AI/rocm-systems
2
0
Vork 0
Code Kwesties Pull-aanvragen Actions Paketten Projecten Publicaties Wiki Activiteit

wsl/libhsakmt/memory: Dynamically allocate static global memory

Bladeren bron 
Signed-off-by: Flora Cui <flora.cui@amd.com>
Reviewed-by: Tianci Yin <tianci.yin@amd.com>
Part-of: <http://10.67.69.192/wsl/rocr-runtime/-/merge_requests/68>
This commit is contained in:
Flora Cui
2025-05-22 14:31:39 +08:00
gecommit door Frank Min
bovenliggende 95ec7d5c27
commit 56553974d0
1 gewijzigde bestanden met toevoegingen van 38 en 36 verwijderingen
Download Patch-bestand Download Diff-bestand Expand all files Collapse all files
memory.cpp
+38 -36
Bestand weergeven
@@ -62,14 +62,16 @@ struct Allocation {
int dmabuf_fd;
};
static std::map<const void *, Allocation> allocation_map_;
static std::unique_ptr<std::mutex> allocation_map_lock_ = std::make_unique<std::mutex>();
static std::map<const void *, Allocation>* allocation_map_ = new std::map<const void *, Allocation>();
static std::mutex* allocation_map_lock_ = new std::mutex();
void clear_allocation_map(void)
{
allocation_map_lock_ = std::make_unique<std::mutex>();
//delete allocation_map_lock_;
allocation_map_lock_ = new std::mutex();
std::lock_guard<std::mutex> lock(*allocation_map_lock_);
allocation_map_.clear();
delete allocation_map_;
allocation_map_ = new std::map<const void *, Allocation>();
}
HSAKMT_STATUS HSAKMTAPI hsaKmtSetMemoryPolicy(HSAuint32 Node,
@@ -249,7 +251,7 @@ after_trim:
else
*MemoryAddress = reinterpret_cast<void *>(gpu_mem->GpuAddress());
allocation_map_[*MemoryAddress] = Allocation(
(*allocation_map_)[*MemoryAddress] = Allocation(
gpu_mem->GetGpuMemoryHandle(), *MemoryAddress, (uint64_t)*MemoryAddress,
create_info.size, false, nullptr, SizeInBytes,
MemFlags.ui32.GTTAccess ? 0 : PreferredNode, MemFlags.Value);
@@ -291,8 +293,8 @@ HSAKMT_STATUS hsaKmtFreeMemoryInternal(void *MemoryAddress,
wsl::thunk::GpuMemory *gpu_mem = nullptr;
{
std::lock_guard<std::mutex> gard(*allocation_map_lock_);
auto it = allocation_map_.find(MemoryAddress);
if (it == allocation_map_.end()) {
auto it = allocation_map_->find(MemoryAddress);
if (it == allocation_map_->end()) {
return HSAKMT_STATUS_ERROR;
}
@@ -301,7 +303,7 @@ HSAKMT_STATUS hsaKmtFreeMemoryInternal(void *MemoryAddress,
close(it->second.dmabuf_fd);
it->second.dmabuf_fd = -1;
}
allocation_map_.erase(it);
allocation_map_->erase(it);
}
if (gpu_mem->IsQueueReferenced())
@@ -323,8 +325,8 @@ bool queue_acquire_buffer(void *MemoryAddress) {
wsl::thunk::GpuMemory *gpu_mem = nullptr;
{
std::lock_guard<std::mutex> gard(*allocation_map_lock_);
auto it = allocation_map_.find(MemoryAddress);
if (it == allocation_map_.end()) {
auto it = allocation_map_->find(MemoryAddress);
if (it == allocation_map_->end()) {
return HSAKMT_STATUS_ERROR;
}
@@ -344,8 +346,8 @@ bool queue_release_buffer(void *MemoryAddress) {
wsl::thunk::GpuMemory *gpu_mem = nullptr;
{
std::lock_guard<std::mutex> gard(*allocation_map_lock_);
auto it = allocation_map_.find(MemoryAddress);
if (it == allocation_map_.end()) {
auto it = allocation_map_->find(MemoryAddress);
if (it == allocation_map_->end()) {
return HSAKMT_STATUS_ERROR;
}
@@ -360,8 +362,8 @@ bool queue_release_buffer(void *MemoryAddress) {
wsl::thunk::GpuMemory *get_gpu_mem(void *MemoryAddress) {
std::lock_guard<std::mutex> gard(*allocation_map_lock_);
auto it = allocation_map_.find(MemoryAddress);
if (it == allocation_map_.end()) {
auto it = allocation_map_->find(MemoryAddress);
if (it == allocation_map_->end()) {
return nullptr;
}
@@ -463,8 +465,8 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtExportDMABufHandle(void *MemoryAddress,
std::lock_guard<std::mutex> gard(*allocation_map_lock_);
auto it = allocation_map_.upper_bound(MemoryAddress);
if (it != allocation_map_.begin()) {
auto it = allocation_map_->upper_bound(MemoryAddress);
if (it != allocation_map_->begin()) {
--it;
if (it->second.dmabuf_fd == -1) {
auto gpu_mem = wsl::thunk::GpuMemory::Convert(it->second.handle);
@@ -524,7 +526,7 @@ HSAKMT_STATUS hsaKmtImportDMABufHandle(int DMABufFd,
* the gpu_mem->Flags() need convert back from GpuMemoryCreateFlags to
* HsaMemFlags, reference hsaKmtAllocMemoryAlign
* */
allocation_map_[MemoryAddress] = Allocation(
(*allocation_map_)[MemoryAddress] = Allocation(
gpu_mem->GetGpuMemoryHandle(), MemoryAddress, (uint64_t)MemoryAddress,
