1 /* $Id: x86_cliptest.S,v 1.10 2002/06/11 01:26:58 brianp Exp $ */
4 * Mesa 3-D graphics library
7 * Copyright (C) 1999-2001 Brian Paul All Rights Reserved.
9 * Permission is hereby granted, free of charge, to any person obtaining a
10 * copy of this software and associated documentation files (the "Software"),
11 * to deal in the Software without restriction, including without limitation
12 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
13 * and/or sell copies of the Software, and to permit persons to whom the
14 * Software is furnished to do so, subject to the following conditions:
16 * The above copyright notice and this permission notice shall be included
17 * in all copies or substantial portions of the Software.
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
22 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
23 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
24 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
28 * NOTE: Avoid using spaces in between '(' ')' and arguments, especially
29 * with macros like CONST, LLBL that expand to CONCAT(...). Putting spaces
30 * in there will break the build on some platforms.
34 #include "clip_args.h"
36 #define SRC0 REGOFF(0, ESI)
37 #define SRC1 REGOFF(4, ESI)
38 #define SRC2 REGOFF(8, ESI)
39 #define SRC3 REGOFF(12, ESI)
40 #define DST0 REGOFF(0, EDI)
41 #define DST1 REGOFF(4, EDI)
42 #define DST2 REGOFF(8, EDI)
43 #define DST3 REGOFF(12, EDI)
44 #define MAT0 REGOFF(0, EDX)
45 #define MAT1 REGOFF(4, EDX)
46 #define MAT2 REGOFF(8, EDX)
47 #define MAT3 REGOFF(12, EDX)
51 * Table for clip test.
55 * bit4 = abs(S(2)) > abs(S(3))
57 * bit2 = abs(S(1)) > abs(S(3))
59 * bit0 = abs(S(0)) > abs(S(3))
65 D_BYTE 0x00, 0x01, 0x00, 0x02, 0x04, 0x05, 0x04, 0x06
66 D_BYTE 0x00, 0x01, 0x00, 0x02, 0x08, 0x09, 0x08, 0x0a
67 D_BYTE 0x20, 0x21, 0x20, 0x22, 0x24, 0x25, 0x24, 0x26
68 D_BYTE 0x20, 0x21, 0x20, 0x22, 0x28, 0x29, 0x28, 0x2a
69 D_BYTE 0x00, 0x01, 0x00, 0x02, 0x04, 0x05, 0x04, 0x06
70 D_BYTE 0x00, 0x01, 0x00, 0x02, 0x08, 0x09, 0x08, 0x0a
71 D_BYTE 0x10, 0x11, 0x10, 0x12, 0x14, 0x15, 0x14, 0x16
72 D_BYTE 0x10, 0x11, 0x10, 0x12, 0x18, 0x19, 0x18, 0x1a
73 D_BYTE 0x3f, 0x3d, 0x3f, 0x3e, 0x37, 0x35, 0x37, 0x36
74 D_BYTE 0x3f, 0x3d, 0x3f, 0x3e, 0x3b, 0x39, 0x3b, 0x3a
75 D_BYTE 0x2f, 0x2d, 0x2f, 0x2e, 0x27, 0x25, 0x27, 0x26
76 D_BYTE 0x2f, 0x2d, 0x2f, 0x2e, 0x2b, 0x29, 0x2b, 0x2a
77 D_BYTE 0x3f, 0x3d, 0x3f, 0x3e, 0x37, 0x35, 0x37, 0x36
78 D_BYTE 0x3f, 0x3d, 0x3f, 0x3e, 0x3b, 0x39, 0x3b, 0x3a
79 D_BYTE 0x1f, 0x1d, 0x1f, 0x1e, 0x17, 0x15, 0x17, 0x16
80 D_BYTE 0x1f, 0x1d, 0x1f, 0x1e, 0x1b, 0x19, 0x1b, 0x1a
86 * _mesa_x86_cliptest_points4
98 #if defined(__ELF__) && defined(__PIC__) && defined(GNU_ASSEMBLER) && !defined(ELFPIC)
103 GLOBL GLNAME( _mesa_x86_cliptest_points4 )
104 GLNAME( _mesa_x86_cliptest_points4 ):
107 #define FRAME_OFFSET 20
109 #define FRAME_OFFSET 16
117 /* store pointer to clip_table on stack */
118 CALL( LLBL(ctp4_get_eip) )
119 ADD_L( CONST(_GLOBAL_OFFSET_TABLE_), EBX )
120 MOV_L( REGOFF(clip_table@GOT, EBX), EBX )
122 JMP( LLBL(ctp4_clip_table_ready) )
125 /* store eip in ebx */
126 MOV_L( REGIND(ESP), EBX )
129 LLBL(ctp4_clip_table_ready):
132 MOV_L( ARG_SOURCE, ESI )
133 MOV_L( ARG_DEST, EDI )
135 MOV_L( ARG_CLIP, EDX )
138 MOV_L( ARG_AND, EBP )
139 MOV_L( REGOFF(V4F_STRIDE, ESI), EAX )
141 MOV_L( REGOFF(V4F_COUNT, ESI), ECX )
142 MOV_L( REGOFF(V4F_START, ESI), ESI )
144 OR_L( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, EDI) )
