, "Cannot close file");
}
}
fname_ = std::string();
fdesc_ = amd::Os::FDescInit();
fsize_ = 0;
image_ = nullptr;
uri_ = std::string();
}
private:
std::string fname_; // File name
amd::Os::FileDesc fdesc_; // File descriptor