1 /**************************************************************************
3 * Copyright 2007 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 * This file implements the st_draw_vbo() function which is called from
30 * Mesa's VBO module. All point/line/triangle rendering is done through
31 * this function whether the user called glBegin/End, glDrawArrays,
32 * glDrawElements, glEvalMesh, or glCalList, etc.
34 * We basically convert the VBO's vertex attribute/array information into
35 * Gallium vertex state, bind the vertex buffer objects and call
39 * Keith Whitwell <keith@tungstengraphics.com>
43 #include "main/imports.h"
44 #include "main/image.h"
45 #include "main/bufferobj.h"
46 #include "main/macros.h"
47 #include "main/mfeatures.h"
51 #include "st_context.h"
53 #include "st_cb_bufferobjects.h"
54 #include "st_cb_xformfb.h"
56 #include "st_program.h"
58 #include "pipe/p_context.h"
59 #include "pipe/p_defines.h"
60 #include "util/u_inlines.h"
61 #include "util/u_format.h"
62 #include "util/u_prim.h"
63 #include "util/u_draw_quad.h"
64 #include "util/u_upload_mgr.h"
65 #include "draw/draw_context.h"
66 #include "cso_cache/cso_context.h"
68 #include "../glsl/ir_uniform.h"
71 static GLuint double_types
[4] = {
72 PIPE_FORMAT_R64_FLOAT
,
73 PIPE_FORMAT_R64G64_FLOAT
,
74 PIPE_FORMAT_R64G64B64_FLOAT
,
75 PIPE_FORMAT_R64G64B64A64_FLOAT
78 static GLuint float_types
[4] = {
79 PIPE_FORMAT_R32_FLOAT
,
80 PIPE_FORMAT_R32G32_FLOAT
,
81 PIPE_FORMAT_R32G32B32_FLOAT
,
82 PIPE_FORMAT_R32G32B32A32_FLOAT
85 static GLuint half_float_types
[4] = {
86 PIPE_FORMAT_R16_FLOAT
,
87 PIPE_FORMAT_R16G16_FLOAT
,
88 PIPE_FORMAT_R16G16B16_FLOAT
,
89 PIPE_FORMAT_R16G16B16A16_FLOAT
92 static GLuint uint_types_norm
[4] = {
93 PIPE_FORMAT_R32_UNORM
,
94 PIPE_FORMAT_R32G32_UNORM
,
95 PIPE_FORMAT_R32G32B32_UNORM
,
96 PIPE_FORMAT_R32G32B32A32_UNORM
99 static GLuint uint_types_scale
[4] = {
100 PIPE_FORMAT_R32_USCALED
,
101 PIPE_FORMAT_R32G32_USCALED
,
102 PIPE_FORMAT_R32G32B32_USCALED
,
103 PIPE_FORMAT_R32G32B32A32_USCALED
106 static GLuint uint_types_int
[4] = {
107 PIPE_FORMAT_R32_UINT
,
108 PIPE_FORMAT_R32G32_UINT
,
109 PIPE_FORMAT_R32G32B32_UINT
,
110 PIPE_FORMAT_R32G32B32A32_UINT
113 static GLuint int_types_norm
[4] = {
114 PIPE_FORMAT_R32_SNORM
,
115 PIPE_FORMAT_R32G32_SNORM
,
116 PIPE_FORMAT_R32G32B32_SNORM
,
117 PIPE_FORMAT_R32G32B32A32_SNORM
120 static GLuint int_types_scale
[4] = {
121 PIPE_FORMAT_R32_SSCALED
,
122 PIPE_FORMAT_R32G32_SSCALED
,
123 PIPE_FORMAT_R32G32B32_SSCALED
,
124 PIPE_FORMAT_R32G32B32A32_SSCALED
127 static GLuint int_types_int
[4] = {
128 PIPE_FORMAT_R32_SINT
,
129 PIPE_FORMAT_R32G32_SINT
,
130 PIPE_FORMAT_R32G32B32_SINT
,
131 PIPE_FORMAT_R32G32B32A32_SINT
134 static GLuint ushort_types_norm
[4] = {
135 PIPE_FORMAT_R16_UNORM
,
136 PIPE_FORMAT_R16G16_UNORM
,
137 PIPE_FORMAT_R16G16B16_UNORM
,
138 PIPE_FORMAT_R16G16B16A16_UNORM
141 static GLuint ushort_types_scale
[4] = {
142 PIPE_FORMAT_R16_USCALED
,
143 PIPE_FORMAT_R16G16_USCALED
,
144 PIPE_FORMAT_R16G16B16_USCALED
,
145 PIPE_FORMAT_R16G16B16A16_USCALED
148 static GLuint ushort_types_int
[4] = {
149 PIPE_FORMAT_R16_UINT
,
150 PIPE_FORMAT_R16G16_UINT
,
151 PIPE_FORMAT_R16G16B16_UINT
,
152 PIPE_FORMAT_R16G16B16A16_UINT
155 static GLuint short_types_norm
[4] = {
156 PIPE_FORMAT_R16_SNORM
,
157 PIPE_FORMAT_R16G16_SNORM
,
158 PIPE_FORMAT_R16G16B16_SNORM
,
159 PIPE_FORMAT_R16G16B16A16_SNORM
162 static GLuint short_types_scale
[4] = {
