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"
43 #include "intel_batchbuffer.h"
45 #define FILE_DEBUG_FLAG DEBUG_PRIMS
47 static GLuint prim_to_hw_prim
[GL_POLYGON
+1] = {
61 static const GLenum reduced_prim
[GL_POLYGON
+1] = {
75 /* When the primitive changes, set a state bit and re-validate. Not
76 * the nicest and would rather deal with this by having all the
77 * programs be immune to the active primitive (ie. cope with all
78 * possibilities). That may not be realistic however.
80 static GLuint
brw_set_prim(struct brw_context
*brw
,
81 const struct _mesa_prim
*prim
)
83 struct gl_context
*ctx
= &brw
->intel
.ctx
;
84 GLenum mode
= prim
->mode
;
86 DBG("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim
->mode
));
88 /* Slight optimization to avoid the GS program when not needed:
90 if (mode
== GL_QUAD_STRIP
&&
91 ctx
->Light
.ShadeModel
!= GL_FLAT
&&
92 ctx
->Polygon
.FrontMode
== GL_FILL
&&
93 ctx
->Polygon
.BackMode
== GL_FILL
)
94 mode
= GL_TRIANGLE_STRIP
;
96 if (prim
->mode
== GL_QUADS
&& prim
->count
== 4 &&
97 ctx
->Light
.ShadeModel
!= GL_FLAT
&&
98 ctx
->Polygon
.FrontMode
== GL_FILL
&&
99 ctx
->Polygon
.BackMode
== GL_FILL
) {
100 mode
= GL_TRIANGLE_FAN
;
103 if (mode
!= brw
->primitive
) {
104 brw
->primitive
= mode
;
105 brw
->state
.dirty
.brw
|= BRW_NEW_PRIMITIVE
;
107 if (reduced_prim
[mode
] != brw
->intel
.reduced_primitive
) {
108 brw
->intel
.reduced_primitive
= reduced_prim
[mode
];
109 brw
->state
.dirty
.brw
|= BRW_NEW_REDUCED_PRIMITIVE
;
113 return prim_to_hw_prim
[mode
];
117 static GLuint
trim(GLenum prim
, GLuint length
)
119 if (prim
== GL_QUAD_STRIP
)
120 return length
> 3 ? (length
- length
% 2) : 0;
121 else if (prim
== GL_QUADS
)
122 return length
- length
% 4;
128 static void brw_emit_prim(struct brw_context
*brw
,
129 const struct _mesa_prim
*prim
,
132 struct brw_3d_primitive prim_packet
;
133 struct intel_context
*intel
= &brw
->intel
;
135 DBG("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim
->mode
),
136 prim
->start
, prim
->count
);
138 prim_packet
.header
.opcode
= CMD_3D_PRIM
;
139 prim_packet
.header
.length
= sizeof(prim_packet
)/4 - 2;
140 prim_packet
.header
.pad
= 0;
141 prim_packet
.header
.topology
= hw_prim
;
142 prim_packet
.header
.indexed
= prim
->indexed
;
144 prim_packet
.verts_per_instance
= trim(prim
->mode
, prim
->count
);
145 prim_packet
.start_vert_location
= prim
->start
;
147 prim_packet
.start_vert_location
+= brw
->ib
.start_vertex_offset
;
149 prim_packet
.start_vert_location
+= brw
->vb
.start_vertex_bias
;
150 prim_packet
.instance_count
= 1;
151 prim_packet
.start_instance_location
= 0;
152 prim_packet
.base_vert_location
= prim
->basevertex
;
154 prim_packet
.base_vert_location
+= brw
->vb
.start_vertex_bias
;
157 /* If we're set to always flush, do it before and after the primitive emit.
158 * We want to catch both missed flushes that hurt instruction/state cache
159 * and missed flushes of the render cache as it heads to other parts of
160 * the besides the draw code.
162 if (intel
->always_flush_cache
) {
163 intel_batchbuffer_emit_mi_flush(intel
);
165 if (prim_packet
.verts_per_instance
) {
166 intel_batchbuffer_data(&brw
->intel
, &prim_packet
,
167 sizeof(prim_packet
), false);
169 if (intel
->always_flush_cache
) {
170 intel_batchbuffer_emit_mi_flush(intel
);
174 static void brw_merge_inputs( struct brw_context
*brw
,
175 const struct gl_client_array
*arrays
[])
177 struct brw_vertex_info old
= brw
->vb
.info
;
180 for (i
= 0; i
< brw
->vb
.nr_buffers
; i
++) {
181 drm_intel_bo_unreference(brw
->vb
.buffers
[i
].bo
);
182 brw
->vb
.buffers
[i
].bo
= NULL
;
184 brw
->vb
.nr_buffers
= 0;
186 memset(&brw
->vb
.info
, 0, sizeof(brw
->vb
.info
));
188 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++) {
189 brw
->vb
.inputs
[i
].buffer
= -1;
190 brw
->vb
.inputs
[i
].glarray
= arrays
[i
];
191 brw
->vb
.inputs
[i
].attrib
= (gl_vert_attrib
) i
;
193 if (arrays
[i
]->StrideB
!= 0)
194 brw
->vb
.info
.sizes
[i
/16] |= (brw
->vb
.inputs
[i
].glarray
->Size
- 1) <<
198 /* Raise statechanges if input sizes have changed. */
199 if (memcmp(brw
->vb
.info
.sizes
, old
.sizes
, sizeof(old
.sizes
)) != 0)
200 brw
->state
.dirty
.brw
|= BRW_NEW_INPUT_DIMENSIONS
;
203 /* XXX: could split the primitive list to fallback only on the
204 * non-conformant primitives.
