2 * Mesa 3-D graphics library
5 * Copyright (C) 1999-2007 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.
26 * Keith Whitwell <keith@tungstengraphics.com>
29 #include "main/glheader.h"
30 #include "main/context.h"
31 #include "main/imports.h"
32 #include "main/mtypes.h"
34 #include "t_context.h"
35 #include "t_pipeline.h"
36 #include "t_vp_build.h"
39 void _tnl_install_pipeline( struct gl_context
*ctx
,
40 const struct tnl_pipeline_stage
**stages
)
42 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
45 tnl
->pipeline
.new_state
= ~0;
47 /* Create a writeable copy of each stage.
49 for (i
= 0 ; i
< MAX_PIPELINE_STAGES
&& stages
[i
] ; i
++) {
50 struct tnl_pipeline_stage
*s
= &tnl
->pipeline
.stages
[i
];
51 memcpy(s
, stages
[i
], sizeof(*s
));
56 tnl
->pipeline
.nr_stages
= i
;
59 void _tnl_destroy_pipeline( struct gl_context
*ctx
)
61 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
64 for (i
= 0 ; i
< tnl
->pipeline
.nr_stages
; i
++) {
65 struct tnl_pipeline_stage
*s
= &tnl
->pipeline
.stages
[i
];
70 tnl
->pipeline
.nr_stages
= 0;
75 static GLuint
check_input_changes( struct gl_context
*ctx
)
77 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
80 for (i
= 0; i
<= _TNL_LAST_MAT
; i
++) {
81 if (tnl
->vb
.AttribPtr
[i
]->size
!= tnl
->pipeline
.last_attrib_size
[i
] ||
82 tnl
->vb
.AttribPtr
[i
]->stride
!= tnl
->pipeline
.last_attrib_stride
[i
]) {
83 tnl
->pipeline
.last_attrib_size
[i
] = tnl
->vb
.AttribPtr
[i
]->size
;
84 tnl
->pipeline
.last_attrib_stride
[i
] = tnl
->vb
.AttribPtr
[i
]->stride
;
85 tnl
->pipeline
.input_changes
|= 1<<i
;
89 return tnl
->pipeline
.input_changes
;
93 static GLuint
check_output_changes( struct gl_context
*ctx
)
96 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
98 for (i
= 0; i
< VARYING_SLOT_MAX
; i
++) {
99 if (tnl
->vb
.ResultPtr
[i
]->size
!= tnl
->last_result_size
[i
] ||
100 tnl
->vb
.ResultPtr
[i
]->stride
!= tnl
->last_result_stride
[i
]) {
101 tnl
->last_result_size
[i
] = tnl
->vb
.ResultPtr
[i
]->size
;
102 tnl
->last_result_stride
[i
] = tnl
->vb
.ResultPtr
[i
]->stride
;
103 tnl
->pipeline
.output_changes
|= 1<<i
;
107 if (tnl
->pipeline
.output_changes
)
108 tnl
->Driver
.NotifyOutputChanges( ctx
, tnl
->pipeline
.output_changes
);
110 return tnl
->pipeline
.output_changes
;
117 void _tnl_run_pipeline( struct gl_context
*ctx
)
119 TNLcontext
*tnl
= TNL_CONTEXT(ctx
);
120 unsigned short __tmp
;
126 /* Check for changed input sizes or change in stride to/from zero
127 * (ie const or non-const).
129 if (check_input_changes( ctx
) || tnl
->pipeline
.new_state
) {
130 if (ctx
->VertexProgram
._MaintainTnlProgram
)
131 _tnl_UpdateFixedFunctionProgram( ctx
);
133 for (i
= 0; i
< tnl
->pipeline
.nr_stages
; i
++) {
134 struct tnl_pipeline_stage
*s
= &tnl
->pipeline
.stages
[i
];
136 s
->validate( ctx
, s
);
139 tnl
->pipeline
.new_state
= 0;
140 tnl
->pipeline
.input_changes
= 0;
142 /* Pipeline can only change its output in response to either a
143 * statechange or an input size/stride change. No other changes
146 if (check_output_changes( ctx
))
147 _tnl_notify_pipeline_output_change( ctx
);
151 /* Don't adjust FPU precision mode in case multiple threads are to be used.
152 * This would require that the additional threads also changed the FPU mode
153 * which is quite a mess as this had to be done in all parallelized sections;
154 * otherwise the master thread and all other threads are running in different
155 * modes, producing inconsistent results.
156 * Note that all x64 implementations don't define/use START_FAST_MATH, so
157 * this is "hack" is only used in i386 mode
159 START_FAST_MATH(__tmp
);
162 for (i
= 0; i
< tnl
->pipeline
.nr_stages
; i
++) {
163 struct tnl_pipeline_stage
*s
= &tnl
->pipeline
.stages
[i
];
164 if (!s
->run( ctx
, s
))
169 END_FAST_MATH(__tmp
);
175 /* The default pipeline. This is useful for software rasterizers, and
176 * simple hardware rasterizers. For customization, I don't recommend
177 * tampering with the internals of these stages in the way that
178 * drivers did in Mesa 3.4. These stages are basically black boxes,
179 * and should be left intact.
181 * To customize the pipeline, consider:
183 * - removing redundant stages (making sure that the software rasterizer
184 * can cope with this on fallback paths). An example is fog
185 * coordinate generation, which is not required in the FX driver.
187 * - replacing general-purpose machine-independent stages with
188 * general-purpose machine-specific stages. There is no example of
189 * this to date, though it must be borne in mind that all subsequent
190 * stages that reference the output of the new stage must cope with
191 * any machine-specific data introduced. This may not be easy
192 * unless there are no such stages (ie the new stage is the last in
195 * - inserting optimized (but specialized) stages ahead of the
196 * general-purpose fallback implementation. For example, the old
197 * fastpath mechanism, which only works when the VB->Elts input is
198 * available, can be duplicated by placing the fastpath stage at the
199 * head of this pipeline. Such specialized stages are currently
200 * constrained to have no outputs (ie. they must either finish the *
201 * pipeline by returning GL_FALSE from run(), or do nothing).
203 * Some work can be done to lift some of the restrictions in the final
204 * case, if it becomes necessary to do so.
206 const struct tnl_pipeline_stage
*_tnl_default_pipeline
[] = {
207 &_tnl_vertex_transform_stage
,
208 &_tnl_normal_transform_stage
,
209 &_tnl_lighting_stage
,
211 &_tnl_texture_transform_stage
,
212 &_tnl_point_attenuation_stage
,
213 &_tnl_vertex_program_stage
,
214 &_tnl_fog_coordinate_stage
,
219 const struct tnl_pipeline_stage
*_tnl_vp_pipeline
[] = {
220 &_tnl_vertex_program_stage
,