1 /* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_vtxfmt_sse.c,v 1.1 2002/10/30 12:51:58 alanh Exp $ */
2 /**************************************************************************
4 Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
5 Tungsten Graphics Inc., Cedar Park, Texas.
9 Permission is hereby granted, free of charge, to any person obtaining
10 a copy of this software and associated documentation files (the
11 "Software"), to deal in the Software without restriction, including
12 without limitation the rights to use, copy, modify, merge, publish,
13 distribute, sublicense, and/or sell copies of the Software, and to
14 permit persons to whom the Software is furnished to do so, subject to
15 the following conditions:
17 The above copyright notice and this permission notice (including the
18 next paragraph) shall be included in all copies or substantial
19 portions of the Software.
21 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
22 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
23 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
24 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
25 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
26 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
27 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
29 **************************************************************************/
33 * Keith Whitwell <keith@tungstengraphics.com>
38 #include "simple_list.h"
39 #include "radeon_vtxfmt.h"
41 #if defined(USE_SSE_ASM)
42 #include "x86/common_x86_asm.h"
44 #define EXTERN( FUNC ) \
45 extern const char *FUNC; \
46 extern const char *FUNC##_end
48 EXTERN( _sse_Attribute2fv
);
49 EXTERN( _sse_Attribute2f
);
50 EXTERN( _sse_Attribute3fv
);
51 EXTERN( _sse_Attribute3f
);
52 EXTERN( _sse_MultiTexCoord2fv
);
53 EXTERN( _sse_MultiTexCoord2f
);
54 EXTERN( _sse_MultiTexCoord2fv_2
);
55 EXTERN( _sse_MultiTexCoord2f_2
);
57 /* Build specialized versions of the immediate calls on the fly for
61 static struct dynfn
*radeon_makeSSEAttribute2fv( struct dynfn
* cache
, int key
,
62 const char * name
, void * dest
)
64 struct dynfn
*dfn
= MALLOC_STRUCT( dynfn
);
66 if (RADEON_DEBUG
& DEBUG_CODEGEN
)
67 fprintf(stderr
, "%s 0x%08x\n", name
, key
);
69 DFN ( _sse_Attribute2fv
, (*cache
) );
70 FIXUP(dfn
->code
, 10, 0x0, (int)dest
);
74 static struct dynfn
*radeon_makeSSEAttribute2f( struct dynfn
* cache
, int key
,
75 const char * name
, void * dest
)
77 struct dynfn
*dfn
= MALLOC_STRUCT( dynfn
);
79 if (RADEON_DEBUG
& DEBUG_CODEGEN
)
80 fprintf(stderr
, "%s 0x%08x\n", name
, key
);
82 DFN ( _sse_Attribute2f
, (*cache
) );
83 FIXUP(dfn
->code
, 8, 0x0, (int)dest
);
87 static struct dynfn
*radeon_makeSSEAttribute3fv( struct dynfn
* cache
, int key
,
88 const char * name
, void * dest
)
90 struct dynfn
*dfn
= MALLOC_STRUCT( dynfn
);
92 if (RADEON_DEBUG
& DEBUG_CODEGEN
)
93 fprintf(stderr
, "%s 0x%08x\n", name
, key
);
95 DFN ( _sse_Attribute3fv
, (*cache
) );
96 FIXUP(dfn
->code
, 13, 0x0, (int)dest
);
97 FIXUP(dfn
->code
, 18, 0x8, 8+(int)dest
);
101 static struct dynfn
*radeon_makeSSEAttribute3f( struct dynfn
* cache
, int key
,
102 const char * name
, void * dest
)
104 struct dynfn
*dfn
= MALLOC_STRUCT( dynfn
);
106 if (RADEON_DEBUG
& DEBUG_CODEGEN
)
107 fprintf(stderr
, "%s 0x%08x\n", name
, key
);
109 DFN ( _sse_Attribute3f
, (*cache
) );
110 FIXUP(dfn
->code
, 12, 0x0, (int)dest
);
111 FIXUP(dfn
->code
, 17, 0x8, 8+(int)dest
);
115 static struct dynfn
* radeon_makeSSENormal3fv( GLcontext
*ctx
, int key
)
117 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
