intel/blorp: Plumb through access to the workaround BO
[mesa.git] / src / intel / vulkan / genX_blorp_exec.c
1 /*
2 * Copyright © 2016 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 <assert.h>
25
26 #include "anv_private.h"
27
28 /* These are defined in anv_private.h and blorp_genX_exec.h */
29 #undef __gen_address_type
30 #undef __gen_user_data
31 #undef __gen_combine_address
32
33 #include "common/gen_l3_config.h"
34 #include "common/gen_sample_positions.h"
35 #include "blorp/blorp_genX_exec.h"
36
37 static void *
38 blorp_emit_dwords(struct blorp_batch *batch, unsigned n)
39 {
40 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
41 return anv_batch_emit_dwords(&cmd_buffer->batch, n);
42 }
43
44 static uint64_t
45 blorp_emit_reloc(struct blorp_batch *batch,
46 void *location, struct blorp_address address, uint32_t delta)
47 {
48 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
49 assert(cmd_buffer->batch.start <= location &&
50 location < cmd_buffer->batch.end);
51 return anv_batch_emit_reloc(&cmd_buffer->batch, location,
52 address.buffer, address.offset + delta);
53 }
54
55 static void
56 blorp_surface_reloc(struct blorp_batch *batch, uint32_t ss_offset,
57 struct blorp_address address, uint32_t delta)
58 {
59 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
60 VkResult result =
61 anv_reloc_list_add(&cmd_buffer->surface_relocs, &cmd_buffer->pool->alloc,
62 ss_offset, address.buffer, address.offset + delta);
63 if (result != VK_SUCCESS)
64 anv_batch_set_error(&cmd_buffer->batch, result);
65 }
66
67 static void *
68 blorp_alloc_dynamic_state(struct blorp_batch *batch,
69 uint32_t size,
70 uint32_t alignment,
71 uint32_t *offset)
72 {
73 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
74
75 struct anv_state state =
76 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, size, alignment);
77
78 *offset = state.offset;
79 return state.map;
80 }
81
82 static void
83 blorp_alloc_binding_table(struct blorp_batch *batch, unsigned num_entries,
84 unsigned state_size, unsigned state_alignment,
85 uint32_t *bt_offset,
86 uint32_t *surface_offsets, void **surface_maps)
87 {
88 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
89
90 uint32_t state_offset;
91 struct anv_state bt_state;
92
93 VkResult result =
94 anv_cmd_buffer_alloc_blorp_binding_table(cmd_buffer, num_entries,
95 &state_offset, &bt_state);
96 if (result != VK_SUCCESS)
97 return;
98
99 uint32_t *bt_map = bt_state.map;
100 *bt_offset = bt_state.offset;
101
102 for (unsigned i = 0; i < num_entries; i++) {
103 struct anv_state surface_state =
104 anv_cmd_buffer_alloc_surface_state(cmd_buffer);
105 bt_map[i] = surface_state.offset + state_offset;
106 surface_offsets[i] = surface_state.offset;
107 surface_maps[i] = surface_state.map;
108 }
109
110 anv_state_flush(cmd_buffer->device, bt_state);
111 }
112
113 static void *
114 blorp_alloc_vertex_buffer(struct blorp_batch *batch, uint32_t size,
115 struct blorp_address *addr)
116 {
117 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
118
119 /* From the Skylake PRM, 3DSTATE_VERTEX_BUFFERS:
120 *
121 * "The VF cache needs to be invalidated before binding and then using
122 * Vertex Buffers that overlap with any previously bound Vertex Buffer
123 * (at a 64B granularity) since the last invalidation. A VF cache
124 * invalidate is performed by setting the "VF Cache Invalidation Enable"
125 * bit in PIPE_CONTROL."
126 *
127 * This restriction first appears in the Skylake PRM but the internal docs
128 * also list it as being an issue on Broadwell. In order to avoid this
129 * problem, we align all vertex buffer allocations to 64 bytes.
130 */
131 struct anv_state vb_state =
132 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, size, 64);
133
134 *addr = (struct blorp_address) {
135 .buffer = &cmd_buffer->device->dynamic_state_pool.block_pool.bo,
136 .offset = vb_state.offset,
137 };
138
139 return vb_state.map;
140 }
141
142 #if GEN_GEN >= 8
143 static struct blorp_address
144 blorp_get_workaround_page(struct blorp_batch *batch)
145 {
146 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
147
148 return (struct blorp_address) {
149 .buffer = &cmd_buffer->device->workaround_bo,
150 };
151 }
152 #endif
153
154 static void
155 blorp_flush_range(struct blorp_batch *batch, void *start, size_t size)
156 {
157 struct anv_device *device = batch->blorp->driver_ctx;
158 if (!device->info.has_llc)
159 anv_flush_range(start, size);
160 }
161
162 static void
163 blorp_emit_urb_config(struct blorp_batch *batch,
164 unsigned vs_entry_size, unsigned sf_entry_size)
165 {
166 struct anv_device *device = batch->blorp->driver_ctx;
167 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
168
169 assert(sf_entry_size == 0);
170
171 const unsigned entry_size[4] = { vs_entry_size, 1, 1, 1 };
172
173 genX(emit_urb_setup)(device, &cmd_buffer->batch,
174 cmd_buffer->state.current_l3_config,
175 VK_SHADER_STAGE_VERTEX_BIT |
176 VK_SHADER_STAGE_FRAGMENT_BIT,
177 entry_size);
178 }
179
180 void
181 genX(blorp_exec)(struct blorp_batch *batch,
182 const struct blorp_params *params)
183 {
184 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
185
186 if (!cmd_buffer->state.current_l3_config) {
187 const struct gen_l3_config *cfg =
188 gen_get_default_l3_config(&cmd_buffer->device->info);
189 genX(cmd_buffer_config_l3)(cmd_buffer, cfg);
190 }
191
192 genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);
193
194 genX(flush_pipeline_select_3d)(cmd_buffer);
195
196 genX(cmd_buffer_emit_gen7_depth_flush)(cmd_buffer);
197
198 /* BLORP doesn't do anything fancy with depth such as discards, so we want
199 * the PMA fix off. Also, off is always the safe option.
200 */
201 genX(cmd_buffer_enable_pma_fix)(cmd_buffer, false);
202
203 /* Disable VF statistics */
204 blorp_emit(batch, GENX(3DSTATE_VF_STATISTICS), vf) {
205 vf.StatisticsEnable = false;
206 }
207
208 blorp_exec(batch, params);
209
210 cmd_buffer->state.vb_dirty = ~0;
211 cmd_buffer->state.dirty = ~0;
212 cmd_buffer->state.push_constants_dirty = ~0;
213 }