floraachy/OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
develop
OpenBLAS/kernel/x86_64/dtrsm_kernel_RT_bulldozer.c
Martin Kroeker f9bb76d29a Fix inline assembly constraints in Bulldozer TRSM kernels 
rework indices to allow marking i,as and bs as both input and output (marked operand n1 as well for simplicity). For #2009
2019-02-16 20:06:48 +01:00

547 lines
15 KiB
C
Raw Permalink Blame History

/*********************************************************************/
/* Copyright 2009, 2010 The University of Texas at Austin. */
/* 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. */
/* */
/* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */
/* AUSTIN ``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 UNIVERSITY OF TEXAS AT */
/* AUSTIN 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. */
/* */
/* The views and conclusions contained in the software and */
/* documentation are those of the authors and should not be */
/* interpreted as representing official policies, either expressed */
/* or implied, of The University of Texas at Austin. */
/*********************************************************************/
#include "common.h"
static FLOAT dm1 = -1.;
#ifdef CONJ
#define GEMM_KERNEL GEMM_KERNEL_R
#else
#define GEMM_KERNEL GEMM_KERNEL_N
#endif
#if GEMM_DEFAULT_UNROLL_M == 1
#define GEMM_UNROLL_M_SHIFT 0
#endif
#if GEMM_DEFAULT_UNROLL_M == 2
#define GEMM_UNROLL_M_SHIFT 1
#endif
#if GEMM_DEFAULT_UNROLL_M == 4
#define GEMM_UNROLL_M_SHIFT 2
#endif
#if GEMM_DEFAULT_UNROLL_M == 6
#define GEMM_UNROLL_M_SHIFT 2
#endif
#if GEMM_DEFAULT_UNROLL_M == 8
#define GEMM_UNROLL_M_SHIFT 3
#endif
#if GEMM_DEFAULT_UNROLL_M == 16
#define GEMM_UNROLL_M_SHIFT 4
#endif
#if GEMM_DEFAULT_UNROLL_N == 1
#define GEMM_UNROLL_N_SHIFT 0
#endif
#if GEMM_DEFAULT_UNROLL_N == 2
#define GEMM_UNROLL_N_SHIFT 1
#endif
#if GEMM_DEFAULT_UNROLL_N == 4
#define GEMM_UNROLL_N_SHIFT 2
#endif
#if GEMM_DEFAULT_UNROLL_N == 8
#define GEMM_UNROLL_N_SHIFT 3
#endif
#if GEMM_DEFAULT_UNROLL_N == 16
#define GEMM_UNROLL_N_SHIFT 4
#endif
static void dtrsm_RT_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, FLOAT *as, FLOAT *bs) __attribute__ ((noinline));
static void dtrsm_RT_solve_opt(BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, FLOAT *as, FLOAT *bs)
{
FLOAT *c1 = c + ldc ;
BLASLONG n1 = n * 8;
BLASLONG i=0;
as += (2 - 1) * 8;
bs += (2 - 1) * 2;
__asm__ __volatile__
(
