Skip to content

Commit 2fb11f8

Browse files
author
Rajalakshmi Srinivasaraghavan
committed
POWER10: Improve copy performance
This patch aligns the stores to 32 byte boundary for scopy and dcopy before entering into vector pair loop. For ccopy, changed the store instructions to stxv to improve performance of unaligned cases.
1 parent 87315e8 commit 2fb11f8

File tree

5 files changed

+152
-23
lines changed

5 files changed

+152
-23
lines changed

‎kernel/power/ccopy_microk_power10.c

Lines changed: 115 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,115 @@
1+
/***************************************************************************
2+
Copyright (c) 2020, 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+
#define HAVE_KERNEL 1
29+
30+
static void copy_kernel (BLASLONG n, FLOAT *x, FLOAT *y)
31+
{
32+
__asm__
33+
(
34+
"lxvp 32, 0(%2) \n\t"
35+
"lxvp 34, 32(%2) \n\t"
36+
"lxvp 36, 64(%2) \n\t"
37+
"lxvp 38, 96(%2) \n\t"
38+
"lxvp 40, 128(%2) \n\t"
39+
"lxvp 42, 160(%2) \n\t"
40+
"lxvp 44, 192(%2) \n\t"
41+
"lxvp 46, 224(%2) \n\t"
42+
43+
"addi %2, %2, 256 \n\t"
44+
"addic. %1, %1, -32 \n\t"
45+
"ble two%= \n\t"
46+
47+
".align 5 \n"
48+
"one%=: \n\t"
49+
50+
"stxv 33, 0(%3) \n\t"
51+
"stxv 32, 16(%3) \n\t"
52+
"stxv 35, 32(%3) \n\t"
53+
"stxv 34, 48(%3) \n\t"
54+
"stxv 37, 64(%3) \n\t"
55+
"stxv 36, 80(%3) \n\t"
56+
"stxv 39, 96(%3) \n\t"
57+
"stxv 38, 112(%3) \n\t"
58+
"lxvp 32, 0(%2) \n\t"
59+
"lxvp 34, 32(%2) \n\t"
60+
"lxvp 36, 64(%2) \n\t"
61+
"lxvp 38, 96(%2) \n\t"
62+
63+
"stxv 41, 128(%3) \n\t"
64+
"stxv 40, 144(%3) \n\t"
65+
"stxv 43, 160(%3) \n\t"
66+
"stxv 42, 176(%3) \n\t"
67+
"stxv 45, 192(%3) \n\t"
68+
"stxv 44, 208(%3) \n\t"
69+
"stxv 47, 224(%3) \n\t"
70+
"stxv 46, 240(%3) \n\t"
71+
"lxvp 40, 128(%2) \n\t"
72+
"lxvp 42, 160(%2) \n\t"
73+
"lxvp 44, 192(%2) \n\t"
74+
"lxvp 46, 224(%2) \n\t"
75+
76+
77+
"addi %3, %3, 256 \n\t"
78+
"addi %2, %2, 256 \n\t"
79+
80+
"addic. %1, %1, -32 \n\t"
81+
"bgt one%= \n"
82+
83+
"two%=: \n\t"
84+
85+
"stxv 33, 0(%3) \n\t"
86+
"stxv 32, 16(%3) \n\t"
87+
"stxv 35, 32(%3) \n\t"
88+
"stxv 34, 48(%3) \n\t"
89+
"stxv 37, 64(%3) \n\t"
90+
"stxv 36, 80(%3) \n\t"
91+
"stxv 39, 96(%3) \n\t"
92+
"stxv 38, 112(%3) \n\t"
93+
"stxv 41, 128(%3) \n\t"
94+
"stxv 40, 144(%3) \n\t"
95+
"stxv 43, 160(%3) \n\t"
96+
"stxv 42, 176(%3) \n\t"
97+
"stxv 45, 192(%3) \n\t"
98+
"stxv 44, 208(%3) \n\t"
99+
"stxv 47, 224(%3) \n\t"
100+
"stxv 46, 240(%3) \n\t"
101+
102+
"#n=%1 x=%4=%2 y=%0=%3"
103+
:
104+
"=m" (*y),
105+
"+r" (n), // 1
106+
"+b" (x), // 2
107+
"+b" (y) // 3
108+
:
109+
"m" (*x)
110+
:
111+
"cr0",
112+
"vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39",
113+
"vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47"
114+
);
115+
}

‎kernel/power/ccopy_power10.c

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -28,7 +28,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
2828
#include "common.h"
2929

3030
#if defined(__VEC__) || defined(__ALTIVEC__)
31-
#include "copy_microk_power10.c"
31+
#include "ccopy_microk_power10.c"
3232
#endif
3333

3434
#ifndef HAVE_KERNEL
@@ -86,7 +86,7 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
8686
if ( (inc_x == 1) && (inc_y == 1 ))
8787
{
8888

