Skip to content

[UR][L0v2] add support for batched queue submissions #19769

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 3 commits into
base: sycl
Choose a base branch
from
Open

[UR][L0v2] add support for batched queue submissions #19769

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
1e10ac6
add support for batched queues in L0v2 adapter
EuphoricThinking Jul 7, 2025
c79c8ea
for tests in CI
EuphoricThinking Oct 17, 2025
89178cc
add batch queues workarounds and skips
EuphoricThinking Oct 17, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion unified-runtime/scripts/templates/queue_api.hpp.mako
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,8 +25,9 @@ from templates import helper as th
#pragma once

#include <ur_api.h>
#include "queue_extensions.hpp"

struct ur_queue_t_ {
struct ur_queue_t_ : ur_queue_extensions {
virtual ~ur_queue_t_();

%for obj in th.get_queue_related_functions(specs, n, tags):
Expand Down
2 changes: 2 additions & 0 deletions unified-runtime/source/adapters/level_zero/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -171,6 +171,7 @@ if(UR_BUILD_ADAPTER_L0_V2)
${CMAKE_CURRENT_SOURCE_DIR}/v2/memory.hpp
${CMAKE_CURRENT_SOURCE_DIR}/v2/lockable.hpp
${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_api.hpp
${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_batched.hpp
${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_immediate_in_order.hpp
${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_immediate_out_of_order.hpp
${CMAKE_CURRENT_SOURCE_DIR}/v2/usm.hpp
Expand All @@ -187,6 +188,7 @@ if(UR_BUILD_ADAPTER_L0_V2)
${CMAKE_CURRENT_SOURCE_DIR}/v2/kernel.cpp
${CMAKE_CURRENT_SOURCE_DIR}/v2/memory.cpp
${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_api.cpp
${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_batched.cpp
${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_create.cpp
${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_immediate_in_order.cpp
${CMAKE_CURRENT_SOURCE_DIR}/v2/queue_immediate_out_of_order.cpp
Expand Down
85 changes: 62 additions & 23 deletions unified-runtime/source/adapters/level_zero/v2/command_buffer.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@
#include "../command_buffer_command.hpp"
#include "../helpers/kernel_helpers.hpp"
#include "../ur_interface_loader.hpp"
#include "command_list_manager.hpp"
#include "logger/ur_logger.hpp"
#include "queue_handle.hpp"

Expand Down Expand Up @@ -323,9 +324,12 @@ ur_result_t urCommandBufferAppendKernelLaunchExp(
auto eventsWaitList = commandBuffer->getWaitListFromSyncPoints(
syncPointWaitList, numSyncPointsInWaitList);

wait_list_view waitListView =
wait_list_view(eventsWaitList, numSyncPointsInWaitList);

UR_CALL(commandListLocked->appendKernelLaunch(
hKernel, workDim, pGlobalWorkOffset, pGlobalWorkSize, pLocalWorkSize, 0,
nullptr, numSyncPointsInWaitList, eventsWaitList,
nullptr, waitListView,
commandBuffer->createEventIfRequested(retSyncPoint)));

return UR_RESULT_SUCCESS;
Expand All @@ -348,8 +352,11 @@ ur_result_t urCommandBufferAppendUSMMemcpyExp(
auto eventsWaitList = hCommandBuffer->getWaitListFromSyncPoints(
pSyncPointWaitList, numSyncPointsInWaitList);

wait_list_view waitListView =
wait_list_view(eventsWaitList, numSyncPointsInWaitList);

UR_CALL(commandListLocked->appendUSMMemcpy(
false, pDst, pSrc, size, numSyncPointsInWaitList, eventsWaitList,
false, pDst, pSrc, size, waitListView,
hCommandBuffer->createEventIfRequested(pSyncPoint)));

return UR_RESULT_SUCCESS;
Expand All @@ -375,9 +382,12 @@ ur_result_t urCommandBufferAppendMemBufferCopyExp(
auto eventsWaitList = hCommandBuffer->getWaitListFromSyncPoints(
pSyncPointWaitList, numSyncPointsInWaitList);

wait_list_view waitListView =
wait_list_view(eventsWaitList, numSyncPointsInWaitList);