" vzeroupper \n\t"
" prefetcht0 (%4) \n\t"
" prefetcht0 (%5) \n\t"
" vxorpd %%xmm8 , %%xmm8 , %%xmm8 \n\t"
" vxorpd %%xmm9 , %%xmm9 , %%xmm9 \n\t"
" vxorpd %%xmm10, %%xmm10, %%xmm10 \n\t"
" vxorpd %%xmm11, %%xmm11, %%xmm11 \n\t"
" vxorpd %%xmm12, %%xmm12, %%xmm12 \n\t"
" vxorpd %%xmm13, %%xmm13, %%xmm13 \n\t"
" vxorpd %%xmm14, %%xmm14, %%xmm14 \n\t"
" vxorpd %%xmm15, %%xmm15, %%xmm15 \n\t"
" cmpq $0, %0 \n\t"
" je 2f \n\t"
" .align 16 \n\t"
"1: \n\t"
" prefetcht0 384(%6,%1,8) \n\t"
" prefetcht0 384(%7,%1,8) \n\t"
" vmovddup (%7,%1,2), %%xmm0 \n\t" // read b
" vmovups (%6,%1,8), %%xmm4 \n\t"
" vmovddup 8(%7,%1,2), %%xmm1 \n\t"
" vmovups 16(%6,%1,8), %%xmm5 \n\t"
" vmovups 32(%6,%1,8), %%xmm6 \n\t"
" vmovups 48(%6,%1,8), %%xmm7 \n\t"
" vfmaddpd %%xmm8 , %%xmm0 , %%xmm4 , %%xmm8 \n\t"
" vfmaddpd %%xmm12, %%xmm1 , %%xmm4 , %%xmm12 \n\t"
" vfmaddpd %%xmm9 , %%xmm0 , %%xmm5 , %%xmm9 \n\t"
" vfmaddpd %%xmm13, %%xmm1 , %%xmm5 , %%xmm13 \n\t"
" vfmaddpd %%xmm10, %%xmm0 , %%xmm6 , %%xmm10 \n\t"
" vfmaddpd %%xmm14, %%xmm1 , %%xmm6 , %%xmm14 \n\t"
" addq $8, %1 \n\t"
" vfmaddpd %%xmm11, %%xmm0 , %%xmm7 , %%xmm11 \n\t"
" vfmaddpd %%xmm15, %%xmm1 , %%xmm7 , %%xmm15 \n\t"
" cmpq %1, %0 \n\t"
" jz 2f \n\t"
" prefetcht0 384(%6,%1,8) \n\t"
" vmovddup (%7,%1,2), %%xmm0 \n\t" // read b
" vmovups (%6,%1,8), %%xmm4 \n\t"
" vmovddup 8(%7,%1,2), %%xmm1 \n\t"
" vmovups 16(%6,%1,8), %%xmm5 \n\t"
" vmovups 32(%6,%1,8), %%xmm6 \n\t"
" vmovups 48(%6,%1,8), %%xmm7 \n\t"
" vfmaddpd %%xmm8 , %%xmm0 , %%xmm4 , %%xmm8 \n\t"
" vfmaddpd %%xmm12, %%xmm1 , %%xmm4 , %%xmm12 \n\t"
" vfmaddpd %%xmm9 , %%xmm0 , %%xmm5 , %%xmm9 \n\t"
" vfmaddpd %%xmm13, %%xmm1 , %%xmm5 , %%xmm13 \n\t"
" vfmaddpd %%xmm10, %%xmm0 , %%xmm6 , %%xmm10 \n\t"
" vfmaddpd %%xmm14, %%xmm1 , %%xmm6 , %%xmm14 \n\t"
" addq $8, %1 \n\t"
" vfmaddpd %%xmm11, %%xmm0 , %%xmm7 , %%xmm11 \n\t"
" vfmaddpd %%xmm15, %%xmm1 , %%xmm7 , %%xmm15 \n\t"
" cmpq %1, %0 \n\t"
" jz 2f \n\t"
" prefetcht0 384(%6,%1,8) \n\t"
" vmovddup (%7,%1,2), %%xmm0 \n\t" // read b
" vmovups (%6,%1,8), %%xmm4 \n\t"
" vmovddup 8(%7,%1,2), %%xmm1 \n\t"
" vmovups 16(%6,%1,8), %%xmm5 \n\t"
" vmovups 32(%6,%1,8), %%xmm6 \n\t"
" vmovups 48(%6,%1,8), %%xmm7 \n\t"
" vfmaddpd %%xmm8 , %%xmm0 , %%xmm4 , %%xmm8 \n\t"
" vfmaddpd %%xmm12, %%xmm1 , %%xmm4 , %%xmm12 \n\t"
" vfmaddpd %%xmm9 , %%xmm0 , %%xmm5 , %%xmm9 \n\t"
" vfmaddpd %%xmm13, %%xmm1 , %%xmm5 , %%xmm13 \n\t"
