1 /**************************************************************************
3 * Copyright 2003 VMware, Inc.
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 VMWARE 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 **************************************************************************/
28 #include "main/glheader.h"
29 #include "main/bufferobj.h"
30 #include "main/context.h"
31 #include "main/enums.h"
32 #include "main/macros.h"
33 #include "main/glformats.h"
36 #include "brw_defines.h"
37 #include "brw_context.h"
38 #include "brw_state.h"
40 #include "intel_batchbuffer.h"
41 #include "intel_buffer_objects.h"
43 static GLuint double_types
[5] = {
45 BRW_SURFACEFORMAT_R64_FLOAT
,
46 BRW_SURFACEFORMAT_R64G64_FLOAT
,
47 BRW_SURFACEFORMAT_R64G64B64_FLOAT
,
48 BRW_SURFACEFORMAT_R64G64B64A64_FLOAT
51 static GLuint float_types
[5] = {
53 BRW_SURFACEFORMAT_R32_FLOAT
,
54 BRW_SURFACEFORMAT_R32G32_FLOAT
,
55 BRW_SURFACEFORMAT_R32G32B32_FLOAT
,
56 BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
59 static GLuint half_float_types
[5] = {
61 BRW_SURFACEFORMAT_R16_FLOAT
,
62 BRW_SURFACEFORMAT_R16G16_FLOAT
,
63 BRW_SURFACEFORMAT_R16G16B16A16_FLOAT
,
64 BRW_SURFACEFORMAT_R16G16B16A16_FLOAT
67 static GLuint fixed_point_types
[5] = {
69 BRW_SURFACEFORMAT_R32_SFIXED
,
70 BRW_SURFACEFORMAT_R32G32_SFIXED
,
71 BRW_SURFACEFORMAT_R32G32B32_SFIXED
,
72 BRW_SURFACEFORMAT_R32G32B32A32_SFIXED
,
75 static GLuint uint_types_direct
[5] = {
77 BRW_SURFACEFORMAT_R32_UINT
,
78 BRW_SURFACEFORMAT_R32G32_UINT
,
79 BRW_SURFACEFORMAT_R32G32B32_UINT
,
80 BRW_SURFACEFORMAT_R32G32B32A32_UINT
83 static GLuint uint_types_norm
[5] = {
85 BRW_SURFACEFORMAT_R32_UNORM
,
86 BRW_SURFACEFORMAT_R32G32_UNORM
,
87 BRW_SURFACEFORMAT_R32G32B32_UNORM
,
88 BRW_SURFACEFORMAT_R32G32B32A32_UNORM
91 static GLuint uint_types_scale
[5] = {
93 BRW_SURFACEFORMAT_R32_USCALED
,
94 BRW_SURFACEFORMAT_R32G32_USCALED
,
95 BRW_SURFACEFORMAT_R32G32B32_USCALED
,
96 BRW_SURFACEFORMAT_R32G32B32A32_USCALED
99 static GLuint int_types_direct
[5] = {
101 BRW_SURFACEFORMAT_R32_SINT
,
102 BRW_SURFACEFORMAT_R32G32_SINT
,
103 BRW_SURFACEFORMAT_R32G32B32_SINT
,
104 BRW_SURFACEFORMAT_R32G32B32A32_SINT
107 static GLuint int_types_norm
[5] = {
109 BRW_SURFACEFORMAT_R32_SNORM
,
110 BRW_SURFACEFORMAT_R32G32_SNORM
,
111 BRW_SURFACEFORMAT_R32G32B32_SNORM
,
112 BRW_SURFACEFORMAT_R32G32B32A32_SNORM
115 static GLuint int_types_scale
[5] = {
117 BRW_SURFACEFORMAT_R32_SSCALED
,
118 BRW_SURFACEFORMAT_R32G32_SSCALED
,
119 BRW_SURFACEFORMAT_R32G32B32_SSCALED
,
