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/bufferobj.h"
31 #include "main/context.h"
32 #include "main/enums.h"
35 #include "brw_defines.h"
36 #include "brw_context.h"
37 #include "brw_state.h"
39 #include "intel_batchbuffer.h"
40 #include "intel_buffer_objects.h"
42 static GLuint double_types
[5] = {
44 BRW_SURFACEFORMAT_R64_FLOAT
,
45 BRW_SURFACEFORMAT_R64G64_FLOAT
,
46 BRW_SURFACEFORMAT_R64G64B64_FLOAT
,
47 BRW_SURFACEFORMAT_R64G64B64A64_FLOAT
50 static GLuint float_types
[5] = {
52 BRW_SURFACEFORMAT_R32_FLOAT
,
53 BRW_SURFACEFORMAT_R32G32_FLOAT
,
54 BRW_SURFACEFORMAT_R32G32B32_FLOAT
,
55 BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
58 static GLuint half_float_types
[5] = {
60 BRW_SURFACEFORMAT_R16_FLOAT
,
61 BRW_SURFACEFORMAT_R16G16_FLOAT
,
62 BRW_SURFACEFORMAT_R16G16B16A16_FLOAT
,
63 BRW_SURFACEFORMAT_R16G16B16A16_FLOAT
66 static GLuint uint_types_norm
[5] = {
68 BRW_SURFACEFORMAT_R32_UNORM
,
69 BRW_SURFACEFORMAT_R32G32_UNORM
,
70 BRW_SURFACEFORMAT_R32G32B32_UNORM
,
71 BRW_SURFACEFORMAT_R32G32B32A32_UNORM
74 static GLuint uint_types_scale
[5] = {
76 BRW_SURFACEFORMAT_R32_USCALED
,
77 BRW_SURFACEFORMAT_R32G32_USCALED
,
78 BRW_SURFACEFORMAT_R32G32B32_USCALED
,
79 BRW_SURFACEFORMAT_R32G32B32A32_USCALED
82 static GLuint int_types_norm
[5] = {
84 BRW_SURFACEFORMAT_R32_SNORM
,
85 BRW_SURFACEFORMAT_R32G32_SNORM
,
86 BRW_SURFACEFORMAT_R32G32B32_SNORM
,
87 BRW_SURFACEFORMAT_R32G32B32A32_SNORM
90 static GLuint int_types_scale
[5] = {
92 BRW_SURFACEFORMAT_R32_SSCALED
,
93 BRW_SURFACEFORMAT_R32G32_SSCALED
,
94 BRW_SURFACEFORMAT_R32G32B32_SSCALED
,
95 BRW_SURFACEFORMAT_R32G32B32A32_SSCALED
98 static GLuint ushort_types_norm
[5] = {
100 BRW_SURFACEFORMAT_R16_UNORM
,
101 BRW_SURFACEFORMAT_R16G16_UNORM
,
102 BRW_SURFACEFORMAT_R16G16B16_UNORM
,
103 BRW_SURFACEFORMAT_R16G16B16A16_UNORM
106 static GLuint ushort_types_scale
[5] = {
108 BRW_SURFACEFORMAT_R16_USCALED
,
109 BRW_SURFACEFORMAT_R16G16_USCALED
,
110 BRW_SURFACEFORMAT_R16G16B16_USCALED
,
111 BRW_SURFACEFORMAT_R16G16B16A16_USCALED
114 static GLuint short_types_norm
[5] = {
116 BRW_SURFACEFORMAT_R16_SNORM
,
117 BRW_SURFACEFORMAT_R16G16_SNORM
,
118 BRW_SURFACEFORMAT_R16G16B16_SNORM
,
119 BRW_SURFACEFORMAT_R16G16B16A16_SNORM
122 static GLuint short_types_scale
[5] = {
124 BRW_SURFACEFORMAT_R16_SSCALED
,
125 BRW_SURFACEFORMAT_R16G16_SSCALED
,
126 BRW_SURFACEFORMAT_R16G16B16_SSCALED
,
127 BRW_SURFACEFORMAT_R16G16B16A16_SSCALED
130 static GLuint ubyte_types_norm
[5] = {
132 BRW_SURFACEFORMAT_R8_UNORM
,
133 BRW_SURFACEFORMAT_R8G8_UNORM
,
134 BRW_SURFACEFORMAT_R8G8B8_UNORM
,
135 BRW_SURFACEFORMAT_R8G8B8A8_UNORM
138 static GLuint ubyte_types_scale
[5] = {
140 BRW_SURFACEFORMAT_R8_USCALED
,
141 BRW_SURFACEFORMAT_R8G8_USCALED
,
142 BRW_SURFACEFORMAT_R8G8B8_USCALED
,
143 BRW_SURFACEFORMAT_R8G8B8A8_USCALED
146 static GLuint byte_types_norm
[5] = {
148 BRW_SURFACEFORMAT_R8_SNORM
,
149 BRW_SURFACEFORMAT_R8G8_SNORM
,
150 BRW_SURFACEFORMAT_R8G8B8_SNORM
,
151 BRW_SURFACEFORMAT_R8G8B8A8_SNORM
154 static GLuint byte_types_scale
[5] = {
156 BRW_SURFACEFORMAT_R8_SSCALED
,
157 BRW_SURFACEFORMAT_R8G8_SSCALED
,
158 BRW_SURFACEFORMAT_R8G8B8_SSCALED
,
159 BRW_SURFACEFORMAT_R8G8B8A8_SSCALED
164 * Given vertex array type/size/format/normalized info, return
165 * the appopriate hardware surface type.
