1 /**************************************************************************
3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
33 #include "api_validate.h"
37 #include "brw_defines.h"
38 #include "brw_attrib.h"
39 #include "brw_context.h"
41 #include "brw_state.h"
42 #include "brw_fallback.h"
44 #include "intel_ioctl.h"
45 #include "intel_batchbuffer.h"
46 #include "intel_buffer_objects.h"
53 static GLuint hw_prim
[GL_POLYGON
+1] = {
67 static const GLenum reduced_prim
[GL_POLYGON
+1] = {
81 /* When the primitive changes, set a state bit and re-validate. Not
82 * the nicest and would rather deal with this by having all the
83 * programs be immune to the active primitive (ie. cope with all
84 * possibilities). That may not be realistic however.
86 static GLuint
brw_set_prim(struct brw_context
*brw
, GLenum prim
)
88 if (INTEL_DEBUG
& DEBUG_PRIMS
)
89 _mesa_printf("PRIM: %s\n", _mesa_lookup_enum_by_nr(prim
));
91 /* Slight optimization to avoid the GS program when not needed:
93 if (prim
== GL_QUAD_STRIP
&&
94 brw
->attribs
.Light
->ShadeModel
!= GL_FLAT
&&
95 brw
->attribs
.Polygon
->FrontMode
== GL_FILL
&&
96 brw
->attribs
.Polygon
->BackMode
== GL_FILL
)
97 prim
= GL_TRIANGLE_STRIP
;
99 if (prim
!= brw
->primitive
) {
100 brw
->primitive
= prim
;
101 brw
->state
.dirty
.brw
|= BRW_NEW_PRIMITIVE
;
103 if (reduced_prim
[prim
] != brw
->intel
.reduced_primitive
) {
104 brw
->intel
.reduced_primitive
= reduced_prim
[prim
];
105 brw
->state
.dirty
.brw
|= BRW_NEW_REDUCED_PRIMITIVE
;
108 brw_validate_state(brw
);
111 return hw_prim
[prim
];
115 static GLuint
trim(GLenum prim
, GLuint length
)
117 if (prim
== GL_QUAD_STRIP
)
118 return length
> 3 ? (length
- length
% 2) : 0;
119 else if (prim
== GL_QUADS
)
120 return length
- length
% 4;
126 static void brw_emit_cliprect( struct brw_context
*brw
,
127 const drm_clip_rect_t
*rect
)
129 struct brw_drawrect bdr
;
131 bdr
.header
.opcode
= CMD_DRAW_RECT
;
132 bdr
.header
.length
= sizeof(bdr
)/4 - 2;
134 bdr
.xmax
= rect
->x2
- 1;
136 bdr
.ymax
= rect
->y2
- 1;
137 bdr
.xorg
= brw
->intel
.drawX
;
138 bdr
.yorg
= brw
->intel
.drawY
;
140 intel_batchbuffer_data( brw
->intel
.batch
, &bdr
, sizeof(bdr
),
141 INTEL_BATCH_NO_CLIPRECTS
);
145 static void brw_emit_prim( struct brw_context
*brw
,
146 const struct brw_draw_prim
*prim
)
149 struct brw_3d_primitive prim_packet
;
151 if (INTEL_DEBUG
& DEBUG_PRIMS
)
152 _mesa_printf("PRIM: %s %d %d\n", _mesa_lookup_enum_by_nr(prim
->mode
),
153 prim
->start
, prim
->count
);
155 prim_packet
.header
.opcode
= CMD_3D_PRIM
;
156 prim_packet
.header
.length
= sizeof(prim_packet
)/4 - 2;
157 prim_packet
.header
.pad
= 0;
158 prim_packet
.header
.topology
= brw_set_prim(brw
, prim
->mode
);
159 prim_packet
.header
.indexed
= prim
->indexed
;
161 prim_packet
.verts_per_instance
= trim(prim
->mode
, prim
->count
);
162 prim_packet
.start_vert_location
= prim
->start
;
163 prim_packet
.instance_count
= 1;
164 prim_packet
.start_instance_location
= 0;
165 prim_packet
.base_vert_location
= 0;
167 if (prim_packet
.verts_per_instance
) {
168 intel_batchbuffer_data( brw
->intel
.batch
, &prim_packet
, sizeof(prim_packet
),
169 INTEL_BATCH_NO_CLIPRECTS
);
175 static void update_current_size( struct gl_client_array
*array
)
177 const GLfloat
*ptr
= (const GLfloat
*)array
->Ptr
;
179 assert(array
->StrideB
== 0);
180 assert(array
->Type
== GL_FLOAT
