src/gallium/drivers/panfrost/pan_context.h

   1 /*
   2  * © Copyright 2018 Alyssa Rosenzweig
   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 FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  *
  23  */
  24
  25 #ifndef __BUILDER_H__
  26 #define __BUILDER_H__
  27
  28 #define _LARGEFILE64_SOURCE 1
  29 #define CACHE_LINE_SIZE 1024 /* TODO */
  30 #include <sys/mman.h>
  31 #include <assert.h>
  32 #include "pan_resource.h"
  33 #include "pan_job.h"
  34
  35 #include "pipe/p_compiler.h"
  36 #include "pipe/p_config.h"
  37 #include "pipe/p_context.h"
  38 #include "pipe/p_defines.h"
  39 #include "pipe/p_format.h"
  40 #include "pipe/p_screen.h"
  41 #include "pipe/p_state.h"
  42 #include "util/u_blitter.h"
  43 #include "util/hash_table.h"
  44
  45 /* Forward declare to avoid extra header dep */
  46 struct prim_convert_context;
  47
  48 #define MAX_DRAW_CALLS 4096
  49 #define MAX_VARYINGS   4096
  50
  51 //#define PAN_DIRTY_CLEAR            (1 << 0)
  52 #define PAN_DIRTY_RASTERIZER (1 << 2)
  53 #define PAN_DIRTY_FS         (1 << 3)
  54 #define PAN_DIRTY_FRAG_CORE  (PAN_DIRTY_FS) /* Dirty writes are tied */
  55 #define PAN_DIRTY_VS         (1 << 4)
  56 #define PAN_DIRTY_VERTEX     (1 << 5)
  57 #define PAN_DIRTY_VERT_BUF   (1 << 6)
  58 //#define PAN_DIRTY_VIEWPORT   (1 << 7)
  59 #define PAN_DIRTY_SAMPLERS   (1 << 8)
  60 #define PAN_DIRTY_TEXTURES   (1 << 9)
  61
  62 #define SET_BIT(lval, bit, cond) \
  63         if (cond) \
  64                 lval |= (bit); \
  65         else \
  66                 lval &= ~(bit);
  67
  68 struct panfrost_constant_buffer {
  69         bool dirty;
  70         size_t size;
  71         void *buffer;
  72 };
  73
  74 struct panfrost_query {
  75         /* Passthrough from Gallium */
  76         unsigned type;
  77         unsigned index;
  78
  79         /* Memory for the GPU to writeback the value of the query */
  80         struct panfrost_transfer transfer;
  81 };
  82
  83 struct panfrost_fence {
  84         struct pipe_reference reference;
  85         int fd;
  86 };
  87
  88 #define PANFROST_MAX_TRANSIENT_ENTRIES 64
  89
  90 struct panfrost_transient_pool {
  91         /* Memory blocks in the pool */
  92         struct panfrost_memory_entry *entries[PANFROST_MAX_TRANSIENT_ENTRIES];
  93
  94         /* Number of entries we own */
  95         unsigned entry_count;
  96
  97         /* Current entry that we are writing to, zero-indexed, strictly less than entry_count */
  98         unsigned entry_index;
  99
 100         /* Number of bytes into the current entry we are */
 101         off_t entry_offset;
 102
 103         /* Entry size (all entries must be homogenous) */
 104         size_t entry_size;
 105 };
 106
 107 struct panfrost_context {
 108         /* Gallium context */
 109         struct pipe_context base;
 110
 111         /* Bound job and map of panfrost_job_key to jobs */
 112         struct panfrost_job *job;
 113         struct hash_table *jobs;
 114
 115         /* Bit mask for supported PIPE_DRAW for this hardware */
 116         unsigned draw_modes;
 117
 118         struct pipe_framebuffer_state pipe_framebuffer;
 119
 120         /* The number of concurrent FBOs allowed depends on the number of pools
 121          * used; pools are ringed for parallelism opportunities */
 122
 123         struct panfrost_transient_pool transient_pools[2];
 124         int cmdstream_i;
 125
 126         struct panfrost_memory cmdstream_persistent;
 127         struct panfrost_memory shaders;
 128         struct panfrost_memory scratchpad;
 129         struct panfrost_memory tiler_heap;
 130         struct panfrost_memory varying_mem;
 131         struct panfrost_memory misc_0;
 132         struct panfrost_memory misc_1;
 133         struct panfrost_memory depth_stencil_buffer;
 134
 135         struct panfrost_query *occlusion_query;
 136
 137         /* Each render job has multiple framebuffer descriptors associated with
 138          * it, used for various purposes with more or less the same format. The
 139          * most obvious is the fragment framebuffer descriptor, which carries
 140          * e.g. clearing information */
 141
 142         union {
 143                 struct {
 144                         struct bifrost_framebuffer fragment_mfbd;
 145                         struct bifrost_fb_extra fragment_extra;
 146                         struct bifrost_render_target fragment_rts[4];
 147                 };
 148         };
 149
 150         /* Each draw has corresponding vertex and tiler payloads */