163 PIPE_FORMAT_R16_SSCALED
,
164 PIPE_FORMAT_R16G16_SSCALED
,
165 PIPE_FORMAT_R16G16B16_SSCALED
,
166 PIPE_FORMAT_R16G16B16A16_SSCALED
169 static GLuint short_types_int
[4] = {
170 PIPE_FORMAT_R16_SINT
,
171 PIPE_FORMAT_R16G16_SINT
,
172 PIPE_FORMAT_R16G16B16_SINT
,
173 PIPE_FORMAT_R16G16B16A16_SINT
176 static GLuint ubyte_types_norm
[4] = {
177 PIPE_FORMAT_R8_UNORM
,
178 PIPE_FORMAT_R8G8_UNORM
,
179 PIPE_FORMAT_R8G8B8_UNORM
,
180 PIPE_FORMAT_R8G8B8A8_UNORM
183 static GLuint ubyte_types_scale
[4] = {
184 PIPE_FORMAT_R8_USCALED
,
185 PIPE_FORMAT_R8G8_USCALED
,
186 PIPE_FORMAT_R8G8B8_USCALED
,
187 PIPE_FORMAT_R8G8B8A8_USCALED
190 static GLuint ubyte_types_int
[4] = {
192 PIPE_FORMAT_R8G8_UINT
,
193 PIPE_FORMAT_R8G8B8_UINT
,
194 PIPE_FORMAT_R8G8B8A8_UINT
197 static GLuint byte_types_norm
[4] = {
198 PIPE_FORMAT_R8_SNORM
,
199 PIPE_FORMAT_R8G8_SNORM
,
200 PIPE_FORMAT_R8G8B8_SNORM
,
201 PIPE_FORMAT_R8G8B8A8_SNORM
204 static GLuint byte_types_scale
[4] = {
205 PIPE_FORMAT_R8_SSCALED
,
206 PIPE_FORMAT_R8G8_SSCALED
,
207 PIPE_FORMAT_R8G8B8_SSCALED
,
208 PIPE_FORMAT_R8G8B8A8_SSCALED
211 static GLuint byte_types_int
[4] = {
213 PIPE_FORMAT_R8G8_SINT
,
214 PIPE_FORMAT_R8G8B8_SINT
,
215 PIPE_FORMAT_R8G8B8A8_SINT
218 static GLuint fixed_types
[4] = {
219 PIPE_FORMAT_R32_FIXED
,
220 PIPE_FORMAT_R32G32_FIXED
,
221 PIPE_FORMAT_R32G32B32_FIXED
,
222 PIPE_FORMAT_R32G32B32A32_FIXED
228 * Return a PIPE_FORMAT_x for the given GL datatype and size.
231 st_pipe_vertex_format(GLenum type
, GLuint size
, GLenum format
,
232 GLboolean normalized
, GLboolean integer
)
234 assert((type
>= GL_BYTE
&& type
<= GL_DOUBLE
) ||
235 type
== GL_FIXED
|| type
== GL_HALF_FLOAT
||
236 type
== GL_INT_2_10_10_10_REV
||
237 type
== GL_UNSIGNED_INT_2_10_10_10_REV
);
240 assert(format
== GL_RGBA
|| format
== GL_BGRA
);
242 if (type
== GL_INT_2_10_10_10_REV
||
243 type
== GL_UNSIGNED_INT_2_10_10_10_REV
) {
247 if (format
== GL_BGRA
) {
248 if (type
== GL_INT_2_10_10_10_REV
) {
250 return PIPE_FORMAT_B10G10R10A2_SNORM
;
252 return PIPE_FORMAT_B10G10R10A2_SSCALED
;
255 return PIPE_FORMAT_B10G10R10A2_UNORM
;
257 return PIPE_FORMAT_B10G10R10A2_USCALED
;
260 if (type
== GL_INT_2_10_10_10_REV
) {
262 return PIPE_FORMAT_R10G10B10A2_SNORM
;
264 return PIPE_FORMAT_R10G10B10A2_SSCALED
;
267 return PIPE_FORMAT_R10G10B10A2_UNORM
;
269 return PIPE_FORMAT_R10G10B10A2_USCALED
;
274 if (format
== GL_BGRA
) {
275 /* this is an odd-ball case */
276 assert(type
== GL_UNSIGNED_BYTE
);
278 return PIPE_FORMAT_B8G8R8A8_UNORM
;
283 case GL_INT
: return int_types_int
[size
-1];
284 case GL_SHORT
: return short_types_int
[size
-1];
285 case GL_BYTE
: return byte_types_int
[size
-1];
286 case GL_UNSIGNED_INT
: return uint_types_int
[size
-1];
287 case GL_UNSIGNED_SHORT
: return ushort_types_int
[size
-1];
288 case GL_UNSIGNED_BYTE
: return ubyte_types_int
[size
-1];
289 default: assert(0); return 0;
292 else if (normalized
) {
294 case GL_DOUBLE
: return double_types
[size
-1];
295 case GL_FLOAT
: return float_types
[size
-1];
296 case GL_HALF_FLOAT
: return half_float_types
[size
-1];
297 case GL_INT
: return int_types_norm
[size
-1];
298 case GL_SHORT
: return short_types_norm
[size
-1];
299 case GL_BYTE
: return byte_types_norm
[size
-1];
300 case GL_UNSIGNED_INT
: return uint_types_norm
[size
-1];
301 case GL_UNSIGNED_SHORT
: return ushort_types_norm
[size
-1];
302 case GL_UNSIGNED_BYTE
: return ubyte_types_norm
[size
-1];
303 case GL_FIXED
