1 ; $Id: tmap_lin.asm,v 1.2 2003-02-18 20:15:48 btb Exp $
2 ;THE COMPUTER CODE CONTAINED HEREIN IS THE SOLE PROPERTY OF PARALLAX
3 ;SOFTWARE CORPORATION ("PARALLAX"). PARALLAX, IN DISTRIBUTING THE CODE TO
4 ;END-USERS, AND SUBJECT TO ALL OF THE TERMS AND CONDITIONS HEREIN, GRANTS A
5 ;ROYALTY-FREE, PERPETUAL LICENSE TO SUCH END-USERS FOR USE BY SUCH END-USERS
6 ;IN USING, DISPLAYING, AND CREATING DERIVATIVE WORKS THEREOF, SO LONG AS
7 ;SUCH USE, DISPLAY OR CREATION IS FOR NON-COMMERCIAL, ROYALTY OR REVENUE
8 ;FREE PURPOSES. IN NO EVENT SHALL THE END-USER USE THE COMPUTER CODE
9 ;CONTAINED HEREIN FOR REVENUE-BEARING PURPOSES. THE END-USER UNDERSTANDS
10 ;AND AGREES TO THE TERMS HEREIN AND ACCEPTS THE SAME BY USE OF THIS FILE.
11 ;COPYRIGHT 1993-1998 PARALLAX SOFTWARE CORPORATION. ALL RIGHTS RESERVED.
14 ; Linearly interpolating texture mapper inner loop
17 ; Revision 1.3 1995/02/20 18:22:57 john
18 ; Put all the externs in the assembly modules into tmap_inc.asm.
19 ; Also, moved all the C versions of the inner loops into a new module,
22 ; Revision 1.2 1995/02/20 17:09:10 john
23 ; Added code so that you can build the tmapper with no assembly!
25 ; Revision 1.1 1994/11/30 12:24:04 mike
28 ; Revision 1.6 1994/11/12 16:39:40 mike
31 ; Revision 1.5 1994/02/10 21:24:43 matt
32 ; Got rid of debug_on set
34 ; Revision 1.4 1994/01/31 15:39:53 mike
35 ; Write additional inner loop for transparency.
37 ; Revision 1.3 1994/01/26 11:50:04 john
38 ; Added transparency to linear unlighted texture mapper.
40 ; Revision 1.2 1993/11/22 10:24:36 mike
41 ; *** empty log message ***
43 ; Revision 1.1 1993/09/08 17:29:50 mike
51 global _asm_tmap_scanline_lin
52 global asm_tmap_scanline_lin
56 %include "tmap_inc.asm"
62 ; --------------------------------------------------------------------------------------------------
64 ; _xleft fixed point left x coordinate
65 ; _xright fixed point right x coordinate
66 ; _y fixed point y coordinate
67 ; _pixptr address of source pixel map
68 ; _u fixed point initial u coordinate
69 ; _v fixed point initial v coordinate
70 ; _du_dx fixed point du/dx
71 ; _dv_dx fixed point dv/dx
73 ; for (x = (int) xleft; x <= (int) xright; x++) {
74 ; _setcolor(read_pixel_from_tmap(srcb,((int) (u/z)) & 63,((int) (v/z)) & 63));
83 _asm_tmap_scanline_lin:
84 asm_tmap_scanline_lin:
87 ; Setup for loop: _loop_count iterations = (int) xright - (int) xleft
88 ; esi source pixel pointer = pixptr
89 ; edi initial row pointer = y*320+x
91 ; set esi = pointer to start of texture map data
94 ; set edi = address of first pixel to modify
96 cmp edi,[_window_bottom]
99 imul edi,[_bytes_per_row]
106 add edi,[_write_buffer]
108 ; set _loop_count = # of iterations
110 cmp eax,[_window_right]
112 mov eax,[_window_right]
113 eax_ok1: cmp eax,[_window_left]
115 mov eax,[_window_left]
121 cmp eax,[_window_width]
123 mov eax,[_window_width]
125 mov [_loop_count],eax
127 ; edi destination pixel pointer
148 test dword [_Transparency_on],-1
149 jne near transparent_texture
151 %define _size (_end1 - _start1)/num_iters
153 sub eax,[_loop_count]
155 inc eax ; sort of a hack, but we can get -1 here and want to be graceful
156 jns j_eax_ok1 ; if we jump, we had -1, which is kind of ok, if not, we int 3
157 int 3 ; oops, going to jump behind _start1, very bad...
