Commit Graph

256 Commits

Author SHA1 Message Date
Bartosz Dunajski b1dea19fbd refactor: move tag initialization to allocator [1/n]
Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
2024-12-17 17:53:13 +01:00
Dominik Dabek e61d04a881 fix(ocl): track buffer pool count per device
Track amount of created buffer pools per device. Do not allocate extra
pools if limit is reached. New contexts will have pooling disabled if
limit is reached on device.

Related-To: NEO-13461

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
2024-12-16 11:38:05 +01:00
Marcel Skierkowski e27a6dc280 refactor: use virtualFileSystem in ULTs
reducing the number of tests that have interactions with filesystem.
writeDataToFile() saves filename and content in std::map.
fileExistsHasSize() checks if file was previously written to virtualFileSystem
loadDataFromVirtualFile() fetches data from std::map based on filename

Related-To: NEO-7006
Signed-off-by: Marcel Skierkowski <marcel.skierkowski@intel.com>
2024-12-13 17:57:07 +01:00
Szymon Morek 6c4eb322b1 performance: introduce staging reads from image
Related-To: NEO-12968

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
2024-12-11 14:43:45 +01:00
Fabian Zwoliński 5f8e761541 fix: HeapAllocator - ensure getBaseAddress returns initial base address
getBaseAddress was incorrectly returning pLeftBound which changes after
memory allocation.
Added baseAddress field to store and return initial address value.

Signed-off-by: Fabian Zwoliński <fabian.zwolinski@intel.com>
2024-12-10 10:52:36 +01:00
Compute-Runtime-Validation 58e45afd39 Revert "fix: HeapAllocator - ensure getBaseAddress returns initial base address"
This reverts commit ffec97acc5.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
2024-12-08 14:35:22 +01:00
Fabian Zwoliński ffec97acc5 fix: HeapAllocator - ensure getBaseAddress returns initial base address
getBaseAddress was incorrectly returning pLeftBound which changes after
memory allocation.
Added baseAddress field to store and return initial address value.

Signed-off-by: Fabian Zwoliński <fabian.zwolinski@intel.com>
2024-12-05 20:42:16 +01:00
Fabian Zwoliński d2ce3badfc fix: bindlessHeapsHelper handle unavailable external heap
This PR handles the situation in which a component
has reserved a front window space for itself in the external heap,
so that the Compute Runtime cannot access this area.

In such a situation, we perform the following steps:
1. reserve 4GB chunk in heapStandard
2. split our chunk into 2 parts: heapFrontWindow, heapRegular
3. from this point on, map all linearStream allocations in reserved 4GB
chunk

Patch applies to Windows and WSL.
Patch only applies when the bindless global allocator is enabled.

Related-To: HSD-16025889919
Signed-off-by: Fabian Zwoliński <fabian.zwolinski@intel.com>
2024-12-05 14:18:01 +01:00
Szymon Morek 16684d8803 fix: reuse staging buffers from other csr
Related-To: NEO-13247

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
2024-11-18 16:39:57 +01:00
Bartosz Dunajski 7bf22ed33e feature: counter based allocation peer sharing
Related-To: NEO-13079

Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
2024-11-13 15:01:32 +01:00
Jaroslaw Warchulski 72efceb8a3 fix: try to lower alignment in heap allocator when it is too big
Related-To: GSD-8948
Signed-off-by: Jaroslaw Warchulski <jaroslaw.warchulski@intel.com>
2024-11-13 09:14:12 +01:00
Szymon Morek 1199e19df5 performance: enable image writes through staging chunks
Related-To: NEO-12968

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
2024-11-07 13:43:25 +01:00
Szymon Morek a25e973205 performance: write image through staging chunks
Related-To: NEO-12968

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
2024-11-06 16:45:22 +01:00
Mrozek, Michal 2febf0597e fix: remove incorrect code
- freed pointer cannot overlap with freed chunks, therefore there is no way
to merge it with freed chunks if they overlap

Signed-off-by: Mrozek, Michal <michal.mrozek@intel.com>
2024-11-05 18:58:41 +01:00
Lukasz Jobczyk fe82425fa9 performance: Prefer to assign new gpu va rather than reuse
Resolves: NEO-10734

