413 lines
12 KiB
C
413 lines
12 KiB
C
/***************************************************************************
|
|
Copyright (c) 2017, The OpenBLAS Project
|
|
All rights reserved.
|
|
Redistribution and use in source and binary forms, with or without
|
|
modification, are permitted provided that the following conditions are
|
|
met:
|
|
1. Redistributions of source code must retain the above copyright
|
|
notice, this list of conditions and the following disclaimer.
|
|
2. 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.
|
|
3. Neither the name of the OpenBLAS project 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 OPENBLAS PROJECT 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 "common.h"
|
|
#include <float.h>
|
|
#include <arm_neon.h>
|
|
|
|
#if defined(SMP)
|
|
extern int blas_level1_thread_with_return_value(int mode, BLASLONG m, BLASLONG n,
|
|
BLASLONG k, void *alpha, void *a, BLASLONG lda, void *b, BLASLONG ldb,
|
|
void *c, BLASLONG ldc, int (*function)(), int nthreads);
|
|
#endif
|
|
|
|
#define N "x0" /* vector length */
|
|
#define X "x1" /* X vector address */
|
|
#define INC_X "x2" /* X stride */
|
|
#define J "x3" /* loop variable */
|
|
#define K "x4" /* loop variable */
|
|
|
|
#if !defined(COMPLEX)
|
|
#define INC_SHIFT "3"
|
|
#define SZ "8"
|
|
#else
|
|
#define INC_SHIFT "4"
|
|
#define SZ "16"
|
|
#endif
|
|
|
|
#define SSQ "d0"
|
|
#define SCALE "d1"
|
|
#define REGZERO "d5"
|
|
#define REGONE "d6"
|
|
#define CUR_MAX "d7"
|
|
#define CUR_MAXINV "d8"
|
|
#define CUR_MAXINV_V "v8.2d"
|
|
#define CUR_MAX_V "v8.2d"
|
|
#define REGINF "d9"
|
|
|
|
static void nrm2_compute(BLASLONG n, FLOAT *x, BLASLONG inc_x,
|
|
double *ssq, double *scale)
|
|
{
|
|
*ssq = 0.0;
|
|
*scale = 0.0;
|
|
|
|
if (n <= 0) return;
|
|
|
|
__asm__ __volatile__ (
|
|
" mov "N", %[N_] \n"
|
|
" mov "X", %[X_] \n"
|
|
" mov "INC_X", %[INCX_] \n"
|
|
" fmov "SCALE", xzr \n"
|
|
" fmov "SSQ", #1.0 \n"
|
|
" cmp "N", xzr \n"
|
|
" ble 9f //nrm2_kernel_L999 \n"
|
|
" cmp "INC_X", xzr \n"
|
|
" beq 9f //nrm2_kernel_L999 \n"
|
|
|
|
"1: //nrm2_kernel_F_BEGIN: \n"
|
|
" mov x6, #0x7FF0000000000000 //+Infinity \n"
|
|
" fmov "REGZERO", xzr \n"
|
|
" fmov "REGONE", #1.0 \n"
|
|
" fmov "REGINF", x6 \n"
|
|
" lsl "INC_X", "INC_X", #"INC_SHIFT" \n"
|
|
" mov "J", "N" \n"
|
|
" cmp "J", xzr \n"
|
|
" beq 9f //nrm2_kernel_L999 \n"
|
|
|
|
"2: //nrm2_kernel_F_ZERO_SKIP: \n"
|
|
" ldr d4, ["X"] \n"
|
|
" fcmp d4, "REGZERO" \n"
|
|
" bne 3f //nrm2_kernel_F_INIT \n"
|
|
#if defined(COMPLEX)
