;;; x86_64 assembly for basic texture mapping functions, based on a.nasm.
;;; For use with System V AMD64 calling convention (rdi rsi rdx rcx r8 r9)
;;; See the included license file "BUILDLIC.TXT" for license info.

SECTION .text

%ifdef UNDERSCORES
%define asm1 _asm1
%define asm2 _asm2

%define a64_bpl _a64_bpl
%define a64_transmode _a64_transmode
%define a64_glogy _a64_glogy
%define a64_gtrans _a64_gtrans
%define a64_paloffs _a64_paloffs

%define tvlineasm1 _tvlineasm1
%define tvlineasm2 _tvlineasm2
%endif

;;; Imports
EXTERN asm1  ; intptr_t
EXTERN asm2  ; intptr_t

EXTERN a64_bpl  ; int32_t
EXTERN a64_transmode  ; int32_t
EXTERN a64_glogy  ; int32_t
EXTERN a64_gtrans  ; char *
EXTERN a64_paloffs  ; intptr_t

;;; Exports
GLOBAL tvlineasm1  ; Masked & transluscent 1-pixel wide vline
GLOBAL tvlineasm2  ; Masked & transluscent 2-pixel wide vline


;;; ========== MACROS ==========

; Construct name referring to low doubleword of GPR, only for r8-r15.
%define REGd(reg) reg %+ d
; Construct name of whole GPR from word-sized register, only for ax-bp.
%define rREG(regw) r %+ regw
; Construct name of low doubleword of GPR from word-sized register, only for ax-bp.
%define eREG(regw) e %+ regw

;;; Multi-line macros for code shared between functions

;; Prologue getters
%macro PGET_rBpl 0
	push rBpl
	mov REGd(rBpl), dword [rel a64_bpl]
%endmacro

%macro PGET_rATransluc 0
	push rATransluc
	mov rATransluc, [rel a64_gtrans]
%endmacro

; Get right shift count 'glogy' into cl
%macro GET_cl 0
	mov cl, byte [rel a64_glogy]
%endmacro

; Look up texel
; <dest_reg_lobyte> := bufplc[vplc>>glogy]
; High bytes are cleared to zero.
%macro gettexel 3  ; <dest_reg_word> <bufplc_reg> <vlpc_reg>
	mov eREG(%1), %3
    shr eREG(%1), cl
	movzx eREG(%1), byte [%2 + rREG(%1)]
%endmacro

; Do palette/shade and translucency lookup
%macro dotranspal 4  ; <tmp_reg> <tmp_reg_lobyte> <texel_reg> <dst_ofs>
	; Get palette index of the pixel in the frame buffer.
	; NOTE: e.g. "mov ah, byte [reg]" is not encodeable.
	movzx %1, byte [rDst + %4]
	shl %1, 8

	mov %2, byte [rAPalookup + %3]  ; palette/shade
	mov %2, byte [rATransluc + %1]	; translucency
%endmacro


;;; ========== TVLINEASM1 ==========
;;; TODO: transmode, nonpow2

;;; Registers used in the loop
%define rBpl r10
%define rATransluc r12
%define rVplc r11d

;;; Registers of input args also used in the loop
%define rVinc edi
%define rAPalookup rsi
%define rCnt edx
%define rABufplc r8
%define rDst r9

;;; int32_t tvlineasm1(vinc, *paloffs, cnt, vplc, *bufplc, *p)
;;; eax                edi   rsi       edx  ecx   r8       r9
	ALIGN 16
tvlineasm1:
	inc rCnt
	;; First, back up callee-saved registers and set up those used in the loop.
	PGET_rBpl
	mov rVplc, ecx
	GET_cl
	PGET_rATransluc
	push rbx
	jmp short tv1_loop

 ALIGN 16
tv1_loop:
	gettexel bx, rABufplc, rVplc
	cmp bl, 255
	je short tv1_skiptrans

	dotranspal rax, al, rbx, 0
	mov byte [rDst], al
tv1_skiptrans:
	add rVplc, rVinc
	add rDst, rBpl
	dec rCnt
	jnz short tv1_loop

	mov eax, rVplc  ; return vplc
	pop rbx
	pop rATransluc
	pop rBpl
	ret

;;; Undefine input arg macros
%undef rVinc
%undef rAPalookup
%undef rCnt
%undef rABufplc
%undef rDst

;;; Udefine rVplc register macro, but keep rBpl and rATransluc (r10 and r12).
;;; Take care not to clash with them in the following.
%undef rVplc


;;; ========== TVLINEASM2 ==========

;;; Registers used in the loop
%define rVinc2 r11d
%define rABufplc2 r13
%define rDstEnd r14
%define rAPalookup r15  ; TODO: second paloffs!

;;; Registers of input args also used in the loop
%define rVplc2 edi
%define rVinc1 esi
%define rABufplc1 rdx
; bufplc2: rcx -> r13
%define rVplc1 r8d
%define rDst r9

;;; Pass: asm1=vinc2, asm2=pend
;;; Return: asm1=vplc1, asm2=vplc2
;;; void tvlineasm2(vplc2, vinc1, *bufplc1, *bufplc2, vplc1, *p)
;;;                 edi    esi    rdx       rcx       r8d    r9

	ALIGN 16
tvlineasm2:
	;; First, back up callee-saved registers and set up those used in the loop.
	mov rVinc2, dword [rel asm1]
	PGET_rBpl
	push rABufplc2
	mov rABufplc2, rcx
	GET_cl
	PGET_rATransluc

	push rDstEnd
	mov rDstEnd, [rel asm2]
	dec rDstEnd
	add rDstEnd, rBpl  ; one more: compare with a-c.c's tvlineasm2()

	push rbx
	push rbp
	push rAPalookup
	mov rAPalookup, [rel a64_paloffs]
	jmp short tv2_loop

 ALIGN 16
tv2_loop:
	gettexel bx, rABufplc1, rVplc1
	gettexel bp, rABufplc2, rVplc2

	mov eax, ebp
	shl eax, 8
	or eax, ebx

	cmp eax, 0xffff
	jz tv2_skiptrans

	dotranspal rax, al, rbx, 0
	dotranspal rbx, bl, rbp, 1
	and rax, 0xff
	shl rbx, 8
	or rax, rbx

	mov word [rDst], ax

tv2_skiptrans:
	add rVplc1, rVinc1
	add rVplc2, rVinc2
	add rDst, rBpl
	cmp rDst, rDstEnd
	jnz tv2_loop

	pop r15
	pop rbp
	pop rbx
	pop rDstEnd
	pop rATransluc
	pop rABufplc2
	pop rBpl
	ret