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 **************************************************************************/
33 #include "api_validate.h"
37 #include "brw_defines.h"
38 #include "brw_context.h"
40 #include "brw_state.h"
41 #include "brw_fallback.h"
43 #include "intel_ioctl.h"
44 #include "intel_batchbuffer.h"
45 #include "intel_buffer_objects.h"
48 #include "vbo/vbo_context.h"
49 #include "swrast/swrast.h"
50 #include "swrast_setup/swrast_setup.h"
54 static GLuint hw_prim
[GL_POLYGON
+1] = {
68 static const GLenum reduced_prim
[GL_POLYGON
+1] = {
82 /* When the primitive changes, set a state bit and re-validate. Not
83 * the nicest and would rather deal with this by having all the
84 * programs be immune to the active primitive (ie. cope with all
85 * possibilities). That may not be realistic however.
87 static GLuint
brw_set_prim(struct brw_context
*brw
, GLenum prim
)
89 if (INTEL_DEBUG
& DEBUG_PRIMS
)
90 _mesa_printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim
));
92 /* Slight optimization to avoid the GS program when not needed:
94 if (prim
== GL_QUAD_STRIP
&&
95 brw
->attribs
.Light
->ShadeModel
!= GL_FLAT
&&
96 brw
->attribs
.Polygon
->FrontMode
== GL_FILL
&&
97 brw
->attribs
.Polygon
->BackMode
== GL_FILL
)
98 prim
= GL_TRIANGLE_STRIP
;
100 if (prim
!= brw
->primitive
) {
101 brw
->primitive
= prim
;
102 brw
->state
.dirty
.brw
|= BRW_NEW_PRIMITIVE
;
104 if (reduced_prim
[prim
] != brw
->intel
.reduced_primitive
) {
105 brw
->intel
.reduced_primitive
= reduced_prim
[prim
];
106 brw
->state
.dirty
.brw
|= BRW_NEW_REDUCED_PRIMITIVE
;
109 brw_validate_state(brw
);
112 return hw_prim
[prim
];
116 static GLuint
trim(GLenum prim
, GLuint length
)
118 if (prim
== GL_QUAD_STRIP
)
119 return length
> 3 ? (length
- length
% 2) : 0;
120 else if (prim
== GL_QUADS
)
121 return length
- length
% 4;
127 static void brw_emit_cliprect( struct brw_context
*brw
,
128 const drm_clip_rect_t
*rect
)
130 struct brw_drawrect bdr
;
132 bdr
.header
.opcode
= CMD_DRAW_RECT
;
133 bdr
.header
.length
= sizeof(bdr
)/4 - 2;
135 bdr
.xmax
= rect
->x2
- 1;
137 bdr
.ymax
= rect
->y2
- 1;
138 bdr
.xorg
= brw
->intel
.drawX
;
139 bdr
.yorg
= brw
->intel
.drawY
;
141 intel_batchbuffer_data( brw
->intel
.batch
, &bdr
, sizeof(bdr
),
142 INTEL_BATCH_NO_CLIPRECTS
);
146 static void brw_emit_prim( struct brw_context
*brw
,
147 const struct _mesa_prim
*prim
)
150 struct brw_3d_primitive prim_packet
;
152 if (INTEL_DEBUG
& DEBUG_PRIMS
)
153 _mesa_printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim
->mode
),
154 prim
->start
, prim
->count
);
156 prim_packet
.header
.opcode
= CMD_3D_PRIM
;
157 prim_packet
.header
.length
= sizeof(prim_packet
)/4 - 2;
158 prim_packet
.header
.pad
= 0;
159 prim_packet
.header
.topology
= brw_set_prim(brw
, prim
->mode
);
160 prim_packet
.header
.indexed
= prim
->indexed
;
162 prim_packet
.verts_per_instance
= trim(prim
->mode
, prim
->count
);
163 prim_packet
.start_vert_location
= prim
->start
;
164 prim_packet
.instance_count
= 1;
165 prim_packet
.start_instance_location
= 0;
166 prim_packet
.base_vert_location
= 0;
168 if (prim_packet
.verts_per_instance
) {
169 intel_batchbuffer_data( brw
->intel
.batch
, &prim_packet
, sizeof(prim_packet
),
170 INTEL_BATCH_NO_CLIPRECTS
);
174 static void brw_merge_inputs( struct brw_context
*brw
,
175 const struct gl_client_array
*arrays
[])
177 struct brw_vertex_element
*inputs
= brw
->vb
.inputs
;
178 struct brw_vertex_info old
= brw
->vb
.info
;
181 memset(inputs
, 0, sizeof(*inputs
));
182 memset(&brw
->vb
.info
, 0, sizeof(brw
->vb
.info
));
184 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
185 brw
->vb
.inputs
[i
].glarray
= arrays
[i
];
187 /* XXX: metaops passes null arrays */
189 if (arrays
[i
]->StrideB
!= 0)
190 brw
->vb
.info
.varying
|= 1 << i
;
192 brw
->vb
.info
.sizes
[i
/16] |= (inputs
[i
].glarray
->Size
- 1) << ((i
%16) * 2);
196 /* Raise statechanges if input sizes and varying have changed:
198 if (memcmp(brw
->vb
.info
.sizes
, old
.sizes
, sizeof(old
.sizes
)) != 0)
199 brw
->state
.dirty
.brw
|= BRW_NEW_INPUT_DIMENSIONS
;
201 if (brw
->vb
.info
.varying
!= old
.varying
)
202 brw
->state
.dirty
.brw
|= BRW_NEW_INPUT_VARYING
;
205 /* XXX: could split the primitive list to fallback only on the
206 * non-conformant primitives.