gpu_mem->Size(), false, nullptr, gpu_mem->ClientSize(),
1, gpu_mem->Flags());
@@ -602,14 +604,14 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtDeregisterMemory(void *MemoryAddress) {
std::lock_guard<std::mutex> gard(*allocation_map_lock_);
// IPC mem(vram) and IPC signal(sys mem)
auto it_ipc = allocation_map_.find(MemoryAddress);
if (it_ipc != allocation_map_.end()) {
auto it_ipc = allocation_map_->find(MemoryAddress);
if (it_ipc != allocation_map_->end()) {
wsl::thunk::GpuMemoryDescFlags flags;
flags.reserved = it_ipc->second.mem_flags_value;
if (flags.is_imported_vram_alloc_va || flags.is_imported_sys_memfd) {
wsl::thunk::GpuMemory *gpu_mem;
gpu_mem = wsl::thunk::GpuMemory::Convert(it_ipc->second.handle);
allocation_map_.erase(it_ipc);
allocation_map_->erase(it_ipc);
delete gpu_mem;
return HSAKMT_STATUS_SUCCESS;
}
@@ -644,8 +646,8 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtMapMemoryToGPU(void *MemoryAddress,
{
std::lock_guard<std::mutex> gard(*allocation_map_lock_);
// IPC mem
auto it_ipc = allocation_map_.find(aligned_ptr);
if (it_ipc != allocation_map_.end()) {
auto it_ipc = allocation_map_->find(aligned_ptr);
if (it_ipc != allocation_map_->end()) {
wsl::thunk::GpuMemoryDescFlags flags;
flags.reserved = it_ipc->second.mem_flags_value;
if (flags.is_imported_vram_alloc_va) {
@@ -669,8 +671,8 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtMapMemoryToGPU(void *MemoryAddress,
}
// GTT mem
auto it_gtt = allocation_map_.find(aligned_ptr);
if (it_gtt != allocation_map_.end()) {
auto it_gtt = allocation_map_->find(aligned_ptr);
if (it_gtt != allocation_map_->end()) {
if (!it_gtt->second.userptr) {
if (it_gtt->second.size >= MemorySizeInBytes) {
*AlternateVAGPU = (uint64_t)MemoryAddress;
@@ -682,8 +684,8 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtMapMemoryToGPU(void *MemoryAddress,
}
// userptr mem
auto it = allocation_map_.find(MemoryAddress);
if (it != allocation_map_.end()) {
auto it = allocation_map_->find(MemoryAddress);
if (it != allocation_map_->end()) {
if (it->second.userptr && it->second.size >= MemorySizeInBytes) {
*AlternateVAGPU =
(uintptr_t)it->second.gpu_addr +
@@ -715,10 +717,10 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtMapMemoryToGPU(void *MemoryAddress,
{
std::lock_guard<std::mutex> guard(*allocation_map_lock_);
allocation_map_[MemoryAddress] =
(*allocation_map_)[MemoryAddress] =
Allocation(handle, aligned_ptr, addr, aligned_size, true, MemoryAddress,
MemorySizeInBytes);
allocation_map_[(void *)addr] =
(*allocation_map_)[(void *)addr] =
Allocation(handle, aligned_ptr, addr, aligned_size, true, nullptr,
MemorySizeInBytes);
}
@@ -755,8 +757,8 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtUnmapMemoryToGPU(void *MemoryAddress) {
std::lock_guard<std::mutex> gard(*allocation_map_lock_);
// IPC mem
auto it_ipc = allocation_map_.find(MemoryAddress);
if (it_ipc != allocation_map_.end()) {
auto it_ipc = allocation_map_->find(MemoryAddress);
if (it_ipc != allocation_map_->end()) {
wsl::thunk::GpuMemoryDescFlags flags;
flags.reserved = it_ipc->second.mem_flags_value;
if (flags.is_imported_vram_alloc_va) {
@@ -772,8 +774,8 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtUnmapMemoryToGPU(void *MemoryAddress) {
}
}
auto it = allocation_map_.find(MemoryAddress);
if (it == allocation_map_.end()) {
auto it = allocation_map_->find(MemoryAddress);
if (it == allocation_map_->end()) {
return HSAKMT_STATUS_ERROR;
}
@@ -783,8 +785,8 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtUnmapMemoryToGPU(void *MemoryAddress) {
handle = it->second.handle;
allocation_map_.erase((void *)it->second.gpu_addr);
allocation_map_.erase(it);
allocation_map_->erase((void *)it->second.gpu_addr);
allocation_map_->erase(it);
}
auto gpu_mem = wsl::thunk::GpuMemory::Convert(handle);
if (gpu_mem->IsQueueReferenced())
@@ -842,8 +844,8 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtQueryPointerInfo(const void *Pointer,
bool found = false;
{
std::lock_guard<std::mutex> gard(*allocation_map_lock_);
auto it = allocation_map_.upper_bound(Pointer);
if (it != allocation_map_.begin()) {
auto it = allocation_map_->upper_bound(Pointer);
if (it != allocation_map_->begin()) {
--it;
if (Pointer >= it->first &&
(Pointer < reinterpret_cast<const uint8_t*>(it->first) + it->second.size_requested)) {