blob: 4446c1c2cee41495937d0ff66aa05c8ee8161f2c [file] [log] [blame]
;
; x86 format converters for HERMES
; Some routines Copyright (c) 1998 Christian Nentwich (brn@eleet.mcb.at)
; This source code is licensed under the GNU LGPL
;
; Please refer to the file COPYING.LIB contained in the distribution for
; licensing conditions
;
; Most routines are (c) Glenn Fiedler (ptc@gaffer.org), used with permission
;
BITS 32
%include "common.inc"
SDL_FUNC _ConvertX86p32_32BGR888
SDL_FUNC _ConvertX86p32_32RGBA888
SDL_FUNC _ConvertX86p32_32BGRA888
SDL_FUNC _ConvertX86p32_24RGB888
SDL_FUNC _ConvertX86p32_24BGR888
SDL_FUNC _ConvertX86p32_16RGB565
SDL_FUNC _ConvertX86p32_16BGR565
SDL_FUNC _ConvertX86p32_16RGB555
SDL_FUNC _ConvertX86p32_16BGR555
SDL_FUNC _ConvertX86p32_8RGB332
SECTION .text
;; _Convert_*
;; Paramters:
;; ESI = source
;; EDI = dest
;; ECX = amount (NOT 0!!! (the _ConvertX86 routine checks for that though))
;; Destroys:
;; EAX, EBX, EDX
_ConvertX86p32_32BGR888:
; check short
cmp ecx,BYTE 32
ja .L3
.L1 ; short loop
mov edx,[esi]
bswap edx
ror edx,8
mov [edi],edx
add esi,BYTE 4
add edi,BYTE 4
dec ecx
jnz .L1
.L2
retn
.L3 ; save ebp
push ebp
; unroll four times
mov ebp,ecx
shr ebp,2
; save count
push ecx
.L4 mov eax,[esi]
mov ebx,[esi+4]
bswap eax
bswap ebx
ror eax,8
mov ecx,[esi+8]
ror ebx,8
mov edx,[esi+12]
bswap ecx
bswap edx
ror ecx,8
mov [edi+0],eax
ror edx,8
mov [edi+4],ebx
mov [edi+8],ecx
mov [edi+12],edx
add esi,BYTE 16
add edi,BYTE 16
dec ebp
jnz .L4
; check tail
pop ecx
and ecx,BYTE 11b
jz .L6
.L5 ; tail loop
mov edx,[esi]
bswap edx
ror edx,8
mov [edi],edx
add esi,BYTE 4
add edi,BYTE 4
dec ecx
jnz .L5
.L6 pop ebp
retn
_ConvertX86p32_32RGBA888:
; check short
cmp ecx,BYTE 32
ja .L3
.L1 ; short loop
mov edx,[esi]
rol edx,8
mov [edi],edx
add esi,BYTE 4
add edi,BYTE 4
dec ecx
jnz .L1
.L2
retn
.L3 ; save ebp
push ebp
; unroll four times
mov ebp,ecx
shr ebp,2
; save count
push ecx
.L4 mov eax,[esi]
mov ebx,[esi+4]
rol eax,8
mov ecx,[esi+8]
rol ebx,8
mov edx,[esi+12]
rol ecx,8
mov [edi+0],eax
rol edx,8
mov [edi+4],ebx
mov [edi+8],ecx
mov [edi+12],edx
add esi,BYTE 16
add edi,BYTE 16
dec ebp
jnz .L4
; check tail
pop ecx
and ecx,BYTE 11b
jz .L6
.L5 ; tail loop
mov edx,[esi]
rol edx,8
mov [edi],edx
add esi,BYTE 4
add edi,BYTE 4
dec ecx
jnz .L5
.L6 pop ebp
retn
_ConvertX86p32_32BGRA888:
; check short
cmp ecx,BYTE 32
ja .L3
