1 ;THE COMPUTER CODE CONTAINED HEREIN IS THE SOLE PROPERTY OF PARALLAX
2 ;SOFTWARE CORPORATION ("PARALLAX"). PARALLAX, IN DISTRIBUTING THE CODE TO
3 ;END-USERS, AND SUBJECT TO ALL OF THE TERMS AND CONDITIONS HEREIN, GRANTS A
4 ;ROYALTY-FREE, PERPETUAL LICENSE TO SUCH END-USERS FOR USE BY SUCH END-USERS
5 ;IN USING, DISPLAYING, AND CREATING DERIVATIVE WORKS THEREOF, SO LONG AS
6 ;SUCH USE, DISPLAY OR CREATION IS FOR NON-COMMERCIAL, ROYALTY OR REVENUE
7 ;FREE PURPOSES. IN NO EVENT SHALL THE END-USER USE THE COMPUTER CODE
8 ;CONTAINED HEREIN FOR REVENUE-BEARING PURPOSES. THE END-USER UNDERSTANDS
9 ;AND AGREES TO THE TERMS HEREIN AND ACCEPTS THE SAME BY USE OF THIS FILE.
10 ;COPYRIGHT 1993-1998 PARALLAX SOFTWARE CORPORATION. ALL RIGHTS RESERVED.
12 ; $Source: /cvs/cvsroot/d2x/texmap/tmap_lin.asm,v $
13 ; $Revision: 1.1.1.1 $
15 ; $Date: 2001-01-19 03:30:16 $
17 ; Linearly interpolating texture mapper inner loop
19 ; $Log: not supported by cvs2svn $
20 ; Revision 1.1.1.1 1999/06/14 22:13:53 donut
21 ; Import of d1x 1.37 source.
23 ; Revision 1.3 1995/02/20 18:22:57 john
24 ; Put all the externs in the assembly modules into tmap_inc.asm.
25 ; Also, moved all the C versions of the inner loops into a new module,
28 ; Revision 1.2 1995/02/20 17:09:10 john
29 ; Added code so that you can build the tmapper with no assembly!
31 ; Revision 1.1 1994/11/30 12:24:04 mike
34 ; Revision 1.6 1994/11/12 16:39:40 mike
37 ; Revision 1.5 1994/02/10 21:24:43 matt
38 ; Got rid of debug_on set
40 ; Revision 1.4 1994/01/31 15:39:53 mike
41 ; Write additional inner loop for transparency.
43 ; Revision 1.3 1994/01/26 11:50:04 john
44 ; Added transparency to linear unlighted texture mapper.
46 ; Revision 1.2 1993/11/22 10:24:36 mike
47 ; *** empty log message ***
49 ; Revision 1.1 1993/09/08 17:29:50 mike
57 global _asm_tmap_scanline_lin
58 global asm_tmap_scanline_lin
62 %include "tmap_inc.asm"
68 ; --------------------------------------------------------------------------------------------------
70 ; _xleft fixed point left x coordinate
71 ; _xright fixed point right x coordinate
72 ; _y fixed point y coordinate
73 ; _pixptr address of source pixel map
74 ; _u fixed point initial u coordinate
75 ; _v fixed point initial v coordinate
76 ; _du_dx fixed point du/dx
77 ; _dv_dx fixed point dv/dx
79 ; for (x = (int) xleft; x <= (int) xright; x++) {
80 ; _setcolor(read_pixel_from_tmap(srcb,((int) (u/z)) & 63,((int) (v/z)) & 63));
89 _asm_tmap_scanline_lin:
90 asm_tmap_scanline_lin:
93 ; Setup for loop: _loop_count iterations = (int) xright - (int) xleft
94 ; esi source pixel pointer = pixptr
95 ; edi initial row pointer = y*320+x
97 ; set esi = pointer to start of texture map data
100 ; set edi = address of first pixel to modify
102 cmp edi,[_window_bottom]
105 imul edi,[_bytes_per_row]
112 add edi,[_write_buffer]
114 ; set _loop_count = # of iterations
116 cmp eax,[_window_right]
118 mov eax,[_window_right]
119 eax_ok1: cmp eax,[_window_left]
121 mov eax,[_window_left]
127 cmp eax,[_window_width]
129 mov eax,[_window_width]
131 mov [_loop_count],eax
133 ; edi destination pixel pointer
154 test dword [_Transparency_on],-1
155 jne near transparent_texture
157 %define _size (_end1 - _start1)/num_iters
159 sub eax,[_loop_count]
161 inc eax ; sort of a hack, but we can get -1 here and want to be graceful
162 jns j_eax_ok1 ; if we jump, we had -1, which is kind of ok, if not, we int 3
163 int 3 ; oops, going to jump behind _start1, very bad...
