Subversion Repositories Kolibri OS

Rev

Rev 4358 | Go to most recent revision | Details | Compare with Previous | Last modification | View Log | RSS feed

Rev Author Line No. Line
4358 Serge 1
 
2
 * Mesa 3-D graphics library
3
 *
4
 * Copyright (C) 1999-2001  Brian Paul   All Rights Reserved.
5
 *
6
 * Permission is hereby granted, free of charge, to any person obtaining a
7
 * copy of this software and associated documentation files (the "Software"),
8
 * to deal in the Software without restriction, including without limitation
9
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10
 * and/or sell copies of the Software, and to permit persons to whom the
11
 * Software is furnished to do so, subject to the following conditions:
12
 *
13
 * The above copyright notice and this permission notice shall be included
14
 * in all copies or substantial portions of the Software.
15
 *
16
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
17
 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
20
 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21
 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22
 * OTHER DEALINGS IN THE SOFTWARE.
23
 */
24
25
 
26
  * - insert PREFETCH instructions to avoid cache-misses !
27
  * - some more optimizations are possible...
28
  * - for 40-50% more performance in the SSE-functions, the
29
  *   data (trans-matrix, src_vert, dst_vert) needs to be 16byte aligned !
30
  */
31
32
 
33
#include "assyntax.h"
34
#include "matypes.h"
35
#include "norm_args.h"
36
37
 
38
39
 
40
#define S(i)	REGOFF(i * 4, ESI)
41
#define D(i)	REGOFF(i * 4, EDI)
42
#define STRIDE  REGOFF(12, ESI)
43
44
 
45
 
46
GLOBL GLNAME(_mesa_sse_transform_rescale_normals_no_rot)
4632 Serge 47
HIDDEN(_mesa_sse_transform_rescale_normals_no_rot)
48
GLNAME(_mesa_sse_transform_rescale_normals_no_rot):
49
4358 Serge 50
 
51
	PUSH_L  ( ESI )
52
	PUSH_L  ( EDI )
53
54
 
55
	MOV_L	( ARG_DEST, EDI )			/* ptr to dest GLvector3f */
56
57
 
58
	MOV_L	( REGOFF(MATRIX_INV, EDX), EDX)		/* matrix->inv */
59
60
 
61
62
 
63
	JZ( LLBL(K_G3TRNNRR_finish) )			/* count was zero; go to finish */
64
65
 
66
	MOV_L	( ECX, REGOFF(V4F_COUNT, EDI) )		/* set dest-count */
67
68
 
69
	MOV_L( REGOFF(V4F_START, ESI), ESI )		/* ptr to first source vertex */
70
71
 
72
	ADD_L( EDI, ECX ) 				/* count += dest ptr */
73
74
 
75
	MOVSS	( M(0), XMM1 )				/* m0 */
76
	MOVSS	( M(5), XMM2 )				/* m5 */
77
	UNPCKLPS( XMM2, XMM1 )				/* m5 | m0 */
78
	MOVSS	( ARG_SCALE, XMM0 )			/* scale */
79
	SHUFPS	( CONST(0x0), XMM0, XMM0 )		/* scale | scale */
80
	MULPS	( XMM0, XMM1 )				/* m5*scale | m0*scale */
81
	MULSS	( M(10), XMM0 )				/* m10*scale */
82
83
 
84
LLBL(K_G3TRNNRR_top):
85
	MOVLPS	( S(0), XMM2 )				/* uy | ux */
86
	MULPS	( XMM1, XMM2 )				/* uy*m5*scale | ux*m0*scale */
87
	MOVLPS	( XMM2, D(0) )				/* ->D(1) | D(0) */
88
89
 
90
	MULSS	( XMM0, XMM2 )				/* uz*m10*scale */
91
	MOVSS	( XMM2, D(2) )				/* ->D(2) */
92
93
 
94
	ADD_L	( CONST(16), EDI )
95
	ADD_L	( EAX, ESI )
96
	CMP_L	( ECX, EDI )
97
	JNE	( LLBL(K_G3TRNNRR_top) )
98
99
 
100
	POP_L	( EDI )
101
	POP_L	( ESI )
102
	RET
103
#undef FRAME_OFFSET
104
105
 
106
 
107
 
108
GLOBL GLNAME(_mesa_sse_transform_rescale_normals)
4632 Serge 109
HIDDEN(_mesa_sse_transform_rescale_normals)
110
GLNAME(_mesa_sse_transform_rescale_normals):
111
4358 Serge 112
 
113
	PUSH_L  ( ESI )
114
	PUSH_L  ( EDI )
115
116
 
117
	MOV_L	( ARG_DEST, EDI )			/* ptr to dest GLvector3f */
118
119
 
120
	MOV_L	( REGOFF(MATRIX_INV, EDX), EDX)		/* matrix->inv */
121
122
 
123
124
 
125
	JZ( LLBL(K_G3TRNR_finish) )			/* count was zero; go to finish */
126
127
 
128
	MOV_L	( ECX, REGOFF(V4F_COUNT, EDI) )		/* set dest-count */
129
130
 
131
	MOV_L( REGOFF(V4F_START, ESI), ESI )		/* ptr to first source vertex */
132
133
 
134
	ADD_L( EDI, ECX ) 				/* count += dest ptr */
135
136
 
137
	MOVSS	( M(0), XMM0 )				/* m0 */
138
	MOVSS	( M(4), XMM1 )				/* m4 */
139
	UNPCKLPS( XMM1, XMM0 )				/* m4 | m0 */
140
141
 
