intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-20 08:53:55 +08:00
Code Issues Packages Projects Releases Wiki Activity
11,963 Commits 42 Branches 655 Tags
5c1b50bccf6df53d97a467bc2b95a3b6e027eee8
Commit Graph

93 Commits

Author SHA1 Message Date
Slawomir Milczarek
5c1b50bccf Add memory prefetch modes for single and multiple subdevices 
Single-subdevice prefetch for cmd list copy-only (with bcs) and acc mode.
Multi-subdevice prefetch (default) for shared allocation with multiple BOs.

Related-To: NEO-6740

Signed-off-by: Slawomir Milczarek <slawomir.milczarek@intel.com>
 2022-12-07 15:25:32 +01:00
Lukasz Jobczyk
7c572b4090 Do not free SVM alloc under SVM manager lock 
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
 2022-12-02 12:36:10 +01:00
Lukasz Jobczyk
dbc6ac2743 Set alignment for SVM_CPU properties on xe_hpc and later 
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
 2022-12-02 12:35:54 +01:00
Compute-Runtime-Validation
083471a158 Revert "Set alignment for SVM_CPU properties" 
This reverts commit db89d2ce6b.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
 2022-12-02 08:17:13 +01:00
Lukasz Jobczyk
db89d2ce6b Set alignment for SVM_CPU properties 
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
 2022-12-01 10:45:47 +01:00
Lukasz Jobczyk
7ea11f9e44 Remove 2MB alignment from host part of shared allocation 
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
 2022-11-30 17:04:24 +01:00
Maciej Plewka
4b42b066f8 Use dedicated using type for TaskCount 
Related-To: NEO-7155

Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
 2022-11-28 16:44:44 +01:00
Compute-Runtime-Validation
7b195d6b8e Revert "Add support for USM shared in WSL for dGPU" 
This reverts commit 44d8a6ae2b.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
 2022-11-25 08:42:15 +01:00
Kamil Diedrich
44d8a6ae2b Add support for USM shared in WSL for dGPU 
This patch force KMD allocation path for USM shared
Additionally we force 64kb page from lock which is
required to properly program GPU VA

Related-To: NEO-6913
Signed-off-by: Kamil Diedrich kamil.diedrich@intel.com
 2022-11-23 12:41:25 +01:00
Lukasz Jobczyk
e197736b6f Ensure base ptr is passed to PageFaultManager::removeAllocation 
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
 2022-11-17 16:12:31 +01:00
Compute-Runtime-Validation
083168dafb Revert "Add support for USM shared in WSL for dGPU" 
This reverts commit b0c97e49ea.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
 2022-11-16 10:09:42 +01:00
Kamil Diedrich
b0c97e49ea Add support for USM shared in WSL for dGPU 
This patch force KMD allocation path for USM shared

Related-To: NEO-6913
Signed-off-by: Kamil Diedrich kamil.diedrich@intel.com
 2022-11-14 13:56:22 +01:00
Lukasz Jobczyk
ff500e0de6 Add support for external hostptr in shared allocations 
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
 2022-11-09 14:48:01 +01:00
Maciej Plewka
1458602efc Store indirect residency at command queue level 
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
 2022-09-22 14:07:19 +02:00
Compute-Runtime-Validation
643e21631c Revert "Store indirect residency at command queue level" 
This reverts commit ffad5c6c09.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
 2022-09-20 18:12:12 +02:00
Maciej Plewka
ffad5c6c09 Store indirect residency at command queue level 
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>

Related-To: NEO-7211
 2022-09-19 17:01:20 +02:00
Dominik Dabek
16798467ac Add api specific config for allocation cache 
Currently disabled for both opencl and level zero

Related-To: NEO-6893

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
 2022-09-05 21:15:18 +02:00
Dominik Dabek
2d949a9f3d Fix device allocation cache 
Do not use aligned size when storing allocation
Trim allocation cache before deleting devices

