src/amd/vulkan/radv_shader.h

   1 /*
   2  * Copyright © 2016 Red Hat.
   3  * Copyright © 2016 Bas Nieuwenhuizen
   4  *
   5  * based in part on anv driver which is:
   6  * Copyright © 2015 Intel Corporation
   7  *
   8  * Permission is hereby granted, free of charge, to any person obtaining a
   9  * copy of this software and associated documentation files (the "Software"),
  10  * to deal in the Software without restriction, including without limitation
  11  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  12  * and/or sell copies of the Software, and to permit persons to whom the
  13  * Software is furnished to do so, subject to the following conditions:
  14  *
  15  * The above copyright notice and this permission notice (including the next
  16  * paragraph) shall be included in all copies or substantial portions of the
  17  * Software.
  18  *
  19  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  20  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  21  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  22  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  23  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  24  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  25  * IN THE SOFTWARE.
  26  */
  27
  28 #ifndef RADV_SHADER_H
  29 #define RADV_SHADER_H
  30
  31 #include "radv_debug.h"
  32 #include "radv_private.h"
  33
  34 #include "nir/nir.h"
  35
  36 /* descriptor index into scratch ring offsets */
  37 #define RING_SCRATCH 0
  38 #define RING_ESGS_VS 1
  39 #define RING_ESGS_GS 2
  40 #define RING_GSVS_VS 3
  41 #define RING_GSVS_GS 4
  42 #define RING_HS_TESS_FACTOR 5
  43 #define RING_HS_TESS_OFFCHIP 6
  44 #define RING_PS_SAMPLE_POSITIONS 7
  45
  46 // Match MAX_SETS from radv_descriptor_set.h
  47 #define RADV_UD_MAX_SETS MAX_SETS
  48
  49 #define RADV_NUM_PHYSICAL_VGPRS 256
  50
  51 struct radv_shader_module {
  52         struct nir_shader *nir;
  53         unsigned char sha1[20];
  54         uint32_t size;
  55         char data[0];
  56 };
  57
  58 enum {
  59         RADV_ALPHA_ADJUST_NONE = 0,
  60         RADV_ALPHA_ADJUST_SNORM = 1,
  61         RADV_ALPHA_ADJUST_SINT = 2,
  62         RADV_ALPHA_ADJUST_SSCALED = 3,
  63 };
  64
  65 struct radv_vs_variant_key {
  66         uint32_t instance_rate_inputs;
  67         uint32_t instance_rate_divisors[MAX_VERTEX_ATTRIBS];
  68
  69         /* For 2_10_10_10 formats the alpha is handled as unsigned by pre-vega HW.
  70          * so we may need to fix it up. */
  71         uint64_t alpha_adjust;
  72
  73         uint32_t as_es:1;
  74         uint32_t as_ls:1;
  75         uint32_t export_prim_id:1;
  76         uint32_t export_layer_id:1;
  77 };
  78
  79 struct radv_tes_variant_key {
  80         uint32_t as_es:1;
  81         uint32_t export_prim_id:1;
  82         uint32_t export_layer_id:1;
  83         uint8_t num_patches;
  84         uint8_t tcs_num_outputs;
  85 };
  86
  87 struct radv_tcs_variant_key {
  88         struct radv_vs_variant_key vs_key;
  89         unsigned primitive_mode;
  90         unsigned input_vertices;
  91         unsigned num_inputs;
  92         uint32_t tes_reads_tess_factors:1;
  93 };
  94
  95 struct radv_fs_variant_key {
  96         uint32_t col_format;
  97         uint8_t log2_ps_iter_samples;
  98         uint8_t log2_num_samples;
  99         uint32_t is_int8;
 100         uint32_t is_int10;
 101 };
 102
 103 struct radv_shader_variant_key {
 104         union {
 105                 struct radv_vs_variant_key vs;
 106                 struct radv_fs_variant_key fs;
 107                 struct radv_tes_variant_key tes;
 108                 struct radv_tcs_variant_key tcs;
 109         };
 110         bool has_multiview_view_index;
 111 };
 112
 113 struct radv_nir_compiler_options {
 114         struct radv_pipeline_layout *layout;
 115         struct radv_shader_variant_key key;
 116         bool unsafe_math;
 117         bool supports_spill;
 118         bool clamp_shadow_reference;
 119         bool dump_shader;
 120         bool dump_preoptir;
 121         bool record_llvm_ir;
 122         bool check_ir;
 123         enum radeon_family family;
 124         enum chip_class chip_class;
 125         uint32_t tess_offchip_block_dw_size;
 126         uint32_t address32_hi;
 127 };
 128
 129 enum radv_ud_index {
 130         AC_UD_SCRATCH_RING_OFFSETS = 0,
 131         AC_UD_PUSH_CONSTANTS = 1,
 132         AC_UD_INDIRECT_DESCRIPTOR_SETS = 2,
 133         AC_UD_VIEW_INDEX = 3,
 134         AC_UD_SHADER_START = 4,
 135         AC_UD_VS_VERTEX_BUFFERS = AC_UD_SHADER_START,
 136         AC_UD_VS_BASE_VERTEX_START_INSTANCE,
 137         AC_UD_VS_MAX_UD,
