2 * Copyright 2003 VMware, Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sublicense, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
13 * The above copyright notice and this permission notice (including the
14 * next paragraph) shall be included in all copies or substantial portions
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
19 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
20 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
21 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
22 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
23 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 #include "main/bufferobj.h"
27 #include "main/context.h"
28 #include "main/enums.h"
29 #include "main/macros.h"
30 #include "main/glformats.h"
33 #include "brw_defines.h"
34 #include "brw_context.h"
35 #include "brw_state.h"
37 #include "intel_batchbuffer.h"
38 #include "intel_buffer_objects.h"
40 static const GLuint double_types_float
[5] = {
42 BRW_SURFACEFORMAT_R64_FLOAT
,
43 BRW_SURFACEFORMAT_R64G64_FLOAT
,
44 BRW_SURFACEFORMAT_R64G64B64_FLOAT
,
45 BRW_SURFACEFORMAT_R64G64B64A64_FLOAT
48 static const GLuint double_types_passthru
[5] = {
50 BRW_SURFACEFORMAT_R64_PASSTHRU
,
51 BRW_SURFACEFORMAT_R64G64_PASSTHRU
,
52 BRW_SURFACEFORMAT_R64G64B64_PASSTHRU
,
53 BRW_SURFACEFORMAT_R64G64B64A64_PASSTHRU
56 static const GLuint float_types
[5] = {
58 BRW_SURFACEFORMAT_R32_FLOAT
,
59 BRW_SURFACEFORMAT_R32G32_FLOAT
,
60 BRW_SURFACEFORMAT_R32G32B32_FLOAT
,
61 BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
64 static const GLuint half_float_types
[5] = {
66 BRW_SURFACEFORMAT_R16_FLOAT
,
67 BRW_SURFACEFORMAT_R16G16_FLOAT
,
68 BRW_SURFACEFORMAT_R16G16B16_FLOAT
,
69 BRW_SURFACEFORMAT_R16G16B16A16_FLOAT
72 static const GLuint fixed_point_types
[5] = {
74 BRW_SURFACEFORMAT_R32_SFIXED
,
75 BRW_SURFACEFORMAT_R32G32_SFIXED
,
76 BRW_SURFACEFORMAT_R32G32B32_SFIXED
,
77 BRW_SURFACEFORMAT_R32G32B32A32_SFIXED
,
80 static const GLuint uint_types_direct
[5] = {
82 BRW_SURFACEFORMAT_R32_UINT
,
83 BRW_SURFACEFORMAT_R32G32_UINT
,
84 BRW_SURFACEFORMAT_R32G32B32_UINT
,
85 BRW_SURFACEFORMAT_R32G32B32A32_UINT
88 static const GLuint uint_types_norm
[5] = {
90 BRW_SURFACEFORMAT_R32_UNORM
,
91 BRW_SURFACEFORMAT_R32G32_UNORM
,
92 BRW_SURFACEFORMAT_R32G32B32_UNORM
,
93 BRW_SURFACEFORMAT_R32G32B32A32_UNORM
96 static const GLuint uint_types_scale
[5] = {
98 BRW_SURFACEFORMAT_R32_USCALED
,
99 BRW_SURFACEFORMAT_R32G32_USCALED
,
100 BRW_SURFACEFORMAT_R32G32B32_USCALED
,
101 BRW_SURFACEFORMAT_R32G32B32A32_USCALED
104 static const GLuint int_types_direct
[5] = {
106 BRW_SURFACEFORMAT_R32_SINT
,
107 BRW_SURFACEFORMAT_R32G32_SINT
,
108 BRW_SURFACEFORMAT_R32G32B32_SINT
,
109 BRW_SURFACEFORMAT_R32G32B32A32_SINT
112 static const GLuint int_types_norm
[5] = {
114 BRW_SURFACEFORMAT_R32_SNORM
,
115 BRW_SURFACEFORMAT_R32G32_SNORM
,
116 BRW_SURFACEFORMAT_R32G32B32_SNORM
,
117 BRW_SURFACEFORMAT_R32G32B32A32_SNORM
120 static const GLuint int_types_scale
[5] = {
122 BRW_SURFACEFORMAT_R32_SSCALED
,
123 BRW_SURFACEFORMAT_R32G32_SSCALED
,
124 BRW_SURFACEFORMAT_R32G32B32_SSCALED
,
125 BRW_SURFACEFORMAT_R32G32B32A32_SSCALED
128 static const GLuint ushort_types_direct
[5] = {
130 BRW_SURFACEFORMAT_R16_UINT
,
131 BRW_SURFACEFORMAT_R16G16_UINT
,
132 BRW_SURFACEFORMAT_R16G16B16_UINT
,
133 BRW_SURFACEFORMAT_R16G16B16A16_UINT
136 static const GLuint ushort_types_norm
[5] = {
138 BRW_SURFACEFORMAT_R16_UNORM
,
139 BRW_SURFACEFORMAT_R16G16_UNORM
,
140 BRW_SURFACEFORMAT_R16G16B16_UNORM
,
141 BRW_SURFACEFORMAT_R16G16B16A16_UNORM
144 static const GLuint ushort_types_scale
[5] = {
146 BRW_SURFACEFORMAT_R16_USCALED
,
147 BRW_SURFACEFORMAT_R16G16_USCALED
,
148 BRW_SURFACEFORMAT_R16G16B16_USCALED
,
149 BRW_SURFACEFORMAT_R16G16B16A16_USCALED
152 static const GLuint short_types_direct
[5] = {
154 BRW_SURFACEFORMAT_R16_SINT
,
155 BRW_SURFACEFORMAT_R16G16_SINT
,
156 BRW_SURFACEFORMAT_R16G16B16_SINT
,
157 BRW_SURFACEFORMAT_R16G16B16A16_SINT
160 static const GLuint short_types_norm
[5] = {
162 BRW_SURFACEFORMAT_R16_SNORM
,
163 BRW_SURFACEFORMAT_R16G16_SNORM
,
164 BRW_SURFACEFORMAT_R16G16B16_SNORM
,