: return fixed_types
[size
-1];
304 default: assert(0); return 0;
309 case GL_DOUBLE
: return double_types
[size
-1];
310 case GL_FLOAT
: return float_types
[size
-1];
311 case GL_HALF_FLOAT
: return half_float_types
[size
-1];
312 case GL_INT
: return int_types_scale
[size
-1];
313 case GL_SHORT
: return short_types_scale
[size
-1];
314 case GL_BYTE
: return byte_types_scale
[size
-1];
315 case GL_UNSIGNED_INT
: return uint_types_scale
[size
-1];
316 case GL_UNSIGNED_SHORT
: return ushort_types_scale
[size
-1];
317 case GL_UNSIGNED_BYTE
: return ubyte_types_scale
[size
-1];
318 case GL_FIXED
: return fixed_types
[size
-1];
319 default: assert(0); return 0;
322 return PIPE_FORMAT_NONE
; /* silence compiler warning */
327 * This is very similar to vbo_all_varyings_in_vbos() but we are
328 * only interested in per-vertex data. See bug 38626.
331 all_varyings_in_vbos(const struct gl_client_array
*arrays
[])
335 for (i
= 0; i
< VERT_ATTRIB_MAX
; i
++)
336 if (arrays
[i
]->StrideB
&&
337 !arrays
[i
]->InstanceDivisor
&&
338 !_mesa_is_bufferobj(arrays
[i
]->BufferObj
))
346 * Examine the active arrays to determine if we have interleaved
347 * vertex arrays all living in one VBO, or all living in user space.
350 is_interleaved_arrays(const struct st_vertex_program
*vp
,
351 const struct st_vp_variant
*vpv
,
352 const struct gl_client_array
**arrays
)
355 const struct gl_buffer_object
*firstBufObj
= NULL
;
356 GLint firstStride
= -1;
357 const GLubyte
*firstPtr
= NULL
;
358 GLboolean userSpaceBuffer
= GL_FALSE
;
360 for (attr
= 0; attr
< vpv
->num_inputs
; attr
++) {
361 const GLuint mesaAttr
= vp
->index_to_input
[attr
];
362 const struct gl_client_array
*array
= arrays
[mesaAttr
];
363 const struct gl_buffer_object
*bufObj
= array
->BufferObj
;
364 const GLsizei stride
= array
->StrideB
; /* in bytes */
367 /* save info about the first array */
368 firstStride
= stride
;
369 firstPtr
= array
->Ptr
;
370 firstBufObj
= bufObj
;
371 userSpaceBuffer
= !bufObj
|| !bufObj
->Name
;
374 /* check if other arrays interleave with the first, in same buffer */
375 if (stride
!= firstStride
)
376 return GL_FALSE
; /* strides don't match */
378 if (bufObj
!= firstBufObj
)
379 return GL_FALSE
; /* arrays in different VBOs */
381 if (abs(array
->Ptr
- firstPtr
) > firstStride
)
382 return GL_FALSE
; /* arrays start too far apart */
384 if ((!_mesa_is_bufferobj(bufObj
)) != userSpaceBuffer
)
385 return GL_FALSE
; /* mix of VBO and user-space arrays */
394 * Set up for drawing interleaved arrays that all live in one VBO
395 * or all live in user space.
396 * \param vbuffer returns vertex buffer info
397 * \param velements returns vertex element info
398 * \return GL_TRUE for success, GL_FALSE otherwise (probably out of memory)
401 setup_interleaved_attribs(struct gl_context
*ctx
,
402 const struct st_vertex_program
*vp
,
403 const struct st_vp_variant
*vpv
,
404 const struct gl_client_array
**arrays
,
405 struct pipe_vertex_buffer
*vbuffer
,
406 struct pipe_vertex_element velements
[],
408 unsigned num_instances
)
410 struct st_context
*st
= st_context(ctx
);
411 struct pipe_context
*pipe
= st
->pipe
;
413 const GLubyte
*low_addr
= NULL
;
414 GLboolean usingVBO
; /* all arrays in a VBO? */
415 struct gl_buffer_object
*bufobj
;
416 GLuint user_buffer_size
= 0;
417 GLuint vertex_size
= 0; /* bytes per vertex, in bytes */
420 /* Find the lowest address of the arrays we're drawing,
421 * Init bufobj and stride.
