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
|
/* ************************************************************************
* Copyright 2013 Advanced Micro Devices, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
* ************************************************************************/
#include <stdio.h>
#include <matrix_props.h>
#include <matrix_dims.h>
#include "trxm_common_legacy.h"
void
declareLdsBasedTrxmVariables(
struct KgenContext *ctx,
DataType dtype,
const SubproblemDim *dims,
const PGranularity *pgran,
bool useLocalC)
{
char tmp[1024];
size_t pitchAB, pitchC;
const char *inTypeName, *outTypeName;
unsigned int nrRegs;
unsigned int vecLen;
inTypeName = dtypeBuiltinType(dtype);
pitchAB = matrBlockPitch(dims, MATRIX_A, dtype, clblasLeft);
pitchC = matrBlockPitch(dims, MATRIX_C, dtype, clblasLeft);
vecLen = sizeof(cl_float4) / dtypeSize(dtype);
sprintf(tmp, "__local %s tempA[%lu];\n"
"__local %s tempB[%lu];\n"
"uint m0, k0;\n"
"uint currM, currN;\n"
"uint2 coordA, coordB;\n"
"uint x, y;\n",
inTypeName, pitchAB * dims->y, inTypeName,
pitchAB * dims->x);
kgenAddStmt(ctx, tmp);
getResultGPRsInfo(dtype, &dims[1], vecLen, &nrRegs, &outTypeName);
if (useLocalC) {
sprintf(tmp, "__local %s tempC[%lu];\n", inTypeName,
pitchC * dims->y);
}
else {
sprintf(tmp, "%s c[%u];\n", outTypeName, nrRegs);
}
kgenAddStmt(ctx, tmp);
kgenDeclareLocalID(ctx, "lid", pgran);
kgenDeclareGroupID(ctx, "gid", pgran);
kgenAddBlankLine(ctx);
}
void
genPrepareTrxmBlockA(
struct KgenContext *ctx,
const SubproblemDim *dim,
DataType dtype,
const CopyBufFuncs *copyFuncs,
const ZeroFuncs *zeroFuncs,
KernelExtraFlags flags,
const char *nameM)
{
char tmp[1024];
size_t pitch;
const char *coordName[2] = {"currM", "k0"};
const char *sizeName[2] = {"y", "x"};
int tra;
pitch = matrBlockPitch(dim, MATRIX_A, dtype, clblasLeft);
tra = isMatrixAccessColMaj(CLBLAS_TRMM, flags, MATRIX_A);
/*
* If the (sub)problem is integrally divisible,
* skip any checks, and just read with optimal blocks,
* otherwise check for tails and then read with a
* fast function in the case of optimal blocks, and with
* the slow one in the case of tails respectively
*/
if (!(flags & KEXTRA_TAILS_M)) {
sprintf(tmp, "%s((LPtr)tempA, (GPtr)A, %s, %s, lda);\n",
copyFuncs->read[MATRIX_A], coordName[tra], coordName[1 - tra]);
}
else {
sprintf(tmp,
"y = (currM + %lu <= M) ? %lu : M - currM;\n"
"x = (k0 + %lu <= %s) ? %lu : %s - k0;\n"
"if ((y == %lu) && (x == %lu)) {\n"
// fast read
" %s((LPtr)tempA, (GPtr)A, %s, %s, lda);\n"
"}\n"
"else {\n"
" %s((__local float4*)tempA);\n" // zeroing
" barrier(CLK_LOCAL_MEM_FENCE);\n"
// slow read
" %s((LPtr)tempA, (GPtr)A, %s, %s, %s, %s, %lu, lda);\n"
"}\n\n",
dim->y, dim->y, dim->bwidth, nameM, dim->bwidth, nameM, dim->y,
dim->bwidth, copyFuncs->read[MATRIX_A], coordName[tra],
coordName[1 - tra], zeroFuncs->names[MATRIX_A],
copyFuncs->readGeneric[MATRIX_A], coordName[tra],
coordName[1 - tra], sizeName[tra], sizeName[1 - tra],
