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

rocr/libhsakmt: Add coarse-grain allocator to GPU

Browse Source 
[ROCm/ROCR-Runtime commit: adbc0495e2]
This commit is contained in:
Tony Gutierrez
2025-02-18 14:02:37 -08:00
committed by Yat Sin, David
parent 945d6da90b
commit 3ebcf3020f
2 changed files with 42 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp
+22 -15
View File
@@ -2456,25 +2456,32 @@ void GpuAgent::InitAllocators() {
}
assert(system_allocator_ && "Nearest NUMA node did not have a kernarg pool.");
// Setup fine-grain allocator
// Setup this GPU's fine-grain and coarse-grain allocators.
for (auto region : regions()) {
const AMD::MemoryRegion* amd_region = (const AMD::MemoryRegion*)region;
if (amd_region->IsLocalMemory() && amd_region->fine_grain()) {
finegrain_allocator_ = [region](size_t size,
MemoryRegion::AllocateFlags alloc_flags) -> void* {
void* ptr = nullptr;
return (HSA_STATUS_SUCCESS ==
core::Runtime::runtime_singleton_->AllocateMemory(region, size, alloc_flags, &ptr))
? ptr
: nullptr;
};
const AMD::MemoryRegion* amd_region = static_cast<const AMD::MemoryRegion*>(region);
finegrain_deallocator_ = [](void* ptr) {
core::Runtime::runtime_singleton_->FreeMemory(ptr);
};
auto region_allocator = [region](size_t size,
MemoryRegion::AllocateFlags alloc_flags) -> void* {
void* ptr = nullptr;
return (HSA_STATUS_SUCCESS ==
core::Runtime::runtime_singleton_->AllocateMemory(region, size, alloc_flags, &ptr))
? ptr
: nullptr;
};
auto region_deallocator = [](void* ptr) { core::Runtime::runtime_singleton_->FreeMemory(ptr); };
if (amd_region->IsLocalMemory() && amd_region->fine_grain()) {
finegrain_allocator_ = region_allocator;
finegrain_deallocator_ = region_deallocator;
} else if (amd_region->IsLocalMemory() &&
!(amd_region->fine_grain() || amd_region->extended_scope_fine_grain())) {
coarsegrain_allocator_ = region_allocator;
coarsegrain_deallocator_ = region_deallocator;
}
}
assert(finegrain_deallocator_ && "Agent does not have a fine-grain allocator");
assert(finegrain_allocator_ && "GPU agent does not have a fine-grain allocator");
assert(coarsegrain_allocator_ && "GPU agent does not have a coarse-grain allocator");
}
core::Agent* GpuAgent::GetNearestCpuAgent() const {