120 BRW_SURFACEFORMAT_R32G32B32A32_SSCALED
123 static GLuint ushort_types_direct
[5] = {
125 BRW_SURFACEFORMAT_R16_UINT
,
126 BRW_SURFACEFORMAT_R16G16_UINT
,
127 BRW_SURFACEFORMAT_R16G16B16A16_UINT
,
128 BRW_SURFACEFORMAT_R16G16B16A16_UINT
131 static GLuint ushort_types_norm
[5] = {
133 BRW_SURFACEFORMAT_R16_UNORM
,
134 BRW_SURFACEFORMAT_R16G16_UNORM
,
135 BRW_SURFACEFORMAT_R16G16B16_UNORM
,
136 BRW_SURFACEFORMAT_R16G16B16A16_UNORM
139 static GLuint ushort_types_scale
[5] = {
141 BRW_SURFACEFORMAT_R16_USCALED
,
142 BRW_SURFACEFORMAT_R16G16_USCALED
,
143 BRW_SURFACEFORMAT_R16G16B16_USCALED
,
144 BRW_SURFACEFORMAT_R16G16B16A16_USCALED
147 static GLuint short_types_direct
[5] = {
149 BRW_SURFACEFORMAT_R16_SINT
,
150 BRW_SURFACEFORMAT_R16G16_SINT
,
151 BRW_SURFACEFORMAT_R16G16B16A16_SINT
,
152 BRW_SURFACEFORMAT_R16G16B16A16_SINT
155 static GLuint short_types_norm
[5] = {
157 BRW_SURFACEFORMAT_R16_SNORM
,
158 BRW_SURFACEFORMAT_R16G16_SNORM
,
159 BRW_SURFACEFORMAT_R16G16B16_SNORM
,
160 BRW_SURFACEFORMAT_R16G16B16A16_SNORM
163 static GLuint short_types_scale
[5] = {
165 BRW_SURFACEFORMAT_R16_SSCALED
,
166 BRW_SURFACEFORMAT_R16G16_SSCALED
,
167 BRW_SURFACEFORMAT_R16G16B16_SSCALED
,
168 BRW_SURFACEFORMAT_R16G16B16A16_SSCALED
171 static GLuint ubyte_types_direct
[5] = {
173 BRW_SURFACEFORMAT_R8_UINT
,
174 BRW_SURFACEFORMAT_R8G8_UINT
,
175 BRW_SURFACEFORMAT_R8G8B8A8_UINT
,
176 BRW_SURFACEFORMAT_R8G8B8A8_UINT
179 static GLuint ubyte_types_norm
[5] = {
181 BRW_SURFACEFORMAT_R8_UNORM
,
182 BRW_SURFACEFORMAT_R8G8_UNORM
,
183 BRW_SURFACEFORMAT_R8G8B8_UNORM
,
184 BRW_SURFACEFORMAT_R8G8B8A8_UNORM
187 static GLuint ubyte_types_scale
[5] = {
189 BRW_SURFACEFORMAT_R8_USCALED
,
190 BRW_SURFACEFORMAT_R8G8_USCALED
,
191 BRW_SURFACEFORMAT_R8G8B8_USCALED
,
192 BRW_SURFACEFORMAT_R8G8B8A8_USCALED
195 static GLuint byte_types_direct
[5] = {
197 BRW_SURFACEFORMAT_R8_SINT
,
198 BRW_SURFACEFORMAT_R8G8_SINT
,
199 BRW_SURFACEFORMAT_R8G8B8A8_SINT
,
200 BRW_SURFACEFORMAT_R8G8B8A8_SINT
203 static GLuint byte_types_norm
[5] = {
205 BRW_SURFACEFORMAT_R8_SNORM
,
206 BRW_SURFACEFORMAT_R8G8_SNORM
,
207 BRW_SURFACEFORMAT_R8G8B8_SNORM
,
208 BRW_SURFACEFORMAT_R8G8B8A8_SNORM
211 static GLuint byte_types_scale
[5] = {
213 BRW_SURFACEFORMAT_R8_SSCALED
,
214 BRW_SURFACEFORMAT_R8G8_SSCALED
,
215 BRW_SURFACEFORMAT_R8G8B8_SSCALED
,
216 BRW_SURFACEFORMAT_R8G8B8A8_SSCALED
221 * Given vertex array type/size/format/normalized info, return
222 * the appopriate hardware surface type.
223 * Format will be GL_RGBA or possibly GL_BGRA for GLubyte[4] color arrays.
