Commit Graph

3388 Commits

Author SHA1 Message Date
Zbigniew Zdanowicz
ef5efeac0c refactor: add kernel properties and move implementations to imp class
Related-To: NEO-15374

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2025-07-15 15:08:44 +02:00
Shreyas Kunder
e4158208ae feature: Update Metrics Driver Code to use runtime enable & Disable APIs
Related-To: NEO-15339

Signed-off-by: Shreyas Kunder <shreyas.kunder@intel.com>
2025-07-15 14:28:49 +02:00
Tomasz Biernacik
2c5cbec033 feature: control post sync completion check
Related-To: NEO-14844

Signed-off-by: Tomasz Biernacik <tomasz.biernacik@intel.com>
2025-07-15 13:51:19 +02:00
Zbigniew Zdanowicz
087d1ecea4 refactor: use encoders instead direct dispatch commands in mcl experimental
Related-To: NEO-10492

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2025-07-15 12:52:13 +02:00
Radoslaw Jablonski
be6fe9294c fix: stop Direct Submission on device teardown
This prevents a race condition, causing page faults, where device related
resources are released, but are still in use by commands (e.g. cache flushes)
submitted by the Direct Submission Controller.

Related-To: NEO-15017

Signed-off-by: Radoslaw Jablonski <radoslaw.jablonski@intel.com>
2025-07-15 12:12:59 +02:00
Zbigniew Zdanowicz
b62813297d refactor: move add kernel arg stateful buffer to experimental implementation
Related-To: NEO-10492

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2025-07-15 01:02:26 +02:00
Mateusz Jablonski
8dc24d9e2c feature: add default command queue descriptor to ze_intel_gpu.h
Related-To: NEO-14560
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2025-07-14 16:24:18 +02:00
Zbigniew Zdanowicz
1aefbd709b performance: reserve space for internal vectors
- reserved space avoids vector memory relocations

Related-To: NEO-13916

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2025-07-14 13:13:55 +02:00
Zbigniew Zdanowicz
d31366794a fix: skip and continue kernel isa mutation when same kernel is provided
Related-To: NEO-10492

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2025-07-11 20:28:51 +02:00
Mateusz Jablonski
762b04cf77 feature: add command queue flag to pass copy offload hint
Related-To: NEO-14560
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2025-07-11 19:42:48 +02:00
Mateusz Jablonski
645de5add8 fix: correct shared cross device alloc capabilities
allow for access in case of multiple exposed devices when
there is only one root device

Related-To: HSD-18042639532
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2025-07-11 18:32:06 +02:00
Zbigniew Zdanowicz
1599e5bafe feature: add missing experimental mcl api calls
Related-To: NEO-10492

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2025-07-11 18:23:37 +02:00
Mateusz Jablonski
8f1903c7dd feature: add zerDriverGetLastErrorDescription API
this function retrieves last error from default driver

Related-To: NEO-14560
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2025-07-11 13:37:50 +02:00
Aravind Gopalakrishnan
c1c8eb59c0 refactor: Cache context settings during init
- cache context related settings once to reuse in subsequent calls

Signed-off-by: Aravind Gopalakrishnan <aravind.gopalakrishnan@intel.com>
2025-07-10 12:06:20 +02:00
Aleksandra Nizio
c9ebf91271 fix: Add -emit-lib-compile-errors flag when library compilation is enabled
Related-To: NEO-8608
Signed-off-by: Aleksandra Nizio <aleksandra.nizio@intel.com>
2025-07-10 10:48:21 +02:00
Mateusz Hoppe
703497b067 fix: read ONEAPI_PVC_SEND_WAR_WA env
- disable optimization with compiler internal option when env is set to
zero

Related-To: NEO-15378, GSD-10884

Signed-off-by: Mateusz Hoppe <mateusz.hoppe@intel.com>
2025-07-09 16:22:18 +02:00
Zbigniew Zdanowicz
fc37b98b69 fix: provide fixes for kernel isa mutation of cooperative kernels
- update correctly mutable residency of cooperative kernel barrier allocation
- fix disable noop patching when mutate into non cooperative kernel
- reset mutation variables for not default kernels in mutable kernel group

