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/state.h"
32 #include "main/enums.h"
34 #include "vbo/vbo_context.h"
35 #include "swrast/swrast.h"
36 #include "swrast_setup/swrast_setup.h"
39 #include "brw_defines.h"
40 #include "brw_context.h"
41 #include "brw_state.h"
42 #include "brw_fallback.h"
44 #include "intel_batchbuffer.h"
45 #include "intel_buffer_objects.h"
47 #define FILE_DEBUG_FLAG DEBUG_BATCH
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 GLuint
brw_set_prim(struct brw_context
*brw
, GLenum prim
)
84 GLcontext
*ctx
= &brw
->intel
.ctx
;
86 if (INTEL_DEBUG
& DEBUG_PRIMS
)
87 _mesa_printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim
));
89 /* Slight optimization to avoid the GS program when not needed:
91 if (prim
== GL_QUAD_STRIP
&&
92 ctx
->Light
.ShadeModel
!= GL_FLAT
&&
93 ctx
->Polygon
.FrontMode
== GL_FILL
&&
94 ctx
->Polygon
.BackMode
== GL_FILL
)
95 prim
= GL_TRIANGLE_STRIP
;
97 if (prim
!= brw
->primitive
) {
98 brw
->primitive
= prim
;
99 brw
->state
.dirty
.brw
|= BRW_NEW_PRIMITIVE
;
101 if (reduced_prim
[prim
] != brw
->intel
.reduced_primitive
) {
102 brw
->intel
.reduced_primitive
= reduced_prim
[prim
];
103 brw
->state
.dirty
.brw
|= BRW_NEW_REDUCED_PRIMITIVE
;
107 return prim_to_hw_prim
[prim
];
111 static GLuint
trim(GLenum prim
, GLuint length
)
113 if (prim
== GL_QUAD_STRIP
)
114 return length
> 3 ? (length
- length
% 2) : 0;
115 else if (prim
== GL_QUADS
)
116 return length
- length
% 4;
122 static void brw_emit_prim(struct brw_context
*brw
,
123 const struct _mesa_prim
*prim
,
126 struct brw_3d_primitive prim_packet
;
127 struct intel_context
*intel
= &brw
->intel
;
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
;
142 prim_packet
.start_vert_location
+= brw
->ib
.start_vertex_offset
;
143 prim_packet
.instance_count
= 1;
144 prim_packet
.start_instance_location
= 0;
145 prim_packet
.base_vert_location
= prim
->basevertex
;
147 /* Can't wrap here, since we rely on the validated state. */
148 brw
->no_batch_wrap
= GL_TRUE
;
150 /* If we're set to always flush, do it before and after the primitive emit.
151 * We want to catch both missed flushes that hurt instruction/state cache
152 * and missed flushes of the render cache as it heads to other parts of
153 * the besides the draw code.
155 if (intel
->always_flush_cache
) {
156 BEGIN_BATCH(1, IGNORE_CLIPRECTS
);
157 OUT_BATCH(intel
->vtbl
.flush_cmd());
160 if (prim_packet
.verts_per_instance
) {
161 intel_batchbuffer_data( brw
->intel
.batch
, &prim_packet
,
162 sizeof(prim_packet
), LOOP_CLIPRECTS
);
164 if (intel
->always_flush_cache
) {
165 BEGIN_BATCH(1, IGNORE_CLIPRECTS
);
166 OUT_BATCH(intel
->vtbl
.flush_cmd());
170 brw
->no_batch_wrap
= GL_FALSE
;
173 static void brw_merge_inputs( struct brw_context
*brw
,
174 const struct gl_client_array
*arrays
[])
176 struct brw_vertex_info old
= brw
->vb
.info
;
179 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++)
180 dri_bo_unreference(brw
->vb
.inputs
[i
].bo
);
182 memset(&brw
->vb
.inputs
, 0, sizeof(brw
->vb
.inputs
));
183 memset(&brw
->vb
.info
, 0, sizeof(brw
->vb
.info
));
185 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
186 brw
->vb
.inputs
[i
].glarray
= arrays
[i
];
187 brw
->vb
.inputs
[i
].attrib
= (gl_vert_attrib
) i
;
189 if (arrays
[i
]->StrideB
!= 0)
190 brw
->vb
.info
.sizes
[i
/16] |= (brw
->vb
.inputs
[i
].glarray
->Size
- 1) <<
194 /* Raise statechanges if input sizes have changed. */
195 if (memcmp(brw
->vb
.info
.sizes
, old
.sizes
, sizeof(old
.sizes
)) != 0)
196 brw
->state
.dirty
.brw
|= BRW_NEW_INPUT_DIMENSIONS
;
199 /* XXX: could split the primitive list to fallback only on the
200 * non-conformant primitives.