" vfmaddpd %%xmm10, %%xmm0 , %%xmm6 , %%xmm10 \n\t"
" vfmaddpd %%xmm14, %%xmm1 , %%xmm6 , %%xmm14 \n\t"
" addq $8, %1 \n\t"
" vfmaddpd %%xmm11, %%xmm0 , %%xmm7 , %%xmm11 \n\t"
" vfmaddpd %%xmm15, %%xmm1 , %%xmm7 , %%xmm15 \n\t"
" cmpq %1, %0 \n\t"
" jz 2f \n\t"
" prefetcht0 384(%6,%1,8) \n\t"
" vmovddup (%7,%1,2), %%xmm0 \n\t" // read b
" vmovddup 8(%7,%1,2), %%xmm1 \n\t"
" vmovups (%6,%1,8), %%xmm4 \n\t"
" vmovups 16(%6,%1,8), %%xmm5 \n\t"
" vmovups 32(%6,%1,8), %%xmm6 \n\t"
" vmovups 48(%6,%1,8), %%xmm7 \n\t"
" vfmaddpd %%xmm8 , %%xmm0 , %%xmm4 , %%xmm8 \n\t"
" vfmaddpd %%xmm12, %%xmm1 , %%xmm4 , %%xmm12 \n\t"
" vfmaddpd %%xmm9 , %%xmm0 , %%xmm5 , %%xmm9 \n\t"
" vfmaddpd %%xmm13, %%xmm1 , %%xmm5 , %%xmm13 \n\t"
" vfmaddpd %%xmm10, %%xmm0 , %%xmm6 , %%xmm10 \n\t"
" vfmaddpd %%xmm14, %%xmm1 , %%xmm6 , %%xmm14 \n\t"
" addq $8, %1 \n\t"
" vfmaddpd %%xmm11, %%xmm0 , %%xmm7 , %%xmm11 \n\t"
" vfmaddpd %%xmm15, %%xmm1 , %%xmm7 , %%xmm15 \n\t"
" cmpq %1, %0 \n\t"
" jnz 1b \n\t"
"2: \n\t"
" vmovups (%4) , %%xmm0 \n\t"
" vmovups 16(%4) , %%xmm1 \n\t"
" vmovups 32(%4) , %%xmm2 \n\t"
" vmovups 48(%4) , %%xmm3 \n\t"
" vmovups (%5) , %%xmm4 \n\t"
" vmovups 16(%5) , %%xmm5 \n\t"
" vmovups 32(%5) , %%xmm6 \n\t"
" vmovups 48(%5) , %%xmm7 \n\t"
" vsubpd %%xmm8 , %%xmm0 , %%xmm8 \n\t"
" vsubpd %%xmm9 , %%xmm1 , %%xmm9 \n\t"
" vsubpd %%xmm10, %%xmm2 , %%xmm10 \n\t"
" vsubpd %%xmm11, %%xmm3 , %%xmm11 \n\t"
" vsubpd %%xmm12, %%xmm4 , %%xmm12 \n\t"
" vsubpd %%xmm13, %%xmm5 , %%xmm13 \n\t"
" vsubpd %%xmm14, %%xmm6 , %%xmm14 \n\t"
" vsubpd %%xmm15, %%xmm7 , %%xmm15 \n\t"
"3: \n\t" // i = 1
" vmovddup (%3), %%xmm1 \n\t" // read b
" vmovddup 8(%3), %%xmm0 \n\t" // read bb
" vmulpd %%xmm12 , %%xmm0 , %%xmm12 \n\t" // aa * bb
" vmulpd %%xmm13 , %%xmm0 , %%xmm13 \n\t" // aa * bb
" vmulpd %%xmm14 , %%xmm0 , %%xmm14 \n\t" // aa * bb
" vmulpd %%xmm15 , %%xmm0 , %%xmm15 \n\t" // aa * bb
" vmovups %%xmm12 , (%2) \n\t" // write a
" vmovups %%xmm13 , 16(%2) \n\t" // write a
" vmovups %%xmm14 , 32(%2) \n\t" // write a
" vmovups %%xmm15 , 48(%2) \n\t" // write a
" vmovups %%xmm12 , (%5) \n\t" // write c1
" vmovups %%xmm13 , 16(%5) \n\t"
" vmovups %%xmm14 , 32(%5) \n\t"
" vmovups %%xmm15 , 48(%5) \n\t"
" vfnmaddpd %%xmm8 , %%xmm12 , %%xmm1 , %%xmm8 \n\t" // c = c - aa * b
" vfnmaddpd %%xmm9 , %%xmm13 , %%xmm1 , %%xmm9 \n\t"
" vfnmaddpd %%xmm10 , %%xmm14 , %%xmm1 , %%xmm10 \n\t"
