nb_kernel204_ia32_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,120 行 · 第 1/5 页

;#
;# $Id: nb_kernel204_ia32_sse2.intel_syntax.s,v 1.1.2.2 2007/11/28 18:35:34 spoel Exp $
;#
;# Gromacs 4.0                         Copyright (c) 1991-2003 
;# David van der Spoel, Erik Lindahl
;#
;# This program is free software; you can redistribute it and/or
;# modify it under the terms of the GNU General Public License
;# as published by the Free Software Foundation; either version 2
;# of the License, or (at your option) any later version.
;#
;# To help us fund GROMACS development, we humbly ask that you cite
;# the research papers on the package. Check out http://www.gromacs.org
;# 
;# And Hey:
;# Gnomes, ROck Monsters And Chili Sauce
;#

;# These files require GNU binutils 2.10 or later, since we
;# use intel syntax for portability, or a recent version 
;# of NASM that understands Extended 3DNow and SSE2 instructions.
;# (NASM is normally only used with MS Visual C++).
;# Since NASM and gnu as disagree on some definitions and use 
;# completely different preprocessing options I have to introduce a
;# trick: NASM uses ';' for comments, while gnu as uses '#' on x86.
;# Gnu as treats ';' as a line break, i.e. ignores it. This is the
;# reason why all comments need both symbols...
;# The source is written for GNU as, with intel syntax. When you use
;# NASM we redefine a couple of things. The false if-statement around 
;# the following code is seen by GNU as, but NASM doesn't see it, so 
;# the code inside is read by NASM but not gcc.

; .if 0    # block below only read by NASM
%define .section	section
%define .long		dd
%define .align		align
%define .globl		global
;# NASM only wants 'dword', not 'dword ptr'.
%define ptr
.equiv          .equiv                  2
   %1 equ %2
%endmacro
; .endif                   # End of NASM-specific block
; .intel_syntax noprefix   # Line only read by gnu as



