src/gallium/drivers/panfrost/pan_job.h

   1 /*
   2  * Copyright (C) 2019 Alyssa Rosenzweig
   3  * Copyright (C) 2014-2017 Broadcom
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   9  * and/or sell copies of the Software, and to permit persons to whom the
  10  * Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  *
  24  */
  25
  26 #ifndef __PAN_JOB_H__
  27 #define __PAN_JOB_H__
  28
  29 #include "util/u_dynarray.h"
  30 #include "pipe/p_state.h"
  31 #include "pan_allocate.h"
  32 #include "pan_resource.h"
  33
  34 /* panfrost_batch_fence is the out fence of a batch that users or other batches
  35  * might want to wait on. The batch fence lifetime is different from the batch
  36  * one as want will certainly want to wait upon the fence after the batch has
  37  * been submitted (which is when panfrost_batch objects are freed).
  38  */
  39 struct panfrost_batch_fence {
  40         /* Refcounting object for the fence. */
  41         struct pipe_reference reference;
  42
  43         /* Batch that created this fence object. Will become NULL at batch
  44          * submission time. This field is mainly here to know whether the
  45          * batch has been flushed or not.
  46          */
  47         struct panfrost_batch *batch;
  48
  49         /* Context this fence is attached to. We need both ctx and batch, as
  50          * the batch will go away after it's been submitted, but the fence
  51          * will stay a bit longer.
  52          */
  53         struct panfrost_context *ctx;
  54
  55         /* Sync object backing this fence. */
  56         uint32_t syncobj;
  57
  58         /* Cached value of the signaled state to avoid calling WAIT_SYNCOBJs
  59          * when we know the fence has already been signaled.
  60          */
  61         bool signaled;
  62 };
  63
  64 #define PAN_REQ_MSAA            (1 << 0)
  65 #define PAN_REQ_DEPTH_WRITE     (1 << 1)
  66
  67 /* A panfrost_batch corresponds to a bound FBO we're rendering to,
  68  * collecting over multiple draws. */
  69
  70 struct panfrost_batch {
  71         struct panfrost_context *ctx;
  72         struct pipe_framebuffer_state key;
  73
  74         /* Buffers cleared (PIPE_CLEAR_* bitmask) */
  75         unsigned clear;
  76
  77         /* Packed clear values, indexed by both render target as well as word.
  78          * Essentially, a single pixel is packed, with some padding to bring it
  79          * up to a 32-bit interval; that pixel is then duplicated over to fill
  80          * all 16-bytes */
  81
  82         uint32_t clear_color[PIPE_MAX_COLOR_BUFS][4];
  83         float clear_depth;
  84         unsigned clear_stencil;
  85
  86         /* Amount of thread local storage required per thread */
  87         unsigned stack_size;
  88
  89         /* Whether this job uses the corresponding requirement (PAN_REQ_*
  90          * bitmask) */
  91         unsigned requirements;
  92
  93         /* The bounding box covered by this job, taking scissors into account.
  94          * Basically, the bounding box we have to run fragment shaders for */
  95
  96         unsigned minx, miny;
  97         unsigned maxx, maxy;
  98
  99         /* CPU pointers to the job descriptor headers. next_job is only
 100          * set at submit time (since only then are all the dependencies
 101          * known). The upshot is that this is append-only.
 102          *
 103          * These arrays contain the headers for the "primary batch", our jargon
 104          * referring to the part of the panfrost_job that actually contains
 105          * meaningful work. In an OpenGL ES setting, that means the
 106          * WRITE_VALUE/VERTEX/TILER jobs. Excluded is specifically the FRAGMENT
 107          * job, which is sent on as a secondary batch containing only a single
 108          * hardware job. Since there's one and only one FRAGMENT job issued per
 109          * panfrost_job, there is no need to do any scoreboarding / management;
 110          * it's easy enough to open-code it and it's not like we can get any
 111          * better anyway. */
 112         struct util_dynarray headers;
 113
 114         /* (And the GPU versions; TODO maybe combine) */
 115         struct util_dynarray gpu_headers;
 116
 117         /* The last job in the primary batch */
 118         struct panfrost_transfer last_job;
 119
 120         /* The first/last tiler job */
 121         struct panfrost_transfer first_tiler;
 122         struct panfrost_transfer last_tiler;
 123
 124         /* The first vertex job used as the input to a tiler job */
 125         struct panfrost_transfer first_vertex_for_tiler;
 126
 127         /* The first job. Notice we've created a linked list */
 128         struct panfrost_transfer first_job;
 129
