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#include "broadcom/common/v3d_util.h"
42
43void
44v3d_flush(struct pipe_context *pctx)
45{
46        struct v3d_context *v3d = v3d_context(pctx);
47
48        hash_table_foreach(v3d->jobs, entry) {
49                struct v3d_job *job = entry->data;
50                v3d_job_submit(v3d, job);
51        }
52}
53
54static void
55v3d_pipe_flush(struct pipe_context *pctx, struct pipe_fence_handle **fence,
56               unsigned flags)
57{
58        struct v3d_context *v3d = v3d_context(pctx);
59
60        v3d_flush(pctx);
61
62        if (fence) {
63                struct pipe_screen *screen = pctx->screen;
64                struct v3d_fence *f = v3d_fence_create(v3d);
65                screen->fence_reference(screen, fence, NULL);
66                *fence = (struct pipe_fence_handle *)f;
67        }
68}
69
70static void
71v3d_memory_barrier(struct pipe_context *pctx, unsigned int flags)
72{
73        struct v3d_context *v3d = v3d_context(pctx);
74
75        /* We only need to flush for SSBOs and images, because for everything
76         * else we flush the job automatically when we needed.
77         */
78        const unsigned int flush_flags = PIPE_BARRIER_SHADER_BUFFER |
79                                         PIPE_BARRIER_IMAGE;
80
81	if (!(flags & flush_flags))
82		return;
83
84        /* We only need to flush jobs writing to SSBOs/images. */
85        perf_debug("Flushing all jobs for glMemoryBarrier(), could do better");
86        v3d_flush(pctx);
87}
88
89static void
90v3d_set_debug_callback(struct pipe_context *pctx,
91                       const struct util_debug_callback *cb)
92{
93        struct v3d_context *v3d = v3d_context(pctx);
94
95        if (cb)
96                v3d->debug = *cb;
97        else
98                memset(&v3d->debug, 0, sizeof(v3d->debug));
99}
100
101static void
102v3d_invalidate_resource(struct pipe_context *pctx, struct pipe_resource *prsc)
103{
104        struct v3d_context *v3d = v3d_context(pctx);
105        struct v3d_resource *rsc = v3d_resource(prsc);
106
107        rsc->initialized_buffers = 0;
108
109        struct hash_entry *entry = _mesa_hash_table_search(v3d->write_jobs,
110                                                           prsc);
111        if (!entry)
112                return;
113
114        struct v3d_job *job = entry->data;
115        if (job->key.zsbuf && job->key.zsbuf->texture == prsc)
116                job->store &= ~(PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL);
117}
118
119/**
120 * Flushes the current job to get up-to-date primitive counts written to the
121 * primitive counts BO, then accumulates the transform feedback primitive count
122 * in the context and the corresponding vertex counts in the bound stream
123 * output targets.
124 */
125void
126v3d_update_primitive_counters(struct v3d_context *v3d)
127{
128        struct v3d_job *job = v3d_get_job_for_fbo(v3d);
129        if (job->draw_calls_queued == 0)
130                return;
131
132        /* In order to get up-to-date primitive counts we need to submit
133         * the job for execution so we get the counts written to memory.
134         * Notice that this will require a sync wait for the buffer write.
135         */
136        uint32_t prims_before = v3d->tf_prims_generated;
137        v3d_job_submit(v3d, job);
138        uint32_t prims_after = v3d->tf_prims_generated;
139        if (prims_before == prims_after)
140                return;
141
142        enum pipe_prim_type prim_type = u_base_prim_type(v3d->prim_mode);
143        uint32_t num_verts = u_vertices_for_prims(prim_type,
144                                                  prims_after - prims_before);
145        for (int i = 0; i < v3d->streamout.num_targets; i++) {
146                struct v3d_stream_output_target *so =
147                        v3d_stream_output_target(v3d->streamout.targets[i]);
148                so->recorded_vertex_count += num_verts;
149        }
150}
151
152bool
153v3d_line_smoothing_enabled(struct v3d_context *v3d)
154{
155        if (!v3d->rasterizer->base.line_smooth)
156                return false;
157
158        /* According to the OpenGL docs, line smoothing shouldn’t be applied
159         * when multisampling
160         */
161        if (v3d->job->msaa || v3d->rasterizer->base.multisample)
162                return false;
163
164        if (v3d->framebuffer.nr_cbufs <= 0)
165                return false;
166
167        struct pipe_surface *cbuf = v3d->framebuffer.cbufs[0];
168        if (!cbuf)
169                return false;
170
171        /* Modifying the alpha for pure integer formats probably
172         * doesn’t make sense because we don’t know how the application
173         * uses the alpha value.