165 BRW_SURFACEFORMAT_R16G16B16A16_SNORM
168 static const GLuint short_types_scale
[5] = {
170 BRW_SURFACEFORMAT_R16_SSCALED
,
171 BRW_SURFACEFORMAT_R16G16_SSCALED
,
172 BRW_SURFACEFORMAT_R16G16B16_SSCALED
,
173 BRW_SURFACEFORMAT_R16G16B16A16_SSCALED
176 static const GLuint ubyte_types_direct
[5] = {
178 BRW_SURFACEFORMAT_R8_UINT
,
179 BRW_SURFACEFORMAT_R8G8_UINT
,
180 BRW_SURFACEFORMAT_R8G8B8_UINT
,
181 BRW_SURFACEFORMAT_R8G8B8A8_UINT
184 static const GLuint ubyte_types_norm
[5] = {
186 BRW_SURFACEFORMAT_R8_UNORM
,
187 BRW_SURFACEFORMAT_R8G8_UNORM
,
188 BRW_SURFACEFORMAT_R8G8B8_UNORM
,
189 BRW_SURFACEFORMAT_R8G8B8A8_UNORM
192 static const GLuint ubyte_types_scale
[5] = {
194 BRW_SURFACEFORMAT_R8_USCALED
,
195 BRW_SURFACEFORMAT_R8G8_USCALED
,
196 BRW_SURFACEFORMAT_R8G8B8_USCALED
,
197 BRW_SURFACEFORMAT_R8G8B8A8_USCALED
200 static const GLuint byte_types_direct
[5] = {
202 BRW_SURFACEFORMAT_R8_SINT
,
203 BRW_SURFACEFORMAT_R8G8_SINT
,
204 BRW_SURFACEFORMAT_R8G8B8_SINT
,
205 BRW_SURFACEFORMAT_R8G8B8A8_SINT
208 static const GLuint byte_types_norm
[5] = {
210 BRW_SURFACEFORMAT_R8_SNORM
,
211 BRW_SURFACEFORMAT_R8G8_SNORM
,
212 BRW_SURFACEFORMAT_R8G8B8_SNORM
,
213 BRW_SURFACEFORMAT_R8G8B8A8_SNORM
216 static const GLuint byte_types_scale
[5] = {
218 BRW_SURFACEFORMAT_R8_SSCALED
,
219 BRW_SURFACEFORMAT_R8G8_SSCALED
,
220 BRW_SURFACEFORMAT_R8G8B8_SSCALED
,
221 BRW_SURFACEFORMAT_R8G8B8A8_SSCALED
225 double_types(struct brw_context
*brw
,
229 /* From the BDW PRM, Volume 2d, page 588 (VERTEX_ELEMENT_STATE):
230 * "When SourceElementFormat is set to one of the *64*_PASSTHRU formats,
231 * 64-bit components are stored in the URB without any conversion."
232 * Also included on BDW PRM, Volume 7, page 470, table "Source Element
233 * Formats Supported in VF Unit"
235 * Previous PRMs don't include those references, so for gen7 we can't use
236 * PASSTHRU formats directly. But in any case, we prefer to return passthru
237 * even in that case, because that reflects what we want to achieve, even
238 * if we would need to workaround on gen < 8.
241 ? double_types_passthru
[size
]
242 : double_types_float
[size
]);
246 is_passthru_format(uint32_t format
)
249 case BRW_SURFACEFORMAT_R64_PASSTHRU
:
250 case BRW_SURFACEFORMAT_R64G64_PASSTHRU
:
251 case BRW_SURFACEFORMAT_R64G64B64_PASSTHRU
:
252 case BRW_SURFACEFORMAT_R64G64B64A64_PASSTHRU
:
260 uploads_needed(uint32_t format
)
262 if (!is_passthru_format(format
))
266 case BRW_SURFACEFORMAT_R64_PASSTHRU
:
267 case BRW_SURFACEFORMAT_R64G64_PASSTHRU
:
269 case BRW_SURFACEFORMAT_R64G64B64_PASSTHRU
:
270 case BRW_SURFACEFORMAT_R64G64B64A64_PASSTHRU
:
273 unreachable("not reached");
278 * Returns the number of componentes associated with a format that is used on
279 * a 64 to 32 format split. See downsize_format()
282 upload_format_size(uint32_t upload_format
)
284 switch (upload_format
) {
285 case BRW_SURFACEFORMAT_R32G32_FLOAT
:
287 case BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
:
290 unreachable("not reached");
295 * Returns the format that we are finally going to use when upload a vertex
296 * element. It will only change if we are using *64*PASSTHRU formats, as for
297 * gen < 8 they need to be splitted on two *32*FLOAT formats.
299 * @upload points in which upload we are. Valid values are [0,1]
302 downsize_format_if_needed(uint32_t format
,
305 assert(upload
== 0 || upload
== 1);
307 if (!is_passthru_format(format
))
311 case BRW_SURFACEFORMAT_R64_PASSTHRU
:
312 return BRW_SURFACEFORMAT_R32G32_FLOAT
;
313 case BRW_SURFACEFORMAT_R64G64_PASSTHRU
:
314 return BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
;
315 case BRW_SURFACEFORMAT_R64G64B64_PASSTHRU
:
316 return !upload
? BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
317 : BRW_SURFACEFORMAT_R32G32_FLOAT
;
318 case BRW_SURFACEFORMAT_R64G64B64A64_PASSTHRU
:
319 return BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
;
321 unreachable("not reached");
326 * Given vertex array type/size/format/normalized info, return
327 * the appopriate hardware surface type.
328 * Format will be GL_RGBA or possibly GL_BGRA for GLubyte[4] color arrays.
