1 /*
2 * Copyright © 2014-2017 Broadcom
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 <xf86drm.h>
25 #include <err.h>
26
27 #include "pipe/p_defines.h"
28 #include "util/hash_table.h"
29 #include "util/ralloc.h"
30 #include "util/u_inlines.h"
31 #include "util/u_memory.h"
32 #include "util/u_blitter.h"
33 #include "util/u_upload_mgr.h"
34 #include "util/u_prim.h"
35 #include "pipe/p_screen.h"
36
37 #include "v3d_screen.h"
38 #include "v3d_context.h"
39 #include "v3d_resource.h"
40 #include "broadcom/compiler/v3d_compiler.h"
41
42 void
v3d_flush(struct pipe_context * pctx)43 v3d_flush(struct pipe_context *pctx)
44 {
45 struct v3d_context *v3d = v3d_context(pctx);
46
47 hash_table_foreach(v3d->jobs, entry) {
48 struct v3d_job *job = entry->data;
49 v3d_job_submit(v3d, job);
50 }
51 }
52
53 static void
v3d_pipe_flush(struct pipe_context * pctx,struct pipe_fence_handle ** fence,unsigned flags)54 v3d_pipe_flush(struct pipe_context *pctx, struct pipe_fence_handle **fence,
55 unsigned flags)
56 {
57 struct v3d_context *v3d = v3d_context(pctx);
58
59 v3d_flush(pctx);
60
61 if (fence) {
62 struct pipe_screen *screen = pctx->screen;
63 struct v3d_fence *f = v3d_fence_create(v3d);
64 screen->fence_reference(screen, fence, NULL);
65 *fence = (struct pipe_fence_handle *)f;
66 }
67 }
68
69 static void
v3d_memory_barrier(struct pipe_context * pctx,unsigned int flags)70 v3d_memory_barrier(struct pipe_context *pctx, unsigned int flags)
71 {
72 struct v3d_context *v3d = v3d_context(pctx);
73
74 /* We only need to flush for SSBOs and images, because for everything
75 * else we flush the job automatically when we needed.
76 */
77 const unsigned int flush_flags = PIPE_BARRIER_SHADER_BUFFER |
78 PIPE_BARRIER_IMAGE;
79
80 if (!(flags & flush_flags))
81 return;
82
83 /* We only need to flush jobs writing to SSBOs/images. */
84 perf_debug("Flushing all jobs for glMemoryBarrier(), could do better");
85 v3d_flush(pctx);
86 }
87
88 static void
v3d_set_debug_callback(struct pipe_context * pctx,const struct pipe_debug_callback * cb)89 v3d_set_debug_callback(struct pipe_context *pctx,
90 const struct pipe_debug_callback *cb)
91 {
92 struct v3d_context *v3d = v3d_context(pctx);
93
94 if (cb)
95 v3d->debug = *cb;
96 else
97 memset(&v3d->debug, 0, sizeof(v3d->debug));
98 }
99
100 static void
v3d_invalidate_resource(struct pipe_context * pctx,struct pipe_resource * prsc)101 v3d_invalidate_resource(struct pipe_context *pctx, struct pipe_resource *prsc)
102 {
103 struct v3d_context *v3d = v3d_context(pctx);
104 struct v3d_resource *rsc = v3d_resource(prsc);
105
106 rsc->initialized_buffers = 0;
107
108 struct hash_entry *entry = _mesa_hash_table_search(v3d->write_jobs,
109 prsc);
110 if (!entry)
111 return;
112
113 struct v3d_job *job = entry->data;
114 if (job->key.zsbuf && job->key.zsbuf->texture == prsc)
115 job->store &= ~(PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL);
116 }
117
118 /**
119 * Flushes the current job to get up-to-date primitive counts written to the
120 * primitive counts BO, then accumulates the transform feedback primitive count
121 * in the context and the corresponding vertex counts in the bound stream
122 * output targets.