158 sub eax,eax ; ok to continue
159 j_eax_ok1: imul eax,eax,_size
166 ; "OPTIMIZATIONS" maybe not worth making
167 ; Getting rid of the esi from the mov al,[esi+eax] instruction.
168 ; This would require moving into eax at the top of the loop, rather than doing the sub eax,eax.
169 ; You would have to align your bitmaps so that the two shlds would create the proper base address.
170 ; In other words, your bitmap data would have to begin at 4096x (for 64x64 bitmaps).
171 ; I did timings without converting the sub to a mov eax,esi and setting esi to the proper value.
172 ; There was a speedup of about 1% to 1.5% without converting the sub to a mov.
173 ; Getting rid of the edi by doing a mov nnnn[edi],al instead of mov [edi],al.
174 ; The problem with this is you would have a dword offset for nnnn. My timings indicate it is slower. (I think.)
175 ; Combining u,v and du,dv into single longwords.
176 ; The problem with this is you then must do a 16 bit operation to extract them, and you don't have enough
177 ; instructions to separate a destination operand from being used by the next instruction. It shaves out one
178 ; register instruction (an add reg,reg), but adds a 16 bit operation, and the setup is more complicated.
185 ; esi pointer to source bitmap
188 mov eax,ebp ; clear for
189 add ebp,edx ; update v coordinate
190 shr eax,26 ; shift in v coordinate
191 shld eax,ebx,6 ; shift in u coordinate while shifting up v coordinate
192 add ebx,ecx ; update u coordinate
193 mov al,[esi+eax] ; get pixel from source bitmap
195 inc edi ; XPARENT ADDED BY JOHN
197 ; inner loop if bitmaps are 256x256
198 ; your register usage is bogus, and you must clear ecx
200 ; this is only about 10% faster in the inner loop
201 ; this method would adapt to writing two pixels at a time better than
202 ; the 64x64 method because you wouldn't run out of registers
203 ; Note that this method assumes that both dv_dx and du_dx are in edx.
205 ; where each field is 8 bits, vi = integer v coordinate, vf = fractional v coordinate, etc.
220 ; ----------------------------------------------------------------------------------------
221 ; if texture map has transparency, use this code.
223 test dword [_loop_count],-1
226 mov eax,ebp ; clear for
227 add ebp,edx ; update v coordinate
228 shr eax,26 ; shift in v coordinate
229 shld eax,ebx,6 ; shift in u coordinate while shifting up v coordinate
230 add ebx,ecx ; update u coordinate
231 mov al,[esi+eax] ; get pixel from source bitmap
235 transp: inc edi ; XPARENT ADDED BY JOHN
237 dec dword [_loop_count]
244 ; This is the inner loop to write two pixels at a time
245 ; This is about 2.5% faster overall (on Mike's 66 MHz 80486 DX2, VLB)
246 ; You must write code to even align edi and do half as many iterations, and write
247 ; the beginning and ending extra pixels, if necessary.
248 ; sub eax,eax ; clear for
249 ; shld eax,ebp,6 ; shift in v coordinate
250 ; add ebp,_fx_dv_dx ; update v coordinate
251 ; shld eax,ebx,6 ; shift in u coordinate while shifting up v coordinate
252 ; add ebx,ecx ; update u coordinate
253 ; mov dl,[esi+eax] ; get pixel from source bitmap
255 ; sub eax,eax ; clear for
256 ; shld eax,ebp,6 ; shift in v coordinate
257 ; add ebp,_fx_dv_dx ; update v coordinate
258 ; shld eax,ebx,6 ; shift in u coordinate while shifting up v coordinate
259 ; add ebx,ecx ; update u coordinate
260 ; mov dh,[esi+eax] ; get pixel from source bitmap