3 * Mesa 3-D graphics library
5 * Copyright (C) 1999-2001 Brian Paul All Rights Reserved.
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
11 * and/or sell copies of the Software, and to permit persons to whom the
12 * Software is furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
21 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
22 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
23 * OTHER DEALINGS IN THE SOFTWARE.
28 #define MATH_ASM_PTR_SIZE 4
29 #include "math/m_vector_asm.h"
30 #include "xform_args.h"
34 #define FRAME_OFFSET 4
38 GLOBL GLNAME( _mesa_3dnow_transform_points1_general )
39 HIDDEN(_mesa_3dnow_transform_points1_general)
40 GLNAME( _mesa_3dnow_transform_points1_general ):
44 MOV_L ( ARG_DEST, ECX )
45 MOV_L ( ARG_MATRIX, ESI )
46 MOV_L ( ARG_SOURCE, EAX )
47 MOV_L ( CONST(4), REGOFF(V4F_SIZE, ECX) )
48 OR_B ( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, ECX) )
49 MOV_L ( REGOFF(V4F_COUNT, EAX), EDX )
50 MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) )
54 MOV_L ( REGOFF(4, ECX), EDX )
56 MOV_L ( REGOFF(V4F_COUNT, EAX), ESI )
57 MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI )
58 MOV_L ( REGOFF(V4F_START, EAX), EAX )
61 JZ ( LLBL( G3TPGR_3 ) )
63 MOVQ ( REGIND(ECX), MM0 ) /* m01 | m00 */
64 MOVQ ( REGOFF(8, ECX), MM1 ) /* m03 | m02 */
66 MOVQ ( REGOFF(48, ECX), MM2 ) /* m31 | m30 */
67 MOVQ ( REGOFF(56, ECX), MM3 ) /* m33 | m32 */
72 MOVD ( REGIND(EAX), MM4 ) /* | x0 */
73 PUNPCKLDQ ( MM4, MM4 ) /* x0 | x0 */
75 MOVQ ( MM4, MM5 ) /* x0 | x0 */
76 PFMUL ( MM0, MM4 ) /* x0*m01 | x0*m00 */
78 PFMUL ( MM1, MM5 ) /* x0*m03 | x0*m02 */
79 PFADD ( MM2, MM4 ) /* x0*m01+m31 | x0*m00+m30 */
81 PFADD ( MM3, MM5 ) /* x0*m03+m33 | x0*m02+m32 */
82 MOVQ ( MM4, REGIND(EDX) ) /* write r1, r0 */
84 MOVQ ( MM5, REGOFF(8, EDX) ) /* write r3, r2 */
85 ADD_L ( EDI, EAX ) /* next vertex */
87 ADD_L ( CONST(16), EDX ) /* next r */
88 DEC_L ( ESI ) /* decrement vertex counter */
90 JNZ ( LLBL( G3TPGR_2 ) ) /* cnt > 0 ? -> process next vertex */
103 GLOBL GLNAME( _mesa_3dnow_transform_points1_identity )
104 HIDDEN(_mesa_3dnow_transform_points1_identity)
105 GLNAME( _mesa_3dnow_transform_points1_identity ):
109 MOV_L ( ARG_DEST, ECX )
110 MOV_L ( ARG_MATRIX, ESI )
111 MOV_L ( ARG_SOURCE, EAX )
112 MOV_L ( CONST(1), REGOFF(V4F_SIZE, ECX) )
113 OR_B ( CONST(VEC_SIZE_1), REGOFF(V4F_FLAGS, ECX) )
114 MOV_L ( REGOFF(V4F_COUNT, EAX), EDX )
115 MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) )
119 MOV_L ( REGOFF(4, ECX), EDX )
121 MOV_L ( REGOFF(V4F_COUNT, EAX), ESI )
122 MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI )
123 MOV_L ( REGOFF(V4F_START, EAX), EAX )
126 JZ ( LLBL( G3TPIR_4) )
131 MOVD ( REGIND(EAX), MM0 ) /* | x0 */
132 ADD_L ( EDI, EAX ) /* next vertex */
134 MOVD ( MM0, REGIND(EDX) ) /* | r0 */
135 ADD_L ( CONST(16), EDX ) /* next r */
137 DEC_L ( ESI ) /* decrement vertex counter */
138 JNZ ( LLBL( G3TPIR_3 ) ) /* cnt > 0 ? -> process next vertex */
151 GLOBL GLNAME( _mesa_3dnow_transform_points1_3d_no_rot )
152 HIDDEN(_mesa_3dnow_transform_points1_3d_no_rot)
