1/*
2 * Copyright © 2011 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
24#include "main/macros.h"
25#include "intel_batchbuffer.h"
26#include "brw_context.h"
27#include "brw_state.h"
28#include "brw_defines.h"
29
30#include "common/gen_l3_config.h"
31
32/**
33 * The following diagram shows how we partition the URB:
34 *
35 *        16kB or 32kB               Rest of the URB space
36 *   __________-__________   _________________-_________________
37 *  /                     \ /                                   \
38 * +-------------------------------------------------------------+
39 * |  VS/HS/DS/GS/FS Push  |           VS/HS/DS/GS URB           |
40 * |       Constants       |               Entries               |
41 * +-------------------------------------------------------------+
42 *
43 * Notably, push constants must be stored at the beginning of the URB
44 * space, while entries can be stored anywhere.  Ivybridge and Haswell
45 * GT1/GT2 have a maximum constant buffer size of 16kB, while Haswell GT3
46 * doubles this (32kB).
47 *
48 * Ivybridge and Haswell GT1/GT2 allow push constants to be located (and
49 * sized) in increments of 1kB.  Haswell GT3 requires them to be located and
50 * sized in increments of 2kB.
51 *
52 * Currently we split the constant buffer space evenly among whatever stages
53 * are active.  This is probably not ideal, but simple.
54 *
55 * Ivybridge GT1 and Haswell GT1 have 128kB of URB space.
56 * Ivybridge GT2 and Haswell GT2 have 256kB of URB space.
57 * Haswell GT3 has 512kB of URB space.
58 *
59 * See "Volume 2a: 3D Pipeline," section 1.8, "Volume 1b: Configurations",
60 * and the documentation for 3DSTATE_PUSH_CONSTANT_ALLOC_xS.
61 */
62static void
63gen7_allocate_push_constants(struct brw_context *brw)
64{
65   /* BRW_NEW_GEOMETRY_PROGRAM */
66   bool gs_present = brw->geometry_program;
67
68   /* BRW_NEW_TESS_PROGRAMS */
69   bool tess_present = brw->tess_eval_program;
70
71   unsigned avail_size = 16;
72   unsigned multiplier =
73      (brw->gen >= 8 || (brw->is_haswell && brw->gt == 3)) ? 2 : 1;
74
75   int stages = 2 + gs_present + 2 * tess_present;
76
77   /* Divide up the available space equally between stages.  Because we
78    * round down (using floor division), there may be some left over
79    * space.  We allocate that to the pixel shader stage.
80    */
81   unsigned size_per_stage = avail_size / stages;
82
83   unsigned vs_size = size_per_stage;
84   unsigned hs_size = tess_present ? size_per_stage : 0;
85   unsigned ds_size = tess_present ? size_per_stage : 0;
86   unsigned gs_size = gs_present ? size_per_stage : 0;
87   unsigned fs_size = avail_size - size_per_stage * (stages - 1);
88
89   gen7_emit_push_constant_state(brw, multiplier * vs_size,
90                                 multiplier * hs_size, multiplier * ds_size,
91                                 multiplier * gs_size, multiplier * fs_size);
92
93   /* From p115 of the Ivy Bridge PRM (3.2.1.4 3DSTATE_PUSH_CONSTANT_ALLOC_VS):
94    *
95    *     Programming Restriction:
96    *
97    *     The 3DSTATE_CONSTANT_VS must be reprogrammed prior to the next
98    *     3DPRIMITIVE command after programming the
99    *     3DSTATE_PUSH_CONSTANT_ALLOC_VS.
100    *
101    * Similar text exists for the other 3DSTATE_PUSH_CONSTANT_ALLOC_*
102    * commands.
103    */
104   brw->ctx.NewDriverState |= BRW_NEW_PUSH_CONSTANT_ALLOCATION;
105}
106
107void
108gen7_emit_push_constant_state(struct brw_context *brw, unsigned vs_size,
109                              unsigned hs_size, unsigned ds_size,
110                              unsigned gs_size, unsigned fs_size)
111{
112   unsigned offset = 0;
113
114   BEGIN_BATCH(10);
115   OUT_BATCH(_3DSTATE_PUSH_CONSTANT_ALLOC_VS << 16 | (2 - 2));
116   OUT_BATCH(vs_size | offset << GEN7_PUSH_CONSTANT_BUFFER_OFFSET_SHIFT);
117   offset += vs_size;
118
119   OUT_BATCH(_3DSTATE_PUSH_CONSTANT_ALLOC_HS << 16 | (2 - 2));
120   OUT_BATCH(hs_size | offset << GEN7_PUSH_CONSTANT_BUFFER_OFFSET_SHIFT);
121   offset += hs_size;
122
123   OUT_BATCH(_3DSTATE_PUSH_CONSTANT_ALLOC_DS << 16 | (2 - 2));
124   OUT_BATCH(ds_size | offset << GEN7_PUSH_CONSTANT_BUFFER_OFFSET_SHIFT);
125   offset += ds_size;
126
127   OUT_BATCH(_3DSTATE_PUSH_CONSTANT_ALLOC_GS << 16 | (2 - 2));
128   OUT_BATCH(gs_size | offset << GEN7_PUSH_CONSTANT_BUFFER_OFFSET_SHIFT);
129   offset += gs_size;
130
131   OUT_BATCH(_3DSTATE_PUSH_CONSTANT_ALLOC_PS << 16 | (2 - 2));
132   OUT_BATCH(fs_size | offset << GEN7_PUSH_CONSTANT_BUFFER_OFFSET_SHIFT);
133   ADVANCE_BATCH();
134
135   /* From p292 of the Ivy Bridge PRM (11.2.4 3DSTATE_PUSH_CONSTANT_ALLOC_PS):
136    *
137    *     A PIPE_CONTROL command with the CS Stall bit set must be programmed
138    *     in the ring after this instruction.