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2024-11-05 10:16:03 +01:00
Szymon Morek 5d62be2bea performance: enable staging buffer for write image
Related-To: NEO-12968

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
2024-11-05 06:06:07 +01:00
Compute-Runtime-Validation aca7875dfd Revert "performance: Prefer to assign new gpu va rather than reuse"
This reverts commit 0f2f3c3764.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
2024-10-31 05:06:43 +01:00
Szymon Morek cf58be4142 performance: use staging buffer when writing to an image
Related-To: NEO-12968

Also, don't import usm/mapped allocations for image
operations

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
2024-10-30 17:02:13 +01:00
Lukasz Jobczyk 0f2f3c3764 performance: Prefer to assign new gpu va rather than reuse
Resolves: NEO-10734

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2024-10-30 16:48:49 +01:00
Maciej Bielski 45e78fea76 fix: use productHelper in getPatIndexInfoString() on Windows
Fix the PAT-index reporting in logger as currently on Windows reported
values are simply wrong.

The changed logic dependends on `RootDeviceEnvironment` and in order to
avoid introducing such dependencies into logger.[ch] the
`logAllocation()` is no longer a member of `FileLogger` but
a free-function instead (and a separate .cpp file). This is important
because the source files `logger.[ch]` are also used by ocloc library
and there is no point to contaminate ocloc code structure with
unnecessary dependencies.

Related-To: NEO-9421
Signed-off-by: Maciej Bielski <maciej.bielski@intel.com>
2024-10-22 19:27:13 +02:00
Mateusz Jablonski 7b54ca131f fix: avoid OOB access in StackVec::resizeImpl
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2024-09-19 13:54:24 +02:00
Mateusz Jablonski ebfa384a74 fix: avoid calling free on non-heap object
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2024-09-13 16:34:23 +02:00
Michal Mrozek dd631610b3 refactor: move memory tracking to memory manager
- remove wddm specific code
- improve total size reported to be in decimal

Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
2024-09-12 17:32:38 +02:00
Michal Mrozek da59b88122 refactor: improve logging of allocations
add capability to measure total amount of allocated memory
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
2024-09-12 11:00:27 +02:00
Michal Mrozek 7cba813c81 refactor: improve logging by adding size.
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
2024-09-11 20:33:55 +02:00
Fabian Zwoliński 38e1614f4a fix: create and use new allocation type for syncBuffer
Related-To: NEO-11533
Related-To: HSD-18039788811
Signed-off-by: Fabian Zwoliński <fabian.zwolinski@intel.com>
2024-09-06 14:11:33 +02:00
Mateusz Jablonski c934877790 refactor: remove not needed function
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2024-08-30 12:18:14 +02:00
Igor Venevtsev fb8921df97 fix: enable file logger for ReleaseInternal configuration
Signed-off-by: Igor Venevtsev <igor.venevtsev@intel.com>
2024-08-26 10:54:41 +02:00
Filip Hazubski 556ba1b7c4 performance: Avoid data copy in getPciStats function
Additionally explicitly remove unused functions of Directory and
StagingBuffer classes to comply with rule of five.

Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
2024-08-22 13:30:01 +02:00
Winston Zhang 0590b34cfa feature: refactor and rewrite setErrorDescription
Related-To: NEO-8379

Signed-off-by: Winston Zhang <winston.zhang@intel.com>
2024-08-21 17:26:25 +02:00
Oskar Hubert Weber 3ccab79ed8 test: dir create
Related-To: NEO-11500

Signed-off-by: Oskar Hubert Weber <oskar.hubert.weber@intel.com>
2024-08-05 13:58:25 +02:00
Bartosz Dunajski c3312f21f7 feature: dont initialize in-order TS nodes
Related-To: NEO-11925

Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
2024-07-23 13:31:23 +02:00
Filip Hazubski f8867ac3ac fix: Minor code improvements
Add explicit pointer checks to CommandQueue::blitEnqueueAllowed.

Explicitly check result of getDeviceArgValueIdx during ocloc compile.