.L1 ; short loop
mov edx,[esi]
bswap edx
mov [edi],edx
add esi,BYTE 4
add edi,BYTE 4
dec ecx
jnz .L1
.L2
retn
.L3 ; save ebp
push ebp
; unroll four times
mov ebp,ecx
shr ebp,2
; save count
push ecx
.L4 mov eax,[esi]
mov ebx,[esi+4]
mov ecx,[esi+8]
mov edx,[esi+12]
bswap eax
bswap ebx
bswap ecx
bswap edx
mov [edi+0],eax
mov [edi+4],ebx
mov [edi+8],ecx
mov [edi+12],edx
add esi,BYTE 16
add edi,BYTE 16
dec ebp
jnz .L4
; check tail
pop ecx
and ecx,BYTE 11b
jz .L6
.L5 ; tail loop
mov edx,[esi]
bswap edx
mov [edi],edx
add esi,BYTE 4
add edi,BYTE 4
dec ecx
jnz .L5
.L6 pop ebp
retn
;; 32 bit RGB 888 to 24 BIT RGB 888
_ConvertX86p32_24RGB888:
; check short
cmp ecx,BYTE 32
ja .L3
.L1 ; short loop
mov al,[esi]
mov bl,[esi+1]
mov dl,[esi+2]
mov [edi],al
mov [edi+1],bl
mov [edi+2],dl
add esi,BYTE 4
add edi,BYTE 3
dec ecx
jnz .L1
.L2
retn
.L3 ; head
mov edx,edi
and edx,BYTE 11b
jz .L4
mov al,[esi]
mov bl,[esi+1]
mov dl,[esi+2]
mov [edi],al
mov [edi+1],bl
mov [edi+2],dl
add esi,BYTE 4
add edi,BYTE 3
dec ecx
jmp SHORT .L3
.L4 ; unroll 4 times
push ebp
mov ebp,ecx
shr ebp,2
; save count
push ecx
.L5 mov eax,[esi] ; first dword eax = [A][R][G][B]
mov ebx,[esi+4] ; second dword ebx = [a][r][g][b]
shl eax,8 ; eax = [R][G][B][.]
mov ecx,[esi+12] ; third dword ecx = [a][r][g][b]
shl ebx,8 ; ebx = [r][g][b][.]
mov al,[esi+4] ; eax = [R][G][B][b]
ror eax,8 ; eax = [b][R][G][B] (done)
mov bh,[esi+8+1] ; ebx = [r][g][G][.]
mov [edi],eax
add edi,BYTE 3*4
shl ecx,8 ; ecx = [r][g][b][.]
mov bl,[esi+8+0] ; ebx = [r][g][G][B]
rol ebx,16 ; ebx = [G][B][r][g] (done)
mov cl,[esi+8+2] ; ecx = [r][g][b][R] (done)
mov [edi+4-3*4],ebx
add esi,BYTE 4*4
mov [edi+8-3*4],ecx
dec ebp
jnz .L5
; check tail
pop ecx
and ecx,BYTE 11b
jz .L7
.L6 ; tail loop
mov al,[esi]
mov bl,[esi+1]
mov dl,[esi+2]
mov [edi],al
mov [edi+1],bl
mov [edi+2],dl
add esi,BYTE 4
add edi,BYTE 3
dec ecx
jnz .L6
.L7 pop ebp
retn
;; 32 bit RGB 888 to 24 bit BGR 888
_ConvertX86p32_24BGR888:
; check short
cmp ecx,BYTE 32
ja .L3
.L1 ; short loop
mov dl,[esi]
mov bl,[esi+1]
mov al,[esi+2]
mov [edi],al
mov [edi+1],bl
mov [edi+2],dl
add esi,BYTE 4
add edi,BYTE 3
dec ecx
jnz .L1
.L2
retn
.L3 ; head
mov edx,edi
and edx,BYTE 11b
jz .L4
mov dl,[esi]
mov bl,[esi+1]
mov al,[esi+2]
mov [edi],al
mov [edi+1],bl
mov [edi+2],dl
add esi,BYTE 4
add edi,BYTE 3
dec ecx
jmp SHORT .L3
.L4 ; unroll 4 times
push ebp
mov ebp,ecx