206 static GLboolean
check_fallbacks( struct brw_context
*brw
,
207 const struct _mesa_prim
*prim
,
210 struct gl_context
*ctx
= &brw
->intel
.ctx
;
213 /* If we don't require strict OpenGL conformance, never
214 * use fallbacks. If we're forcing fallbacks, always
217 if (brw
->intel
.conformance_mode
== 0)
220 if (brw
->intel
.conformance_mode
== 2)
223 if (ctx
->Polygon
.SmoothFlag
) {
224 for (i
= 0; i
< nr_prims
; i
++)
225 if (reduced_prim
[prim
[i
].mode
] == GL_TRIANGLES
)
229 /* BRW hardware will do AA lines, but they are non-conformant it
230 * seems. TBD whether we keep this fallback:
232 if (ctx
->Line
.SmoothFlag
) {
233 for (i
= 0; i
< nr_prims
; i
++)
234 if (reduced_prim
[prim
[i
].mode
] == GL_LINES
)
238 /* Stipple -- these fallbacks could be resolved with a little
241 if (ctx
->Line
.StippleFlag
) {
242 for (i
= 0; i
< nr_prims
; i
++) {
243 /* GS doesn't get enough information to know when to reset
244 * the stipple counter?!?
246 if (prim
[i
].mode
== GL_LINE_LOOP
|| prim
[i
].mode
== GL_LINE_STRIP
)
249 if (prim
[i
].mode
== GL_POLYGON
&&
250 (ctx
->Polygon
.FrontMode
== GL_LINE
||
251 ctx
->Polygon
.BackMode
== GL_LINE
))
256 if (ctx
->Point
.SmoothFlag
) {
257 for (i
= 0; i
< nr_prims
; i
++)
258 if (prim
[i
].mode
== GL_POINTS
)
262 /* BRW hardware doesn't handle GL_CLAMP texturing correctly;
263 * brw_wm_sampler_state:translate_wrap_mode() treats GL_CLAMP
264 * as GL_CLAMP_TO_EDGE instead. If we're using GL_CLAMP, and
265 * we want strict conformance, force the fallback.
266 * Right now, we only do this for 2D textures.
270 for (u
= 0; u
< ctx
->Const
.MaxTextureCoordUnits
; u
++) {
271 struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[u
];
272 if (texUnit
->Enabled
) {
273 if (texUnit
->Enabled
& TEXTURE_1D_BIT
) {
274 if (texUnit
->CurrentTex
[TEXTURE_1D_INDEX
]->Sampler
.WrapS
== GL_CLAMP
) {
278 if (texUnit
->Enabled
& TEXTURE_2D_BIT
) {
279 if (texUnit
->CurrentTex
[TEXTURE_2D_INDEX
]->Sampler
.WrapS
== GL_CLAMP
||
280 texUnit
->CurrentTex
[TEXTURE_2D_INDEX
]->Sampler
.WrapT
== GL_CLAMP
) {
284 if (texUnit
->Enabled
& TEXTURE_3D_BIT
) {
285 if (texUnit
->CurrentTex
[TEXTURE_3D_INDEX
]->Sampler
.WrapS
== GL_CLAMP
||
286 texUnit
->CurrentTex
[TEXTURE_3D_INDEX
]->Sampler
.WrapT
== GL_CLAMP
||
287 texUnit
->CurrentTex
[TEXTURE_3D_INDEX
]->Sampler
.WrapR
== GL_CLAMP
) {
295 /* Nothing stopping us from the fast path now */
299 /* May fail if out of video memory for texture or vbo upload, or on
300 * fallback conditions.
302 static GLboolean
brw_try_draw_prims( struct gl_context
*ctx
,
303 const struct gl_client_array
*arrays
[],
304 const struct _mesa_prim
*prim
,
306 const struct _mesa_index_buffer
*ib
,
310 struct intel_context
*intel
= intel_context(ctx
);
311 struct brw_context
*brw
= brw_context(ctx
);
312 GLboolean retval
= GL_FALSE
;
313 GLboolean warn
= GL_FALSE
;
317 _mesa_update_state( ctx
);
319 /* We have to validate the textures *before* checking for fallbacks;
320 * otherwise, the software fallback won't be able to rely on the
321 * texture state, the firstLevel and lastLevel fields won't be
322 * set in the intel texture object (they'll both be 0), and the
323 * software fallback will segfault if it attempts to access any
324 * texture level other than level 0.