|| array
->Type
== GL_UNSIGNED_BYTE
);
184 else if (ptr
[2] != 0.0)
186 else if (ptr
[1] != 0.0)
194 /* Fill in any gaps in passed arrays with pointers to current
197 static void brw_merge_inputs( struct brw_context
*brw
,
198 const struct gl_client_array
*arrays
[])
200 struct gl_client_array
*current_values
= brw
->vb
.current_values
;
201 struct brw_vertex_element
*inputs
= brw
->vb
.inputs
;
202 struct brw_vertex_info old
= brw
->vb
.info
;
205 memset(inputs
, 0, sizeof(*inputs
));
206 memset(&brw
->vb
.info
, 0, sizeof(brw
->vb
.info
));
208 for (i
= 0; i
< BRW_ATTRIB_MAX
; i
++) {
209 if (arrays
[i
] && arrays
[i
]->Enabled
)
211 brw
->vb
.inputs
[i
].glarray
= arrays
[i
];
212 brw
->vb
.info
.varying
|= (GLuint64EXT
) 1 << i
;
216 brw
->vb
.inputs
[i
].glarray
= ¤t_values
[i
];
217 update_current_size(¤t_values
[i
]);
220 brw
->vb
.info
.sizes
[i
/16] |= (inputs
[i
].glarray
->Size
- 1) << ((i
%16) * 2);
223 /* Raise statechanges if input sizes and varying have changed:
225 if (memcmp(brw
->vb
.info
.sizes
, old
.sizes
, sizeof(old
.sizes
)) != 0)
226 brw
->state
.dirty
.brw
|= BRW_NEW_INPUT_DIMENSIONS
;
228 if (brw
->vb
.info
.varying
!= old
.varying
)
229 brw
->state
.dirty
.brw
|= BRW_NEW_INPUT_VARYING
;
232 static GLboolean
check_fallbacks( struct brw_context
*brw
,
233 const struct brw_draw_prim
*prim
,
238 if (!brw
->intel
.strict_conformance
)
241 if (brw
->attribs
.Polygon
->SmoothFlag
) {
242 for (i
= 0; i
< nr_prims
; i
++)
243 if (reduced_prim
[prim
[i
].mode
] == GL_TRIANGLES
)
247 /* BRW hardware will do AA lines, but they are non-conformant it
248 * seems. TBD whether we keep this fallback:
250 if (brw
->attribs
.Line
->SmoothFlag
) {
251 for (i
= 0; i
< nr_prims
; i
++)
252 if (reduced_prim
[prim
[i
].mode
] == GL_LINES
)
256 /* Stipple -- these fallbacks could be resolved with a little
259 if (brw
->attribs
.Line
->StippleFlag
) {
260 for (i
= 0; i
< nr_prims
; i
++) {
261 /* GS doesn't get enough information to know when to reset
262 * the stipple counter?!?
264 if (prim
[i
].mode
== GL_LINE_LOOP
)
267 if (prim
[i
].mode
== GL_POLYGON
&&
268 (brw
->attribs
.Polygon
->FrontMode
== GL_LINE
||
269 brw
->attribs
.Polygon
->BackMode
== GL_LINE
))
275 if (brw
->attribs
.Point
->SmoothFlag
) {
276 for (i
= 0; i
< nr_prims
; i
++)
277 if (prim
[i
].mode
== GL_POINTS
)
285 static GLboolean
brw_try_draw_prims( GLcontext
*ctx
,
286 const struct gl_client_array
*arrays
[],
287 const struct brw_draw_prim
*prim
,
289 const struct brw_draw_index_buffer
*ib
,
294 struct intel_context
*intel
= intel_context(ctx
);
295 struct brw_context
*brw
= brw_context(ctx
);
296 GLboolean retval
= GL_FALSE
;
300 _mesa_update_state( ctx
);
302 /* Bind all inputs, derive varying and size information:
304 brw_merge_inputs( brw
, arrays
);
306 /* Have to validate state quite late. Will rebuild tnl_program,
307 * which depends on varying information.
309 * Note this is where brw->vs->prog_data.inputs_read is calculated,
310 * so can't access it earlier.
313 LOCK_HARDWARE(intel
);
315 if (brw
->intel
.numClipRects
== 0) {
316 assert(intel
->batch
->ptr
== intel
->batch
->map
+ intel
->batch
->offset
);
317 UNLOCK_HARDWARE(intel
);
322 assert(intel
->locked
);
326 /* Set the first primitive early, ahead of validate_state:
328 brw_set_prim(brw
, prim
[0].mode
);
330 /* XXX: Need to separate validate and upload of state.
