I thought I'd post this here as well, might be of use to someone. It does conversions in 20 to 50 cycles for 1 to 10 digit integers.
I'll post the FastProc version which should convert easily to other assemblers.
itoaFastProc.inc
#Include This Once
' Usage
' Local buffer as StringZ * 12
' itoa 1234567890, ByRef buffer
AsmData itoa_cvt
DW &h3030, &h3130, &h3230, &h3330, &h3430, &h3530, &h3630, &h3730, &h3830, &h3930
DW &h3031, &h3131, &h3231, &h3331, &h3431, &h3531, &h3631, &h3731, &h3831, &h3931
DW &h3032, &h3132, &h3232, &h3332, &h3432, &h3532, &h3632, &h3732, &h3832, &h3932
DW &h3033, &h3133, &h3233, &h3333, &h3433, &h3533, &h3633, &h3733, &h3833, &h3933
DW &h3034, &h3134, &h3234, &h3334, &h3434, &h3534, &h3634, &h3734, &h3834, &h3934
DW &h3035, &h3135, &h3235, &h3335, &h3435, &h3535, &h3635, &h3735, &h3835, &h3935
DW &h3036, &h3136, &h3236, &h3336, &h3436, &h3536, &h3636, &h3736, &h3836, &h3936
DW &h3037, &h3137, &h3237, &h3337, &h3437, &h3537, &h3637, &h3737, &h3837, &h3937
DW &h3038, &h3138, &h3238, &h3338, &h3438, &h3538, &h3638, &h3738, &h3838, &h3938
DW &h3039, &h3139, &h3239, &h3339, &h3439, &h3539, &h3639, &h3739, &h3839, &h3939
End AsmData
FastProc itoa( ByVal value As Long, ByVal b As Long )
#Register None
Prefix "!"
mov ebx, esi ; eax = value
mov esi, edi ; esi = VarPtr( b )
lea edi, itoa_cvt ; edi = cvt
xor edx, edx ; edx = 0 (initial 0)
cmp ebx, edx ; if value
jge itoa_init ; >= 0 Then itoa_init
mov byte ptr[esi], 45 ; @esi = '-'
inc esi ; incr esi
neg eax ; eax = -eax
itoa_init:
cmp eax, 10000
jl itoa_digit8
cmp eax, 10000000
jl itoa_digit5
cmp eax, 100000000
jl itoa_digit4
cmp eax, 1000000000
jl itoa_digit3
;===============================
; 10 digits
mov byte ptr[esi+10], 0
jmp itoa_10
itoa_digit3:
;===============================
; 9 digits
mov eax, 1717986919 ; 66666667H
mov byte ptr [esi+9], 0
imul ebx
sar edx, 2
mov ebx, edx
shr ebx, 31 ; 0000001fH
add ebx, edx
mov al, bl
shl al, 2
lea ecx, DWORD PTR [eax+ebx]
mov eax, value
add cl, cl
sub al, cl
add al, 48 ; 00000030H
mov byte ptr [esi+8], al
jmp itoa_8
itoa_digit4:
;===============================
; 8 digits
mov byte ptr[esi+8], 0
jmp itoa_8
itoa_digit5:
cmp eax, 100000
jl itoa_digit7
cmp eax, 1000000
jl itoa_6
;===============================
; 7 digits
mov eax, 1717986919 ; 66666667H
mov byte ptr [esi+7], 0
imul ebx
sar edx, 2
mov ebx, edx
shr ebx, 31 ; 0000001fH
add ebx, edx
mov al, bl
shl al, 2
lea ecx, DWORD PTR [eax+ebx]
mov eax, value
add cl, cl
sub al, cl
add al, 48 ; 00000030H
mov byte ptr [esi+6], al
jmp itoa_6
itoa_digit6:
;===============================
; 6 digits
mov byte ptr[esi+6], 0
jmp itoa_6
itoa_digit7:
;===============================
; 5 digits
mov eax, 1717986919 ; 66666667H
mov byte ptr [esi+5], 0
imul ebx
sar edx, 2
mov ebx, edx
shr ebx, 31 ; 0000001fH
add ebx, edx
mov al, bl
shl al, 2
lea ecx, DWORD PTR [eax+ebx]
mov eax, value
add cl, cl
sub al, cl
add al, 48 ; 00000030H
mov byte ptr [esi+4], al
jmp itoa_4
itoa_digit8:
cmp eax, 100
jl itoa_digit10
cmp eax, 1000
jl itoa_digit9
;===============================
