1 /**************************************************************************
3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * 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, sub license, 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 portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
30 #include "main/glheader.h"
31 #include "main/context.h"
32 #include "main/state.h"
33 #include "main/api_validate.h"
34 #include "main/enums.h"
37 #include "brw_defines.h"
38 #include "brw_context.h"
39 #include "brw_state.h"
40 #include "brw_fallback.h"
42 #include "intel_batchbuffer.h"
43 #include "intel_buffer_objects.h"
46 #include "vbo/vbo_context.h"
47 #include "swrast/swrast.h"
48 #include "swrast_setup/swrast_setup.h"
50 #define FILE_DEBUG_FLAG DEBUG_BATCH
52 static GLuint prim_to_hw_prim
[GL_POLYGON
+1] = {
66 static const GLenum reduced_prim
[GL_POLYGON
+1] = {
80 /* When the primitive changes, set a state bit and re-validate. Not
81 * the nicest and would rather deal with this by having all the
82 * programs be immune to the active primitive (ie. cope with all
83 * possibilities). That may not be realistic however.
85 static GLuint
brw_set_prim(struct brw_context
*brw
, GLenum prim
)
87 if (INTEL_DEBUG
& DEBUG_PRIMS
)
88 _mesa_printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim
));
90 /* Slight optimization to avoid the GS program when not needed:
92 if (prim
== GL_QUAD_STRIP
&&
93 brw
->attribs
.Light
->ShadeModel
!= GL_FLAT
&&
94 brw
->attribs
.Polygon
->FrontMode
== GL_FILL
&&
95 brw
->attribs
.Polygon
->BackMode
== GL_FILL
)
96 prim
= GL_TRIANGLE_STRIP
;
98 if (prim
!= brw
->primitive
) {
99 brw
->primitive
= prim
;
100 brw
->state
.dirty
.brw
|= BRW_NEW_PRIMITIVE
;
102 if (reduced_prim
[prim
] != brw
->intel
.reduced_primitive
) {
103 brw
->intel
.reduced_primitive
= reduced_prim
[prim
];
104 brw
->state
.dirty
.brw
|= BRW_NEW_REDUCED_PRIMITIVE
;
108 return prim_to_hw_prim
[prim
];
112 static GLuint
trim(GLenum prim
, GLuint length
)
114 if (prim
== GL_QUAD_STRIP
)
115 return length
> 3 ? (length
- length
% 2) : 0;
116 else if (prim
== GL_QUADS
)
117 return length
- length
% 4;
123 static void brw_emit_prim(struct brw_context
*brw
,
124 const struct _mesa_prim
*prim
,
127 struct brw_3d_primitive prim_packet
;
129 if (INTEL_DEBUG
& DEBUG_PRIMS
)
130 _mesa_printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim
->mode
),
131 prim
->start
, prim
->count
);
133 prim_packet
.header
.opcode
= CMD_3D_PRIM
;
134 prim_packet
.header
.length
= sizeof(prim_packet
)/4 - 2;
135 prim_packet
.header
.pad
= 0;
136 prim_packet
.header
.topology
= hw_prim
;
137 prim_packet
.header
.indexed
= prim
->indexed
;
139 prim_packet
.verts_per_instance
= trim(prim
->mode
, prim
->count
);
140 prim_packet
.start_vert_location
= prim
->start
;
141 prim_packet
.instance_count
= 1;
142 prim_packet
.start_instance_location
= 0;
143 prim_packet
.base_vert_location
= 0;
145 /* Can't wrap here, since we rely on the validated state. */
146 brw
->no_batch_wrap
= GL_TRUE
;
147 if (prim_packet
.verts_per_instance
) {
148 intel_batchbuffer_data( brw
->intel
.batch
, &prim_packet
,
149 sizeof(prim_packet
), LOOP_CLIPRECTS
);
151 brw
->no_batch_wrap
= GL_FALSE
;
154 static void brw_merge_inputs( struct brw_context
*brw
,
155 const struct gl_client_array
*arrays
[])
157 struct brw_vertex_info old
= brw
->vb
.info
;
160 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++)
161 dri_bo_unreference(brw
->vb
.inputs
[i
].bo
);
163 memset(&brw
->vb
.inputs
, 0, sizeof(brw
->vb
.inputs
));
164 memset(&brw
->vb
.info
, 0, sizeof(brw
->vb
.info
));
166 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
167 brw
->vb
.inputs
[i
].glarray
= arrays
[i
];
169 if (arrays
[i
]->StrideB
!= 0)
170 brw
->vb
.info
.varying
|= 1 << i
;
172 brw
->vb
.info
.sizes
[i
/16] |= (brw
->vb
.inputs
[i
].glarray
->Size
- 1) <<
176 /* Raise statechanges if input sizes and varying have changed:
178 if (memcmp(brw
->vb
.info
.sizes
, old
.sizes
, sizeof(old
.sizes
)) != 0)
179 brw
->state
.dirty
.brw
|= BRW_NEW_INPUT_DIMENSIONS
;
181 if (brw
->vb
.info
.varying
!= old
.varying
)
182 brw
->state
.dirty
.brw
|= BRW_NEW_INPUT_VARYING
;
185 /* XXX: could split the primitive list to fallback only on the
186 * non-conformant primitives.
