2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **********************************************************************/
29 * Keith Whitwell <keith@tungstengraphics.com>
34 #include "brw_context.h"
35 #include "brw_state.h"
36 #include "brw_defines.h"
37 #include "main/macros.h"
38 #include "intel_fbo.h"
40 static void upload_sf_vp(struct brw_context
*brw
)
42 GLcontext
*ctx
= &brw
->intel
.ctx
;
43 const GLfloat depth_scale
= 1.0F
/ ctx
->DrawBuffer
->_DepthMaxF
;
44 struct brw_sf_viewport sfv
;
45 GLfloat y_scale
, y_bias
;
46 const GLboolean render_to_fbo
= (ctx
->DrawBuffer
->Name
!= 0);
47 const GLfloat
*v
= ctx
->Viewport
._WindowMap
.m
;
49 memset(&sfv
, 0, sizeof(sfv
));
57 y_bias
= ctx
->DrawBuffer
->Height
;
62 sfv
.viewport
.m00
= v
[MAT_SX
];
63 sfv
.viewport
.m11
= v
[MAT_SY
] * y_scale
;
64 sfv
.viewport
.m22
= v
[MAT_SZ
] * depth_scale
;
65 sfv
.viewport
.m30
= v
[MAT_TX
];
66 sfv
.viewport
.m31
= v
[MAT_TY
] * y_scale
+ y_bias
;
67 sfv
.viewport
.m32
= v
[MAT_TZ
] * depth_scale
;
71 /* The scissor only needs to handle the intersection of drawable and
72 * scissor rect. Clipping to the boundaries of static shared buffers
73 * for front/back/depth is covered by looping over cliprects in brw_draw.c.
75 * Note that the hardware's coordinates are inclusive, while Mesa's min is
76 * inclusive but max is exclusive.
79 /* texmemory: Y=0=bottom */
80 sfv
.scissor
.xmin
= ctx
->DrawBuffer
->_Xmin
;
81 sfv
.scissor
.xmax
= ctx
->DrawBuffer
->_Xmax
- 1;
82 sfv
.scissor
.ymin
= ctx
->DrawBuffer
->_Ymin
;
83 sfv
.scissor
.ymax
= ctx
->DrawBuffer
->_Ymax
- 1;
87 sfv
.scissor
.xmin
= ctx
->DrawBuffer
->_Xmin
;
88 sfv
.scissor
.xmax
= ctx
->DrawBuffer
->_Xmax
- 1;
89 sfv
.scissor
.ymin
= ctx
->DrawBuffer
->Height
- ctx
->DrawBuffer
->_Ymax
;
90 sfv
.scissor
.ymax
= ctx
->DrawBuffer
->Height
- ctx
->DrawBuffer
->_Ymin
- 1;
93 dri_bo_unreference(brw
->sf
.vp_bo
);
94 brw
->sf
.vp_bo
= brw_cache_data( &brw
->cache
, BRW_SF_VP
, &sfv
, NULL
, 0 );
97 const struct brw_tracked_state brw_sf_vp
= {
99 .mesa
= (_NEW_VIEWPORT
|
104 .prepare
= upload_sf_vp
107 struct brw_sf_unit_key
{
108 unsigned int total_grf
;
109 unsigned int urb_entry_read_length
;
111 unsigned int nr_urb_entries
, urb_size
, sfsize
;
113 GLenum front_face
, cull_face
;
115 unsigned line_smooth
:1;
116 unsigned point_sprite
:1;
117 unsigned point_attenuated
:1;
118 unsigned render_to_fbo
:1;
124 sf_unit_populate_key(struct brw_context
*brw
, struct brw_sf_unit_key
*key
)
126 GLcontext
*ctx
= &brw
->intel
.ctx
;
127 memset(key
, 0, sizeof(*key
));
129 /* CACHE_NEW_SF_PROG */
130 key
->total_grf
= brw
->sf
.prog_data
->total_grf
;
131 key
->urb_entry_read_length
= brw
->sf
.prog_data
->urb_read_length
;
133 /* BRW_NEW_URB_FENCE */
134 key
->nr_urb_entries
= brw
->urb
.nr_sf_entries
;
135 key
->urb_size
= brw
->urb
.vsize
;
136 key
->sfsize