Related-To: NEO-6893

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
 2022-08-26 11:15:56 +02:00
Mateusz Jablonski
3d9b5d441a Add UNRECOVERABLE_IF to avoid nullptr dereference 
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
 2022-08-09 15:44:49 +02:00
Milczarek, Slawomir
3e22f30ed5 clEnqueueMigrateMemINTEL with memory prefetch for KMD migrated shared allocation 
Related-To: NEO-6740

Signed-off-by: Milczarek, Slawomir <slawomir.milczarek@intel.com>
 2022-07-14 17:35:22 +02:00
Dominik Dabek
cdfe2ce8ad Feature: Flag for device usm allocation reusing 
With flag enabled, when app calls freeSVMAlloc on device usm allocation,
don't free it immediately but save it,
and try to use it on subsequent allocations.

This allocation cache will be trimmed if an allocation fails.

Related-To: NEO-6893

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
 2022-06-13 20:02:52 +02:00
Artur Harasimiuk
819e0f5515 style: configure readability-identifier-naming.LocalVariableCase 
Signed-off-by: Artur Harasimiuk <artur.harasimiuk@intel.com>
 2022-05-16 12:39:44 +02:00
Dominik Dabek
8d1ad5a4f3 Refactor: use stack vector for root device indices 
Stack vector will not cause dynamic allocations in most circumstances
ie. number of root device indices not more than 16

Related-To: NEO-6837

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
 2022-04-14 14:05:42 +02:00
Compute-Runtime-Validation
7a3976ad64 Revert "Force 64KB page size for cpu alignment in dual storage allocation" 
This reverts commit 7ff6a5c1fa.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
 2022-04-07 00:45:07 +02:00
Kamil Diedrich
7ff6a5c1fa Force 64KB page size for cpu alignment in dual storage allocation 
There is no need to force 2MB alignment for CPU allocation in dual
storage usage. Additionaly for WSL this will allow to avoid usage of
malloc in driver path.

Relates-To: NEO-6620
Signed-off-by: Kamil Diedrich <kamil.diedrich@intel.com>
 2022-04-06 15:28:17 +02:00
Jaime Arteaga
c0e2251ceb Skip adding allocations to remote devices if not allocated there 
When making graphics allocations resident in multi-GPU scenarios,
we should make them resident only if there's an allocation for that
device. So return appropriate null pointer and skip it.

Signed-off-by: Jaime Arteaga <jaime.a.arteaga.molina@intel.com>
 2022-03-16 23:58:07 +01:00
Jaime Arteaga
a010fb3634 Dont close shared handle on imported allocations 
Related-To: LOCI-2272

Signed-off-by: Jaime Arteaga <jaime.a.arteaga.molina@intel.com>
 2022-03-06 04:35:47 +01:00
Lukasz Jobczyk
0634aa3f1b Create resource with given address 
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
 2022-02-24 17:06:19 +01:00
Lukasz Jobczyk
c405fb5c2c Use shared mutex for unified memory manager 
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
 2022-02-23 19:00:38 +01:00
Bartosz Dunajski
4b0d986876 Move AllocationType enum out of GraphicsAllocation class 
Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
 2022-02-04 17:49:09 +01:00
Milczarek, Slawomir
2be98a1e62 Create kmd migrated allocation with initial placement 
Implements ZE_HOST_MEM_ALLOC_FLAG_BIAS_INITIAL_PLACEMENT
for zeMemAllocShared with KMD migrated allocation.

Signed-off-by: Milczarek, Slawomir <slawomir.milczarek@intel.com>
 2022-02-01 15:42:10 +01:00
Dominik Dabek
a5067d6b0b Remove duplicate check in residency container 
Remove find in SVMAllocsManager
addInternalAllocationsToResidencyContainer,
not needed, CSR resolves duplicates at makeResident calls

Co-authored-by: Michal Mrozek <michal.mrozek@intel.com>

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
 2022-01-26 17:09:31 +01:00
Jaime Arteaga
dbf0f90186 Return pageSize in getMemAllocProperties 
Signed-off-by: Jaime Arteaga <jaime.a.arteaga.molina@intel.com>
 2022-01-25 18:26:13 +01:00
Michal Mrozek
6df17f5a30 [3/n] Optimize indirect allocations handling. 
Add new debug variable to trigger new mode.

Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
 2022-01-25 16:40:56 +01:00
Michal Mrozek
52d636394c [1/n] Improve indirect allocations handling. 
Add new functions that would treat all indirect allocations as single pack.

Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
 2022-01-25 06:27:33 +01:00
Michal Mrozek
151aaf7678 Fix alignment for host allocations. 
- it is 4k not 64k.

Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
 2022-01-24 13:24:03 +01:00
Compute-Runtime-Validation
6082865eb4 Revert "Optimize Level Zero indirect allocations handling." 
This reverts commit 3ecbc55ba9.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
 2022-01-20 11:41:13 +01:00
Michal Mrozek
3ecbc55ba9 Optimize Level Zero indirect allocations handling. 
Make them resident directly instead of populating residency container
Remove finds, not needed, CSR resolves duplicates at makeResident calls
Observed gain is 32x for 10k indirect allocations.


Co-authored-by: Michal Mrozek <michal.mrozek@intel.com>

Co-authored-by: Dominik Dabek <dominik.dabek@intel.com>

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
 2022-01-19 13:08:35 +01:00
Bartosz Dunajski
e040c8c6bb Initial compression support for L0 Device allocations 
Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
 2021-12-16 15:55:13 +01:00
Bartosz Dunajski
dfdd3c597a Remove redundant BUFFER_COMPRESSED allocation type 
Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
 2021-12-07 13:35:49 +01:00
Bartosz Dunajski
9481d8a88b Compression refactor [2/n] 
Dont use allocation type for compression preference

Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
 2021-12-03 08:47:46 +01:00
Mateusz Jablonski
b1df167632 Move enable_product.inl to shared 
extract api agnostic validators to shared
remove not needed opencl includes from neo shared


Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
 2021-10-12 22:02:16 +02:00
Michal Mrozek
06561a06b6 Restore 2MB size & alignment for CPU SVM. 
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
 2021-10-11 23:16:42 +02:00
Michal Mrozek
c8e79b7447 Change shared allocations required alignment to 64KB. 
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
 2021-10-08 13:08:11 +02:00
Mateusz Jablonski
48f01f28f5 Move core part of MemoryPropertiesHelpers to shared 
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
 2021-10-07 23:14:38 +02:00
Jaime Arteaga
522d2550ee Add allocation ID to be retrieved by L0 getMemAllocProperties 
zeMemGetAllocProperties() specifies an ID for the allocation can be
retrieved by the application. This can be used by applications to
identify the allocation throughout the application lifetime.

We were currently returning as id the same allocation address. This
patch fixes it to return a true unique identifier for the allocation.

Signed-off-by: Jaime Arteaga <jaime.a.arteaga.molina@intel.com>
 2021-09-15 19:22:50 +02:00
Rafal Maziejuk
bbfbf19a02 Move allowStatelessCompression from HwHelper to HwInfoConfig 
Signed-off-by: Rafal Maziejuk <rafal.maziejuk@intel.com>
Related-To: NEO-4541
 2021-09-09 11:37:47 +02:00
Kamil Diedrich
4390590831 Add shareable allocation for Windows 
Signed-off-by: Kamil Diedrich <kamil.diedrich@intel.com>
 2021-08-26 13:43:55 +02:00
Milczarek, Slawomir
eb14d8458b Add helper function to enable stateless compression 
Related-To: NEO-5107

Signed-off-by: Milczarek, Slawomir <slawomir.milczarek@intel.com>
 2021-08-02 14:34:33 +02:00
Jaime Arteaga
b057935c7c Associate shared-allocations with device in multi-device systems 
When a shared-allocation is created in a multi-device system and
no device is passed during the call, then the allocation needs to be
associated with all the devices. For now, that means we default to
creating the shared-allocation in the host, with no device-backing
memory, so all devices can access it.

However, if a device is passed, then the shared-allocation is
expected to be accessed only by that device, and there is no need
to allocate it solely in the host.

Signed-off-by: Jaime Arteaga <jaime.a.arteaga.molina@intel.com>
 2021-07-29 20:28:27 +02:00