331 brw_get_vertex_surface_type(struct brw_context
*brw
,
332 const struct gl_vertex_array
*glarray
)
334 int size
= glarray
->Size
;
335 const bool is_ivybridge_or_older
=
336 brw
->gen
<= 7 && !brw
->is_baytrail
&& !brw
->is_haswell
;
338 if (unlikely(INTEL_DEBUG
& DEBUG_VERTS
))
339 fprintf(stderr
, "type %s size %d normalized %d\n",
340 _mesa_enum_to_string(glarray
->Type
),
341 glarray
->Size
, glarray
->Normalized
);
343 if (glarray
->Integer
) {
344 assert(glarray
->Format
== GL_RGBA
); /* sanity check */
345 switch (glarray
->Type
) {
346 case GL_INT
: return int_types_direct
[size
];
348 if (is_ivybridge_or_older
&& size
== 3)
349 return short_types_direct
[4];
351 return short_types_direct
[size
];
353 if (is_ivybridge_or_older
&& size
== 3)
354 return byte_types_direct
[4];
356 return byte_types_direct
[size
];
357 case GL_UNSIGNED_INT
: return uint_types_direct
[size
];
358 case GL_UNSIGNED_SHORT
:
359 if (is_ivybridge_or_older
&& size
== 3)
360 return ushort_types_direct
[4];
362 return ushort_types_direct
[size
];
363 case GL_UNSIGNED_BYTE
:
364 if (is_ivybridge_or_older
&& size
== 3)
365 return ubyte_types_direct
[4];
367 return ubyte_types_direct
[size
];
368 default: unreachable("not reached");
370 } else if (glarray
->Type
== GL_UNSIGNED_INT_10F_11F_11F_REV
) {
371 return BRW_SURFACEFORMAT_R11G11B10_FLOAT
;
372 } else if (glarray
->Normalized
) {
373 switch (glarray
->Type
) {
374 case GL_DOUBLE
: return double_types(brw
, size
, glarray
->Doubles
);
375 case GL_FLOAT
: return float_types
[size
];
377 case GL_HALF_FLOAT_OES
:
378 if (brw
->gen
< 6 && size
== 3)
379 return half_float_types
[4];
381 return half_float_types
[size
];
382 case GL_INT
: return int_types_norm
[size
];
383 case GL_SHORT
: return short_types_norm
[size
];
384 case GL_BYTE
: return byte_types_norm
[size
];
385 case GL_UNSIGNED_INT
: return uint_types_norm
[size
];
386 case GL_UNSIGNED_SHORT
: return ushort_types_norm
[size
];
387 case GL_UNSIGNED_BYTE
:
388 if (glarray
->Format
== GL_BGRA
) {
389 /* See GL_EXT_vertex_array_bgra */
391 return BRW_SURFACEFORMAT_B8G8R8A8_UNORM
;
394 return ubyte_types_norm
[size
];
397 if (brw
->gen
>= 8 || brw
->is_haswell
)
398 return fixed_point_types
[size
];
400 /* This produces GL_FIXED inputs as values between INT32_MIN and
401 * INT32_MAX, which will be scaled down by 1/65536 by the VS.
403 return int_types_scale
[size
];
404 /* See GL_ARB_vertex_type_2_10_10_10_rev.
405 * W/A: Pre-Haswell, the hardware doesn't really support the formats we'd
406 * like to use here, so upload everything as UINT and fix
409 case GL_INT_2_10_10_10_REV
:
411 if (brw
->gen
>= 8 || brw
->is_haswell
) {
412 return glarray
->Format
== GL_BGRA
413 ? BRW_SURFACEFORMAT_B10G10R10A2_SNORM
414 : BRW_SURFACEFORMAT_R10G10B10A2_SNORM
;
416 return BRW_SURFACEFORMAT_R10G10B10A2_UINT
;
417 case GL_UNSIGNED_INT_2_10_10_10_REV
:
419 if (brw
->gen
>= 8 || brw
->is_haswell
) {
420 return glarray
->Format
== GL_BGRA
421 ? BRW_SURFACEFORMAT_B10G10R10A2_UNORM
422 : BRW_SURFACEFORMAT_R10G10B10A2_UNORM
;
424 return BRW_SURFACEFORMAT_R10G10B10A2_UINT
;
425 default: unreachable("not reached");
429 /* See GL_ARB_vertex_type_2_10_10_10_rev.
430 * W/A: the hardware doesn't really support the formats we'd
431 * like to use here, so upload everything as UINT and fix
434 if (glarray
->Type
== GL_INT_2_10_10_10_REV
) {
436 if (brw
->gen
>= 8 || brw
->is_haswell
) {
437 return glarray
->Format
== GL_BGRA
438 ? BRW_SURFACEFORMAT_B10G10R10A2_SSCALED
439 : BRW_SURFACEFORMAT_R10G10B10A2_SSCALED
;
441 return BRW_SURFACEFORMAT_R10G10B10A2_UINT
;
442 } else if (glarray
->Type
== GL_UNSIGNED_INT_2_10_10_10_REV
) {
444 if (brw
->gen
>= 8 || brw
->is_haswell
) {
445 return glarray
->Format
== GL_BGRA
446 ? BRW_SURFACEFORMAT_B10G10R10A2_USCALED
447 : BRW_SURFACEFORMAT_R10G10B10A2_USCALED
;
449 return BRW_SURFACEFORMAT_R10G10B10A2_UINT
;
451 assert(glarray
->Format
== GL_RGBA
); /* sanity check */
452 switch (glarray
->Type
) {
453 case GL_DOUBLE
: return double_types(brw
, size
, glarray
->Doubles
);
454 case GL_FLOAT
: return float_types
[size
];
456 case GL_HALF_FLOAT_OES
:
457 if (brw
->gen
< 6 && size
== 3)
458 return half_float_types
[4];
460 return half_float_types
[size
];
461 case GL_INT
: return int_types_scale
[size
];
462 case GL_SHORT
: return short_types_scale
[size
];
463 case GL_BYTE
: return byte_types_scale
[size
];
464 case GL_UNSIGNED_INT
: return uint_types_scale
[size
];
465 case GL_UNSIGNED_SHORT
: return ushort_types_scale
[size
];
466 case GL_UNSIGNED_BYTE
: return ubyte_types_scale
[size
];
468 if (brw
->gen
>= 8 || brw
->is_haswell
)
469 return fixed_point_types
[size
];
471 /* This produces GL_FIXED inputs as values between INT32_MIN and
472 * INT32_MAX, which will be scaled down by 1/65536 by the VS.
474 return int_types_scale
[size
];
475 default: unreachable("not reached");
481 copy_array_to_vbo_array(struct brw_context
*brw
,
482 struct brw_vertex_element
*element
,
484 struct brw_vertex_buffer
*buffer
,
487 const int src_stride
= element
->glarray
->StrideB
;
489 /* If the source stride is zero, we just want to upload the current
490 * attribute once and set the buffer's stride to 0. There's no need
491 * to replicate it out.
