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

rocr/driver: add memory residency management interface in driver

Browse Source 
This commit introduces MakeMemoryResident and MakeMemoryUnresident
functions to KfdDriver and XdnaDriver classes.

- Added implementations in amd_kfd_driver.cpp
- Added stubs in amd_xdna_driver.cpp returning HSA_STATUS_ERROR
- Updated header files amd_kfd_driver.h and amd_xdna_driver.h
- Removed MakeKfdMemoryResident/Unresident from amd_memory_region.cpp

Signed-off-by: Honglei Huang <Honglei1.Huang@amd.com>


[ROCm/ROCR-Runtime commit: 6c87f5b5ce]
This commit is contained in:
Honglei Huang
2025-07-11 13:54:39 +08:00
committed by Huang, Honglei1
parent b61df004ff
commit 7c29c36f7e
8 changed files with 77 additions and 40 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
+7 -7
View File
@@ -556,7 +556,7 @@ void GpuAgent::ReserveScratch()
void* reserved_base = scratch_pool_.alloc(reserved_sz);
assert(reserved_base && "Could not allocate reserved memory");
if (HSAKMT_CALL(hsaKmtMapMemoryToGPU(reserved_base, reserved_sz, &alt_va)) == HSAKMT_STATUS_SUCCESS)
if (driver().MakeMemoryResident(reserved_base, reserved_sz, &alt_va) == HSA_STATUS_SUCCESS)
scratch_cache_.reserve(reserved_sz, reserved_base);
else
throw AMD::hsa_exception(HSA_STATUS_ERROR_OUT_OF_RESOURCES, "Reserve scratch memory failed.");
@@ -1887,8 +1887,8 @@ void GpuAgent::AcquireQueueMainScratch(ScratchInfo& scratch) {
if (scratch.main_queue_base != nullptr) {
HSAuint64 alternate_va;
if ((profile_ == HSA_PROFILE_FULL) ||
(HSAKMT_CALL(hsaKmtMapMemoryToGPU(scratch.main_queue_base, scratch.main_size, &alternate_va)) ==
HSAKMT_STATUS_SUCCESS)) {
(driver().MakeMemoryResident(scratch.main_queue_base, scratch.main_size,
&alternate_va) == HSA_STATUS_SUCCESS)) {
if (scratch.large) scratch_used_large_ += scratch.main_size;
scratch_cache_.insertMain(scratch);
return;
@@ -1940,7 +1940,7 @@ void GpuAgent::AcquireQueueMainScratch(ScratchInfo& scratch) {
HSAuint64 alternate_va;
if ((base != nullptr) &&
((profile_ == HSA_PROFILE_FULL) ||
(HSAKMT_CALL(hsaKmtMapMemoryToGPU(base, size, &alternate_va)) == HSAKMT_STATUS_SUCCESS))) {
(driver().MakeMemoryResident(base, size, &alternate_va) == HSA_STATUS_SUCCESS))) {
// Scratch allocated and either full profile or map succeeded.
scratch.main_queue_base = base;
scratch.main_size = size;
@@ -2020,8 +2020,8 @@ void GpuAgent::AcquireQueueAltScratch(ScratchInfo& scratch) {
if (scratch.alt_queue_base != nullptr) {
HSAuint64 alternate_va;
if ((profile_ == HSA_PROFILE_FULL) ||
(HSAKMT_CALL(hsaKmtMapMemoryToGPU(scratch.alt_queue_base, scratch.alt_size, &alternate_va)) ==
HSAKMT_STATUS_SUCCESS)) {
(driver().MakeMemoryResident(scratch.alt_queue_base, scratch.alt_size, &alternate_va) ==
HSA_STATUS_SUCCESS)) {
scratch_cache_.insertAlt(scratch);
return;
}
@@ -2061,7 +2061,7 @@ void GpuAgent::ReleaseQueueAltScratch(ScratchInfo& scratch) {
void GpuAgent::ReleaseScratch(void* base, size_t size, bool large) {
if (profile_ == HSA_PROFILE_BASE) {
if (HSAKMT_STATUS_SUCCESS != HSAKMT_CALL(hsaKmtUnmapMemoryToGPU(base))) {
