Skip to content

Commit 87bebdb

Browse files
fenrus75fenrus75
committed
Add an AVX512 enabled DGEMV (n) function
written in C intrinsics for best readability. (the same C code works for Haswell as well) For logistical reasons the code falls back to the existing haswell AVX2 implementation if the GCC or LLVM compiler is not new enough
1 parent 1a00ef3 commit 87bebdb

File tree

2 files changed

+129
-1
lines changed

2 files changed

+129
-1
lines changed

‎kernel/x86_64/dgemv_n_4.c

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -31,8 +31,10 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
3131

3232
#if defined(NEHALEM)
3333
#include "dgemv_n_microk_nehalem-4.c"
34-
#elif defined(HASWELL) || defined(ZEN) || defined(STEAMROLLER) || defined(EXCAVATOR) || defined (SKYLAKEX)
34+
#elif defined(HASWELL) || defined(ZEN) || defined(STEAMROLLER) || defined(EXCAVATOR)
3535
#include "dgemv_n_microk_haswell-4.c"
36+
#elif defined (SKYLAKEX)
37+
#include "dgemv_n_microk_skylakex-4.c"
3638
#endif
3739

3840

‎kernel/x86_64/dgemv_n_microk_skylakex-4.c

Lines changed: 126 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,126 @@
1+
/***************************************************************************
2+
Copyright (c) 2014, 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+
/* need a new enough GCC for avx512 support */
29+
#if (( defined(__GNUC__) && __GNUC__ > 6 && defined(__AVX2__)) || (defined(__clang__) && __clang_major__ >= 6))
30+
31+
#define HAVE_KERNEL_4x4 1
32+
33+
#include <immintrin.h>
34+
35+
static void dgemv_kernel_4x4( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha)
36+
{
37+
38+
int i = 0;
39+
40+
__m256d x0, x1, x2, x3;
41+
__m256d __alpha;
42+
43+
x0 = _mm256_broadcastsd_pd(_mm_load_sd(&x[0]));
44+
x1 = _mm256_broadcastsd_pd(_mm_load_sd(&x[1]));
45+
x2 = _mm256_broadcastsd_pd(_mm_load_sd(&x[2]));
46+
x3 = _mm256_broadcastsd_pd(_mm_load_sd(&x[3]));
47+
48+
__alpha = _mm256_broadcastsd_pd(_mm_load_sd(alpha));
49+
50+
#ifdef __AVX512CD__
51+
int n5;
52+
__m512d x05, x15, x25, x35;
53+
__m512d __alpha5;
54+
n5 = n & ~7;
55+
56+
x05 = _mm512_broadcastsd_pd(_mm_load_sd(&x[0]));
57+
x15 = _mm512_broadcastsd_pd(_mm_load_sd(&x[1]));
58+
x25 = _mm512_broadcastsd_pd(_mm_load_sd(&x[2]));
59+
x35 = _mm512_broadcastsd_pd(_mm_load_sd(&x[3]));
60+
61+
__alpha5 = _mm512_broadcastsd_pd(_mm_load_sd(alpha));
62+
63+
for (; i < n5; i+= 8) {
64+
__m512d tempY;
65+
__m512d sum;
66+
67+
sum = _mm512_loadu_pd(&ap[0][i]) * x05 +
68+
_mm512_loadu_pd(&ap[1][i]) * x15 +
69+
_mm512_loadu_pd(&ap[2][i]) * x25 +
70+
_mm512_loadu_pd(&ap[3][i]) * x35;
71+
72+
tempY = _mm512_loadu_pd(&y[i]);
73+
tempY += sum * __alpha5;
74+
_mm512_storeu_pd(&y[i], tempY);
75+
}
76+
#endif
77+
78+
for (; i < n; i+= 4) {
79+
__m256d tempY;
80+
__m256d sum;
81+
82+
sum = _mm256_loadu_pd(&ap[0][i]) * x0 +
83+
_mm256_loadu_pd(&ap[1][i]) * x1 +
84+
_mm256_loadu_pd(&ap[2][i]) * x2 +
85+
_mm256_loadu_pd(&ap[3][i]) * x3;
86+
87+
tempY = _mm256_loadu_pd(&y[i]);
88+
tempY += sum * __alpha;
89+
_mm256_storeu_pd(&y[i], tempY);
90+
}
91+
92+
}
93+
94+
95+
#define HAVE_KERNEL_4x2
96+
97+
static void dgemv_kernel_4x2( BLASLONG n, FLOAT **ap, FLOAT *x, FLOAT *y, FLOAT *alpha)
98+
{
99+
100+
int i = 0;
101+
102+
__m256d x0, x1;
103+
__m256d __alpha;
104+
105+
x0 = _mm256_broadcastsd_pd(_mm_load_sd(&x[0]));
106+
x1 = _mm256_broadcastsd_pd(_mm_load_sd(&x[1]));
107+
108+
__alpha = _mm256_broadcastsd_pd(_mm_load_sd(alpha));
109+
110+
111+
for (i = 0; i < n; i+= 4) {
112+
__m256d tempY;
113+
__m256d sum;
114+
115+
sum = _mm256_loadu_pd(&ap[0][i]) * x0 + _mm256_loadu_pd(&ap[1][i]) * x1;
116+
117+
tempY = _mm256_loadu_pd(&y[i]);
118+
tempY += sum * __alpha;
119+
_mm256_storeu_pd(&y[i], tempY);
120+
}
121+
122+
}
123+
124+
#else
125+
#include "dgemv_n_microk_haswell-4.c"
126+
#endif

0 commit comments

Comments
 (0)