New Getline in assembly
in [ASM]
|
Prev: popen
Next: cpu type idea
From: rio on 12 Jul 2008 06:39 Getline() function is born today, what do you say? not have debug it GetLine_m(b, i, pf) in b there is a pointer of a pointer of one array i there is its len pf is a pointer to file return 1 error of parameter CF=1 2 error of memory 3 error memory is not of malloc -1 EOF ; ; b i j ; u32 GetLine_m(u8** s, u32* len, u32* pf) ; (0k,4j,8i,12b,16ra,20P_s,24P_len,28P_pf) GetLine_m: push ebx push esi push edi push ebp mov eax, 1 mov ebx, [esp+20] mov esi, [esp+24] mov edi, [esp+28] cmp ebx, 0 je .e1 cmp esi, 0 je .e1 cmp edi, 0 jne .1 ..e1: ; err di parametro stc jmp .f ..e2: ; err di memoria mov eax, 2 jmp short .e1 ..e3: ; err il vettore non di malloc mov eax, 3 jmp short .e1 ..e4: ; incontrato EOF clc jmp .f ..1: mov eax, dword[ebx] cmp eax, 0 jne .2 cmp dword[esi], 0 jne .e1 push 128 call Malloc_m cmp eax, 0 je .e2 mov dword[ebx], eax mov ebp, 128 jmp short .3 ..2: push eax call GetSizeFast jc .e3 mov ebp, eax ..3: xor edx, edx mov ebx, [ebx] mov byte[ebx], 0 jmp short .6 ..4: mov ecx, edx shl ecx, 1 mov ebp, ecx mov esi, eax push edx push ecx push ebx call Realloc_m pop edx cmp eax, 0 jne .5 dec edx mov byte[ebx+edx], 0 mov esi, [esp+24] mov [esi], edx jmp short .e2 ..5: mov ebx, eax mov ecx, [esp+20] mov eax, esi mov [ecx], ebx ..6: ; Fgetc preserve all but eax and CF push edi call Fgetc_m cmp eax, -1 je .e4 mov [ebx+edx], al inc edx dec ebp jz .4 cmp al, 10 jne .4 mov byte[ebx+edx], 0 mov esi, [esp+24] mov [esi], edx xor eax, eax clc ..f: pop ebp pop edi pop esi pop ebx ret 12 ; ritorna il numero di char riservati al programma ; puntati dall'argomento ap che deve essere un ; puntatore ritornato da malloc-realloc ; 0ra, 4P_ap GetSizeFast: mov eax, [esp+4] cmp eax, 0 je .e cmp eax, [eax-4] je .1 ..e: xor eax, eax stc jmp short .f ..ee: xor eax, eax clc jmp short .f ..1: mov eax, [eax-8] test eax, 0xF0000000 jnz .e shl eax, 4 ; 12+15=27 sub eax, 27 cmp eax, -11 jl .e ; 16-12=4 4-15=-11=16-27 cmp eax, 0 jl .ee clc ..f: ret 4 --------------------------------- /* /* b i j /* u32 GetLine_m(u8** s, u32* len, u32* pf) /* (0k,4j,8i,12b,16ra,20P_s,24P_len,28P_pf) GetLine_m: <b,i,j,k a=1|b=[s+20]|i=[s+24]|j=[s+28] b==0#.e1|i==0#.e1|j#.1 ..e1: stc| ##.f /* err di parametro ..e2: a=2| #.e1 /* err di memoria ..e3: a=3| #.e1 /* err il vettore non di malloc ..e4: clc| ##.f /* incontrato EOF ..1: a=D[b] a#.2|D[i]#.e1|Malloc_m(128)|a!#.e2 D[b]=a|k=128| #.3 ..2: GetSizeFast(a) | jc .e3 | k=a ..3: r^=r | b=[b] | B[b]=0 | #.6 ..4: c=r|c<<=1|k=c|i=a <r |Realloc_m(b, c)| >r a==0!#.5|--r|B[b+r]=0|i=[s+24]| *i=r| #.e2 ..5: b=a|c=[s+20]|a=i|[c]=b ..6: Fgetc_m(j)| a==-1#.e4 /* Fgetc preserve all but eax and CF [b+r]=al| ++r| --k jz .4| al!=10#.4 B[b+r]=0 i=[s+24]| *i=r|a^=a|clc ..f: >b,i,j,k ret 12 /* ritorna il numero di char riservati al programma /* puntati dall'argomento ap che deve essere un /* puntatore ritornato da malloc-realloc /* 0ra, 4P_ap GetSizeFast: a=[s+4]|a==0#.e|a==[a-4]#.1 ..e: a^=a |stc| #.f ..ee: a^=a |clc| #.f ..1: a=[a-8]|a&0xF0000000#.e a<<=4 |a-=27 /* 12+15=27 a< -11?#.e|a<0?#.ee /* 16-12=4 4-15=-11=16-27 clc ..f: ret 4