Related-To: NEO-10492

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2025-07-09 16:22:00 +02:00
Zbigniew Zdanowicz
ec0e493ca8 fix: add initial calculation max thread group count for cooperative
Related-To: NEO-10492

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2025-07-08 21:23:24 +02:00
Mateusz Jablonski
2661fd9522 feature: define default descriptors for counter based event and USM
user can use default descriptor to simplify creation

Related-To: NEO-14560
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2025-07-08 16:10:23 +02:00
Mateusz Jablonski
a345fa0783 fix: remove not needed printf from production code
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2025-07-08 12:01:52 +02:00
Zbigniew Zdanowicz
c1ec5db4fc refactor: simplify conditions and add unit tests for mcl
- remove always true conditions
- move experimental code to experimental implementations
- add missing unit tests
- extend existing unit tests

Related-To: NEO-10492

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2025-07-08 11:07:10 +02:00
Maciej Plewka
6951ff1a07 fix: store whole global timestamp before context timestamp
Related-To: NEO-15192
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
2025-07-07 21:34:08 +02:00
Compute-Runtime-Validation
d952298321 Revert "fix: skip cleanup in L0 DllMain when terminating process"
This reverts commit b70c7fd078.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
2025-07-07 17:42:50 +02:00
Chodor, Jaroslaw
58228a36fe feature: Adding basic record and replay mechanism
Related-To: NEO-15373

Signed-off-by: Chodor, Jaroslaw <jaroslaw.chodor@intel.com>
2025-07-07 15:12:57 +02:00
Lukasz Jobczyk
67462c4356 fix: Correct alignment check for immediate fill v2
Resolves: HSD-18042731538

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2025-07-07 11:43:53 +02:00
Compute-Runtime-Validation
aa37604e7c Revert "fix: skip teardown clean up when terminating process on Windows"
This reverts commit 66bf978a13.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
2025-07-05 06:28:05 +02:00
Bartosz Dunajski
07fb7ac02e fix: copy only mi_atomic signaling for aggregated events
Related-To: NEO-14557

Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
2025-07-04 14:46:47 +02:00
Oskar Hubert Weber
66bf978a13 fix: skip teardown clean up when terminating process on Windows
Per https://learn.microsoft.com/en-us/windows/win32/dlls/dllmain,
it's not safe to clean up resources in DllMain when
fdwReason == DLL_PROCESS_DETACH and lpvReserved != NULL (the process is
terminating), e.g. can result in a hang in some cases.

Related-To: NEO-14121

Signed-off-by: Oskar Hubert Weber <oskar.hubert.weber@intel.com>
2025-07-04 14:05:39 +02:00
Lukasz Jobczyk
c2e6f5daa3 fix: Move in order counter flush one layer down to include all API calls
Resolves: GSD-11304

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2025-07-03 13:42:35 +02:00
Bartosz Dunajski
fba837a7ca fix: disable in-order counter patching
Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
2025-07-03 07:47:02 +02:00
Mateusz Jablonski
42826b562d fix: respect pNext extensions in zeCommandListAppendLaunchKernelWithArguments
Related-To: NEO-14560
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2025-07-02 15:27:49 +02:00
Oskar Hubert Weber
b70c7fd078 fix: skip cleanup in L0 DllMain when terminating process
Per https://learn.microsoft.com/en-us/windows/win32/dlls/dllmain,
it's not safe to cleanup resources in DllMain when
fdwReason == DLL_PROCESS_DETACH and lpvReserved != NULL (the process is
terminating), e.g. can result in a hang in some cases.

Related-To: NEO-14121

Signed-off-by: Oskar Hubert Weber <oskar.hubert.weber@intel.com>
2025-07-02 11:19:45 +02:00
Neil R. Spruit
1f862955a8 fix: Set isValidFlag for DDI tables during DriverDispatch constructor
- Fix to setting of isValidFlag such that the Driver Dispatch
constructor correctly sets the isValidFlag during setup.