226 brw_get_vertex_surface_type(struct brw_context
*brw
,
227 const struct gl_client_array
*glarray
)
229 int size
= glarray
->Size
;
231 if (unlikely(INTEL_DEBUG
& DEBUG_VERTS
))
232 fprintf(stderr
, "type %s size %d normalized %d\n",
233 _mesa_lookup_enum_by_nr(glarray
->Type
),
234 glarray
->Size
, glarray
->Normalized
);
236 if (glarray
->Integer
) {
237 assert(glarray
->Format
== GL_RGBA
); /* sanity check */
238 switch (glarray
->Type
) {
239 case GL_INT
: return int_types_direct
[size
];
240 case GL_SHORT
: return short_types_direct
[size
];
241 case GL_BYTE
: return byte_types_direct
[size
];
242 case GL_UNSIGNED_INT
: return uint_types_direct
[size
];
243 case GL_UNSIGNED_SHORT
: return ushort_types_direct
[size
];
244 case GL_UNSIGNED_BYTE
: return ubyte_types_direct
[size
];
245 default: assert(0); return 0;
247 } else if (glarray
->Type
== GL_UNSIGNED_INT_10F_11F_11F_REV
) {
248 return BRW_SURFACEFORMAT_R11G11B10_FLOAT
;
249 } else if (glarray
->Normalized
) {
250 switch (glarray
->Type
) {
251 case GL_DOUBLE
: return double_types
[size
];
252 case GL_FLOAT
: return float_types
[size
];
253 case GL_HALF_FLOAT
: return half_float_types
[size
];
254 case GL_INT
: return int_types_norm
[size
];
255 case GL_SHORT
: return short_types_norm
[size
];
256 case GL_BYTE
: return byte_types_norm
[size
];
257 case GL_UNSIGNED_INT
: return uint_types_norm
[size
];
258 case GL_UNSIGNED_SHORT
: return ushort_types_norm
[size
];
259 case GL_UNSIGNED_BYTE
:
260 if (glarray
->Format
== GL_BGRA
) {
261 /* See GL_EXT_vertex_array_bgra */
263 return BRW_SURFACEFORMAT_B8G8R8A8_UNORM
;
266 return ubyte_types_norm
[size
];
269 if (brw
->gen
>= 8 || brw
->is_haswell
)
270 return fixed_point_types
[size
];
272 /* This produces GL_FIXED inputs as values between INT32_MIN and
273 * INT32_MAX, which will be scaled down by 1/65536 by the VS.
275 return int_types_scale
[size
];
276 /* See GL_ARB_vertex_type_2_10_10_10_rev.
277 * W/A: Pre-Haswell, the hardware doesn't really support the formats we'd
278 * like to use here, so upload everything as UINT and fix
281 case GL_INT_2_10_10_10_REV
:
283 if (brw
->gen
>= 8 || brw
->is_haswell
) {
284 return glarray
->Format
== GL_BGRA
285 ? BRW_SURFACEFORMAT_B10G10R10A2_SNORM
286 : BRW_SURFACEFORMAT_R10G10B10A2_SNORM
;
288 return BRW_SURFACEFORMAT_R10G10B10A2_UINT
;
289 case GL_UNSIGNED_INT_2_10_10_10_REV
:
291 if (brw
->gen
>= 8 || brw
->is_haswell
) {
292 return glarray
->Format
== GL_BGRA
293 ? BRW_SURFACEFORMAT_B10G10R10A2_UNORM
294 : BRW_SURFACEFORMAT_R10G10B10A2_UNORM
;
296 return BRW_SURFACEFORMAT_R10G10B10A2_UINT
;
297 default: assert(0); return 0;
301 /* See GL_ARB_vertex_type_2_10_10_10_rev.
302 * W/A: the hardware doesn't really support the formats we'd
303 * like to use here, so upload everything as UINT and fix
306 if (glarray
->Type
== GL_INT_2_10_10_10_REV
) {
308 if (brw
->gen
>= 8 || brw
->is_haswell
) {
309 return glarray
->Format
== GL_BGRA
310 ? BRW_SURFACEFORMAT_B10G10R10A2_SSCALED
311 : BRW_SURFACEFORMAT_R10G10B10A2_SSCALED
;
313 return BRW_SURFACEFORMAT_R10G10B10A2_UINT
;
314 } else if (glarray
->Type
== GL_UNSIGNED_INT_2_10_10_10_REV
) {
316 if (brw
->gen
>= 8 || brw
->is_haswell
) {
317 return glarray
->Format
== GL_BGRA
318 ? BRW_SURFACEFORMAT_B10G10R10A2_USCALED
319 : BRW_SURFACEFORMAT_R10G10B10A2_USCALED
;
321 return BRW_SURFACEFORMAT_R10G10B10A2_UINT
;
323 assert(glarray
->Format
== GL_RGBA
); /* sanity check */
324 switch (glarray
->Type
) {
325 case GL_DOUBLE
: return double_types
[size
];
326 case GL_FLOAT
: return float_types
[size
];
327 case GL_HALF_FLOAT
: return half_float_types
[size
];
328 case GL_INT
: return int_types_scale
[size
];
329 case GL_SHORT
: return short_types_scale
[size
];
330 case GL_BYTE
: return byte_types_scale
[size
];
331 case GL_UNSIGNED_INT
: return uint_types_scale
[size
];
332 case GL_UNSIGNED_SHORT
: return ushort_types_scale
[size
];
333 case GL_UNSIGNED_BYTE
: return ubyte_types_scale
[size
];
335 if (brw
->gen
>= 8 || brw
->is_haswell
)
336 return fixed_point_types
[size
];
338 /* This produces GL_FIXED inputs as values between INT32_MIN and
339 * INT32_MAX, which will be scaled down by 1/65536 by the VS.
