Subversion Repositories shark

Rev

Details | Last modification | View Log | RSS feed

Rev Author Line No. Line
70 giacomo 1
/* $Id: sse_xform4.s,v 1.1 2003-03-13 12:11:49 giacomo Exp $ */
2
 
3
/*
4
 * Mesa 3-D graphics library
5
 * Version:  3.5
6
 *
7
 * Copyright (C) 1999-2001  Brian Paul   All Rights Reserved.
8
 *
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:
15
 *
16
 * The above copyright notice and this permission notice shall be included
17
 * in all copies or substantial portions of the Software.
18
 *
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.
25
 */
26
 
27
#include "matypes.h"
28
#include "xform_args.h"
29
 
30
	SEG_TEXT
31
 
32
#define FRAME_OFFSET	8
33
 
34
#define SRC(i)		REGOFF(i * 4, ESI)
35
#define DST(i)		REGOFF(i * 4, EDI)
36
#define MAT(i)		REGOFF(i * 4, EDX)
37
 
38
#define SELECT(r0, r1, r2, r3)	CONST( r0 * 64 + r1 * 16 + r2 * 4 + r3 )
39
 
40
 
41
ALIGNTEXT16
42
GLOBL GLNAME( _mesa_sse_transform_points4_general )
43
GLNAME( _mesa_sse_transform_points4_general ):
44
 
45
	PUSH_L( ESI )
46
	PUSH_L( EDI )
47
 
48
	MOV_L( ARG_SOURCE, ESI )
49
	MOV_L( ARG_DEST, EDI )
50
 
51
	MOV_L( ARG_MATRIX, EDX )
52
	MOV_L( REGOFF(V4F_COUNT, ESI), ECX )
53
 
54
	TEST_L( ECX, ECX )			/* verify non-zero count */
55
	JE( LLBL( sse_general_done ) )
56
 
57
	MOV_L( REGOFF(V4F_STRIDE, ESI), EAX )	/* stride */
58
	OR_L( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, EDI) ) /* set dest flags */
59
 
60
	MOV_L( ECX, REGOFF(V4F_COUNT, EDI) )	/* set dest count */
61
	MOV_L( CONST(4), REGOFF(V4F_SIZE, EDI) )/* set dest size */
62
 
63
	MOV_L( REGOFF(V4F_START, ESI), ESI )	/* ptr to first source vertex */
64
	MOV_L( REGOFF(V4F_START, EDI), EDI )	/* ptr to first dest vertex */
65
 
66
	PREFETCHT0( REGIND(ESI) )
67
 
68
	MOVAPS( MAT(0), XMM4 )			/* m3  | m2  | m1  | m0  */
69
	MOVAPS( MAT(4), XMM5 )			/* m7  | m6  | m5  | m4  */
70
	MOVAPS( MAT(8), XMM6 )			/* m11 | m10 | m9  | m8  */
71
	MOVAPS( MAT(12), XMM7 )			/* m15 | m14 | m13 | m12 */
72
 
73
ALIGNTEXT16
74
LLBL( sse_general_loop ):
75
 
76
	MOVSS( SRC(0), XMM0 )			/* ox */
77
	SHUFPS( CONST(0x0), XMM0, XMM0 )	/* ox | ox | ox | ox */
78
	MULPS( XMM4, XMM0 )			/* ox*m3 | ox*m2 | ox*m1 | ox*m0 */
79
 
80
	MOVSS( SRC(1), XMM1 )			/* oy */
81
	SHUFPS( CONST(0x0), XMM1, XMM1 )	/* oy | oy | oy | oy */
82
	MULPS( XMM5, XMM1 )			/* oy*m7 | oy*m6 | oy*m5 | oy*m4 */
83
 
84
	MOVSS( SRC(2), XMM2 )			/* oz */
85
	SHUFPS( CONST(0x0), XMM2, XMM2 )	/* oz | oz | oz | oz */
86
	MULPS( XMM6, XMM2 )			/* oz*m11 | oz*m10 | oz*m9 | oz*m8 */
87
 