174         */
175        if (util_format_is_pure_integer(cbuf->format))
176                return false;
177
178        return true;
179}
180
181float
182v3d_get_real_line_width(struct v3d_context *v3d)
183{
184        float width = v3d->rasterizer->base.line_width;
185
186        if (v3d_line_smoothing_enabled(v3d)) {
187                /* If line smoothing is enabled then we want to add some extra
188                 * pixels to the width in order to have some semi-transparent
189                 * edges.
190                 */
191                width = floorf(M_SQRT2 * width) + 3;
192        }
193
194        return width;
195}
196
197void
198v3d_ensure_prim_counts_allocated(struct v3d_context *ctx)
199{
200        if (ctx->prim_counts)
201                return;
202
203        /* Init all 7 counters and 1 padding to 0 */
204        uint32_t zeroes[8] = { 0 };
205        u_upload_data(ctx->uploader,
206                      0, sizeof(zeroes), 32, zeroes,
207                      &ctx->prim_counts_offset,
208                      &ctx->prim_counts);
209}
210
211void
212v3d_flag_dirty_sampler_state(struct v3d_context *v3d,
213                             enum pipe_shader_type shader)
214{
215        switch (shader) {
216        case PIPE_SHADER_VERTEX:
217                v3d->dirty |= V3D_DIRTY_VERTTEX;
218                break;
219        case PIPE_SHADER_GEOMETRY:
220                v3d->dirty |= V3D_DIRTY_GEOMTEX;
221                break;
222        case PIPE_SHADER_FRAGMENT:
223                v3d->dirty |= V3D_DIRTY_FRAGTEX;
224                break;
225        case PIPE_SHADER_COMPUTE:
226                v3d->dirty |= V3D_DIRTY_COMPTEX;
227                break;
228        default:
229                unreachable("Unsupported shader stage");
230        }
231}
232
233void
234v3d_create_texture_shader_state_bo(struct v3d_context *v3d,
235                                   struct v3d_sampler_view *so)
236{
237        if (v3d->screen->devinfo.ver >= 41)
238                v3d41_create_texture_shader_state_bo(v3d, so);
239        else
240                v3d33_create_texture_shader_state_bo(v3d, so);
241}
242
243void
244v3d_get_tile_buffer_size(bool is_msaa,
245                         bool double_buffer,
246                         uint32_t nr_cbufs,
247                         struct pipe_surface **cbufs,
248                         struct pipe_surface *bbuf,
249                         uint32_t *tile_width,
250                         uint32_t *tile_height,
251                         uint32_t *max_bpp)
252{
253        assert(!is_msaa || !double_buffer);
254
255        uint32_t max_cbuf_idx = 0;
256        *max_bpp = 0;
257        for (int i = 0; i < nr_cbufs; i++) {
258                if (cbufs[i]) {
259                        struct v3d_surface *surf = v3d_surface(cbufs[i]);
260                        *max_bpp = MAX2(*max_bpp, surf->internal_bpp);
261                        max_cbuf_idx = MAX2(i, max_cbuf_idx);
262                }
263        }
264
265        if (bbuf) {
266                struct v3d_surface *bsurf = v3d_surface(bbuf);
267                assert(bbuf->texture->nr_samples <= 1 || is_msaa);
268                *max_bpp = MAX2(*max_bpp, bsurf->internal_bpp);
269        }
270
271        v3d_choose_tile_size(max_cbuf_idx + 1, *max_bpp,
272                             is_msaa, double_buffer,
273                             tile_width, tile_height);
274}
275
276static void
277v3d_context_destroy(struct pipe_context *pctx)
278{
279        struct v3d_context *v3d = v3d_context(pctx);
280
281        v3d_flush(pctx);
282
283        if (v3d->blitter)
284                util_blitter_destroy(v3d->blitter);
285
286        if (v3d->uploader)
287                u_upload_destroy(v3d->uploader);
288        if (v3d->state_uploader)
289                u_upload_destroy(v3d->state_uploader);
290
291        if (v3d->prim_counts)
292                pipe_resource_reference(&v3d->prim_counts, NULL);
293
294        slab_destroy_child(&v3d->transfer_pool);
295