= brw
->urb
.sfsize
;
138 key
->scissor
= ctx
->Scissor
.Enabled
;
139 key
->front_face
= ctx
->Polygon
.FrontFace
;
141 if (ctx
->Polygon
.CullFlag
)
142 key
->cull_face
= ctx
->Polygon
.CullFaceMode
;
144 key
->cull_face
= GL_NONE
;
146 key
->line_width
= ctx
->Line
.Width
;
147 key
->line_smooth
= ctx
->Line
.SmoothFlag
;
149 key
->point_sprite
= ctx
->Point
.PointSprite
;
150 key
->point_size
= ctx
->Point
.Size
;
151 key
->point_attenuated
= ctx
->Point
._Attenuated
;
153 key
->render_to_fbo
= brw
->intel
.ctx
.DrawBuffer
->Name
!= 0;
157 sf_unit_create_from_key(struct brw_context
*brw
, struct brw_sf_unit_key
*key
,
160 struct brw_sf_unit_state sf
;
163 memset(&sf
, 0, sizeof(sf
));
165 sf
.thread0
.grf_reg_count
= ALIGN(key
->total_grf
, 16) / 16 - 1;
166 sf
.thread0
.kernel_start_pointer
= brw
->sf
.prog_bo
->offset
>> 6; /* reloc */
168 sf
.thread1
.floating_point_mode
= BRW_FLOATING_POINT_NON_IEEE_754
;
170 sf
.thread3
.dispatch_grf_start_reg
= 3;
171 sf
.thread3
.urb_entry_read_offset
= 1;
172 sf
.thread3
.urb_entry_read_length
= key
->urb_entry_read_length
;
174 sf
.thread4
.nr_urb_entries
= key
->nr_urb_entries
;
175 sf
.thread4
.urb_entry_allocation_size
= key
->sfsize
- 1;
176 /* Each SF thread produces 1 PUE, and there can be up to 24 threads */
177 sf
.thread4
.max_threads
= MIN2(24, key
->nr_urb_entries
) - 1;
179 if (INTEL_DEBUG
& DEBUG_SINGLE_THREAD
)
180 sf
.thread4
.max_threads
= 0;
182 if (INTEL_DEBUG
& DEBUG_STATS
)
183 sf
.thread4
.stats_enable
= 1;
185 /* CACHE_NEW_SF_VP */
186 sf
.sf5
.sf_viewport_state_offset
= brw
->sf
.vp_bo
->offset
>> 5; /* reloc */
188 sf
.sf5
.viewport_transform
= 1;
195 if (key
->front_face
== GL_CCW
)
196 sf
.sf5
.front_winding
= BRW_FRONTWINDING_CCW
;
198 sf
.sf5
.front_winding
= BRW_FRONTWINDING_CW
;
200 /* The viewport is inverted for rendering to a FBO, and that inverts
201 * polygon front/back orientation.
203 sf
.sf5
.front_winding
^= key
->render_to_fbo
;
205 switch (key
->cull_face
) {
207 sf
.sf6
.cull_mode
= BRW_CULLMODE_FRONT
;
210 sf
.sf6
.cull_mode
= BRW_CULLMODE_BACK
;
212 case GL_FRONT_AND_BACK
:
213 sf
.sf6
.cull_mode
= BRW_CULLMODE_BOTH
;
216 sf
.sf6
.cull_mode
= BRW_CULLMODE_NONE
;
224 /* XXX use ctx->Const.Min/MaxLineWidth here */
225 sf
.sf6
.line_width
= CLAMP(key
->line_width
, 1.0, 5.0) * (1<<1);
227 sf
.sf6
.line_endcap_aa_region_width
= 1;
228 if (key
->line_smooth
)
229 sf
.sf6
.aa_enable
= 1;
230 else if (sf
.sf6
.line_width
<= 0x2)
231 sf
.sf6
.line_width
= 0;
234 key
->render_to_fbo
= brw
->intel
.ctx
.DrawBuffer
->Name
!= 0;
235 if (!key
->render_to_fbo
) {
236 /* Rendering to an OpenGL window */
237 sf
.sf6
.point_rast_rule
= BRW_RASTRULE_UPPER_RIGHT
;
240 /* If rendering to an FBO, the pixel coordinate system is
241 * inverted with respect to the normal OpenGL coordinate
242 * system, so BRW_RASTRULE_LOWER_RIGHT is correct.
