intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 06:49:52 +08:00
Code Issues Packages Projects Releases Wiki Activity
12,303 Commits 43 Branches 655 Tags
172ea34fc28e31e3a3942ca2f914ea70355f4d07
Commit Graph

1630 Commits

Author SHA1 Message Date
Kamil Kopryk
1e0a2b897f Introduce getHelper - helper getter for ClHwHelper and L0HwHelper 
Related-To: NEO-6853
Signed-off-by: Kamil Kopryk <kamil.kopryk@intel.com>
 2022-11-14 15:54:45 +01:00
Lukasz Jobczyk
0a07ea94c7 Enable bcs split for H2H transfers 
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
 2022-11-14 14:23:02 +01:00
Kamil Kopryk
4aa1697e3c Move hwInfoConfig ownership to RootDeviceEnvironment 2/n 
Related-To: NEO-6853
Signed-off-by: Kamil Kopryk <kamil.kopryk@intel.com>

Use RootDeviceEnvironment getHelper<ProductHelper> for
- adjustSamplerState
- adjustPlatformForProductFamily.
 2022-11-14 13:04:31 +01:00
Compute-Runtime-Validation
77b6918f30 Revert "LOCI-3365: Cleanup MediaInterfaceDescriptorLoad logic in command enco... 
This reverts commit cb3f7234f0.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
 2022-11-12 16:05:11 +01:00
Tratnack, Geoffrey
cb3f7234f0 LOCI-3365: Cleanup MediaInterfaceDescriptorLoad logic in command encoder 
Add a patch to command encoder for samplers when DSH is dirty.

Signed-off-by: Tratnack, Geoffrey <geoffrey.tratnack@intel.com>
Related-To: LOCI-3365
 2022-11-10 23:28:58 +01:00
Zbigniew Zdanowicz
be1c1e89cb Add option to signal and synchronize using all available event packets 
Related-To: NEO-7469

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
 2022-11-10 23:28:43 +01:00
Zbigniew Zdanowicz
5d79a84448 Enable front end tracking on xe_hp+ platforms 
Related-To: NEO-5019

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
 2022-11-10 21:14:46 +01:00
Dunajski, Bartosz
5266a95c03 Debug flag to control auto grf mode 
Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
 2022-11-10 19:25:58 +01:00
Zbigniew Zdanowicz
596c9026f0 Enable state compute mode tracking on xe_hp+ platforms 
Related-To: NEO-5019

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
 2022-11-10 18:57:14 +01:00
Dunajski, Bartosz
002184586c Add command buffer helpers: Conditional BB_START and GPR Inc/Dec 
Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
 2022-11-10 18:56:24 +01:00
Kamil Kopryk
002a90c717 Move hwHelper ownership to RootDeviceEnvironment 2/n 
Related-To: NEO-6853
Signed-off-by: Kamil Kopryk <kamil.kopryk@intel.com>

UseRootDeviceEnvironment getHelper<CoreHelper> for:
- getMaxBarrierRegisterPerSlice
- getPaddingForISAAllocation
 2022-11-10 16:39:39 +01:00
Warchulski, Jaroslaw
e4d10e5460 Cleanup includes 4 
Related-To: NEO-5548
Signed-off-by: Warchulski, Jaroslaw <jaroslaw.warchulski@intel.com>
 2022-11-10 14:22:18 +01:00
Zbigniew Zdanowicz
b7203d9ce5 Enable pipeline select tracking for all xe_hp+ platforms 
Related-To: NEO-5019

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
 2022-11-10 13:19:53 +01:00
Jaime A Arteaga Molina
192f02785d Use PRELIM_DRM_I915_QUERY_FABRIC_INFO for canAccessPeer when available 
When available, PRELIM_DRM_I915_QUERY_FABRIC_INFO is used to query
connectivity between two devices. If not, then a copy is performed.

Signed-off-by: Jaime A Arteaga Molina <jaime.a.arteaga.molina@intel.com>
 2022-11-10 07:30:32 +01:00
Kamil Kopryk
af3f370ff0 Correct event reset 
Related-To: NEO-7391
Signed-off-by: Kamil Kopryk <kamil.kopryk@intel.com>
 2022-11-09 17:20:22 +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
Kacper Nowak
481cf36ad1 Add recompilation warning to debug messages 
In case of recompilation from IR, apart from updating build log,
print debug message with recompilation warning.

Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
 2022-11-09 14:19:54 +01:00
John Falkowski
6b64c1b04b Return error from DeviceImp::createImage call when images are not supported 
Signed-off-by: John Falkowski <john.falkowski@intel.com>
 2022-11-09 13:36:08 +01:00
Zbigniew Zdanowicz
3169b4afd6 Enable command list heap sharing on xe_hp+ platforms 
Related-To: NEO-7142

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
 2022-11-09 12:58:23 +01:00
Maciej Bielski
c06ddfc7b8 Allocate kernel private memory for xehp and later 
Add missing allocation of kernel private memory for the scenario when
the private memory is not allocated within `KernelImp::initialize()` but
deferred until `appendLaunchKernelWithParams()` instead.

One kernel can never allocate more private/scratch memory than
`globalMemorySize`, that ends up in `ZE_RESULT_ERROR_OUT_OF_DEVICE_MEMORY`
being returned. However, several separate kernels can exceed the
`globalMemorySize` and then, the private region of each such kernel is
allocated at later stage, in `appendLaunchKernelWithParams()`.

Such mechanism was present on pre-xehp platforms and it is now added to
xehp-and-later.

See:
* ModuleImp::checkIfPrivateMemoryPerDispatchIsNeeded()
* Module::shouldAllocatePrivateMemoryPerDispatch()

Related-To: NEO-7398
Signed-off-by: Maciej Bielski <maciej.bielski@intel.com>
 2022-11-08 19:10:26 +01:00
Dominik Dabek
30fe24aa79 Avoid cmpexchg due to CPU Hardware limitation 
Limit the amount of times compare_exchange_weak is called,
to avoid issues with contention when multiple cpu cores request
the same address.

Related-To: NEO-7030

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
 2022-11-08 17:11:52 +01:00
Jim Snow
48ba0554db Allocate RTDispatchGlobals as array-of-structures. 
This fixes several bugs in previous (reverted) implementation.
We use correct RTStack pointer offset, and a larger RTStack size.

Related-To: LOCI-2966

Signed-off-by: Jim Snow <jim.m.snow@intel.com>
 2022-11-07 21:25:32 +01:00
Mateusz Jablonski
5b11a4a5fa Return submission status from flushHandler function 
Related-To: NEO-7412
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
 2022-11-07 15:09:33 +01:00
Warchulski, Jaroslaw
6cbb3cfb05 Cleanup includes 3 
Related-To: NEO-5548
Signed-off-by: Warchulski, Jaroslaw <jaroslaw.warchulski@intel.com>
 2022-11-07 14:52:31 +01:00
Yates, Brandon
99ef6c499b L0 debug - fix windows bitmask decode 
Keep threads created in EU range (0,7)

Signed-off-by: Yates, Brandon <brandon.yates@intel.com>
 2022-11-07 14:41:29 +01:00
Dunajski, Bartosz
67af920281 Unify programming of ending commands in direct submission path 
Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
 2022-11-07 10:56:36 +01:00
Warchulski, Jaroslaw
fb25f96081 Cleanup includes 2 
Related-To: NEO-5548
Signed-off-by: Warchulski, Jaroslaw <jaroslaw.warchulski@intel.com>
 2022-11-07 10:36:50 +01:00
Compute-Runtime-Validation
ddbaa5e8c9 Revert "Cleanup MediaInterfaceDescriptorLoad logic in command encoder" 
This reverts commit 349af0bd5e.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
 2022-11-07 02:20:27 +01:00
Tratnack, Geoffrey
349af0bd5e Cleanup MediaInterfaceDescriptorLoad logic in command encoder 
Add a patch to command encoder for samplers when DSH is dirty.

Signed-off-by: Tratnack, Geoffrey <geoffrey.tratnack@intel.com>
Related-To: LOCI-3365
 2022-11-06 03:27:23 +01:00
Zbigniew Zdanowicz
27d042107a Add feature to estimate number of event packets 
Related-To: NEO-7469

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
 2022-11-04 18:38:17 +01:00
Luzynski, Sebastian Jozef
91c69e0fe7 Patch global variables and constants once 
This change replaces mechanism of patching global constants and
variables in kernel per relocation to patching them only once. This
would improve linking time performance for kernels with multiple global
symbols.