493 if (src_stride
== 0) {
494 intel_upload_data(brw
, element
->glarray
->Ptr
,
495 element
->glarray
->_ElementSize
,
496 element
->glarray
->_ElementSize
,
497 &buffer
->bo
, &buffer
->offset
);
500 buffer
->size
= element
->glarray
->_ElementSize
;
504 const unsigned char *src
= element
->glarray
->Ptr
+ min
* src_stride
;
505 int count
= max
- min
+ 1;
506 GLuint size
= count
* dst_stride
;
507 uint8_t *dst
= intel_upload_space(brw
, size
, dst_stride
,
508 &buffer
->bo
, &buffer
->offset
);
510 /* The GL 4.5 spec says:
511 * "If any enabled array’s buffer binding is zero when DrawArrays or
512 * one of the other drawing commands defined in section 10.4 is called,
513 * the result is undefined."
515 * In this case, let's the dst with undefined values
518 if (dst_stride
== src_stride
) {
519 memcpy(dst
, src
, size
);
522 memcpy(dst
, src
, dst_stride
);
528 buffer
->stride
= dst_stride
;
533 brw_prepare_vertices(struct brw_context
*brw
)
535 struct gl_context
*ctx
= &brw
->ctx
;
536 /* BRW_NEW_VS_PROG_DATA */
537 const struct brw_vs_prog_data
*vs_prog_data
=
538 brw_vs_prog_data(brw
->vs
.base
.prog_data
);
539 GLbitfield64 vs_inputs
= vs_prog_data
->inputs_read
;
540 const unsigned char *ptr
= NULL
;
541 GLuint interleaved
= 0;
542 unsigned int min_index
= brw
->vb
.min_index
+ brw
->basevertex
;
543 unsigned int max_index
= brw
->vb
.max_index
+ brw
->basevertex
;
547 struct brw_vertex_element
*upload
[VERT_ATTRIB_MAX
];
548 GLuint nr_uploads
= 0;
552 * On gen6+, edge flags don't end up in the VUE (either in or out of the
553 * VS). Instead, they're uploaded as the last vertex element, and the data
554 * is passed sideband through the fixed function units. So, we need to
555 * prepare the vertex buffer for it, but it's not present in inputs_read.
557 if (brw
->gen
>= 6 && (ctx
->Polygon
.FrontMode
!= GL_FILL
||
558 ctx
->Polygon
.BackMode
!= GL_FILL
)) {
559 vs_inputs
|= VERT_BIT_EDGEFLAG
;
563 fprintf(stderr
, "%s %d..%d\n", __func__
, min_index
, max_index
);
565 /* Accumulate the list of enabled arrays. */
566 brw
->vb
.nr_enabled
= 0;
568 GLuint first
= ffsll(vs_inputs
) - 1;
570 first
- DIV_ROUND_UP(_mesa_bitcount_64(vs_prog_data
->double_inputs_read
&
571 BITFIELD64_MASK(first
)), 2);
572 struct brw_vertex_element
*input
= &brw
->vb
.inputs
[index
];
573 input
->is_dual_slot
= (vs_prog_data
->double_inputs_read
& BITFIELD64_BIT(first
)) != 0;
574 vs_inputs
&= ~BITFIELD64_BIT(first
);
575 if (input
->is_dual_slot
)
576 vs_inputs
&= ~BITFIELD64_BIT(first
+ 1);
577 brw
->vb
.enabled
[brw
->vb
.nr_enabled
++] = input
;
580 if (brw
->vb
.nr_enabled
== 0)
583 if (brw
->vb
.nr_buffers
)
586 /* The range of data in a given buffer represented as [min, max) */
587 struct intel_buffer_object
*enabled_buffer
[VERT_ATTRIB_MAX
];
588 uint32_t buffer_range_start
[VERT_ATTRIB_MAX
];
589 uint32_t buffer_range_end
[VERT_ATTRIB_MAX
];
591 for (i
= j
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
592 struct brw_vertex_element
*input
= brw
->vb
.enabled
[i
];
593 const struct gl_vertex_array
*glarray
= input
->glarray
;
595 if (_mesa_is_bufferobj(glarray
->BufferObj
)) {
596 struct intel_buffer_object
*intel_buffer
=
597 intel_buffer_object(glarray
->BufferObj
);
599 const uint32_t offset
= (uintptr_t)glarray
->Ptr
;
601 /* Start with the worst case */
603 uint32_t range
= intel_buffer
->Base
.Size
;
604 if (glarray
->InstanceDivisor
) {
605 if (brw
->num_instances
) {
606 start
= offset
+ glarray
->StrideB
* brw
->baseinstance
;
607 range
= (glarray
->StrideB
* ((brw
->num_instances
- 1) /
608 glarray
->InstanceDivisor
) +
609 glarray
->_ElementSize
);
612 if (brw
->vb
.index_bounds_valid
) {
613 start
= offset
+ min_index
* glarray
->StrideB
;
614 range
= (glarray
->StrideB
* (max_index
- min_index
) +
615 glarray
->_ElementSize
);
619 /* If we have a VB set to be uploaded for this buffer object
620 * already, reuse that VB state so that we emit fewer
624 for (k
= 0; k
< i
; k
++) {
625 const struct gl_vertex_array
*other
= brw
->vb
.enabled
[k
]->glarray
;
626 if (glarray
->BufferObj
== other
->BufferObj
&&
627 glarray
->StrideB
== other
->StrideB
&&
628 glarray
->InstanceDivisor
== other
->InstanceDivisor
&&
629 (uintptr_t)(glarray
->Ptr
- other
->Ptr
) < glarray
->StrideB
)
631 input
->buffer
= brw
->vb
.enabled
[k
]->buffer
;
632 input
->offset
= glarray
->Ptr
- other
->Ptr
;
634 buffer_range_start
[input
->buffer
] =
635 MIN2(buffer_range_start
[input
->buffer
], start
);
636 buffer_range_end
[input
->buffer
] =
637 MAX2(buffer_range_end
[input
->buffer
], start
+ range
);
642 struct brw_vertex_buffer
*buffer
= &brw
->vb
.buffers
[j
];
644 /* Named buffer object: Just reference its contents directly. */
645 buffer
->offset
= offset
;
646 buffer
->stride
= glarray
->StrideB
;
647 buffer
->step_rate
= glarray
->InstanceDivisor
;
648 buffer
->size
= glarray
->BufferObj
->Size
- offset
;
650 enabled_buffer
[j
] = intel_buffer
;
651 buffer_range_start
[j
] = start
;
652 buffer_range_end
[j
] = start
+ range
;
658 /* Queue the buffer object up to be uploaded in the next pass,
659 * when we've decided if we're doing interleaved or not.