145 MOV_L( EAX, ARG_SOURCE ) /* put stride in ARG_SOURCE */
147 MOV_L( CONST(4), REGOFF(V4F_SIZE, EDI) )
148 MOV_L( ECX, REGOFF(V4F_COUNT, EDI) )
150 MOV_L( REGOFF(V4F_START, EDI), EDI )
153 MOV_L( ECX, ARG_CLIP ) /* put clipmask + count in ARG_CLIP */
156 MOV_B( REGIND(EBX), AL )
157 MOV_B( REGIND(EBP), AH )
159 JZ( LLBL(ctp4_finish) )
165 FDIV_S( SRC3 ) /* GH: don't care about div-by-zero */
171 ADD_L( EBP, EBP ) /* ebp = abs(S(3))*2 ; carry = sign of S(3) */
174 ADD_L( EBX, EBX ) /* ebx = abs(S(2))*2 ; carry = sign of S(2) */
177 CMP_L( EBX, EBP ) /* carry = abs(S(2))*2 > abs(S(3))*2 */
182 ADD_L( EBX, EBX ) /* ebx = abs(S(1))*2 ; carry = sign of S(1) */
185 CMP_L( EBX, EBP ) /* carry = abs(S(1))*2 > abs(S(3))*2 */
190 ADD_L( EBX, EBX ) /* ebx = abs(S(0))*2 ; carry = sign of S(0) */
193 CMP_L( EBX, EBP ) /* carry = abs(S(0))*2 > abs(S(3))*2 */
198 MOV_L( REGIND(ESP), EBP ) /* clip_table */
200 MOV_B( REGBI(EBP, ECX), CL )
202 MOV_B( REGOFF(clip_table,ECX), CL )
209 MOV_B( CL, REGIND(EDX) )
211 JZ( LLBL(ctp4_proj) )
217 MOV_L( CONST(0), DST0 )
218 MOV_L( CONST(0), DST1 )
219 MOV_L( CONST(0), DST2 )
220 MOV_L( CONST(0x3f800000), DST3 )
222 JMP( LLBL(ctp4_next) )
226 FLD_S( SRC0 ) /* F0 F3 */
229 FLD_S( SRC1 ) /* F1 F0 F3 */
232 FLD_S( SRC2 ) /* F2 F1 F0 F3 */
235 FXCH( ST(2) ) /* F0 F1 F2 F3 */
236 FSTP_S( DST0 ) /* F1 F2 F3 */
237 FSTP_S( DST1 ) /* F2 F3 */
238 FSTP_S( DST2 ) /* F3 */
244 ADD_L( CONST(16), EDI )
246 ADD_L( ARG_SOURCE, ESI )
247 CMP_L( EDX, ARG_CLIP )
249 JNZ( LLBL(ctp4_top) )
252 MOV_L( ARG_AND, EDX )
254 MOV_B( AL, REGIND(ECX) )
255 MOV_B( AH, REGIND(EDX) )
259 MOV_L( ARG_DEST, EAX )
261 POP_L( ESI ) /* discard ptr to clip_table */
277 GLOBL GLNAME( _mesa_x86_cliptest_points4_np )
278 GLNAME( _mesa_x86_cliptest_points4_np ):
281 #define FRAME_OFFSET 20
283 #define FRAME_OFFSET 16
291 /* store pointer to clip_table on stack */
292 CALL( LLBL(ctp4_np_get_eip) )
293 ADD_L( CONST(_GLOBAL_OFFSET_TABLE_), EBX )
294 MOV_L( REGOFF(clip_table@GOT, EBX), EBX )
296 JMP( LLBL(ctp4_np_clip_table_ready) )
298 LLBL(ctp4_np_get_eip):
299 /* store eip in ebx */
300 MOV_L( REGIND(ESP), EBX )
303 LLBL(ctp4_np_clip_table_ready):
306 MOV_L( ARG_SOURCE, ESI )
309 MOV_L( ARG_CLIP, EDX )
312 MOV_L( ARG_AND, EBP )
313 MOV_L( REGOFF(V4F_STRIDE, ESI), EAX )
315 MOV_L( REGOFF(V4F_COUNT, ESI), ECX )
316 MOV_L( REGOFF(V4F_START, ESI), ESI )
318 MOV_L( EAX, ARG_DEST ) /* put stride in ARG_DEST */
321 MOV_L( ECX, EDI ) /* put clipmask + count in EDI */
324 MOV_B( REGIND(EBX), AL )
325 MOV_B( REGIND(EBP), AH )
327 JZ( LLBL(ctp4_np_finish) )
336 ADD_L( EBP, EBP ) /* ebp = abs(S(3))*2 ; carry = sign of S(3) */
339 ADD_L( EBX, EBX ) /* ebx = abs(S(2))*2 ; carry = sign of S(2) */
342 CMP_L( EBX, EBP ) /* carry = abs(S(2))*2 > abs(S(3))*2 */
347 ADD_L( EBX, EBX ) /* ebx = abs(S(1))*2 ; carry = sign of S(1) */
350 CMP_L( EBX, EBP ) /* carry = abs(S(1))*2 > abs(S(3))*2 */
355 ADD_L( EBX, EBX ) /* ebx = abs(S(0))*2 ; carry = sign of S(0) */
358 CMP_L( EBX, EBP ) /* carry = abs(S(0))*2 > abs(S(3))*2 */
363 MOV_L( REGIND(ESP), EBP ) /* clip_table */
365 MOV_B( REGBI(EBP, ECX), CL )
367 MOV_B( REGOFF(clip_table,ECX), CL )
374 MOV_B( CL, REGIND(EDX) )
379 ADD_L( ARG_DEST, ESI )
382 JNZ( LLBL(ctp4_np_top) )
385 MOV_L( ARG_AND, EDX )
387 MOV_B( AL, REGIND(ECX) )
388 MOV_B( AH, REGIND(EDX) )
390 LLBL(ctp4_np_finish):
392 MOV_L( ARG_SOURCE, EAX )
394 POP_L( ESI ) /* discard ptr to clip_table */