142
	SHUFPS	( CONST(0x0), XMM4, XMM4 )		/* scale | scale */
143
144
 
145
	MOVSS	( M(1), XMM1 )				/* m1 */
146
	MOVSS	( M(5), XMM2 )				/* m5 */
147
	UNPCKLPS( XMM2, XMM1 )				/* m5 | m1 */
148
	MULPS	( XMM4, XMM1 )				/* m5*scale | m1*scale */
149
	MOVSS	( M(2), XMM2 )				/* m2 */
150
	MOVSS	( M(6), XMM3 )				/* m6 */
151
	UNPCKLPS( XMM3, XMM2 )				/* m6 | m2 */
152
	MULPS	( XMM4, XMM2 )				/* m6*scale | m2*scale */
153
154
 
155
	MULSS	( ARG_SCALE, XMM6 )			/* m8*scale */
156
	MOVSS	( M(9), XMM7 )				/* m9 */
157
	MULSS	( ARG_SCALE, XMM7 )			/* m9*scale */
158
159
 
160
LLBL(K_G3TRNR_top):
161
	MOVSS	( S(0), XMM3 )				/* ux */
162
	SHUFPS	( CONST(0x0), XMM3, XMM3 )		/* ux | ux */
163
	MULPS	( XMM0, XMM3 )				/* ux*m4 | ux*m0 */
164
	MOVSS	( S(1), XMM4 )				/* uy */
165
	SHUFPS	( CONST(0x0), XMM4, XMM4 )		/* uy | uy */
166
	MULPS	( XMM1, XMM4 )				/* uy*m5 | uy*m1 */
167
	MOVSS	( S(2), XMM5 )				/* uz */
168
	SHUFPS	( CONST(0x0), XMM5, XMM5 )		/* uz | uz */
169
	MULPS	( XMM2, XMM5 )				/* uz*m6 | uz*m2 */
170
171
 
172
	ADDPS	( XMM5, XMM3 )
173
	MOVLPS	( XMM3, D(0) )
174
175
 
176
	MULSS	( ARG_SCALE, XMM3 )			/* m10*scale */
177
	MULSS	( S(2), XMM3 )				/* m10*scale*uz */
178
	MOVSS	( S(1), XMM4 )				/* uy */
179
	MULSS	( XMM7, XMM4 )				/* uy*m9*scale */
180
	MOVSS	( S(0), XMM5 )				/* ux */
181
	MULSS	( XMM6, XMM5 )				/* ux*m8*scale */
182
183
 
184
	ADDSS	( XMM5, XMM3 )
185
	MOVSS	( XMM3, D(2) )
186
187
 
188
	ADD_L	( CONST(16), EDI )
189
	ADD_L	( EAX, ESI )
190
	CMP_L	( ECX, EDI )
191
	JNE	( LLBL(K_G3TRNR_top) )
192
193
 
194
	POP_L	( EDI )
195
	POP_L	( ESI )
196
	RET
197
#undef FRAME_OFFSET
198
199
 
200
 
201
GLOBL GLNAME(_mesa_sse_transform_normals_no_rot)
4632 Serge 202
HIDDEN(_mesa_sse_transform_normals_no_rot)
203
GLNAME(_mesa_sse_transform_normals_no_rot):
204
4358 Serge 205
 
206
	PUSH_L  ( ESI )
207
	PUSH_L  ( EDI )
208
209
 
210
	MOV_L	( ARG_DEST, EDI )			/* ptr to dest GLvector3f */
211
212
 
213
	MOV_L	( REGOFF(MATRIX_INV, EDX), EDX)		/* matrix->inv */
214
215
 
216
217
 
218
	JZ( LLBL(K_G3TNNRR_finish) )			/* count was zero; go to finish */
219
220
 
221
	MOV_L	( ECX, REGOFF(V4F_COUNT, EDI) )		/* set dest-count */
222
223
 
224
	MOV_L( REGOFF(V4F_START, ESI), ESI )		/* ptr to first source vertex */
225
226
 
227
	ADD_L( EDI, ECX ) 				/* count += dest ptr */
228
229
 
230
	MOVSS( M(0), XMM0 )				/* m0 */
231
	MOVSS( M(5), XMM1 )				/* m5 */
232
	UNPCKLPS( XMM1, XMM0 )				/* m5 | m0 */
233
	MOVSS( M(10), XMM1 )				/* m10 */
234
235
 
236
LLBL(K_G3TNNRR_top):
237
	MOVLPS( S(0), XMM2 )				/* uy | ux */
238
	MULPS( XMM0, XMM2 )				/* uy*m5 | ux*m0 */
239
	MOVLPS( XMM2, D(0) )
240
241
 
242
	MULSS( XMM1, XMM2 )				/* uz*m10 */
243
	MOVSS( XMM2, D(2) )
244
245
 
246
	ADD_L	( CONST(16), EDI )
247
	ADD_L	( EAX, ESI )
248
	CMP_L	( ECX, EDI )
249
	JNE	( LLBL(K_G3TNNRR_top) )
250
251
 
252
	POP_L	( EDI )
253
	POP_L	( ESI )
254
	RET
255
#undef FRAME_OFFSET
256
#endif
257
258
 
259
	.section .note.GNU-stack,"",%progbits
260
#endif
261