mesa/i965/i915/r200: eliminate gl_vertex_program
[mesa.git] / src / mesa / drivers / dri / i965 / gen6_urb.c
1 /*
2 * Copyright © 2009 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 *
23 * Authors:
24 * Eric Anholt <eric@anholt.net>
25 *
26 */
27
28 #include "main/macros.h"
29 #include "intel_batchbuffer.h"
30 #include "brw_context.h"
31 #include "brw_state.h"
32 #include "brw_defines.h"
33
34 /**
35 * When the GS is not in use, we assign the entire URB space to the VS. When
36 * the GS is in use, we split the URB space evenly between the VS and the GS.
37 * This is not ideal, but it's simple.
38 *
39 * URB size / 2 URB size / 2
40 * _____________-______________ _____________-______________
41 * / \ / \
42 * +-------------------------------------------------------------+
43 * | Vertex Shader Entries | Geometry Shader Entries |
44 * +-------------------------------------------------------------+
45 *
46 * Sandybridge GT1 has 32kB of URB space, while GT2 has 64kB.
47 * (See the Sandybridge PRM, Volume 2, Part 1, Section 1.4.7: 3DSTATE_URB.)
48 */
49 void
50 gen6_upload_urb(struct brw_context *brw, unsigned vs_size,
51 bool gs_present, unsigned gs_size)
52 {
53 int nr_vs_entries, nr_gs_entries;
54 int total_urb_size = brw->urb.size * 1024; /* in bytes */
55 const struct gen_device_info *devinfo = &brw->screen->devinfo;
56
57 /* Calculate how many entries fit in each stage's section of the URB */
58 if (gs_present) {
59 nr_vs_entries = (total_urb_size/2) / (vs_size * 128);
60 nr_gs_entries = (total_urb_size/2) / (gs_size * 128);
61 } else {
62 nr_vs_entries = total_urb_size / (vs_size * 128);
63 nr_gs_entries = 0;
64 }
65
66 /* Then clamp to the maximum allowed by the hardware */
67 if (nr_vs_entries > devinfo->urb.max_vs_entries)
68 nr_vs_entries = devinfo->urb.max_vs_entries;
69
70 if (nr_gs_entries > devinfo->urb.max_gs_entries)
71 nr_gs_entries = devinfo->urb.max_gs_entries;
72
73 /* Finally, both must be a multiple of 4 (see 3DSTATE_URB in the PRM). */
74 brw->urb.nr_vs_entries = ROUND_DOWN_TO(nr_vs_entries, 4);
75 brw->urb.nr_gs_entries = ROUND_DOWN_TO(nr_gs_entries, 4);
76
77 assert(brw->urb.nr_vs_entries >= devinfo->urb.min_vs_entries);
78 assert(brw->urb.nr_vs_entries % 4 == 0);
79 assert(brw->urb.nr_gs_entries % 4 == 0);
80 assert(vs_size <= 5);
81 assert(gs_size <= 5);
82
83 BEGIN_BATCH(3);
84 OUT_BATCH(_3DSTATE_URB << 16 | (3 - 2));
85 OUT_BATCH(((vs_size - 1) << GEN6_URB_VS_SIZE_SHIFT) |
86 ((brw->urb.nr_vs_entries) << GEN6_URB_VS_ENTRIES_SHIFT));
87 OUT_BATCH(((gs_size - 1) << GEN6_URB_GS_SIZE_SHIFT) |
88 ((brw->urb.nr_gs_entries) << GEN6_URB_GS_ENTRIES_SHIFT));
89 ADVANCE_BATCH();
90
91 /* From the PRM Volume 2 part 1, section 1.4.7:
92 *
93 * Because of a urb corruption caused by allocating a previous gsunit’s
94 * urb entry to vsunit software is required to send a "GS NULL
95 * Fence"(Send URB fence with VS URB size == 1 and GS URB size == 0) plus
96 * a dummy DRAW call before any case where VS will be taking over GS URB
97 * space.
98 *
99 * It is not clear exactly what this means ("URB fence" is a command that
100 * doesn't exist on Gen6). So for now we just do a full pipeline flush as
101 * a workaround.
102 */
103 if (brw->urb.gs_present && !gs_present)
104 brw_emit_mi_flush(brw);
105 brw->urb.gs_present = gs_present;
106 }
107
108 static void
109 upload_urb(struct brw_context *brw)
110 {
111 /* BRW_NEW_VS_PROG_DATA */
112 const struct brw_vue_prog_data *vs_vue_prog_data =
113 brw_vue_prog_data(brw->vs.base.prog_data);
114 const unsigned vs_size = MAX2(vs_vue_prog_data->urb_entry_size, 1);
115
116 /* BRW_NEW_GEOMETRY_PROGRAM, BRW_NEW_GS_PROG_DATA */
117 const bool gs_present = brw->ff_gs.prog_active || brw->geometry_program;
118
119 /* Whe using GS to do transform feedback only we use the same VUE layout for
120 * VS outputs and GS outputs (as it's what the SF and Clipper expect), so we
121 * can simply make the GS URB entry size the same as for the VS. This may
122 * technically be too large in cases where we have few vertex attributes and
123 * a lot of varyings, since the VS size is determined by the larger of the
124 * two. For now, it's safe.
125 *
126 * For user-provided GS the assumption above does not hold since the GS
127 * outputs can be different from the VS outputs.
128 */
129 unsigned gs_size = vs_size;
130 if (brw->geometry_program) {
131 const struct brw_vue_prog_data *gs_vue_prog_data =
132 brw_vue_prog_data(brw->gs.base.prog_data);
133 gs_size = gs_vue_prog_data->urb_entry_size;
134 assert(gs_size >= 1);
135 }
136
137 gen6_upload_urb(brw, vs_size, gs_present, gs_size);
138 }
139
140 const struct brw_tracked_state gen6_urb = {
141 .dirty = {
142 .mesa = 0,
143 .brw = BRW_NEW_BLORP |
144 BRW_NEW_CONTEXT |
145 BRW_NEW_FF_GS_PROG_DATA |
146 BRW_NEW_GEOMETRY_PROGRAM |
147 BRW_NEW_GS_PROG_DATA |
148 BRW_NEW_VS_PROG_DATA,
149 },
150 .emit = upload_urb,
151 };