|
|
" ldr d4, ["X", #8] \n"
|
|
" fcmp d4, "REGZERO" \n"
|
|
" bne 4f //nrm2_kernel_F_INIT_I \n"
|
|
#endif
|
|
" add "X", "X", "INC_X" \n"
|
|
" subs "J", "J", #1 \n"
|
|
" beq 9f //nrm2_kernel_L999 \n"
|
|
" b 2b //nrm2_kernel_F_ZERO_SKIP \n"
|
|
|
|
"3: //nrm2_kernel_F_INIT: \n"
|
|
" ldr d4, ["X"] \n"
|
|
" fabs d4, d4 \n"
|
|
" fmax "CUR_MAX", "SCALE", d4 \n"
|
|
" fcmp "CUR_MAX", "REGINF" \n"
|
|
" beq 10f \n"
|
|
" fdiv "SCALE", "SCALE", "CUR_MAX" \n"
|
|
" fmul "SCALE", "SCALE", "SCALE" \n"
|
|
" fmul "SSQ", "SSQ", "SCALE" \n"
|
|
" fdiv d4, d4, "CUR_MAX" \n"
|
|
" fmul d4, d4, d4 \n"
|
|
" fadd "SSQ", "SSQ", d4 \n"
|
|
" fmov "SCALE", "CUR_MAX" \n"
|
|
#if defined(COMPLEX)
|
|
"4: //nrm2_kernel_F_INIT_I: \n"
|
|
" ldr d3, ["X", #8] \n"
|
|
" fabs d3, d3 \n"
|
|
" fmax "CUR_MAX", "SCALE", d3 \n"
|
|
" fcmp "CUR_MAX", "REGINF" \n"
|
|
" beq 10f \n"
|
|
" fdiv "SCALE", "SCALE", "CUR_MAX" \n"
|
|
" fmul "SCALE", "SCALE", "SCALE" \n"
|
|
" fmul "SSQ", "SSQ", "SCALE" \n"
|
|
" fdiv d3, d3, "CUR_MAX" \n"
|
|
" fmul d3, d3, d3 \n"
|
|
" fadd "SSQ", "SSQ", d3 \n"
|
|
" fmov "SCALE", "CUR_MAX" \n"
|
|
#endif
|
|
" add "X", "X", "INC_X" \n"
|
|
" subs "J", "J", #1 \n"
|
|
" beq 9f //nrm2_kernel_L999 \n"
|
|
|
|
"5: //nrm2_kernel_F_START: \n"
|
|
" cmp "INC_X", #"SZ" \n"
|
|
" bne 8f //nrm2_kernel_F1 \n"
|
|
" asr "K", "J", #4 \n"
|
|
" cmp "K", xzr \n"
|
|
" beq 8f //nrm2_kernel_F1 \n"
|
|
|
|
"6: //nrm2_kernel_F: \n"
|
|
" ldp q16, q17, ["X"] \n"
|
|
" ldp q18, q19, ["X", #32] \n"
|
|
" ldp q20, q21, ["X", #64] \n"
|
|
" ldp q22, q23, ["X", #96] \n"
|
|
" add "X", "X", #128 \n"
|
|
" fabs v16.2d, v16.2d \n"
|
|
" fabs v17.2d, v17.2d \n"
|
|
" fabs v18.2d, v18.2d \n"
|
|
" fabs v19.2d, v19.2d \n"
|
|
" fabs v20.2d, v20.2d \n"
|
|
" fabs v21.2d, v21.2d \n"
|
|
" fabs v22.2d, v22.2d \n"
|
|
" fabs v23.2d, v23.2d \n"
|
|
" fmaxp v24.2d, v16.2d, v17.2d \n"
|
|
" fmaxp v25.2d, v18.2d, v19.2d \n"
|
|
" fmaxp v26.2d, v20.2d, v21.2d \n"
|
|
" fmaxp v27.2d, v22.2d, v23.2d \n"
|
|
" fmaxp v24.2d, v24.2d, v25.2d \n"
|
|
" fmaxp v26.2d, v26.2d, v27.2d \n"
|
|
" fmaxp v24.2d, v24.2d, v26.2d \n"
|
|
" fmaxp v24.2d, v24.2d, v24.2d \n"
|
|
" fmax "CUR_MAX", "SCALE", d24 \n"
|
|
" fcmp "CUR_MAX", "REGINF" \n"
|
|
" beq 10f \n"
|
|
" fdiv "CUR_MAXINV", "REGONE", "CUR_MAX" \n"
|
|
" //dup "CUR_MAX_V", v7.d[0] \n"
|
|
" fdiv "SCALE", "SCALE", "CUR_MAX" \n"
|
|
" fmul "SCALE", "SCALE", "SCALE" \n"
|
|
" fmul "SSQ", "SSQ", "SCALE" \n"
|
|
" dup "CUR_MAXINV_V", v8.d[0] \n"
|
|