423 if (vpv
->num_inputs
) {
424 const GLuint mesaAttr0
= vp
->index_to_input
[0];
425 const struct gl_client_array
*array
= arrays
[mesaAttr0
];
427 /* Since we're doing interleaved arrays, we know there'll be at most
428 * one buffer object and the stride will be the same for all arrays.
431 bufobj
= array
->BufferObj
;
432 stride
= array
->StrideB
;
434 low_addr
= arrays
[vp
->index_to_input
[0]]->Ptr
;
436 for (attr
= 1; attr
< vpv
->num_inputs
; attr
++) {
437 const GLubyte
*start
= arrays
[vp
->index_to_input
[attr
]]->Ptr
;
438 low_addr
= MIN2(low_addr
, start
);
442 /* not sure we'll ever have zero inputs, but play it safe */
448 /* are the arrays in user space? */
449 usingVBO
= _mesa_is_bufferobj(bufobj
);
451 for (attr
= 0; attr
< vpv
->num_inputs
; attr
++) {
452 const GLuint mesaAttr
= vp
->index_to_input
[attr
];
453 const struct gl_client_array
*array
= arrays
[mesaAttr
];
454 unsigned src_offset
= (unsigned) (array
->Ptr
- low_addr
);
455 GLuint element_size
= array
->_ElementSize
;
457 assert(element_size
== array
->Size
* _mesa_sizeof_type(array
->Type
));
459 velements
[attr
].src_offset
= src_offset
;
460 velements
[attr
].instance_divisor
= array
->InstanceDivisor
;
461 velements
[attr
].vertex_buffer_index
= 0;
462 velements
[attr
].src_format
= st_pipe_vertex_format(array
->Type
,
467 assert(velements
[attr
].src_format
);
470 /* how many bytes referenced by this attribute array? */
471 uint divisor
= array
->InstanceDivisor
;
472 uint last_index
= divisor
? num_instances
/ divisor
: max_index
;
473 uint bytes
= src_offset
+ stride
* last_index
+ element_size
;
475 user_buffer_size
= MAX2(user_buffer_size
, bytes
);
477 /* update vertex size */
478 vertex_size
= MAX2(vertex_size
, src_offset
+ element_size
);
483 * Return the vbuffer info and setup user-space attrib info, if needed.
485 if (vpv
->num_inputs
== 0) {
486 /* just defensive coding here */
487 vbuffer
->buffer
= NULL
;
488 vbuffer
->buffer_offset
= 0;
490 st
->num_user_attribs
= 0;
493 /* all interleaved arrays in a VBO */
494 struct st_buffer_object
*stobj
= st_buffer_object(bufobj
);
496 if (!stobj
|| !stobj
->buffer
) {
497 /* probably out of memory (or zero-sized buffer) */
501 vbuffer
->buffer
= NULL
;
502 pipe_resource_reference(&vbuffer
->buffer
, stobj
->buffer
);
503 vbuffer
->buffer_offset
= pointer_to_offset(low_addr
);
504 vbuffer
->stride
= stride
;
505 st
->num_user_attribs
= 0;
508 /* all interleaved arrays in user memory */
509 vbuffer
->buffer
= pipe_user_buffer_create(pipe
->screen
,
512 PIPE_BIND_VERTEX_BUFFER
);
513 vbuffer
->buffer_offset
= 0;
514 vbuffer
->stride
= stride
;
516 /* Track user vertex buffers. */
517 pipe_resource_reference(&st
->user_attrib
[0].buffer
, vbuffer
->buffer
);
518 st
->user_attrib
[0].element_size
= vertex_size
;
519 st
->user_attrib
[0].stride
= stride
;
520 st
->num_user_attribs
= 1;
528 * Set up a separate pipe_vertex_buffer and pipe_vertex_element for each
530 * \param vbuffer returns vertex buffer info
531 * \param velements returns vertex element info
532 * \return GL_TRUE for success, GL_FALSE otherwise (probably out of memory)
535 setup_non_interleaved_attribs(struct gl_context
*ctx
,
536 const struct st_vertex_program
*vp
,
537 const struct st_vp_variant
*vpv
,
538 const struct gl_client_array
**arrays
,
539 struct pipe_vertex_buffer vbuffer
[],
540 struct pipe_vertex_element velements
[],
542 unsigned num_instances
)
544 struct st_context
*st
= st_context(ctx
);
545 struct pipe_context
*pipe
= st
->pipe
;
548 for (attr
= 0; attr
< vpv
->num_inputs
; attr
++) {
549 const GLuint mesaAttr
= vp
->index_to_input
[attr
];
550 const struct gl_client_array
*array
= arrays
[mesaAttr
];
551 struct gl_buffer_object
*bufobj
= array
->BufferObj
;
552 GLuint element_size
= array
->_ElementSize
;
553 GLsizei stride
= array
->StrideB
;
555 assert(element_size
== array
->Size
* _mesa_sizeof_type(array
->Type
));
557 if (_mesa_is_bufferobj(bufobj
)) {
558 /* Attribute data is in a VBO.