188 static GLboolean
check_fallbacks( struct brw_context
*brw
,
189 const struct _mesa_prim
*prim
,
194 if (!brw
->intel
.strict_conformance
)
197 if (brw
->attribs
.Polygon
->SmoothFlag
) {
198 for (i
= 0; i
< nr_prims
; i
++)
199 if (reduced_prim
[prim
[i
].mode
] == GL_TRIANGLES
)
203 /* BRW hardware will do AA lines, but they are non-conformant it
204 * seems. TBD whether we keep this fallback:
206 if (brw
->attribs
.Line
->SmoothFlag
) {
207 for (i
= 0; i
< nr_prims
; i
++)
208 if (reduced_prim
[prim
[i
].mode
] == GL_LINES
)
212 /* Stipple -- these fallbacks could be resolved with a little
215 if (brw
->attribs
.Line
->StippleFlag
) {
216 for (i
= 0; i
< nr_prims
; i
++) {
217 /* GS doesn't get enough information to know when to reset
218 * the stipple counter?!?
220 if (prim
[i
].mode
== GL_LINE_LOOP
)
223 if (prim
[i
].mode
== GL_POLYGON
&&
224 (brw
->attribs
.Polygon
->FrontMode
== GL_LINE
||
225 brw
->attribs
.Polygon
->BackMode
== GL_LINE
))
231 if (brw
->attribs
.Point
->SmoothFlag
) {
232 for (i
= 0; i
< nr_prims
; i
++)
233 if (prim
[i
].mode
== GL_POINTS
)
240 /* May fail if out of video memory for texture or vbo upload, or on
241 * fallback conditions.
243 static GLboolean
brw_try_draw_prims( GLcontext
*ctx
,
244 const struct gl_client_array
*arrays
[],
245 const struct _mesa_prim
*prim
,
247 const struct _mesa_index_buffer
*ib
,
251 struct intel_context
*intel
= intel_context(ctx
);
252 struct brw_context
*brw
= brw_context(ctx
);
253 GLboolean retval
= GL_FALSE
;
254 GLboolean warn
= GL_FALSE
;
255 GLboolean first_time
= GL_TRUE
;
259 _mesa_update_state( ctx
);
261 if (check_fallbacks(brw
, prim
, nr_prims
))
264 brw_validate_textures( brw
);
266 /* Bind all inputs, derive varying and size information:
268 brw_merge_inputs( brw
, arrays
);
271 brw
->state
.dirty
.brw
|= BRW_NEW_INDICES
;
273 brw
->vb
.min_index
= min_index
;
274 brw
->vb
.max_index
= max_index
;
275 brw
->state
.dirty
.brw
|= BRW_NEW_VERTICES
;
277 /* Have to validate state quite late. Will rebuild tnl_program,
278 * which depends on varying information.
280 * Note this is where brw->vs->prog_data.inputs_read is calculated,
281 * so can't access it earlier.