119 return radeon_makeSSEAttribute3fv( & rmesa
->vb
.dfn_cache
.Normal3fv
, key
,
120 __FUNCTION__
, rmesa
->vb
.normalptr
);
123 static struct dynfn
*radeon_makeSSENormal3f( GLcontext
*ctx
, int key
)
125 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
127 return radeon_makeSSEAttribute3f( & rmesa
->vb
.dfn_cache
.Normal3f
, key
,
128 __FUNCTION__
, rmesa
->vb
.normalptr
);
131 static struct dynfn
*radeon_makeSSEColor3fv( GLcontext
*ctx
, int key
)
133 if (key
& (RADEON_CP_VC_FRMT_PKCOLOR
|RADEON_CP_VC_FRMT_FPALPHA
))
137 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
139 return radeon_makeSSEAttribute3fv( & rmesa
->vb
.dfn_cache
.Color3fv
, key
,
140 __FUNCTION__
, rmesa
->vb
.floatcolorptr
);
144 static struct dynfn
*radeon_makeSSEColor3f( GLcontext
*ctx
, int key
)
146 if (key
& (RADEON_CP_VC_FRMT_PKCOLOR
|RADEON_CP_VC_FRMT_FPALPHA
))
150 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
152 return radeon_makeSSEAttribute3f( & rmesa
->vb
.dfn_cache
.Color3f
, key
,
153 __FUNCTION__
, rmesa
->vb
.floatcolorptr
);
157 static struct dynfn
*radeon_makeSSETexCoord2fv( GLcontext
*ctx
, int key
)
159 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
161 return radeon_makeSSEAttribute2fv( & rmesa
->vb
.dfn_cache
.TexCoord2fv
, key
,
162 __FUNCTION__
, rmesa
->vb
.texcoordptr
[0] );
165 static struct dynfn
*radeon_makeSSETexCoord2f( GLcontext
*ctx
, int key
)
167 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
169 return radeon_makeSSEAttribute2f( & rmesa
->vb
.dfn_cache
.TexCoord2f
, key
,
170 __FUNCTION__
, rmesa
->vb
.texcoordptr
[0] );
173 static struct dynfn
*radeon_makeSSEMultiTexCoord2fv( GLcontext
*ctx
, int key
)
175 struct dynfn
*dfn
= MALLOC_STRUCT( dynfn
);
176 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
178 if (RADEON_DEBUG
& DEBUG_CODEGEN
)
179 fprintf(stderr
, "%s 0x%08x\n", __FUNCTION__
, key
);
181 if ((key
& (RADEON_CP_VC_FRMT_ST0
|RADEON_CP_VC_FRMT_ST1
)) ==
182 (RADEON_CP_VC_FRMT_ST0
|RADEON_CP_VC_FRMT_ST1
)) {
183 DFN ( _sse_MultiTexCoord2fv
, rmesa
->vb
.dfn_cache
.MultiTexCoord2fvARB
);
184 FIXUP(dfn
->code
, 18, 0xdeadbeef, (int)rmesa
->vb
.texcoordptr
[0]);
186 DFN ( _sse_MultiTexCoord2fv_2
, rmesa
->vb
.dfn_cache
.MultiTexCoord2fvARB
);
187 FIXUP(dfn
->code
, 14, 0x0, (int)rmesa
->vb
.texcoordptr
);
192 static struct dynfn
*radeon_makeSSEMultiTexCoord2f( GLcontext
*ctx
, int key
)
194 struct dynfn
*dfn
= MALLOC_STRUCT( dynfn
);
195 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
197 if (RADEON_DEBUG
& DEBUG_CODEGEN
)
198 fprintf(stderr
, "%s 0x%08x\n", __FUNCTION__
, key
);
200 if ((key
& (RADEON_CP_VC_FRMT_ST0
|RADEON_CP_VC_FRMT_ST1
)) ==
201 (RADEON_CP_VC_FRMT_ST0
|RADEON_CP_VC_FRMT_ST1
)) {
202 DFN ( _sse_MultiTexCoord2f
, rmesa
->vb
.dfn_cache
.MultiTexCoord2fARB
);
203 FIXUP(dfn
->code
, 16, 0xdeadbeef, (int)rmesa
->vb
.texcoordptr
[0]);
205 DFN ( _sse_MultiTexCoord2f_2
, rmesa
->vb
.dfn_cache
.MultiTexCoord2fARB
);
206 FIXUP(dfn
->code
, 15, 0x0, (int)rmesa
->vb
.texcoordptr
);
211 void radeonInitSSECodegen( struct dfn_generators
*gen
)
214 gen
->Normal3fv
= (void *) radeon_makeSSENormal3fv
;
215 gen
->Normal3f
= (void *) radeon_makeSSENormal3f
;
216 gen
->Color3fv
= (void *) radeon_makeSSEColor3fv
;
217 gen
->Color3f
= (void *) radeon_makeSSEColor3f
;
218 gen
->TexCoord2fv
= (void *) radeon_makeSSETexCoord2fv
;
219 gen
->TexCoord2f
= (void *) radeon_makeSSETexCoord2f
;
220 gen
->MultiTexCoord2fvARB
= (void *) radeon_makeSSEMultiTexCoord2fv
;
221 gen
->MultiTexCoord2fARB
= (void *) radeon_makeSSEMultiTexCoord2f
;
227 void radeonInitSSECodegen( struct dfn_generators
*gen
)