208 static GLboolean
check_fallbacks( struct brw_context
*brw
,
209 const struct _mesa_prim
*prim
,
214 if (!brw
->intel
.strict_conformance
)
217 if (brw
->attribs
.Polygon
->SmoothFlag
) {
218 for (i
= 0; i
< nr_prims
; i
++)
219 if (reduced_prim
[prim
[i
].mode
] == GL_TRIANGLES
)
223 /* BRW hardware will do AA lines, but they are non-conformant it
224 * seems. TBD whether we keep this fallback:
226 if (brw
->attribs
.Line
->SmoothFlag
) {
227 for (i
= 0; i
< nr_prims
; i
++)
228 if (reduced_prim
[prim
[i
].mode
] == GL_LINES
)
232 /* Stipple -- these fallbacks could be resolved with a little
235 if (brw
->attribs
.Line
->StippleFlag
) {
236 for (i
= 0; i
< nr_prims
; i
++) {
237 /* GS doesn't get enough information to know when to reset
238 * the stipple counter?!?
240 if (prim
[i
].mode
== GL_LINE_LOOP
)
243 if (prim
[i
].mode
== GL_POLYGON
&&
244 (brw
->attribs
.Polygon
->FrontMode
== GL_LINE
||
245 brw
->attribs
.Polygon
->BackMode
== GL_LINE
))
251 if (brw
->attribs
.Point
->SmoothFlag
) {
252 for (i
= 0; i
< nr_prims
; i
++)
253 if (prim
[i
].mode
== GL_POINTS
)
260 /* May fail if out of video memory for texture or vbo upload, or on
261 * fallback conditions.
263 static GLboolean
brw_try_draw_prims( GLcontext
*ctx
,
264 const struct gl_client_array
*arrays
[],
265 const struct _mesa_prim
*prim
,
267 const struct _mesa_index_buffer
*ib
,
271 struct intel_context
*intel
= intel_context(ctx
);
272 struct brw_context
*brw
= brw_context(ctx
);
273 GLboolean retval
= GL_FALSE
;
277 _mesa_update_state( ctx
);
279 /* Bind all inputs, derive varying and size information:
281 brw_merge_inputs( brw
, arrays
);
283 /* Have to validate state quite late. Will rebuild tnl_program,
284 * which depends on varying information.
286 * Note this is where brw->vs->prog_data.inputs_read is calculated,
287 * so can't access it earlier.
290 LOCK_HARDWARE(intel
);
292 if (brw
->intel
.numClipRects
== 0) {
293 assert(intel
->batch
->ptr
== intel
->batch
->map
+ intel
->batch
->offset
);
294 UNLOCK_HARDWARE(intel
);
299 /* Set the first primitive early, ahead of validate_state:
301 brw_set_prim(brw
, prim
[0].mode
);
303 /* XXX: Need to separate validate and upload of state.
305 brw_validate_state( brw
);
307 /* Various fallback checks:
309 if (brw
->intel
.Fallback
)
312 if (check_fallbacks( brw
, prim
, nr_prims
))
315 /* Upload index, vertex data:
318 brw_upload_indices( brw
, ib
);
320 if (!brw_upload_vertices( brw
, min_index
, max_index
)) {
324 /* For single cliprect, state is already emitted:
326 if (brw
->intel
.numClipRects
== 1) {
327 for (i
= 0; i
< nr_prims
; i
++) {
328 brw_emit_prim(brw
, &prim
[i
]);
332 /* Otherwise, explicitly do the cliprects at this point:
334 for (j
= 0; j
< brw
->intel
.numClipRects
; j
++) {
335 brw_emit_cliprect(brw
, &brw
->intel
.pClipRects
[j
]);
337 /* Emit prims to batchbuffer:
339 for (i
= 0; i
< nr_prims
; i
++) {
340 brw_emit_prim(brw
, &prim
[i
]);
345 intel
->need_flush
= GL_TRUE
;
351 /* Currently have to do this to synchronize with the map/unmap of
352 * the vertex buffer in brw_exec_api.c. Not sure if there is any
353 * way around this, as not every flush is due to a buffer filling
356 if (!intel_batchbuffer_flush( brw
->intel
.batch
)) {
357 DBG("%s intel_batchbuffer_flush failed\n", __FUNCTION__
);
361 if (retval
&& intel
->thrashing
) {
365 /* Free any old data so it doesn't clog up texture memory - we
366 * won't be referencing it again.