326 brw_validate_textures( brw
);
328 if (check_fallbacks(brw
, prim
, nr_prims
))
331 /* Bind all inputs, derive varying and size information:
333 brw_merge_inputs( brw
, arrays
);
336 brw
->state
.dirty
.brw
|= BRW_NEW_INDICES
;
338 brw
->vb
.min_index
= min_index
;
339 brw
->vb
.max_index
= max_index
;
340 brw
->state
.dirty
.brw
|= BRW_NEW_VERTICES
;
342 /* Have to validate state quite late. Will rebuild tnl_program,
343 * which depends on varying information.
345 * Note this is where brw->vs->prog_data.inputs_read is calculated,
346 * so can't access it earlier.
349 intel_prepare_render(intel
);
351 for (i
= 0; i
< nr_prims
; i
++) {
354 /* Flush the batch if it's approaching full, so that we don't wrap while
355 * we've got validated state that needs to be in the same batch as the
356 * primitives. This fraction is just a guess (minimal full state plus
357 * a primitive is around 512 bytes), and would be better if we had
358 * an upper bound of how much we might emit in a single
359 * brw_try_draw_prims().
361 intel_batchbuffer_require_space(intel
, 1024, false);
363 hw_prim
= brw_set_prim(brw
, &prim
[i
]);
364 if (brw
->state
.dirty
.brw
) {
365 brw_validate_state(brw
);
367 /* Various fallback checks: */
368 if (brw
->intel
.Fallback
)
371 /* Check that we can fit our state in with our existing batchbuffer, or
374 if (dri_bufmgr_check_aperture_space(brw
->state
.validated_bos
,
375 brw
->state
.validated_bo_count
)) {
376 static GLboolean warned
;
377 intel_batchbuffer_flush(intel
);
379 /* Validate the state after we flushed the batch (which would have
380 * changed the set of dirty state). If we still fail to
381 * check_aperture, warn of what's happening, but attempt to continue
382 * on since it may succeed anyway, and the user would probably rather
383 * see a failure and a warning than a fallback.
385 brw_validate_state(brw
);
387 dri_bufmgr_check_aperture_space(brw
->state
.validated_bos
,
388 brw
->state
.validated_bo_count
)) {
394 intel
->no_batch_wrap
= GL_TRUE
;
395 brw_upload_state(brw
);
398 brw_emit_prim(brw
, &prim
[i
], hw_prim
);
400 intel
->no_batch_wrap
= GL_FALSE
;
405 if (intel
->always_flush_batch
)
406 intel_batchbuffer_flush(intel
);
409 brw_state_cache_check_size(brw
);
412 fprintf(stderr
, "i965: Single primitive emit potentially exceeded "
413 "available aperture space\n");
416 DBG("%s failed\n", __FUNCTION__
);
421 void brw_draw_prims( struct gl_context
*ctx
,
422 const struct gl_client_array
*arrays
[],
423 const struct _mesa_prim
*prim
,
425 const struct _mesa_index_buffer
*ib
,
426 GLboolean index_bounds_valid
,
432 if (!vbo_all_varyings_in_vbos(arrays
)) {
433 if (!index_bounds_valid
)
434 vbo_get_minmax_index(ctx
, prim
, ib
, &min_index
, &max_index
);
436 /* Decide if we want to rebase. If so we end up recursing once
437 * only into this function.
439 if (min_index
!= 0 && !vbo_any_varyings_in_vbos(arrays
)) {
440 vbo_rebase_prims(ctx
, arrays
,
442 ib
, min_index
, max_index
,
448 /* Make a first attempt at drawing:
450 retval
= brw_try_draw_prims(ctx
, arrays
, prim
, nr_prims
, ib
, min_index
, max_index
);
452 /* Otherwise, we really are out of memory. Pass the drawing
453 * command to the software tnl module and which will in turn call
454 * swrast to do the drawing.
457 _swsetup_Wakeup(ctx
);
458 _tnl_draw_prims(ctx
, arrays
, prim
, nr_prims
, ib
, min_index
, max_index
);
463 void brw_draw_init( struct brw_context
*brw
)
465 struct gl_context
*ctx
= &brw
->intel
.ctx
;
466 struct vbo_context
*vbo
= vbo_context(ctx
);
469 /* Register our drawing function:
471 vbo
->draw_prims
= brw_draw_prims
;
473 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++)
474 brw
->vb
.inputs
[i
].buffer
= -1;
475 brw
->vb
.nr_buffers
= 0;
476 brw
->vb
.nr_enabled
= 0;
479 void brw_draw_destroy( struct brw_context
*brw
)
483 for (i
= 0; i
< brw
->vb
.nr_buffers
; i
++) {
484 drm_intel_bo_unreference(brw
->vb
.buffers
[i
].bo
);
485 brw
->vb
.buffers
[i
].bo
= NULL
;
487 brw
->vb
.nr_buffers
= 0;
489 for (i
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
490 brw
->vb
.enabled
[i
]->buffer
= -1;
492 brw
->vb
.nr_enabled
= 0;
494 drm_intel_bo_unreference(brw
->ib
.bo
);