; 4 digits
mov byte ptr[esi+4], 0
jmp itoa_4
itoa_digit9:
;===============================
; 3 digits
mov eax, 1717986919 ; 66666667H
mov byte ptr [esi+3], 0
imul ebx
sar edx, 2
mov ebx, edx
shr ebx, 31 ; 0000001fH
add ebx, edx
mov al, bl
shl al, 2
lea ecx, DWORD PTR [eax+ebx]
mov eax, value
add cl, cl
sub al, cl
add al, 48 ; 00000030H
mov byte ptr [esi+2], al
jmp itoa_2
itoa_digit10:
;===============================
; 2 digits
cmp eax, 10
jl itoa_digit11
mov byte ptr[esi+2], 0
jmp itoa_2
itoa_digit11:
;===============================
; 1 digits
mov eax, 1717986919 ; 66666667H
mov byte ptr [esi+1], 0
imul ebx
sar edx, 2
mov ebx, edx
shr ebx, 31 ; 0000001fH
add ebx, edx
mov al, bl
shl al, 2
lea ecx, DWORD PTR [eax+ebx]
mov eax, value
add cl, cl
sub al, cl
add al, 48 ; 00000030H
mov byte ptr [esi], al
jmp itoa_0
; Convert 10 digits
itoa_10:
mov eax, 1374389535 ; 51eb851fH
imul ebx
sar edx, 5
mov ecx, edx
shr ecx, 31 ; 0000001fH
add ecx, edx
imul eax, ecx, 100
sub ebx, eax ; ebx = value mod 100
mov ax, word ptr [edi+ebx*2]
mov ebx, ecx ; ebx = value \ 100
mov word ptr [esi+8], ax
; Convert 8 digits
itoa_8:
mov eax, 1374389535 ; 51eb851fH
imul ebx
sar edx, 5
mov ecx, edx
shr ecx, 31 ; 0000001fH
add ecx, edx
imul eax, ecx, 100
sub ebx, eax ; ebx = value mod 100
mov ax, word ptr [edi+ebx*2]
mov ebx, ecx ; ebx = value \ 100
mov word ptr [esi+6], ax
; Convert 6 digits
itoa_6:
mov eax, 1374389535 ; 51eb851fH
imul ebx
sar edx, 5
mov ecx, edx
shr ecx, 31 ; 0000001fH
add ecx, edx
imul eax, ecx, 100
sub ebx, eax ; ebx = value mod 100
mov ax, word ptr [edi+ebx*2]
mov ebx, ecx ; ebx = value \ 100
mov word ptr [esi+4], ax
; Convert 4 digits
itoa_4:
mov eax, 1374389535 ; 51eb851fH
imul ebx
sar edx, 5
mov ecx, edx
shr ecx, 31 ; 0000001fH
add ecx, edx
imul eax, ecx, 100
sub ebx, eax ; ebx = value mod 100
mov ax, word ptr [edi+ebx*2]
mov ebx, ecx ; ebx = value \ 100
mov word ptr [esi+2], ax
; Convert 2 digits
itoa_2:
mov eax, 1374389535 ; 51eb851fH
imul ebx
sar edx, 5
mov ecx, edx
shr ecx, 31 ; 0000001fH
add ecx, edx
imul eax, ecx, 100
sub ebx, eax ; ebx = value mod 100
mov ax, word ptr [edi+ebx*2]
mov ebx, ecx ; ebx = value \ 100
mov word ptr [esi], ax
itoa_0:
End Prefix
End FastProc
MakeCvt.bas
#Compile Exe
#Dim All
Function PBMain () As Long
MakeCvt
End Function
Sub MakeCvt()
Local i As Long
Local msg As String
msg = "AsmData cvt"
For i=0 To 99
If i Mod 10 = 0 Then msg += $CrLf + " DW " Else msg += ", "
msg += "&h" + Right$("0000" + Hex$( (i\10 + 48) + ((i Mod 10)+48 )*256 ), 4)
Next
msg += $CrLf + "End AsmData" + $CrLf
Clipboard Reset
Clipboard Set Text msg
End Sub
test_itoaFastProc.bas
#Compile Exe
#Dim All
#Optimize Speed
#Include "itoaFastProc.inc"
%COUNT = 100000000
Function PBMain () As Long
Local i As Long
Local v As StringZ * 12
Local q As Quad
Local s, e As Single
s = Timer
Tix q
For i=1 To %COUNT
itoa 1234567890, ByRef v
Next
Tix End q
e = Timer
? "FastProc atoi: " + Format$( q/%COUNT, "#,##0.0" ) + " tix : " + Format$( %COUNT/(e-s), "#,##0" )+ "/sec : " + v
End Function
Thanks Larry, it looks good.
