intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-25 05:24:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add support for zero-copy r/w buffer

Browse Source 
Change-Id: Ie9f3f2211d107eb338bd97692d36e9c7d7a0feab
This commit is contained in:
mplewka
2018-01-11 15:42:55 +01:00
parent 79e4b2c104
commit 2c2bbbcdbb
13 changed files with 694 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

163
unit_tests/command_queue/enqueue_read_buffer_event_tests.cpp
View File

@@ -22,6 +22,7 @@
#include "runtime/command_queue/command_queue.h"
#include "runtime/event/event.h"
#include "unit_tests/helpers/debug_manager_state_restore.h"
#include "unit_tests/fixtures/hello_world_fixture.h"
#include "unit_tests/fixtures/buffer_fixture.h"
#include "gtest/gtest.h"
@@ -108,3 +109,165 @@ TEST_F(EnqueueReadBuffer, eventReturnedShouldBeMaxOfInputEventsAndCmdQPlus1) {
delete pEvent;
}
TEST_F(EnqueueReadBuffer, givenInOrderQueueAndEnabledSupportCpuCopiesAndDstPtrEqualSrcPtrWithEventsWhenReadBufferIsExecutedThenTaskLevelShouldNotBeIncreased) {
DebugManagerStateRestore dbgRestore;
DebugManager.flags.DoCpuCopyOnReadBuffer.set(true);
cl_int retVal = CL_SUCCESS;
uint32_t taskLevelCmdQ = 17;
pCmdQ->taskLevel = taskLevelCmdQ;
uint32_t taskLevelEvent1 = 8;
uint32_t taskLevelEvent2 = 19;
Event event1(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, taskLevelEvent1, 4);
Event event2(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, taskLevelEvent2, 10);
cl_bool blockingRead = CL_TRUE;
size_t size = sizeof(cl_float);
cl_event eventWaitList[] =
{
&event1,
&event2};
cl_uint numEventsInWaitList = sizeof(eventWaitList) / sizeof(eventWaitList[0]);
cl_event event = nullptr;
auto srcBuffer = std::unique_ptr<Buffer>(BufferHelper<>::create());
void *ptr = srcBuffer->getCpuAddressForMemoryTransfer();
retVal = pCmdQ->enqueueReadBuffer(srcBuffer.get(),
blockingRead,
0,
size,
ptr,
numEventsInWaitList,
eventWaitList,
&event);
EXPECT_EQ(CL_SUCCESS, retVal);
ASSERT_NE(nullptr, event);
auto pEvent = (Event *)event;
EXPECT_EQ(19u, pEvent->taskLevel);
EXPECT_EQ(17u, pCmdQ->taskLevel);
pEvent->release();
}
TEST_F(EnqueueReadBuffer, givenOutOfOrderQueueAndEnabledSupportCpuCopiesAndDstPtrEqualSrcPtrWithEventsWhenReadBufferIsExecutedThenTaskLevelShouldNotBeIncreased) {
DebugManagerStateRestore dbgRestore;
DebugManager.flags.DoCpuCopyOnReadBuffer.set(true);
std::unique_ptr<CommandQueue> pCmdOOQ(createCommandQueue(pDevice, CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE));
cl_int retVal = CL_SUCCESS;
uint32_t taskLevelCmdQ = 17;
pCmdOOQ->taskLevel = taskLevelCmdQ;
uint32_t taskLevelEvent1 = 8;
uint32_t taskLevelEvent2 = 19;
Event event1(pCmdOOQ.get(), CL_COMMAND_NDRANGE_KERNEL, taskLevelEvent1, 4);
Event event2(pCmdOOQ.get(), CL_COMMAND_NDRANGE_KERNEL, taskLevelEvent2, 10);
cl_bool blockingRead = CL_TRUE;
size_t size = sizeof(cl_float);
cl_event eventWaitList[] =
{
&event1,
&event2};
cl_uint numEventsInWaitList = sizeof(eventWaitList) / sizeof(eventWaitList[0]);