123 */
124 void
v3d_update_primitive_counters(struct v3d_context * v3d)125 v3d_update_primitive_counters(struct v3d_context *v3d)
126 {
127 struct v3d_job *job = v3d_get_job_for_fbo(v3d);
128 if (job->draw_calls_queued == 0)
129 return;
130
131 /* In order to get up-to-date primitive counts we need to submit
132 * the job for execution so we get the counts written to memory.
133 * Notice that this will require a sync wait for the buffer write.
134 */
135 uint32_t prims_before = v3d->tf_prims_generated;
136 v3d_job_submit(v3d, job);
137 uint32_t prims_after = v3d->tf_prims_generated;
138 if (prims_before == prims_after)
139 return;
140
141 enum pipe_prim_type prim_type = u_base_prim_type(v3d->prim_mode);
142 uint32_t num_verts = u_vertices_for_prims(prim_type,
143 prims_after - prims_before);
144 for (int i = 0; i < v3d->streamout.num_targets; i++) {
145 struct v3d_stream_output_target *so =
146 v3d_stream_output_target(v3d->streamout.targets[i]);
147 so->recorded_vertex_count += num_verts;
148 }
149 }
150
151 bool
v3d_line_smoothing_enabled(struct v3d_context * v3d)152 v3d_line_smoothing_enabled(struct v3d_context *v3d)
153 {
154 if (!v3d->rasterizer->base.line_smooth)
155 return false;
156
157 /* According to the OpenGL docs, line smoothing shouldn’t be applied
158 * when multisampling
159 */
160 if (v3d->job->msaa || v3d->rasterizer->base.multisample)
161 return false;
162
163 if (v3d->framebuffer.nr_cbufs <= 0)
164 return false;
165
166 struct pipe_surface *cbuf = v3d->framebuffer.cbufs[0];
167 if (!cbuf)
168 return false;
169
170 /* Modifying the alpha for pure integer formats probably
171 * doesn’t make sense because we don’t know how the application
172 * uses the alpha value.
173 */
174 if (util_format_is_pure_integer(cbuf->format))
175 return false;
176
177 return true;
178 }
179
180 float
v3d_get_real_line_width(struct v3d_context * v3d)181 v3d_get_real_line_width(struct v3d_context *v3d)
182 {
183 float width = v3d->rasterizer->base.line_width;
184
185 if (v3d_line_smoothing_enabled(v3d)) {
186 /* If line smoothing is enabled then we want to add some extra
187 * pixels to the width in order to have some semi-transparent
188 * edges.
189 */
190 width = floorf(M_SQRT2 * width) + 3;
191 }
192
193 return width;
194 }
195
196 void
v3d_ensure_prim_counts_allocated(struct v3d_context * ctx)197 v3d_ensure_prim_counts_allocated(struct v3d_context *ctx)
198 {
199 if (ctx->prim_counts)
200 return;
201
202 /* Init all 7 counters and 1 padding to 0 */
203 uint32_t zeroes[8] = { 0 };
204 u_upload_data(ctx->uploader,
205 0, sizeof(zeroes), 32, zeroes,
206 &ctx->prim_counts_offset,
207 &ctx->prim_counts);
208 }
209
210 void
v3d_flag_dirty_sampler_state(struct v3d_context * v3d,enum pipe_shader_type shader)211 v3d_flag_dirty_sampler_state(struct v3d_context *v3d,
212 enum pipe_shader_type shader)
213 {
214 switch (shader) {
215 case PIPE_SHADER_VERTEX:
216 v3d->dirty |= V3D_DIRTY_VERTTEX;
217 break;
218 case PIPE_SHADER_GEOMETRY:
219 v3d->dirty |= V3D_DIRTY_GEOMTEX;
220 break;
221 case PIPE_SHADER_FRAGMENT:
222 v3d->dirty |= V3D_DIRTY_FRAGTEX;
223 break;
224 case PIPE_SHADER_COMPUTE:
225 v3d->dirty |= V3D_DIRTY_COMPTEX;
226 break;
227 default:
228 unreachable("Unsupported shader stage");
