Commit Graph

110 Commits

Author SHA1 Message Date
9b2ad0c5df Detect GPU hangs in flushBcsTask()
This change introduces detection of GPU hangs in flushBcsTask()
function. The new code has been covered with ULTs.

Related-To: NEO-6681
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
2022-05-12 16:17:16 +02:00
910871a706 add multi-tile support for completion diagnostic
Related-To: NEO-6871

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2022-05-11 13:37:32 +02:00
ffd1c430b1 feature: use completion fence value from direct submission when available
Related-To: NEO-6643
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2022-04-28 15:38:11 +02:00
a3745c28a3 Create peekGmmHelper function in CSR
Related-To: NEO-6523
Signed-off-by: Krzysztof Gibala <krzysztof.gibala@intel.com>
2022-04-28 12:54:32 +02:00
72f6ef7d2b Do not submit to gpgpu during blit in immediate dispatch mode
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2022-04-28 11:36:42 +02:00
03185f7111 feature direct submission: use tag allocation as a completion fence
use tag allocation address as a completion address in exec call
wait for completion value before destroying drm direct submission

Related-To: NEO-6643
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2022-04-25 13:30:55 +02:00
8f38f4ee3c fix command stream receiver: stop calling virtual methods in dtor
make getCompletionAddress and getCompletionValue non-virtual methods

Related-To: NEO-6643
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2022-04-20 17:41:54 +02:00
b5b8762faf refactor direct submission: Pass command stream receiver to ctor
Related-To: NEO-6643

Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2022-04-19 23:02:55 +02:00
b67b1bd6fc Download tag allocation only if was submitted
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2022-04-13 16:30:47 +02:00
fd45ac133d Unify append calls
Related-To: NEO-6242

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2022-04-07 10:18:55 +02:00
4044320301 Limit allocations in execute command list
Reserve space in ResidencyContainer

Related-To: NEO-6837

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
2022-04-06 15:00:34 +02:00
9d502dea25 Refactor waitOnCompletionFence method in DrmMemoryManager
get completion address and value from command stream receiver

Related-To: NEO-6643

Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
2022-03-31 19:30:16 +02:00
308f9ccfff Do not allocate dynamic state heap if not needed
Dynamic state heap is only used for sampler data.

Related-To: NEO-6821

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
2022-03-31 12:13:13 +02:00
a285712cc4 Add missing download allocation calls
Signed-off-by: Jobczyk, Lukasz <lukasz.jobczyk@intel.com>
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2022-03-31 09:49:22 +02:00
a230f267e1 Poll task count indefinitely on high throttle command queue
Resolves: NEO-6781

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2022-03-25 10:06:16 +01:00
32b0f7b014 Remove redundant value CsrSizeRequestFlags::numGrfRequiredChanged
Related-To: NEO-5995

Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
2022-03-15 15:08:15 +01:00
999c6424a4 While enqueue blit do not flush gpgpu if already flushed
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2022-03-03 13:01:57 +01:00
7f729b7f89 Detect GPU hang in clWaitForEvents
This change:
- moves NEO::WaitStatus to a separate file
- enables detection of GPU hang in clWaitForEvents
- adjusts most of blocking calls in CommandStreamReceiver to return WaitStatus
- adds ULTs to cover the new code

Related-To: NEO-6681
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
2022-02-23 13:33:09 +01:00
1275c4e200 Detect GPU hang in remaining blocking calls of L0
This change introduces detection of GPU hangs in
zeEventHostSynchronize and zeFenceHostSynchronize.
Furthermore, if CommandQueueHw::executeCommandLists
uses ZE_COMMAND_QUEUE_MODE_SYNCHRONOUS and hang occurs,
the information about it is propagated to the caller.

Related-To: NEO-6681
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
2022-02-16 14:47:29 +01:00
52c6973e89 Rename blitBuffer method
Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
2022-02-03 15:02:55 +01:00
9d8ce7aace Command container appends BB_END on cmd buffer allocation end
When linear stream created for command container has not enough space
for command and BB_END it will program BB_END and allocate new command
buffer allocation. Pointer returned from getSpace in this case will
return storage from new command buffer allocation.

Related-To: NEO-5707

Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
2022-01-31 16:15:37 +01:00
498cf5e871 Implement GPU hang detection
This change uses DRM_IOCTL_I915_GET_RESET_STATS to detect
GPU hangs. When such situation is encountered, then
zeCommandQueueSynchronize returns ZE_RESULT_ERROR_DEVICE_LOST.

