52 lines
2.0 KiB
C++
52 lines
2.0 KiB
C++
/*
|
|
* Copyright (C) 2017-2019 Intel Corporation
|
|
*
|
|
* SPDX-License-Identifier: MIT
|
|
*
|
|
*/
|
|
|
|
#include "runtime/helpers/per_thread_data.h"
|
|
|
|
#include "core/command_stream/linear_stream.h"
|
|
#include "core/helpers/debug_helpers.h"
|
|
|
|
#include <array>
|
|
|
|
namespace NEO {
|
|
|
|
size_t PerThreadDataHelper::sendPerThreadData(
|
|
LinearStream &indirectHeap,
|
|
uint32_t simd,
|
|
uint32_t numChannels,
|
|
const size_t localWorkSizes[3],
|
|
const std::array<uint8_t, 3> &workgroupWalkOrder,
|
|
bool hasKernelOnlyImages) {
|
|
auto offsetPerThreadData = indirectHeap.getUsed();
|
|
if (numChannels) {
|
|
auto localWorkSize = localWorkSizes[0] * localWorkSizes[1] * localWorkSizes[2];
|
|
auto sizePerThreadDataTotal = getPerThreadDataSizeTotal(simd, numChannels, localWorkSize);
|
|
auto pDest = indirectHeap.getSpace(sizePerThreadDataTotal);
|
|
|
|
// Generate local IDs
|
|
DEBUG_BREAK_IF(numChannels != 3);
|
|
generateLocalIDs(pDest, static_cast<uint16_t>(simd),
|
|
std::array<uint16_t, 3>{{static_cast<uint16_t>(localWorkSizes[0]),
|
|
static_cast<uint16_t>(localWorkSizes[1]),
|
|
static_cast<uint16_t>(localWorkSizes[2])}},
|
|
std::array<uint8_t, 3>{{workgroupWalkOrder[0], workgroupWalkOrder[1], workgroupWalkOrder[2]}},
|
|
hasKernelOnlyImages);
|
|
}
|
|
return offsetPerThreadData;
|
|
}
|
|
|
|
uint32_t PerThreadDataHelper::getThreadPayloadSize(const iOpenCL::SPatchThreadPayload &threadPayload, uint32_t simd) {
|
|
uint32_t multiplier = static_cast<uint32_t>(getGRFsPerThread(simd));
|
|
uint32_t threadPayloadSize = 0;
|
|
threadPayloadSize = getNumLocalIdChannels(threadPayload) * multiplier * sizeof(GRF);
|
|
threadPayloadSize += (threadPayload.HeaderPresent) ? sizeof(GRF) : 0;
|
|
threadPayloadSize += (threadPayload.LocalIDFlattenedPresent) ? (sizeof(GRF) * multiplier) : 0;
|
|
threadPayloadSize += (threadPayload.UnusedPerThreadConstantPresent) ? (sizeof(GRF)) : 0;
|
|
return threadPayloadSize;
|
|
}
|
|
} // namespace NEO
|