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 #include "midgard/midgard_compile.h"
  46
  47 /* Forward declare to avoid extra header dep */
  48 struct prim_convert_context;
  49
  50 #define MAX_VARYINGS   4096
  51
  52 //#define PAN_DIRTY_CLEAR            (1 << 0)
  53 #define PAN_DIRTY_RASTERIZER (1 << 2)
  54 #define PAN_DIRTY_FS         (1 << 3)
  55 #define PAN_DIRTY_FRAG_CORE  (PAN_DIRTY_FS) /* Dirty writes are tied */
  56 #define PAN_DIRTY_VS         (1 << 4)
  57 #define PAN_DIRTY_VERTEX     (1 << 5)
  58 #define PAN_DIRTY_VERT_BUF   (1 << 6)
  59 //#define PAN_DIRTY_VIEWPORT   (1 << 7)
  60 #define PAN_DIRTY_SAMPLERS   (1 << 8)
  61 #define PAN_DIRTY_TEXTURES   (1 << 9)
  62
  63 #define SET_BIT(lval, bit, cond) \
  64         if (cond) \
  65                 lval |= (bit); \
  66         else \
  67                 lval &= ~(bit);
  68
  69 struct panfrost_constant_buffer {
  70         struct pipe_constant_buffer cb[PIPE_MAX_CONSTANT_BUFFERS];
  71         uint32_t enabled_mask;
  72         uint32_t dirty_mask;
  73 };
  74
  75 struct panfrost_query {
  76         /* Passthrough from Gallium */
  77         unsigned type;
  78         unsigned index;
  79
  80         /* Memory for the GPU to writeback the value of the query */
  81         struct panfrost_transfer transfer;
  82 };
  83
  84 struct panfrost_fence {
  85         struct pipe_reference reference;
  86         int fd;
  87 };
  88
  89 #define PANFROST_MAX_TRANSIENT_ENTRIES 64
  90
  91 struct panfrost_transient_pool {
  92         /* Memory blocks in the pool */
  93         struct panfrost_memory_entry *entries[PANFROST_MAX_TRANSIENT_ENTRIES];
  94
  95         /* Number of entries we own */
  96         unsigned entry_count;
  97
  98         /* Current entry that we are writing to, zero-indexed, strictly less than entry_count */
  99         unsigned entry_index;
 100
 101         /* Number of bytes into the current entry we are */
 102         off_t entry_offset;
 103
 104         /* Entry size (all entries must be homogenous) */
 105         size_t entry_size;
 106 };
 107
 108 struct panfrost_context {
 109         /* Gallium context */
 110         struct pipe_context base;
 111
 112         /* Bound job and map of panfrost_job_key to jobs */
 113         struct panfrost_job *job;
 114         struct hash_table *jobs;
 115
 116         /* panfrost_resource -> panfrost_job */
 117         struct hash_table *write_jobs;
 118
 119         /* Bit mask for supported PIPE_DRAW for this hardware */
 120         unsigned draw_modes;
 121
 122         struct pipe_framebuffer_state pipe_framebuffer;
 123
 124         /* The number of concurrent FBOs allowed depends on the number of pools
 125          * used; pools are ringed for parallelism opportunities */
 126
 127         struct panfrost_transient_pool transient_pools[2];
 128         int cmdstream_i;
 129
 130         struct panfrost_memory cmdstream_persistent;
 131         struct panfrost_memory shaders;
 132         struct panfrost_memory scratchpad;
 133         struct panfrost_memory tiler_heap;
 134         struct panfrost_memory varying_mem;
 135         struct panfrost_memory tiler_polygon_list;
 136         struct panfrost_memory tiler_dummy;
 137         struct panfrost_memory depth_stencil_buffer;
 138
 139         struct panfrost_query *occlusion_query;
 140
 141         /* Each draw has corresponding vertex and tiler payloads */
 142         struct midgard_payload_vertex_tiler payload_vertex;
 143         struct midgard_payload_vertex_tiler payload_tiler;
 144
 145         /* The fragment shader binary itself is pointed here (for the tripipe) but
 146          * also everything else in the shader core, including blending, the
 147          * stencil/depth tests, etc. Refer to the presentations. */
 148
 149         struct mali_shader_meta fragment_shader_core;