559 * Recall that for VBOs, the gl_client_array->Ptr field is
560 * really an offset from the start of the VBO, not a pointer.
562 struct st_buffer_object
*stobj
= st_buffer_object(bufobj
);
564 if (!stobj
|| !stobj
->buffer
) {
565 /* probably out of memory (or zero-sized buffer) */
569 vbuffer
[attr
].buffer
= NULL
;
570 pipe_resource_reference(&vbuffer
[attr
].buffer
, stobj
->buffer
);
571 vbuffer
[attr
].buffer_offset
= pointer_to_offset(array
->Ptr
);
579 uint divisor
= array
->InstanceDivisor
;
580 uint last_index
= divisor
? num_instances
/ divisor
: max_index
;
582 bytes
= stride
* last_index
+ element_size
;
584 ptr
= (void *) array
->Ptr
;
587 /* no array, use ctx->Current.Attrib[] value */
588 bytes
= element_size
= sizeof(ctx
->Current
.Attrib
[0]);
589 ptr
= (void *) ctx
->Current
.Attrib
[mesaAttr
];
596 vbuffer
[attr
].buffer
=
597 pipe_user_buffer_create(pipe
->screen
, ptr
, bytes
,
598 PIPE_BIND_VERTEX_BUFFER
);
600 vbuffer
[attr
].buffer_offset
= 0;
602 /* Track user vertex buffers. */
603 pipe_resource_reference(&st
->user_attrib
[attr
].buffer
, vbuffer
[attr
].buffer
);
604 st
->user_attrib
[attr
].element_size
= element_size
;
605 st
->user_attrib
[attr
].stride
= stride
;
606 st
->num_user_attribs
= MAX2(st
->num_user_attribs
, attr
+ 1);
608 if (!vbuffer
[attr
].buffer
) {
609 /* probably ran out of memory */
614 /* common-case setup */
615 vbuffer
[attr
].stride
= stride
; /* in bytes */
617 velements
[attr
].src_offset
= 0;
618 velements
[attr
].instance_divisor
= array
->InstanceDivisor
;
619 velements
[attr
].vertex_buffer_index
= attr
;
620 velements
[attr
].src_format
= st_pipe_vertex_format(array
->Type
,
625 assert(velements
[attr
].src_format
);
633 setup_index_buffer(struct st_context
*st
,
634 const struct _mesa_index_buffer
*ib
,
635 struct pipe_index_buffer
*ibuffer
)
637 struct pipe_context
*pipe
= st
->pipe
;
638 struct gl_buffer_object
*bufobj
= ib
->obj
;
640 ibuffer
->index_size
= vbo_sizeof_ib_type(ib
->type
);
642 /* get/create the index buffer object */
643 if (_mesa_is_bufferobj(bufobj
)) {
644 /* indices are in a real VBO */
645 struct st_buffer_object
*stobj
= st_buffer_object(bufobj
);
646 pipe_resource_reference(&ibuffer
->buffer
, stobj
->buffer
);
647 ibuffer
->offset
= pointer_to_offset(ib
->ptr
);
649 else if (st
->indexbuf_uploader
) {
650 u_upload_data(st
->indexbuf_uploader
, 0, ib
->count
* ibuffer
->index_size
,
651 ib
->ptr
, &ibuffer
->offset
, &ibuffer
->buffer
);
654 /* indices are in user space memory */
656 pipe_user_buffer_create(pipe
->screen
, (void *) ib
->ptr
,
657 ib
->count
* ibuffer
->index_size
,
658 PIPE_BIND_INDEX_BUFFER
);
661 cso_set_index_buffer(st
->cso_context
, ibuffer
);
666 * Prior to drawing, check that any uniforms referenced by the
667 * current shader have been set. If a uniform has not been set,
671 check_uniforms(struct gl_context
*ctx
)
673 struct gl_shader_program
*shProg
[3] = {
674 ctx
->Shader
.CurrentVertexProgram
,
675 ctx
->Shader
.CurrentGeometryProgram
,
676 ctx
->Shader
.CurrentFragmentProgram
,
680 for (j
= 0; j
< 3; j
++) {
683 if (shProg
[j
] == NULL
|| !shProg
[j
]->LinkStatus
)
686 for (i
= 0; i
< shProg
[j
]->NumUserUniformStorage
; i
++) {
687 const struct gl_uniform_storage
*u
= &shProg
[j
]->UniformStorage
[i
];
688 if (!u
->initialized
) {
690 "Using shader with uninitialized uniform: %s",
699 * Notes on primitive restart:
700 * The code below is used when the gallium driver does not support primitive
701 * restart itself. We map the index buffer, find the restart indexes, unmap
702 * the index buffer then draw the sub-primitives delineated by the restarts.