341 return int_types_scale
[size
];
342 default: assert(0); return 0;
348 brw_get_index_type(GLenum type
)
351 case GL_UNSIGNED_BYTE
: return BRW_INDEX_BYTE
;
352 case GL_UNSIGNED_SHORT
: return BRW_INDEX_WORD
;
353 case GL_UNSIGNED_INT
: return BRW_INDEX_DWORD
;
354 default: assert(0); return 0;
359 copy_array_to_vbo_array(struct brw_context
*brw
,
360 struct brw_vertex_element
*element
,
362 struct brw_vertex_buffer
*buffer
,
365 const int src_stride
= element
->glarray
->StrideB
;
367 /* If the source stride is zero, we just want to upload the current
368 * attribute once and set the buffer's stride to 0. There's no need
369 * to replicate it out.
371 if (src_stride
== 0) {
372 intel_upload_data(brw
, element
->glarray
->Ptr
,
373 element
->glarray
->_ElementSize
,
374 element
->glarray
->_ElementSize
,
375 &buffer
->bo
, &buffer
->offset
);
381 const unsigned char *src
= element
->glarray
->Ptr
+ min
* src_stride
;
382 int count
= max
- min
+ 1;
383 GLuint size
= count
* dst_stride
;
385 if (dst_stride
== src_stride
) {
386 intel_upload_data(brw
, src
, size
, dst_stride
,
387 &buffer
->bo
, &buffer
->offset
);
389 char * const map
= intel_upload_map(brw
, size
, dst_stride
);
393 memcpy(dst
, src
, dst_stride
);
397 intel_upload_unmap(brw
, map
, size
, dst_stride
,
398 &buffer
->bo
, &buffer
->offset
);
400 buffer
->stride
= dst_stride
;
404 brw_prepare_vertices(struct brw_context
*brw
)
406 struct gl_context
*ctx
= &brw
->ctx
;
407 /* CACHE_NEW_VS_PROG */
408 GLbitfield64 vs_inputs
= brw
->vs
.prog_data
->inputs_read
;
409 const unsigned char *ptr
= NULL
;
410 GLuint interleaved
= 0;
411 unsigned int min_index
= brw
->vb
.min_index
+ brw
->basevertex
;
412 unsigned int max_index
= brw
->vb
.max_index
+ brw
->basevertex
;
415 struct brw_vertex_element
*upload
[VERT_ATTRIB_MAX
];
416 GLuint nr_uploads
= 0;
420 * On gen6+, edge flags don't end up in the VUE (either in or out of the
421 * VS). Instead, they're uploaded as the last vertex element, and the data
422 * is passed sideband through the fixed function units. So, we need to
423 * prepare the vertex buffer for it, but it's not present in inputs_read.
425 if (brw
->gen
>= 6 && (ctx
->Polygon
.FrontMode
!= GL_FILL
||
426 ctx
->Polygon
.BackMode
!= GL_FILL
)) {
427 vs_inputs
|= VERT_BIT_EDGEFLAG
;
431 fprintf(stderr
, "%s %d..%d\n", __FUNCTION__
, min_index
, max_index
);
433 /* Accumulate the list of enabled arrays. */
434 brw
->vb
.nr_enabled
= 0;
436 GLuint i
= ffsll(vs_inputs
) - 1;
437 struct brw_vertex_element
*input
= &brw
->vb
.inputs
[i
];
439 vs_inputs
&= ~BITFIELD64_BIT(i
);
440 brw
->vb
.enabled
[brw
->vb
.nr_enabled
++] = input
;
443 if (brw
->vb
.nr_enabled
== 0)
446 if (brw
->vb
.nr_buffers
)
449 for (i
= j
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
450 struct brw_vertex_element
*input
= brw
->vb
.enabled
[i
];
451 const struct gl_client_array
*glarray
= input
->glarray
;
453 if (_mesa_is_bufferobj(glarray
->BufferObj
)) {
454 struct intel_buffer_object
*intel_buffer
=
455 intel_buffer_object(glarray
->BufferObj
);
458 /* If we have a VB set to be uploaded for this buffer object
459 * already, reuse that VB state so that we emit fewer
462 for (k
= 0; k
< i
; k
++) {
463 const struct gl_client_array
*other
= brw
->vb
.enabled
[k
]->glarray
;
464 if (glarray
->BufferObj
== other
->BufferObj
&&
465 glarray
->StrideB
== other
->StrideB
&&
466 glarray
->InstanceDivisor
== other
->InstanceDivisor
&&
467 (uintptr_t)(glarray
->Ptr
- other
->Ptr
) < glarray
->StrideB
)
469 input
->buffer
= brw
->vb
.enabled
[k
]->buffer
;
470 input
->offset
= glarray
->Ptr
- other
->Ptr
;
475 struct brw_vertex_buffer
*buffer
= &brw
->vb
.buffers
[j
];
477 /* Named buffer object: Just reference its contents directly. */
478 buffer
->offset
= (uintptr_t)glarray
