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 **************************************************************************/
29 #include "main/glheader.h"
30 #include "main/context.h"
31 #include "main/condrender.h"
32 #include "main/samplerobj.h"
33 #include "main/state.h"
34 #include "main/enums.h"
36 #include "vbo/vbo_context.h"
37 #include "swrast/swrast.h"
38 #include "swrast_setup/swrast_setup.h"
41 #include "brw_defines.h"
42 #include "brw_context.h"
43 #include "brw_state.h"
45 #include "intel_batchbuffer.h"
47 #define FILE_DEBUG_FLAG DEBUG_PRIMS
49 static GLuint prim_to_hw_prim
[GL_POLYGON
+1] = {
63 static const GLenum reduced_prim
[GL_POLYGON
+1] = {
77 /* When the primitive changes, set a state bit and re-validate. Not
78 * the nicest and would rather deal with this by having all the
79 * programs be immune to the active primitive (ie. cope with all
80 * possibilities). That may not be realistic however.
82 static void brw_set_prim(struct brw_context
*brw
,
83 const struct _mesa_prim
*prim
)
85 struct gl_context
*ctx
= &brw
->intel
.ctx
;
86 uint32_t hw_prim
= prim_to_hw_prim
[prim
->mode
];
88 DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim
->mode
));
90 /* Slight optimization to avoid the GS program when not needed:
92 if (prim
->mode
== GL_QUAD_STRIP
&&
93 ctx
->Light
.ShadeModel
!= GL_FLAT
&&
94 ctx
->Polygon
.FrontMode
== GL_FILL
&&
95 ctx
->Polygon
.BackMode
== GL_FILL
)
96 hw_prim
= _3DPRIM_TRISTRIP
;
98 if (prim
->mode
== GL_QUADS
&& prim
->count
== 4 &&
99 ctx
->Light
.ShadeModel
!= GL_FLAT
&&
100 ctx
->Polygon
.FrontMode
== GL_FILL
&&
101 ctx
->Polygon
.BackMode
== GL_FILL
) {
102 hw_prim
= _3DPRIM_TRIFAN
;
105 if (hw_prim
!= brw
->primitive
) {
106 brw
->primitive
= hw_prim
;
107 brw
->state
.dirty
.brw
|= BRW_NEW_PRIMITIVE
;
109 if (reduced_prim
[prim
->mode
] != brw
->intel
.reduced_primitive
) {
110 brw
->intel
.reduced_primitive
= reduced_prim
[prim
->mode
];
111 brw
->state
.dirty
.brw
|= BRW_NEW_REDUCED_PRIMITIVE
;
116 static void gen6_set_prim(struct brw_context
*brw
,
117 const struct _mesa_prim
*prim
)
119 uint32_t hw_prim
= prim_to_hw_prim
[prim
->mode
];
121 DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim
->mode
));
123 if (hw_prim
!= brw
->primitive
) {
124 brw
->primitive
= hw_prim
;
125 brw
->state
.dirty
.brw
|= BRW_NEW_PRIMITIVE
;
130 static GLuint
trim(GLenum prim
, GLuint length
)
132 if (prim
== GL_QUAD_STRIP
)
133 return length
> 3 ? (length
- length
% 2) : 0;
134 else if (prim
== GL_QUADS
)
135 return length
- length
% 4;
141 static void brw_emit_prim(struct brw_context
*brw
,
142 const struct _mesa_prim
*prim
,
145 struct intel_context
*intel
= &brw
->intel
;
146 int verts_per_instance
;
147 int vertex_access_type
;
148 int start_vertex_location
;
149 int base_vertex_location
;
151 DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim
->mode
),
152 prim
->start
, prim
->count
);
154 start_vertex_location
= prim
->start
;
155 base_vertex_location
= prim
->basevertex
;
157 vertex_access_type
= GEN4_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM
;
158 start_vertex_location
+= brw
->ib
.start_vertex_offset
;
159 base_vertex_location
+= brw
->vb
.start_vertex_bias
;
161 vertex_access_type
= GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL
;
162 start_vertex_location
+= brw
->vb
.start_vertex_bias
;
165 verts_per_instance
= trim(prim
->mode
, prim
->count
);
167 /* If nothing to emit, just return. */
168 if (verts_per_instance
== 0)
171 /* If we're set to always flush, do it before and after the primitive emit.
172 * We want to catch both missed flushes that hurt instruction/state cache
173 * and missed flushes of the render cache as it heads to other parts of
174 * the besides the draw code.