Signed-off-by: Luzynski, Sebastian Jozef <sebastian.jozef.luzynski@intel.com>
 2022-11-03 17:45:09 +01:00
Zbigniew Zdanowicz
6a6ab80113 Add option to compact event L3 flush packet 
Related-To: NEO-7434

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
 2022-11-03 14:23:42 +01:00
Zbigniew Zdanowicz
b284b727e9 Fix number of available events in IPC event pools 
Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
 2022-11-03 13:06:28 +01:00
Zbigniew Zdanowicz
29eee35491 pass event host scope flag to walker when single packet events are used 
Related-To: NEO-7434

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
 2022-11-03 13:06:07 +01:00
Maciej Plewka
8a9ea9afd7 Make migration for indirect allocations 
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
 2022-11-02 14:53:14 +01:00
Maciej Plewka
7f38c5e633 Revert "Return error code for unsuported image arg in gen12lp" 
This reverts commit bbc31e6aac


Signed-off-by: Maciej Plewka maciej.plewka@intel.com
 2022-11-02 12:57:16 +01:00
Maciej Plewka
ff01b9361e Return error code when there is no space for scratch/private 
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
 2022-11-02 11:55:18 +01:00
Mateusz Jablonski
51fa04fc60 L0: Add experimental extensions for wait and write on memory 
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
 2022-11-02 10:47:39 +01:00
Mateusz Jablonski
9816f815f3 Propagate exec buffer error to L0 API level on Xe HPC 
This change makes that drm file is opened in nonblocking mode for prelim
kernels. In such case when calling exec buffer ioctl and get
EAGAIN (aka EWOULDBLOCK) we may return error to API level

Related-To: NEO-7144

Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
 2022-10-31 10:09:13 +01:00
Compute-Runtime-Validation
33d9c5979a Revert "Enable flush task for immediate command lists" 
This reverts commit 95c3ef28fc.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
 2022-10-29 15:48:44 +02:00
Aravind Gopalakrishnan
95c3ef28fc Enable flush task for immediate command lists 
Previously enabled for: DG2, PVC.
With this commit enabling for Gen9 onwards.

Related-To: LOCI-3379

Signed-off-by: Aravind Gopalakrishnan <aravind.gopalakrishnan@intel.com>
 2022-10-28 19:17:42 +02:00
Compute-Runtime-Validation
d8a66cd14d Revert "image create check" 
This reverts commit 333906e278.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
 2022-10-28 03:25:32 +02:00
Dominik Dabek
526ba1bde5 Fix l0 kernel set arg buffer caching 
Fix for incorrect cache hit if alloc id was uninitialized
and allocations counter was the same.

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
 2022-10-27 17:25:10 +02:00
Zbigniew Zdanowicz
daa26701e4 Use pipe control to signal event of multi kernel operations 
Related-To: NEO-7434

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
 2022-10-27 15:08:09 +02:00
Jaime A Arteaga Molina
5446b9ca0d Update Fabric Latency to Edge Properties (2) 
- Only initialize vertexes when queried.
- Return also bandwidth when calling PRELIM_DRM_I915_QUERY_FABRIC_INFO.

Related-To: LOCI-3464

Signed-off-by: Jaime A Arteaga Molina <jaime.a.arteaga.molina@intel.com>
 2022-10-26 08:15:35 +02:00
John Falkowski
333906e278 image create check 
Signed-off-by: John Falkowski <john.falkowski@intel.com>
 2022-10-25 18:33:37 +02:00
Szymon Morek
f5355912ab Synchronize before CPU memcpy after wait on events 
Currently we track dependencies only in append
barrier call. This commit adds tracking for
append waitOnEvents so CPU memcpy correctly
waits for such dependency.

Signed-off-by: Szymon Morek <szymon.morek@intel.com>
 2022-10-25 11:51:21 +02:00
Zbigniew Zdanowicz
44d7420893 Enable state compute mode tracking for XE_HPG 
Related-To: NEO-5019

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
 2022-10-25 09:31:14 +02:00
Joshua Santosh Ranjan
83b4edddf0 Update Fabric Latency to Edge Properties 
Related-To: LOCI-3464

Signed-off-by: Joshua Santosh Ranjan <joshua.santosh.ranjan@intel.com>
 2022-10-25 03:23:53 +02:00
Aravind Gopalakrishnan
fcd0dd95a1 Activate metric group for immediate command list 
Related-To: LOCI-3518

Signed-off-by: Aravind Gopalakrishnan <aravind.gopalakrishnan@intel.com>
 2022-10-25 00:05:33 +02:00