->Ptr
;
479 buffer
->stride
= glarray
->StrideB
;
480 buffer
->step_rate
= glarray
->InstanceDivisor
;
482 uint32_t offset
, size
;
483 if (glarray
->InstanceDivisor
) {
484 offset
= buffer
->offset
;
485 size
= (buffer
->stride
* ((brw
->num_instances
/
486 glarray
->InstanceDivisor
) - 1) +
487 glarray
->_ElementSize
);
489 if (min_index
== -1) {
491 size
= intel_buffer
->Base
.Size
;
493 offset
= buffer
->offset
+ min_index
* buffer
->stride
;
494 size
= (buffer
->stride
* (max_index
- min_index
) +
495 glarray
->_ElementSize
);
498 buffer
->bo
= intel_bufferobj_buffer(brw
, intel_buffer
,
500 drm_intel_bo_reference(buffer
->bo
);
506 /* This is a common place to reach if the user mistakenly supplies
507 * a pointer in place of a VBO offset. If we just let it go through,
508 * we may end up dereferencing a pointer beyond the bounds of the
509 * GTT. We would hope that the VBO's max_index would save us, but
510 * Mesa appears to hand us min/max values not clipped to the
511 * array object's _MaxElement, and _MaxElement frequently appears
512 * to be wrong anyway.
514 * The VBO spec allows application termination in this case, and it's
515 * probably a service to the poor programmer to do so rather than
516 * trying to just not render.
518 assert(input
->offset
< brw
->vb
.buffers
[input
->buffer
].bo
->size
);
520 /* Queue the buffer object up to be uploaded in the next pass,
521 * when we've decided if we're doing interleaved or not.
523 if (nr_uploads
== 0) {
524 interleaved
= glarray
->StrideB
;
527 else if (interleaved
!= glarray
->StrideB
||
528 glarray
->Ptr
< ptr
||
529 (uintptr_t)(glarray
->Ptr
- ptr
) + glarray
->_ElementSize
> interleaved
)
531 /* If our stride is different from the first attribute's stride,
532 * or if the first attribute's stride didn't cover our element,
533 * disable the interleaved upload optimization. The second case
534 * can most commonly occur in cases where there is a single vertex
535 * and, for example, the data is stored on the application's
538 * NOTE: This will also disable the optimization in cases where
539 * the data is in a different order than the array indices.
543 * glVertexAttribPointer(0, 4, GL_FLOAT, 32, &data[4]);
544 * glVertexAttribPointer(1, 4, GL_FLOAT, 32, &data[0]);
549 upload
[nr_uploads
++] = input
;
553 /* If we need to upload all the arrays, then we can trim those arrays to
554 * only the used elements [min_index, max_index] so long as we adjust all
555 * the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
557 brw
->vb
.start_vertex_bias
= 0;
559 if (nr_uploads
== brw
->vb
.nr_enabled
) {
560 brw
->vb
.start_vertex_bias
= -delta
;
564 /* Handle any arrays to be uploaded. */
565 if (nr_uploads
> 1) {
567 struct brw_vertex_buffer
*buffer
= &brw
->vb
.buffers
[j
];
568 /* All uploads are interleaved, so upload the arrays together as
569 * interleaved. First, upload the contents and set up upload[0].
571 copy_array_to_vbo_array(brw
, upload
[0], min_index
, max_index
,
572 buffer
, interleaved
);
573 buffer
->offset
-= delta
* interleaved
;
575 for (i
= 0; i
< nr_uploads
; i
++) {
576 /* Then, just point upload[i] at upload[0]'s buffer. */
578 ((const unsigned char *)upload
[i
]->glarray
->Ptr
- ptr
);
579 upload
[i
]->buffer
= j
;
586 /* Upload non-interleaved arrays */
587 for (i
= 0; i
< nr_uploads
; i
++) {
588 struct brw_vertex_buffer
*buffer
= &brw
->vb
.buffers
[j
];
589 if (upload
[i
]->glarray
->InstanceDivisor
== 0) {
590 copy_array_to_vbo_array(brw
, upload
[i
], min_index
, max_index
,
591 buffer
, upload
[i
]->glarray
->_ElementSize
);
593 /* This is an instanced attribute, since its InstanceDivisor
594 * is not zero. Therefore, its data will be stepped after the
595 * instanced draw has been run InstanceDivisor times.