164 sub eax,eax ; ok to continue
165 j_eax_ok1: imul eax,eax,_size
172 ; "OPTIMIZATIONS" maybe not worth making
173 ; Getting rid of the esi from the mov al,[esi+eax] instruction.
174 ; This would require moving into eax at the top of the loop, rather than doing the sub eax,eax.
175 ; You would have to align your bitmaps so that the two shlds would create the proper base address.
176 ; In other words, your bitmap data would have to begin at 4096x (for 64x64 bitmaps).
177 ; I did timings without converting the sub to a mov eax,esi and setting esi to the proper value.
178 ; There was a speedup of about 1% to 1.5% without converting the sub to a mov.
179 ; Getting rid of the edi by doing a mov nnnn[edi],al instead of mov [edi],al.
180 ; The problem with this is you would have a dword offset for nnnn. My timings indicate it is slower. (I think.)
181 ; Combining u,v and du,dv into single longwords.
182 ; The problem with this is you then must do a 16 bit operation to extract them, and you don't have enough
183 ; instructions to separate a destination operand from being used by the next instruction. It shaves out one
184 ; register instruction (an add reg,reg), but adds a 16 bit operation, and the setup is more complicated.
191 ; esi pointer to source bitmap
194 mov eax,ebp ; clear for
195 add ebp,edx ; update v coordinate
196 shr eax,26 ; shift in v coordinate
197 shld eax,ebx,6 ; shift in u coordinate while shifting up v coordinate
198 add ebx,ecx ; update u coordinate
199 mov al,[esi+eax] ; get pixel from source bitmap
201 inc edi ; XPARENT ADDED BY JOHN
203 ; inner loop if bitmaps are 256x256
204 ; your register usage is bogus, and you must clear ecx
206 ; this is only about 10% faster in the inner loop
207 ; this method would adapt to writing two pixels at a time better than
208 ; the 64x64 method because you wouldn't run out of registers
209 ; Note that this method assumes that both dv_dx and du_dx are in edx.
211 ; where each field is 8 bits, vi = integer v coordinate, vf = fractional v coordinate, etc.
226 ; ----------------------------------------------------------------------------------------
227 ; if texture map has transparency, use this code.
229 test dword [_loop_count],-1
232 mov eax,ebp ; clear for
233 add ebp,edx ; update v coordinate
234 shr eax,26 ; shift in v coordinate
235 shld eax,ebx,6 ; shift in u coordinate while shifting up v coordinate
236 add ebx,ecx ; update u coordinate
237 mov al,[esi+eax] ; get pixel from source bitmap
241 transp: inc edi ; XPARENT ADDED BY JOHN
243 dec dword [_loop_count]
250 ; This is the inner loop to write two pixels at a time
251 ; This is about 2.5% faster overall (on Mike's 66 MHz 80486 DX2, VLB)
252 ; You must write code to even align edi and do half as many iterations, and write
253 ; the beginning and ending extra pixels, if necessary.
254 ; sub eax,eax ; clear for
255 ; shld eax,ebp,6 ; shift in v coordinate
256 ; add ebp,_fx_dv_dx ; update v coordinate
257 ; shld eax,ebx,6 ; shift in u coordinate while shifting up v coordinate
258 ; add ebx,ecx ; update u coordinate
259 ; mov dl,[esi+eax] ; get pixel from source bitmap
261 ; sub eax,eax ; clear for
262 ; shld eax,ebp,6 ; shift in v coordinate
263 ; add ebp,_fx_dv_dx ; update v coordinate
264 ; shld eax,ebx,6 ; shift in u coordinate while shifting up v coordinate
265 ; add ebx,ecx ; update u coordinate
266 ; mov dh,[esi+eax] ; get pixel from source bitmap