703 * A couple possible optimizations:
704 * 1. Save the list of sub-primitive (start, count) values in a list attached
705 * to the index buffer for re-use in subsequent draws. The list would be
706 * invalidated when the contents of the buffer changed.
707 * 2. If drawing triangle strips or quad strips, create a new index buffer
708 * that uses duplicated vertices to render the disjoint strips as one
709 * long strip. We'd have to be careful to avoid using too much memory
711 * Finally, some apps might perform better if they don't use primitive restart
712 * at all rather than this fallback path. Set MESA_EXTENSION_OVERRIDE to
713 * "-GL_NV_primitive_restart" to test that.
719 unsigned start
, count
;
724 * Scan the elements array to find restart indexes. Return a list
725 * of primitive (start,count) pairs to indicate how to draw the sub-
726 * primitives delineated by the restart index.
728 static struct sub_primitive
*
729 find_sub_primitives(const void *elements
, unsigned element_size
,
730 unsigned start
, unsigned end
, unsigned restart_index
,
731 unsigned *num_sub_prims
)
733 const unsigned max_prims
= end
- start
;
734 struct sub_primitive
*sub_prims
;
735 unsigned i
, cur_start
, cur_count
, num
;
737 sub_prims
= (struct sub_primitive
*)
738 malloc(max_prims
* sizeof(struct sub_primitive
));
749 #define SCAN_ELEMENTS(TYPE) \
750 for (i = start; i < end; i++) { \
751 if (((const TYPE *) elements)[i] == restart_index) { \
752 if (cur_count > 0) { \
753 assert(num < max_prims); \
754 sub_prims[num].start = cur_start; \
755 sub_prims[num].count = cur_count; \
765 if (cur_count > 0) { \
766 assert(num < max_prims); \
767 sub_prims[num].start = cur_start; \
768 sub_prims[num].count = cur_count; \
772 switch (element_size
) {
774 SCAN_ELEMENTS(ubyte
);
777 SCAN_ELEMENTS(ushort
);
783 assert(0 && "bad index_size in find_sub_primitives()");
788 *num_sub_prims
= num
;
795 * For gallium drivers that don't support the primitive restart
796 * feature, handle it here by breaking up the indexed primitive into
800 handle_fallback_primitive_restart(struct cso_context
*cso
,
801 struct pipe_context
*pipe
,
802 const struct _mesa_index_buffer
*ib
,
803 struct pipe_index_buffer
*ibuffer
,
804 struct pipe_draw_info
*orig_info
)
806 const unsigned start
= orig_info
->start
;
807 const unsigned count
= orig_info
->count
;
808 struct pipe_draw_info info
= *orig_info
;
809 struct pipe_transfer
*transfer
= NULL
;
810 unsigned instance
, i
;
811 const void *ptr
= NULL
;
812 struct sub_primitive
*sub_prims
;
813 unsigned num_sub_prims
;
815 assert(info
.indexed
);
816 assert(ibuffer
->buffer
);
819 if (!ibuffer
->buffer
|| !ib
)
822 info
.primitive_restart
= FALSE
;
823 info
.instance_count
= 1;
825 if (_mesa_is_bufferobj(ib
->obj
)) {
826 ptr
= pipe_buffer_map_range(pipe
, ibuffer
->buffer
,
827 start
* ibuffer
->index_size
, /* start */
828 count
* ibuffer
->index_size
, /* length */
829 PIPE_TRANSFER_READ
, &transfer
);
833 ptr
= (uint8_t*)ptr
+ (ibuffer
->offset
- start
* ibuffer
->index_size
);
841 sub_prims
= find_sub_primitives(ptr
, ibuffer
->index_size
,
842 0, count
, orig_info
->restart_index
,
846 pipe_buffer_unmap(pipe
, transfer
);
848 /* Now draw the sub primitives.
849 * Need to loop over instances as well to preserve draw order.
851 for (instance
= 0; instance
< orig_info
->instance_count
; instance
++) {
852 info
.start_instance
= instance
+ orig_info
->start_instance
;
853 for (i
= 0; i
< num_sub_prims
; i
++) {
854 info
.start
= sub_prims
[i
].start
;
855 info
.count
= sub_prims
[i
].count
;
856 if (u_trim_pipe_prim(info
.mode
, &info
.count
)) {
857 cso_draw_vbo(cso
, &info
);
868 * Translate OpenGL primtive type (GL_POINTS, GL_TRIANGLE_STRIP, etc) to
869 * the corresponding Gallium type.
872 translate_prim(const struct gl_context
*ctx
, unsigned prim
)
874 /* GL prims should match Gallium prims, spot-check a few */
875 assert(GL_POINTS
== PIPE_PRIM_POINTS
);
876 assert(GL_QUADS
== PIPE_PRIM_QUADS
);
877 assert(GL_TRIANGLE_STRIP_ADJACENCY
== PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
);
879 /* Avoid quadstrips if it's easy to do so:
880 * Note: it's important to do the correct trimming if we change the
881 * prim type! We do that wherever this function is called.