166 * Format will be GL_RGBA or possibly GL_BGRA for GLubyte[4] color arrays.
168 static GLuint
get_surface_type( GLenum type
, GLuint size
,
169 GLenum format
, GLboolean normalized
)
171 if (INTEL_DEBUG
& DEBUG_VERTS
)
172 printf("type %s size %d normalized %d\n",
173 _mesa_lookup_enum_by_nr(type
), size
, normalized
);
177 case GL_DOUBLE
: return double_types
[size
];
178 case GL_FLOAT
: return float_types
[size
];
179 case GL_HALF_FLOAT
: return half_float_types
[size
];
180 case GL_INT
: return int_types_norm
[size
];
181 case GL_SHORT
: return short_types_norm
[size
];
182 case GL_BYTE
: return byte_types_norm
[size
];
183 case GL_UNSIGNED_INT
: return uint_types_norm
[size
];
184 case GL_UNSIGNED_SHORT
: return ushort_types_norm
[size
];
185 case GL_UNSIGNED_BYTE
:
186 if (format
== GL_BGRA
) {
187 /* See GL_EXT_vertex_array_bgra */
189 return BRW_SURFACEFORMAT_B8G8R8A8_UNORM
;
192 return ubyte_types_norm
[size
];
194 default: assert(0); return 0;
198 assert(format
== GL_RGBA
); /* sanity check */
200 case GL_DOUBLE
: return double_types
[size
];
201 case GL_FLOAT
: return float_types
[size
];
202 case GL_HALF_FLOAT
: return half_float_types
[size
];
203 case GL_INT
: return int_types_scale
[size
];
204 case GL_SHORT
: return short_types_scale
[size
];
205 case GL_BYTE
: return byte_types_scale
[size
];
206 case GL_UNSIGNED_INT
: return uint_types_scale
[size
];
207 case GL_UNSIGNED_SHORT
: return ushort_types_scale
[size
];
208 case GL_UNSIGNED_BYTE
: return ubyte_types_scale
[size
];
209 default: assert(0); return 0;
215 static GLuint
get_size( GLenum type
)
218 case GL_DOUBLE
: return sizeof(GLdouble
);
219 case GL_FLOAT
: return sizeof(GLfloat
);
220 case GL_HALF_FLOAT
: return sizeof(GLhalfARB
);
221 case GL_INT
: return sizeof(GLint
);
222 case GL_SHORT
: return sizeof(GLshort
);
223 case GL_BYTE
: return sizeof(GLbyte
);
224 case GL_UNSIGNED_INT
: return sizeof(GLuint
);
225 case GL_UNSIGNED_SHORT
: return sizeof(GLushort
);
226 case GL_UNSIGNED_BYTE
: return sizeof(GLubyte
);
231 static GLuint
get_index_type(GLenum type
)
234 case GL_UNSIGNED_BYTE
: return BRW_INDEX_BYTE
;
235 case GL_UNSIGNED_SHORT
: return BRW_INDEX_WORD
;
236 case GL_UNSIGNED_INT
: return BRW_INDEX_DWORD
;
237 default: assert(0); return 0;
241 static void wrap_buffers( struct brw_context
*brw
,
244 if (size
< BRW_UPLOAD_INIT_SIZE
)
245 size
= BRW_UPLOAD_INIT_SIZE
;
247 brw
->vb
.upload
.offset
= 0;
249 if (brw
->vb
.upload
.bo
!= NULL
)
250 drm_intel_bo_unreference(brw
->vb
.upload
.bo
);
251 brw
->vb
.upload
.bo
= drm_intel_bo_alloc(brw
->intel
.bufmgr