shr ebp,2
; save count
push ecx
.L5
mov eax,[esi] ; first dword eax = [A][R][G][B]
mov ebx,[esi+4] ; second dword ebx = [a][r][g][b]
bswap eax ; eax = [B][G][R][A]
bswap ebx ; ebx = [b][g][r][a]
mov al,[esi+4+2] ; eax = [B][G][R][r]
mov bh,[esi+4+4+1] ; ebx = [b][g][G][a]
ror eax,8 ; eax = [r][B][G][R] (done)
mov bl,[esi+4+4+2] ; ebx = [b][g][G][R]
ror ebx,16 ; ebx = [G][R][b][g] (done)
mov [edi],eax
mov [edi+4],ebx
mov ecx,[esi+12] ; third dword ecx = [a][r][g][b]
bswap ecx ; ecx = [b][g][r][a]
mov cl,[esi+8] ; ecx = [b][g][r][B] (done)
add esi,BYTE 4*4
mov [edi+8],ecx
add edi,BYTE 3*4
dec ebp
jnz .L5
; check tail
pop ecx
and ecx,BYTE 11b
jz .L7
.L6 ; tail loop
mov dl,[esi]
mov bl,[esi+1]
mov al,[esi+2]
mov [edi],al
mov [edi+1],bl
mov [edi+2],dl
add esi,BYTE 4
add edi,BYTE 3
dec ecx
jnz .L6
.L7
pop ebp
retn
;; 32 bit RGB 888 to 16 BIT RGB 565
_ConvertX86p32_16RGB565:
; check short
cmp ecx,BYTE 16
ja .L3
.L1 ; short loop
mov bl,[esi+0] ; blue
mov al,[esi+1] ; green
mov ah,[esi+2] ; red
shr ah,3
and al,11111100b
shl eax,3
shr bl,3
add al,bl
mov [edi+0],al
mov [edi+1],ah
add esi,BYTE 4
add edi,BYTE 2
dec ecx
jnz .L1
.L2: ; End of short loop
retn
.L3 ; head
mov ebx,edi
and ebx,BYTE 11b
jz .L4
mov bl,[esi+0] ; blue
mov al,[esi+1] ; green
mov ah,[esi+2] ; red
shr ah,3
and al,11111100b
shl eax,3
shr bl,3
add al,bl
mov [edi+0],al
mov [edi+1],ah
add esi,BYTE 4
add edi,BYTE 2
dec ecx
.L4:
; save count
push ecx
; unroll twice
shr ecx,1
; point arrays to end
lea esi,[esi+ecx*8]
lea edi,[edi+ecx*4]
; negative counter
neg ecx
jmp SHORT .L6
.L5:
mov [edi+ecx*4-4],eax
.L6:
mov eax,[esi+ecx*8]
shr ah,2
mov ebx,[esi+ecx*8+4]
shr eax,3
mov edx,[esi+ecx*8+4]
shr bh,2
mov dl,[esi+ecx*8+2]
shl ebx,13
and eax,000007FFh
shl edx,8
and ebx,07FF0000h
and edx,0F800F800h
add eax,ebx
add eax,edx
inc ecx
jnz .L5
mov [edi+ecx*4-4],eax
; tail
pop ecx
test cl,1
jz .L7
mov bl,[esi+0] ; blue
mov al,[esi+1] ; green
mov ah,[esi+2] ; red
shr ah,3
and al,11111100b
shl eax,3
shr bl,3
add al,bl
mov [edi+0],al
mov [edi+1],ah
add esi,BYTE 4
add edi,BYTE 2
.L7:
retn
;; 32 bit RGB 888 to 16 BIT BGR 565
_ConvertX86p32_16BGR565:
; check short
cmp ecx,BYTE 16
ja .L3
.L1 ; short loop
mov ah,[esi+0] ; blue
mov al,[esi+1] ; green
mov bl,[esi+2] ; red
shr ah,3
and al,11111100b
shl eax,3
shr bl,3
add al,bl
mov [edi+0],al
mov [edi+1],ah
add esi,BYTE 4
add edi,BYTE 2
