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 draw has corresponding vertex and tiler payloads */
 138         struct midgard_payload_vertex_tiler payload_vertex;
 139         struct midgard_payload_vertex_tiler payload_tiler;
 140
 141         /* The fragment shader binary itself is pointed here (for the tripipe) but
 142          * also everything else in the shader core, including blending, the
 143          * stencil/depth tests, etc. Refer to the presentations. */
 144
 145         struct mali_shader_meta fragment_shader_core;
 146
 147         /* A frame is composed of a starting set value job, a number of vertex
 148          * and tiler jobs, linked to the fragment job at the end. See the
 149          * presentations for more information how this works */
 150
 151         unsigned draw_count;
 152
 153         mali_ptr set_value_job;
 154         mali_ptr vertex_jobs[MAX_DRAW_CALLS];
 155         mali_ptr tiler_jobs[MAX_DRAW_CALLS];
 156
 157         struct mali_job_descriptor_header *u_set_value_job;
 158         struct mali_job_descriptor_header *u_vertex_jobs[MAX_DRAW_CALLS];
 159         struct mali_job_descriptor_header *u_tiler_jobs[MAX_DRAW_CALLS];
 160
 161         unsigned vertex_job_count;
 162         unsigned tiler_job_count;
 163
 164         /* Per-draw Dirty flags are setup like any other driver */
 165         int dirty;
 166
 167         unsigned vertex_count;
 168
 169         union mali_attr attributes[PIPE_MAX_ATTRIBS];
 170
 171         unsigned varying_height;
 172
 173         struct mali_viewport *viewport;
 174         struct mali_single_framebuffer vt_framebuffer_sfbd;
 175         struct bifrost_framebuffer vt_framebuffer_mfbd;
 176
 177         /* TODO: Multiple uniform buffers (index =/= 0), finer updates? */
 178
 179         struct panfrost_constant_buffer constant_buffer[PIPE_SHADER_TYPES];
 180
 181         /* CSOs */
 182         struct panfrost_rasterizer *rasterizer;
 183
 184         struct panfrost_shader_variants *vs;
 185         struct panfrost_shader_variants *fs;
 186
 187         struct panfrost_vertex_state *vertex;
 188
 189         struct pipe_vertex_buffer *vertex_buffers;
 190         unsigned vertex_buffer_count;
 191
 192         struct panfrost_sampler_state *samplers[PIPE_SHADER_TYPES][PIPE_MAX_SAMPLERS];
 193         unsigned sampler_count[PIPE_SHADER_TYPES];
 194
 195         struct panfrost_sampler_view *sampler_views[PIPE_SHADER_TYPES][PIPE_MAX_SHADER_SAMPLER_VIEWS];
 196         unsigned sampler_view_count[PIPE_SHADER_TYPES];
 197
 198         struct primconvert_context *primconvert;
 199         struct blitter_context *blitter;
 200
 201         struct panfrost_blend_state *blend;
 202
 203         struct pipe_viewport_state pipe_viewport;
 204         struct pipe_scissor_state scissor;
 205         struct pipe_blend_color blend_color;
 206         struct pipe_depth_stencil_alpha_state *depth_stencil;
 207         struct pipe_stencil_ref stencil_ref;
 208
 209         /* True for t6XX, false for t8xx. */
 210         bool is_t6xx;
 211
 212         /* If set, we'll require the use of single render-target framebuffer
 213          * descriptors (SFBD), for older hardware -- specifically, <T760 hardware, If
 214          * false, we'll use the MFBD no matter what. New hardware -does- retain support
 215          * for SFBD, and in theory we could flip between them on a per-RT basis, but
 216          * there's no real advantage to doing so */
 217         bool require_sfbd;
 218
 219         uint32_t out_sync;
 220 };
 221
 222 /* Corresponds to the CSO */
 223
 224 struct panfrost_rasterizer {
 225         struct pipe_rasterizer_state base;
 226
 227         /* Bitmask of front face, etc */
 228         unsigned tiler_gl_enables;
 229 };
 230
 231 struct panfrost_blend_state {
 232         struct pipe_blend_state base;
 233
 234         /* Whether a blend shader is in use */
 235         bool has_blend_shader;
 236
 237         /* Compiled fixed function command */
 238         struct mali_blend_equation equation;
 239
 240         /* Compiled blend shader */
 241         mali_ptr blend_shader;
 242         int blend_work_count;
 243 };
 244
 245 /* Internal varyings descriptor */
 246 struct panfrost_varyings {
 247         /* Varyings information: stride of each chunk of memory used for
 248          * varyings (similar structure with attributes). Count is just the
 249          * number of vec4's. Buffer count is the number of varying chunks (<=
 250          * count). Height is used to calculate gl_Position's position ("it's
 251          * not a pun, Alyssa!"). Vertex-only varyings == descriptor for
 252          * gl_Position and something else apparently occupying the same space.