, "temporary VBO",
255 static void get_space( struct brw_context
*brw
,
257 drm_intel_bo
**bo_return
,
258 GLuint
*offset_return
)
260 size
= ALIGN(size
, 64);
262 if (brw
->vb
.upload
.bo
== NULL
||
263 brw
->vb
.upload
.offset
+ size
> brw
->vb
.upload
.bo
->size
) {
264 wrap_buffers(brw
, size
);
267 assert(*bo_return
== NULL
);
268 drm_intel_bo_reference(brw
->vb
.upload
.bo
);
269 *bo_return
= brw
->vb
.upload
.bo
;
270 *offset_return
= brw
->vb
.upload
.offset
;
271 brw
->vb
.upload
.offset
+= size
;
275 copy_array_to_vbo_array( struct brw_context
*brw
,
276 struct brw_vertex_element
*element
,
279 GLuint size
= element
->count
* dst_stride
;
281 get_space(brw
, size
, &element
->bo
, &element
->offset
);
283 if (element
->glarray
->StrideB
== 0) {
284 assert(element
->count
== 1);
287 element
->stride
= dst_stride
;
290 if (dst_stride
== element
->glarray
->StrideB
) {
291 drm_intel_gem_bo_map_gtt(element
->bo
);
292 memcpy((char *)element
->bo
->virtual + element
->offset
,
293 element
->glarray
->Ptr
, size
);
294 drm_intel_gem_bo_unmap_gtt(element
->bo
);
297 const unsigned char *src
= element
->glarray
->Ptr
;
300 drm_intel_gem_bo_map_gtt(element
->bo
);
301 dest
= element
->bo
->virtual;
302 dest
+= element
->offset
;
304 for (i
= 0; i
< element
->count
; i
++) {
305 memcpy(dest
, src
, dst_stride
);
306 src
+= element
->glarray
->StrideB
;
310 drm_intel_gem_bo_unmap_gtt(element
->bo
);
314 static void brw_prepare_vertices(struct brw_context
*brw
)
316 struct gl_context
*ctx
= &brw
->intel
.ctx
;
317 struct intel_context
*intel
= intel_context(ctx
);
318 GLbitfield vs_inputs
= brw
->vs
.prog_data
->inputs_read
;
320 const unsigned char *ptr
= NULL
;
321 GLuint interleave
= 0;
322 unsigned int min_index
= brw
->vb
.min_index
;
323 unsigned int max_index
= brw
->vb
.max_index
;
325 struct brw_vertex_element
*upload
[VERT_ATTRIB_MAX
];
326 GLuint nr_uploads
= 0;
328 /* First build an array of pointers to ve's in vb.inputs_read
331 printf("%s %d..%d\n", __FUNCTION__
, min_index
, max_index
);
333 /* Accumulate the list of enabled arrays. */
334 brw
->vb
.nr_enabled
= 0;
336 GLuint i
= _mesa_ffsll(vs_inputs
) - 1;
337 struct brw_vertex_element
*input
= &brw
->vb
.inputs
[i
];
339 vs_inputs
&= ~(1 << i
);
340 brw
->vb
.enabled
[brw
->vb
.nr_enabled
++] = input
;
343 /* XXX: In the rare cases where this happens we fallback all
344 * the way to software rasterization, although a tnl fallback
345 * would be sufficient. I don't know of *any* real world
346 * cases with > 17 vertex attributes enabled, so it probably
347 * isn't an issue at this point.