88
	MOVSS( SRC(3), XMM3 )			/* ow */
89
	SHUFPS( CONST(0x0), XMM3, XMM3 )	/* ow | ow | ow | ow */
90
	MULPS( XMM7, XMM3 )			/* ow*m15 | ow*m14 | ow*m13 | ow*m12 */
91
 
92
	ADDPS( XMM1, XMM0 )			/* ox*m3+oy*m7 | ... */
93
	ADDPS( XMM2, XMM0 )			/* ox*m3+oy*m7+oz*m11 | ... */
94
	ADDPS( XMM3, XMM0 )			/* ox*m3+oy*m7+oz*m11+ow*m15 | ... */
95
	MOVAPS( XMM0, DST(0) )			/* ->D(3) | ->D(2) | ->D(1) | ->D(0) */
96
 
97
	ADD_L( CONST(16), EDI )
98
	ADD_L( EAX, ESI )
99
 
100
	DEC_L( ECX )
101
	JNZ( LLBL( sse_general_loop ) )
102
 
103
LLBL( sse_general_done ):
104
 
105
	POP_L( EDI )
106
	POP_L( ESI )
107
	RET
108
 
109
 
110
 
111
 
112
ALIGNTEXT4
113
GLOBL GLNAME( _mesa_sse_transform_points4_3d )
114
GLNAME( _mesa_sse_transform_points4_3d ):
115
 
116
	PUSH_L( ESI )
117
	PUSH_L( EDI )
118
 
119
	MOV_L( ARG_SOURCE, ESI )		/* ptr to source GLvector4f */
120
	MOV_L( ARG_DEST, EDI )			/* ptr to dest GLvector4f */
121
 
122
	MOV_L( ARG_MATRIX, EDX )		/* ptr to matrix */
123
	MOV_L( REGOFF(V4F_COUNT, ESI), ECX )	/* source count */
124
 
125
	TEST_L( ECX, ECX)
126
	JZ( LLBL(K_GTP43P3DR_finish) )		/* count was zero; go to finish */
127
 
128
	MOV_L( REGOFF(V4F_STRIDE, ESI), EAX )	/* stride */
129
	OR_L( CONST(VEC_SIZE_3), REGOFF(V4F_FLAGS, EDI) ) /* set dest flags */
130
 
131
	MOV_L( ECX, REGOFF(V4F_COUNT, EDI) )	/* set dest count */
132
	MOV_L( CONST(3), REGOFF(V4F_SIZE, EDI) )/* set dest size */
133
 
134
	SHL_L( CONST(4), ECX )			/* count *= 16 */
135
	MOV_L( REGOFF(V4F_START, ESI), ESI )	/* ptr to first source vertex */
136
 
137
	MOV_L( REGOFF(V4F_START, EDI), EDI )	/* ptr to first dest vertex */
138
	ADD_L( EDI, ECX )			/* count += dest ptr */
139
 
140
	MOVAPS( MAT(0), XMM0 )			/* m3  | m2  | m1  |  m0 */
141
	MOVAPS( MAT(4), XMM1 )			/* m7  | m6  | m5  |  m4 */
142
	MOVAPS( MAT(8), XMM2 )			/* m11 | m10 | m9  |  m8 */
143
	MOVAPS( MAT(12), XMM3 )			/* m15 | m14 | m13 | m12 */
144
 
145
ALIGNTEXT32
146
LLBL( K_GTP43P3DR_top ):
147
	MOVSS( SRC(0), XMM4 )			/* ox */
148
	SHUFPS( CONST(0x0), XMM4, XMM4 )	/* ox | ox | ox | ox */
149
	MULPS( XMM0, XMM4 )			/* ox*m3 | ox*m2 | ox*m1 | ox*m0 */
150
 