dec ecx
jnz .L1
.L2
retn
.L3 ; head
mov ebx,edi
and ebx,BYTE 11b
jz .L4
mov ah,[esi+0] ; blue
mov al,[esi+1] ; green
mov bl,[esi+2] ; red
shr ah,3
and al,11111100b
shl eax,3
shr bl,3
add al,bl
mov [edi+0],al
mov [edi+1],ah
add esi,BYTE 4
add edi,BYTE 2
dec ecx
.L4 ; save count
push ecx
; unroll twice
shr ecx,1
; point arrays to end
lea esi,[esi+ecx*8]
lea edi,[edi+ecx*4]
; negative count
neg ecx
jmp SHORT .L6
.L5
mov [edi+ecx*4-4],eax
.L6
mov edx,[esi+ecx*8+4]
mov bh,[esi+ecx*8+4]
mov ah,[esi+ecx*8]
shr bh,3
mov al,[esi+ecx*8+1]
shr ah,3
mov bl,[esi+ecx*8+5]
shl eax,3
mov dl,[esi+ecx*8+2]
shl ebx,19
and eax,0000FFE0h
shr edx,3
and ebx,0FFE00000h
and edx,001F001Fh
add eax,ebx
add eax,edx
inc ecx
jnz .L5
mov [edi+ecx*4-4],eax
; tail
pop ecx
and ecx,BYTE 1
jz .L7
mov ah,[esi+0] ; blue
mov al,[esi+1] ; green
mov bl,[esi+2] ; red
shr ah,3
and al,11111100b
shl eax,3
shr bl,3
add al,bl
mov [edi+0],al
mov [edi+1],ah
add esi,BYTE 4
add edi,BYTE 2
.L7
retn
;; 32 BIT RGB TO 16 BIT RGB 555
_ConvertX86p32_16RGB555:
; check short
cmp ecx,BYTE 16
ja .L3
.L1 ; short loop
mov bl,[esi+0] ; blue
mov al,[esi+1] ; green
mov ah,[esi+2] ; red
shr ah,3
and al,11111000b
shl eax,2
shr bl,3
add al,bl
mov [edi+0],al
mov [edi+1],ah
add esi,BYTE 4
add edi,BYTE 2
dec ecx
jnz .L1
.L2
retn
.L3 ; head
mov ebx,edi
and ebx,BYTE 11b
jz .L4
mov bl,[esi+0] ; blue
mov al,[esi+1] ; green
mov ah,[esi+2] ; red
shr ah,3
and al,11111000b
shl eax,2
shr bl,3
add al,bl
mov [edi+0],al
mov [edi+1],ah
add esi,BYTE 4
add edi,BYTE 2
dec ecx
.L4 ; save count
push ecx
; unroll twice
shr ecx,1
; point arrays to end
lea esi,[esi+ecx*8]
lea edi,[edi+ecx*4]
; negative counter
neg ecx
jmp SHORT .L6
.L5
mov [edi+ecx*4-4],eax
.L6
mov eax,[esi+ecx*8]
shr ah,3
mov ebx,[esi+ecx*8+4]
shr eax,3
mov edx,[esi+ecx*8+4]
shr bh,3
mov dl,[esi+ecx*8+2]
shl ebx,13
and eax,000007FFh
shl edx,7
and ebx,07FF0000h
and edx,07C007C00h
add eax,ebx
add eax,edx
inc ecx
jnz .L5
mov [edi+ecx*4-4],eax
; tail
pop ecx
and ecx,BYTE 1
jz .L7
mov bl,[esi+0] ; blue
mov al,[esi+1] ; green
mov ah,[esi+2] ; red
shr ah,3
and al,11111000b
shl eax,2
shr bl,3
add al,bl
mov [edi+0],al
mov [edi+1],ah
add esi,BYTE 4
add edi,BYTE 2
.L7
retn
;; 32 BIT RGB TO 16 BIT BGR 555
_ConvertX86p32_16BGR555:
; check short
cmp ecx,BYTE 16
ja .L3
.L1 ; short loop
mov ah,[esi+0] ; blue
mov al,[esi+1] ; green
mov bl,[esi+2] ; red
shr ah,3