332 brw_validate_state( brw
);
334 /* Various fallback checks:
336 if (brw
->intel
.Fallback
)
339 if (check_fallbacks( brw
, prim
, nr_prims
))
342 /* Upload index, vertex data:
345 brw_upload_indices( brw
, ib
);
347 if (!brw_upload_vertices( brw
, min_index
, max_index
)) {
351 /* For single cliprect, state is already emitted:
353 if (brw
->intel
.numClipRects
== 1) {
354 for (i
= 0; i
< nr_prims
; i
++) {
355 brw_emit_prim(brw
, &prim
[i
]);
359 /* Otherwise, explicitly do the cliprects at this point:
361 for (j
= 0; j
< brw
->intel
.numClipRects
; j
++) {
362 brw_emit_cliprect(brw
, &brw
->intel
.pClipRects
[j
]);
364 /* Emit prims to batchbuffer:
366 for (i
= 0; i
< nr_prims
; i
++) {
367 brw_emit_prim(brw
, &prim
[i
]);
372 intel
->need_flush
= GL_TRUE
;
378 /* Currently have to do this to synchronize with the map/unmap of
379 * the vertex buffer in brw_exec_api.c. Not sure if there is any
380 * way around this, as not every flush is due to a buffer filling
383 if (!intel_batchbuffer_flush( brw
->intel
.batch
)) {
384 DBG("%s intel_batchbuffer_flush failed\n", __FUNCTION__
);
388 if (retval
&& intel
->thrashing
) {
392 /* Free any old data so it doesn't clog up texture memory - we
393 * won't be referencing it again.
395 while (brw
->vb
.upload
.wrap
!= brw
->vb
.upload
.buf
) {
396 ctx
->Driver
.BufferData(ctx
,
398 BRW_UPLOAD_INIT_SIZE
,
401 brw
->vb
.upload
.vbo
[brw
->vb
.upload
.wrap
]);
402 brw
->vb
.upload
.wrap
++;
403 brw
->vb
.upload
.wrap
%= BRW_NR_UPLOAD_BUFS
;
406 UNLOCK_HARDWARE(intel
);
409 DBG("%s failed\n", __FUNCTION__
);
415 GLboolean
brw_draw_prims( GLcontext
*ctx
,
416 const struct gl_client_array
*arrays
[],
417 const struct brw_draw_prim
*prim
,
419 const struct brw_draw_index_buffer
*ib
,
424 struct intel_context
*intel
= intel_context(ctx
);
427 retval
= brw_try_draw_prims(ctx
, arrays
, prim
, nr_prims
, ib
, min_index
, max_index
, flags
);
430 if (!retval
&& bmError(intel
)) {
433 /* This looks like out-of-memory but potentially we have
434 * situation where there is enough memory but it has become
435 * fragmented. Clear out all heaps and start from scratch by
436 * faking a contended lock event: (done elsewhere)
439 /* Then try a second time only to upload textures and draw the
442 retval
= brw_try_draw_prims(ctx
, arrays
, prim
, nr_prims
, ib
, min_index
, max_index
, flags
);
445 if (intel
->aub_file
&& (INTEL_DEBUG
& DEBUG_SYNC
)) {
446 intelFinish( &intel
->ctx
);
455 static void brw_invalidate_vbo_cb( struct intel_context
*intel
, void *ptr
)
457 /* nothing to do, we don't rely on the contents being preserved */
461 void brw_draw_init( struct brw_context
*brw
)
463 GLcontext
*ctx
= &brw
->intel
.ctx
;
466 brw
->vb
.upload
.size
= BRW_UPLOAD_INIT_SIZE
;
468 for (i
= 0; i
< BRW_NR_UPLOAD_BUFS
; i
++) {
469 brw
->vb
.upload
.vbo
[i
] = ctx
->Driver
.NewBufferObject(ctx
, 1, GL_ARRAY_BUFFER_ARB
);
471 /* XXX: Set these to no-backing-store
473 bmBufferSetInvalidateCB(&brw
->intel
,
474 intel_bufferobj_buffer(intel_buffer_object(brw
->vb
.upload
.vbo
[i
])),
475 brw_invalidate_vbo_cb
,
481 ctx
->Driver
.BufferData( ctx
,
483 BRW_UPLOAD_INIT_SIZE
,
486 brw
->vb
.upload
.vbo
[0] );
489 brw_init_current_values(ctx
, brw
->vb
.current_values
);
492 void brw_draw_destroy( struct brw_context
*brw
)
494 GLcontext
*ctx
= &brw
->intel
.ctx
;
497 for (i
= 0; i
< BRW_NR_UPLOAD_BUFS
; i
++)
498 ctx
->Driver
.DeleteBuffer(ctx
, brw
->vb
.upload
.vbo
[i
]);