	
.globl nb_kernel204_ia32_sse2
.globl _nb_kernel204_ia32_sse2
nb_kernel204_ia32_sse2:	
_nb_kernel204_ia32_sse2:	
.equiv          nb204_p_nri,            8
.equiv          nb204_iinr,             12
.equiv          nb204_jindex,           16
.equiv          nb204_jjnr,             20
.equiv          nb204_shift,            24
.equiv          nb204_shiftvec,         28
.equiv          nb204_fshift,           32
.equiv          nb204_gid,              36
.equiv          nb204_pos,              40
.equiv          nb204_faction,          44
.equiv          nb204_charge,           48
.equiv          nb204_p_facel,          52
.equiv          nb204_argkrf,           56
.equiv          nb204_argcrf,           60
.equiv          nb204_Vc,               64
.equiv          nb204_type,             68
.equiv          nb204_p_ntype,          72
.equiv          nb204_vdwparam,         76
.equiv          nb204_Vvdw,             80
.equiv          nb204_p_tabscale,       84
.equiv          nb204_VFtab,            88
.equiv          nb204_invsqrta,         92
.equiv          nb204_dvda,             96
.equiv          nb204_p_gbtabscale,     100
.equiv          nb204_GBtab,            104
.equiv          nb204_p_nthreads,       108
.equiv          nb204_count,            112
.equiv          nb204_mtx,              116
.equiv          nb204_outeriter,        120
.equiv          nb204_inneriter,        124
.equiv          nb204_work,             128
	;# stack offsets for local variables  
	;# bottom of stack is cache-aligned for sse2 use 
.equiv          nb204_ixM,              0
.equiv          nb204_iyM,              16
.equiv          nb204_izM,              32
.equiv          nb204_ixH1,             48
.equiv          nb204_iyH1,             64
.equiv          nb204_izH1,             80
.equiv          nb204_ixH2,             96
.equiv          nb204_iyH2,             112
.equiv          nb204_izH2,             128
.equiv          nb204_jxM,              144
.equiv          nb204_jyM,              160
.equiv          nb204_jzM,              176
.equiv          nb204_jxH1,             192
.equiv          nb204_jyH1,             208
.equiv          nb204_jzH1,             224
.equiv          nb204_jxH2,             240
.equiv          nb204_jyH2,             256
.equiv          nb204_jzH2,             272
.equiv          nb204_dxMM,             288
.equiv          nb204_dyMM,             304
.equiv          nb204_dzMM,             320
.equiv          nb204_dxMH1,            336
.equiv          nb204_dyMH1,            352
.equiv          nb204_dzMH1,            368
.equiv          nb204_dxMH2,            384
.equiv          nb204_dyMH2,            400
.equiv          nb204_dzMH2,            416
.equiv          nb204_dxH1M,            432
.equiv          nb204_dyH1M,            448
.equiv          nb204_dzH1M,            464
.equiv          nb204_dxH1H1,           480
.equiv          nb204_dyH1H1,           496
.equiv          nb204_dzH1H1,           512
.equiv          nb204_dxH1H2,           528
.equiv          nb204_dyH1H2,           544
.equiv          nb204_dzH1H2,           560
.equiv          nb204_dxH2M,            576
.equiv          nb204_dyH2M,            592
.equiv          nb204_dzH2M,            608
.equiv          nb204_dxH2H1,           624
.equiv          nb204_dyH2H1,           640
.equiv          nb204_dzH2H1,           656
.equiv          nb204_dxH2H2,           672
.equiv          nb204_dyH2H2,           688
.equiv          nb204_dzH2H2,           704
.equiv          nb204_qqMM,             720
.equiv          nb204_qqMH,             736
.equiv          nb204_qqHH,             752
.equiv          nb204_vctot,            768
.equiv          nb204_fixM,             784
.equiv          nb204_fiyM,             800
.equiv          nb204_fizM,             816
.equiv          nb204_fixH1,            832
.equiv          nb204_fiyH1,            848
.equiv          nb204_fizH1,            864
.equiv          nb204_fixH2,            880
.equiv          nb204_fiyH2,            896
.equiv          nb204_fizH2,            912
.equiv          nb204_fjxM,             928
.equiv          nb204_fjyM,             944
.equiv          nb204_fjzM,             960
.equiv          nb204_fjxH1,            976
.equiv          nb204_fjyH1,            992
.equiv          nb204_fjzH1,            1008
.equiv          nb204_fjxH2,            1024
.equiv          nb204_fjyH2,            1040
.equiv          nb204_fjzH2,            1056
.equiv          nb204_half,             1072
.equiv          nb204_three,            1088
.equiv          nb204_rsqMM,            1104
.equiv          nb204_rsqMH1,           1120
.equiv          nb204_rsqMH2,           1136
.equiv          nb204_rsqH1M,           1152
.equiv          nb204_rsqH1H1,          1168
.equiv          nb204_rsqH1H2,          1184
.equiv          nb204_rsqH2M,           1200
.equiv          nb204_rsqH2H1,          1216
.equiv          nb204_rsqH2H2,          1232
.equiv          nb204_rinvMM,           1248
.equiv          nb204_rinvMH1,          1264
.equiv          nb204_rinvMH2,          1280
.equiv          nb204_rinvH1M,          1296
.equiv          nb204_rinvH1H1,         1312
.equiv          nb204_rinvH1H2,         1328
.equiv          nb204_rinvH2M,          1344
.equiv          nb204_rinvH2H1,         1360
.equiv          nb204_rinvH2H2,         1376
.equiv          nb204_two,              1392
.equiv          nb204_krf,              1408
.equiv          nb204_crf,              1424
.equiv          nb204_is3,              1440
.equiv          nb204_ii3,              1444
.equiv          nb204_innerjjnr,        1448
.equiv          nb204_innerk,           1452
.equiv          nb204_n,                1456
.equiv          nb204_nn1,              1460
.equiv          nb204_nri,              1464
.equiv          nb204_nouter,           1468
.equiv          nb204_ninner,           1472
.equiv          nb204_salign,           1476
	push ebp
	mov ebp,esp	
    push eax
    push ebx
    push ecx
    push edx
	push esi
	push edi
	sub esp, 1480		;# local stack space 
	mov  eax, esp
	and  eax, 0xf
	sub esp, eax
	mov [esp + nb204_salign], eax

	emms

	;# Move args passed by reference to stack
	mov ecx, [ebp + nb204_p_nri]
	mov ecx, [ecx]
	mov [esp + nb204_nri], ecx

	;# zero iteration counters
	mov eax, 0
	mov [esp + nb204_nouter], eax
	mov [esp + nb204_ninner], eax


	;# create constant floating-point factors on stack
	mov eax, 0x00000000     ;# lower half of double 0.5 IEEE (hex)
	mov ebx, 0x3fe00000
	mov [esp + nb204_half], eax
	mov [esp + nb204_half+4], ebx
	movsd xmm1, [esp + nb204_half]
	shufpd xmm1, xmm1, 0    ;# splat to all elements
	movapd xmm3, xmm1
	addpd  xmm3, xmm3       ;# 1.0
	movapd xmm2, xmm3
	addpd  xmm2, xmm2       ;# 2.0
	addpd  xmm3, xmm2	;# 3.0
	movapd [esp + nb204_half], xmm1
	movapd [esp + nb204_two], xmm2
	movapd [esp + nb204_three], xmm3

	mov esi, [ebp + nb204_argkrf]
	mov edi, [ebp + nb204_argcrf]
	movsd xmm5, [esi]
	movsd xmm6, [edi]
	shufpd xmm5, xmm5, 0
	shufpd xmm6, xmm6, 0
	movapd [esp + nb204_krf], xmm5
	movapd [esp + nb204_crf], xmm6
	