153 GLNAME( _mesa_3dnow_transform_points1_3d_no_rot ):
157 MOV_L ( ARG_DEST, ECX )
158 MOV_L ( ARG_MATRIX, ESI )
159 MOV_L ( ARG_SOURCE, EAX )
160 MOV_L ( CONST(3), REGOFF(V4F_SIZE, ECX) )
161 OR_B ( CONST(VEC_SIZE_3), REGOFF(V4F_FLAGS, ECX) )
162 MOV_L ( REGOFF(V4F_COUNT, EAX), EDX )
163 MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) )
167 MOV_L ( REGOFF(4, ECX), EDX )
169 MOV_L ( REGOFF(V4F_COUNT, EAX), ESI )
170 MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI )
171 MOV_L ( REGOFF(V4F_START, EAX), EAX )
174 JZ ( LLBL( G3TP3NRR_3 ) )
176 MOVD ( REGIND(ECX), MM0 ) /* | m00 */
177 MOVQ ( REGOFF(48, ECX), MM2 ) /* m31 | m30 */
179 MOVD ( REGOFF(56, ECX), MM3 ) /* | m32 */
184 MOVD ( REGIND(EAX), MM4 ) /* | x0 */
185 PFMUL ( MM0, MM4 ) /* | x0*m00 */
187 PFADD ( MM2, MM4 ) /* m31 | x0*m00+m30 */
188 MOVQ ( MM4, REGIND(EDX) ) /* write r1, r0 */
190 MOVD ( MM3, REGOFF(8, EDX) ) /* write r2 */
191 ADD_L ( EDI, EAX ) /* next vertex */
193 ADD_L ( CONST(16), EDX ) /* next r */
194 DEC_L ( ESI ) /* decrement vertex counter */
196 JNZ ( LLBL( G3TP3NRR_2 ) ) /* cnt > 0 ? -> process next vertex */
209 GLOBL GLNAME( _mesa_3dnow_transform_points1_perspective )
210 HIDDEN(_mesa_3dnow_transform_points1_perspective)
211 GLNAME( _mesa_3dnow_transform_points1_perspective ):
215 MOV_L ( ARG_DEST, ECX )
216 MOV_L ( ARG_MATRIX, ESI )
217 MOV_L ( ARG_SOURCE, EAX )
218 MOV_L ( CONST(4), REGOFF(V4F_SIZE, ECX) )
219 OR_B ( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, ECX) )
220 MOV_L ( REGOFF(V4F_COUNT, EAX), EDX )
221 MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) )
225 MOV_L ( REGOFF(4, ECX), EDX )
227 MOV_L ( REGOFF(V4F_COUNT, EAX), ESI )
228 MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI )
229 MOV_L ( REGOFF(V4F_START, EAX), EAX )
232 JZ ( LLBL( G3TPPR_3 ) )
234 MOVD ( REGIND(ECX), MM0 ) /* | m00 */
235 MOVD ( REGOFF(56, ECX), MM3 ) /* | m32 */
240 MOVD ( REGIND(EAX), MM4 ) /* 0 | x0 */
241 PFMUL ( MM0, MM4 ) /* 0 | x0*m00 */
243 MOVQ ( MM4, REGIND(EDX) ) /* write r1, r0 */
244 MOVQ ( MM3, REGOFF(8, EDX) ) /* write r2 (=m32), r3 (=0) */
246 ADD_L ( EDI, EAX ) /* next vertex */
247 ADD_L ( CONST(16), EDX ) /* next r */
249 DEC_L ( ESI ) /* decrement vertex counter */
250 JNZ ( LLBL( G3TPPR_2 ) ) /* cnt > 0 ? -> process next vertex */
263 GLOBL GLNAME( _mesa_3dnow_transform_points1_2d )
264 HIDDEN(_mesa_3dnow_transform_points1_2d)
265 GLNAME( _mesa_3dnow_transform_points1_2d ):
269 MOV_L ( ARG_DEST, ECX )
270 MOV_L ( ARG_MATRIX, ESI )
271 MOV_L ( ARG_SOURCE, EAX )
272 MOV_L ( CONST(2), REGOFF(V4F_SIZE, ECX) )
273 OR_B ( CONST(VEC_SIZE_2), REGOFF(V4F_FLAGS, ECX) )
274 MOV_L ( REGOFF(V4F_COUNT, EAX), EDX )
275 MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) )
279 MOV_L ( REGOFF(4, ECX), EDX )
281 MOV_L ( REGOFF(V4F_COUNT, EAX), ESI )
282 MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI )
283 MOV_L ( REGOFF(V4F_START, EAX), EAX )
286 JZ ( LLBL( G3TP2R_3 ) )
288 MOVQ ( REGIND(ECX), MM0 ) /* m01 | m00 */
289 MOVQ ( REGOFF(48, ECX), MM2 ) /* m31 | m30 */
