2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
6 Permission is hereby granted, free of charge, to any person obtaining
7 a 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, sublicense, 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
16 portions of the Software.
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **********************************************************************/
29 * Keith Whitwell <keith@tungstengraphics.com>
34 #include "brw_context.h"
35 #include "brw_state.h"
36 #include "brw_defines.h"
37 #include "main/macros.h"
39 struct brw_vs_unit_key
{
40 unsigned int total_grf
;
41 unsigned int urb_entry_read_length
;
42 unsigned int curb_entry_read_length
;
44 unsigned int curbe_offset
;
46 unsigned int nr_urb_entries
, urb_size
;
48 unsigned int nr_surfaces
;
52 vs_unit_populate_key(struct brw_context
*brw
, struct brw_vs_unit_key
*key
)
54 GLcontext
*ctx
= &brw
->intel
.ctx
;
56 memset(key
, 0, sizeof(*key
));
58 /* CACHE_NEW_VS_PROG */
59 key
->total_grf
= brw
->vs
.prog_data
->total_grf
;
60 key
->urb_entry_read_length
= brw
->vs
.prog_data
->urb_read_length
;
61 key
->curb_entry_read_length
= brw
->vs
.prog_data
->curb_read_length
;
63 /* BRW_NEW_URB_FENCE */
64 key
->nr_urb_entries
= brw
->urb
.nr_vs_entries
;
65 key
->urb_size
= brw
->urb
.vsize
;
67 /* BRW_NEW_NR_VS_SURFACES */
68 key
->nr_surfaces
= brw
->vs
.nr_surfaces
;
70 /* BRW_NEW_CURBE_OFFSETS, _NEW_TRANSFORM */
71 if (ctx
->Transform
.ClipPlanesEnabled
) {
72 /* Note that we read in the userclip planes as well, hence
75 key
->curbe_offset
= brw
->curbe
.clip_start
;
78 key
->curbe_offset
= brw
->curbe
.vs_start
;
83 vs_unit_create_from_key(struct brw_context
*brw
, struct brw_vs_unit_key
*key
)
85 struct intel_context
*intel
= &brw
->intel
;
86 struct brw_vs_unit_state vs
;
88 int chipset_max_threads
;
90 memset(&vs
, 0, sizeof(vs
));
92 vs
.thread0
.kernel_start_pointer
= brw
->vs
.prog_bo
->offset
>> 6; /* reloc */
93 vs
.thread0
.grf_reg_count
= ALIGN(key
->total_grf
, 16) / 16 - 1;
94 vs
.thread1
.floating_point_mode
= BRW_FLOATING_POINT_NON_IEEE_754
;
95 /* Choosing multiple program flow means that we may get 2-vertex threads,
96 * which will have the channel mask for dwords 4-7 enabled in the thread,
97 * and those dwords will be written to the second URB handle when we
98 * brw_urb_WRITE() results.
100 vs
.thread1
.single_program_flow
= 0;
102 if (intel
->is_ironlake
)
103 vs
.thread1
.binding_table_entry_count
= 0; /* hardware requirement */
105 vs
.thread1
.binding_table_entry_count
= key
->nr_surfaces
;
107 vs
.thread3
.urb_entry_read_length
= key
->urb_entry_read_length
;
108 vs
.thread3
.const_urb_entry_read_length
= key
->curb_entry_read_length
;
109 vs
.thread3
.dispatch_grf_start_reg
= 1;
110 vs
.thread3
.urb_entry_read_offset
= 0;
111 vs
.thread3
.const_urb_entry_read_offset
= key
->curbe_offset
* 2;
113 if (intel
->is_ironlake
) {
114 switch (key
->nr_urb_entries
) {
126 vs
.thread4
.nr_urb_entries
= key
->nr_urb_entries
>> 2;
132 switch (key
->nr_urb_entries
) {
139 assert(BRW_IS_G4X(brw
));
144 vs
.thread4
.nr_urb_entries
= key
->nr_urb_entries
;
147 vs
.thread4
.urb_entry_allocation_size
= key
->urb_size
- 1;
149 if (intel
->is_ironlake
)
150 chipset_max_threads
= 72;
151 else if (BRW_IS_G4X(brw
))
152 chipset_max_threads
= 32;
154 chipset_max_threads
= 16;
155 vs
.thread4
.max_threads
= CLAMP(key
->nr_urb_entries
/ 2,
156 1, chipset_max_threads
) - 1;
158 if (INTEL_DEBUG
& DEBUG_SINGLE_THREAD
)
159 vs
.thread4
.max_threads
= 0;
161 /* No samplers for ARB_vp programs:
163 /* It has to be set to 0 for IGDNG
165 vs
.vs5
.sampler_count
= 0;
167 if (INTEL_DEBUG
& DEBUG_STATS
)
168 vs
.thread4
.stats_enable
= 1;
170 /* Vertex program always enabled:
172 vs
.vs6
.vs_enable
= 1;
174 bo
= brw_upload_cache(&brw
->cache
, BRW_VS_UNIT
,
180 /* Emit VS program relocation */
181 dri_bo_emit_reloc(bo
,
182 I915_GEM_DOMAIN_INSTRUCTION
, 0,
183 vs
.thread0
.grf_reg_count
<< 1,
184 offsetof(struct brw_vs_unit_state
, thread0
),
190 static void prepare_vs_unit(struct brw_context
*brw
)
192 struct brw_vs_unit_key key
;
194 vs_unit_populate_key(brw
, &key
);
196 dri_bo_unreference(brw
->vs
.state_bo
);
197 brw
->vs
.state_bo
= brw_search_cache(&brw
->cache
, BRW_VS_UNIT
,
201 if (brw
->vs
.state_bo
== NULL
) {
202 brw
->vs
.state_bo
= vs_unit_create_from_key(brw
, &key
);
206 const struct brw_tracked_state brw_vs_unit
= {
208 .mesa
= _NEW_TRANSFORM
,
209 .brw
= (BRW_NEW_CURBE_OFFSETS
|
210 BRW_NEW_NR_VS_SURFACES
|
212 .cache
= CACHE_NEW_VS_PROG
214 .prepare
= prepare_vs_unit
,