661 if (nr_uploads
== 0) {
662 interleaved
= glarray
->StrideB
;
665 else if (interleaved
!= glarray
->StrideB
||
666 glarray
->Ptr
< ptr
||
667 (uintptr_t)(glarray
->Ptr
- ptr
) + glarray
->_ElementSize
> interleaved
)
669 /* If our stride is different from the first attribute's stride,
670 * or if the first attribute's stride didn't cover our element,
671 * disable the interleaved upload optimization. The second case
672 * can most commonly occur in cases where there is a single vertex
673 * and, for example, the data is stored on the application's
676 * NOTE: This will also disable the optimization in cases where
677 * the data is in a different order than the array indices.
681 * glVertexAttribPointer(0, 4, GL_FLOAT, 32, &data[4]);
682 * glVertexAttribPointer(1, 4, GL_FLOAT, 32, &data[0]);
687 upload
[nr_uploads
++] = input
;
691 /* Now that we've set up all of the buffers, we walk through and reference
692 * each of them. We do this late so that we get the right size in each
693 * buffer and don't reference too little data.
695 for (i
= 0; i
< j
; i
++) {
696 struct brw_vertex_buffer
*buffer
= &brw
->vb
.buffers
[i
];
700 const uint32_t start
= buffer_range_start
[i
];
701 const uint32_t range
= buffer_range_end
[i
] - buffer_range_start
[i
];
703 buffer
->bo
= intel_bufferobj_buffer(brw
, enabled_buffer
[i
], start
, range
);
704 drm_intel_bo_reference(buffer
->bo
);
707 /* If we need to upload all the arrays, then we can trim those arrays to
708 * only the used elements [min_index, max_index] so long as we adjust all
709 * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
711 brw
->vb
.start_vertex_bias
= 0;
713 if (nr_uploads
== brw
->vb
.nr_enabled
) {
714 brw
->vb
.start_vertex_bias
= -delta
;
718 /* Handle any arrays to be uploaded. */
719 if (nr_uploads
> 1) {
721 struct brw_vertex_buffer
*buffer
= &brw
->vb
.buffers
[j
];
722 /* All uploads are interleaved, so upload the arrays together as
723 * interleaved. First, upload the contents and set up upload[0].
725 copy_array_to_vbo_array(brw
, upload
[0], min_index
, max_index
,
726 buffer
, interleaved
);
727 buffer
->offset
-= delta
* interleaved
;
728 buffer
->size
+= delta
* interleaved
;
730 for (i
= 0; i
< nr_uploads
; i
++) {
731 /* Then, just point upload[i] at upload[0]'s buffer. */
733 ((const unsigned char *)upload
[i
]->glarray
->Ptr
- ptr
);
734 upload
[i
]->buffer
= j
;
741 /* Upload non-interleaved arrays */
742 for (i
= 0; i
< nr_uploads
; i
++) {
743 struct brw_vertex_buffer
*buffer
= &brw
->vb
.buffers
[j
];
744 if (upload
[i
]->glarray
->InstanceDivisor
== 0) {
745 copy_array_to_vbo_array(brw
, upload
[i
], min_index
, max_index
,
746 buffer
, upload
[i
]->glarray
->_ElementSize
);
748 /* This is an instanced attribute, since its InstanceDivisor
749 * is not zero. Therefore, its data will be stepped after the
750 * instanced draw has been run InstanceDivisor times.
752 uint32_t instanced_attr_max_index
=
753 (brw
->num_instances
- 1) / upload
[i
]->glarray
->InstanceDivisor
;
754 copy_array_to_vbo_array(brw
, upload
[i
], 0, instanced_attr_max_index
,
755 buffer
, upload
[i
]->glarray
->_ElementSize
);
757 buffer
->offset
-= delta
* buffer
->stride
;
758 buffer
->size
+= delta
* buffer
->stride
;
759 buffer
->step_rate
= upload
[i
]->glarray
->InstanceDivisor
;
760 upload
[i
]->buffer
= j
++;
761 upload
[i
]->offset
= 0;
764 brw
->vb
.nr_buffers
= j
;
768 brw_prepare_shader_draw_parameters(struct brw_context
*brw
)
770 const struct brw_vs_prog_data
*vs_prog_data
=
771 brw_vs_prog_data(brw
->vs
.base
.prog_data
);
773 /* For non-indirect draws, upload gl_BaseVertex. */
774 if ((vs_prog_data
->uses_basevertex
|| vs_prog_data
->uses_baseinstance
) &&
775 brw
->draw
.draw_params_bo
== NULL
) {
776 intel_upload_data(brw
, &brw
->draw
.params
, sizeof(brw
->draw
.params
), 4,
777 &brw
->draw
.draw_params_bo
,
778 &brw
->draw
.draw_params_offset
);
781 if (vs_prog_data
->uses_drawid
) {
782 intel_upload_data(brw
, &brw
->draw
.gl_drawid
, sizeof(brw
->draw
.gl_drawid
), 4,
783 &brw
->draw
.draw_id_bo
,
784 &brw
->draw
.draw_id_offset
);
789 * Emit a VERTEX_BUFFER_STATE entry (part of 3DSTATE_VERTEX_BUFFERS).
792 brw_emit_vertex_buffer_state(struct brw_context
*brw
,
795 unsigned start_offset
,
801 struct gl_context
*ctx
= &brw
->ctx
;
805 dw0
= buffer_nr
<< GEN6_VB0_INDEX_SHIFT
;
806 } else if (brw
->gen
>= 6) {
807 dw0
= (buffer_nr
<< GEN6_VB0_INDEX_SHIFT
) |
808 (step_rate
? GEN6_VB0_ACCESS_INSTANCEDATA
809 : GEN6_VB0_ACCESS_VERTEXDATA
);
811 dw0
= (buffer_nr
<< BRW_VB0_INDEX_SHIFT
) |
812 (step_rate
? BRW_VB0_ACCESS_INSTANCEDATA
813 : BRW_VB0_ACCESS_VERTEXDATA
);
817 dw0
|= GEN7_VB0_ADDRESS_MODIFYENABLE
;
821 dw0
|= GEN7_MOCS_L3
<< 16;
824 dw0
|= BDW_MOCS_WB
<< 16;
827 dw0
|= SKL_MOCS_WB
<< 16;
831 WARN_ONCE(stride
>= (brw
->gen
>= 5 ? 2048 : 2047),
832 "VBO stride %d too large, bad rendering may occur\n",
834 OUT_BATCH(dw0
| (stride
<< BRW_VB0_PITCH_SHIFT
));
836 OUT_RELOC64(bo
, I915_GEM_DOMAIN_VERTEX
, 0, start_offset
);
837 /* From the BSpec: 3D Pipeline Stages - 3D Pipeline Geometry -
838 * Vertex Fetch (VF) Stage - State
840 * Instead of "VBState.StartingBufferAddress + VBState.MaxIndex x
841 * VBState.BufferPitch", the address of the byte immediately beyond the
842 * last valid byte of the buffer is determined by
843 * "VBState.StartingBufferAddress + VBState.BufferSize".
