src/gallium/drivers/v3d/v3d_context.c

   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 "indices/u_primconvert.h"
  36 #include "pipe/p_screen.h"
  37
  38 #include "v3d_screen.h"
  39 #include "v3d_context.h"
  40 #include "v3d_resource.h"
  41 #include "broadcom/compiler/v3d_compiler.h"
  42
  43 void
  44 v3d_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
  54 static void
  55 v3d_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
  70 static void
  71 v3d_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
  89 static void
  90 v3d_set_debug_callback(struct pipe_context *pctx,
  91                        const struct pipe_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
 101 static void
 102 v3d_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 primive 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  */
 125 void
 126 v3d_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
 152 bool
 153 v3d_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
 181 float
 182 v3d_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
 197 static void
 198 v3d_context_destroy(struct pipe_context *pctx)
 199 {
 200         struct v3d_context *v3d = v3d_context(pctx);
 201
 202         v3d_flush(pctx);
 203
 204         if (v3d->blitter)
 205                 util_blitter_destroy(v3d->blitter);
 206
 207         if (v3d->primconvert)
 208                 util_primconvert_destroy(v3d->primconvert);
 209
 210         if (v3d->uploader)
 211                 u_upload_destroy(v3d->uploader);
 212         if (v3d->state_uploader)
 213                 u_upload_destroy(v3d->state_uploader);
 214
 215         if (v3d->prim_counts)
 216                 pipe_resource_reference(&v3d->prim_counts, NULL);
 217
 218         slab_destroy_child(&v3d->transfer_pool);
 219
 220         pipe_surface_reference(&v3d->framebuffer.cbufs[0], NULL);
 221         pipe_surface_reference(&v3d->framebuffer.zsbuf, NULL);
 222
 223         v3d_program_fini(pctx);
 224
 225         ralloc_free(v3d);
 226 }
 227
 228 static void
 229 v3d_get_sample_position(struct pipe_context *pctx,
 230                         unsigned sample_count, unsigned sample_index,
 231                         float *xy)
 232 {
 233         struct v3d_context *v3d = v3d_context(pctx);
 234
 235         if (sample_count <= 1) {
 236                 xy[0] = 0.5;
 237                 xy[1] = 0.5;
 238         } else {
 239                 static const int xoffsets_v33[] = { 1, -3, 3, -1 };
 240                 static const int xoffsets_v42[] = { -1, 3, -3, 1 };
 241                 const int *xoffsets = (v3d->screen->devinfo.ver >= 42 ?
 242                                        xoffsets_v42 : xoffsets_v33);
 243
 244                 xy[0] = 0.5 + xoffsets[sample_index] * .125;
 245                 xy[1] = .125 + sample_index * .25;
 246         }
 247 }
 248
 249 struct pipe_context *
 250 v3d_context_create(struct pipe_screen *pscreen, void *priv, unsigned flags)
 251 {
 252         struct v3d_screen *screen = v3d_screen(pscreen);
 253         struct v3d_context *v3d;
 254
 255         /* Prevent dumping of the shaders built during context setup. */
 256         uint32_t saved_shaderdb_flag = V3D_DEBUG & V3D_DEBUG_SHADERDB;
 257         V3D_DEBUG &= ~V3D_DEBUG_SHADERDB;
 258
 259         v3d = rzalloc(NULL, struct v3d_context);
 260         if (!v3d)
 261                 return NULL;
 262         struct pipe_context *pctx = &v3d->base;
 263
 264         v3d->screen = screen;
 265
 266         int ret = drmSyncobjCreate(screen->fd, DRM_SYNCOBJ_CREATE_SIGNALED,
 267                                    &v3d->out_sync);
 268         if (ret) {
 269                 ralloc_free(v3d);
 270                 return NULL;
 271         }
 272
 273         pctx->screen = pscreen;
 274         pctx->priv = priv;
 275         pctx->destroy = v3d_context_destroy;
 276         pctx->flush = v3d_pipe_flush;
 277         pctx->memory_barrier = v3d_memory_barrier;
 278         pctx->set_debug_callback = v3d_set_debug_callback;
 279         pctx->invalidate_resource = v3d_invalidate_resource;
 280         pctx->get_sample_position = v3d_get_sample_position;
 281
 282         if (screen->devinfo.ver >= 41) {
 283                 v3d41_draw_init(pctx);
 284                 v3d41_state_init(pctx);
 285         } else {
 286                 v3d33_draw_init(pctx);
 287                 v3d33_state_init(pctx);
 288         }
 289         v3d_program_init(pctx);
 290         v3d_query_init(pctx);
 291         v3d_resource_context_init(pctx);
 292
 293         v3d_job_init(v3d);
 294
 295         v3d->fd = screen->fd;
 296
 297         slab_create_child(&v3d->transfer_pool, &screen->transfer_pool);
 298
 299         v3d->uploader = u_upload_create_default(&v3d->base);
 300         v3d->base.stream_uploader = v3d->uploader;
 301         v3d->base.const_uploader = v3d->uploader;
 302         v3d->state_uploader = u_upload_create(&v3d->base,
 303                                               4096,
 304                                               PIPE_BIND_CONSTANT_BUFFER,
 305                                               PIPE_USAGE_STREAM, 0);
 306
 307         v3d->blitter = util_blitter_create(pctx);
 308         if (!v3d->blitter)
 309                 goto fail;
 310         v3d->blitter->use_index_buffer = true;
 311
 312         v3d->primconvert = util_primconvert_create(pctx,
 313                                                    (1 << PIPE_PRIM_QUADS) - 1);
 314         if (!v3d->primconvert)
 315                 goto fail;
 316
 317         V3D_DEBUG |= saved_shaderdb_flag;
 318
 319         v3d->sample_mask = (1 << V3D_MAX_SAMPLES) - 1;
 320         v3d->active_queries = true;
 321
 322         return &v3d->base;
 323
 324 fail:
 325         pctx->destroy(pctx);
 326         return NULL;
 327 }