229 }
230 }
231
232 void
v3d_create_texture_shader_state_bo(struct v3d_context * v3d,struct v3d_sampler_view * so)233 v3d_create_texture_shader_state_bo(struct v3d_context *v3d,
234 struct v3d_sampler_view *so)
235 {
236 if (v3d->screen->devinfo.ver >= 41)
237 v3d41_create_texture_shader_state_bo(v3d, so);
238 else
239 v3d33_create_texture_shader_state_bo(v3d, so);
240 }
241
242 void
v3d_get_tile_buffer_size(bool is_msaa,uint32_t nr_cbufs,struct pipe_surface ** cbufs,struct pipe_surface * bbuf,uint32_t * tile_width,uint32_t * tile_height,uint32_t * max_bpp)243 v3d_get_tile_buffer_size(bool is_msaa,
244 uint32_t nr_cbufs,
245 struct pipe_surface **cbufs,
246 struct pipe_surface *bbuf,
247 uint32_t *tile_width,
248 uint32_t *tile_height,
249 uint32_t *max_bpp)
250 {
251 static const uint8_t tile_sizes[] = {
252 64, 64,
253 64, 32,
254 32, 32,
255 32, 16,
256 16, 16,
257 };
258 int tile_size_index = 0;
259 if (is_msaa)
260 tile_size_index += 2;
261
262 if (cbufs[3] || cbufs[2])
263 tile_size_index += 2;
264 else if (cbufs[1])
265 tile_size_index++;
266
267 *max_bpp = 0;
268 for (int i = 0; i < nr_cbufs; i++) {
269 if (cbufs[i]) {
270 struct v3d_surface *surf = v3d_surface(cbufs[i]);
271 *max_bpp = MAX2(*max_bpp, surf->internal_bpp);
272 }
273 }
274
275 if (bbuf) {
276 struct v3d_surface *bsurf = v3d_surface(bbuf);
277 assert(bbuf->texture->nr_samples <= 1 || is_msaa);
278 *max_bpp = MAX2(*max_bpp, bsurf->internal_bpp);
279 }
280
281 tile_size_index += *max_bpp;
282
283 assert(tile_size_index < ARRAY_SIZE(tile_sizes));
284 *tile_width = tile_sizes[tile_size_index * 2 + 0];
285 *tile_height = tile_sizes[tile_size_index * 2 + 1];
286 }
287
288 static void
v3d_context_destroy(struct pipe_context * pctx)289 v3d_context_destroy(struct pipe_context *pctx)
290 {
291 struct v3d_context *v3d = v3d_context(pctx);
292
293 v3d_flush(pctx);
294
295 if (v3d->blitter)
296 util_blitter_destroy(v3d->blitter);
297
298 if (v3d->uploader)
299 u_upload_destroy(v3d->uploader);
300 if (v3d->state_uploader)
301 u_upload_destroy(v3d->state_uploader);
302
303 if (v3d->prim_counts)
304 pipe_resource_reference(&v3d->prim_counts, NULL);
305
306 slab_destroy_child(&v3d->transfer_pool);
307
308 pipe_surface_reference(&v3d->framebuffer.cbufs[0], NULL);
309 pipe_surface_reference(&v3d->framebuffer.zsbuf, NULL);
310
311 if (v3d->sand8_blit_vs)
312 pctx->delete_vs_state(pctx, v3d->sand8_blit_vs);
313 if (v3d->sand8_blit_fs_luma)
314 pctx->delete_fs_state(pctx, v3d->sand8_blit_fs_luma);
315 if (v3d->sand8_blit_fs_chroma)
316 pctx->delete_fs_state(pctx, v3d->sand8_blit_fs_chroma);
317
318 v3d_program_fini(pctx);
319
320 ralloc_free(v3d);
321 }
322
323 static void
v3d_get_sample_position(struct pipe_context * pctx,unsigned sample_count,unsigned sample_index,float * xy)324 v3d_get_sample_position(struct pipe_context *pctx,
325 unsigned sample_count, unsigned sample_index,
326 float *xy)
327 {
328 struct v3d_context *v3d = v3d_context(pctx);
329
330 if (sample_count <= 1) {
331 xy[0] = 0.5;
332 xy[1] = 0.5;
333 } else {
334 static const int xoffsets_v33[] = { 1, -3, 3, -1 };
335 static const int xoffsets_v42[] = { -1, 3, -3, 1 };
336 const int *xoffsets = (v3d->screen->devinfo.ver >= 42 ?
