📄 zswap.s
字号:
/*********************************************************************//* *//* Optimized BLAS libraries *//* By Kazushige Goto <kgoto@tacc.utexas.edu> *//* *//* Copyright (c) The University of Texas, 2005. All rights reserved. *//* UNIVERSITY EXPRESSLY DISCLAIMS ANY AND ALL WARRANTIES CONCERNING *//* THIS SOFTWARE AND DOCUMENTATION, INCLUDING ANY WARRANTIES OF *//* MERCHANTABILITY, FITNESS FOR ANY PARTICULAR PURPOSE, *//* NON-INFRINGEMENT AND WARRANTIES OF PERFORMANCE, AND ANY WARRANTY *//* THAT MIGHT OTHERWISE ARISE FROM COURSE OF DEALING OR USAGE OF *//* TRADE. NO WARRANTY IS EITHER EXPRESS OR IMPLIED WITH RESPECT TO *//* THE USE OF THE SOFTWARE OR DOCUMENTATION. *//* Under no circumstances shall University be liable for incidental, *//* special, indirect, direct or consequential damages or loss of *//* profits, interruption of business, or related expenses which may *//* arise from use of Software or Documentation, including but not *//* limited to those resulting from defects in Software and/or *//* Documentation, or loss or inaccuracy of data of any kind. *//*********************************************************************/#define ASSEMBLER#include "common.h"#ifdef XDOUBLE#define PREFETCH_SIZE ( 8 * 16)#elif defined(DOUBLE)#define PREFETCH_SIZE (16 * 16)#else#define PREFETCH_SIZE (32 * 16)#endif#define SP r12#ifdef XDOUBLE#define N r32#define X r14#define INCX r15#define Y r16#define INCY r17#else#define N r32#define X r37#define INCX r38#define Y r39#define INCY r36#endif#define PRE1 r2#define PRE2 r3#define I r18#define J r19#define YY r20#define XX r21#define INCXM1 r22#define INCYM1 r23#define INCX8 r24#define INCY8 r25#define PR r30#define ARLC r31 PROLOGUE .prologue PROFCODE { .mmi adds r14 = 16, SP adds r15 = 24, SP adds r16 = 32, SP } { .mmb adds r17 = 40, SP cmp.gt p15, p0 = r0, N (p15) br.ret.sptk.many b0 } ;; #ifdef XDOUBLE { .mmi ld8 X = [r14] ld8 INCX = [r15] nop __LINE__ } { .mmi ld8 Y = [r16] ld8 INCY = [r17] nop __LINE__ } ;;#else { .mmi ld8 INCY = [r14] nop __LINE__ nop __LINE__ } ;;#endif { .mii .save ar.lc, ARLC mov ARLC = ar.lc shl INCX = INCX, ZBASE_SHIFT } ;; .body { .mii and J = 7, N mov PR = pr shl INCY = INCY, ZBASE_SHIFT } ;; { .mmi mov XX = X mov YY = Y shr I = N, 3 } ;; { .mmi adds I = -1, I cmp.eq p9, p0 = r0, J mov pr.rot = 0 } ;; { .mmi shladd INCX8 = INCX, 3, r0 shladd INCY8 = INCY, 3, r0 mov ar.ec= 3 } { .mmi adds INCXM1 = -SIZE, INCX adds INCYM1 = -SIZE, INCY cmp.eq p16, p0 = r0, r0 } ;; { .mmi adds PRE1 = PREFETCH_SIZE * SIZE, X adds PRE2 = PREFETCH_SIZE * SIZE, Y mov ar.lc = I } { .mib cmp.eq p8 ,p0 = -1, I tbit.z p0, p12 = J, 2 (p8) br.cond.dpnt .L55 } ;; .align 32.L52: { .mmi (p18) STFD [XX] = f37, 1 * SIZE (p18) STFD [YY] = f34, 1 * SIZE } { .mmi (p16) LDFD f32 = [X], 1 * SIZE (p16) LDFD f35 = [Y], 1 * SIZE } ;; { .mmi (p18) STFD [XX] = f43 (p18) STFD [YY] = f40 (p18) add XX = XX, INCXM1 } { .mmi (p16) LDFD f38 = [X], INCXM1 (p16) LDFD f41 = [Y], INCYM1 (p18) add YY = YY, INCYM1 } ;; { .mmi (p18) STFD [XX] = f49, 1 * SIZE (p18) STFD [YY] = f46, 1 * SIZE } { .mmi (p16) LDFD f44 = [X], 1 * SIZE (p16) LDFD f47 = [Y], 1 * SIZE } ;; { .mmi (p18) STFD [XX] = f55 (p18) STFD [YY] = f52 (p18) add XX = XX, INCXM1 } { .mmi (p16) LDFD f50 = [X], INCXM1 (p16) LDFD f53 = [Y], INCYM1 (p18) add YY = YY, INCYM1 } ;; { .mmi (p18) STFD [XX] = f61, 1 * SIZE (p18) STFD [YY] = f58, 1 * SIZE } { .mmi (p16) LDFD f56 = [X], 1 * SIZE (p16) LDFD f59 = [Y], 1 * SIZE } ;; { .mmi (p18) STFD [XX] = f67 (p18) STFD [YY] = f64 (p18) add XX = XX, INCXM1 } { .mmi (p16) LDFD f62 = [X], INCXM1 (p16) LDFD f65 = [Y], INCYM1 (p18) add YY = YY, INCYM1 } ;; { .mmi (p18) STFD [XX] = f73, 1 * SIZE (p18) STFD [YY] = f70, 1 * SIZE } { .mmi (p16) LDFD f68 = [X], 1 * SIZE (p16) LDFD f71 = [Y], 1 * SIZE } ;; { .mmi (p18) STFD [XX] = f79 (p18) STFD [YY] = f76 (p18) add XX = XX, INCXM1 } { .mmi (p16) LDFD f74 = [X], INCXM1 (p16) LDFD f77 = [Y], INCYM1 (p18) add YY = YY, INCYM1 } ;; { .mmi (p18) STFD [XX] = f85, 1 * SIZE (p18) STFD [YY] = f82, 1 * SIZE } { .mmi (p16) LDFD f80 = [X], 1 * SIZE (p16) LDFD f83 = [Y], 1 * SIZE } ;; { .mmi (p18) STFD [XX] = f91 (p18) STFD [YY] = f88 (p18) add XX = XX, INCXM1 } { .mmi (p16) LDFD f86 = [X], INCXM1 (p16) LDFD f89 = [Y], INCYM1 (p18) add YY = YY, INCYM1 } ;; { .mmi (p18) STFD [XX] = f97, 1 * SIZE (p18) STFD [YY] = f94, 1 * SIZE } { .mmi (p16) LDFD f92 = [X], 1 * SIZE (p16) LDFD f95 = [Y], 1 * SIZE } ;; { .mmi (p18) STFD [XX] = f103 (p18) STFD [YY] = f100 (p18) add XX = XX, INCXM1 } { .mmi (p16) LDFD f98 = [X], INCXM1 (p16) LDFD f101 = [Y], INCYM1 (p18) add YY = YY, INCYM1 } ;; { .mmi (p18) STFD [XX] = f109, 1 * SIZE (p18) STFD [YY] = f106, 1 * SIZE } { .mmi (p16) LDFD f104 = [X], 1 * SIZE (p16) LDFD f107 = [Y], 1 * SIZE } ;; { .mmi (p18) STFD [XX] = f115 (p18) STFD [YY] = f112 (p18) add XX = XX, INCXM1 } { .mmi (p16) LDFD f110 = [X], INCXM1 (p16) LDFD f113 = [Y], INCYM1 (p18) add YY = YY, INCYM1 } ;; { .mmi (p18) STFD [XX] = f121, 1 * SIZE (p18) STFD [YY] = f118, 1 * SIZE } { .mmi (p16) LDFD f116 = [X], 1 * SIZE (p16) LDFD f119 = [Y], 1 * SIZE } ;; { .mmi (p18) STFD [XX] = f127 (p18) STFD [YY] = f124 (p18) add XX = XX, INCXM1 } { .mmi (p16) LDFD f122 = [X], INCXM1 (p16) LDFD f125 = [Y], INCYM1 (p18) add YY = YY, INCYM1 } { .mmb (p16) lfetch.excl.nt1 [PRE1], INCX8 (p16) lfetch.excl.nt1 [PRE2], INCY8 br.ctop.sptk.few .L52 } ;; .align 32.L55: { .mmi (p12) LDFD f32 = [X], 1 * SIZE (p12) LDFD f80 = [Y], 1 * SIZE mov ar.lc = ARLC } ;; { .mmi (p12) LDFD f33 = [X], INCXM1 (p12) LDFD f81 = [Y], INCYM1 mov pr = PR, -65474 } ;; { .mmb (p12) LDFD f34 = [X], 1 * SIZE (p12) LDFD f82 = [Y], 1 * SIZE (p9) br.ret.sptk.many b0 } ;; { .mmi (p12) LDFD f35 = [X], INCXM1 (p12) LDFD f83 = [Y], INCYM1 tbit.z p0, p13 = N, 1 } ;; { .mmi (p12) LDFD f36 = [X], 1 * SIZE (p12) LDFD f84 = [Y], 1 * SIZE tbit.z p0, p14 = N, 0 } ;; { .mmi (p12) LDFD f37 = [X], INCXM1 (p12) LDFD f85 = [Y], INCYM1 } ;; { .mmi (p12) STFD [XX] = f80, 1 * SIZE (p12) STFD [YY] = f32, 1 * SIZE } { .mmi (p12) LDFD f38 = [X], 1 * SIZE (p12) LDFD f86 = [Y], 1 * SIZE } ;; { .mmi (p12) STFD [XX] = f81 (p12) STFD [YY] = f33 (p12) add XX = XX, INCXM1 } { .mmi (p12) LDFD f39 = [X], INCXM1 (p12) LDFD f87 = [Y], INCYM1 (p12) add YY = YY, INCYM1 } ;; { .mmi (p12) STFD [XX] = f82, 1 * SIZE (p12) STFD [YY] = f34, 1 * SIZE } { .mmi (p13) LDFD f40 = [X], 1 * SIZE (p13) LDFD f88 = [Y], 1 * SIZE } ;; { .mmi (p12) STFD [XX] = f83 (p12) STFD [YY] = f35 (p12) add XX = XX, INCXM1 } { .mmi (p13) LDFD f41 = [X], INCXM1 (p13) LDFD f89 = [Y], INCYM1 (p12) add YY = YY, INCYM1 } ;; { .mmi (p12) STFD [XX] = f84, 1 * SIZE (p12) STFD [YY] = f36, 1 * SIZE } { .mmi (p13) LDFD f42 = [X], 1 * SIZE (p13) LDFD f90 = [Y], 1 * SIZE } ;; { .mmi (p12) STFD [XX] = f85 (p12) STFD [YY] = f37 (p12) add XX = XX, INCXM1 } { .mmi (p13) LDFD f43 = [X], INCXM1 (p13) LDFD f91 = [Y], INCYM1 (p12) add YY = YY, INCYM1 } ;; { .mmi (p12) STFD [XX] = f86, 1 * SIZE (p12) STFD [YY] = f38, 1 * SIZE } { .mmi (p14) LDFD f44 = [X], 1 * SIZE (p14) LDFD f92 = [Y], 1 * SIZE } ;; { .mmi (p12) STFD [XX] = f87 (p12) STFD [YY] = f39 (p12) add XX = XX, INCXM1 } { .mmi (p14) LDFD f45 = [X] (p14) LDFD f93 = [Y] (p12) add YY = YY, INCYM1 } ;; { .mmi (p13) STFD [XX] = f88, 1 * SIZE (p13) STFD [YY] = f40, 1 * SIZE } ;; (p13) STFD [XX] = f89 (p13) add XX = XX, INCXM1 (p13) STFD [YY] = f41 (p13) add YY = YY, INCYM1 ;; (p13) STFD [XX] = f90, 1 * SIZE (p13) STFD [YY] = f42, 1 * SIZE ;; (p13) STFD [XX] = f91 (p13) add XX = XX, INCXM1 (p13) STFD [YY] = f43 (p13) add YY = YY, INCYM1 ;; (p14) STFD [XX] = f92, 1 * SIZE (p14) STFD [YY] = f44, 1 * SIZE ;; (p14) STFD [XX] = f93 (p14) STFD [YY] = f45 br.ret.sptk.many b0 ;; EPILOGUE
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -