Skip to content

Commit 29416cb

Browse files
authored
[ZARCH] Add Z13 version for max/min functions
1 parent 48b9b94 commit 29416cb

File tree

7 files changed

+1216
-6
lines changed

7 files changed

+1216
-6
lines changed

kernel/zarch/KERNEL.Z13

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -1,18 +1,18 @@
11
SAMAXKERNEL = ../arm/amax.c
2-
DAMAXKERNEL = damax.c
2+
DAMAXKERNEL = damax_z13.c
33
CAMAXKERNEL = ../arm/zamax.c
4-
ZAMAXKERNEL = zamax.c
4+
ZAMAXKERNEL = zamax_z13.c
55

66
SAMINKERNEL = ../arm/amin.c
7-
DAMINKERNEL = damin.c
7+
DAMINKERNEL = damin_z13.c
88
CAMINKERNEL = ../arm/zamin.c
9-
ZAMINKERNEL = zamin.c
9+
ZAMINKERNEL = zamin_z13.c
1010

1111
SMAXKERNEL = ../arm/max.c
12-
DMAXKERNEL = dmax.c
12+
DMAXKERNEL = dmax_z13.c
1313

1414
SMINKERNEL = ../arm/min.c
15-
DMINKERNEL = dmin.c
15+
DMINKERNEL = dmin_z13.c
1616

1717
ISAMAXKERNEL = ../arm/iamax.c
1818
IDAMAXKERNEL = idamax.c

kernel/zarch/damax_z13.c

Lines changed: 204 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,204 @@
1+
/***************************************************************************
2+
Copyright (c) 2013-2017, The OpenBLAS Project
3+
All rights reserved.
4+
Redistribution and use in source and binary forms, with or without
5+
modification, are permitted provided that the following conditions are
6+
met:
7+
1. Redistributions of source code must retain the above copyright
8+
notice, this list of conditions and the following disclaimer.
9+
2. Redistributions in binary form must reproduce the above copyright
10+
notice, this list of conditions and the following disclaimer in
11+
the documentation and/or other materials provided with the
12+
distribution.
13+
3. Neither the name of the OpenBLAS project nor the names of
14+
its contributors may be used to endorse or promote products
15+
derived from this software without specific prior written permission.
16+
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
17+
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18+
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19+
ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE
20+
LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21+
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
22+
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
23+
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
24+
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
25+
USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26+
*****************************************************************************/
27+
28+
#include "common.h"
29+
#include <math.h>
30+
31+
#if defined(DOUBLE)
32+
#define ABS fabs
33+
#else
34+
#define ABS fabsf
35+
#endif
36+
37+
static FLOAT damax_kernel_32(BLASLONG n, FLOAT *x)
38+
{
39+
FLOAT amax;
40+
41+
__asm__ volatile (
42+
"vl %%v0,0(%2) \n\t"
43+
"vflpdb %%v0,%%v0 \n\t"
44+
"srlg %%r0,%1,5 \n\t"
45+
"xgr %%r1,%%r1 \n\t"
46+
"0: \n\t"
47+
"pfd 1, 1024(%%r1,%2) \n\t"
48+
49+
"vl %%v16,0(%%r1,%2) \n\t"
50+
"vl %%v17,16(%%r1,%2) \n\t"
51+
"vl %%v18,32(%%r1,%2) \n\t"
52+
"vl %%v19,48(%%r1,%2) \n\t"
53+
"vl %%v20,64(%%r1,%2) \n\t"
54+
"vl %%v21,80(%%r1,%2) \n\t"
55+
"vl %%v22,96(%%r1,%2) \n\t"
56+
"vl %%v23,112(%%r1,%2) \n\t"
57+
"vflpdb %%v16, %%v16 \n\t"
58+
"vflpdb %%v17, %%v17 \n\t"
59+
"vflpdb %%v18, %%v18 \n\t"
60+
"vflpdb %%v19, %%v19 \n\t"
61+
"vflpdb %%v20, %%v20 \n\t"
62+
"vflpdb %%v21, %%v21 \n\t"
63+
