1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
|
/*
* Copyright (C) 2017-2020 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#pragma once
#include "shared/source/built_ins/built_ins.h"
#include "shared/source/command_stream/command_stream_receiver.h"
#include "opencl/source/command_queue/command_queue_hw.h"
#include "opencl/source/helpers/hardware_commands_helper.h"
#include "opencl/source/mem_obj/buffer.h"
#include "opencl/source/memory_manager/mem_obj_surface.h"
#include <new>
namespace NEO {
template <typename GfxFamily>
cl_int CommandQueueHw<GfxFamily>::enqueueWriteBufferRect(
Buffer *buffer,
cl_bool blockingWrite,
const size_t *bufferOrigin,
const size_t *hostOrigin,
const size_t *region,
size_t bufferRowPitch,
size_t bufferSlicePitch,
size_t hostRowPitch,
size_t hostSlicePitch,
const void *ptr,
cl_uint numEventsInWaitList,
const cl_event *eventWaitList,
cl_event *event) {
const cl_command_type cmdType = CL_COMMAND_WRITE_BUFFER_RECT;
auto isMemTransferNeeded = true;
if (buffer->isMemObjZeroCopy()) {
size_t bufferOffset;
size_t hostOffset;
computeOffsetsValueForRectCommands(&bufferOffset, &hostOffset, bufferOrigin, hostOrigin, region, bufferRowPitch, bufferSlicePitch, hostRowPitch, hostSlicePitch);
isMemTransferNeeded = buffer->checkIfMemoryTransferIsRequired(bufferOffset, hostOffset, ptr, cmdType);
}
if (!isMemTransferNeeded) {
return enqueueMarkerForReadWriteOperation(buffer, const_cast<void *>(ptr), cmdType, blockingWrite,
numEventsInWaitList, eventWaitList, event);
}
auto eBuiltInOps = EBuiltInOps::CopyBufferRect;
if (forceStateless(buffer->getSize())) {
eBuiltInOps = EBuiltInOps::CopyBufferRectStateless;
}
size_t hostPtrSize = Buffer::calculateHostPtrSize(hostOrigin, region, hostRowPitch, hostSlicePitch);
void *srcPtr = const_cast<void *>(ptr);
MemObjSurface dstBufferSurf(buffer);
HostPtrSurface hostPtrSurf(srcPtr, hostPtrSize, true);
Surface *surfaces[] = {&dstBufferSurf, &hostPtrSurf};
if (region[0] != 0 &&
region[1] != 0 &&
region[2] != 0) {
auto &csr = getCommandStreamReceiverByCommandType(cmdType);
bool status = csr.createAllocationForHostSurface(hostPtrSurf, false);
if (!status) {
return CL_OUT_OF_RESOURCES;
}
srcPtr = reinterpret_cast<void *>(hostPtrSurf.getAllocation()->getGpuAddress());
}
void *alignedSrcPtr = alignDown(srcPtr, 4);
size_t srcPtrOffset = ptrDiff(srcPtr, alignedSrcPtr);
BuiltinOpParams dc;
dc.srcPtr = alignedSrcPtr;
dc.dstMemObj = buffer;
dc.srcOffset = hostOrigin;
dc.srcOffset.x += srcPtrOffset;
dc.dstOffset = bufferOrigin;
dc.transferAllocation = hostPtrSurf.getAllocation();
dc.size = region;
dc.srcRowPitch = hostRowPitch;
dc.srcSlicePitch = hostSlicePitch;
dc.dstRowPitch = bufferRowPitch;
dc.dstSlicePitch = bufferSlicePitch;
MultiDispatchInfo dispatchInfo(dc);
dispatchBcsOrGpgpuEnqueue<CL_COMMAND_WRITE_BUFFER_RECT>(dispatchInfo, surfaces, eBuiltInOps, numEventsInWaitList, eventWaitList, event, blockingWrite);
if (context->isProvidingPerformanceHints()) {
context->providePerformanceHint(CL_CONTEXT_DIAGNOSTICS_LEVEL_NEUTRAL_INTEL, CL_ENQUEUE_WRITE_BUFFER_RECT_REQUIRES_COPY_DATA, static_cast<cl_mem>(buffer));
}
return CL_SUCCESS;
}
} // namespace NEO
|