Explicitly remove unused StagingBufferManager functions.
Move chunkCopyFunc by reference.

Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
2024-07-19 09:24:37 +02:00
Morek, Szymon ed7fc9acc9 performance: use staging buffer as a pool
Related-To: NEO-11501

Currently whole staging buffer is consumed even if
size of the transfer is smaller than that buffer.
This commit changes that, so single staging buffer
might be utilized by several smaller transfers
as long as they don't exceed total size

Signed-off-by: Morek, Szymon <szymon.morek@intel.com>
2024-07-18 12:54:00 +02:00
Szymon Morek dbd96372be performance: adjust staging buffer usage
Related-To: NEO-11928

Don't copy through staging buffer if dst usm allocation
was not used before and transfer would be splitted.
Also, don't use staging buffers for mapped ocl buffers.

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
2024-07-10 10:19:18 +02:00
Mateusz Hoppe 96c9956d50 refactor: remove debug break
Signed-off-by: Mateusz Hoppe <mateusz.hoppe@intel.com>
2024-07-04 11:00:54 +02:00
Oskar Hubert Weber a7ddf7848b fix: fatbinary generation with "-out_dir"
Directories are created if needed before generating fatbinary.

Related-To: NEO-11500

Signed-off-by: Oskar Hubert Weber <oskar.hubert.weber@intel.com>
2024-07-02 13:33:53 +02:00
Szymon Morek 29e3eb512c performance: non-usm copy through staging buffers
Related-To: NEO-11501

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
2024-06-25 07:18:53 +02:00
Szymon Morek 34e5e18fa6 fix: delete move assignment operator
Signed-off-by: Szymon Morek <szymon.morek@intel.com>
2024-05-31 18:59:51 +02:00
Compute-Runtime-Validation de789ac7e5 Revert "performance: remove page size limit for sharing ISAs"
This reverts commit e7c036a91b.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
2024-05-12 09:00:32 +02:00
Szymon Morek e7c036a91b performance: remove page size limit for sharing ISAs
Related-To: NEO-9403

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
2024-05-10 08:56:49 +02:00
Szymon Morek 10ed479b16 performance: share inter-module ISA allocations
Related-To: NEO-10258

Currently each module has it's own GA
for kernel ISA's. This change allows new modules to
reuse existing allocation.

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
2024-05-09 08:43:55 +02:00
Andrzej Koska b499973658 refactor: removing unnecessary lock
Related-To: NEO-8196

Lock in the constructor should not affect the flow of the code

Signed-off-by: Andrzej Koska <andrzej.koska@intel.com>
2024-04-23 17:39:50 +02:00
Bartosz Dunajski 74dcb5a9e3 feature: initial support for sync dispatch token allocation
Related-To: NEO-8171

Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
2024-03-26 12:07:58 +01:00
Zbigniew Zdanowicz fe377b6597 feature: add out cmd pointer to store data imm encoder
Related-To: NEO-10385

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2024-03-07 12:52:08 +01:00
Dunajski, Bartosz bbe1043f08 feature: initial support for pooling in-order counter allocations
Related-To: NEO-10507

Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
2024-03-01 12:08:02 +01:00
Maciej Plewka 964f0166d7 fix: Reuse freed chunk from left even if not aligned
Related-To: NEO-10416, NEO-10418, NEO-10437

Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
2024-02-21 12:31:54 +01:00
Chodor, Jaroslaw b58717b9e3 feature: Add support for legacy acronyms in ocloc's fatbinary
Expands support for deprecated acronyms to fatbinary. Previously,
these were allowed only in single-target builds.

Related-To: NEO-10190

Signed-off-by: Chodor, Jaroslaw <jaroslaw.chodor@intel.com>
2024-02-15 10:14:49 +01:00
Maciej Plewka 233f898b59 fix: split freed chunk from right
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>

Related-To: NEO-10280
2024-02-13 16:14:39 +01:00
Lukasz Jobczyk 486cc71b76 refactor: Add GDI profiling
Resolves: NEO-9236
Related-To: NEO-10036

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2024-02-07 18:44:11 +01:00