" fmul v16.2d, v16.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v17.2d, v17.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v18.2d, v18.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v19.2d, v19.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v20.2d, v20.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v21.2d, v21.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v22.2d, v22.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v23.2d, v23.2d, "CUR_MAXINV_V" \n"
|
|
" //fdiv v16.2d, v16.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v17.2d, v17.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v18.2d, v18.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v19.2d, v19.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v20.2d, v20.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v21.2d, v21.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v22.2d, v22.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v23.2d, v23.2d, "CUR_MAX_V" \n"
|
|
" fmul v24.2d, v16.2d, v16.2d \n"
|
|
" fmul v25.2d, v17.2d, v17.2d \n"
|
|
" fmul v26.2d, v18.2d, v18.2d \n"
|
|
" fmul v27.2d, v19.2d, v19.2d \n"
|
|
" fmla v24.2d, v20.2d, v20.2d \n"
|
|
" fmla v25.2d, v21.2d, v21.2d \n"
|
|
" fmla v26.2d, v22.2d, v22.2d \n"
|
|
" fmla v27.2d, v23.2d, v23.2d \n"
|
|
" fadd v24.2d, v24.2d, v25.2d \n"
|
|
" fadd v26.2d, v26.2d, v27.2d \n"
|
|
" fadd v24.2d, v24.2d, v26.2d \n"
|
|
" faddp d24, v24.2d \n"
|
|
" fadd "SSQ", "SSQ", d24 \n"
|
|
" fmov "SCALE", "CUR_MAX" \n"
|
|
#if defined(COMPLEX)
|
|
" ldp q16, q17, ["X"] \n"
|
|
" ldp q18, q19, ["X", #32] \n"
|
|
" ldp q20, q21, ["X", #64] \n"
|
|
" ldp q22, q23, ["X", #96] \n"
|
|
" add "X", "X", #128 \n"
|
|
" fabs v16.2d, v16.2d \n"
|
|
" fabs v17.2d, v17.2d \n"
|
|
" fabs v18.2d, v18.2d \n"
|
|
" fabs v19.2d, v19.2d \n"
|
|
" fabs v20.2d, v20.2d \n"
|
|
" fabs v21.2d, v21.2d \n"
|
|
" fabs v22.2d, v22.2d \n"
|
|
" fabs v23.2d, v23.2d \n"
|
|
" fmaxp v24.2d, v16.2d, v17.2d \n"
|
|
" fmaxp v25.2d, v18.2d, v19.2d \n"
|
|
" fmaxp v26.2d, v20.2d, v21.2d \n"
|
|
" fmaxp v27.2d, v22.2d, v23.2d \n"
|
|
" fmaxp v24.2d, v24.2d, v25.2d \n"
|
|
" fmaxp v26.2d, v26.2d, v27.2d \n"
|
|
" fmaxp v24.2d, v24.2d, v26.2d \n"
|
|
" fmaxp v24.2d, v24.2d, v24.2d \n"
|
|
" fmax "CUR_MAX", "SCALE", d24 \n"
|
|
" fcmp "CUR_MAX", "REGINF" \n"
|
|
" beq 10f \n"
|
|
" fdiv "CUR_MAXINV", "REGONE", "CUR_MAX" \n"
|
|
" //dup "CUR_MAX_V", v7.d[0] \n"
|
|
" fdiv "SCALE", "SCALE", "CUR_MAX" \n"
|
|
" fmul "SCALE", "SCALE", "SCALE" \n"
|
|
" fmul "SSQ", "SSQ", "SCALE" \n"
|
|