202 static GLboolean
check_fallbacks( struct brw_context
*brw
,
203 const struct _mesa_prim
*prim
,
206 GLcontext
*ctx
= &brw
->intel
.ctx
;
209 /* If we don't require strict OpenGL conformance, never
210 * use fallbacks. If we're forcing fallbacks, always
213 if (brw
->intel
.conformance_mode
== 0)
216 if (brw
->intel
.conformance_mode
== 2)
219 if (ctx
->Polygon
.SmoothFlag
) {
220 for (i
= 0; i
< nr_prims
; i
++)
221 if (reduced_prim
[prim
[i
].mode
] == GL_TRIANGLES
)
225 /* BRW hardware will do AA lines, but they are non-conformant it
226 * seems. TBD whether we keep this fallback:
228 if (ctx
->Line
.SmoothFlag
) {
229 for (i
= 0; i
< nr_prims
; i
++)
230 if (reduced_prim
[prim
[i
].mode
] == GL_LINES
)
234 /* Stipple -- these fallbacks could be resolved with a little
237 if (ctx
->Line
.StippleFlag
) {
238 for (i
= 0; i
< nr_prims
; i
++) {
239 /* GS doesn't get enough information to know when to reset
240 * the stipple counter?!?
242 if (prim
[i
].mode
== GL_LINE_LOOP
|| prim
[i
].mode
== GL_LINE_STRIP
)
245 if (prim
[i
].mode
== GL_POLYGON
&&
246 (ctx
->Polygon
.FrontMode
== GL_LINE
||
247 ctx
->Polygon
.BackMode
== GL_LINE
))
252 if (ctx
->Point
.SmoothFlag
) {
253 for (i
= 0; i
< nr_prims
; i
++)
254 if (prim
[i
].mode
== GL_POINTS
)
258 /* BRW hardware doesn't handle GL_CLAMP texturing correctly;
259 * brw_wm_sampler_state:translate_wrap_mode() treats GL_CLAMP
260 * as GL_CLAMP_TO_EDGE instead. If we're using GL_CLAMP, and
261 * we want strict conformance, force the fallback.
262 * Right now, we only do this for 2D textures.
266 for (u
= 0; u
< ctx
->Const
.MaxTextureCoordUnits
; u
++) {
267 struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[u
];
268 if (texUnit
->Enabled
) {
269 if (texUnit
->Enabled
& TEXTURE_1D_BIT
) {
270 if (texUnit
->CurrentTex
[TEXTURE_1D_INDEX
]->WrapS
== GL_CLAMP
) {
274 if (texUnit
->Enabled
& TEXTURE_2D_BIT
) {
275 if (texUnit
->CurrentTex
[TEXTURE_2D_INDEX
]->WrapS
== GL_CLAMP
||
276 texUnit
->CurrentTex
[TEXTURE_2D_INDEX
]->WrapT
== GL_CLAMP
) {
280 if (texUnit
->Enabled
& TEXTURE_3D_BIT
) {
281 if (texUnit
->CurrentTex
[TEXTURE_3D_INDEX
]->WrapS
== GL_CLAMP
||
282 texUnit
->CurrentTex
[TEXTURE_3D_INDEX
]->WrapT
== GL_CLAMP
||
283 texUnit
->CurrentTex
[TEXTURE_3D_INDEX
]->WrapR
== GL_CLAMP
) {
291 /* Nothing stopping us from the fast path now */
295 /* May fail if out of video memory for texture or vbo upload, or on
296 * fallback conditions.
298 static GLboolean
brw_try_draw_prims( GLcontext
*ctx
,
299 const struct gl_client_array
*arrays
[],
300 const struct _mesa_prim
*prim
,
302 const struct _mesa_index_buffer
*ib
,
306 struct intel_context
*intel
= intel_context(ctx
);
307 struct brw_context
*brw
= brw_context(ctx
);
308 GLboolean retval
= GL_FALSE
;
309 GLboolean warn
= GL_FALSE
;
310 GLboolean first_time
= GL_TRUE
;
314 _mesa_update_state( ctx
);
316 /* We have to validate the textures *before* checking for fallbacks;
317 * otherwise, the software fallback won't be able to rely on the
318 * texture state, the firstLevel and lastLevel fields won't be
319 * set in the intel texture object (they'll both be 0), and the
320 * software fallback will segfault if it attempts to access any
321 * texture level other than level 0.
323 brw_validate_textures( brw
);
325 if (check_fallbacks(brw
, prim
, nr_prims
))
328 /* Bind all inputs, derive varying and size information:
330 brw_merge_inputs( brw
, arrays
);
333 brw
->state
.dirty
.brw
|= BRW_NEW_INDICES
;
335 brw
->vb
.min_index
= min_index
;
336 brw
->vb
.max_index
= max_index
;
337 brw
->state
.dirty
.brw
|= BRW_NEW_VERTICES
;
339 /* Have to validate state quite late. Will rebuild tnl_program,
340 * which depends on varying information.
