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
|
/*
* Copyright (C) 2018-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
#pragma once
#include "shared/source/helpers/simd_helper.h"
#include <array>
#include <cstddef>
#include <cstdint>
namespace NEO {
struct RootDeviceEnvironment;
inline uint32_t getNumGrfsPerLocalIdCoordinate(uint32_t simd, uint32_t grfSize) {
return (simd == 32 && grfSize == 32) ? 2 : 1;
}
inline uint32_t getThreadsPerWG(uint32_t simd, uint32_t lws) {
if (isSimd1(simd)) {
return lws;
}
auto result = lws + simd - 1;
// Original logic:
// result = (lws + simd - 1) / simd;
// This sequence is meant to avoid an CPU DIV instruction.
result >>= simd == 32
? 5
: simd == 16
? 4
: 3; // for SIMD 8
return result;
}
inline uint32_t getPerThreadSizeLocalIDs(uint32_t simd, uint32_t grfSize, uint32_t numChannels = 3) {
if (isSimd1(simd)) {
return grfSize;
}
auto numGRFSPerLocalIdCoord = getNumGrfsPerLocalIdCoordinate(simd, grfSize);
uint32_t returnSize = numGRFSPerLocalIdCoord * grfSize * numChannels;
return returnSize;
}
struct LocalIDHelper {
static void (*generateSimd8)(void *buffer, const std::array<uint16_t, 3> &localWorkgroupSize, uint16_t threadsPerWorkGroup, const std::array<uint8_t, 3> &dimensionsOrder, bool chooseMaxRowSize);
static void (*generateSimd16)(void *buffer, const std::array<uint16_t, 3> &localWorkgroupSize, uint16_t threadsPerWorkGroup, const std::array<uint8_t, 3> &dimensionsOrder, bool chooseMaxRowSize);
static void (*generateSimd32)(void *buffer, const std::array<uint16_t, 3> &localWorkgroupSize, uint16_t threadsPerWorkGroup, const std::array<uint8_t, 3> &dimensionsOrder, bool chooseMaxRowSize);
static LocalIDHelper initializer;
private:
LocalIDHelper();
};
extern const uint16_t initialLocalID[];
template <typename Vec, int simd>
void generateLocalIDsSimd(void *b, const std::array<uint16_t, 3> &localWorkgroupSize, uint16_t threadsPerWorkGroup,
const std::array<uint8_t, 3> &dimensionsOrder, bool chooseMaxRowSize);
void generateLocalIDs(void *buffer, uint16_t simd, const std::array<uint16_t, 3> &localWorkgroupSize,
const std::array<uint8_t, 3> &dimensionsOrder, bool isImageOnlyKernel, uint32_t grfSize, uint32_t grfCount, const RootDeviceEnvironment &rootDeviceEnvironment);
void generateLocalIDsWithLayoutForImages(void *b, const std::array<uint16_t, 3> &localWorkgroupSize, uint16_t simd);
bool isCompatibleWithLayoutForImages(const std::array<uint16_t, 3> &localWorkgroupSize, const std::array<uint8_t, 3> &dimensionsOrder, uint16_t simd);
void generateLocalIDsForSimdOne(void *b, const std::array<uint16_t, 3> &localWorkgroupSize,
const std::array<uint8_t, 3> &dimensionsOrder, uint32_t grfSize);
} // namespace NEO
|