349 if (brw
->vb
.nr_enabled
>= BRW_VEP_MAX
) {
350 intel
->Fallback
= GL_TRUE
; /* boolean, not bitfield */
354 for (i
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
355 struct brw_vertex_element
*input
= brw
->vb
.enabled
[i
];
357 input
->element_size
= get_size(input
->glarray
->Type
) * input
->glarray
->Size
;
359 if (_mesa_is_bufferobj(input
->glarray
->BufferObj
)) {
360 struct intel_buffer_object
*intel_buffer
=
361 intel_buffer_object(input
->glarray
->BufferObj
);
363 /* Named buffer object: Just reference its contents directly. */
364 drm_intel_bo_unreference(input
->bo
);
365 input
->bo
= intel_bufferobj_buffer(intel
, intel_buffer
,
367 drm_intel_bo_reference(input
->bo
);
368 input
->offset
= (unsigned long)input
->glarray
->Ptr
;
369 input
->stride
= input
->glarray
->StrideB
;
370 input
->count
= input
->glarray
->_MaxElement
;
372 /* This is a common place to reach if the user mistakenly supplies
373 * a pointer in place of a VBO offset. If we just let it go through,
374 * we may end up dereferencing a pointer beyond the bounds of the
375 * GTT. We would hope that the VBO's max_index would save us, but
376 * Mesa appears to hand us min/max values not clipped to the
377 * array object's _MaxElement, and _MaxElement frequently appears
378 * to be wrong anyway.
380 * The VBO spec allows application termination in this case, and it's
381 * probably a service to the poor programmer to do so rather than
382 * trying to just not render.
384 assert(input
->offset
< input
->bo
->size
);
386 input
->count
= input
->glarray
->StrideB
? max_index
+ 1 : 1;
387 if (input
->bo
!= NULL
) {
388 /* Already-uploaded vertex data is present from a previous
389 * prepare_vertices, but we had to re-validate state due to
390 * check_aperture failing and a new batch being produced.
395 /* Queue the buffer object up to be uploaded in the next pass,
396 * when we've decided if we're doing interleaved or not.
398 if (input
->attrib
== VERT_ATTRIB_POS
) {
399 /* Position array not properly enabled:
401 if (input
->glarray
->StrideB
== 0) {
402 intel
->Fallback
= GL_TRUE
; /* boolean, not bitfield */
406 interleave
= input
->glarray
->StrideB
;
407 ptr
= input
->glarray
->Ptr
;
409 else if (interleave
!= input
->glarray
->StrideB
||
410 (const unsigned char *)input
->glarray
->Ptr
- ptr
< 0 ||
411 (const unsigned char *)input
->glarray
->Ptr
- ptr
> interleave
)
416 upload
[nr_uploads
++] = input
;
420 /* Handle any arrays to be uploaded. */
421 if (nr_uploads
> 1 && interleave
&& interleave
<= 256) {
422 /* All uploads are interleaved, so upload the arrays together as
423 * interleaved. First, upload the contents and set up upload[0].
425 copy_array_to_vbo_array(brw
, upload
[0], interleave
);
427 for (i
= 1; i
< nr_uploads
; i
++) {
428 /* Then, just point upload[i] at upload[0]'s buffer. */
429 upload
[i
]->stride
= interleave
;
430 upload
[i
]->offset
= upload
[0]->offset
+
431 ((const unsigned char *)upload
[i
]->glarray
->Ptr
- ptr
);
432 upload
[i
]->bo
= upload
[0]->bo
;
433 drm_intel_bo_reference(upload
[i
]->bo
);
437 /* Upload non-interleaved arrays */
438 for (i
= 0; i
< nr_uploads
; i
++) {
439 copy_array_to_vbo_array(brw
, upload
[i
], upload
[i
]->element_size
);
443 brw_prepare_query_begin(brw
);
445 for (i
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
446 struct brw_vertex_element
*input
= brw
->vb
.enabled
[i
];
448 brw_add_validated_bo(brw
, input
->bo
);
452 static void brw_emit_vertices(struct brw_context
*brw
)
454 struct gl_context
*ctx
= &brw
->intel
.ctx
;
455 struct intel_context
*intel
= intel_context(ctx
);
458 brw_emit_query_begin(brw
);
460 /* If the VS doesn't read any inputs (calculating vertex position from
461 * a state variable for some reason, for example), emit a single pad
462 * VERTEX_ELEMENT struct and bail.
464 * The stale VB state stays in place, but they don't do anything unless
465 * a VE loads from them.
467 if (brw
->vb
.nr_enabled
== 0) {
469 OUT_BATCH((CMD_VERTEX_ELEMENT
<< 16) | 1);
470 if (intel
->gen
>= 6) {
471 OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT
) |
473 (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
<< BRW_VE0_FORMAT_SHIFT
) |
474 (0 << BRW_VE0_SRC_OFFSET_SHIFT
));
476 OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT
) |
478 (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
<< BRW_VE0_FORMAT_SHIFT
) |
479 (0 << BRW_VE0_SRC_OFFSET_SHIFT
));
481 OUT_BATCH((BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_0_SHIFT
) |
482 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_1_SHIFT
) |
483 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_2_SHIFT
) |
484 (BRW_VE1_COMPONENT_STORE_1_FLT
<< BRW_VE1_COMPONENT_3_SHIFT
));
489 /* Now emit VB and VEP state packets.
