SWDEV-311270 - Clean-up IPC path in ROCclr
- Make IPC interfaces generic between devices and rely on the IPC buffer
for attach/detach logic
Change-Id: Id3c18d122030329b7ee532bbb6317de9dd6a0bbe
[ROCm/clr commit: 9aa6f25f1c]
이 커밋은 다음에 포함됨:
@@ -2220,142 +2220,6 @@ void Device::updateFreeMemory(size_t size, bool free) {
|
||||
ClPrint(amd::LOG_INFO, amd::LOG_MEM, "device=0x%lx, freeMem_ = 0x%zx", this, freeMem_.load());
|
||||
}
|
||||
|
||||
bool Device::IpcCreate(void* dev_ptr, size_t* mem_size, void* handle, size_t* mem_offset) const {
|
||||
hsa_status_t hsa_status = HSA_STATUS_SUCCESS;
|
||||
|
||||
amd::Memory* amd_mem_obj = amd::MemObjMap::FindMemObj(dev_ptr);
|
||||
if (amd_mem_obj == nullptr) {
|
||||
DevLogPrintfError("Cannot retrieve amd_mem_obj for dev_ptr: 0x%x", dev_ptr);
|
||||
return false;
|
||||
}
|
||||
|
||||
// Get the original pointer from the amd::Memory object
|
||||
void* orig_dev_ptr = nullptr;
|
||||
if (amd_mem_obj->getSvmPtr() != nullptr) {
|
||||
orig_dev_ptr = amd_mem_obj->getSvmPtr();
|
||||
} else if (amd_mem_obj->getHostMem() != nullptr) {
|
||||
orig_dev_ptr = amd_mem_obj->getHostMem();
|
||||
} else {
|
||||
ShouldNotReachHere();
|
||||
}
|
||||
|
||||
// Check if the dev_ptr is lesser than original dev_ptr
|
||||
if (orig_dev_ptr > dev_ptr) {
|
||||
//If this happens, then revisit FindMemObj logic
|
||||
DevLogPrintfError("Original dev_ptr: 0x%x cannot be greater than dev_ptr: 0x%x",
|
||||
orig_dev_ptr, dev_ptr);
|
||||
return false;
|
||||
}
|
||||
|
||||
//Calculate the memory offset from the original base ptr
|
||||
*mem_offset = reinterpret_cast<address>(dev_ptr) - reinterpret_cast<address>(orig_dev_ptr);
|
||||
*mem_size = amd_mem_obj->getSize();
|
||||
|
||||
//Check if the dev_ptr is greater than memory allocated
|
||||
if (*mem_offset > *mem_size) {
|
||||
DevLogPrintfError("Memory offset: %u cannot be greater than size of "
|
||||
"original memory allocated: %u", *mem_size, *mem_offset);
|
||||
return false;
|
||||
}
|
||||
|
||||
hsa_status = hsa_amd_ipc_memory_create(orig_dev_ptr, *mem_size,
|
||||
reinterpret_cast<hsa_amd_ipc_memory_t*>(handle));
|
||||
|
||||
if (hsa_status != HSA_STATUS_SUCCESS) {
|
||||
LogPrintfError("Failed to create memory for IPC, failed with hsa_status: %d \n", hsa_status);
|
||||
return false;
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
bool Device::IpcAttach(const void* handle, size_t mem_size, size_t mem_offset,
|
||||
unsigned int flags, void** dev_ptr) const {
|
||||
amd::Memory* amd_mem_obj = nullptr;
|
||||
void* orig_dev_ptr = nullptr;
|
||||
|
||||
// Retrieve the devPtr from the handle
|
||||
hsa_status_t hsa_status =
|
||||
hsa_amd_ipc_memory_attach(reinterpret_cast<const hsa_amd_ipc_memory_t*>(handle),
|
||||
mem_size, (1 + p2p_agents_.size()), p2p_agents_list_,
|
||||
&orig_dev_ptr);
|
||||
|
||||
if (hsa_status != HSA_STATUS_SUCCESS) {
|
||||
LogPrintfError("HSA failed to attach IPC memory with status: %d \n", hsa_status);
|
||||
return false;
|
||||
}
|
||||
|
||||
amd_mem_obj = amd::MemObjMap::FindMemObj(orig_dev_ptr);
|
||||
if (amd_mem_obj == nullptr) {
|
||||
|
||||
// Memory does not exist, create an amd Memory object for the pointer
|
||||
amd_mem_obj = new (context()) amd::Buffer(context(), flags, mem_size, orig_dev_ptr);
|
||||
if (amd_mem_obj == nullptr) {
|
||||
LogError("failed to create a mem object!");
|
||||
return false;
|
||||
}
|
||||
|
||||
if (!amd_mem_obj->create(nullptr)) {
|
||||
LogError("failed to create a svm hidden buffer!");
|
||||
amd_mem_obj->release();
|
||||
return false;
|
||||
}
|
||||
amd_mem_obj->setIpcShared(true);
|
||||
// Add the original mem_ptr to the MemObjMap with newly created amd_mem_obj
|
||||
amd::MemObjMap::AddMemObj(orig_dev_ptr, amd_mem_obj);
|
||||
|
||||
} else {
|
||||
//Memory already exists, just retain the old one.
|
||||
amd_mem_obj->retain();
|
||||
}
|
||||
|
||||
//Make sure the mem_offset doesnt overflow the allocated memory
|
||||
guarantee((mem_offset < mem_size), "IPC mem offset greater than allocated size");
|
||||
|
||||
// Return orig_dev_ptr
|
||||
*dev_ptr = reinterpret_cast<address>(orig_dev_ptr);
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
bool Device::IpcDetach (void* dev_ptr) const {
|
||||
hsa_status_t hsa_status = HSA_STATUS_SUCCESS;
|
||||
|
||||
amd::Memory* amd_mem_obj = amd::MemObjMap::FindMemObj(dev_ptr);
|
||||
if (amd_mem_obj == nullptr) {
|
||||
DevLogPrintfError("Memory object for the ptr: 0x%x cannot be null \n", dev_ptr);
|
||||
return false;
|
||||
}
|
||||
|
||||
if (!amd_mem_obj->ipcShared()) {
|
||||
DevLogPrintfError("Memory object for the ptr: 0x%x is not ipcShared \n", dev_ptr);
|
||||
return false;
|
||||
}
|
||||
|
||||
// Get the original pointer from the amd::Memory object
|
||||
void* orig_dev_ptr = nullptr;
|
||||
if (amd_mem_obj->getSvmPtr() != nullptr) {
|
||||
orig_dev_ptr = amd_mem_obj->getSvmPtr();
|
||||
} else if (amd_mem_obj->getHostMem() != nullptr) {
|
||||
orig_dev_ptr = amd_mem_obj->getHostMem();
|
||||
} else {
|
||||
ShouldNotReachHere();
|
||||
}
|
||||
|
||||
if (amd_mem_obj->release() == 0) {
|
||||
amd::MemObjMap::RemoveMemObj(orig_dev_ptr);
|
||||
|
||||
// Detach the memory from HSA
|
||||
hsa_status = hsa_amd_ipc_memory_detach(orig_dev_ptr);
|
||||
if (hsa_status != HSA_STATUS_SUCCESS) {
|
||||
LogPrintfError("HSA failed to detach memory with status: %d \n", hsa_status);
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
// ================================================================================================
|
||||
void* Device::svmAlloc(amd::Context& context, size_t size, size_t alignment, cl_svm_mem_flags flags,
|
||||
void* svmPtr) const {
|
||||
|
||||
새 이슈에서 참조
사용자 차단