368 while (brw
->vb
.upload
.wrap
!= brw
->vb
.upload
.buf
) {
369 ctx
->Driver
.BufferData(ctx
,
371 BRW_UPLOAD_INIT_SIZE
,
374 brw
->vb
.upload
.vbo
[brw
->vb
.upload
.wrap
]);
375 brw
->vb
.upload
.wrap
++;
376 brw
->vb
.upload
.wrap
%= BRW_NR_UPLOAD_BUFS
;
379 UNLOCK_HARDWARE(intel
);
382 DBG("%s failed\n", __FUNCTION__
);
387 static GLboolean
brw_need_rebase( GLcontext
*ctx
,
388 const struct gl_client_array
*arrays
[],
389 const struct _mesa_index_buffer
*ib
,
396 if (!vbo_all_varyings_in_vbos(arrays
))
402 /* Hmm. This isn't quite what I wanted. BRW can actually
403 * handle the mixed case well enough that we shouldn't need to
404 * rebase. However, it's probably not very common, nor hugely
405 * expensive to do it this way:
407 if (!vbo_all_varyings_in_vbos(arrays
))
415 void brw_draw_prims( GLcontext
*ctx
,
416 const struct gl_client_array
*arrays
[],
417 const struct _mesa_prim
*prim
,
419 const struct _mesa_index_buffer
*ib
,
423 struct intel_context
*intel
= intel_context(ctx
);
426 /* Decide if we want to rebase. If so we end up recursing once
427 * only into this function.
429 if (brw_need_rebase( ctx
, arrays
, ib
, min_index
)) {
430 vbo_rebase_prims( ctx
, arrays
,
432 ib
, min_index
, max_index
,
439 /* Make a first attempt at drawing:
441 retval
= brw_try_draw_prims(ctx
, arrays
, prim
, nr_prims
, ib
, min_index
, max_index
);
444 /* This looks like out-of-memory but potentially we have
445 * situation where there is enough memory but it has become
446 * fragmented. Clear out all heaps and start from scratch by
447 * faking a contended lock event: (done elsewhere)
449 if (!retval
&& !intel
->Fallback
&& bmError(intel
)) {
451 /* Then try a second time only to upload textures and draw the
454 retval
= brw_try_draw_prims(ctx
, arrays
, prim
, nr_prims
, ib
, min_index
, max_index
);
457 /* Otherwise, we really are out of memory. Pass the drawing
458 * command to the software tnl module and which will in turn call
459 * swrast to do the drawing.
462 _swsetup_Wakeup(ctx
);
463 _tnl_draw_prims(ctx
, arrays
, prim
, nr_prims
, ib
, min_index
, max_index
);
466 if (intel
->aub_file
&& (INTEL_DEBUG
& DEBUG_SYNC
)) {
467 intelFinish( &intel
->ctx
);
473 static void brw_invalidate_vbo_cb( struct intel_context
*intel
, void *ptr
)
475 /* nothing to do, we don't rely on the contents being preserved */
479 void brw_draw_init( struct brw_context
*brw
)
481 GLcontext
*ctx
= &brw
->intel
.ctx
;
482 struct vbo_context
*vbo
= vbo_context(ctx
);
485 /* Register our drawing function:
487 vbo
->draw_prims
= brw_draw_prims
;
489 brw
->vb
.upload
.size
= BRW_UPLOAD_INIT_SIZE
;
491 for (i
= 0; i
< BRW_NR_UPLOAD_BUFS
; i
++) {
492 brw
->vb
.upload
.vbo
[i
] = ctx
->Driver
.NewBufferObject(ctx
, 1, GL_ARRAY_BUFFER_ARB
);
494 /* NOTE: These are set to no-backing-store.
496 bmBufferSetInvalidateCB(&brw
->intel
,
497 intel_bufferobj_buffer(intel_buffer_object(brw
->vb
.upload
.vbo
[i
])),
498 brw_invalidate_vbo_cb
,
503 ctx
->Driver
.BufferData( ctx
,
505 BRW_UPLOAD_INIT_SIZE
,
508 brw
->vb
.upload
.vbo
[0] );
511 void brw_draw_destroy( struct brw_context
*brw
)
513 GLcontext
*ctx
= &brw
->intel
.ctx
;
516 for (i
= 0; i
< BRW_NR_UPLOAD_BUFS
; i
++)
517 ctx
->Driver
.DeleteBuffer(ctx
, brw
->vb
.upload
.vbo
[i
]);