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
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
|
// =================================================================================================
// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This
// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max-
// width of 100 characters per line.
//
// Author(s):
// Cedric Nugteren <www.cedricnugteren.nl>
//
// This file contains the Xgemv kernel (fast versions) for matrix-vector multiplication.
//
// =================================================================================================
// Enables loading of this file using the C++ pre-processor's #include (C++11 standard raw string
// literal). Comment-out this line for syntax-highlighting when developing.
R"(
// =================================================================================================
// Parameters set by the tuner or by the database. Here they are given a basic default value in case
// this kernel file is used outside of the CLBlast library.
// 1: For the full version, see 'xgemv.opencl'
// 2: For the fast version
#ifndef WGS2
#define WGS2 64 // The local work-group size
#endif
#ifndef WPT2
#define WPT2 1 // The amount of work-per-thread
#endif
#ifndef VW2
#define VW2 1 // Vector width of matrix A loads
#endif
// 3: For the fast rotated version
#ifndef WGS3
#define WGS3 64 // The local work-group size
#endif
#ifndef WPT3
#define WPT3 1 // The tile-size
#endif
#ifndef VW3
#define VW3 1 // Vector width of matrix A loads
#endif
// =================================================================================================
// Data-widths for the 'fast' kernel
#if VW2 == 1
typedef real realVF;
#elif VW2 == 2
typedef real2 realVF;
#elif VW2 == 4
typedef real4 realVF;
#elif VW2 == 8
typedef real8 realVF;
#elif VW2 == 16
typedef real16 realVF;
#endif
// Data-widths for the 'fast' kernel with rotated matrix
#if VW3 == 1
typedef real realVFR;
#elif VW3 == 2
typedef real2 realVFR;
#elif VW3 == 4
typedef real4 realVFR;
#elif VW3 == 8
typedef real8 realVFR;
#elif VW3 == 16
typedef real16 realVFR;
#endif
// =================================================================================================
// Loads a vector input value
INLINE_FUNC realVF LoadMatrixAVF(const __global realVF* restrict agm, const int x, const int y,
const int a_ld) {
return agm[a_ld*y + x];
}
// =================================================================================================
// Faster version of the kernel, assuming that:
// --> 'm' and 'n' are multiples of WGS2
// --> 'a_offset' is 0
// --> 'a_ld' is a multiple of VW2
// --> 'a_rotated' is 0
// --> 'do_conjugate' is 0
#if RELAX_WORKGROUP_SIZE == 1
__kernel
#else
__kernel __attribute__((reqd_work_group_size(WGS2, 1, 1)))
#endif
void XgemvFast(const int m, const int n,
const real_arg arg_alpha,
const real_arg arg_beta,
const int a_rotated,
const __global realVF* restrict agm, const int a_offset, const int a_ld,
const __global real* restrict xgm, const int x_offset, const int x_inc,
__global real* ygm, const int y_offset, const int y_inc,
const int do_conjugate, const int parameter,