UR_CALL(commandListLocked->appendMemBufferCopy(
hSrcMem, hDstMem, srcOffset, dstOffset, size, numSyncPointsInWaitList,
eventsWaitList, hCommandBuffer->createEventIfRequested(pSyncPoint)));
hSrcMem, hDstMem, srcOffset, dstOffset, size, waitListView,
hCommandBuffer->createEventIfRequested(pSyncPoint)));

return UR_RESULT_SUCCESS;
} catch (...) {
Expand All @@ -402,9 +412,12 @@ ur_result_t urCommandBufferAppendMemBufferWriteExp(
auto eventsWaitList = hCommandBuffer->getWaitListFromSyncPoints(
pSyncPointWaitList, numSyncPointsInWaitList);

wait_list_view waitListView =
wait_list_view(eventsWaitList, numSyncPointsInWaitList);

UR_CALL(commandListLocked->appendMemBufferWrite(
hBuffer, false, offset, size, pSrc, numSyncPointsInWaitList,
eventsWaitList, hCommandBuffer->createEventIfRequested(pSyncPoint)));
hBuffer, false, offset, size, pSrc, waitListView,
hCommandBuffer->createEventIfRequested(pSyncPoint)));

return UR_RESULT_SUCCESS;
} catch (...) {
Expand All @@ -427,9 +440,12 @@ ur_result_t urCommandBufferAppendMemBufferReadExp(
auto eventsWaitList = hCommandBuffer->getWaitListFromSyncPoints(
pSyncPointWaitList, numSyncPointsInWaitList);

wait_list_view waitListView =
wait_list_view(eventsWaitList, numSyncPointsInWaitList);

UR_CALL(commandListLocked->appendMemBufferRead(
hBuffer, false, offset, size, pDst, numSyncPointsInWaitList,
eventsWaitList, hCommandBuffer->createEventIfRequested(pSyncPoint)));
hBuffer, false, offset, size, pDst, waitListView,
hCommandBuffer->createEventIfRequested(pSyncPoint)));

return UR_RESULT_SUCCESS;
} catch (...) {
Expand All @@ -456,10 +472,13 @@ ur_result_t urCommandBufferAppendMemBufferCopyRectExp(
auto eventsWaitList = hCommandBuffer->getWaitListFromSyncPoints(
pSyncPointWaitList, numSyncPointsInWaitList);

wait_list_view waitListView =
wait_list_view(eventsWaitList, numSyncPointsInWaitList);

UR_CALL(commandListLocked->appendMemBufferCopyRect(
hSrcMem, hDstMem, srcOrigin, dstOrigin, region, srcRowPitch,
srcSlicePitch, dstRowPitch, dstSlicePitch, numSyncPointsInWaitList,
eventsWaitList, hCommandBuffer->createEventIfRequested(pSyncPoint)));
srcSlicePitch, dstRowPitch, dstSlicePitch, waitListView,
hCommandBuffer->createEventIfRequested(pSyncPoint)));

return UR_RESULT_SUCCESS;
} catch (...) {
Expand All @@ -486,10 +505,12 @@ ur_result_t urCommandBufferAppendMemBufferWriteRectExp(
auto eventsWaitList = hCommandBuffer->getWaitListFromSyncPoints(
pSyncPointWaitList, numSyncPointsInWaitList);

wait_list_view waitListView =
wait_list_view(eventsWaitList, numSyncPointsInWaitList);

UR_CALL(commandListLocked->appendMemBufferWriteRect(
hBuffer, false, bufferOffset, hostOffset, region, bufferRowPitch,
bufferSlicePitch, hostRowPitch, hostSlicePitch, pSrc,
numSyncPointsInWaitList, eventsWaitList,
bufferSlicePitch, hostRowPitch, hostSlicePitch, pSrc, waitListView,
hCommandBuffer->createEventIfRequested(pSyncPoint)));

return UR_RESULT_SUCCESS;
Expand Down Expand Up @@ -517,10 +538,12 @@ ur_result_t urCommandBufferAppendMemBufferReadRectExp(
auto eventsWaitList = hCommandBuffer->getWaitListFromSyncPoints(
pSyncPointWaitList, numSyncPointsInWaitList);

wait_list_view waitListView =
wait_list_view(eventsWaitList, numSyncPointsInWaitList);