 150
 151         /* Per-draw Dirty flags are setup like any other driver */
 152         int dirty;
 153
 154         unsigned vertex_count;
 155
 156         union mali_attr attributes[PIPE_MAX_ATTRIBS];
 157
 158         unsigned varying_height;
 159
 160         struct mali_single_framebuffer vt_framebuffer_sfbd;
 161         struct bifrost_framebuffer vt_framebuffer_mfbd;
 162
 163         /* TODO: Multiple uniform buffers (index =/= 0), finer updates? */
 164
 165         struct panfrost_constant_buffer constant_buffer[PIPE_SHADER_TYPES];
 166
 167         /* CSOs */
 168         struct panfrost_rasterizer *rasterizer;
 169
 170         struct panfrost_shader_variants *vs;
 171         struct panfrost_shader_variants *fs;
 172
 173         struct panfrost_vertex_state *vertex;
 174
 175         struct pipe_vertex_buffer vertex_buffers[PIPE_MAX_ATTRIBS];
 176         uint32_t vb_mask;
 177
 178         struct panfrost_sampler_state *samplers[PIPE_SHADER_TYPES][PIPE_MAX_SAMPLERS];
 179         unsigned sampler_count[PIPE_SHADER_TYPES];
 180
 181         struct panfrost_sampler_view *sampler_views[PIPE_SHADER_TYPES][PIPE_MAX_SHADER_SAMPLER_VIEWS];
 182         unsigned sampler_view_count[PIPE_SHADER_TYPES];
 183
 184         struct primconvert_context *primconvert;
 185         struct blitter_context *blitter;
 186
 187         /* Blitting the wallpaper (the old contents of the framebuffer back to
 188          * itself) uses a dedicated u_blitter instance versus general blit()
 189          * callbacks from Gallium, as the blit() callback can trigger
 190          * wallpapering without Gallium realising, which in turns u_blitter
 191          * errors due to unsupported reucrsion */
 192
 193         struct blitter_context *blitter_wallpaper;
 194         struct panfrost_job *wallpaper_batch;
 195
 196         struct panfrost_blend_state *blend;
 197
 198         struct pipe_viewport_state pipe_viewport;
 199         struct pipe_scissor_state scissor;
 200         struct pipe_blend_color blend_color;
 201         struct pipe_depth_stencil_alpha_state *depth_stencil;
 202         struct pipe_stencil_ref stencil_ref;
 203
 204         /* True for t6XX, false for t8xx. */
 205         bool is_t6xx;
 206
 207         /* If set, we'll require the use of single render-target framebuffer
 208          * descriptors (SFBD), for older hardware -- specifically, <T760 hardware, If
 209          * false, we'll use the MFBD no matter what. New hardware -does- retain support
 210          * for SFBD, and in theory we could flip between them on a per-RT basis, but
 211          * there's no real advantage to doing so */
 212         bool require_sfbd;
 213
 214         uint32_t out_sync;
 215 };
 216
 217 /* Corresponds to the CSO */
 218
 219 struct panfrost_rasterizer {
 220         struct pipe_rasterizer_state base;
 221
 222         /* Bitmask of front face, etc */
 223         unsigned tiler_gl_enables;
 224 };
 225
 226 struct panfrost_blend_state {
 227         struct pipe_blend_state base;
 228
 229         /* Whether a blend shader is in use */
 230         bool has_blend_shader;
 231
 232         /* Compiled fixed function command */
 233         struct mali_blend_equation equation;
 234         float constant;
 235
 236         /* Compiled blend shader */
 237         mali_ptr blend_shader;
 238         int blend_work_count;
 239 };
 240
 241 /* Variants bundle together to form the backing CSO, bundling multiple
 242  * shaders with varying emulated features baked in (alpha test
 243  * parameters, etc) */
 244 #define MAX_SHADER_VARIANTS 8
 245
 246 /* A shader state corresponds to the actual, current variant of the shader */
 247 struct panfrost_shader_state {
 248         struct pipe_shader_state *base;
 249
 250         /* Compiled, mapped descriptor, ready for the hardware */
 251         bool compiled;