if (HSA_STATUS_SUCCESS != driver().MakeMemoryUnresident(base)) {
assert(false && "Unmap scratch subrange failed!");
}
}
projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_memory_region.cpp
+11 -26
View File
@@ -58,24 +58,6 @@ namespace AMD {
size_t MemoryRegion::max_sysmem_alloc_size_ = 0;
const size_t MemoryRegion::kPageSize_ = sysconf(_SC_PAGESIZE);
bool MemoryRegion::MakeKfdMemoryResident(size_t num_node, const uint32_t* nodes, const void* ptr,
size_t size, uint64_t* alternate_va,
HsaMemMapFlags map_flag) {
assert(num_node > 0);
assert(nodes != NULL);
*alternate_va = 0;
const HSAKMT_STATUS status = HSAKMT_CALL(hsaKmtMapMemoryToGPUNodes(
const_cast<void*>(ptr), size, alternate_va, map_flag, num_node, const_cast<uint32_t*>(nodes)));
return (status == HSAKMT_STATUS_SUCCESS);
}
bool MemoryRegion::MakeKfdMemoryUnresident(const void* ptr) {
const HSAKMT_STATUS status = HSAKMT_CALL(hsaKmtUnmapMemoryToGPU(const_cast<void*>(ptr)));
return (status == HSAKMT_STATUS_SUCCESS);
}
MemoryRegion::MemoryRegion(bool fine_grain, bool kernarg, bool full_profile,
bool extended_scope_fine_grain, bool user_visible, core::Agent* owner,
const HsaMemoryProperties& mem_props)
@@ -508,7 +490,7 @@ hsa_status_t MemoryRegion::AllowAccess(uint32_t num_agents,
assert(cpu_in_list);
// This is a system region and only CPU agents in the whitelist.
// Remove old mappings.
AMD::MemoryRegion::MakeKfdMemoryUnresident(ptr);
owner()->driver().MakeMemoryUnresident(ptr);
return HSA_STATUS_SUCCESS;
}
@@ -528,10 +510,10 @@ hsa_status_t MemoryRegion::AllowAccess(uint32_t num_agents,
ScopedAcquire<KernelSharedMutex::Shared> lock(
core::Runtime::runtime_singleton_->memory_lock_.shared());
uint64_t alternate_va = 0;
if (!AMD::MemoryRegion::MakeKfdMemoryResident(
whitelist_nodes.size(), &whitelist_nodes[0], ptr,
size, &alternate_va, map_flag)) {
return HSA_STATUS_ERROR_OUT_OF_RESOURCES;
if (owner()->driver().MakeMemoryResident(ptr, size, &alternate_va, &map_flag,
whitelist_nodes.size(),
whitelist_nodes.data()) != HSA_STATUS_SUCCESS) {
return HSA_STATUS_ERROR_OUT_OF_RESOURCES;
}
}
@@ -599,8 +581,9 @@ hsa_status_t MemoryRegion::Lock(uint32_t num_agents, const hsa_agent_t* agents,
if (owner()->driver().RegisterMemory(host_ptr, size, const_cast<HsaMemFlags&>(mem_flag_)) ==
HSA_STATUS_SUCCESS) {
uint64_t alternate_va = 0;
if (MakeKfdMemoryResident(whitelist_nodes.size(), &whitelist_nodes[0],
host_ptr, size, &alternate_va, map_flag_)) {
if (owner()->driver().MakeMemoryResident(host_ptr, size, &alternate_va, &map_flag_,
whitelist_nodes.size(),
whitelist_nodes.data()) == HSA_STATUS_SUCCESS) {
if (alternate_va != 0) {
*agent_ptr = reinterpret_cast<void*>(alternate_va);
} else {
@@ -625,7 +608,9 @@ hsa_status_t MemoryRegion::Unlock(void* host_ptr) const {
return HSA_STATUS_SUCCESS;
}
MakeKfdMemoryUnresident(host_ptr);
if (owner()->driver().MakeMemoryUnresident(host_ptr) != HSA_STATUS_SUCCESS) {
assert(false && "Failed to unmap host pointer");
}
if (owner()->driver().DeregisterMemory(host_ptr) != HSA_STATUS_SUCCESS) {
assert(false && "Failed to deregister host pointer");
}