UR_CALL(commandListLocked->appendMemBufferReadRect(
hBuffer, false, bufferOffset, hostOffset, region, bufferRowPitch,
bufferSlicePitch, hostRowPitch, hostSlicePitch, pDst,
numSyncPointsInWaitList, eventsWaitList,
bufferSlicePitch, hostRowPitch, hostSlicePitch, pDst, waitListView,
hCommandBuffer->createEventIfRequested(pSyncPoint)));

return UR_RESULT_SUCCESS;
Expand All @@ -543,9 +566,12 @@ ur_result_t urCommandBufferAppendUSMFillExp(
auto eventsWaitList = hCommandBuffer->getWaitListFromSyncPoints(
pSyncPointWaitList, numSyncPointsInWaitList);

wait_list_view waitListView =
wait_list_view(eventsWaitList, numSyncPointsInWaitList);

UR_CALL(commandListLocked->appendUSMFill(
pMemory, patternSize, pPattern, size, numSyncPointsInWaitList,
eventsWaitList, hCommandBuffer->createEventIfRequested(pSyncPoint)));
pMemory, patternSize, pPattern, size, waitListView,
hCommandBuffer->createEventIfRequested(pSyncPoint)));
return UR_RESULT_SUCCESS;
} catch (...) {
return exceptionToResult(std::current_exception());
Expand All @@ -567,9 +593,12 @@ ur_result_t urCommandBufferAppendMemBufferFillExp(
auto eventsWaitList = hCommandBuffer->getWaitListFromSyncPoints(
pSyncPointWaitList, numSyncPointsInWaitList);

wait_list_view waitListView =
wait_list_view(eventsWaitList, numSyncPointsInWaitList);

UR_CALL(commandListLocked->appendMemBufferFill(
hBuffer, pPattern, patternSize, offset, size, numSyncPointsInWaitList,
eventsWaitList, hCommandBuffer->createEventIfRequested(pSyncPoint)));
hBuffer, pPattern, patternSize, offset, size, waitListView,
hCommandBuffer->createEventIfRequested(pSyncPoint)));

return UR_RESULT_SUCCESS;
} catch (...) {
Expand All @@ -593,8 +622,11 @@ ur_result_t urCommandBufferAppendUSMPrefetchExp(
auto eventsWaitList = hCommandBuffer->getWaitListFromSyncPoints(
pSyncPointWaitList, numSyncPointsInWaitList);

wait_list_view waitListView =
wait_list_view(eventsWaitList, numSyncPointsInWaitList);

UR_CALL(commandListLocked->appendUSMPrefetch(
pMemory, size, flags, numSyncPointsInWaitList, eventsWaitList,
pMemory, size, flags, waitListView,
hCommandBuffer->createEventIfRequested(pSyncPoint)));

return UR_RESULT_SUCCESS;
Expand All @@ -617,8 +649,11 @@ ur_result_t urCommandBufferAppendUSMAdviseExp(
auto eventsWaitList = hCommandBuffer->getWaitListFromSyncPoints(
pSyncPointWaitList, numSyncPointsInWaitList);

wait_list_view waitListView =
wait_list_view(eventsWaitList, numSyncPointsInWaitList);

UR_CALL(commandListLocked->appendUSMAdvise(
pMemory, size, advice, numSyncPointsInWaitList, eventsWaitList,
pMemory, size, advice, waitListView,
hCommandBuffer->createEventIfRequested(pSyncPoint)));

return UR_RESULT_SUCCESS;
Expand Down Expand Up @@ -667,15 +702,19 @@ ur_result_t urCommandBufferAppendNativeCommandExp(
auto eventsWaitList = hCommandBuffer->getWaitListFromSyncPoints(
pSyncPointWaitList, numSyncPointsInWaitList);

UR_CALL(commandListLocked->appendEventsWaitWithBarrier(
numSyncPointsInWaitList, eventsWaitList, nullptr));
wait_list_view waitListView =
wait_list_view(eventsWaitList, numSyncPointsInWaitList);

UR_CALL(
commandListLocked->appendEventsWaitWithBarrier(waitListView, nullptr));

// Call user-defined function immediately
pfnNativeCommand(pData);

wait_list_view emptyWaitList = wait_list_view(nullptr, 0);
// Barrier on all commands after user defined commands.
UR_CALL(commandListLocked->appendEventsWaitWithBarrier(
0, nullptr, hCommandBuffer->createEventIfRequested(pSyncPoint)));
emptyWaitList, hCommandBuffer->createEventIfRequested(pSyncPoint)));

return UR_RESULT_SUCCESS;
}
Expand Down
Loading
Loading