491 * This still defines a hardware VB for each input, even if they
492 * are interleaved or from the same VBO. TBD if this makes a
493 * performance difference.
495 BEGIN_BATCH(1 + brw
->vb
.nr_enabled
* 4);
496 OUT_BATCH((CMD_VERTEX_BUFFER
<< 16) |
497 ((1 + brw
->vb
.nr_enabled
* 4) - 2));
499 for (i
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
500 struct brw_vertex_element
*input
= brw
->vb
.enabled
[i
];
503 if (intel
->gen
>= 6) {
504 dw0
= GEN6_VB0_ACCESS_VERTEXDATA
|
505 (i
<< GEN6_VB0_INDEX_SHIFT
);
507 dw0
= BRW_VB0_ACCESS_VERTEXDATA
|
508 (i
<< BRW_VB0_INDEX_SHIFT
);
512 (input
->stride
<< BRW_VB0_PITCH_SHIFT
));
514 I915_GEM_DOMAIN_VERTEX
, 0,
516 if (intel
->gen
>= 5) {
518 I915_GEM_DOMAIN_VERTEX
, 0,
519 input
->bo
->size
- 1);
521 OUT_BATCH(input
->stride
? input
->count
: 0);
522 OUT_BATCH(0); /* Instance data step rate */
526 BEGIN_BATCH(1 + brw
->vb
.nr_enabled
* 2);
527 OUT_BATCH((CMD_VERTEX_ELEMENT
<< 16) | ((1 + brw
->vb
.nr_enabled
* 2) - 2));
528 for (i
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
529 struct brw_vertex_element
*input
= brw
->vb
.enabled
[i
];
530 uint32_t format
= get_surface_type(input
->glarray
->Type
,
531 input
->glarray
->Size
,
532 input
->glarray
->Format
,
533 input
->glarray
->Normalized
);
534 uint32_t comp0
= BRW_VE1_COMPONENT_STORE_SRC
;
535 uint32_t comp1
= BRW_VE1_COMPONENT_STORE_SRC
;
536 uint32_t comp2
= BRW_VE1_COMPONENT_STORE_SRC
;
537 uint32_t comp3
= BRW_VE1_COMPONENT_STORE_SRC
;
539 switch (input
->glarray
->Size
) {
540 case 0: comp0
= BRW_VE1_COMPONENT_STORE_0
;
541 case 1: comp1
= BRW_VE1_COMPONENT_STORE_0
;
542 case 2: comp2
= BRW_VE1_COMPONENT_STORE_0
;
543 case 3: comp3
= BRW_VE1_COMPONENT_STORE_1_FLT
;
547 if (intel
->gen
>= 6) {
548 OUT_BATCH((i
<< GEN6_VE0_INDEX_SHIFT
) |
550 (format
<< BRW_VE0_FORMAT_SHIFT
) |
551 (0 << BRW_VE0_SRC_OFFSET_SHIFT
));
553 OUT_BATCH((i
<< BRW_VE0_INDEX_SHIFT
) |
555 (format
<< BRW_VE0_FORMAT_SHIFT
) |
556 (0 << BRW_VE0_SRC_OFFSET_SHIFT
));
560 OUT_BATCH((comp0
<< BRW_VE1_COMPONENT_0_SHIFT
) |
561 (comp1
<< BRW_VE1_COMPONENT_1_SHIFT
) |
562 (comp2
<< BRW_VE1_COMPONENT_2_SHIFT
) |
563 (comp3
<< BRW_VE1_COMPONENT_3_SHIFT
));
565 OUT_BATCH((comp0
<< BRW_VE1_COMPONENT_0_SHIFT
) |
566 (comp1
<< BRW_VE1_COMPONENT_1_SHIFT
) |
567 (comp2
<< BRW_VE1_COMPONENT_2_SHIFT
) |
568 (comp3
<< BRW_VE1_COMPONENT_3_SHIFT
) |
569 ((i
* 4) << BRW_VE1_DST_OFFSET_SHIFT
));
574 const struct brw_tracked_state brw_vertices
= {
577 .brw
= BRW_NEW_BATCH
| BRW_NEW_VERTICES
,
580 .prepare
= brw_prepare_vertices
,
581 .emit
= brw_emit_vertices
,
584 static void brw_prepare_indices(struct brw_context
*brw
)
586 struct gl_context
*ctx
= &brw
->intel
.ctx
;
587 struct intel_context
*intel
= &brw
->intel
;
588 const struct _mesa_index_buffer
