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>
32
#include "util/u_math.h"
34
#include "brw_context.h"
36
#include "brw_state.h"
37
#include "brw_defines.h"
38
#include "brw_debug.h"
40
struct brw_clip_unit_key {
41
unsigned int total_grf;
42
unsigned int urb_entry_read_length;
43
unsigned int curb_entry_read_length;
44
unsigned int clip_mode;
46
unsigned int curbe_offset;
48
unsigned int nr_urb_entries, urb_size;
50
GLboolean depth_clamp;
54
clip_unit_populate_key(struct brw_context *brw, struct brw_clip_unit_key *key)
56
memset(key, 0, sizeof(*key));
58
/* CACHE_NEW_CLIP_PROG */
59
key->total_grf = brw->clip.prog_data->total_grf;
60
key->urb_entry_read_length = brw->clip.prog_data->urb_read_length;
61
key->curb_entry_read_length = brw->clip.prog_data->curb_read_length;
62
key->clip_mode = brw->clip.prog_data->clip_mode;
64
/* BRW_NEW_CURBE_OFFSETS */
65
key->curbe_offset = brw->curbe.clip_start;
67
/* BRW_NEW_URB_FENCE */
68
key->nr_urb_entries = brw->urb.nr_clip_entries;
69
key->urb_size = brw->urb.vsize;
72
key->depth_clamp = 0; /* XXX: add this to gallium: ctx->Transform.DepthClamp; */
75
static enum pipe_error
76
clip_unit_create_from_key(struct brw_context *brw,
77
struct brw_clip_unit_key *key,
78
struct brw_winsys_reloc *reloc,
79
struct brw_winsys_buffer **bo_out)
81
struct brw_clip_unit_state clip;
84
memset(&clip, 0, sizeof(clip));
86
clip.thread0.grf_reg_count = align(key->total_grf, 16) / 16 - 1;
88
clip.thread0.kernel_start_pointer = 0;
90
clip.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
91
clip.thread1.single_program_flow = 1;
93
clip.thread3.urb_entry_read_length = key->urb_entry_read_length;
94
clip.thread3.const_urb_entry_read_length = key->curb_entry_read_length;
95
clip.thread3.const_urb_entry_read_offset = key->curbe_offset * 2;
96
clip.thread3.dispatch_grf_start_reg = 1;
97
clip.thread3.urb_entry_read_offset = 0;
99
clip.thread4.nr_urb_entries = key->nr_urb_entries;
100
clip.thread4.urb_entry_allocation_size = key->urb_size - 1;
101
/* If we have enough clip URB entries to run two threads, do so.
103
if (key->nr_urb_entries >= 10) {
104
/* Half of the URB entries go to each thread, and it has to be an
107
assert(key->nr_urb_entries % 2 == 0);
109
/* Although up to 16 concurrent Clip threads are allowed on IGDNG,
110
* only 2 threads can output VUEs at a time.
113
clip.thread4.max_threads = 16 - 1;
115
clip.thread4.max_threads = 2 - 1;
117
assert(key->nr_urb_entries >= 5);
118
clip.thread4.max_threads = 1 - 1;
121
if (BRW_DEBUG & DEBUG_SINGLE_THREAD)
122
clip.thread4.max_threads = 0;
124
if (BRW_DEBUG & DEBUG_STATS)
125
clip.thread4.stats_enable = 1;
127
clip.clip5.userclip_enable_flags = 0x7f;
128
clip.clip5.userclip_must_clip = 1;
129
clip.clip5.guard_band_enable = 0;
130
if (!key->depth_clamp)
131
clip.clip5.viewport_z_clip_enable = 1;
132
clip.clip5.viewport_xy_clip_enable = 1;
133
clip.clip5.vertex_position_space = BRW_CLIP_NDCSPACE;
134
clip.clip5.api_mode = BRW_CLIP_API_OGL;
135
clip.clip5.clip_mode = key->clip_mode;
138
clip.clip5.negative_w_clip_test = 1;
140
clip.clip6.clipper_viewport_state_ptr = 0;
141
clip.viewport_xmin = -1;
142
clip.viewport_xmax = 1;
143
clip.viewport_ymin = -1;
144
clip.viewport_ymax = 1;
146
ret = brw_upload_cache(&brw->cache, BRW_CLIP_UNIT,
158
static int upload_clip_unit( struct brw_context *brw )
160
struct brw_clip_unit_key key;
161
struct brw_winsys_reloc reloc[1];
162
unsigned grf_reg_count;
165
clip_unit_populate_key(brw, &key);
167
grf_reg_count = align(key.total_grf, 16) / 16 - 1;
169
/* clip program relocation
171
* XXX: these reloc structs are long lived and only need to be
172
* updated when the bound BO changes. Hopefully the stuff mixed in
173
* in the delta's is non-orthogonal.
175
assert(brw->clip.prog_bo);
176
make_reloc(&reloc[0],
179
offsetof(struct brw_clip_unit_state, thread0),
183
if (brw_search_cache(&brw->cache, BRW_CLIP_UNIT,
187
&brw->clip.state_bo))
192
ret = clip_unit_create_from_key(brw, &key,
194
&brw->clip.state_bo);
201
const struct brw_tracked_state brw_clip_unit = {
204
.brw = (BRW_NEW_CURBE_OFFSETS |
206
.cache = CACHE_NEW_CLIP_PROG
208
.prepare = upload_clip_unit,