	;# assume we have at least one i particle - start directly 
	mov   ecx, [ebp + nb204_iinr]       ;# ecx = pointer into iinr[] 	
	mov   ebx, [ecx]	    ;# ebx =ii 

	mov   edx, [ebp + nb204_charge]
	movsd xmm3, [edx + ebx*8 + 24]	
	movsd xmm4, xmm3	
	movsd xmm5, [edx + ebx*8 + 8]	
	mov esi, [ebp + nb204_p_facel]
	movsd xmm6, [esi]
	mulsd  xmm3, xmm3
	mulsd  xmm4, xmm5
	mulsd  xmm5, xmm5
	mulsd  xmm3, xmm6
	mulsd  xmm4, xmm6
	mulsd  xmm5, xmm6
	shufpd xmm3, xmm3, 0
	shufpd xmm4, xmm4, 0
	shufpd xmm5, xmm5, 0
	movapd [esp + nb204_qqMM], xmm3
	movapd [esp + nb204_qqMH], xmm4
	movapd [esp + nb204_qqHH], xmm5
	
.nb204_threadloop:
        mov   esi, [ebp + nb204_count]          ;# pointer to sync counter
        mov   eax, [esi]
.nb204_spinlock:
        mov   ebx, eax                          ;# ebx=*count=nn0
        add   ebx, 1                           ;# ebx=nn1=nn0+10
        lock
        cmpxchg [esi], ebx                      ;# write nn1 to *counter,
                                                ;# if it hasnt changed.
                                                ;# or reread *counter to eax.
        pause                                   ;# -> better p4 performance
        jnz .nb204_spinlock

        ;# if(nn1>nri) nn1=nri
        mov ecx, [esp + nb204_nri]
        mov edx, ecx
        sub ecx, ebx
        cmovle ebx, edx                         ;# if(nn1>nri) nn1=nri
        ;# Cleared the spinlock if we got here.
        ;# eax contains nn0, ebx contains nn1.
        mov [esp + nb204_n], eax
        mov [esp + nb204_nn1], ebx
        sub ebx, eax                            ;# calc number of outer lists
	mov esi, eax				;# copy n to esi
        jg  .nb204_outerstart
        jmp .nb204_end

.nb204_outerstart:
	;# ebx contains number of outer iterations
	add ebx, [esp + nb204_nouter]
	mov [esp + nb204_nouter], ebx

.nb204_outer:
	mov   eax, [ebp + nb204_shift]      ;# eax = pointer into shift[] 
	mov   ebx, [eax + esi*4]		;# ebx=shift[n] 
	
	lea   ebx, [ebx + ebx*2]    ;# ebx=3*is 
	mov   [esp + nb204_is3],ebx    	;# store is3 

	mov   eax, [ebp + nb204_shiftvec]   ;# eax = base of shiftvec[] 

	movsd xmm0, [eax + ebx*8]
	movsd xmm1, [eax + ebx*8 + 8]
	movsd xmm2, [eax + ebx*8 + 16] 

	mov   ecx, [ebp + nb204_iinr]       ;# ecx = pointer into iinr[] 	
	mov   ebx, [ecx+esi*4]	    ;# ebx =ii 

	lea   ebx, [ebx + ebx*2]	;# ebx = 3*ii=ii3 
	mov   eax, [ebp + nb204_pos]    ;# eax = base of pos[]  
	mov   [esp + nb204_ii3], ebx	
	
	movapd xmm3, xmm0
	movapd xmm4, xmm1
	movapd xmm5, xmm2
	addsd xmm3, [eax + ebx*8 + 24]
	addsd xmm4, [eax + ebx*8 + 32]
	addsd xmm5, [eax + ebx*8 + 40]		
	shufpd xmm3, xmm3, 0
	shufpd xmm4, xmm4, 0
	shufpd xmm5, xmm5, 0
	movapd [esp + nb204_ixH1], xmm3
	movapd [esp + nb204_iyH1], xmm4
	movapd [esp + nb204_izH1], xmm5

	movsd xmm3, xmm0
	movsd xmm4, xmm1
	movsd xmm5, xmm2
	addsd xmm0, [eax + ebx*8 + 48]
	addsd xmm1, [eax + ebx*8 + 56]
	addsd xmm2, [eax + ebx*8 + 64]		
	addsd xmm3, [eax + ebx*8 + 72]
	addsd xmm4, [eax + ebx*8 + 80]
	addsd xmm5, [eax + ebx*8 + 88]		