296        for (int i = 0; i < v3d->framebuffer.nr_cbufs; i++)
297                pipe_surface_reference(&v3d->framebuffer.cbufs[i], NULL);
298
299        pipe_surface_reference(&v3d->framebuffer.zsbuf, NULL);
300
301        if (v3d->sand8_blit_vs)
302                pctx->delete_vs_state(pctx, v3d->sand8_blit_vs);
303        if (v3d->sand8_blit_fs_luma)
304                pctx->delete_fs_state(pctx, v3d->sand8_blit_fs_luma);
305        if (v3d->sand8_blit_fs_chroma)
306                pctx->delete_fs_state(pctx, v3d->sand8_blit_fs_chroma);
307
308        v3d_program_fini(pctx);
309
310        ralloc_free(v3d);
311}
312
313static void
314v3d_get_sample_position(struct pipe_context *pctx,
315                        unsigned sample_count, unsigned sample_index,
316                        float *xy)
317{
318        struct v3d_context *v3d = v3d_context(pctx);
319
320        if (sample_count <= 1) {
321                xy[0] = 0.5;
322                xy[1] = 0.5;
323        } else {
324                static const int xoffsets_v33[] = { 1, -3, 3, -1 };
325                static const int xoffsets_v42[] = { -1, 3, -3, 1 };
326                const int *xoffsets = (v3d->screen->devinfo.ver >= 42 ?
327                                       xoffsets_v42 : xoffsets_v33);
328
329                xy[0] = 0.5 + xoffsets[sample_index] * .125;
330                xy[1] = .125 + sample_index * .25;
331        }
332}
333
334struct pipe_context *
335v3d_context_create(struct pipe_screen *pscreen, void *priv, unsigned flags)
336{
337        struct v3d_screen *screen = v3d_screen(pscreen);
338        struct v3d_context *v3d;
339
340        /* Prevent dumping of the shaders built during context setup. */
341        uint32_t saved_shaderdb_flag = V3D_DEBUG & V3D_DEBUG_SHADERDB;
342        V3D_DEBUG &= ~V3D_DEBUG_SHADERDB;
343
344        v3d = rzalloc(NULL, struct v3d_context);
345        if (!v3d)
346                return NULL;
347        struct pipe_context *pctx = &v3d->base;
348
349        v3d->screen = screen;
350
351        int ret = drmSyncobjCreate(screen->fd, DRM_SYNCOBJ_CREATE_SIGNALED,
352                                   &v3d->out_sync);
353        if (ret) {
354                ralloc_free(v3d);
355                return NULL;
356        }
357
358        pctx->screen = pscreen;
359        pctx->priv = priv;
360        pctx->destroy = v3d_context_destroy;
361        pctx->flush = v3d_pipe_flush;
362        pctx->memory_barrier = v3d_memory_barrier;
363        pctx->set_debug_callback = v3d_set_debug_callback;
364        pctx->invalidate_resource = v3d_invalidate_resource;
365        pctx->get_sample_position = v3d_get_sample_position;
366
367        if (screen->devinfo.ver >= 41) {
368                v3d41_draw_init(pctx);
369                v3d41_state_init(pctx);
370        } else {
371                v3d33_draw_init(pctx);
372                v3d33_state_init(pctx);
373        }
374        v3d_program_init(pctx);
375        v3d_query_init(pctx);
376        v3d_resource_context_init(pctx);
377
378        v3d_job_init(v3d);
379
380        v3d->fd = screen->fd;
381
382        slab_create_child(&v3d->transfer_pool, &screen->transfer_pool);
383
384        v3d->uploader = u_upload_create_default(&v3d->base);
385        v3d->base.stream_uploader = v3d->uploader;
386        v3d->base.const_uploader = v3d->uploader;
387        v3d->state_uploader = u_upload_create(&v3d->base,
388                                              4096,
389                                              PIPE_BIND_CONSTANT_BUFFER,
390                                              PIPE_USAGE_STREAM, 0);
391
392        v3d->blitter = util_blitter_create(pctx);
393        if (!v3d->blitter)
394                goto fail;
395        v3d->blitter->use_index_buffer = true;
396
397        V3D_DEBUG |= saved_shaderdb_flag;
398
399        v3d->sample_mask = (1 << V3D_MAX_SAMPLES) - 1;
400        v3d->active_queries = true;
401
402        return &v3d->base;
403
404fail:
405        pctx->destroy(pctx);
406        return NULL;
407}
408