845 OUT_BATCH(end_offset
- start_offset
);
846 } else if (brw
->gen
>= 5) {
847 OUT_RELOC(bo
, I915_GEM_DOMAIN_VERTEX
, 0, start_offset
);
848 /* From the BSpec: 3D Pipeline Stages - 3D Pipeline Geometry -
849 * Vertex Fetch (VF) Stage - State
851 * Instead of "VBState.StartingBufferAddress + VBState.MaxIndex x
852 * VBState.BufferPitch", the address of the byte immediately beyond the
853 * last valid byte of the buffer is determined by
854 * "VBState.EndAddress + 1".
856 OUT_RELOC(bo
, I915_GEM_DOMAIN_VERTEX
, 0, end_offset
- 1);
857 OUT_BATCH(step_rate
);
859 OUT_RELOC(bo
, I915_GEM_DOMAIN_VERTEX
, 0, start_offset
);
861 OUT_BATCH(step_rate
);
868 brw_emit_vertices(struct brw_context
*brw
)
872 brw_prepare_vertices(brw
);
873 brw_prepare_shader_draw_parameters(brw
);
875 brw_emit_query_begin(brw
);
877 const struct brw_vs_prog_data
*vs_prog_data
=
878 brw_vs_prog_data(brw
->vs
.base
.prog_data
);
880 unsigned nr_elements
= brw
->vb
.nr_enabled
;
881 if (vs_prog_data
->uses_vertexid
|| vs_prog_data
->uses_instanceid
||
882 vs_prog_data
->uses_basevertex
|| vs_prog_data
->uses_baseinstance
)
884 if (vs_prog_data
->uses_drawid
)
887 /* If any of the formats of vb.enabled needs more that one upload, we need
888 * to add it to nr_elements */
889 unsigned extra_uploads
= 0;
890 for (unsigned i
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
891 struct brw_vertex_element
*input
= brw
->vb
.enabled
[i
];
892 uint32_t format
= brw_get_vertex_surface_type(brw
, input
->glarray
);
894 if (uploads_needed(format
) > 1)
897 nr_elements
+= extra_uploads
;
899 /* If the VS doesn't read any inputs (calculating vertex position from
900 * a state variable for some reason, for example), emit a single pad
901 * VERTEX_ELEMENT struct and bail.
903 * The stale VB state stays in place, but they don't do anything unless
904 * a VE loads from them.
906 if (nr_elements
== 0) {
908 OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS
<< 16) | 1);
910 OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT
) |
912 (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
<< BRW_VE0_FORMAT_SHIFT
) |
913 (0 << BRW_VE0_SRC_OFFSET_SHIFT
));
915 OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT
) |
917 (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
<< BRW_VE0_FORMAT_SHIFT
) |
918 (0 << BRW_VE0_SRC_OFFSET_SHIFT
));
920 OUT_BATCH((BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_0_SHIFT
) |
921 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_1_SHIFT
) |
922 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_2_SHIFT
) |
923 (BRW_VE1_COMPONENT_STORE_1_FLT
<< BRW_VE1_COMPONENT_3_SHIFT
));
928 /* Now emit VB and VEP state packets.
931 const bool uses_draw_params
=
932 vs_prog_data
->uses_basevertex
||
933 vs_prog_data
->uses_baseinstance
;
934 const unsigned nr_buffers
= brw
->vb
.nr_buffers
+
935 uses_draw_params
+ vs_prog_data
->uses_drawid
;
939 assert(nr_buffers
<= 33);
941 assert(nr_buffers
<= 17);
944 BEGIN_BATCH(1 + 4 * nr_buffers
);
945 OUT_BATCH((_3DSTATE_VERTEX_BUFFERS
<< 16) | (4 * nr_buffers
- 1));
946 for (i
= 0; i
< brw
->vb
.nr_buffers
; i
++) {
947 struct brw_vertex_buffer
*buffer
= &brw
->vb
.buffers
[i
];
948 /* Prior to Haswell and Bay Trail we have to use 4-component formats
949 * to fake 3-component ones. In particular, we do this for
950 * half-float and 8 and 16-bit integer formats. This means that the
951 * vertex element may poke over the end of the buffer by 2 bytes.
954 (brw
->gen
<= 7 && !brw
->is_baytrail
&& !brw
->is_haswell
) * 2;
955 EMIT_VERTEX_BUFFER_STATE(brw
, i
, buffer
->bo
, buffer
->offset
,
956 buffer
->offset
+ buffer
->size
+ padding
,
957 buffer
->stride
, buffer
->step_rate
);
961 if (uses_draw_params
) {
962 EMIT_VERTEX_BUFFER_STATE(brw
, brw
->vb
.nr_buffers
,
963 brw
->draw
.draw_params_bo
,
964 brw
->draw
.draw_params_offset
,
965 brw
->draw
.draw_params_bo
->size
,
970 if (vs_prog_data
->uses_drawid
) {
971 EMIT_VERTEX_BUFFER_STATE(brw
, brw
->vb
.nr_buffers
+ 1,
972 brw
->draw
.draw_id_bo
,
973 brw
->draw
.draw_id_offset
,
974 brw
->draw
.draw_id_bo
->size
,
982 /* The hardware allows one more VERTEX_ELEMENTS than VERTEX_BUFFERS, presumably
983 * for VertexID/InstanceID.