294 MOVD ( REGIND(EAX), MM4 ) /* | x0 */
295 PUNPCKLDQ ( MM4, MM4 ) /* x0 | x0 */
297 PFMUL ( MM0, MM4 ) /* x0*m01 | x0*m00 */
298 PFADD ( MM2, MM4 ) /* x0*m01+m31 | x0*m00+m30 */
300 MOVQ ( MM4, REGIND(EDX) ) /* write r1, r0 */
301 ADD_L ( EDI, EAX ) /* next vertex */
303 ADD_L ( CONST(16), EDX ) /* next r */
304 DEC_L ( ESI ) /* decrement vertex counter */
306 JNZ ( LLBL( G3TP2R_2 ) ) /* cnt > 0 ? -> process next vertex */
319 GLOBL GLNAME( _mesa_3dnow_transform_points1_2d_no_rot )
320 HIDDEN(_mesa_3dnow_transform_points1_2d_no_rot)
321 GLNAME( _mesa_3dnow_transform_points1_2d_no_rot ):
325 MOV_L ( ARG_DEST, ECX )
326 MOV_L ( ARG_MATRIX, ESI )
327 MOV_L ( ARG_SOURCE, EAX )
328 MOV_L ( CONST(2), REGOFF(V4F_SIZE, ECX) )
329 OR_B ( CONST(VEC_SIZE_2), REGOFF(V4F_FLAGS, ECX) )
330 MOV_L ( REGOFF(V4F_COUNT, EAX), EDX )
331 MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) )
335 MOV_L ( REGOFF(4, ECX), EDX )
337 MOV_L ( REGOFF(V4F_COUNT, EAX), ESI )
338 MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI )
339 MOV_L ( REGOFF(V4F_START, EAX), EAX )
342 JZ ( LLBL( G3TP2NRR_3 ) )
344 MOVD ( REGIND(ECX), MM0 ) /* | m00 */
345 MOVQ ( REGOFF(48, ECX), MM2 ) /* m31 | m30 */
350 MOVD ( REGIND(EAX), MM4 ) /* | x0 */
351 ADD_L ( EDI, EAX ) /* next vertex */
353 PFMUL ( MM0, MM4 ) /* | x0*m00 */
354 PFADD ( MM2, MM4 ) /* m31 | x0*m00+m30 */
356 MOVQ ( MM4, REGIND(EDX) ) /* write r1, r0 */
357 ADD_L ( CONST(16), EDX ) /* next r */
359 DEC_L ( ESI ) /* decrement vertex counter */
360 JNZ ( LLBL( G3TP2NRR_2 ) ) /* cnt > 0 ? -> process next vertex */
373 GLOBL GLNAME( _mesa_3dnow_transform_points1_3d )
374 HIDDEN(_mesa_3dnow_transform_points1_3d)
375 GLNAME( _mesa_3dnow_transform_points1_3d ):
379 MOV_L ( ARG_DEST, ECX )
380 MOV_L ( ARG_MATRIX, ESI )
381 MOV_L ( ARG_SOURCE, EAX )
382 MOV_L ( CONST(3), REGOFF(V4F_SIZE, ECX) )
383 OR_B ( CONST(VEC_SIZE_3), REGOFF(V4F_FLAGS, ECX) )
384 MOV_L ( REGOFF(V4F_COUNT, EAX), EDX )
385 MOV_L ( EDX, REGOFF(V4F_COUNT, ECX) )
389 MOV_L ( REGOFF(4, ECX), EDX )
391 MOV_L ( REGOFF(V4F_COUNT, EAX), ESI )
392 MOV_L ( REGOFF(V4F_STRIDE, EAX), EDI )
393 MOV_L ( REGOFF(V4F_START, EAX), EAX )
396 JZ ( LLBL( G3TP3R_3 ) )
398 MOVQ ( REGIND(ECX), MM0 ) /* m01 | m00 */
399 MOVD ( REGOFF(8, ECX), MM1 ) /* | m02 */
401 MOVQ ( REGOFF(48, ECX), MM2 ) /* m31 | m30 */
402 MOVD ( REGOFF(56, ECX), MM3 ) /* | m32 */
407 MOVD ( REGIND(EAX), MM4 ) /* | x0 */
408 PUNPCKLDQ ( MM4, MM4 ) /* x0 | x0 */
410 MOVQ ( MM4, MM5 ) /* | x0 */
411 PFMUL ( MM0, MM4 ) /* x0*m01 | x0*m00 */
413 PFMUL ( MM1, MM5 ) /* | x0*m02 */
414 PFADD ( MM2, MM4 ) /* x0*m01+m31 | x0*m00+m30 */
416 PFADD ( MM3, MM5 ) /* | x0*m02+m32 */
417 MOVQ ( MM4, REGIND(EDX) ) /* write r1, r0 */
419 MOVD ( MM5, REGOFF(8, EDX) ) /* write r2 */
420 ADD_L ( EDI, EAX ) /* next vertex */
422 ADD_L ( CONST(16), EDX ) /* next r */
423 DEC_L ( ESI ) /* decrement vertex counter */
425 JNZ ( LLBL( G3TP3R_2 ) ) /* cnt > 0 ? -> process next vertex */
436 #if defined (__ELF__) && defined (__linux__)
437 .section .note.GNU-stack,"",%progbits