" vfnmaddpd %%xmm11 , %%xmm15 , %%xmm1 , %%xmm11 \n\t"
" \n\t" // i = 0
" subq $16 , %3 \n\t" // b = b - 2
" subq $64 , %2 \n\t" // a = a - 8
" vmovddup (%3), %%xmm0 \n\t" // read bb
" vmulpd %%xmm8 , %%xmm0 , %%xmm8 \n\t" // aa * bb
" vmulpd %%xmm9 , %%xmm0 , %%xmm9 \n\t"
" vmulpd %%xmm10 , %%xmm0 , %%xmm10 \n\t"
" vmulpd %%xmm11 , %%xmm0 , %%xmm11 \n\t"
" vmovups %%xmm8 , (%2) \n\t" // write a
" vmovups %%xmm9 , 16(%2) \n\t"
" vmovups %%xmm10 , 32(%2) \n\t"
" vmovups %%xmm11 , 48(%2) \n\t"
" vmovups %%xmm8 , (%4) \n\t" // write c0
" vmovups %%xmm9 , 16(%4) \n\t"
" vmovups %%xmm10 , 32(%4) \n\t"
" vmovups %%xmm11 , 48(%4) \n\t"
" vzeroupper \n\t"
:
"+r" (n1), // 0
"+a" (i), // 1
"+r" (as), // 2
"+r" (bs) // 3
:
"r" (c), // 4
"r" (c1), // 5
"r" (a), // 6
"r" (b) // 7
: "cc",
"%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6", "%xmm7",
"%xmm8", "%xmm9", "%xmm10", "%xmm11",
"%xmm12", "%xmm13", "%xmm14", "%xmm15",
"memory"
);
}
#ifndef COMPLEX
static inline void solve(BLASLONG m, BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc) {
FLOAT aa, bb;
int i, j, k;
a += (n - 1) * m;
b += (n - 1) * n;
for (i = n - 1; i >= 0; i--) {
bb = *(b + i);
for (j = 0; j < m; j ++) {
aa = *(c + j + i * ldc);
aa *= bb;
*a = aa;
*(c + j + i * ldc) = aa;
a ++;
for (k = 0; k < i; k ++){
*(c + j + k * ldc) -= aa * *(b + k);
}
}
b -= n;
a -= 2 * m;
}
}
#else
static inline void solve(BLASLONG m, BLASLONG n, FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc) {
FLOAT aa1, aa2;
FLOAT bb1, bb2;
FLOAT cc1, cc2;
int i, j, k;
ldc *= 2;
a += (n - 1) * m * 2;
b += (n - 1) * n * 2;
for (i = n - 1; i >= 0; i--) {
bb1 = *(b + i * 2 + 0);
bb2 = *(b + i * 2 + 1);
for (j = 0; j < m; j ++) {
aa1 = *(c + j * 2 + 0 + i * ldc);
aa2 = *(c + j * 2 + 1 + i * ldc);
#ifndef CONJ
cc1 = aa1 * bb1 - aa2 * bb2;
cc2 = aa1 * bb2 + aa2 * bb1;
#else
cc1 = aa1 * bb1 + aa2 * bb2;
cc2 = - aa1 * bb2 + aa2 * bb1;
#endif
*(a + 0) = cc1;
*(a + 1) = cc2;
*(c + j * 2 + 0 + i * ldc) = cc1;
*(c + j * 2 + 1 + i * ldc) = cc2;
a += 2;
for (k = 0; k < i; k ++){
#ifndef CONJ
*(c + j * 2 + 0 + k * ldc) -= cc1 * *(b + k * 2 + 0) - cc2 * *(b + k * 2 + 1);
*(c + j * 2 + 1 + k * ldc) -= cc1 * *(b + k * 2 + 1) + cc2 * *(b + k * 2 + 0);
#else
*(c + j * 2 + 0 + k * ldc) -= cc1 * *(b + k * 2 + 0) + cc2 * *(b + k * 2 + 1);
*(c + j * 2 + 1 + k * ldc) -= -cc1 * *(b + k * 2 + 1) + cc2 * *(b + k * 2 + 0);
#endif
}
}