 252         struct mali_shader_meta *tripipe;
 253         mali_ptr tripipe_gpu;
 254
 255         /* Non-descript information */
 256         int uniform_count;
 257         bool can_discard;
 258         bool writes_point_size;
 259         bool reads_point_coord;
 260
 261         struct mali_attr_meta varyings[PIPE_MAX_ATTRIBS];
 262         gl_varying_slot varyings_loc[PIPE_MAX_ATTRIBS];
 263
 264         unsigned sysval_count;
 265         unsigned sysval[MAX_SYSVAL_COUNT];
 266
 267         /* Information on this particular shader variant */
 268         struct pipe_alpha_state alpha_state;
 269
 270         uint16_t point_sprite_mask;
 271         unsigned point_sprite_upper_left : 1;
 272 };
 273
 274 /* A collection of varyings (the CSO) */
 275 struct panfrost_shader_variants {
 276         struct pipe_shader_state base;
 277
 278         struct panfrost_shader_state variants[MAX_SHADER_VARIANTS];
 279         unsigned variant_count;
 280
 281         /* The current active variant */
 282         unsigned active_variant;
 283 };
 284
 285 struct panfrost_vertex_state {
 286         unsigned num_elements;
 287
 288         struct pipe_vertex_element pipe[PIPE_MAX_ATTRIBS];
 289         struct mali_attr_meta hw[PIPE_MAX_ATTRIBS];
 290 };
 291
 292 struct panfrost_sampler_state {
 293         struct pipe_sampler_state base;
 294         struct mali_sampler_descriptor hw;
 295 };
 296
 297 /* Misnomer: Sampler view corresponds to textures, not samplers */
 298
 299 struct panfrost_sampler_view {
 300         struct pipe_sampler_view base;
 301         struct mali_texture_descriptor hw;
 302 };
 303
 304 static inline struct panfrost_context *
 305 pan_context(struct pipe_context *pcontext)
 306 {
 307         return (struct panfrost_context *) pcontext;
 308 }
 309
 310 struct pipe_context *
 311 panfrost_create_context(struct pipe_screen *screen, void *priv, unsigned flags);
 312
 313 void
 314 panfrost_emit_for_draw(struct panfrost_context *ctx, bool with_vertex_data);
 315
 316 struct panfrost_transfer
 317 panfrost_vertex_tiler_job(struct panfrost_context *ctx, bool is_tiler);
 318
 319 unsigned
 320 panfrost_get_default_swizzle(unsigned components);
 321
 322 void
 323 panfrost_flush(
 324         struct pipe_context *pipe,
 325         struct pipe_fence_handle **fence,
 326         unsigned flags);
 327
 328 bool
 329 panfrost_is_scanout(struct panfrost_context *ctx);
 330
 331 mali_ptr panfrost_sfbd_fragment(struct panfrost_context *ctx, bool has_draws);
 332 mali_ptr panfrost_mfbd_fragment(struct panfrost_context *ctx, bool has_draws);
 333
 334 struct bifrost_framebuffer
 335 panfrost_emit_mfbd(struct panfrost_context *ctx, unsigned vertex_count);
 336
 337 struct mali_single_framebuffer
 338 panfrost_emit_sfbd(struct panfrost_context *ctx, unsigned vertex_count);
 339
 340 mali_ptr
 341 panfrost_fragment_job(struct panfrost_context *ctx, bool has_draws);
 342
 343 void
 344 panfrost_shader_compile(struct panfrost_context *ctx, struct mali_shader_meta *meta, const char *src, int type, struct panfrost_shader_state *state);
 345
 346 void
 347 panfrost_pack_work_groups_compute(
 348                 struct mali_vertex_tiler_prefix *out,
 349                 unsigned num_x,
 350                 unsigned num_y,
 351                 unsigned num_z,
 352                 unsigned size_x,
 353                 unsigned size_y,
 354                 unsigned size_z);
 355
 356 void
 357 panfrost_pack_work_groups_fused(
 358                 struct mali_vertex_tiler_prefix *vertex,
 359                 struct mali_vertex_tiler_prefix *tiler,
 360                 unsigned num_x,
 361                 unsigned num_y,
 362                 unsigned num_z,
 363                 unsigned size_x,
 364                 unsigned size_y,
 365                 unsigned size_z);
 366
 367
 368
 369 #endif