Related-To: NEO-15419

Signed-off-by: Neil R. Spruit <neil.r.spruit@intel.com>
2025-07-02 00:13:53 +02:00
Neil R. Spruit
1abaf40799 feature: RTAS EXT support
- Added Support for the RTAs Extension support replacing the Exp support
which will remain for backwards compatability.

Related-To: NEO-15257

Signed-off-by: Neil R. Spruit <neil.r.spruit@intel.com>
2025-07-01 21:36:40 +02:00
Bartosz Dunajski
e2dff82741 feature: remote copy support for bcs split
Related-To: NEO-14557

Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
2025-07-01 15:57:02 +02:00
Chodor, Jaroslaw
f35bdeb2fe feature: Adding Record and Replay API entrypoints
Related-To: NEO-15373

Signed-off-by: Chodor, Jaroslaw <jaroslaw.chodor@intel.com>
2025-07-01 14:43:13 +02:00
Szymon Morek
ca760712f6 fix: flush cache when waiting for event from barrier
Related-To: NEO-15387

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
2025-07-01 13:33:39 +02:00
Lukasz Jobczyk
64a80aef2e fix: Correct alignment check for immediate fill
Resolves: HSD-18042728025

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2025-06-30 16:33:50 +02:00
Lukasz Jobczyk
4010ff6908 performance: Use tag allocator for fill pattern
Resolves: NEO-9729

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2025-06-30 15:24:08 +02:00
Bartosz Dunajski
78c6751be1 feature: initial support for multitile bcs split
Related-To: NEO-14557

Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
2025-06-30 12:38:58 +02:00
Lukasz Jobczyk
4bc1cf0c7a refactor: Simplify alignment check for immediate fill
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2025-06-30 11:35:09 +02:00
Bartosz Dunajski
76678964d6 fix: disable external CB event completion caching
Related-To: NEO-14557

Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
2025-06-26 16:33:02 +02:00
Lukasz Jobczyk
198762baa8 performance: Store pattern allocations on reset and sync
Related-To: NEO-9729

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2025-06-26 14:00:44 +02:00
Lukasz Jobczyk
100bec3fa8 performance: Use immediate fill for pattern sizes <= 4
Related-To: NEO-9729

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2025-06-26 14:00:29 +02:00
Bartosz Dunajski
72532d035e feature: marker event support for aggregated bcs split
Related-To: NEO-14557

Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
2025-06-26 12:18:16 +02:00
Mateusz Hoppe
f1249a9e2b refactor: validateLaunchParams
Signed-off-by: Mateusz Hoppe <mateusz.hoppe@intel.com>
2025-06-26 10:27:25 +02:00
Zbigniew Zdanowicz
442c11b5ba feature: add mutable command list implementation
Related-To: NEO-10492

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2025-06-25 16:32:08 +02:00
Dominik Dabek
5c7fd72d5f performance: enable l0 usm reuse
Do not reuse internal or imported allocations.
Mark as aub writable on getting from reuse.

Related-To: NEO-6893

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
2025-06-25 15:49:15 +02:00
Naklicki, Mateusz
9603d1f2a7 fix: initialize CommandToPatch member
Also: adjusted tests for optional scratch, reordered members for best
alignment and removed unused variable

Related-To: NEO-14130
Signed-off-by: Naklicki, Mateusz <mateusz.naklicki@intel.com>
2025-06-25 13:07:17 +02:00
Neil R. Spruit
532c663da6 feature: Add Initial Support for Get Vector Width Properties
- Added support for L0 to query Vector Width Properties which allows for
querying all the supported vector widths and their prefferred and
native vector widths per data type.
- Updated shared code to support querying the same values thru
getPreferredVectorWidth* and getNativeVectorWidth* helper functions
for OpenCL and Level Zero.
- Initial support for reporting base values for the vector
widths with future support for reporting different values per platform and
based on the SIMD/SIMT size used.

Related-To: NEO-13745

Signed-off-by: Neil R. Spruit <neil.r.spruit@intel.com>
2025-06-25 01:11:30 +02:00