*index_buffer
= brw
->ib
.ib
;
590 drm_intel_bo
*bo
= NULL
;
591 struct gl_buffer_object
*bufferobj
;
595 if (index_buffer
== NULL
)
598 ib_type_size
= get_size(index_buffer
->type
);
599 ib_size
= ib_type_size
* index_buffer
->count
;
600 bufferobj
= index_buffer
->obj
;;
602 /* Turn into a proper VBO:
604 if (!_mesa_is_bufferobj(bufferobj
)) {
605 brw
->ib
.start_vertex_offset
= 0;
607 /* Get new bufferobj, offset:
609 get_space(brw
, ib_size
, &bo
, &offset
);
613 drm_intel_gem_bo_map_gtt(bo
);
614 memcpy((char *)bo
->virtual + offset
, index_buffer
->ptr
, ib_size
);
615 drm_intel_gem_bo_unmap_gtt(bo
);
617 offset
= (GLuint
) (unsigned long) index_buffer
->ptr
;
618 brw
->ib
.start_vertex_offset
= 0;
620 /* If the index buffer isn't aligned to its element size, we have to
621 * rebase it into a temporary.
623 if ((get_size(index_buffer
->type
) - 1) & offset
) {
624 GLubyte
*map
= ctx
->Driver
.MapBuffer(ctx
,
625 GL_ELEMENT_ARRAY_BUFFER_ARB
,
630 get_space(brw
, ib_size
, &bo
, &offset
);
632 drm_intel_bo_subdata(bo
, offset
, ib_size
, map
);
634 ctx
->Driver
.UnmapBuffer(ctx
, GL_ELEMENT_ARRAY_BUFFER_ARB
, bufferobj
);
636 bo
= intel_bufferobj_buffer(intel
, intel_buffer_object(bufferobj
),
638 drm_intel_bo_reference(bo
);
640 /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading
641 * the index buffer state when we're just moving the start index
644 brw
->ib
.start_vertex_offset
= offset
/ ib_type_size
;
650 if (brw
->ib
.bo
!= bo
||
651 brw
->ib
.offset
!= offset
||
652 brw
->ib
.size
!= ib_size
)
654 drm_intel_bo_unreference(brw
->ib
.bo
);
656 brw
->ib
.offset
= offset
;
657 brw
->ib
.size
= ib_size
;
659 brw
->state
.dirty
.brw
|= BRW_NEW_INDEX_BUFFER
;
661 drm_intel_bo_unreference(bo
);
664 brw_add_validated_bo(brw
, brw
->ib
.bo
);
667 const struct brw_tracked_state brw_indices
= {
670 .brw
= BRW_NEW_INDICES
,
673 .prepare
= brw_prepare_indices
,
676 static void brw_emit_index_buffer(struct brw_context
*brw
)
678 struct intel_context
*intel
= &brw
->intel
;
679 const struct _mesa_index_buffer
*index_buffer
= brw
->ib
.ib
;
681 if (index_buffer
== NULL
)
684 /* Emit the indexbuffer packet:
687 struct brw_indexbuffer ib
;
689 memset(&ib
, 0, sizeof(ib
));
691 ib
.header
.bits
.opcode
= CMD_INDEX_BUFFER
;
692 ib
.header
.bits
.length
= sizeof(ib
)/4 - 2;
693 ib
.header
.bits
.index_format
= get_index_type(index_buffer
->type
);
694 ib
.header
.bits
.cut_index_enable
= 0;
697 OUT_BATCH( ib
.header
.dword
);
698 OUT_RELOC(brw
->ib
.bo
,
699 I915_GEM_DOMAIN_VERTEX
, 0,
701 OUT_RELOC(brw
->ib
.bo
,
702 I915_GEM_DOMAIN_VERTEX
, 0,
703 brw
->ib
.offset
+ brw
->ib
.size
- 1);
709 const struct brw_tracked_state brw_index_buffer
= {
712 .brw
= BRW_NEW_BATCH
| BRW_NEW_INDEX_BUFFER
,
715 .emit
= brw_emit_index_buffer
,