" dup "CUR_MAXINV_V", v8.d[0] \n"
|
|
" fmul v16.2d, v16.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v17.2d, v17.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v18.2d, v18.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v19.2d, v19.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v20.2d, v20.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v21.2d, v21.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v22.2d, v22.2d, "CUR_MAXINV_V" \n"
|
|
" fmul v23.2d, v23.2d, "CUR_MAXINV_V" \n"
|
|
" //fdiv v16.2d, v16.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v17.2d, v17.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v18.2d, v18.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v19.2d, v19.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v20.2d, v20.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v21.2d, v21.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v22.2d, v22.2d, "CUR_MAX_V" \n"
|
|
" //fdiv v23.2d, v23.2d, "CUR_MAX_V" \n"
|
|
" fmul v24.2d, v16.2d, v16.2d \n"
|
|
" fmul v25.2d, v17.2d, v17.2d \n"
|
|
" fmul v26.2d, v18.2d, v18.2d \n"
|
|
" fmul v27.2d, v19.2d, v19.2d \n"
|
|
" fmla v24.2d, v20.2d, v20.2d \n"
|
|
" fmla v25.2d, v21.2d, v21.2d \n"
|
|
" fmla v26.2d, v22.2d, v22.2d \n"
|
|
" fmla v27.2d, v23.2d, v23.2d \n"
|
|
" fadd v24.2d, v24.2d, v25.2d \n"
|
|
" fadd v26.2d, v26.2d, v27.2d \n"
|
|
" fadd v24.2d, v24.2d, v26.2d \n"
|
|
" faddp d24, v24.2d \n"
|
|
" fadd "SSQ", "SSQ", d24 \n"
|
|
" fmov "SCALE", "CUR_MAX" \n"
|
|
#endif
|
|
" subs "K", "K", #1 \n"
|
|
" bne 6b //nrm2_kernel_F \n"
|
|
|
|
"7: //nrm2_kernel_F_DONE: \n"
|
|
" ands "J", "J", #15 \n"
|
|
" beq 9f //nrm2_kernel_L999 \n"
|
|
|
|
"8: //nrm2_kernel_F1: \n"
|
|
" ldr d4, ["X"] \n"
|
|
" fabs d4, d4 \n"
|
|
" fmax "CUR_MAX", "SCALE", d4 \n"
|
|
" fcmp "CUR_MAX", "REGINF" \n"
|
|
" beq 10f \n"
|
|
" fdiv "SCALE", "SCALE", "CUR_MAX" \n"
|
|
" fmul "SCALE", "SCALE", "SCALE" \n"
|
|
" fmul "SSQ", "SSQ", "SCALE" \n"
|
|
" fdiv d4, d4, "CUR_MAX" \n"
|
|
" fmul d4, d4, d4 \n"
|
|
" fadd "SSQ", "SSQ", d4 \n"
|
|
" fmov "SCALE", "CUR_MAX" \n"
|
|
#if defined(COMPLEX)
|
|
" ldr d3, ["X", #8] \n"
|
|
" fabs d3, d3 \n"
|
|
" fmax "CUR_MAX", "SCALE", d3 \n"
|
|
" fcmp "CUR_MAX", "REGINF" \n"
|
|
" beq 10f \n"
|
|
" fdiv "SCALE", "SCALE", "CUR_MAX" \n"
|
|
" fmul "SCALE", "SCALE", "SCALE" \n"
|
|
" fmul "SSQ", "SSQ", "SCALE" \n"
|
|
" fdiv d3, d3, "CUR_MAX" \n"
|
|
" fmul d3, d3, d3 \n"
|
|
" fadd "SSQ", "SSQ", d3 \n"
|
|
" fmov "SCALE", "CUR_MAX" \n"
|
|
#endif
|
|