"vflpdb %%v22, %%v22 \n\t"
64+
"vflpdb %%v23, %%v23 \n\t"
65+
66+
"vfchdb %%v24,%%v16,%%v17 \n\t"
67+
"vfchdb %%v25,%%v18,%%v19 \n\t"
68+
"vfchdb %%v26,%%v20,%%v21 \n\t"
69+
"vfchdb %%v27,%%v22,%%v23 \n\t"
70+
"vsel %%v24,%%v16,%%v17,%%v24 \n\t"
71+
"vsel %%v25,%%v18,%%v19,%%v25 \n\t"
72+
"vsel %%v26,%%v20,%%v21,%%v26 \n\t"
73+
"vsel %%v27,%%v22,%%v23,%%v27 \n\t"
74+
75+
"vfchdb %%v28,%%v24,%%v25 \n\t"
76+
"vfchdb %%v29,%%v26,%%v27 \n\t"
77+
"vsel %%v28,%%v24,%%v25,%%v28 \n\t"
78+
"vsel %%v29,%%v26,%%v27,%%v29 \n\t"
79+
80+
"vfchdb %%v30,%%v28,%%v29 \n\t"
81+
"vsel %%v30,%%v28,%%v29,%%v30 \n\t"
82+
83+
"vfchdb %%v31,%%v30,%%v0 \n\t"
84+
"vsel %%v0,%%v30,%%v0,%%v31 \n\t"
85+
86+
"vl %%v16,128(%%r1,%2) \n\t"
87+
"vl %%v17,144(%%r1,%2) \n\t"
88+
"vl %%v18,160(%%r1,%2) \n\t"
89+
"vl %%v19,176(%%r1,%2) \n\t"
90+
"vl %%v20,192(%%r1,%2) \n\t"
91+
"vl %%v21,208(%%r1,%2) \n\t"
92+
"vl %%v22,224(%%r1,%2) \n\t"
93+
"vl %%v23,240(%%r1,%2) \n\t"
94+
"vflpdb %%v16, %%v16 \n\t"
95+
"vflpdb %%v17, %%v17 \n\t"
96+
"vflpdb %%v18, %%v18 \n\t"
97+
"vflpdb %%v19, %%v19 \n\t"
98+
"vflpdb %%v20, %%v20 \n\t"
99+
"vflpdb %%v21, %%v21 \n\t"
100+
"vflpdb %%v22, %%v22 \n\t"
101+
"vflpdb %%v23, %%v23 \n\t"
102+
103+
"vfchdb %%v24,%%v16,%%v17 \n\t"
104+
"vfchdb %%v25,%%v18,%%v19 \n\t"
105+
"vfchdb %%v26,%%v20,%%v21 \n\t"
106+
"vfchdb %%v27,%%v22,%%v23 \n\t"
107+
"vsel %%v24,%%v16,%%v17,%%v24 \n\t"
108+
"vsel %%v25,%%v18,%%v19,%%v25 \n\t"
109+
"vsel %%v26,%%v20,%%v21,%%v26 \n\t"
110+
"vsel %%v27,%%v22,%%v23,%%v27 \n\t"
111+
112+
"vfchdb %%v28,%%v24,%%v25 \n\t"
113+
"vfchdb %%v29,%%v26,%%v27 \n\t"
114+
"vsel %%v28,%%v24,%%v25,%%v28 \n\t"
115+
"vsel %%v29,%%v26,%%v27,%%v29 \n\t"
116+
117+
"vfchdb %%v30,%%v28,%%v29 \n\t"
118+
"vsel %%v30,%%v28,%%v29,%%v30 \n\t"
119+
120+
"vfchdb %%v31,%%v30,%%v0 \n\t"
121+
"vsel %%v0,%%v30,%%v0,%%v31 \n\t"
122+
123+
"agfi %%r1, 256 \n\t"
124+
"brctg %%r0, 0b \n\t"
125+
126+
"vrepg %%v16,%%v0,1 \n\t"
127+
"wfchdb %%v17,%%v0,%%v16 \n\t"
128+
"vsel %%v0,%%v0,%%v16,%%v17 \n\t"
129+
"ldr %0,%%f0 "
130+
:"=f"(amax)
131+
:"r"(n),"ZR"((const FLOAT (*)[n])x)
132+
:"memory","cc","r0","r1","v0","v16","v17","v18","v19","v20","v21","v22","v23","v24","v25","v26","v27","v28","v29","v30","v31"
133+
);
134+
135+
return amax;
136+
}
137+
138+
FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x) {
139+
BLASLONG i = 0;
140+
BLASLONG j = 0;
141+
FLOAT maxf = 0.0;
142+
143+
if (n <= 0 || inc_x <= 0) return (maxf);
144+
145+
if (inc_x == 1) {
146+
147+
BLASLONG n1 = n & -32;
148+
if (n1 > 0) {
149+
150+
maxf = damax_kernel_32(n1, x);
151+
152+
i = n1;
153+
}
154+
else
155+
{
156+
maxf=ABS(x[0]);
157+
i++;
158+
}
159+
160+
while (i < n) {
161+
if (ABS(x[i]) > maxf) {
162+
maxf = ABS(x[i]);
163+
}
164+
i++;
165+
}
166+
return (maxf);
167+
168+
} else {
169+
170+
maxf=ABS(x[0]);
171+
172+
BLASLONG n1 = n & -4;
173+
while (j < n1) {
174+
175+
if (ABS(x[i]) > maxf) {
176+
maxf = ABS(x[i]);
177+
}
178+
if (ABS(x[i + inc_x]) > maxf) {
179+
maxf = ABS(x[i + inc_x]);
180+
}
181+
if (ABS(x[i + 2 * inc_x]) > maxf) {
182+
maxf = ABS(x[i + 2 * inc_x]);
183+
}
184+
if (ABS(x[i + 3 * inc_x]) > maxf) {
185+
maxf = ABS(x[i + 3 * inc_x]);
186+
}
187+
188+
i += inc_x * 4;
189+
190+
j += 4;
191+
192+
}
193+
194+
195+
while (j < n) {
196+
if (ABS(x[i]) > maxf) {
197+
maxf = ABS(x[i]);
198+
}
199+
i += inc_x;
200+
j++;
201+
}
202+
return (maxf);
203+
}
204+
}

0 commit comments

Comments
 (0)