151
	MOVSS( SRC(1), XMM5 )			/* oy */
152
	SHUFPS( CONST(0x0), XMM5, XMM5 )	/* oy | oy | oy | oy */
153
	MULPS( XMM1, XMM5 )			/* oy*m7 | oy*m6 | oy*m5 | oy*m4 */
154
 
155
	MOVSS( SRC(2), XMM6 )			/* oz */
156
	SHUFPS( CONST(0x0), XMM6, XMM6 )	/* oz | oz | oz | oz */
157
	MULPS( XMM2, XMM6 )			/* oz*m11 | oz*m10 | oz*m9 | oz*m8 */
158
 
159
	MOVSS( SRC(3), XMM7 )			/* ow */
160
	SHUFPS( CONST(0x0), XMM7, XMM7 )	/* ow | ow | ow | ow */
161
	MULPS( XMM3, XMM7 )			/* ow*m15 | ow*m14 | ow*m13 | ow*m12 */
162
 
163
	ADDPS( XMM5, XMM4 )			/* ox*m3+oy*m7 | ... */
164
	ADDPS( XMM6, XMM4 )			/* ox*m3+oy*m7+oz*m11 | ... */
165
	ADDPS( XMM7, XMM4 )			/* ox*m3+oy*m7+oz*m11+ow*m15 | ... */
166
	MOVAPS( XMM4, DST(0) )			/* ->D(3) | ->D(2) | ->D(1) | ->D(0) */
167
 
168
	MOVSS( SRC(3), XMM4 )			/* ow */
169
	MOVSS( XMM4, DST(3) )			/* ->D(3) */
170
 
171
LLBL( K_GTP43P3DR_skip ):
172
	ADD_L( CONST(16), EDI )
173
	ADD_L( EAX, ESI )
174
	CMP_L( ECX, EDI )
175
	JNE( LLBL(K_GTP43P3DR_top) )
176
 
177
LLBL( K_GTP43P3DR_finish ):
178
	POP_L( EDI )
179
	POP_L( ESI )
180
	RET
181
 
182
 
183
ALIGNTEXT16
184
GLOBL GLNAME( _mesa_sse_transform_points4_identity )
185
GLNAME( _mesa_sse_transform_points4_identity ):
186
 
187
	PUSH_L( ESI )
188
	PUSH_L( EDI )
189
 
190
	MOV_L( ARG_SOURCE, ESI )
191
	MOV_L( ARG_DEST, EDI )
192
 
193
	MOV_L( ARG_MATRIX, EDX )
194
	MOV_L( REGOFF(V4F_COUNT, ESI), ECX )
195
 
196
	TEST_L( ECX, ECX )			/* verify non-zero count */
197
	JE( LLBL( sse_identity_done ) )
198
 
199
	MOV_L( REGOFF(V4F_STRIDE, ESI), EAX )	/* stride */
200
	OR_L( CONST(VEC_SIZE_4), REGOFF(V4F_FLAGS, EDI) ) /* set dest flags */
201
 
202
	MOV_L( ECX, REGOFF(V4F_COUNT, EDI) )	/* set dest count */
203
	MOV_L( CONST(4), REGOFF(V4F_SIZE, EDI) )/* set dest size */
204
 
205
	MOV_L( REGOFF(V4F_START, ESI), ESI )	/* ptr to first source vertex */
206
	MOV_L( REGOFF(V4F_START, EDI), EDI )	/* ptr to first dest vertex */
207
 
208
ALIGNTEXT16
209
LLBL( sse_identity_loop ):
210
 
211
	PREFETCHNTA( REGOFF(32, ESI) )
212
 
213
	MOVAPS( REGIND(ESI), XMM0 )
214
	ADD_L( EAX, ESI )
215
 
216
	MOVAPS( XMM0, REGIND(EDI) )
217
	ADD_L( CONST(16), EDI )
218
 
219
	DEC_L( ECX )
220
	JNZ( LLBL( sse_identity_loop ) )
221
 
222
LLBL( sse_identity_done ):
223
 
224
	POP_L( EDI )
225
	POP_L( ESI )
226
	RET