 151         struct midgard_payload_vertex_tiler payload_vertex;
 152         struct midgard_payload_vertex_tiler payload_tiler;
 153
 154         /* The fragment shader binary itself is pointed here (for the tripipe) but
 155          * also everything else in the shader core, including blending, the
 156          * stencil/depth tests, etc. Refer to the presentations. */
 157
 158         struct mali_shader_meta fragment_shader_core;
 159
 160         /* A frame is composed of a starting set value job, a number of vertex
 161          * and tiler jobs, linked to the fragment job at the end. See the
 162          * presentations for more information how this works */
 163
 164         unsigned draw_count;
 165
 166         mali_ptr set_value_job;
 167         mali_ptr vertex_jobs[MAX_DRAW_CALLS];
 168         mali_ptr tiler_jobs[MAX_DRAW_CALLS];
 169
 170         struct mali_job_descriptor_header *u_set_value_job;
 171         struct mali_job_descriptor_header *u_vertex_jobs[MAX_DRAW_CALLS];
 172         struct mali_job_descriptor_header *u_tiler_jobs[MAX_DRAW_CALLS];
 173
 174         unsigned vertex_job_count;
 175         unsigned tiler_job_count;
 176
 177         /* Per-draw Dirty flags are setup like any other driver */
 178         int dirty;
 179
 180         unsigned vertex_count;
 181
 182         union mali_attr attributes[PIPE_MAX_ATTRIBS];
 183
 184         unsigned varying_height;
 185
 186         struct mali_viewport *viewport;
 187         struct mali_single_framebuffer vt_framebuffer_sfbd;
 188         struct bifrost_framebuffer vt_framebuffer_mfbd;
 189
 190         /* TODO: Multiple uniform buffers (index =/= 0), finer updates? */
 191
 192         struct panfrost_constant_buffer constant_buffer[PIPE_SHADER_TYPES];
 193
 194         /* CSOs */
 195         struct panfrost_rasterizer *rasterizer;
 196
 197         struct panfrost_shader_variants *vs;
 198         struct panfrost_shader_variants *fs;
 199
 200         struct panfrost_vertex_state *vertex;
 201
 202         struct pipe_vertex_buffer *vertex_buffers;
 203         unsigned vertex_buffer_count;
 204
 205         struct panfrost_sampler_state *samplers[PIPE_SHADER_TYPES][PIPE_MAX_SAMPLERS];
 206         unsigned sampler_count[PIPE_SHADER_TYPES];
 207
 208         struct panfrost_sampler_view *sampler_views[PIPE_SHADER_TYPES][PIPE_MAX_SHADER_SAMPLER_VIEWS];
 209         unsigned sampler_view_count[PIPE_SHADER_TYPES];
 210
 211         struct primconvert_context *primconvert;
 212         struct blitter_context *blitter;
 213
 214         struct panfrost_blend_state *blend;
 215
 216         struct pipe_viewport_state pipe_viewport;
 217         struct pipe_scissor_state scissor;
 218         struct pipe_blend_color blend_color;
 219         struct pipe_depth_stencil_alpha_state *depth_stencil;
 220         struct pipe_stencil_ref stencil_ref;
 221
 222         /* True for t6XX, false for t8xx. */
 223         bool is_t6xx;
 224
 225         /* If set, we'll require the use of single render-target framebuffer
 226          * descriptors (SFBD), for older hardware -- specifically, <T760 hardware, If
 227          * false, we'll use the MFBD no matter what. New hardware -does- retain support
 228          * for SFBD, and in theory we could flip between them on a per-RT basis, but
 229          * there's no real advantage to doing so */
 230         bool require_sfbd;
 231
 232         uint32_t out_sync;
 233 };
 234
 235 /* Corresponds to the CSO */
 236
 237 struct panfrost_rasterizer {
 238         struct pipe_rasterizer_state base;
 239
 240         /* Bitmask of front face, etc */
 241         unsigned tiler_gl_enables;
 242 };
 243
 244 struct panfrost_blend_state {
 245         struct pipe_blend_state base;
 246
 247         /* Whether a blend shader is in use */
 248         bool has_blend_shader;
 249
 250         /* Compiled fixed function command */
 251         struct mali_blend_equation equation;
 252
 253         /* Compiled blend shader */
 254         mali_ptr blend_shader;
 255         int blend_work_count;
 256 };
 257
 258 /* Internal varyings descriptor */
 259 struct panfrost_varyings {
 260         /* Varyings information: stride of each chunk of memory used for
 261          * varyings (similar structure with attributes). Count is just the
 262          * number of vec4's. Buffer count is the number of varying chunks (<=
 263          * count). Height is used to calculate gl_Position's position ("it's
 264          * not a pun, Alyssa!"). Vertex-only varyings == descriptor for
 265          * gl_Position and something else apparently occupying the same space.