176 if (intel
->always_flush_cache
) {
177 intel_batchbuffer_emit_mi_flush(intel
);
181 OUT_BATCH(CMD_3D_PRIM
<< 16 | (6 - 2) |
182 hw_prim
<< GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT
|
184 OUT_BATCH(verts_per_instance
);
185 OUT_BATCH(start_vertex_location
);
186 OUT_BATCH(1); // instance count
187 OUT_BATCH(0); // start instance location
188 OUT_BATCH(base_vertex_location
);
191 intel
->batch
.need_workaround_flush
= true;
193 if (intel
->always_flush_cache
) {
194 intel_batchbuffer_emit_mi_flush(intel
);
198 static void gen7_emit_prim(struct brw_context
*brw
,
199 const struct _mesa_prim
*prim
,
202 struct intel_context
*intel
= &brw
->intel
;
203 int verts_per_instance
;
204 int vertex_access_type
;
205 int start_vertex_location
;
206 int base_vertex_location
;
208 DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim
->mode
),
209 prim
->start
, prim
->count
);
211 start_vertex_location
= prim
->start
;
212 base_vertex_location
= prim
->basevertex
;
214 vertex_access_type
= GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM
;
215 start_vertex_location
+= brw
->ib
.start_vertex_offset
;
216 base_vertex_location
+= brw
->vb
.start_vertex_bias
;
218 vertex_access_type
= GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL
;
219 start_vertex_location
+= brw
->vb
.start_vertex_bias
;
222 verts_per_instance
= trim(prim
->mode
, prim
->count
);
224 /* If nothing to emit, just return. */
225 if (verts_per_instance
== 0)
228 /* If we're set to always flush, do it before and after the primitive emit.
229 * We want to catch both missed flushes that hurt instruction/state cache
230 * and missed flushes of the render cache as it heads to other parts of
231 * the besides the draw code.
233 if (intel
->always_flush_cache
) {
234 intel_batchbuffer_emit_mi_flush(intel
);
238 OUT_BATCH(CMD_3D_PRIM
<< 16 | (7 - 2));
239 OUT_BATCH(hw_prim
| vertex_access_type
);
240 OUT_BATCH(verts_per_instance
);
241 OUT_BATCH(start_vertex_location
);
242 OUT_BATCH(1); // instance count
243 OUT_BATCH(0); // start instance location
244 OUT_BATCH(base_vertex_location
);
247 if (intel
->always_flush_cache
) {
248 intel_batchbuffer_emit_mi_flush(intel
);
253 static void brw_merge_inputs( struct brw_context
*brw
,
254 const struct gl_client_array
*arrays
[])
256 struct brw_vertex_info old
= brw
->vb
.info
;
259 for (i
= 0; i
< brw
->vb
.nr_buffers
; i
++) {
260 drm_intel_bo_unreference(brw
->vb
.buffers
[i
].bo
);
261 brw
->vb
.buffers
[i
].bo
= NULL
;
263 brw
->vb
.nr_buffers
= 0;
265 memset(&brw
->vb
.info
, 0, sizeof(brw
->vb
.info
));
267 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
268 brw
->vb
.inputs
[i
].buffer
= -1;
269 brw
->vb
.inputs
[i
].glarray
= arrays
[i
];
270 brw
->vb
.inputs
[i
].attrib
= (gl_vert_attrib
) i
;
272 if (arrays
[i
]->StrideB
!= 0)
273 brw
->vb
.info
.sizes
[i
/16] |= (brw
->vb
.inputs
[i
].glarray
->Size
- 1) <<
277 /* Raise statechanges if input sizes have changed. */
278 if (memcmp(brw
->vb
.info
.sizes
, old
.sizes
, sizeof(old
.sizes
)) != 0)
279 brw
->state
.dirty
.brw
|= BRW_NEW_INPUT_DIMENSIONS
;
282 /* May fail if out of video memory for texture or vbo upload, or on
283 * fallback conditions.
285 static bool brw_try_draw_prims( struct gl_context
*ctx
,
286 const struct gl_client_array
*arrays
[],
287 const struct _mesa_prim
*prim
,
289 const struct _mesa_index_buffer
*ib
,
293 struct intel_context
*intel
= intel_context(ctx
);
294 struct brw_context
*brw
= brw_context(ctx
);
300 _mesa_update_state( ctx
);
302 /* We have to validate the textures *before* checking for fallbacks;
303 * otherwise, the software fallback won't be able to rely on the
304 * texture state, the firstLevel and lastLevel fields won't be
305 * set in the intel texture object (they'll both be 0), and the
306 * software fallback will segfault if it attempts to access any
307 * texture level other than level 0.
309 brw_validate_textures( brw
);
311 /* Bind all inputs, derive varying and size information:
313 brw_merge_inputs( brw
, arrays
);
316 brw
->state
.dirty
.brw
|= BRW_NEW_INDICES
;
318 brw
->vb
.min_index
= min_index
;
319 brw
->vb
.max_index
= max_index
;
320 brw
->state
.dirty
.brw
|= BRW_NEW_VERTICES
;
322 /* Have to validate state quite late. Will rebuild tnl_program,
323 * which depends on varying information.
325 * Note this is where brw->vs->prog_data.inputs_read is calculated,
326 * so can't access it earlier.