 130         /* The number of jobs in the primary batch, essentially */
 131         unsigned job_index;
 132
 133         /* BOs referenced -- will be used for flushing logic */
 134         struct hash_table *bos;
 135
 136         /* Current transient BO */
 137         struct panfrost_bo *transient_bo;
 138
 139         /* Within the topmost transient BO, how much has been used? */
 140         unsigned transient_offset;
 141
 142         /* Polygon list bound to the batch, or NULL if none bound yet */
 143         struct panfrost_bo *polygon_list;
 144
 145         /* Scratchpath BO bound to the batch, or NULL if none bound yet */
 146         struct panfrost_bo *scratchpad;
 147
 148         /* Tiler heap BO bound to the batch, or NULL if none bound yet */
 149         struct panfrost_bo *tiler_heap;
 150
 151         /* Dummy tiler BO bound to the batch, or NULL if none bound yet */
 152         struct panfrost_bo *tiler_dummy;
 153
 154         /* Framebuffer descriptor. */
 155         struct panfrost_transfer framebuffer;
 156
 157         /* Output sync object. Only valid when submitted is true. */
 158         struct panfrost_batch_fence *out_sync;
 159
 160         /* Batch dependencies */
 161         struct util_dynarray dependencies;
 162 };
 163
 164 /* Functions for managing the above */
 165
 166 void
 167 panfrost_batch_fence_unreference(struct panfrost_batch_fence *fence);
 168
 169 void
 170 panfrost_batch_fence_reference(struct panfrost_batch_fence *batch);
 171
 172 struct panfrost_batch *
 173 panfrost_get_batch_for_fbo(struct panfrost_context *ctx);
 174
 175 struct panfrost_batch *
 176 panfrost_get_fresh_batch_for_fbo(struct panfrost_context *ctx);
 177
 178 void
 179 panfrost_batch_init(struct panfrost_context *ctx);
 180
 181 void
 182 panfrost_batch_add_bo(struct panfrost_batch *batch, struct panfrost_bo *bo,
 183                       uint32_t flags);
 184
 185 void panfrost_batch_add_fbo_bos(struct panfrost_batch *batch);
 186
 187 struct panfrost_bo *
 188 panfrost_batch_create_bo(struct panfrost_batch *batch, size_t size,
 189                          uint32_t create_flags, uint32_t access_flags);
 190
 191 void
 192 panfrost_flush_all_batches(struct panfrost_context *ctx, bool wait);
 193
 194 bool
 195 panfrost_pending_batches_access_bo(struct panfrost_context *ctx,
 196                                    const struct panfrost_bo *bo);
 197
 198 void
 199 panfrost_flush_batches_accessing_bo(struct panfrost_context *ctx,
 200                                     struct panfrost_bo *bo, uint32_t flags);
 201
 202 void
 203 panfrost_batch_set_requirements(struct panfrost_batch *batch);
 204
 205 struct panfrost_bo *
 206 panfrost_batch_get_scratchpad(struct panfrost_batch *batch, unsigned shift, unsigned thread_tls_alloc, unsigned core_count);
 207
 208 mali_ptr
 209 panfrost_batch_get_polygon_list(struct panfrost_batch *batch, unsigned size);
 210
 211 struct panfrost_bo *
 212 panfrost_batch_get_tiler_heap(struct panfrost_batch *batch);
 213
 214 struct panfrost_bo *
 215 panfrost_batch_get_tiler_dummy(struct panfrost_batch *batch);
 216
 217 void
 218 panfrost_batch_clear(struct panfrost_batch *batch,
 219                      unsigned buffers,
 220                      const union pipe_color_union *color,
 221                      double depth, unsigned stencil);
 222
 223 void
 224 panfrost_batch_union_scissor(struct panfrost_batch *batch,
 225                              unsigned minx, unsigned miny,
 226                              unsigned maxx, unsigned maxy);
 227
 228 void
 229 panfrost_batch_intersection_scissor(struct panfrost_batch *batch,
 230                                     unsigned minx, unsigned miny,
 231                                     unsigned maxx, unsigned maxy);
 232
 233 /* Scoreboarding */
 234
 235 void
 236 panfrost_scoreboard_queue_compute_job(
 237         struct panfrost_batch *batch,
 238         struct panfrost_transfer job);
 239
 240 void
 241 panfrost_scoreboard_queue_vertex_job(
 242         struct panfrost_batch *batch,
 243         struct panfrost_transfer vertex,
 244         bool requires_tiling);
 245
 246 void
 247 panfrost_scoreboard_queue_tiler_job(
 248         struct panfrost_batch *batch,
 249         struct panfrost_transfer tiler);
 250
 251 void
 252 panfrost_scoreboard_queue_fused_job(
 253         struct panfrost_batch *batch,
 254         struct panfrost_transfer vertex,
 255         struct panfrost_transfer tiler);
 256 void
 257 panfrost_scoreboard_queue_fused_job_prepend(
 258         struct panfrost_batch *batch,
 259         struct panfrost_transfer vertex,
 260         struct panfrost_transfer tiler);
 261
 262 void
 263 panfrost_scoreboard_link_batch(struct panfrost_batch *batch);
 264
 265 bool
 266 panfrost_batch_is_scanout(struct panfrost_batch *batch);
 267
 268 #endif