cl_event event = nullptr;
auto srcBuffer = std::unique_ptr<Buffer>(BufferHelper<>::create());
void *ptr = srcBuffer->getCpuAddressForMemoryTransfer();
retVal = pCmdOOQ->enqueueReadBuffer(srcBuffer.get(),
blockingRead,
0,
size,
ptr,
numEventsInWaitList,
eventWaitList,
&event);
EXPECT_EQ(CL_SUCCESS, retVal);
ASSERT_NE(nullptr, event);
auto pEvent = (Event *)event;
EXPECT_EQ(19u, pEvent->taskLevel);
EXPECT_EQ(17u, pCmdOOQ->taskLevel);
pEvent->release();
}
TEST_F(EnqueueReadBuffer, givenInOrderQueueAndDisabledSupportCpuCopiesAndDstPtrEqualSrcPtrWithEventsWhenReadBufferIsExecutedThenTaskLevelShouldNotBeIncreased) {
DebugManagerStateRestore dbgRestore;
DebugManager.flags.DoCpuCopyOnReadBuffer.set(false);
cl_int retVal = CL_SUCCESS;
uint32_t taskLevelCmdQ = 17;
pCmdQ->taskLevel = taskLevelCmdQ;
uint32_t taskLevelEvent1 = 8;
uint32_t taskLevelEvent2 = 19;
Event event1(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, taskLevelEvent1, 4);
Event event2(pCmdQ, CL_COMMAND_NDRANGE_KERNEL, taskLevelEvent2, 10);
cl_bool blockingRead = CL_TRUE;
size_t size = sizeof(cl_float);
cl_event eventWaitList[] =
{
&event1,
&event2};
cl_uint numEventsInWaitList = sizeof(eventWaitList) / sizeof(eventWaitList[0]);
cl_event event = nullptr;
auto srcBuffer = std::unique_ptr<Buffer>(BufferHelper<>::create());
void *ptr = srcBuffer->getCpuAddressForMemoryTransfer();
retVal = pCmdQ->enqueueReadBuffer(srcBuffer.get(),
blockingRead,
0,
size,
ptr,
numEventsInWaitList,
eventWaitList,
&event);
EXPECT_EQ(CL_SUCCESS, retVal);
ASSERT_NE(nullptr, event);
auto pEvent = (Event *)event;
EXPECT_EQ(19u, pEvent->taskLevel);
EXPECT_EQ(19u, pCmdQ->taskLevel);
pEvent->release();
}
TEST_F(EnqueueReadBuffer, givenOutOfOrderQueueAndDisabledSupportCpuCopiesAndDstPtrEqualSrcPtrWithEventsWhenReadBufferIsExecutedThenTaskLevelShouldNotBeIncreased) {
DebugManagerStateRestore dbgRestore;
DebugManager.flags.DoCpuCopyOnReadBuffer.set(false);
std::unique_ptr<CommandQueue> pCmdOOQ(createCommandQueue(pDevice, CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE));
cl_int retVal = CL_SUCCESS;
uint32_t taskLevelCmdQ = 17;
pCmdOOQ->taskLevel = taskLevelCmdQ;
uint32_t taskLevelEvent1 = 8;
uint32_t taskLevelEvent2 = 19;
Event event1(pCmdOOQ.get(), CL_COMMAND_NDRANGE_KERNEL, taskLevelEvent1, 4);
Event event2(pCmdOOQ.get(), CL_COMMAND_NDRANGE_KERNEL, taskLevelEvent2, 10);
cl_bool blockingRead = CL_TRUE;
size_t size = sizeof(cl_float);
cl_event eventWaitList[] =
{
&event1,
&event2};
cl_uint numEventsInWaitList = sizeof(eventWaitList) / sizeof(eventWaitList[0]);
cl_event event = nullptr;
auto srcBuffer = std::unique_ptr<Buffer>(BufferHelper<>::create());
void *ptr = srcBuffer->getCpuAddressForMemoryTransfer();
retVal = pCmdOOQ->enqueueReadBuffer(srcBuffer.get(),
blockingRead,
0,
size,
ptr,
numEventsInWaitList,
eventWaitList,
&event);
EXPECT_EQ(CL_SUCCESS, retVal);
ASSERT_NE(nullptr, event);
auto pEvent = (Event *)event;
EXPECT_EQ(19u, pEvent->taskLevel);
EXPECT_EQ(19u, pCmdOOQ->taskLevel);
pEvent->release();
}