597 uint32_t instanced_attr_max_index
=
598 (brw
->num_instances
- 1) / upload
[i
]->glarray
->InstanceDivisor
;
599 copy_array_to_vbo_array(brw
, upload
[i
], 0, instanced_attr_max_index
,
600 buffer
, upload
[i
]->glarray
->_ElementSize
);
602 buffer
->offset
-= delta
* buffer
->stride
;
603 buffer
->step_rate
= upload
[i
]->glarray
->InstanceDivisor
;
604 upload
[i
]->buffer
= j
++;
605 upload
[i
]->offset
= 0;
608 brw
->vb
.nr_buffers
= j
;
611 static void brw_emit_vertices(struct brw_context
*brw
)
613 struct gl_context
*ctx
= &brw
->ctx
;
614 GLuint i
, nr_elements
;
616 brw_prepare_vertices(brw
);
618 brw_emit_query_begin(brw
);
620 nr_elements
= brw
->vb
.nr_enabled
+ brw
->vs
.prog_data
->uses_vertexid
;
622 /* If the VS doesn't read any inputs (calculating vertex position from
623 * a state variable for some reason, for example), emit a single pad
624 * VERTEX_ELEMENT struct and bail.
626 * The stale VB state stays in place, but they don't do anything unless
627 * a VE loads from them.
629 if (nr_elements
== 0) {
631 OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS
<< 16) | 1);
633 OUT_BATCH((0 << GEN6_VE0_INDEX_SHIFT
) |
635 (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
<< BRW_VE0_FORMAT_SHIFT
) |
636 (0 << BRW_VE0_SRC_OFFSET_SHIFT
));
638 OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT
) |
640 (BRW_SURFACEFORMAT_R32G32B32A32_FLOAT
<< BRW_VE0_FORMAT_SHIFT
) |
641 (0 << BRW_VE0_SRC_OFFSET_SHIFT
));
643 OUT_BATCH((BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_0_SHIFT
) |
644 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_1_SHIFT
) |
645 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_2_SHIFT
) |
646 (BRW_VE1_COMPONENT_STORE_1_FLT
<< BRW_VE1_COMPONENT_3_SHIFT
));
651 /* Now emit VB and VEP state packets.
654 if (brw
->vb
.nr_buffers
) {
656 assert(brw
->vb
.nr_buffers
<= 33);
658 assert(brw
->vb
.nr_buffers
<= 17);
661 BEGIN_BATCH(1 + 4*brw
->vb
.nr_buffers
);
662 OUT_BATCH((_3DSTATE_VERTEX_BUFFERS
<< 16) | (4*brw
->vb
.nr_buffers
- 1));
663 for (i
= 0; i
< brw
->vb
.nr_buffers
; i
++) {
664 struct brw_vertex_buffer
*buffer
= &brw
->vb
.buffers
[i
];
668 dw0
= buffer
->step_rate
669 ? GEN6_VB0_ACCESS_INSTANCEDATA
670 : GEN6_VB0_ACCESS_VERTEXDATA
;
671 dw0
|= i
<< GEN6_VB0_INDEX_SHIFT
;
673 dw0
= buffer
->step_rate
674 ? BRW_VB0_ACCESS_INSTANCEDATA
675 : BRW_VB0_ACCESS_VERTEXDATA
;
676 dw0
|= i
<< BRW_VB0_INDEX_SHIFT
;
680 dw0
|= GEN7_VB0_ADDRESS_MODIFYENABLE
;
683 dw0
|= GEN7_MOCS_L3
<< 16;
685 WARN_ONCE(buffer
->stride
>= (brw
->gen
>= 5 ? 2048 : 2047),
686 "VBO stride %d too large, bad rendering may occur\n",
688 OUT_BATCH(dw0
| (buffer
->stride
<< BRW_VB0_PITCH_SHIFT
));
689 OUT_RELOC(buffer
->bo
, I915_GEM_DOMAIN_VERTEX
, 0, buffer
->offset
);
691 OUT_RELOC(buffer
->bo
, I915_GEM_DOMAIN_VERTEX
, 0, buffer
->bo
->size
- 1);
694 OUT_BATCH(buffer
->step_rate
);
699 /* The hardware allows one more VERTEX_ELEMENTS than VERTEX_BUFFERS, presumably
700 * for VertexID/InstanceID.