342 * Note this is where brw->vs->prog_data.inputs_read is calculated,
343 * so can't access it earlier.
346 LOCK_HARDWARE(intel
);
348 if (!intel
->constant_cliprect
&& intel
->driDrawable
->numClipRects
== 0) {
349 UNLOCK_HARDWARE(intel
);
353 for (i
= 0; i
< nr_prims
; i
++) {
356 /* Flush the batch if it's approaching full, so that we don't wrap while
357 * we've got validated state that needs to be in the same batch as the
358 * primitives. This fraction is just a guess (minimal full state plus
359 * a primitive is around 512 bytes), and would be better if we had
360 * an upper bound of how much we might emit in a single
361 * brw_try_draw_prims().
363 intel_batchbuffer_require_space(intel
->batch
, intel
->batch
->size
/ 4,
366 hw_prim
= brw_set_prim(brw
, prim
[i
].mode
);
368 if (first_time
|| (brw
->state
.dirty
.brw
& BRW_NEW_PRIMITIVE
)) {
369 first_time
= GL_FALSE
;
371 brw_validate_state(brw
);
373 /* Various fallback checks: */
374 if (brw
->intel
.Fallback
)
377 /* Check that we can fit our state in with our existing batchbuffer, or
380 if (dri_bufmgr_check_aperture_space(brw
->state
.validated_bos
,
381 brw
->state
.validated_bo_count
)) {
382 static GLboolean warned
;
383 intel_batchbuffer_flush(intel
->batch
);
385 /* Validate the state after we flushed the batch (which would have
386 * changed the set of dirty state). If we still fail to
387 * check_aperture, warn of what's happening, but attempt to continue
388 * on since it may succeed anyway, and the user would probably rather
389 * see a failure and a warning than a fallback.
391 brw_validate_state(brw
);
393 dri_bufmgr_check_aperture_space(brw
->state
.validated_bos
,
394 brw
->state
.validated_bo_count
)) {
400 brw_upload_state(brw
);
403 brw_emit_prim(brw
, &prim
[i
], hw_prim
);
408 if (intel
->always_flush_batch
)
409 intel_batchbuffer_flush(intel
->batch
);
411 UNLOCK_HARDWARE(intel
);
413 brw_state_cache_check_size(brw
);
416 fprintf(stderr
, "i965: Single primitive emit potentially exceeded "
417 "available aperture space\n");
420 DBG("%s failed\n", __FUNCTION__
);
425 void brw_draw_prims( GLcontext
*ctx
,
426 const struct gl_client_array
*arrays
[],
427 const struct _mesa_prim
*prim
,
429 const struct _mesa_index_buffer
*ib
,
430 GLboolean index_bounds_valid
,
436 if (!vbo_all_varyings_in_vbos(arrays
)) {
437 if (!index_bounds_valid
)
438 vbo_get_minmax_index(ctx
, prim
, ib
, &min_index
, &max_index
);
440 /* Decide if we want to rebase. If so we end up recursing once
441 * only into this function.
443 if (min_index
!= 0) {
444 vbo_rebase_prims(ctx
, arrays
,
446 ib
, min_index
, max_index
,
452 /* Make a first attempt at drawing:
454 retval
= brw_try_draw_prims(ctx
, arrays
, prim
, nr_prims
, ib
, min_index
, max_index
);
456 /* Otherwise, we really are out of memory. Pass the drawing
457 * command to the software tnl module and which will in turn call
458 * swrast to do the drawing.
461 _swsetup_Wakeup(ctx
);
462 _tnl_draw_prims(ctx
, arrays
, prim
, nr_prims
, ib
, min_index
, max_index
);
467 void brw_draw_init( struct brw_context
*brw
)
469 GLcontext
*ctx
= &brw
->intel
.ctx
;
470 struct vbo_context
*vbo
= vbo_context(ctx
);
472 /* Register our drawing function:
474 vbo
->draw_prims
= brw_draw_prims
;
477 void brw_draw_destroy( struct brw_context
*brw
)
481 if (brw
->vb
.upload
.bo
!= NULL
) {
482 dri_bo_unreference(brw
->vb
.upload
.bo
);
483 brw
->vb
.upload
.bo
= NULL
;
486 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
487 dri_bo_unreference(brw
->vb
.inputs
[i
].bo
);
488 brw
->vb
.inputs
[i
].bo
= NULL
;
491 dri_bo_unreference(brw
->ib
.bo
);