	shufpd xmm0, xmm0, 0
	shufpd xmm1, xmm1, 0
	shufpd xmm2, xmm2, 0
	shufpd xmm3, xmm3, 0
	shufpd xmm4, xmm4, 0
	shufpd xmm5, xmm5, 0
	movapd [esp + nb204_ixH2], xmm0
	movapd [esp + nb204_iyH2], xmm1
	movapd [esp + nb204_izH2], xmm2
	movapd [esp + nb204_ixM], xmm3
	movapd [esp + nb204_iyM], xmm4
	movapd [esp + nb204_izM], xmm5

	;# clear vctot and i forces 
	xorpd xmm4, xmm4
	movapd [esp + nb204_vctot], xmm4
	movapd [esp + nb204_fixM], xmm4
	movapd [esp + nb204_fiyM], xmm4
	movapd [esp + nb204_fizM], xmm4
	movapd [esp + nb204_fixH1], xmm4
	movapd [esp + nb204_fiyH1], xmm4
	movapd [esp + nb204_fizH1], xmm4
	movapd [esp + nb204_fixH2], xmm4
	movapd [esp + nb204_fiyH2], xmm4
	movapd [esp + nb204_fizH2], xmm4
	
	mov   eax, [ebp + nb204_jindex]
	mov   ecx, [eax + esi*4]	     ;# jindex[n] 
	mov   edx, [eax + esi*4 + 4]	     ;# jindex[n+1] 
	sub   edx, ecx               ;# number of innerloop atoms 

	mov   esi, [ebp + nb204_pos]
	mov   edi, [ebp + nb204_faction]	
	mov   eax, [ebp + nb204_jjnr]
	shl   ecx, 2
	add   eax, ecx
	mov   [esp + nb204_innerjjnr], eax     ;# pointer to jjnr[nj0] 
	mov   ecx, edx
	sub   edx,  2
	add   ecx, [esp + nb204_ninner]
	mov   [esp + nb204_ninner], ecx
	add   edx, 0
	mov   [esp + nb204_innerk], edx    ;# number of innerloop atoms 
	jge   .nb204_unroll_loop
	jmp   .nb204_checksingle
.nb204_unroll_loop:
	;# twice unrolled innerloop here 
	mov   edx, [esp + nb204_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [edx]	
	mov   ebx, [edx + 4] 
	
	add dword ptr [esp + nb204_innerjjnr],  8	;# advance pointer (unrolled 2) 

	mov esi, [ebp + nb204_pos]       ;# base of pos[] 

	lea   eax, [eax + eax*2]     ;# replace jnr with j3 
	lea   ebx, [ebx + ebx*2]	
	
	;# move j coordinates to local temp variables 
	movlpd xmm2, [esi + eax*8 + 24]
	movlpd xmm3, [esi + eax*8 + 32]
	movlpd xmm4, [esi + eax*8 + 40]
	movlpd xmm5, [esi + eax*8 + 48]
	movlpd xmm6, [esi + eax*8 + 56]
	movlpd xmm7, [esi + eax*8 + 64]
	movhpd xmm2, [esi + ebx*8 + 24]
	movhpd xmm3, [esi + ebx*8 + 32]
	movhpd xmm4, [esi + ebx*8 + 40]
	movhpd xmm5, [esi + ebx*8 + 48]
	movhpd xmm6, [esi + ebx*8 + 56]
	movhpd xmm7, [esi + ebx*8 + 64]
	movapd 	[esp + nb204_jxH1], xmm2
	movapd 	[esp + nb204_jyH1], xmm3
	movapd 	[esp + nb204_jzH1], xmm4
	movapd 	[esp + nb204_jxH2], xmm5
	movapd 	[esp + nb204_jyH2], xmm6
	movapd 	[esp + nb204_jzH2], xmm7
	movlpd xmm2, [esi + eax*8 + 72]
	movlpd xmm3, [esi + eax*8 + 80]
	movlpd xmm4, [esi + eax*8 + 88]
	movhpd xmm2, [esi + ebx*8 + 72]
	movhpd xmm3, [esi + ebx*8 + 80]
	movhpd xmm4, [esi + ebx*8 + 88]
	movapd 	[esp + nb204_jxM], xmm2
	movapd 	[esp + nb204_jyM], xmm3
	movapd 	[esp + nb204_jzM], xmm4
	
	movapd xmm0, [esp + nb204_ixM]
	movapd xmm1, [esp + nb204_iyM]