703 assert(nr_elements
<= 34);
705 assert(nr_elements
<= 18);
708 struct brw_vertex_element
*gen6_edgeflag_input
= NULL
;
710 BEGIN_BATCH(1 + nr_elements
* 2);
711 OUT_BATCH((_3DSTATE_VERTEX_ELEMENTS
<< 16) | (2 * nr_elements
- 1));
712 for (i
= 0; i
< brw
->vb
.nr_enabled
; i
++) {
713 struct brw_vertex_element
*input
= brw
->vb
.enabled
[i
];
714 uint32_t format
= brw_get_vertex_surface_type(brw
, input
->glarray
);
715 uint32_t comp0
= BRW_VE1_COMPONENT_STORE_SRC
;
716 uint32_t comp1
= BRW_VE1_COMPONENT_STORE_SRC
;
717 uint32_t comp2
= BRW_VE1_COMPONENT_STORE_SRC
;
718 uint32_t comp3
= BRW_VE1_COMPONENT_STORE_SRC
;
720 /* The gen4 driver expects edgeflag to come in as a float, and passes
721 * that float on to the tests in the clipper. Mesa's current vertex
722 * attribute value for EdgeFlag is stored as a float, which works out.
723 * glEdgeFlagPointer, on the other hand, gives us an unnormalized
724 * integer ubyte. Just rewrite that to convert to a float.
726 if (input
->attrib
== VERT_ATTRIB_EDGEFLAG
) {
727 /* Gen6+ passes edgeflag as sideband along with the vertex, instead
728 * of in the VUE. We have to upload it sideband as the last vertex
729 * element according to the B-Spec.
732 gen6_edgeflag_input
= input
;
736 if (format
== BRW_SURFACEFORMAT_R8_UINT
)
737 format
= BRW_SURFACEFORMAT_R8_SSCALED
;
740 switch (input
->glarray
->Size
) {
741 case 0: comp0
= BRW_VE1_COMPONENT_STORE_0
;
742 case 1: comp1
= BRW_VE1_COMPONENT_STORE_0
;
743 case 2: comp2
= BRW_VE1_COMPONENT_STORE_0
;
744 case 3: comp3
= input
->glarray
->Integer
? BRW_VE1_COMPONENT_STORE_1_INT
745 : BRW_VE1_COMPONENT_STORE_1_FLT
;
750 OUT_BATCH((input
->buffer
<< GEN6_VE0_INDEX_SHIFT
) |
752 (format
<< BRW_VE0_FORMAT_SHIFT
) |
753 (input
->offset
<< BRW_VE0_SRC_OFFSET_SHIFT
));
755 OUT_BATCH((input
->buffer
<< BRW_VE0_INDEX_SHIFT
) |
757 (format
<< BRW_VE0_FORMAT_SHIFT
) |
758 (input
->offset
<< BRW_VE0_SRC_OFFSET_SHIFT
));
762 OUT_BATCH((comp0
<< BRW_VE1_COMPONENT_0_SHIFT
) |
763 (comp1
<< BRW_VE1_COMPONENT_1_SHIFT
) |
764 (comp2
<< BRW_VE1_COMPONENT_2_SHIFT
) |
765 (comp3
<< BRW_VE1_COMPONENT_3_SHIFT
));
767 OUT_BATCH((comp0
<< BRW_VE1_COMPONENT_0_SHIFT
) |
768 (comp1
<< BRW_VE1_COMPONENT_1_SHIFT
) |
769 (comp2
<< BRW_VE1_COMPONENT_2_SHIFT
) |
770 (comp3
<< BRW_VE1_COMPONENT_3_SHIFT
) |
771 ((i
* 4) << BRW_VE1_DST_OFFSET_SHIFT
));
774 if (brw
->gen
>= 6 && gen6_edgeflag_input
) {
776 brw_get_vertex_surface_type(brw
, gen6_edgeflag_input
->glarray
);
778 OUT_BATCH((gen6_edgeflag_input
->buffer
<< GEN6_VE0_INDEX_SHIFT
) |
780 GEN6_VE0_EDGE_FLAG_ENABLE
|
781 (format
<< BRW_VE0_FORMAT_SHIFT
) |
782 (gen6_edgeflag_input
->offset
<< BRW_VE0_SRC_OFFSET_SHIFT
));
783 OUT_BATCH((BRW_VE1_COMPONENT_STORE_SRC
<< BRW_VE1_COMPONENT_0_SHIFT
) |
784 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_1_SHIFT
) |
785 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_2_SHIFT
) |
786 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_3_SHIFT
));
789 if (brw
->vs
.prog_data
->uses_vertexid
) {
790 uint32_t dw0
= 0, dw1
= 0;
792 dw1
= ((BRW_VE1_COMPONENT_STORE_VID
<< BRW_VE1_COMPONENT_0_SHIFT
) |
793 (BRW_VE1_COMPONENT_STORE_IID
<< BRW_VE1_COMPONENT_1_SHIFT
) |
794 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_2_SHIFT
) |
795 (BRW_VE1_COMPONENT_STORE_0
<< BRW_VE1_COMPONENT_3_SHIFT
));
798 dw0
|= GEN6_VE0_VALID
;
800 dw0
|= BRW_VE0_VALID
;
801 dw1
|= (i
* 4) << BRW_VE1_DST_OFFSET_SHIFT
;
804 /* Note that for gl_VertexID, gl_InstanceID, and gl_PrimitiveID values,
805 * the format is ignored and the value is always int.