From: Wolfgang Kern on 12 Jul 2008 11:12 "rio" posted: > Getline() function is born today, what do you say? not have debug it Haven't checked in detail if it may work at all, and beside that I think you could save on many branches by reordering your code, I marked a few lines (*). Not sure for it if your array contains variable sized items. __ wolfgang _____________________ > GetLine_m(b, i, pf) > in b there is a pointer of a pointer of one array > i there is its len > pf is a pointer to file > return 1 error of parameter CF=1 > 2 error of memory > 3 error memory is not of malloc > -1 EOF > ; > ; b i j > ; u32 GetLine_m(u8** s, u32* len, u32* pf) > ; (0k,4j,8i,12b,16ra,20P_s,24P_len,28P_pf) * you sure got a reason for using C-styled 'procedures' * I'd had this three arguments in registers. > GetLine_m: > push ebx > push esi > push edi > push ebp > mov eax, 1 > mov ebx, [esp+20] > mov esi, [esp+24] > mov edi, [esp+28] > cmp ebx, 0 * why not OR ebx,ebx ? > je .e1 > cmp esi, 0 * > je .e1 > cmp edi, 0 * > jne .1 > .e1: ; err di parametro > stc > jmp .f > .e2: ; err di memoria > mov eax, 2 > jmp short .e1 > .e3: ; err il vettore non di malloc > mov eax, 3 > jmp short .e1 > .e4: ; incontrato EOF > clc > jmp .f > .1: > mov eax, dword[ebx] > cmp eax, 0 * > jne .2 > cmp dword[esi], 0 > jne .e1 > push 128 > call Malloc_m > cmp eax, 0 * > je .e2 > mov dword[ebx], eax > mov ebp, 128 > jmp short .3 > .2: > push eax > call GetSizeFast > jc .e3 > mov ebp, eax > .3: > xor edx, edx > mov ebx, [ebx] > mov byte[ebx], 0 * MOV [ebx],DL ? > jmp short .6 > .4: > mov ecx, edx > shl ecx, 1 > mov ebp, ecx > mov esi, eax > push edx > push ecx > push ebx > call Realloc_m > pop edx > cmp eax, 0 > jne .5 > dec edx > mov byte[ebx+edx], 0 * MOV [ebx+edx],AL ? > mov esi, [esp+24] > mov [esi], edx > jmp short .e2 > .5: > mov ebx, eax > mov ecx, [esp+20] > mov eax, esi > mov [ecx], ebx > .6: ; Fgetc preserve all but eax and CF > push edi > call Fgetc_m > cmp eax, -1 > je .e4 > mov [ebx+edx], al > inc edx > dec ebp > jz .4 > cmp al, 10 > jne .4 > mov byte[ebx+edx], 0 > mov esi, [esp+24] > mov [esi], edx > xor eax, eax > clc * ;XOR clears the Carry anyway > .f: > pop ebp > pop edi > pop esi > pop ebx > ret 12 > > > ; ritorna il numero di char riservati al programma > ; puntati dall'argomento ap che deve essere un > ; puntatore ritornato da malloc-realloc > ; 0ra, 4P_ap > GetSizeFast: > mov eax, [esp+4] > cmp eax, 0 * > je .e > cmp eax, [eax-4] > je .1 > .e: > xor eax, eax > stc > jmp short .f > .ee: > xor eax, eax > clc * ;XOR ... > jmp short .f > .1: > mov eax, [eax-8] > test eax, 0xF0000000 > jnz .e > shl eax, 4 ; 12+15=27 > sub eax, 27 > cmp eax, -11 > jl .e ; 16-12=4 4-15=-11=16-27 > cmp eax, 0 > jl .ee > clc > .f: > ret 4 > > > --------------------------------- * [snipped the almost unreadable 'to A or not to bee' source] :)