Just an small update on Larry´s algo. Fixed negative values to show the correct values, preserved the registers, eax now returns the lenght of the string
He was right, the function takes about 40 clock cycles on my I7 (including the modifications). On his original version (with the negative errors and no preservation of registers) it varies from 7 to 25 clock cycles.
[itoa_cvt: W$ 03030, 03130, 03230, 03330, 03430, 03530, 03630, 03730, 03830, 03930
W$ 03031, 03131, 03231, 03331, 03431, 03531, 03631, 03731, 03831, 03931
W$ 03032, 03132, 03232, 03332, 03432, 03532, 03632, 03732, 03832, 03932
W$ 03033, 03133, 03233, 03333, 03433, 03533, 03633, 03733, 03833, 03933
W$ 03034, 03134, 03234, 03334, 03434, 03534, 03634, 03734, 03834, 03934
W$ 03035, 03135, 03235, 03335, 03435, 03535, 03635, 03735, 03835, 03935
W$ 03036, 03136, 03236, 03336, 03436, 03536, 03636, 03736, 03836, 03936
W$ 03037, 03137, 03237, 03337, 03437, 03537, 03637, 03737, 03837, 03937
W$ 03038, 03138, 03238, 03338, 03438, 03538, 03638, 03738, 03838, 03938
W$ 03039, 03139, 03239, 03339, 03439, 03539, 03639, 03739, 03839, 03939]
Proc itoa:
Arguments @value, @bValue
Local @Return, @Sign
Uses ebx, ecx, edx, edi, esi
mov eax D@Value
mov edi D@bValue
mov D@Sign 0
mov ebx, eax;esi ; eax = value
mov esi, edi ; esi = VarPtr( b )
mov edi itoa_cvt ; edi = cvt
xor edx, edx ; edx = 0 (initial 0)
cmp ebx, edx ; if value
jge @itoa_init ; >= 0 Then itoa_init
mov D@Sign 1
mov B$esi '-';45 ; @esi = '-'
inc esi ; incr esi
neg eax ; eax = -eax
neg ebx
mov D@Value eax
@itoa_init:
cmp eax, 10000
jl @itoa_digit8
cmp eax, 10000000
jl @itoa_digit5
cmp eax, 100000000
jl @itoa_digit4
cmp eax, 1000000000
jl @itoa_digit3
mov D@Return 10
;===============================
; 10 digits
mov B$esi+10, 0
jmp @itoa_10
@itoa_digit3:
mov D@Return 9
;===============================
; 9 digits
mov eax, 1717986919 ; 66666667H
mov B$esi+9, 0
imul ebx
sar edx, 2
mov ebx, edx
shr ebx, 31 ; 0000001fH
add ebx, edx
mov al, bl
shl al, 2
lea ecx, D$eax+ebx
mov eax D@value
add cl, cl
sub al, cl
add al, 48 ; 00000030H
mov B$esi+8, al
jmp @itoa_8
@itoa_digit4:
mov D@Return 8
;===============================
; 8 digits
mov B$esi+8, 0
jmp @itoa_8
@itoa_digit5:
cmp eax, 100000
jl @itoa_digit7
mov D@Return 6
cmp eax, 1000000
jl @itoa_6
mov D@Return 7
;===============================
; 7 digits
mov eax, 1717986919 ; 66666667H
mov B$esi+7, 0
imul ebx
sar edx, 2
mov ebx, edx
shr ebx, 31 ; 0000001fH
add ebx, edx
mov al, bl
shl al, 2
lea ecx, D$eax+ebx
mov eax, D@value
add cl, cl
sub al, cl
add al, 48 ; 00000030H
mov B$esi+6, al
jmp @itoa_6
@itoa_digit6:
;===============================
; 6 digits
mov B$esi+6, 0
jmp @itoa_6
@itoa_digit7:
;===============================
; 5 digits
mov D@Return 5
mov eax, 1717986919 ; 66666667H
mov B$esi+5, 0
imul ebx
sar edx, 2
mov ebx, edx
shr ebx, 31 ; 0000001fH
add ebx, edx
mov al, bl
shl al, 2
lea ecx, D$eax+ebx
mov eax, D@value
add cl, cl
sub al, cl
add al, 48 ; 00000030H
mov B$esi+4, al
jmp @itoa_4
@itoa_digit8:
cmp eax, 100
jl @itoa_digit10
cmp eax, 1000
jl @itoa_digit9
mov D@Return 4
;===============================
; 4 digits
mov B$esi+4, 0
jmp @itoa_4
@itoa_digit9:
mov D@Return 3
;===============================
; 3 digits
mov eax, 1717986919 ; 66666667H
mov B$esi+3, 0
imul ebx
sar edx, 2
mov ebx, edx
shr ebx, 31 ; 0000001fH
add ebx, edx
mov al, bl
shl al, 2
lea ecx, D$eax+ebx
mov eax, D@value
add cl, cl
sub al, cl
add al, 48 ; 00000030H
mov B$esi+2, al
jmp @itoa_2
@itoa_digit10:
mov D@Return 2
;===============================
; 2 digits
cmp eax, 10
jl @itoa_digit11
mov B$esi+2, 0
jmp @itoa_2
@itoa_digit11:
mov D@Return 1
;===============================
; 1 digits
mov eax, 1717986919 ; 66666667H
mov B$esi+1, 0
imul ebx
sar edx, 2
mov ebx, edx
shr ebx, 31 ; 0000001fH
add ebx, edx
mov al, bl
shl al, 2
lea ecx, D$eax+ebx
mov eax, D@value
add cl, cl
sub al, cl
add al, 48 ; 00000030H
mov B$esi, al
jmp @itoa_0
; Convert 10 digits
@itoa_10:
mov eax, 1374389535 ; 51eb851fH
imul ebx; 1374389535
sar edx, 5
mov ecx, edx
shr ecx, 31 ; 0000001fH
add ecx, edx
imul eax, ecx, 100
sub ebx, eax ; ebx = value mod 100
mov ax, W$edi+ebx*2
mov ebx, ecx ; ebx = value \ 100
mov W$esi+8, ax
; Convert 8 digits
@itoa_8:
mov eax, 1374389535 ; 51eb851fH
imul ebx
sar edx, 5
mov ecx, edx
shr ecx, 31 ; 0000001fH
add ecx, edx
imul eax, ecx, 100
sub ebx, eax ; ebx = value mod 100
mov ax, W$edi+ebx*2
mov ebx, ecx ; ebx = value \ 100
mov W$esi+6, ax
; Convert 6 digits
@itoa_6:
mov eax, 1374389535 ; 51eb851fH
imul ebx
sar edx, 5
mov ecx, edx
shr ecx, 31 ; 0000001fH
add ecx, edx
imul eax, ecx, 100
sub ebx, eax ; ebx = value mod 100
mov ax, W$edi+ebx*2
mov ebx, ecx ; ebx = value \ 100
mov W$esi+4, ax
; Convert 4 digits
@itoa_4:
mov eax, 1374389535 ; 51eb851fH
imul ebx
sar edx, 5
mov ecx, edx
shr ecx, 31 ; 0000001fH
add ecx, edx
imul eax, ecx, 100
sub ebx, eax ; ebx = value mod 100
mov ax, W$edi+ebx*2
mov ebx, ecx ; ebx = value \ 100
mov W$esi+2, ax
; Convert 2 digits
@itoa_2:
mov eax, 1374389535 ; 51eb851fH
imul ebx
sar edx, 5
mov ecx, edx
shr ecx, 31 ; 0000001fH
add ecx, edx
imul eax, ecx, 100
sub ebx, eax ; ebx = value mod 100
mov ax, W$edi+ebx*2
mov ebx, ecx ; ebx = value \ 100
mov W$esi, ax
@itoa_0:
mov eax D@Return
add eax D@Sign
EndP
Steve, i believe this can be optimized a bit more. Perhaps, using less registers instead of all of them. Or changing those cmp to test instructions.