const int kl_unused, const int ku_unused) {
const real alpha = GetRealArg(arg_alpha);
const real beta = GetRealArg(arg_beta);
// Local memory for the vector X
__local real xlm[WGS2];
// Initializes the accumulation registers
#pragma promote_to_registers
real acc2[WPT2];
#pragma unroll
for (int _w = 0; _w < WPT2; _w += 1) {
SetToZero(acc2[_w]);
}
// Loops over work-group sized portions of the work
for (int kwg=0; kwg<n; kwg+=WGS2) {
// Loads the vector X into local memory
const int lid = get_local_id(0);
xlm[lid] = xgm[(kwg + lid)*x_inc + x_offset];
// Synchronizes all threads in a workgroup
barrier(CLK_LOCAL_MEM_FENCE);
// The multiply-add function (not rotated)
#pragma unroll
for (int _kl = 0; _kl < WGS2; _kl += 1) {
const int k = kwg + _kl;
#pragma unroll
for (int _w = 0; _w < WPT2/VW2; _w += 1) {
const int gid = (WPT2/VW2)*get_global_id(0) + _w;
realVF avec = agm[(a_ld/VW2)*k + gid];
#if VW2 == 1
MultiplyAdd(acc2[VW2*_w+0], xlm[_kl], avec);
#elif VW2 == 2
MultiplyAdd(acc2[VW2*_w+0], xlm[_kl], avec.x);
MultiplyAdd(acc2[VW2*_w+1], xlm[_kl], avec.y);
#elif VW2 == 4
MultiplyAdd(acc2[VW2*_w+0], xlm[_kl], avec.x);
MultiplyAdd(acc2[VW2*_w+1], xlm[_kl], avec.y);
MultiplyAdd(acc2[VW2*_w+2], xlm[_kl], avec.z);
MultiplyAdd(acc2[VW2*_w+3], xlm[_kl], avec.w);
#elif VW2 == 8
MultiplyAdd(acc2[VW2*_w+0], xlm[_kl], avec.s0);
MultiplyAdd(acc2[VW2*_w+1], xlm[_kl], avec.s1);
MultiplyAdd(acc2[VW2*_w+2], xlm[_kl], avec.s2);
MultiplyAdd(acc2[VW2*_w+3], xlm[_kl], avec.s3);
MultiplyAdd(acc2[VW2*_w+4], xlm[_kl], avec.s4);
MultiplyAdd(acc2[VW2*_w+5], xlm[_kl], avec.s5);
MultiplyAdd(acc2[VW2*_w+6], xlm[_kl], avec.s6);
MultiplyAdd(acc2[VW2*_w+7], xlm[_kl], avec.s7);
#elif VW2 == 16
MultiplyAdd(acc2[VW2*_w+0], xlm[_kl], avec.s0);
MultiplyAdd(acc2[VW2*_w+1], xlm[_kl], avec.s1);
MultiplyAdd(acc2[VW2*_w+2], xlm[_kl], avec.s2);
MultiplyAdd(acc2[VW2*_w+3], xlm[_kl], avec.s3);
MultiplyAdd(acc2[VW2*_w+4], xlm[_kl], avec.s4);
MultiplyAdd(acc2[VW2*_w+5], xlm[_kl], avec.s5);
MultiplyAdd(acc2[VW2*_w+6], xlm[_kl], avec.s6);
MultiplyAdd(acc2[VW2*_w+7], xlm[_kl], avec.s7);
MultiplyAdd(acc2[VW2*_w+8], xlm[_kl], avec.s8);
MultiplyAdd(acc2[VW2*_w+9], xlm[_kl], avec.s9);
MultiplyAdd(acc2[VW2*_w+10], xlm[_kl], avec.sA);
MultiplyAdd(acc2[VW2*_w+11], xlm[_kl], avec.sB);
MultiplyAdd(acc2[VW2*_w+12], xlm[_kl], avec.sC);
MultiplyAdd(acc2[VW2*_w+13], xlm[_kl], avec.sD);
MultiplyAdd(acc2[VW2*_w+14], xlm[_kl], avec.sE);
MultiplyAdd(acc2[VW2*_w+15], xlm[_kl], avec.sF);
#endif
}
}
// Synchronizes all threads in a workgroup
barrier(CLK_LOCAL_MEM_FENCE);
}
// Stores the final result
#pragma unroll
for (int _w = 0; _w < WPT2; _w += 1) {
const int gid = WPT2*get_global_id(0) + _w;
real yval = ygm[gid*y_inc + y_offset];
AXPBY(ygm[gid*y_inc + y_offset], alpha, acc2[_w], beta, yval);
}
}
// =================================================================================================
// Faster version of the kernel, assuming that:
// --> 'm' and 'n' are multiples of WGS3
// --> 'a_offset' is 0
// --> 'a_ld' is a multiple of VW3
// --> 'a_rotated' is 1
// --> 'do_conjugate' is 0
#if RELAX_WORKGROUP_SIZE == 1
__kernel
#else
__kernel __attribute__((reqd_work_group_size(WGS3, 1, 1)))
#endif
void XgemvFastRot(const int m, const int n,