986 assert(nr_elements
<= 34);
988 assert(nr_elements
<= 18);
991 struct brw_vertex_element
*gen6_edgeflag_input
= NULL
;
993 BEGIN_BATCH(1 + nr_elements
* 2);
994 OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS
<< 16) | (2 * nr_elements
- 1));
995 for (i
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
996 struct brw_vertex_element
*input
= brw
->vb
.enabled
[i
];
997 uint32_t format
= brw_get_vertex_surface_type(brw
, input
->glarray
);
998 uint32_t comp0
= BRW_VE1_COMPONENT_STORE_SRC
;
999 uint32_t comp1
= BRW_VE1_COMPONENT_STORE_SRC
;
1000 uint32_t comp2
= BRW_VE1_COMPONENT_STORE_SRC
;
1001 uint32_t comp3
= BRW_VE1_COMPONENT_STORE_SRC
;
1002 unsigned num_uploads
= 1;
1005 num_uploads
= uploads_needed(format
);
1007 if (input
== &brw
->vb
.inputs
[VERT_ATTRIB_EDGEFLAG
]) {
1008 /* Gen6+ passes edgeflag as sideband along with the vertex, instead
1009 * of in the VUE. We have to upload it sideband as the last vertex
1010 * element according to the B-Spec.
1012 if (brw
->gen
>= 6) {
1013 gen6_edgeflag_input
= input
;
1018 for (c
= 0; c
< num_uploads
; c
++) {
1019 uint32_t upload_format
= downsize_format_if_needed(format
, c
);
1020 /* If we need more that one upload, the offset stride would be 128
1021 * bits (16 bytes), as for previous uploads we are using the full
1023 unsigned int offset
= input
->offset
+ c
* 16;
1024 int size
= input
->glarray
->Size
;
1026 if (is_passthru_format(format
))
1027 size
= upload_format_size(upload_format
);
1030 case 0: comp0
= BRW_VE1_COMPONENT_STORE_0
;
1031 case 1: comp1
= BRW_VE1_COMPONENT_STORE_0
;
1032 case 2: comp2
= BRW_VE1_COMPONENT_STORE_0
;
1033 case 3: comp3
= input
->glarray
->Integer
1034 ? BRW_VE1_COMPONENT_STORE_1_INT
1035 : BRW_VE1_COMPONENT_STORE_1_FLT
;
1039 if (brw
->gen
>= 6) {
1040 OUT_BATCH((input
->buffer
<< GEN6_VE0_INDEX_SHIFT
) |
1042 (upload_format
<< BRW_VE0_FORMAT_SHIFT
) |
1043 (offset
<< BRW_VE0_SRC_OFFSET_SHIFT
));
1045 OUT_BATCH((input
->buffer
<< BRW_VE0_INDEX_SHIFT
) |
1047 (upload_format
<< BRW_VE0_FORMAT_SHIFT
) |
1048 (offset
<< BRW_VE0_SRC_OFFSET_SHIFT
));
1052 OUT_BATCH((comp0
<< BRW_VE1_COMPONENT_0_SHIFT
) |
1053 (comp1
<< BRW_VE1_COMPONENT_1_SHIFT
) |
1054 (comp2
<< BRW_VE1_COMPONENT_2_SHIFT
) |
1055 (comp3
<< BRW_VE1_COMPONENT_3_SHIFT
));
1057 OUT_BATCH((comp0
<< BRW_VE1_COMPONENT_0_SHIFT
) |
1058 (comp1
<< BRW_VE1_COMPONENT_1_SHIFT
) |
1059 (comp2
<< BRW_VE1_COMPONENT_2_SHIFT
) |
1060 (comp3
<< BRW_VE1_COMPONENT_3_SHIFT
) |
1061 ((i
* 4) << BRW_VE1_DST_OFFSET_SHIFT
));
1065 if (vs_prog_data
->uses_vertexid
|| vs_prog_data
->uses_instanceid
||
1066 vs_prog_data
->uses_basevertex
|| vs_prog_data
->uses_baseinstance
) {
1067 uint32_t dw0
= 0, dw1
= 0;
1068 uint32_t comp0
= BRW_VE1_COMPONENT_STORE_0
;
1069 uint32_t comp1
= BRW_VE1_COMPONENT_STORE_0
;
1070 uint32_t comp2
= BRW_VE1_COMPONENT_STORE_0
;
1071 uint32_t comp3
= BRW_VE1_COMPONENT_STORE_0
;
1073 if (vs_prog_data
->uses_basevertex
)
1074 comp0
= BRW_VE1_COMPONENT_STORE_SRC
;
1076 if (vs_prog_data
->uses_baseinstance
)
1077 comp1
= BRW_VE1_COMPONENT_STORE_SRC
;
1079 if (vs_prog_data
->uses_vertexid
)
1080 comp2
= BRW_VE1_COMPONENT_STORE_VID
;
1082 if (vs_prog_data
->uses_instanceid
)
1083 comp3
= BRW_VE1_COMPONENT_STORE_IID
;
1085 dw1
= (comp0
<< BRW_VE1_COMPONENT_0_SHIFT
) |
1086 (comp1
<< BRW_VE1_COMPONENT_1_SHIFT
) |
1087 (comp2
<< BRW_VE1_COMPONENT_2_SHIFT
) |
1088 (comp3
<< BRW_VE1_COMPONENT_3_SHIFT
);
1090 if (brw
->gen
>= 6) {
1091 dw0
|= GEN6_VE0_VALID
|
1092 brw
->vb
.nr_buffers
<< GEN6_VE0_INDEX_SHIFT
|
1093 BRW_SURFACEFORMAT_R32G32_UINT
<< BRW_VE0_FORMAT_SHIFT
;
1095 dw0
|= BRW_VE0_VALID
|
1096 brw
->vb
.nr_buffers
<< BRW_VE0_INDEX_SHIFT
|
1097 BRW_SURFACEFORMAT_R32G32_UINT
<< BRW_VE0_FORMAT_SHIFT
;
1098 dw1
|= (i
* 4) << BRW_VE1_DST_OFFSET_SHIFT
;
1101 /* Note that for gl_VertexID, gl_InstanceID, and gl_PrimitiveID values,
1102 * the format is ignored and the value is always int.
1109 if (vs_prog_data
->uses_drawid
) {
1110 uint32_t dw0
= 0, dw1
= 0;
1112 dw1
= (BRW_VE1_COMPONENT_STORE_SRC
<< BRW_VE1_COMPONENT_0_SHIFT
) |
1113 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_1_SHIFT
) |
1114 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_2_SHIFT
) |
1115 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_3_SHIFT
);
1117 if (brw
->gen
>= 6) {
1118 dw0
|= GEN6_VE0_VALID
|
1119 ((brw
->vb
.nr_buffers
+ 1) << GEN6_VE0_INDEX_SHIFT
) |
1120 (BRW_SURFACEFORMAT_R32_UINT
<< BRW_VE0_FORMAT_SHIFT
);
1122 dw0
|= BRW_VE0_VALID
|
1123 ((brw
->vb
.nr_buffers
+ 1) << BRW_VE0_INDEX_SHIFT
) |
1124 (BRW_SURFACEFORMAT_R32_UINT
<< BRW_VE0_FORMAT_SHIFT
);
1126 dw1
|= (i
* 4) << BRW_VE1_DST_OFFSET_SHIFT
;
1133 if (brw
->gen
>= 6 && gen6_edgeflag_input
) {
1135 brw_get_vertex_surface_type(brw
, gen6_edgeflag_input
->glarray
);
1137 OUT_BATCH((gen6_edgeflag_input
->buffer
<< GEN6_VE0_INDEX_SHIFT
) |
1139 GEN6_VE0_EDGE_FLAG_ENABLE
|
1140 (format
<< BRW_VE0_FORMAT_SHIFT
) |
1141 (gen6_edgeflag_input
->offset
<< BRW_VE0_SRC_OFFSET_SHIFT
));
1142 OUT_BATCH((BRW_VE1_COMPONENT_STORE_SRC
<< BRW_VE1_COMPONENT_0_SHIFT
) |
1143 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_1_SHIFT
) |
1144 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_2_SHIFT
) |
1145 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_3_SHIFT
));
1151 const struct brw_tracked_state brw_vertices
= {
1153 .mesa
= _NEW_POLYGON
,
1154 .brw
= BRW_NEW_BATCH
|
1157 BRW_NEW_VS_PROG_DATA
,
1159 .emit
= brw_emit_vertices
,
1163 brw_upload_indices(struct brw_context
*brw
)
1165 struct gl_context
*ctx
= &brw
->ctx
;
1166 const struct _mesa_index_buffer
*index_buffer
= brw
->ib
.ib
;
1168 drm_intel_bo
*old_bo
= brw
->ib
.bo
;
1169 struct gl_buffer_object
*bufferobj
;
1171 GLuint ib_type_size
;
1173 if (index_buffer
== NULL
)
1176 ib_type_size
= _mesa_sizeof_type(index_buffer
->type
);
1177 ib_size
= index_buffer
->count
? ib_type_size
* index_buffer
->count
:
1178 index_buffer
->obj
->Size
;
1179 bufferobj
= index_buffer
->obj
;
1181 /* Turn into a proper VBO:
1183 if (!_mesa_is_bufferobj(bufferobj
)) {
1184 /* Get new bufferobj, offset:
1186 intel_upload_data(brw
, index_buffer
->ptr
, ib_size
, ib_type_size
,
1187 &brw
->ib
.bo
, &offset
);
1188 brw
->ib
.size
= brw
->ib
.bo
->size
;
1190 offset
= (GLuint
) (unsigned long) index_buffer
->ptr
;
1192 /* If the index buffer isn't aligned to its element size, we have to
1193 * rebase it into a temporary.
1195 if ((ib_type_size
- 1) & offset
) {
1196 perf_debug("copying index buffer to a temporary to work around "
1197 "misaligned offset %d\n", offset
);
1199 GLubyte
*map
= ctx
->Driver
.MapBufferRange(ctx
,
1206 intel_upload_data(brw
, map
, ib_size
, ib_type_size
,
1207 &brw
->ib
.bo
, &offset
);
1208 brw
->ib
.size
= brw
->ib
.bo
->size
;
1210 ctx
->Driver
.UnmapBuffer(ctx
, bufferobj
, MAP_INTERNAL
);
1213 intel_bufferobj_buffer(brw
, intel_buffer_object(bufferobj
),
1215 if (bo
!= brw
->ib
.bo
) {
1216 drm_intel_bo_unreference(brw
->ib
.bo
);
1218 brw
->ib
.size
= bufferobj
->Size
;
1219 drm_intel_bo_reference(bo
);
1224 /* Use 3DPRIMITIVE's start_vertex_offset to avoid re-uploading
1225 * the index buffer state when we're just moving the start index
1228 brw
->ib
.start_vertex_offset
= offset
/ ib_type_size
;
1230 if (brw
->ib
.bo
!= old_bo
)
1231 brw
->ctx
.NewDriverState
|= BRW_NEW_INDEX_BUFFER
;
1233 if (index_buffer
->type
!= brw
->ib
.type
) {
1234 brw
->ib
.type
= index_buffer
->type
;
1235 brw
->ctx
.NewDriverState
|= BRW_NEW_INDEX_BUFFER
;
1239 const struct brw_tracked_state brw_indices
= {
1242 .brw
= BRW_NEW_BLORP
|
1245 .emit
= brw_upload_indices
,
1249 brw_emit_index_buffer(struct brw_context
*brw
)
1251 const struct _mesa_index_buffer
*index_buffer
= brw
->ib
.ib
;
1252 GLuint cut_index_setting
;
1254 if (index_buffer
== NULL
)
1257 if (brw
->prim_restart
.enable_cut_index
&& !brw
->is_haswell
) {
1258 cut_index_setting
= BRW_CUT_INDEX_ENABLE
;
1260 cut_index_setting
= 0;
1264 OUT_BATCH(CMD_INDEX_BUFFER
<< 16 |
1266 brw_get_index_type(index_buffer
->type
) |
1268 OUT_RELOC(brw
->ib
.bo
,
1269 I915_GEM_DOMAIN_VERTEX
, 0,
1271 OUT_RELOC(brw
->ib
.bo
,
1272 I915_GEM_DOMAIN_VERTEX
, 0,
1277 const struct brw_tracked_state brw_index_buffer
= {
1280 .brw
= BRW_NEW_BATCH
|
1282 BRW_NEW_INDEX_BUFFER
,
1284 .emit
= brw_emit_index_buffer
,