AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Correct legacy copy path.

Browse Source 
Legacy p2p copy path incorrectly transfered in whole pages rather than
the requested size.

Change-Id: I9aa7337754f9e32f587a0cc5305f8ffeb6196f10


[ROCm/ROCR-Runtime commit: 34ac62274a]
This commit is contained in:
Sean Keely
2021-02-02 02:13:54 -06:00
parent 4047b1c3a8
commit 622dc89e98
2 changed files with 11 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rocr-runtime/runtime/hsa-runtime/core/inc/runtime.h
+2 -3
View File
@@ -314,7 +314,7 @@ class Runtime {
amd::hsa::code::AmdHsaCodeManager* code_manager() { return &code_manager_; }
std::function<void*(size_t, size_t, MemoryRegion::AllocateFlags)>&
std::function<void*(size_t size, size_t align, MemoryRegion::AllocateFlags flags)>&
system_allocator() {
return system_allocator_;
}
@@ -482,8 +482,7 @@ class Runtime {
std::map<const void*, AllocationRegion> allocation_map_;
// Allocator using ::system_region_
std::function<void*(size_t, size_t, MemoryRegion::AllocateFlags)>
system_allocator_;
std::function<void*(size_t size, size_t align, MemoryRegion::AllocateFlags flags)> system_allocator_;
// Deallocator using ::system_region_
std::function<void(void*)> system_deallocator_;
projects/rocr-runtime/runtime/hsa-runtime/core/runtime/runtime.cpp
+9 -16
View File
@@ -173,20 +173,14 @@ void Runtime::RegisterAgent(Agent* agent) {
if (cpu_agents_.size() == 1) {
// Might need memory pooling to cover allocation that
// requires less than 4096 bytes.
system_allocator_ =
[&](size_t size, size_t alignment,
MemoryRegion::AllocateFlags alloc_flags) -> void* {
assert(alignment <= 4096);
void* ptr = NULL;
return (HSA_STATUS_SUCCESS ==
core::Runtime::runtime_singleton_->AllocateMemory(
system_regions_fine_[0], size, alloc_flags, &ptr))
? ptr
: NULL;
};
system_allocator_ = [this](size_t size, size_t align, MemoryRegion::AllocateFlags alloc_flags) -> void* {
assert(align <= 4096);
void* ptr = nullptr;
core::Runtime::runtime_singleton_->AllocateMemory(system_regions_fine_[0], size, alloc_flags, &ptr);
return ptr;
};
system_deallocator_ =
[](void* ptr) { core::Runtime::runtime_singleton_->FreeMemory(ptr); };
system_deallocator_ = [](void* ptr) { core::Runtime::runtime_singleton_->FreeMemory(ptr); };
BaseShared::SetAllocateAndFree(system_allocator_, system_deallocator_);
}
@@ -451,9 +445,8 @@ hsa_status_t Runtime::CopyMemory(void* dst, const void* src, size_t size) {
requires the caller to specify all allowed agents we can't assume that a peer mapped pointer
would remain mapped for the duration of the copy.
*/
void* temp = nullptr;
system_region->Allocate(size, core::MemoryRegion::AllocateNoFlags, &temp);
MAKE_SCOPE_GUARD([&]() { system_region->Free(temp, size); });
void* temp = system_allocator_(size, 0, core::MemoryRegion::AllocateNoFlags);
MAKE_SCOPE_GUARD([&]() { system_deallocator_(temp); });
hsa_status_t err = src_agent->DmaCopy(temp, source, size);
if (err == HSA_STATUS_SUCCESS) err = dst_agent->DmaCopy(dst, temp, size);
return err;