const real_arg arg_alpha,
const real_arg arg_beta,
const int a_rotated,
const __global realVFR* restrict agm, const int a_offset, const int a_ld,
const __global real* restrict xgm, const int x_offset, const int x_inc,
__global real* ygm, const int y_offset, const int y_inc,
const int do_conjugate, const int parameter,
const int kl_unused, const int ku_unused) {
const real alpha = GetRealArg(arg_alpha);
const real beta = GetRealArg(arg_beta);
// Local memory to store a tile of the matrix (for coalescing)
__local real tile[WPT3][WGS3];
const int lid = get_local_id(0);
const int lid_mod = lid % (WPT3/VW3);
const int lid_div = lid / (WPT3/VW3);
// Local memory for the vector X
__local real xlm[WPT3];
// Initializes the accumulation register
real acc3;
SetToZero(acc3);
// Loops over tile-sized portions of the work
for (int kwg=0; kwg<n; kwg+=WPT3) {
// Loads the vector X into local memory
if (lid < WPT3) {
xlm[lid] = xgm[(kwg + lid) * x_inc + x_offset];
}
// Loads the matrix A into local memory
#pragma unroll
for (int _kl = 0; _kl < WPT3/VW3; _kl += 1) {
const int x = (kwg/VW3) + lid_mod;
const int y = get_group_id(0) * WGS3 + lid_div * (WPT3/VW3) + _kl;
realVFR avec = agm[(a_ld/VW3) * y + x];
#if VW3 == 1
tile[_kl*VW3 + 0][lid] = avec;
#elif VW3 == 2
tile[_kl*VW3 + 0][lid] = avec.x;
tile[_kl*VW3 + 1][lid] = avec.y;
#elif VW3 == 4
tile[_kl*VW3 + 0][lid] = avec.x;
tile[_kl*VW3 + 1][lid] = avec.y;
tile[_kl*VW3 + 2][lid] = avec.z;
tile[_kl*VW3 + 3][lid] = avec.w;
#elif VW3 == 8
tile[_kl*VW3 + 0][lid] = avec.s0;
tile[_kl*VW3 + 1][lid] = avec.s1;
tile[_kl*VW3 + 2][lid] = avec.s2;
tile[_kl*VW3 + 3][lid] = avec.s3;
tile[_kl*VW3 + 4][lid] = avec.s4;
tile[_kl*VW3 + 5][lid] = avec.s5;
tile[_kl*VW3 + 6][lid] = avec.s6;
tile[_kl*VW3 + 7][lid] = avec.s7;
#elif VW3 == 16
tile[_kl*VW3 + 0][lid] = avec.s0;
tile[_kl*VW3 + 1][lid] = avec.s1;
tile[_kl*VW3 + 2][lid] = avec.s2;
tile[_kl*VW3 + 3][lid] = avec.s3;
tile[_kl*VW3 + 4][lid] = avec.s4;
tile[_kl*VW3 + 5][lid] = avec.s5;
tile[_kl*VW3 + 6][lid] = avec.s6;
tile[_kl*VW3 + 7][lid] = avec.s7;
tile[_kl*VW3 + 8][lid] = avec.s8;
tile[_kl*VW3 + 9][lid] = avec.s9;
tile[_kl*VW3 + 10][lid] = avec.sA;
tile[_kl*VW3 + 11][lid] = avec.sB;
tile[_kl*VW3 + 12][lid] = avec.sC;
tile[_kl*VW3 + 13][lid] = avec.sD;
tile[_kl*VW3 + 14][lid] = avec.sE;
tile[_kl*VW3 + 15][lid] = avec.sF;
#endif
}
// Synchronizes all threads in a workgroup
barrier(CLK_LOCAL_MEM_FENCE);
// The multiply-add function (rotated)
#pragma unroll
for (int _kl = 0; _kl < WPT3/VW3; _kl += 1) {
#pragma unroll
for (int _v = 0; _v < VW3; _v += 1) {
real aval = tile[lid_mod*VW3 + _v][lid_div * (WPT3/VW3) + _kl];
real xval = xlm[_kl*VW3 + _v];
MultiplyAdd(acc3, xval, aval);
}
}
// Synchronizes all threads in a workgroup
barrier(CLK_LOCAL_MEM_FENCE);
}
// Stores the final result
const int gid = get_global_id(0);
real yval = ygm[gid * y_inc + y_offset];
AXPBY(ygm[gid * y_inc + y_offset], alpha, acc3, beta, yval);
}
// =================================================================================================
// End of the C++11 raw string literal
)"
// =================================================================================================
|