and al,11111000b
shl eax,2
shr bl,3
add al,bl
mov [edi+0],al
mov [edi+1],ah
add esi,BYTE 4
add edi,BYTE 2
dec ecx
jnz .L1
.L2
retn
.L3 ; head
mov ebx,edi
and ebx,BYTE 11b
jz .L4
mov ah,[esi+0] ; blue
mov al,[esi+1] ; green
mov bl,[esi+2] ; red
shr ah,3
and al,11111000b
shl eax,2
shr bl,3
add al,bl
mov [edi+0],al
mov [edi+1],ah
add esi,BYTE 4
add edi,BYTE 2
dec ecx
.L4 ; save count
push ecx
; unroll twice
shr ecx,1
; point arrays to end
lea esi,[esi+ecx*8]
lea edi,[edi+ecx*4]
; negative counter
neg ecx
jmp SHORT .L6
.L5
mov [edi+ecx*4-4],eax
.L6
mov edx,[esi+ecx*8+4]
mov bh,[esi+ecx*8+4]
mov ah,[esi+ecx*8]
shr bh,3
mov al,[esi+ecx*8+1]
shr ah,3
mov bl,[esi+ecx*8+5]
shl eax,2
mov dl,[esi+ecx*8+2]
shl ebx,18
and eax,00007FE0h
shr edx,3
and ebx,07FE00000h
and edx,001F001Fh
add eax,ebx
add eax,edx
inc ecx
jnz .L5
mov [edi+ecx*4-4],eax
; tail
pop ecx
and ecx,BYTE 1
jz .L7
mov ah,[esi+0] ; blue
mov al,[esi+1] ; green
mov bl,[esi+2] ; red
shr ah,3
and al,11111000b
shl eax,2
shr bl,3
add al,bl
mov [edi+0],al
mov [edi+1],ah
add esi,BYTE 4
add edi,BYTE 2
.L7
retn
;; FROM 32 BIT RGB to 8 BIT RGB (rrrgggbbb)
;; This routine writes FOUR pixels at once (dword) and then, if they exist
;; the trailing three pixels
_ConvertX86p32_8RGB332:
.L_ALIGNED
push ecx
shr ecx,2 ; We will draw 4 pixels at once
jnz .L1
jmp .L2 ; short jump out of range :(
.L1:
mov eax,[esi] ; first pair of pixels
mov edx,[esi+4]
shr dl,6
mov ebx,eax
shr al,6
and ah,0e0h
shr ebx,16
and dh,0e0h
shr ah,3
and bl,0e0h
shr dh,3
or al,bl
mov ebx,edx
or al,ah
shr ebx,16
or dl,dh
and bl,0e0h
or dl,bl
mov ah,dl
mov ebx,[esi+8] ; second pair of pixels
mov edx,ebx
and bh,0e0h
shr bl,6
and edx,0e00000h
shr edx,16
shr bh,3
ror eax,16
or bl,dl
mov edx,[esi+12]
or bl,bh
mov al,bl
mov ebx,edx
and dh,0e0h
shr dl,6
and ebx,0e00000h
shr dh,3
mov ah,dl
shr ebx,16
or ah,dh
or ah,bl
rol eax,16
add esi,BYTE 16
mov [edi],eax
add edi,BYTE 4
dec ecx
jz .L2 ; L1 out of range for short jump :(
jmp .L1
.L2:
pop ecx
and ecx,BYTE 3 ; mask out number of pixels to draw
jz .L4 ; Nothing to do anymore
.L3:
mov eax,[esi] ; single pixel conversion for trailing pixels
mov ebx,eax
shr al,6
and ah,0e0h
shr ebx,16
shr ah,3
and bl,0e0h
or al,ah
or al,bl
mov [edi],al
inc edi
add esi,BYTE 4
dec ecx
jnz .L3
.L4:
retn
%ifidn __OUTPUT_FORMAT__,elf
section .note.GNU-stack noalloc noexec nowrite progbits
%endif