b -= n * 2;
a -= 4 * m;
}
}
#endif
int CNAME(BLASLONG m, BLASLONG n, BLASLONG k, FLOAT dummy1,
#ifdef COMPLEX
FLOAT dummy2,
#endif
FLOAT *a, FLOAT *b, FLOAT *c, BLASLONG ldc, BLASLONG offset){
BLASLONG i, j;
FLOAT *aa, *cc;
BLASLONG kk;
#if 0
fprintf(stderr, "TRSM RT KERNEL m = %3ld n = %3ld k = %3ld offset = %3ld\n",
m, n, k, offset);
#endif
kk = n - offset;
c += n * ldc * COMPSIZE;
b += n * k * COMPSIZE;
if (n & (GEMM_UNROLL_N - 1)) {
j = 1;
while (j < GEMM_UNROLL_N) {
if (n & j) {
aa = a;
b -= j * k * COMPSIZE;
c -= j * ldc* COMPSIZE;
cc = c;
i = (m >> GEMM_UNROLL_M_SHIFT);
if (i > 0) {
do {
if (k - kk > 0) {
GEMM_KERNEL(GEMM_UNROLL_M, j, k - kk, dm1,
#ifdef COMPLEX
ZERO,
#endif
aa + GEMM_UNROLL_M * kk * COMPSIZE,
b + j * kk * COMPSIZE,
cc,
ldc);
}
solve(GEMM_UNROLL_M, j,
aa + (kk - j) * GEMM_UNROLL_M * COMPSIZE,
b + (kk - j) * j * COMPSIZE,
cc, ldc);
aa += GEMM_UNROLL_M * k * COMPSIZE;
cc += GEMM_UNROLL_M * COMPSIZE;
i --;
} while (i > 0);
}
if (m & (GEMM_UNROLL_M - 1)) {
i = (GEMM_UNROLL_M >> 1);
do {
if (m & i) {
if (k - kk > 0) {
GEMM_KERNEL(i, j, k - kk, dm1,
#ifdef COMPLEX
ZERO,
#endif
aa + i * kk * COMPSIZE,
b + j * kk * COMPSIZE,
cc, ldc);
}
solve(i, j,
aa + (kk - j) * i * COMPSIZE,
b + (kk - j) * j * COMPSIZE,
cc, ldc);
aa += i * k * COMPSIZE;
cc += i * COMPSIZE;
}
i >>= 1;
} while (i > 0);
}
kk -= j;
}
j <<= 1;
}
}
j = (n >> GEMM_UNROLL_N_SHIFT);
if (j > 0) {
do {
aa = a;
b -= GEMM_UNROLL_N * k * COMPSIZE;
c -= GEMM_UNROLL_N * ldc * COMPSIZE;
cc = c;
i = (m >> GEMM_UNROLL_M_SHIFT);
if (i > 0) {
do {
dtrsm_RT_solve_opt(k - kk, aa + GEMM_UNROLL_M * kk * COMPSIZE, b + GEMM_UNROLL_N * kk * COMPSIZE, cc, ldc,
aa + (kk - GEMM_UNROLL_N) * GEMM_UNROLL_M * COMPSIZE , b + (kk - GEMM_UNROLL_N) * GEMM_UNROLL_N * COMPSIZE );
aa += GEMM_UNROLL_M * k * COMPSIZE;
cc += GEMM_UNROLL_M * COMPSIZE;
i --;
} while (i > 0);
}
if (m & (GEMM_UNROLL_M - 1)) {
i = (GEMM_UNROLL_M >> 1);
do {
if (m & i) {
if (k - kk > 0) {
GEMM_KERNEL(i, GEMM_UNROLL_N, k - kk, dm1,
#ifdef COMPLEX
ZERO,
#endif
aa + i * kk * COMPSIZE,
b + GEMM_UNROLL_N * kk * COMPSIZE,
cc,
ldc);
}
solve(i, GEMM_UNROLL_N,
aa + (kk - GEMM_UNROLL_N) * i * COMPSIZE,
b + (kk - GEMM_UNROLL_N) * GEMM_UNROLL_N * COMPSIZE,
cc, ldc);
aa += i * k * COMPSIZE;
cc += i * COMPSIZE;
}
i >>= 1;
} while (i > 0);
}
kk -= GEMM_UNROLL_N;
j --;
} while (j > 0);
}
return 0;
}
Reference in New Issue View Git Blame Copy Permalink