" add "X", "X", "INC_X" \n"
|
|
" subs "J", "J", #1 \n"
|
|
" bne 8b //nrm2_kernel_F1 \n"
|
|
|
|
"9: //nrm2_kernel_L999: \n"
|
|
" str "SSQ", [%[SSQ_]] \n"
|
|
" str "SCALE", [%[SCALE_]] \n"
|
|
" b 11f \n"
|
|
"10: \n"
|
|
" str "REGINF", [%[SSQ_]] \n"
|
|
" str "REGINF", [%[SCALE_]] \n"
|
|
"11: \n"
|
|
|
|
:
|
|
: [SSQ_] "r" (ssq), //%0
|
|
[SCALE_] "r" (scale), //%1
|
|
[N_] "r" (n), //%2
|
|
[X_] "r" (x), //%3
|
|
[INCX_] "r" (inc_x) //%4
|
|
: "cc",
|
|
"memory",
|
|
"x0", "x1", "x2", "x3", "x4", "x5", "x6",
|
|
"d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "d8", REGINF
|
|
);
|
|
|
|
}
|
|
|
|
#if defined(SMP)
|
|
static int nrm2_thread_function(BLASLONG n, BLASLONG dummy0,
|
|
BLASLONG dummy1, FLOAT dummy2, FLOAT *x, BLASLONG inc_x, FLOAT *dummy3,
|
|
BLASLONG dummy4, FLOAT *result, BLASLONG dummy5)
|
|
{
|
|
nrm2_compute(n, x, inc_x, result, result + 1);
|
|
|
|
return 0;
|
|
}
|
|
#endif
|
|
|
|
FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x)
|
|
{
|
|
#if defined(SMP)
|
|
int nthreads;
|
|
FLOAT dummy_alpha[2];
|
|
#endif
|
|
FLOAT ssq, scale;
|
|
|
|
if (n <= 0 || inc_x == 0) return 0.0;
|
|
|
|
#if defined(SMP)
|
|
if (n <= 10000)
|
|
nthreads = 1;
|
|
else
|
|
nthreads = num_cpu_avail(1);
|
|
|
|
if (nthreads == 1) {
|
|
nrm2_compute(n, x, inc_x, &ssq, &scale);
|
|
} else {
|
|
int mode, i;
|
|
char result[MAX_CPU_NUMBER * sizeof(double) * 2];
|
|
double *ptr;
|
|
|
|
#if !defined(COMPLEX)
|
|
mode = BLAS_DOUBLE | BLAS_REAL;
|
|
#else
|
|
mode = BLAS_DOUBLE | BLAS_COMPLEX;
|
|
#endif
|
|
|
|
blas_level1_thread_with_return_value(mode, n, 0, 0, &dummy_alpha,
|
|
x, inc_x, NULL, 0, result, 0,
|
|
( void *)nrm2_thread_function, nthreads);
|
|
|
|
scale = 0.0;
|
|
ssq = 1.0;
|
|
ptr = (double *)result;
|
|
for (i = 0; i < nthreads; i++) {
|
|
FLOAT cur_scale, cur_ssq;
|
|
|
|
cur_ssq = *ptr;
|
|
cur_scale = *(ptr + 1);
|
|
|
|
if (cur_ssq == INFINITY) {
|
|
ssq = INFINITY;
|
|
scale = INFINITY;
|
|
break;
|
|
}
|
|
|
|
if (cur_scale != 0) {
|
|
if (cur_scale > scale) {
|
|
scale = (scale / cur_scale);
|
|
ssq = ssq * scale * scale;
|
|
ssq += cur_ssq;
|
|
scale = cur_scale;
|
|
} else {
|
|
cur_scale = (cur_scale / scale);
|
|
cur_ssq = cur_ssq * cur_scale * cur_scale;
|
|
ssq += cur_ssq;
|
|
}
|
|
}
|
|
|
|
ptr = (double *)(((char *)ptr) + sizeof(double) * 2);
|
|
}
|
|
}
|
|
#else
|
|
nrm2_compute(n, x, inc_x, &ssq, &scale);
|
|
#endif
|
|
volatile FLOAT sca = fabs(scale);
|
|
if (sca < DBL_MIN) return 0.;
|
|
ssq = sqrt(ssq) * scale;
|
|
|
|
return ssq;
|
|
}
|