329 intel_prepare_render(intel
);
331 for (i
= 0; i
< nr_prims
; i
++) {
332 int estimated_max_prim_size
;
334 estimated_max_prim_size
= 512; /* batchbuffer commands */
335 estimated_max_prim_size
+= (BRW_MAX_TEX_UNIT
*
336 (sizeof(struct brw_sampler_state
) +
337 sizeof(struct gen5_sampler_default_color
)));
338 estimated_max_prim_size
+= 1024; /* gen6 VS push constants */
339 estimated_max_prim_size
+= 1024; /* gen6 WM push constants */
340 estimated_max_prim_size
+= 512; /* misc. pad */
342 /* Flush the batch if it's approaching full, so that we don't wrap while
343 * we've got validated state that needs to be in the same batch as the
346 intel_batchbuffer_require_space(intel
, estimated_max_prim_size
, false);
349 brw_set_prim(brw
, &prim
[i
]);
351 gen6_set_prim(brw
, &prim
[i
]);
353 if (brw
->state
.dirty
.brw
) {
354 brw_validate_state(brw
);
356 /* Various fallback checks: */
357 if (brw
->intel
.Fallback
)
360 /* Check that we can fit our state in with our existing batchbuffer, or
363 if (dri_bufmgr_check_aperture_space(brw
->state
.validated_bos
,
364 brw
->state
.validated_bo_count
)) {
366 intel_batchbuffer_flush(intel
);
368 /* Validate the state after we flushed the batch (which would have
369 * changed the set of dirty state). If we still fail to
370 * check_aperture, warn of what's happening, but attempt to continue
371 * on since it may succeed anyway, and the user would probably rather
372 * see a failure and a warning than a fallback.
374 brw_validate_state(brw
);
376 dri_bufmgr_check_aperture_space(brw
->state
.validated_bos
,
377 brw
->state
.validated_bo_count
)) {
383 intel
->no_batch_wrap
= true;
384 brw_upload_state(brw
);
388 gen7_emit_prim(brw
, &prim
[i
], brw
->primitive
);
390 brw_emit_prim(brw
, &prim
[i
], brw
->primitive
);
392 intel
->no_batch_wrap
= false;
397 if (intel
->always_flush_batch
)
398 intel_batchbuffer_flush(intel
);
401 brw_state_cache_check_size(brw
);
404 fprintf(stderr
, "i965: Single primitive emit potentially exceeded "
405 "available aperture space\n");
408 DBG("%s failed\n", __FUNCTION__
);
413 void brw_draw_prims( struct gl_context
*ctx
,
414 const struct gl_client_array
*arrays
[],
415 const struct _mesa_prim
*prim
,
417 const struct _mesa_index_buffer
*ib
,
418 GLboolean index_bounds_valid
,
424 if (!_mesa_check_conditional_render(ctx
))
427 if (!vbo_all_varyings_in_vbos(arrays
)) {
428 if (!index_bounds_valid
)
429 vbo_get_minmax_index(ctx
, prim
, ib
, &min_index
, &max_index
);
431 /* Decide if we want to rebase. If so we end up recursing once
432 * only into this function.
434 if (min_index
!= 0 && !vbo_any_varyings_in_vbos(arrays
)) {
435 vbo_rebase_prims(ctx
, arrays
,
437 ib
, min_index
, max_index
,
443 /* Make a first attempt at drawing:
445 retval
= brw_try_draw_prims(ctx
, arrays
, prim
, nr_prims
, ib
, min_index
, max_index
);
447 /* Otherwise, we really are out of memory. Pass the drawing
448 * command to the software tnl module and which will in turn call
449 * swrast to do the drawing.
452 _swsetup_Wakeup(ctx
);
454 _tnl_draw_prims(ctx
, arrays
, prim
, nr_prims
, ib
, min_index
, max_index
);
459 void brw_draw_init( struct brw_context
*brw
)
461 struct gl_context
*ctx
= &brw
->intel
.ctx
;
462 struct vbo_context
*vbo
= vbo_context(ctx
);
465 /* Register our drawing function:
467 vbo
->draw_prims
= brw_draw_prims
;
469 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++)
470 brw
->vb
.inputs
[i
].buffer
= -1;
471 brw
->vb
.nr_buffers
= 0;
472 brw
->vb
.nr_enabled
= 0;
475 void brw_draw_destroy( struct brw_context
*brw
)
479 for (i
= 0; i
< brw
->vb
.nr_buffers
; i
++) {
480 drm_intel_bo_unreference(brw
->vb
.buffers
[i
].bo
);
481 brw
->vb
.buffers
[i
].bo
= NULL
;
483 brw
->vb
.nr_buffers
= 0;
485 for (i
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
486 brw
->vb
.enabled
[i
]->buffer
= -1;
488 brw
->vb
.nr_enabled
= 0;
490 drm_intel_bo_unreference(brw
->ib
.bo
);