243 * But this value is listed as "Reserved, but not seen as useful"
244 * in Intel documentation (page 212, "Point Rasterization Rule",
245 * section 7.4 "SF Pipeline State Summary", of document
246 * "IntelĀ® 965 Express Chipset Family and IntelĀ® G35 Express
247 * Chipset Graphics Controller Programmer's Reference Manual,
248 * Volume 2: 3D/Media", Revision 1.0b as of January 2008,
250 * http://intellinuxgraphics.org/documentation.html
251 * at the time of this writing).
253 * It does work on at least some devices, if not all;
254 * if devices that don't support it can be identified,
255 * the likely failure case is that points are rasterized
256 * incorrectly, which is no worse than occurs without
257 * the value, so we're using it here.
259 sf
.sf6
.point_rast_rule
= BRW_RASTRULE_LOWER_RIGHT
;
261 /* XXX clamp max depends on AA vs. non-AA */
263 sf
.sf7
.sprite_point
= key
->point_sprite
;
264 sf
.sf7
.point_size
= CLAMP(rint(key
->point_size
), 1, 255) * (1<<3);
265 sf
.sf7
.use_point_size_state
= !key
->point_attenuated
;
266 sf
.sf7
.aa_line_distance_mode
= 0;
268 /* might be BRW_NEW_PRIMITIVE if we have to adjust pv for polygons:
270 sf
.sf7
.trifan_pv
= 2;
271 sf
.sf7
.linestrip_pv
= 1;
272 sf
.sf7
.tristrip_pv
= 2;
273 sf
.sf7
.line_last_pixel_enable
= 0;
275 /* Set bias for OpenGL rasterization rules:
277 sf
.sf6
.dest_org_vbias
= 0x8;
278 sf
.sf6
.dest_org_hbias
= 0x8;
280 bo
= brw_upload_cache(&brw
->cache
, BRW_SF_UNIT
,
286 /* Emit SF program relocation */
287 dri_bo_emit_reloc(bo
,
288 I915_GEM_DOMAIN_INSTRUCTION
, 0,
289 sf
.thread0
.grf_reg_count
<< 1,
290 offsetof(struct brw_sf_unit_state
, thread0
),
293 /* Emit SF viewport relocation */
294 dri_bo_emit_reloc(bo
,
295 I915_GEM_DOMAIN_INSTRUCTION
, 0,
296 sf
.sf5
.front_winding
| (sf
.sf5
.viewport_transform
<< 1),
297 offsetof(struct brw_sf_unit_state
, sf5
),
303 static void upload_sf_unit( struct brw_context
*brw
)
305 struct brw_sf_unit_key key
;
306 dri_bo
*reloc_bufs
[2];
308 sf_unit_populate_key(brw
, &key
);
310 reloc_bufs
[0] = brw
->sf
.prog_bo
;
311 reloc_bufs
[1] = brw
->sf
.vp_bo
;
313 dri_bo_unreference(brw
->sf
.state_bo
);
314 brw
->sf
.state_bo
= brw_search_cache(&brw
->cache
, BRW_SF_UNIT
,
318 if (brw
->sf
.state_bo
== NULL
) {
319 brw
->sf
.state_bo
= sf_unit_create_from_key(brw
, &key
, reloc_bufs
);
323 const struct brw_tracked_state brw_sf_unit
= {
325 .mesa
= (_NEW_POLYGON
|
329 .brw
= BRW_NEW_URB_FENCE
,
330 .cache
= (CACHE_NEW_SF_VP
|
333 .prepare
= upload_sf_unit
,