From: rio on 13 Jul 2008 03:36 "Wolfgang Kern" <nowhere(a)never.at> ha scritto nel messaggio news:g5ai74$g3s$1(a)newsreader2.utanet.at... > > "rio" posted: > >> Getline() function is born today, what do you say? not have debug it > > Haven't checked in detail if it may work at all, and beside that > I think you could save on many branches by reordering your code, > I marked a few lines (*). the branches (==jumps) have to be, because there are many cases. i don't understand, how this can save from a branch: cmp ebx, 0 * why not OR ebx,ebx ? je .e1 why is it better "OR ebx,ebx|jz .e1" on "cmp ebx, 0|je .e1"?
From: rio on 13 Jul 2008 03:37 "Wolfgang Kern" <nowhere(a)never.at> ha scritto nel messaggio news:g5ai74$g3s$1(a)newsreader2.utanet.at... > >> xor eax, eax >> clc > * ;XOR ... Would you say that "xor eax, eax" clear the carry flag? Saluti
From: Wolfgang Kern on 13 Jul 2008 06:04
"rio" asked: >>> Getline() function is born today, what do you say? not have debug it >> Haven't checked in detail if it may work at all, and beside that >> I think you could save on many branches by reordering your code, >> I marked a few lines (*). > the branches (==jumps) have to be, because there are many cases. > i don't understand, how this can save from a branch: > cmp ebx, 0 > * why not OR ebx,ebx ? > je .e1 Sure, not all jumps can be avoided, but by using CMOVcc, SETcc, ORing bits, ADC,0 and shift you can get rid of many spaghetti- code and gain speed and often also size. an not optimised example: or ebx,ebx setz ah or al,ah add al,al ;shift left 1 or esi,esi setz ah or al,ah add al,al ;shif left 1 or edi,edi setz ah or al,ah movzx eax,al ;eax = 0 if none of the three is zero 000 * = 1 if edi=0 001 * = 2 if esi=0 010 * = 3 if edi=esi=0 011 = 4 if ebx=0 100 * = 5 if ebx=edi=0 101 = 6 if ebx=esi=0 110 = 7 if all three are zero 111 If you only need the cases 0,1,2,4 then just look at the bits*. and if you have (like me) error text 2**n fixsized and aligned you can use this eax to direct point to an error message string: ;shl eax,... ;if more than 8 bytes lea edx,[err_txt_array+eax*8] ;edx points to the error text yet but if you just want to detect if any of the three is zero: MOV eax,ebx OR eax,esi OR eax,edi CMOVz eax,[err] CMOVnz eax,[ok] > why is it better "OR ebx,ebx|jz .e1" on > "cmp ebx, 0|je .e1"? It is shorter, two bytes 09 DB OR ebx,ebx instead of three 83 FB 00 CMP ebx,+0 or six bytes 81 FB 00 00 00 00 CMP ebx,0 and it works identical if just exact zero is of interest. __ wolfgang |