337 xoffsets_v42 : xoffsets_v33);
338
339 xy[0] = 0.5 + xoffsets[sample_index] * .125;
340 xy[1] = .125 + sample_index * .25;
341 }
342 }
343
344 struct pipe_context *
v3d_context_create(struct pipe_screen * pscreen,void * priv,unsigned flags)345 v3d_context_create(struct pipe_screen *pscreen, void *priv, unsigned flags)
346 {
347 struct v3d_screen *screen = v3d_screen(pscreen);
348 struct v3d_context *v3d;
349
350 /* Prevent dumping of the shaders built during context setup. */
351 uint32_t saved_shaderdb_flag = V3D_DEBUG & V3D_DEBUG_SHADERDB;
352 V3D_DEBUG &= ~V3D_DEBUG_SHADERDB;
353
354 v3d = rzalloc(NULL, struct v3d_context);
355 if (!v3d)
356 return NULL;
357 struct pipe_context *pctx = &v3d->base;
358
359 v3d->screen = screen;
360
361 int ret = drmSyncobjCreate(screen->fd, DRM_SYNCOBJ_CREATE_SIGNALED,
362 &v3d->out_sync);
363 if (ret) {
364 ralloc_free(v3d);
365 return NULL;
366 }
367
368 pctx->screen = pscreen;
369 pctx->priv = priv;
370 pctx->destroy = v3d_context_destroy;
371 pctx->flush = v3d_pipe_flush;
372 pctx->memory_barrier = v3d_memory_barrier;
373 pctx->set_debug_callback = v3d_set_debug_callback;
374 pctx->invalidate_resource = v3d_invalidate_resource;
375 pctx->get_sample_position = v3d_get_sample_position;
376
377 if (screen->devinfo.ver >= 41) {
378 v3d41_draw_init(pctx);
379 v3d41_state_init(pctx);
380 } else {
381 v3d33_draw_init(pctx);
382 v3d33_state_init(pctx);
383 }
384 v3d_program_init(pctx);
385 v3d_query_init(pctx);
386 v3d_resource_context_init(pctx);
387
388 v3d_job_init(v3d);
389
390 v3d->fd = screen->fd;
391
392 slab_create_child(&v3d->transfer_pool, &screen->transfer_pool);
393
394 v3d->uploader = u_upload_create_default(&v3d->base);
395 v3d->base.stream_uploader = v3d->uploader;
396 v3d->base.const_uploader = v3d->uploader;
397 v3d->state_uploader = u_upload_create(&v3d->base,
398 4096,
399 PIPE_BIND_CONSTANT_BUFFER,
400 PIPE_USAGE_STREAM, 0);
401
402 v3d->blitter = util_blitter_create(pctx);
403 if (!v3d->blitter)
404 goto fail;
405 v3d->blitter->use_index_buffer = true;
406
407 V3D_DEBUG |= saved_shaderdb_flag;
408
409 v3d->sample_mask = (1 << V3D_MAX_SAMPLES) - 1;
410 v3d->active_queries = true;
411
412 return &v3d->base;
413
414 fail:
415 pctx->destroy(pctx);
416 return NULL;
417 }
418