89-
BLASLONG n1 = n & -64;
89+
BLASLONG n1 = n & -32;
9090
if ( n1 > 0 )
9191
{
9292
copy_kernel(n1, x, y);

‎kernel/power/copy_microk_power10.c

Lines changed: 13 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -62,38 +62,39 @@ static void copy_kernel (BLASLONG n, FLOAT *x, FLOAT *y)
6262
"one%=: \n\t"
6363

6464
"stxvp 32, 0(%3) \n\t"
65-
"lxvp 32, 0(%2) \n\t"
6665
"stxvp 34, 32(%3) \n\t"
67-
"lxvp 34, 32(%2) \n\t"
6866
"stxvp 36, 64(%3) \n\t"
69-
"lxvp 36, 64(%2) \n\t"
7067
"stxvp 38, 96(%3) \n\t"
68+
"lxvp 32, 0(%2) \n\t"
69+
"lxvp 34, 32(%2) \n\t"
70+
"lxvp 36, 64(%2) \n\t"
7171
"lxvp 38, 96(%2) \n\t"
7272

7373
"stxvp 40, 128(%3) \n\t"
74-
"lxvp 40, 128(%2) \n\t"
7574
"stxvp 42, 160(%3) \n\t"
76-
"lxvp 42, 160(%2) \n\t"
7775
"stxvp 44, 192(%3) \n\t"
78-
"lxvp 44, 192(%2) \n\t"
7976
"stxvp 46, 224(%3) \n\t"
77+
"lxvp 40, 128(%2) \n\t"
78+
"lxvp 42, 160(%2) \n\t"
79+
"lxvp 44, 192(%2) \n\t"
8080
"lxvp 46, 224(%2) \n\t"
8181

8282
"stxvp 48, 256(%3) \n\t"
83-
"lxvp 48, 256(%2) \n\t"
8483
"stxvp 50, 288(%3) \n\t"
85-
"lxvp 50, 288(%2) \n\t"
8684
"stxvp 52, 320(%3) \n\t"
87-
"lxvp 52, 320(%2) \n\t"
8885
"stxvp 54, 352(%3) \n\t"
86+
"lxvp 48, 256(%2) \n\t"
87+
"lxvp 50, 288(%2) \n\t"
88+
"lxvp 52, 320(%2) \n\t"
8989
"lxvp 54, 352(%2) \n\t"
90+
9091
"stxvp 56, 384(%3) \n\t"
91-
"lxvp 56, 384(%2) \n\t"
9292
"stxvp 58, 416(%3) \n\t"
93-
"lxvp 58, 416(%2) \n\t"
9493
"stxvp 60, 448(%3) \n\t"
95-
"lxvp 60, 448(%2) \n\t"
9694
"stxvp 62, 480(%3) \n\t"
95+
"lxvp 56, 384(%2) \n\t"
96+
"lxvp 58, 416(%2) \n\t"
97+
"lxvp 60, 448(%2) \n\t"
9798
"lxvp 62, 480(%2) \n\t"
9899

99100
"addi %3, %3, 512 \n\t"

‎kernel/power/dcopy_power10.c

Lines changed: 11 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -85,12 +85,18 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
8585

8686
if ( (inc_x == 1) && (inc_y == 1 ))
8787
{
88-
89-
BLASLONG n1 = n & -64;
90-
if ( n1 > 0 )
88+
if ( n >= 64 )
89+
{
90+
BLASLONG align = ((32 - ((uintptr_t)y & (uintptr_t)0x1F)) >> 3) & 0x3;
91+
for (i = 0; i < align; i++) {
92+
y[i] = x[i] ;
93+
}
94+
}
95+
BLASLONG n1 = (n-i) & -64;
96+
if ( n1 )
9197
{
92-
copy_kernel(n1, x, y);
93-
i=n1;
98+
copy_kernel(n1, &x[i], &y[i]);
99+
i += n1;
94100
}
95101

96102
while(i < n)

‎kernel/power/scopy_power10.c

Lines changed: 11 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -86,11 +86,18 @@ int CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLASLONG inc_y)
8686
if ( (inc_x == 1) && (inc_y == 1 ))
8787
{
8888

89-
BLASLONG n1 = n & -128;
90-
if ( n1 > 0 )
89+
if ( n >= 128 )
9190
{
92-
copy_kernel (n1, x, y);
93-
i=n1;
91+
BLASLONG align = ((32 - ((uintptr_t)y & (uintptr_t)0x1F)) >> 2) & 0x7;
92+
for (i = 0; i < align; i++) {
93+
y[i] = x[i] ;
94+
}
95+
}
96+
BLASLONG n1 = (n-i) & -128;
97+
if ( n1 )
98+
{
99+
copy_kernel(n1, &x[i], &y[i]);
100+
i += n1;
94101
}
95102

96103
while(i < n)

0 commit comments

Comments
 (0)