815 const struct brw_tracked_state brw_vertices
= {
817 .mesa
= _NEW_POLYGON
,
818 .brw
= BRW_NEW_BATCH
| BRW_NEW_VERTICES
,
819 .cache
= CACHE_NEW_VS_PROG
,
821 .emit
= brw_emit_vertices
,
824 static void brw_upload_indices(struct brw_context
*brw
)
826 struct gl_context
*ctx
= &brw
->ctx
;
827 const struct _mesa_index_buffer
*index_buffer
= brw
->ib
.ib
;
829 drm_intel_bo
*bo
= NULL
;
830 struct gl_buffer_object
*bufferobj
;
834 if (index_buffer
== NULL
)
837 ib_type_size
= _mesa_sizeof_type(index_buffer
->type
);
838 ib_size
= ib_type_size
* index_buffer
->count
;
839 bufferobj
= index_buffer
->obj
;
841 /* Turn into a proper VBO:
843 if (!_mesa_is_bufferobj(bufferobj
)) {
845 /* Get new bufferobj, offset:
847 intel_upload_data(brw
, index_buffer
->ptr
, ib_size
, ib_type_size
,
849 brw
->ib
.start_vertex_offset
= offset
/ ib_type_size
;
851 offset
= (GLuint
) (unsigned long) index_buffer
->ptr
;
853 /* If the index buffer isn't aligned to its element size, we have to
854 * rebase it into a temporary.
856 if ((ib_type_size
- 1) & offset
) {
857 perf_debug("copying index buffer to a temporary to work around "
858 "misaligned offset %d\n", offset
);
860 GLubyte
*map
= ctx
->Driver
.MapBufferRange(ctx
,
867 intel_upload_data(brw
, map
, ib_size
, ib_type_size
, &bo
, &offset
);
868 brw
->ib
.start_vertex_offset
= offset
/ ib_type_size
;
870 ctx
->Driver
.UnmapBuffer(ctx
, bufferobj
, MAP_INTERNAL
);
872 /* Use CMD_3D_PRIM's start_vertex_offset to avoid re-uploading
873 * the index buffer state when we're just moving the start index
876 brw
->ib
.start_vertex_offset
= offset
/ ib_type_size
;
878 bo
= intel_bufferobj_buffer(brw
, intel_buffer_object(bufferobj
),
880 drm_intel_bo_reference(bo
);
884 if (brw
->ib
.bo
!= bo
) {
885 drm_intel_bo_unreference(brw
->ib
.bo
);
888 brw
->state
.dirty
.brw
|= BRW_NEW_INDEX_BUFFER
;
890 drm_intel_bo_unreference(bo
);
893 if (index_buffer
->type
!= brw
->ib
.type
) {
894 brw
->ib
.type
= index_buffer
->type
;
895 brw
->state
.dirty
.brw
|= BRW_NEW_INDEX_BUFFER
;
899 const struct brw_tracked_state brw_indices
= {
902 .brw
= BRW_NEW_INDICES
,
905 .emit
= brw_upload_indices
,
908 static void brw_emit_index_buffer(struct brw_context
*brw
)
910 const struct _mesa_index_buffer
*index_buffer
= brw
->ib
.ib
;
911 GLuint cut_index_setting
;
913 if (index_buffer
== NULL
)
916 if (brw
->prim_restart
.enable_cut_index
&& !brw
->is_haswell
) {
917 cut_index_setting
= BRW_CUT_INDEX_ENABLE
;
919 cut_index_setting
= 0;
923 OUT_BATCH(CMD_INDEX_BUFFER
<< 16 |
925 brw_get_index_type(index_buffer
->type
) << 8 |
927 OUT_RELOC(brw
->ib
.bo
,
928 I915_GEM_DOMAIN_VERTEX
, 0,
930 OUT_RELOC(brw
->ib
.bo
,
931 I915_GEM_DOMAIN_VERTEX
, 0,
932 brw
->ib
.bo
->size
- 1);
936 const struct brw_tracked_state brw_index_buffer
= {
939 .brw
= BRW_NEW_BATCH
| BRW_NEW_INDEX_BUFFER
,
942 .emit
= brw_emit_index_buffer
,