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
|
/*
BLIS
An object-based framework for developing high-performance BLAS-like
libraries.
Copyright (C) 2021-2022, Advanced Micro Devices, Inc. All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
- Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
- Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
- Neither the name(s) of the copyright holder(s) nor the names of its
contributors may be used to endorse or promote products derived
from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include "immintrin.h"
#include "blis.h"
void bli_caxpyf_zen_int_4
(
conj_t conja,
conj_t conjx,
dim_t m,
dim_t b_n,
scomplex* restrict alpha,
scomplex* restrict a, inc_t inca, inc_t lda,
scomplex* restrict x, inc_t incx,
scomplex* restrict y, inc_t incy,
cntx_t* restrict cntx
)
{
inc_t fuse_fac = 4;
inc_t i;
__m256 ymm0, ymm1, ymm2, ymm3;
__m256 ymm4, ymm5, ymm6, ymm7;
__m256 ymm8, ymm10;
__m256 ymm12, ymm13;
float* ap[4];
float* y0 = (float*)y;
scomplex chi0;
scomplex chi1;
scomplex chi2;
scomplex chi3;
dim_t setPlusOne = 1;
if ( bli_is_conj(conja) )
{
setPlusOne = -1;
}
// If either dimension is zero, or if alpha is zero, return early.
if ( bli_zero_dim2( m, b_n ) || bli_ceq0( *alpha ) ) return;
// If b_n is not equal to the fusing factor, then perform the entire
// operation as a loop over axpyv.
if ( b_n != fuse_fac )
{
if ( cntx == NULL ) cntx = bli_gks_query_cntx();
caxpyv_ker_ft f = bli_cntx_get_l1v_ker_dt( BLIS_SCOMPLEX, BLIS_AXPYV_KER, cntx );
for ( i = 0; i < b_n; ++i )
{
scomplex* a1 = a + (0 )*inca + (i )*lda;
scomplex* chi1 = x + (i )*incx;
scomplex* y1 = y + (0 )*incy;
scomplex alpha_chi1;
bli_ccopycjs( conjx, *chi1, alpha_chi1 );
bli_cscals( *alpha, alpha_chi1 );
f
(
conja,
m,
&alpha_chi1,
a1, inca,
y1, incy,
cntx
);
}
return;
}
// At this point, we know that b_n is exactly equal to the fusing factor.
if(bli_is_noconj(conjx))
{
chi0 = *( x + 0*incx );
chi1 = *( x + 1*incx );
chi2 = *( x + 2*incx );
chi3 = *( x + 3*incx );
}
else
{
scomplex *pchi0 = x + 0*incx ;
scomplex *pchi1 = x + 1*incx ;
scomplex *pchi2 = x + 2*incx ;
scomplex *pchi3 = x + 3*incx ;
bli_ccopycjs( conjx, *pchi0, chi0 );
bli_ccopycjs( conjx, *pchi1, chi1 );
bli_ccopycjs( conjx, *pchi2, chi2 );
bli_ccopycjs( conjx, *pchi3, chi3 );
}
// Scale each chi scalar by alpha.
bli_cscals( *alpha, chi0 );
bli_cscals( *alpha, chi1 );
bli_cscals( *alpha, chi2 );
bli_cscals( *alpha, chi3 );
lda *= 2;
incx *= 2;
incy *= 2;
inca *= 2;
ap[0] = (float*)a;
ap[1] = (float*)a + lda;
ap[2] = ap[1] + lda;
ap[3] = ap[2] + lda;
if( inca == 2 && incy == 2 )
{
inc_t n1 = m/4;
inc_t n2 = m%4;
ymm12 = _mm256_setzero_ps();
ymm13 = _mm256_setzero_ps();
// broadcast real & imag parts of 4 elements of x
ymm0 = _mm256_broadcast_ss(&chi0.real); // real part of x0
ymm1 = _mm256_broadcast_ss(&chi0.imag); // imag part of x0
ymm2 = _mm256_broadcast_ss(&chi1.real); // real part of x1
ymm3 = _mm256_broadcast_ss(&chi1.imag); // imag part of x1
ymm4 = _mm256_broadcast_ss(&chi2.real); // real part of x2
ymm5 = _mm256_broadcast_ss(&chi2.imag); // imag part of x2