 266          * Varyings == main varyings descriptors following typical mali_attr
 267          * conventions. */
 268
 269         unsigned varyings_stride[MAX_VARYINGS];
 270         unsigned varying_count;
 271         unsigned varying_buffer_count;
 272
 273         /* Map of the actual varyings buffer */
 274         uint8_t *varyings_buffer_cpu;
 275         mali_ptr varyings_descriptor;
 276         mali_ptr varyings_descriptor_fragment;
 277 };
 278
 279 /* Variants bundle together to form the backing CSO, bundling multiple
 280  * shaders with varying emulated features baked in (alpha test
 281  * parameters, etc) */
 282 #define MAX_SHADER_VARIANTS 8
 283
 284 /* A shader state corresponds to the actual, current variant of the shader */
 285 struct panfrost_shader_state {
 286         struct pipe_shader_state *base;
 287
 288         /* Compiled, mapped descriptor, ready for the hardware */
 289         bool compiled;
 290         struct mali_shader_meta *tripipe;
 291         mali_ptr tripipe_gpu;
 292
 293         /* Non-descript information */
 294         int uniform_count;
 295         bool can_discard;
 296         bool writes_point_size;
 297
 298         /* Valid for vertex shaders only due to when this is calculated */
 299         struct panfrost_varyings varyings;
 300
 301         /* Information on this particular shader variant */
 302         struct pipe_alpha_state alpha_state;
 303 };
 304
 305 /* A collection of varyings (the CSO) */
 306 struct panfrost_shader_variants {
 307         struct pipe_shader_state base;
 308
 309         struct panfrost_shader_state variants[MAX_SHADER_VARIANTS];
 310         unsigned variant_count;
 311
 312         /* The current active variant */
 313         unsigned active_variant;
 314 };
 315
 316 struct panfrost_vertex_state {
 317         unsigned num_elements;
 318
 319         struct pipe_vertex_element pipe[PIPE_MAX_ATTRIBS];
 320         int nr_components[PIPE_MAX_ATTRIBS];
 321
 322         /* The actual attribute meta, prebaked and GPU mapped. TODO: Free memory */
 323         struct mali_attr_meta *hw;
 324         mali_ptr descriptor_ptr;
 325 };
 326
 327 struct panfrost_sampler_state {
 328         struct pipe_sampler_state base;
 329         struct mali_sampler_descriptor hw;
 330 };
 331
 332 /* Misnomer: Sampler view corresponds to textures, not samplers */
 333
 334 struct panfrost_sampler_view {
 335         struct pipe_sampler_view base;
 336         struct mali_texture_descriptor hw;
 337 };
 338
 339 static inline struct panfrost_context *
 340 pan_context(struct pipe_context *pcontext)
 341 {
 342         return (struct panfrost_context *) pcontext;
 343 }
 344
 345 static inline struct panfrost_screen *
 346 pan_screen(struct pipe_screen *p)
 347 {
 348    return (struct panfrost_screen *)p;
 349 }
 350
 351 struct pipe_context *
 352 panfrost_create_context(struct pipe_screen *screen, void *priv, unsigned flags);
 353
 354 void
 355 panfrost_emit_for_draw(struct panfrost_context *ctx, bool with_vertex_data);
 356
 357 struct panfrost_transfer
 358 panfrost_vertex_tiler_job(struct panfrost_context *ctx, bool is_tiler, bool is_elided_tiler);
 359
 360 unsigned
 361 panfrost_get_default_swizzle(unsigned components);
 362
 363 void
 364 panfrost_flush(
 365         struct pipe_context *pipe,
 366         struct pipe_fence_handle **fence,
 367         unsigned flags);
 368
 369 bool
 370 panfrost_is_scanout(struct panfrost_context *ctx);
 371
 372 mali_ptr
 373 panfrost_sfbd_fragment(struct panfrost_context *ctx, bool flip_y);
 374
 375 mali_ptr
 376 panfrost_mfbd_fragment(struct panfrost_context *ctx, bool flip_y);
 377
 378 struct bifrost_framebuffer
 379 panfrost_emit_mfbd(struct panfrost_context *ctx);
 380
 381 struct mali_single_framebuffer
 382 panfrost_emit_sfbd(struct panfrost_context *ctx);
 383
 384 mali_ptr
 385 panfrost_fragment_job(struct panfrost_context *ctx);
 386
 387 void
 388 panfrost_shader_compile(struct panfrost_context *ctx, struct mali_shader_meta *meta, const char *src, int type, struct panfrost_shader_state *state);
 389
 390 #endif