intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 14:55:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

performance: add early return for single byte path

Browse Source 
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
This commit is contained in:
Michal Mrozek
2025-08-08 10:04:02 +00:00
committed by Compute-Runtime-Automation
parent 1ff9c45b90
commit e9ff1260f0
1 changed files with 10 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
level_zero/core/source/driver/driver_handle_imp.cpp
View File

@@ -479,28 +479,20 @@ ze_result_t DriverHandleImp::getDevice(uint32_t *pCount, ze_device_handle_t *phD
bool DriverHandleImp::findAllocationDataForRange(const void *buffer,
size_t size,
NEO::SvmAllocationData *&allocData) {
size_t offset = 0;
if (size > 0) {
offset = size - 1;
}
// Make sure the host buffer does not overlap any existing allocation
const char *baseAddress = reinterpret_cast<const char *>(buffer);
uint64_t baseAddress = castToUint64(buffer);
NEO::SvmAllocationData *beginAllocData = svmAllocsManager->getSVMAlloc(buffer);
NEO::SvmAllocationData *endAllocData = offset == 0 ? beginAllocData : svmAllocsManager->getSVMAlloc(static_cast<const void *>(baseAddress + offset));
if (beginAllocData) {
allocData = beginAllocData;
} else {
allocData = endAllocData;
if (!beginAllocData) {
allocData = nullptr;
return false;
}
// Return true if the whole range requested is covered by the same allocation
if (beginAllocData && endAllocData &&
(beginAllocData->gpuAllocations.getDefaultGraphicsAllocation() == endAllocData->gpuAllocations.getDefaultGraphicsAllocation())) {
// for size <= 1 we do not need to do bounds checking as pointer is confirmed to be valid
if (size <= 1 || (baseAddress + size <= beginAllocData->gpuAllocations.getDefaultGraphicsAllocation()->getGpuAddress() + beginAllocData->size)) {
allocData = beginAllocData;
return true;
}
// we have base allocation, but it doesn't cover whole size, it means we deal with virtual memory pointers, check if we have whole range covered
bool allocationRangeCovered = false;
// If memory accessed is part of a virtual reservation, then return the first allocation data within the range.
auto allocDataVec = findAllocationsWithinRange(buffer, size, &allocationRangeCovered);