ymm6 = _mm256_broadcast_ss(&chi3.real); // real part of x3
ymm7 = _mm256_broadcast_ss(&chi3.imag); // imag part of x3
for(i = 0; i < n1; i++)
{
//load first two columns of A
ymm8 = _mm256_loadu_ps(ap[0] + 0);
ymm10 = _mm256_loadu_ps(ap[1] + 0);
ymm12 = _mm256_mul_ps(ymm8, ymm0);
ymm13 = _mm256_mul_ps(ymm8, ymm1);
ymm12 = _mm256_fmadd_ps(ymm10, ymm2, ymm12);
ymm13 = _mm256_fmadd_ps(ymm10, ymm3, ymm13);
//load 3rd and 4th columns of A
ymm8 = _mm256_loadu_ps(ap[2] + 0);
ymm10 = _mm256_loadu_ps(ap[3] + 0);
ymm12 = _mm256_fmadd_ps(ymm8, ymm4, ymm12);
ymm13 = _mm256_fmadd_ps(ymm8, ymm5, ymm13);
ymm12 = _mm256_fmadd_ps(ymm10, ymm6, ymm12);
ymm13 = _mm256_fmadd_ps(ymm10, ymm7, ymm13);
//load Y vector
ymm10 = _mm256_loadu_ps(y0 + 0);
if(bli_is_noconj(conja))
{
//printf("Inside no conj if\n");
ymm13 = _mm256_permute_ps(ymm13, 0xB1);
ymm8 = _mm256_addsub_ps(ymm12, ymm13);
}
else
{
ymm12 = _mm256_permute_ps(ymm12, 0xB1);
ymm8 = _mm256_addsub_ps(ymm13, ymm12);
ymm8 = _mm256_permute_ps(ymm8, 0xB1);
}
ymm12 = _mm256_add_ps(ymm8, ymm10);
_mm256_storeu_ps((float*)(y0), ymm12);
y0 += 8;
ap[0] += 8;
ap[1] += 8;
ap[2] += 8;
ap[3] += 8;
}
// If there are leftover iterations, perform them with scalar code.
for ( i = 0; (i + 0) < n2 ; ++i )
{
scomplex y0c = *(scomplex*)y0;
const scomplex a0c = *(scomplex*)ap[0];
const scomplex a1c = *(scomplex*)ap[1];
const scomplex a2c = *(scomplex*)ap[2];
const scomplex a3c = *(scomplex*)ap[3];
y0c.real += chi0.real * a0c.real - chi0.imag * a0c.imag * setPlusOne;
y0c.real += chi1.real * a1c.real - chi1.imag * a1c.imag * setPlusOne;
y0c.real += chi2.real * a2c.real - chi2.imag * a2c.imag * setPlusOne;
y0c.real += chi3.real * a3c.real - chi3.imag * a3c.imag * setPlusOne;
y0c.imag += chi0.imag * a0c.real + chi0.real * a0c.imag * setPlusOne;
y0c.imag += chi1.imag * a1c.real + chi1.real * a1c.imag * setPlusOne;
y0c.imag += chi2.imag * a2c.real + chi2.real * a2c.imag * setPlusOne;
y0c.imag += chi3.imag * a3c.real + chi3.real * a3c.imag * setPlusOne;
*(scomplex*)y0 = y0c;
ap[0] += 2;
ap[1] += 2;
ap[2] += 2;
ap[3] += 2;
y0 += 2;
}
//PASTEMAC(c,fprintm)(stdout, "Y after A*x in axpyf",m, 1, (scomplex*)y, 1, 1, "%4.1f", "");
}
else
{
for (i = 0 ; (i + 0) < m ; ++i )
{
scomplex y0c = *(scomplex*)y0;
const scomplex a0c = *(scomplex*)ap[0];
const scomplex a1c = *(scomplex*)ap[1];
const scomplex a2c = *(scomplex*)ap[2];
const scomplex a3c = *(scomplex*)ap[3];
y0c.real += chi0.real * a0c.real - chi0.imag * a0c.imag * setPlusOne;
y0c.real += chi1.real * a1c.real - chi1.imag * a1c.imag * setPlusOne;
y0c.real += chi2.real * a2c.real - chi2.imag * a2c.imag * setPlusOne;
y0c.real += chi3.real * a3c.real - chi3.imag * a3c.imag * setPlusOne;
y0c.imag += chi0.imag * a0c.real + chi0.real * a0c.imag * setPlusOne;
y0c.imag += chi1.imag * a1c.real + chi1.real * a1c.imag * setPlusOne;
y0c.imag += chi2.imag * a2c.real + chi2.real * a2c.imag * setPlusOne;
y0c.imag += chi3.imag * a3c.real + chi3.real * a3c.imag * setPlusOne;
*(scomplex*)y0 = y0c;
ap[0] += inca;
ap[1] += inca;
ap[2] += inca;
ap[3] += inca;
y0 += incy;
}
}
}
|
