intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-06 02:18:05 +08:00
Code Issues Packages Projects Releases Wiki Activity

Create definition of tag allocation layout

Browse Source 
we use tag allocation for multiple purposes, therefore we should define
all offsets in one place

Resolves: NEO-7559
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This commit is contained in:
Mateusz Jablonski
2022-12-06 07:32:34 +00:00
committed by Compute-Runtime-Automation
parent ba9ea6fabf
commit 103f522f18
13 changed files with 67 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
shared/source/direct_submission/linux/drm_direct_submission.inl
View File

@@ -7,6 +7,7 @@
#include "shared/source/command_container/implicit_scaling.h"
#include "shared/source/command_stream/linear_stream.h"
#include "shared/source/command_stream/tag_allocation_layout.h"
#include "shared/source/device/device.h"
#include "shared/source/direct_submission/linux/drm_direct_submission.h"
#include "shared/source/os_interface/linux/drm_allocation.h"
@@ -62,8 +63,8 @@ DrmDirectSubmission<GfxFamily, Dispatcher>::DrmDirectSubmission(const DirectSubm
if (DebugManager.flags.PrintCompletionFenceUsage.get()) {
std::cout << "Completion fence for DirectSubmission:"
<< " GPU address: " << std::hex << (this->completionFenceAllocation->getGpuAddress() + Drm::completionFenceOffset)
<< ", CPU address: " << (castToUint64(this->completionFenceAllocation->getUnderlyingBuffer()) + Drm::completionFenceOffset)
<< " GPU address: " << std::hex << (this->completionFenceAllocation->getGpuAddress() + TagAllocationLayout::completionFenceOffset)
<< ", CPU address: " << (castToUint64(this->completionFenceAllocation->getUnderlyingBuffer()) + TagAllocationLayout::completionFenceOffset)
<< std::dec << std::endl;
}
}
@@ -79,7 +80,7 @@ inline DrmDirectSubmission<GfxFamily, Dispatcher>::~DrmDirectSubmission() {
if (this->isCompletionFenceSupported()) {
auto osContextLinux = static_cast<OsContextLinux *>(&this->osContext);
auto &drm = osContextLinux->getDrm();
auto completionFenceCpuAddress = reinterpret_cast<uint64_t>(this->completionFenceAllocation->getUnderlyingBuffer()) + Drm::completionFenceOffset;
auto completionFenceCpuAddress = reinterpret_cast<uint64_t>(this->completionFenceAllocation->getUnderlyingBuffer()) + TagAllocationLayout::completionFenceOffset;
drm.waitOnUserFences(*osContextLinux, completionFenceCpuAddress, this->completionFenceValue, this->activeTiles, this->postSyncOffset);
}
this->deallocateResources();
@@ -124,7 +125,7 @@ bool DrmDirectSubmission<GfxFamily, Dispatcher>::submit(uint64_t gpuAddress, siz
uint64_t completionFenceGpuAddress = 0u;
if (this->isCompletionFenceSupported()) {
completionValue = ++completionFenceValue;
completionFenceGpuAddress = this->completionFenceAllocation->getGpuAddress() + Drm::completionFenceOffset;
completionFenceGpuAddress = this->completionFenceAllocation->getGpuAddress() + TagAllocationLayout::completionFenceOffset;
}
for (auto drmIterator = 0u; drmIterator < osContextLinux->getDeviceBitfield().size(); drmIterator++) {