📄 zdot_sse2_core2.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"#if defined(F_INTERFACE) && defined(F_INTERFACE_F2C)#define RESULT ARG1 /* rdi */#define N ARG2 /* rsi */#define X ARG3 /* rdx */#define INCX ARG4 /* rcx */#ifndef WINDOWS_ABI#define Y ARG5 /* r8 */#define INCY ARG6 /* r9 */#else#define Y %r10#define INCY %r11#endif#else#define N ARG1 /* rdi */#define X ARG2 /* rsi */#define INCX ARG3 /* rdx */#define Y ARG4 /* rcx */#ifndef WINDOWS_ABI#define INCY ARG5 /* r8 */#else#define INCY %r10#endif#endif#define MOVDDUP(a, b, c) movddup a(b), c#define MOVDDUP2(a, b, c) movddup a##b, c#define PREFETCH prefetcht0#define PREFETCH_SIZE (8 * 5) PROLOGUE PROFCODE#ifdef WINDOWS_ABI#if defined(F_INTERFACE) && defined(F_INTERFACE_F2C) movq 40(%rsp), Y movq 48(%rsp), INCY#else movq 40(%rsp), INCY#endif#endif SAVEREGISTERS#ifdef F_INTERFACE#ifndef USE64BITINT movslq (N), N # N movslq (INCX),INCX # INCX movslq (INCY),INCY # INCY#else movq (N), N # N movq (INCX),INCX # INCX movq (INCY),INCY # INCY#endif#endif salq $ZBASE_SHIFT, INCX salq $ZBASE_SHIFT, INCY pxor %xmm0, %xmm0 pxor %xmm1, %xmm1 pxor %xmm2, %xmm2 pxor %xmm3, %xmm3 cmpq $0, N jle .L49 pxor %xmm6, %xmm6 pxor %xmm7, %xmm7 testq $SIZE, X jne .L30 testq $SIZE, Y jne .L50 cmpq $2 * SIZE, INCX jne .L20 cmpq $2 * SIZE, INCY jne .L20 subq $-16 * SIZE, X subq $-16 * SIZE, Y movq N, %rax sarq $3, %rax jle .L15 ALIGN_3.L11: movapd -16 * SIZE(Y), %xmm5 addpd %xmm6, %xmm2 movapd -16 * SIZE(X), %xmm4 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 movapd -14 * SIZE(Y), %xmm7 addpd %xmm6, %xmm0 movapd -14 * SIZE(X), %xmm4 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 mulpd %xmm4, %xmm6 PREFETCH (PREFETCH_SIZE + 0)(Y) addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 movapd -12 * SIZE(Y), %xmm5 addpd %xmm6, %xmm2 movapd -12 * SIZE(X), %xmm4 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 movapd -10 * SIZE(Y), %xmm7 addpd %xmm6, %xmm0 movapd -10 * SIZE(X), %xmm4 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 movapd -8 * SIZE(Y), %xmm5 addpd %xmm6, %xmm2 movapd -8 * SIZE(X), %xmm4 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 movapd -6 * SIZE(Y), %xmm7 addpd %xmm6, %xmm0 movapd -6 * SIZE(X), %xmm4 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 movapd -4 * SIZE(Y), %xmm5 addpd %xmm6, %xmm2 movapd -4 * SIZE(X), %xmm4 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 movapd -2 * SIZE(Y), %xmm7 subq $-16 * SIZE, Y addpd %xmm6, %xmm0 movapd -2 * SIZE(X), %xmm4 movapd %xmm7, %xmm6 subq $-16 * SIZE, X SHUFPD_1 %xmm7, %xmm7 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 subq $1, %rax jg,pt .L11 ALIGN_3.L15: testq $4, N jle .L16 addpd %xmm6, %xmm2 movapd -16 * SIZE(Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movapd -16 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addpd %xmm6, %xmm0 movapd -14 * SIZE(Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movapd -14 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addpd %xmm6, %xmm2 movapd -12 * SIZE(Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movapd -12 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addpd %xmm6, %xmm0 movapd -10 * SIZE(Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movapd -10 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq $8 * SIZE, X addq $8 * SIZE, Y ALIGN_3.L16: testq $2, N jle .L17 addpd %xmm6, %xmm2 movapd -16 * SIZE(Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movapd -16 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addpd %xmm6, %xmm0 movapd -14 * SIZE(Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movapd -14 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq $4 * SIZE, X addq $4 * SIZE, Y ALIGN_3.L17: addpd %xmm6, %xmm2 addpd %xmm7, %xmm3 testq $1, N jle .L49 movapd -16 * SIZE(Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movapd -16 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm6, %xmm0 mulpd %xmm4, %xmm7 addpd %xmm7, %xmm1 jmp .L49 ALIGN_3.L20:#ifdef F_INTERFACE testq INCX, INCX # if (incx < 0) jge .L21 movq N, %rax # n decq %rax # n - 1 imulq INCX, %rax # (n - 1) * incx subq %rax, X ALIGN_3.L21: testq INCY, INCY # if (incy < 0) jge .L22 movq N, %rax decq %rax # (n - 1) imulq INCY, %rax # (n - 1) * incy subq %rax, Y ALIGN_3.L22:#endif movq N, %rax sarq $3, %rax jle .L25 ALIGN_3.L23: addpd %xmm6, %xmm2 movapd (Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addq INCX, X addq INCY, Y addpd %xmm6, %xmm0 movapd (Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq INCX, X addq INCY, Y addpd %xmm6, %xmm2 movapd (Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addq INCX, X addq INCY, Y addpd %xmm6, %xmm0 movapd (Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq INCX, X addq INCY, Y addpd %xmm6, %xmm2 movapd (Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addq INCX, X addq INCY, Y addpd %xmm6, %xmm0 movapd (Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq INCX, X addq INCY, Y addpd %xmm6, %xmm2 movapd (Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addq INCX, X addq INCY, Y addpd %xmm6, %xmm0 movapd (Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq INCX, X addq INCY, Y decq %rax jg .L23 ALIGN_3.L25: testq $4, N jle .L26 addpd %xmm6, %xmm2 movapd (Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addq INCX, X addq INCY, Y addpd %xmm6, %xmm0 movapd (Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq INCX, X addq INCY, Y addpd %xmm6, %xmm2 movapd (Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addq INCX, X addq INCY, Y addpd %xmm6, %xmm0 movapd (Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq INCX, X addq INCY, Y ALIGN_3.L26: testq $2, N jle .L27 addpd %xmm6, %xmm2 movapd (Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addq INCX, X addq INCY, Y addpd %xmm6, %xmm0 movapd (Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq INCX, X addq INCY, Y ALIGN_3.L27: addpd %xmm6, %xmm2 addpd %xmm7, %xmm3 testq $1, N jle .L49 movapd (Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movapd (X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm6, %xmm0 mulpd %xmm4, %xmm7 addpd %xmm7, %xmm1 jmp .L49 ALIGN_3.L30: testq $SIZE, Y jne .L70 /* Y is aligned */ cmpq $2 * SIZE, INCX jne .L40 cmpq $2 * SIZE, INCY jne .L40 subq $-16 * SIZE, X subq $-16 * SIZE, Y movq N, %rax sarq $3, %rax jle .L35 ALIGN_3.L31: addpd %xmm6, %xmm2 movapd -16 * SIZE(Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movsd -16 * SIZE(X), %xmm4 movhpd -15 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 PREFETCH (PREFETCH_SIZE + 0)(X) addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addpd %xmm6, %xmm0 movapd -14 * SIZE(Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movsd -14 * SIZE(X), %xmm4 movhpd -13 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addpd %xmm6, %xmm2 movapd -12 * SIZE(Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movsd -12 * SIZE(X), %xmm4 movhpd -11 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addpd %xmm6, %xmm0 movapd -10 * SIZE(Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movsd -10 * SIZE(X), %xmm4 movhpd -9 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addpd %xmm6, %xmm2 movapd -8 * SIZE(Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movsd -8 * SIZE(X), %xmm4 movhpd -7 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addpd %xmm6, %xmm0 movapd -6 * SIZE(Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movsd -6 * SIZE(X), %xmm4 movhpd -5 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addpd %xmm6, %xmm2 movapd -4 * SIZE(Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movsd -4 * SIZE(X), %xmm4 movhpd -3 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addpd %xmm6, %xmm0 movapd -2 * SIZE(Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movsd -2 * SIZE(X), %xmm4 movhpd -1 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 subq $-16 * SIZE, X subq $-16 * SIZE, Y decq %rax jg .L31 ALIGN_3.L35: testq $4, N jle .L36 addpd %xmm6, %xmm2 movapd -16 * SIZE(Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movsd -16 * SIZE(X), %xmm4 movhpd -15 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addpd %xmm6, %xmm0 movapd -14 * SIZE(Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movsd -14 * SIZE(X), %xmm4 movhpd -13 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addpd %xmm6, %xmm2 movapd -12 * SIZE(Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movsd -12 * SIZE(X), %xmm4 movhpd -11 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addpd %xmm6, %xmm0 movapd -10 * SIZE(Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movsd -10 * SIZE(X), %xmm4 movhpd -9 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq $8 * SIZE, X addq $8 * SIZE, Y ALIGN_3.L36: testq $2, N jle .L37 addpd %xmm6, %xmm2 movapd -16 * SIZE(Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movsd -16 * SIZE(X), %xmm4 movhpd -15 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addpd %xmm6, %xmm0 movapd -14 * SIZE(Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movsd -14 * SIZE(X), %xmm4 movhpd -13 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq $4 * SIZE, X addq $4 * SIZE, Y ALIGN_3.L37: addpd %xmm6, %xmm2 addpd %xmm7, %xmm3 testq $1, N jle .L49 movapd -16 * SIZE(Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movsd -16 * SIZE(X), %xmm4 movhpd -15 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm6, %xmm0 mulpd %xmm4, %xmm7 addpd %xmm7, %xmm1 jmp .L49 ALIGN_3.L40:#ifdef F_INTERFACE testq INCX, INCX jge .L41 movq N, %rax decq %rax imulq INCX, %rax subq %rax, X ALIGN_3.L41: testq INCY, INCY jge .L42 movq N, %rax decq %rax imulq INCY, %rax subq %rax, Y ALIGN_3.L42:#endif movq N, %rax sarq $3, %rax jle .L45 ALIGN_3.L43: addpd %xmm6, %xmm2 movapd (Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movsd (X), %xmm4 movhpd 1 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addq INCX, X addq INCY, Y addpd %xmm6, %xmm0 movapd (Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movsd (X), %xmm4 movhpd 1 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq INCX, X addq INCY, Y addpd %xmm6, %xmm2 movapd (Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movsd (X), %xmm4 movhpd 1 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addq INCX, X addq INCY, Y addpd %xmm6, %xmm0 movapd (Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movsd (X), %xmm4 movhpd 1 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq INCX, X addq INCY, Y addpd %xmm6, %xmm2 movapd (Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movsd (X), %xmm4 movhpd 1 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addq INCX, X addq INCY, Y addpd %xmm6, %xmm0 movapd (Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movsd (X), %xmm4 movhpd 1 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq INCX, X addq INCY, Y addpd %xmm6, %xmm2 movapd (Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movsd (X), %xmm4 movhpd 1 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5 addq INCX, X addq INCY, Y addpd %xmm6, %xmm0 movapd (Y), %xmm7 movapd %xmm7, %xmm6 SHUFPD_1 %xmm7, %xmm7 movsd (X), %xmm4 movhpd 1 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm5, %xmm1 mulpd %xmm4, %xmm7 addq INCX, X addq INCY, Y decq %rax jg .L43 ALIGN_3.L45: testq $4, N jle .L46 addpd %xmm6, %xmm2 movapd (Y), %xmm5 movapd %xmm5, %xmm6 SHUFPD_1 %xmm5, %xmm5 movsd (X), %xmm4 movhpd 1 * SIZE(X), %xmm4 mulpd %xmm4, %xmm6 addpd %xmm7, %xmm3 mulpd %xmm4, %xmm5⌨️ 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 + -