Related-To: NEO-5313
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
2022-01-31 13:48:17 +01:00
f8c104feaa Use fw declaration of IndirectHeap in CommandContainer
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
2022-01-26 13:30:26 +01:00
1b7949432f Add shareable allocation on windows dGPUs
Add default initialization for object members

Related-To: LOCI-2665

Signed-off-by: Kamil Diedrich <kamil.diedrich@intel.com>
2022-01-19 19:03:18 +01:00
4238679078 Refactor implicit scaling device support
Related-To: NEO-6589

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2022-01-18 13:08:43 +01:00
394c0e90e1 Return error when failing on submission
Signed-off-by: Raiyan Latif <raiyan.latif@intel.com>
2022-01-12 16:42:30 +01:00
b59b0b6b36 Download timestamps before checking completion
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2021-12-28 08:14:27 +01:00
f4c151cce5 Refactor PipeControlArgs struct
Remove struct PipeControlArgsBase

Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
2021-12-22 17:13:16 +01:00
461e6d841d Remove redundant thread arbitration policy variables for SCM
Remove CommandStreamReceiver::lastSentThreadArbitrationPolicy
Remove CommandStreamReceiver::requiredThreadArbitrationPolicy

Related-To: NEO-5995

Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
2021-12-17 19:47:59 +01:00
1f0c58d0bf Refactor timestamp wait mechanism
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2021-12-16 13:22:19 +01:00
01348451db Add multi tile barrier to marker command on multi tile device
Related-To: NEO-6262

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2021-12-13 18:14:39 +01:00
3511b4755f Remove redundant coherency variables for SCM
Remove CommandStreamReceiver::lastSentCoherencyRequest
Remove CsrSizeRequestFlags::coherencyRequestChanged

Related-To: NEO-5995

Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
2021-12-09 14:43:21 +01:00
3e1023fa1a Unify memory layout for all multi tile post sync operations
Related-To: NEO-6262

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2021-12-02 18:00:40 +01:00
36efe72e3e Add wait on timestamp mechanism
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2021-12-01 08:48:00 +01:00
f56773d166 Setup partition registers once per context
Related-To: NEO-6262

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2021-11-22 22:41:09 +01:00
c4e802f01b WSL - fixing resource cleanup on process shutdown
Resolves issues with coexistance of NEO L0 and NEO OCL libraries
within a single process running in WSL and using WDDM GPU PV

Signed-off-by: Jaroslaw Chodor <jaroslaw.chodor@intel.com>
2021-11-20 21:51:51 +01:00
ec1d092b0b Add isAnyDirectSubmissionEnabled method
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2021-11-16 17:15:56 +01:00
60677e2f9a Update latest flushed task count in CSR with AUB dump
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2021-11-12 13:01:23 +01:00
61641bb70a Add implicit barriers capability to opencl
Related-To: NEO-6262

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2021-11-10 08:19:50 +01:00
4ea452040d Defer direct submission control until first submit
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2021-10-29 14:27:25 +02:00
c7d5a96dfd Flush small task adjustments
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2021-10-28 11:39:55 +02:00
8c6fed2149 Revert "Flush small task adjustments"
This reverts commit 32ae9555f1.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
2021-10-28 09:19:32 +02:00
5fdcb14d57 WSL fallback to staging buffer on blocking reads
Signed-off-by: Kamil Diedrich <kamil.diedrich@intel.com>
2021-10-28 03:34:58 +02:00
32ae9555f1 Flush small task adjustments
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
2021-10-27 16:43:06 +02:00
c097f405f1 Stop using virtual function in tests
Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2021-10-20 13:15:10 +02:00
24ff191e61 Verify completion on all tiles
Related-To: NEO-6244

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2021-10-15 15:43:28 +02:00
a51b385e80 Wait on all tiles when clearing allocation list
Related-To: NEO-6244

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
2021-10-12 11:55:43 +02:00
0859f99d64 Enable zeDeviceCanAccessPeer
- Propagate return value up from Batch Buffer exec
- Add zeDeviceCanAccessPeer functionality

Signed-off-by: Raiyan Latif <raiyan.latif@intel.com>
2021-10-12 09:38:22 +02:00
59a1adc41b Poll csr completion at CmdQueue
Related-To: NEO-6090

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
2021-10-07 14:25:28 +02:00
f1c24ddcbb WDDM - guard against duplicated host ptr allocs
Signed-off-by: Jaroslaw Chodor <jaroslaw.chodor@intel.com>
2021-10-06 22:33:40 +02:00