📄 qgemv_t.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 PENTIUM#define P 88#endif#ifndef P#define P 1000#endif#define STACK 16#define ARGS 24 #define NLDA 0 + STACK(%esp)#define XP 4 + STACK(%esp)#define MIN_M 8 + STACK(%esp)#define J 12 + STACK(%esp)#define IS 16 + STACK(%esp)#define M 4 + STACK + ARGS(%esp)#define N 8 + STACK + ARGS(%esp)#define K 12 + STACK + ARGS(%esp)#define ALPHA 16 + STACK + ARGS(%esp)#define A 32 + STACK + ARGS(%esp)#define LDA 36 + STACK + ARGS(%esp)#define X 40 + STACK + ARGS(%esp)#define INCX 44 + STACK + ARGS(%esp)#define Y 48 + STACK + ARGS(%esp)#define INCY 52 + STACK + ARGS(%esp)#define BUFFER 56 + STACK + ARGS(%esp) PROLOGUE subl $ARGS, %esp pushl %ebp pushl %edi pushl %esi pushl %ebx PROFCODE FLD ALPHA movl X, %edi # X movl $0, IS movl M, %ebx movl N, %eax testl %ebx, %ebx jle .L79 testl %eax, %eax jle .L79 movl INCX, %esi sall $BASE_SHIFT, %esi movl %esi, INCX movl INCY, %esi sall $BASE_SHIFT, %esi movl %esi, INCY movl LDA, %ebx imull %ebx, %eax movl $P, %esi subl %eax, %esi sall $BASE_SHIFT, %esi movl %esi, NLDA movl %ebx, %esi sall $BASE_SHIFT, %esi movl %esi, LDA ALIGN_2.L32: movl IS, %esi movl $P, %edx movl M, %eax subl %esi, %eax cmpl %edx, %eax#ifdef PENTIUM jle .L33 movl %edx, %eax.L33:#else cmovg %edx, %eax#endif movl %eax, MIN_M movl IS, %ecx sall $BASE_SHIFT, %ecx leal (%edi,%ecx, 1), %ecx movl INCX, %ebx movl %ecx, XP cmpl $SIZE, %ebx je .L34 movl BUFFER, %esi movl MIN_M, %ecx movl %esi, XP sarl $2, %ecx jle .L35 ALIGN_3.L36: FLD (%edi) addl %ebx, %edi FST 0 * SIZE(%esi) FLD (%edi) addl %ebx, %edi FST 1 * SIZE(%esi) FLD (%edi) addl %ebx, %edi FST 2 * SIZE(%esi) FLD (%edi) addl %ebx, %edi FST 3 * SIZE(%esi) addl $4 * SIZE, %esi decl %ecx jg .L36 ALIGN_3.L35: movl MIN_M, %ecx andl $3,%ecx jle .L34 ALIGN_2.L42: FLD (%edi) addl %ebx, %edi FST (%esi) addl $SIZE, %esi decl %ecx jg .L42 ALIGN_3/* Main Routine */.L34: movl Y, %ebp # coffset = y movl N, %esi sarl $2, %esi movl %esi, J jle .L47 ALIGN_3.L48: movl A, %ebx # a_offset = a fldz movl LDA, %edx fldz leal (%ebx, %edx), %ecx # a_offset2 = a + lda fldz leal (%ebx, %edx, 4), %eax fldz movl %eax, A movl XP, %esi FLD (%esi) movl MIN_M, %eax sarl $2,%eax jle .L51 ALIGN_3#define PRESIZE 8.L80:#ifdef PENTIUM3 prefetcht0 PRESIZE * SIZE(%ebx, %edx, 2) FLD 0 * SIZE(%ebx) # at = *(a_offset + 0 * lda) fmul %st(1),%st # at1 *= bt1 prefetcht0 PRESIZE * SIZE(%ecx) faddp %st,%st(2) # ct1 += at1 FLD 0 * SIZE(%ecx) # at1 = *(a_offset2 + 0 * lda) prefetcht0 PRESIZE * SIZE(%ecx, %edx, 2) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(3) # ct2 += at1 prefetcht0 PRESIZE * SIZE(%ebx) FLD 0 * SIZE(%ebx, %edx, 2) # at = *(a_offset + 2 * lda) fmul %st(1),%st faddp %st,%st(4) FLD 0 * SIZE(%ecx, %edx, 2) # at1 = *(a_offset2 + 2 * lda) fmulp %st, %st(1) faddp %st,%st(4) FLD 1 * SIZE(%esi) FLD 1 * SIZE(%ebx) # at = *(a_offset + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(2) # ct1 += at1 FLD 1 * SIZE(%ecx) # at1 = *(a_offset2 + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(3) # ct2 += at1 FLD 1 * SIZE(%ebx, %edx, 2) # at = *(a_offset + 2 * lda) fmul %st(1),%st faddp %st,%st(4) FLD 1 * SIZE(%ecx, %edx, 2) # at1 = *(a_offset2 + 2 * lda) fmulp %st, %st(1) faddp %st,%st(4) FLD 2 * SIZE(%esi) FLD 2 * SIZE(%ebx) # at = *(a_offset + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(2) # ct1 += at1 FLD 2 * SIZE(%ecx) # at1 = *(a_offset2 + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(3) # ct2 += at1 FLD 2 * SIZE(%ebx, %edx, 2) # at = *(a_offset + 2 * lda) fmul %st(1),%st faddp %st,%st(4) FLD 2 * SIZE(%ecx, %edx, 2) # at1 = *(a_offset2 + 2 * lda) fmulp %st, %st(1) faddp %st,%st(4) FLD 3 * SIZE(%esi) FLD 3 * SIZE(%ebx) # at = *(a_offset + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(2) # ct1 += at1 FLD 3 * SIZE(%ecx) # at1 = *(a_offset2 + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(3) # ct2 += at1 FLD 3 * SIZE(%ebx, %edx, 2) # at = *(a_offset + 2 * lda) fmul %st(1),%st faddp %st,%st(4) FLD 3 * SIZE(%ecx, %edx, 2) # at1 = *(a_offset2 + 2 * lda) fmulp %st, %st(1) addl $4 * SIZE, %ebx faddp %st,%st(4) addl $4 * SIZE, %ecx FLD 4 * SIZE(%esi) addl $4 * SIZE, %esi#else #if defined(HAS_PREFETCH) prefetcht0 PRESIZE * SIZE(%ebx) prefetcht0 PRESIZE * SIZE(%ebx, %edx, 2) prefetcht0 PRESIZE * SIZE(%ecx) prefetcht0 PRESIZE * SIZE(%ecx, %edx, 2)#endif FLD 0 * SIZE(%ebx) # at = *(a_offset + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(2) # ct1 += at1 FLD 0 * SIZE(%ecx) # at1 = *(a_offset2 + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(3) # ct2 += at1 FLD 0 * SIZE(%ebx, %edx, 2) # at = *(a_offset + 2 * lda) fmul %st(1),%st faddp %st,%st(4) FLD 0 * SIZE(%ecx, %edx, 2) # at1 = *(a_offset2 + 2 * lda) fmulp %st, %st(1) faddp %st,%st(4) FLD 1 * SIZE(%esi) FLD 1 * SIZE(%ebx) # at = *(a_offset + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(2) # ct1 += at1 FLD 1 * SIZE(%ecx) # at1 = *(a_offset2 + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(3) # ct2 += at1 FLD 1 * SIZE(%ebx, %edx, 2) # at = *(a_offset + 2 * lda) fmul %st(1),%st faddp %st,%st(4) FLD 1 * SIZE(%ecx, %edx, 2) # at1 = *(a_offset2 + 2 * lda) fmulp %st, %st(1) faddp %st,%st(4) FLD 2 * SIZE(%esi) FLD 2 * SIZE(%ebx) # at = *(a_offset + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(2) # ct1 += at1 FLD 2 * SIZE(%ecx) # at1 = *(a_offset2 + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(3) # ct2 += at1 FLD 2 * SIZE(%ebx, %edx, 2) # at = *(a_offset + 2 * lda) fmul %st(1),%st faddp %st,%st(4) FLD 2 * SIZE(%ecx, %edx, 2) # at1 = *(a_offset2 + 2 * lda) fmulp %st, %st(1) faddp %st,%st(4) FLD 3 * SIZE(%esi) FLD 3 * SIZE(%ebx) # at = *(a_offset + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(2) # ct1 += at1 FLD 3 * SIZE(%ecx) # at1 = *(a_offset2 + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(3) # ct2 += at1 FLD 3 * SIZE(%ebx, %edx, 2) # at = *(a_offset + 2 * lda) fmul %st(1),%st faddp %st,%st(4) FLD 3 * SIZE(%ecx, %edx, 2) # at1 = *(a_offset2 + 2 * lda) fmulp %st, %st(1) faddp %st,%st(4) FLD 4 * SIZE(%esi) addl $4 * SIZE, %ebx addl $4 * SIZE, %ecx addl $4 * SIZE, %esi#endif decl %eax jg .L80 ALIGN_3.L51: movl MIN_M, %eax andl $3, %eax je .L81 ALIGN_3.L52: FLD (%ebx) # at = *(a_offset + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(2) # ct1 += at1 FLD (%ecx) # at1 = *(a_offset2 + 0 * lda) fmul %st(1),%st # at1 *= bt1 faddp %st,%st(3) # ct2 += at1 FLD (%ebx, %edx, 2) # at = *(a_offset + 2 * lda) fmul %st(1),%st faddp %st,%st(4) FLD (%ecx, %edx, 2) # at1 = *(a_offset2 + 2 * lda) fmulp %st, %st(1) faddp %st,%st(4) FLD 1 * SIZE(%esi) addl $SIZE, %ebx addl $SIZE, %ecx addl $SIZE, %esi decl %eax jg .L52 ALIGN_3.L81: ffreep %st(0) fxch %st(4) fmul %st, %st(4) fmul %st, %st(1) fmul %st, %st(2) fmul %st, %st(3) fxch %st(4) movl INCY, %eax FLD (%ebp) faddp %st, %st(1) FST (%ebp) addl %eax, %ebp FLD (%ebp) faddp %st, %st(1) FST (%ebp) addl %eax, %ebp FLD (%ebp) faddp %st, %st(1) FST (%ebp) addl %eax, %ebp FLD (%ebp) faddp %st, %st(1) FST (%ebp) addl %eax, %ebp decl J jg .L48 ALIGN_3.L47: movl N, %esi andl $3,%esi movl %esi, J jle .L60 ALIGN_2.L61: movl A, %ebx # a_offset = a fldz # ct1 = ZERO movl LDA, %edx fldz # ct1 = ZERO addl %ebx, %edx fldz # ct1 = ZERO movl %edx, A fldz # ct1 = ZERO movl XP, %esi movl MIN_M, %eax sarl $3,%eax jle .L64 ALIGN_3.L65:#ifdef HAS_PREFETCH prefetcht0 PRESIZE * 2 * SIZE(%ebx) prefetcht0 PRESIZE * 2 * SIZE(%ebx)#endif FLD 0 * SIZE(%esi) FLD 0 * SIZE(%ebx) fmulp %st, %st(1) faddp %st,%st(1) FLD 1 * SIZE(%esi) FLD 1 * SIZE(%ebx) fmulp %st, %st(1) faddp %st,%st(2) FLD 2 * SIZE(%esi) FLD 2 * SIZE(%ebx) fmulp %st, %st(1) faddp %st,%st(3) FLD 3 * SIZE(%esi) FLD 3 * SIZE(%ebx) fmulp %st, %st(1) faddp %st,%st(4) FLD 4 * SIZE(%esi) FLD 4 * SIZE(%ebx) fmulp %st, %st(1) faddp %st,%st(1) FLD 5 * SIZE(%esi) FLD 5 * SIZE(%ebx) fmulp %st, %st(1) faddp %st,%st(2) FLD 6 * SIZE(%esi) FLD 6 * SIZE(%ebx) fmulp %st, %st(1) faddp %st,%st(3) FLD 7 * SIZE(%esi) FLD 7 * SIZE(%ebx) fmulp %st, %st(1) faddp %st,%st(4) addl $8 * SIZE, %esi addl $8 * SIZE, %ebx decl %eax jg .L65 ALIGN_3.L64: movl MIN_M, %eax andl $7, %eax jle .L70 ALIGN_3.L71: FLD (%esi) FLD (%ebx) fmulp %st, %st(1) faddp %st,%st(1) addl $SIZE, %esi addl $SIZE, %ebx decl %eax jg .L71 ALIGN_3.L70: faddp %st, %st(1) faddp %st, %st(1) faddp %st, %st(1) fmul %st(1),%st FLD (%ebp) faddp %st, %st(1) FST (%ebp) addl INCY, %ebp decl J jg .L61 ALIGN_3.L60: movl A, %ebx addl NLDA, %ebx movl %ebx, A addl $P, IS movl M, %esi cmpl %esi, IS jl .L32 ALIGN_3.L79: ffreep %st(0) popl %ebx popl %esi popl %edi popl %ebp addl $ARGS, %esp ret 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 + -