883 if (prim
== GL_QUAD_STRIP
&&
884 ctx
->Light
.ShadeModel
!= GL_FLAT
&&
885 ctx
->Polygon
.FrontMode
== GL_FILL
&&
886 ctx
->Polygon
.BackMode
== GL_FILL
)
887 prim
= GL_TRIANGLE_STRIP
;
894 * Setup vertex arrays and buffers prior to drawing.
895 * \return GL_TRUE for success, GL_FALSE otherwise (probably out of memory)
898 st_validate_varrays(struct gl_context
*ctx
,
899 const struct gl_client_array
**arrays
,
901 unsigned num_instances
)
903 struct st_context
*st
= st_context(ctx
);
904 const struct st_vertex_program
*vp
;
905 const struct st_vp_variant
*vpv
;
906 struct pipe_vertex_buffer vbuffer
[PIPE_MAX_SHADER_INPUTS
];
907 struct pipe_vertex_element velements
[PIPE_MAX_ATTRIBS
];
908 unsigned num_vbuffers
, num_velements
;
912 /* must get these after state validation! */
914 vpv
= st
->vp_variant
;
916 memset(velements
, 0, sizeof(struct pipe_vertex_element
) * vpv
->num_inputs
);
918 /* Unreference any user vertex buffers. */
919 for (i
= 0; i
< st
->num_user_attribs
; i
++) {
920 pipe_resource_reference(&st
->user_attrib
[i
].buffer
, NULL
);
922 st
->num_user_attribs
= 0;
925 * Setup the vbuffer[] and velements[] arrays.
927 if (is_interleaved_arrays(vp
, vpv
, arrays
)) {
928 if (!setup_interleaved_attribs(ctx
, vp
, vpv
, arrays
, vbuffer
, velements
,
929 max_index
, num_instances
)) {
934 num_velements
= vpv
->num_inputs
;
935 if (num_velements
== 0)
939 if (!setup_non_interleaved_attribs(ctx
, vp
, vpv
, arrays
,
940 vbuffer
, velements
, max_index
,
945 num_vbuffers
= vpv
->num_inputs
;
946 num_velements
= vpv
->num_inputs
;
949 cso_set_vertex_buffers(st
->cso_context
, num_vbuffers
, vbuffer
);
950 cso_set_vertex_elements(st
->cso_context
, num_velements
, velements
);
952 /* unreference buffers (frees wrapped user-space buffer objects)
953 * This is OK, because the pipe driver should reference buffers by itself
954 * in set_vertex_buffers. */
955 for (attr
= 0; attr
< num_vbuffers
; attr
++) {
956 pipe_resource_reference(&vbuffer
[attr
].buffer
, NULL
);
957 assert(!vbuffer
[attr
].buffer
);
965 * This function gets plugged into the VBO module and is called when
966 * we have something to render.
967 * Basically, translate the information into the format expected by gallium.
970 st_draw_vbo(struct gl_context
*ctx
,
971 const struct gl_client_array
**arrays
,
972 const struct _mesa_prim
*prims
,
974 const struct _mesa_index_buffer
*ib
,
975 GLboolean index_bounds_valid
,
978 struct gl_transform_feedback_object
*tfb_vertcount
)
980 struct st_context
*st
= st_context(ctx
);
981 struct pipe_context
*pipe
= st
->pipe
;
982 struct pipe_index_buffer ibuffer
= {0};
983 struct pipe_draw_info info
;
984 unsigned i
, num_instances
= 1;
985 unsigned max_index_plus_base
;
986 GLboolean new_array
=
988 (st
->dirty
.mesa
& (_NEW_ARRAY
| _NEW_PROGRAM
| _NEW_BUFFER_OBJECT
)) != 0;
990 /* Mesa core state should have been validated already */
991 assert(ctx
->NewState
== 0x0);
994 int max_base_vertex
= 0;
996 /* Gallium probably doesn't want this in some cases. */
997 if (!index_bounds_valid
)
998 if (!all_varyings_in_vbos(arrays
))
999 vbo_get_minmax_indices(ctx
, prims
, ib
, &min_index
, &max_index
,
1002 for (i
= 0; i
< nr_prims
; i
++) {
1003 num_instances
= MAX2(num_instances
, prims
[i
].num_instances
);
1004 max_base_vertex
= MAX2(max_base_vertex
, prims
[i
].basevertex
);
1007 /* Compute the sum of max_index and max_base_vertex. That's the value
1008 * we need to use when creating buffers.