139    *
140    * No such restriction exists for Haswell or Baytrail.
141    */
142   if (brw->gen < 8 && !brw->is_haswell && !brw->is_baytrail)
143      gen7_emit_cs_stall_flush(brw);
144}
145
146const struct brw_tracked_state gen7_push_constant_space = {
147   .dirty = {
148      .mesa = 0,
149      .brw = BRW_NEW_CONTEXT |
150             BRW_NEW_GEOMETRY_PROGRAM |
151             BRW_NEW_TESS_PROGRAMS,
152   },
153   .emit = gen7_allocate_push_constants,
154};
155
156static void
157upload_urb(struct brw_context *brw)
158{
159   /* BRW_NEW_VS_PROG_DATA */
160   const struct brw_vue_prog_data *vs_vue_prog_data =
161      brw_vue_prog_data(brw->vs.base.prog_data);
162   const unsigned vs_size = MAX2(vs_vue_prog_data->urb_entry_size, 1);
163   /* BRW_NEW_GS_PROG_DATA */
164   const bool gs_present = brw->gs.base.prog_data;
165   /* BRW_NEW_TES_PROG_DATA */
166   const bool tess_present = brw->tes.base.prog_data;
167
168   gen7_upload_urb(brw, vs_size, gs_present, tess_present);
169}
170
171void
172gen7_upload_urb(struct brw_context *brw, unsigned vs_size,
173                bool gs_present, bool tess_present)
174{
175   const struct gen_device_info *devinfo = &brw->screen->devinfo;
176   const int push_size_kB =
177      (brw->gen >= 8 || (brw->is_haswell && brw->gt == 3)) ? 32 : 16;
178
179   /* BRW_NEW_{VS,TCS,TES,GS}_PROG_DATA */
180   struct brw_vue_prog_data *prog_data[4] = {
181      [MESA_SHADER_VERTEX] =
182         brw_vue_prog_data(brw->vs.base.prog_data),
183      [MESA_SHADER_TESS_CTRL] =
184         tess_present ? brw_vue_prog_data(brw->tcs.base.prog_data) : NULL,
185      [MESA_SHADER_TESS_EVAL] =
186         tess_present ? brw_vue_prog_data(brw->tes.base.prog_data) : NULL,
187      [MESA_SHADER_GEOMETRY] =
188         gs_present ? brw_vue_prog_data(brw->gs.base.prog_data) : NULL,
189   };
190
191   unsigned entry_size[4];
192   entry_size[MESA_SHADER_VERTEX] = vs_size;
193   for (int i = MESA_SHADER_TESS_CTRL; i <= MESA_SHADER_GEOMETRY; i++) {
194      entry_size[i] = prog_data[i] ? prog_data[i]->urb_entry_size : 1;
195   }
196
197   /* If we're just switching between programs with the same URB requirements,
198    * skip the rest of the logic.
199    */
200   if (!(brw->ctx.NewDriverState & BRW_NEW_CONTEXT) &&
201       !(brw->ctx.NewDriverState & BRW_NEW_URB_SIZE) &&
202       brw->urb.vsize == entry_size[MESA_SHADER_VERTEX] &&
203       brw->urb.gs_present == gs_present &&
204       brw->urb.gsize == entry_size[MESA_SHADER_GEOMETRY] &&
205       brw->urb.tess_present == tess_present &&
206       brw->urb.hsize == entry_size[MESA_SHADER_TESS_CTRL] &&
207       brw->urb.dsize == entry_size[MESA_SHADER_TESS_EVAL]) {
208      return;
209   }
210   brw->urb.vsize = entry_size[MESA_SHADER_VERTEX];
211   brw->urb.gs_present = gs_present;
212   brw->urb.gsize = entry_size[MESA_SHADER_GEOMETRY];
213   brw->urb.tess_present = tess_present;
214   brw->urb.hsize = entry_size[MESA_SHADER_TESS_CTRL];
215   brw->urb.dsize = entry_size[MESA_SHADER_TESS_EVAL];
216
217   unsigned entries[4];
218   unsigned start[4];
219   gen_get_urb_config(devinfo, 1024 * push_size_kB, 1024 * brw->urb.size,
220                      tess_present, gs_present, entry_size, entries, start);
221
222   if (brw->gen == 7 && !brw->is_haswell && !brw->is_baytrail)
223      gen7_emit_vs_workaround_flush(brw);
224
225   BEGIN_BATCH(8);
226   for (int i = MESA_SHADER_VERTEX; i <= MESA_SHADER_GEOMETRY; i++) {
227      OUT_BATCH((_3DSTATE_URB_VS + i) << 16 | (2 - 2));
228      OUT_BATCH(entries[i] |
229                ((entry_size[i] - 1) << GEN7_URB_ENTRY_SIZE_SHIFT) |
230                (start[i] << GEN7_URB_STARTING_ADDRESS_SHIFT));
231   }
232   ADVANCE_BATCH();
233}
234
235const struct brw_tracked_state gen7_urb = {
236   .dirty = {
237      .mesa = 0,
238      .brw = BRW_NEW_CONTEXT |
239             BRW_NEW_URB_SIZE |
240             BRW_NEW_GS_PROG_DATA |
241             BRW_NEW_TCS_PROG_DATA |
242             BRW_NEW_TES_PROG_DATA |
243             BRW_NEW_VS_PROG_DATA,
244   },
245   .emit = upload_urb,
246};
247