pitch);
}
kgenAddStmt(ctx, tmp);
}
void
genPrepareTrxmBlockB(
struct KgenContext *ctx,
const SubproblemDim *dim,
DataType dtype,
const CopyBufFuncs *copyFuncs,
const ZeroFuncs *zeroFuncs,
KernelExtraFlags flags)
{
char tmp[1024];
size_t pitch;
const char *coordName[2] = {"currN", "k0"};
const char *sizeName[2] = {"y", "x"};
int trb;
trb = isMatrixAccessColMaj(CLBLAS_TRMM, flags, MATRIX_B);
pitch = matrBlockPitch(dim, MATRIX_B, dtype, clblasLeft);
if (!(flags & (KEXTRA_TAILS_N | KEXTRA_TAILS_K))) {
sprintf(tmp, "%s((LPtr)tempB, (GPtr)B, %s, %s, ldb);\n",
copyFuncs->read[MATRIX_B], coordName[trb],
coordName[1 - trb]);
}
else {
sprintf(tmp,
"y = (currN + %lu <= N) ? %lu : N - currN;\n"
"x = (k0 + %lu <= M) ? %lu : M - k0;\n"
"if ((y == %lu) && (x == %lu)) {\n"
// fast read
" %s((LPtr)tempB, (GPtr)B, %s, %s, ldb);\n"
"}\n"
"else {\n"
" %s((__local float4*)tempB);\n" // zeroing
" barrier(CLK_LOCAL_MEM_FENCE);\n" // barrier if it's needed
// slow read
" %s((LPtr)tempB, (GPtr)B, %s, %s, %s, %s, %lu, ldb);\n"
"}\n\n",
dim->x, dim->x, dim->bwidth, dim->bwidth, dim->x, dim->bwidth,
copyFuncs->read[MATRIX_B], coordName[trb], coordName[1 - trb],
zeroFuncs->names[MATRIX_B],
copyFuncs->readGeneric[MATRIX_B], coordName[trb],
coordName[1 - trb], sizeName[trb], sizeName[1 - trb], pitch);
}
kgenAddStmt(ctx, tmp);
}
void
genTriangMatrBlock(
struct KgenContext *ctx,
const SubproblemDim *dim,
DataType dtype,
KernelExtraFlags kflags)
{
char tmp[1024], tmp1[512];
const char *one;
size_t pitch;
pitch = matrBlockPitch(dim, MATRIX_A, dtype, clblasLeft);
one = strOne(dtype);
strcpy(tmp1, "");
// staring diagonal coordinates
kgenAddStmt(ctx, "y = (k0 < currM) ? 0 : (k0 - currM);\n"
"x = (k0 < currM) ? (currM - k0) : 0;\n\n");
if (isMatrixUpper(kflags)) {
/*
* resulting block is upper diagonal, zeroing everything
* below the diagonal and set "1" on the diagonal for the
* unit diagonal matrix
*/
if (kflags & KEXTRA_UNIT_DIAGONAL) {
sprintf(tmp1, "\n"
" if (x < %lu) {\n"
" tempA[lid * %lu + x] = %s;\n"
" }\n",
dim->bwidth, pitch, one);
}
sprintf(tmp, "if (lid >= y && lid < %lu) {\n"
" uint i;\n"
"\n"
" x = x + lid - y;\n"
" x = (x > %lu) ? %lu : x;\n"
"\n"
" for (i = 0; i < x; i++) {\n"
" tempA[lid * %lu + i] = 0;\n"
" }\n"
"%s"
"}\n",
dim->y, dim->bwidth, dim->bwidth, pitch, tmp1);
}
else {
/*
* resulting block is lower diagonal, zeroing everything
* above the diagonal and set "1" on the diagonal for the
* unit diagonal matrix
*/
if (kflags & KEXTRA_UNIT_DIAGONAL) {
sprintf(tmp1, "\n"
" if (y < %lu) {\n"
" tempA[y * %lu + lid] = %s;\n"
" }\n",
dim->y, pitch, one);
}
sprintf(tmp, "if (lid >= x && lid < %lu) {\n"
" uint i;\n"
"\n"
" y = y + lid - x;\n"
" y = (y > %lu) ? %lu : y;\n"
"\n"
" for (i = 0; i < y; i++) {\n"
" tempA[i * %lu + lid] = 0;\n"
" }\n"
"%s"
"}\n",
dim->bwidth, dim->y, dim->y, pitch, tmp1);
}
kgenAddStmt(ctx, tmp);
kgenAddBarrier(ctx, CLK_LOCAL_MEM_FENCE);
kgenAddBlankLine(ctx);
}
|