 253          * Varyings == main varyings descriptors following typical mali_attr
 254          * conventions. */
 255
 256         unsigned varyings_stride[MAX_VARYINGS];
 257         unsigned varying_count;
 258         unsigned varying_buffer_count;
 259
 260         /* Map of the actual varyings buffer */
 261         uint8_t *varyings_buffer_cpu;
 262         mali_ptr varyings_descriptor;
 263         mali_ptr varyings_descriptor_fragment;
 264 };
 265
 266 /* Variants bundle together to form the backing CSO, bundling multiple
 267  * shaders with varying emulated features baked in (alpha test
 268  * parameters, etc) */
 269 #define MAX_SHADER_VARIANTS 8
 270
 271 /* A shader state corresponds to the actual, current variant of the shader */
 272 struct panfrost_shader_state {
 273         struct pipe_shader_state *base;
 274
 275         /* Compiled, mapped descriptor, ready for the hardware */
 276         bool compiled;
 277         struct mali_shader_meta *tripipe;
 278         mali_ptr tripipe_gpu;
 279
 280         /* Non-descript information */
 281         int uniform_count;
 282         bool can_discard;
 283         bool writes_point_size;
 284
 285         /* Valid for vertex shaders only due to when this is calculated */
 286         struct panfrost_varyings varyings;
 287
 288         /* Information on this particular shader variant */
 289         struct pipe_alpha_state alpha_state;
 290 };
 291
 292 /* A collection of varyings (the CSO) */
 293 struct panfrost_shader_variants {
 294         struct pipe_shader_state base;
 295
 296         struct panfrost_shader_state variants[MAX_SHADER_VARIANTS];
 297         unsigned variant_count;
 298
 299         /* The current active variant */
 300         unsigned active_variant;
 301 };
 302
 303 struct panfrost_vertex_state {
 304         unsigned num_elements;
 305
 306         struct pipe_vertex_element pipe[PIPE_MAX_ATTRIBS];
 307         int nr_components[PIPE_MAX_ATTRIBS];
 308
 309         /* The actual attribute meta, prebaked and GPU mapped. TODO: Free memory */
 310         struct mali_attr_meta *hw;
 311         mali_ptr descriptor_ptr;
 312 };
 313
 314 struct panfrost_sampler_state {
 315         struct pipe_sampler_state base;
 316         struct mali_sampler_descriptor hw;
 317 };
 318
 319 /* Misnomer: Sampler view corresponds to textures, not samplers */
 320
 321 struct panfrost_sampler_view {
 322         struct pipe_sampler_view base;
 323         struct mali_texture_descriptor hw;
 324 };
 325
 326 static inline struct panfrost_context *
 327 pan_context(struct pipe_context *pcontext)
 328 {
 329         return (struct panfrost_context *) pcontext;
 330 }
 331
 332 static inline struct panfrost_screen *
 333 pan_screen(struct pipe_screen *p)
 334 {
 335    return (struct panfrost_screen *)p;
 336 }
 337
 338 struct pipe_context *
 339 panfrost_create_context(struct pipe_screen *screen, void *priv, unsigned flags);
 340
 341 void
 342 panfrost_emit_for_draw(struct panfrost_context *ctx, bool with_vertex_data);
 343
 344 struct panfrost_transfer
 345 panfrost_vertex_tiler_job(struct panfrost_context *ctx, bool is_tiler, bool is_elided_tiler);
 346
 347 unsigned
 348 panfrost_get_default_swizzle(unsigned components);
 349
 350 void
 351 panfrost_flush(
 352         struct pipe_context *pipe,
 353         struct pipe_fence_handle **fence,
 354         unsigned flags);
 355
 356 bool
 357 panfrost_is_scanout(struct panfrost_context *ctx);
 358
 359 mali_ptr
 360 panfrost_sfbd_fragment(struct panfrost_context *ctx, bool flip_y);
 361
 362 mali_ptr
 363 panfrost_mfbd_fragment(struct panfrost_context *ctx, bool flip_y);
 364
 365 struct bifrost_framebuffer
 366 panfrost_emit_mfbd(struct panfrost_context *ctx);
 367
 368 struct mali_single_framebuffer
 369 panfrost_emit_sfbd(struct panfrost_context *ctx);
 370
 371 mali_ptr
 372 panfrost_fragment_job(struct panfrost_context *ctx);
 373
 374 void
 375 panfrost_shader_compile(struct panfrost_context *ctx, struct mali_shader_meta *meta, const char *src, int type, struct panfrost_shader_state *state);
 376
 377 #endif