284 LOCK_HARDWARE(intel
);
286 if (!intel
->constant_cliprect
&& intel
->driDrawable
->numClipRects
== 0) {
287 UNLOCK_HARDWARE(intel
);
291 for (i
= 0; i
< nr_prims
; i
++) {
294 /* Flush the batch if it's approaching full, so that we don't wrap while
295 * we've got validated state that needs to be in the same batch as the
296 * primitives. This fraction is just a guess (minimal full state plus
297 * a primitive is around 512 bytes), and would be better if we had
298 * an upper bound of how much we might emit in a single
299 * brw_try_draw_prims().
301 intel_batchbuffer_require_space(intel
->batch
, intel
->batch
->size
/ 4,
304 hw_prim
= brw_set_prim(brw
, prim
[i
].mode
);
306 if (first_time
|| (brw
->state
.dirty
.brw
& BRW_NEW_PRIMITIVE
)) {
307 first_time
= GL_FALSE
;
309 brw_validate_state(brw
);
311 /* Various fallback checks: */
312 if (brw
->intel
.Fallback
)
315 /* Check that we can fit our state in with our existing batchbuffer, or
318 if (dri_bufmgr_check_aperture_space(brw
->state
.validated_bos
,
319 brw
->state
.validated_bo_count
)) {
320 static GLboolean warned
;
321 intel_batchbuffer_flush(intel
->batch
);
323 /* Validate the state after we flushed the batch (which would have
324 * changed the set of dirty state). If we still fail to
325 * check_aperture, warn of what's happening, but attempt to continue
326 * on since it may succeed anyway, and the user would probably rather
327 * see a failure and a warning than a fallback.
329 brw_validate_state(brw
);
331 dri_bufmgr_check_aperture_space(brw
->state
.validated_bos
,
332 brw
->state
.validated_bo_count
)) {
338 brw_upload_state(brw
);
341 brw_emit_prim(brw
, &prim
[i
], hw_prim
);
347 UNLOCK_HARDWARE(intel
);
350 fprintf(stderr
, "i965: Single primitive emit potentially exceeded "
351 "available aperture space\n");
354 DBG("%s failed\n", __FUNCTION__
);
359 static GLboolean
brw_need_rebase( GLcontext
*ctx
,
360 const struct gl_client_array
*arrays
[],
361 const struct _mesa_index_buffer
*ib
,
368 if (!vbo_all_varyings_in_vbos(arrays
))
374 /* Hmm. This isn't quite what I wanted. BRW can actually
375 * handle the mixed case well enough that we shouldn't need to
376 * rebase. However, it's probably not very common, nor hugely
377 * expensive to do it this way:
379 if (!vbo_all_varyings_in_vbos(arrays
))
387 void brw_draw_prims( GLcontext
*ctx
,
388 const struct gl_client_array
*arrays
[],
389 const struct _mesa_prim
*prim
,
391 const struct _mesa_index_buffer
*ib
,
397 /* Decide if we want to rebase. If so we end up recursing once
398 * only into this function.
400 if (brw_need_rebase( ctx
, arrays
, ib
, min_index
)) {
401 vbo_rebase_prims( ctx
, arrays
,
403 ib
, min_index
, max_index
,
409 /* Make a first attempt at drawing:
411 retval
= brw_try_draw_prims(ctx
, arrays
, prim
, nr_prims
, ib
, min_index
, max_index
);
413 /* Otherwise, we really are out of memory. Pass the drawing
414 * command to the software tnl module and which will in turn call
415 * swrast to do the drawing.
418 _swsetup_Wakeup(ctx
);
419 _tnl_draw_prims(ctx
, arrays
, prim
, nr_prims
, ib
, min_index
, max_index
);
424 void brw_draw_init( struct brw_context
*brw
)
426 GLcontext
*ctx
= &brw
->intel
.ctx
;
427 struct vbo_context
*vbo
= vbo_context(ctx
);
429 /* Register our drawing function:
431 vbo
->draw_prims
= brw_draw_prims
;
434 void brw_draw_destroy( struct brw_context
*brw
)
438 if (brw
->vb
.upload
.bo
!= NULL
) {
439 dri_bo_unreference(brw
->vb
.upload
.bo
);
440 brw
->vb
.upload
.bo
= NULL
;
443 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
444 dri_bo_unreference(brw
->vb
.inputs
[i
].bo
);
445 brw
->vb
.inputs
[i
].bo
= NULL
;
448 dri_bo_unreference(brw
->ib
.bo
);