1010 if (max_index
== ~0)
1011 max_index_plus_base
= max_index
;
1013 max_index_plus_base
= max_index
+ max_base_vertex
;
1016 /* Get min/max index for non-indexed drawing. */
1020 for (i
= 0; i
< nr_prims
; i
++) {
1021 min_index
= MIN2(min_index
, prims
[i
].start
);
1022 max_index
= MAX2(max_index
, prims
[i
].start
+ prims
[i
].count
- 1);
1023 num_instances
= MAX2(num_instances
, prims
[i
].num_instances
);
1026 /* The base vertex offset only applies to indexed drawing */
1027 max_index_plus_base
= max_index
;
1030 /* Validate state. */
1032 GLboolean vertDataEdgeFlags
;
1034 vertDataEdgeFlags
= arrays
[VERT_ATTRIB_EDGEFLAG
]->BufferObj
&&
1035 arrays
[VERT_ATTRIB_EDGEFLAG
]->BufferObj
->Name
;
1036 if (vertDataEdgeFlags
!= st
->vertdata_edgeflags
) {
1037 st
->vertdata_edgeflags
= vertDataEdgeFlags
;
1038 st
->dirty
.st
|= ST_NEW_EDGEFLAGS_DATA
;
1041 st_validate_state(st
);
1044 if (!st_validate_varrays(ctx
, arrays
, max_index_plus_base
,
1046 /* probably out of memory, no-op the draw call */
1052 if (MESA_VERBOSE
& VERBOSE_GLSL
) {
1053 check_uniforms(ctx
);
1056 (void) check_uniforms
;
1060 /* Notify the driver that the content of user buffers may have been
1062 assert(max_index
>= min_index
);
1063 if (!new_array
&& st
->num_user_attribs
) {
1064 for (i
= 0; i
< st
->num_user_attribs
; i
++) {
1065 if (st
->user_attrib
[i
].buffer
) {
1066 unsigned element_size
= st
->user_attrib
[i
].element_size
;
1067 unsigned stride
= st
->user_attrib
[i
].stride
;
1068 unsigned min_offset
= min_index
* stride
;
1069 unsigned max_offset
= max_index_plus_base
* stride
+ element_size
;
1071 assert(max_offset
> min_offset
);
1073 pipe
->redefine_user_buffer(pipe
, st
->user_attrib
[i
].buffer
,
1075 max_offset
- min_offset
);
1080 util_draw_init_info(&info
);
1082 setup_index_buffer(st
, ib
, &ibuffer
);
1084 info
.indexed
= TRUE
;
1085 if (min_index
!= ~0 && max_index
!= ~0) {
1086 info
.min_index
= min_index
;
1087 info
.max_index
= max_index
;
1090 /* The VBO module handles restart for the non-indexed GLDrawArrays
1091 * so we only set these fields for indexed drawing:
1093 info
.primitive_restart
= ctx
->Array
.PrimitiveRestart
;
1094 info
.restart_index
= ctx
->Array
.RestartIndex
;
1097 /* Set info.count_from_stream_output. */
1098 if (tfb_vertcount
) {
1099 st_transform_feedback_draw_init(tfb_vertcount
, &info
);
1102 /* do actual drawing */
1103 for (i
= 0; i
< nr_prims
; i
++) {
1104 info
.mode
= translate_prim( ctx
, prims
[i
].mode
);
1105 info
.start
= prims
[i
].start
;
1106 info
.count
= prims
[i
].count
;
1107 info
.instance_count
= prims
[i
].num_instances
;
1108 info
.index_bias
= prims
[i
].basevertex
;
1110 info
.min_index
= info
.start
;
1111 info
.max_index
= info
.start
+ info
.count
- 1;
1114 if (info
.count_from_stream_output
) {
1115 cso_draw_vbo(st
->cso_context
, &info
);
1117 else if (info
.primitive_restart
) {
1118 if (st
->sw_primitive_restart
) {
1119 /* Handle primitive restart for drivers that doesn't support it */
1120 handle_fallback_primitive_restart(st
->cso_context
, pipe
, ib
,
1124 /* don't trim, restarts might be inside index list */
1125 cso_draw_vbo(st
->cso_context
, &info
);
1128 else if (u_trim_pipe_prim(info
.mode
, &info
.count
))
1129 cso_draw_vbo(st
->cso_context
, &info
);
1132 pipe_resource_reference(&ibuffer
.buffer
, NULL
);
1137 st_init_draw(struct st_context
*st
)
1139 struct gl_context
*ctx
= st
->ctx
;
1141 vbo_set_draw_func(ctx
, st_draw_vbo
);
1143 #if FEATURE_feedback || FEATURE_rastpos
1144 st
->draw
= draw_create(st
->pipe
); /* for selection/feedback */
1146 /* Disable draw options that might convert points/lines to tris, etc.
1147 * as that would foul-up feedback/selection mode.
1149 draw_wide_line_threshold(st
->draw
, 1000.0f
);
1150 draw_wide_point_threshold(st
->draw
, 1000.0f
);
1151 draw_enable_line_stipple(st
->draw
, FALSE
);
1152 draw_enable_point_sprites(st
->draw
, FALSE
);
1158 st_destroy_draw(struct st_context
*st
)
1160 #if FEATURE_feedback || FEATURE_rastpos
1161 draw_destroy(st
->draw
);