 138         AC_UD_PS_SAMPLE_POS_OFFSET = AC_UD_SHADER_START,
 139         AC_UD_PS_MAX_UD,
 140         AC_UD_CS_GRID_SIZE = AC_UD_SHADER_START,
 141         AC_UD_CS_MAX_UD,
 142         AC_UD_GS_MAX_UD,
 143         AC_UD_TCS_MAX_UD,
 144         AC_UD_TES_MAX_UD,
 145         AC_UD_MAX_UD = AC_UD_TCS_MAX_UD,
 146 };
 147 struct radv_shader_info {
 148         bool loads_push_constants;
 149         uint32_t desc_set_used_mask;
 150         bool needs_multiview_view_index;
 151         bool uses_invocation_id;
 152         bool uses_prim_id;
 153         struct {
 154                 uint64_t ls_outputs_written;
 155                 uint8_t input_usage_mask[VERT_ATTRIB_MAX];
 156                 uint8_t output_usage_mask[VARYING_SLOT_VAR31 + 1];
 157                 bool has_vertex_buffers; /* needs vertex buffers and base/start */
 158                 bool needs_draw_id;
 159                 bool needs_instance_id;
 160         } vs;
 161         struct {
 162                 uint8_t output_usage_mask[VARYING_SLOT_VAR31 + 1];
 163         } gs;
 164         struct {
 165                 uint8_t output_usage_mask[VARYING_SLOT_VAR31 + 1];
 166         } tes;
 167         struct {
 168                 bool force_persample;
 169                 bool needs_sample_positions;
 170                 bool uses_input_attachments;
 171                 bool writes_memory;
 172                 bool writes_z;
 173                 bool writes_stencil;
 174                 bool writes_sample_mask;
 175                 bool has_pcoord;
 176                 bool prim_id_input;
 177                 bool layer_input;
 178         } ps;
 179         struct {
 180                 bool uses_grid_size;
 181                 bool uses_block_id[3];
 182                 bool uses_thread_id[3];
 183                 bool uses_local_invocation_idx;
 184         } cs;
 185         struct {
 186                 uint64_t outputs_written;
 187                 uint64_t patch_outputs_written;
 188         } tcs;
 189 };
 190
 191 struct radv_userdata_info {
 192         int8_t sgpr_idx;
 193         uint8_t num_sgprs;
 194         bool indirect;
 195         uint32_t indirect_offset;
 196 };
 197
 198 struct radv_userdata_locations {
 199         struct radv_userdata_info descriptor_sets[RADV_UD_MAX_SETS];
 200         struct radv_userdata_info shader_data[AC_UD_MAX_UD];
 201 };
 202
 203 struct radv_vs_output_info {
 204         uint8_t vs_output_param_offset[VARYING_SLOT_MAX];
 205         uint8_t clip_dist_mask;
 206         uint8_t cull_dist_mask;
 207         uint8_t param_exports;
 208         bool writes_pointsize;
 209         bool writes_layer;
 210         bool writes_viewport_index;
 211         bool export_prim_id;
 212         unsigned pos_exports;
 213 };
 214
 215 struct radv_es_output_info {
 216         uint32_t esgs_itemsize;
 217 };
 218
 219 struct radv_shader_variant_info {
 220         struct radv_userdata_locations user_sgprs_locs;
 221         struct radv_shader_info info;
 222         unsigned num_user_sgprs;
 223         unsigned num_input_sgprs;
 224         unsigned num_input_vgprs;
 225         unsigned private_mem_vgprs;
 226         bool need_indirect_descriptor_sets;
 227         struct {
 228                 struct {
 229                         struct radv_vs_output_info outinfo;
 230                         struct radv_es_output_info es_info;
 231                         unsigned vgpr_comp_cnt;
 232                         bool as_es;
 233                         bool as_ls;
 234                 } vs;
 235                 struct {
 236                         unsigned num_interp;
 237                         uint32_t input_mask;
 238                         uint32_t flat_shaded_mask;
 239                         bool can_discard;
 240                         bool early_fragment_test;
 241                 } fs;
 242                 struct {
 243                         unsigned block_size[3];
 244                 } cs;
 245                 struct {
 246                         unsigned vertices_in;
 247                         unsigned vertices_out;
 248                         unsigned output_prim;
 249                         unsigned invocations;
 250                         unsigned gsvs_vertex_size;
 251                         unsigned max_gsvs_emit_size;
 252                         unsigned es_type; /* GFX9: VS or TES */
 253                 } gs;
 254                 struct {
 255                         unsigned tcs_vertices_out;
 256                         uint32_t num_patches;
 257                         uint32_t lds_size;
 258                 } tcs;
 259                 struct {
 260                         struct radv_vs_output_info outinfo;
 261                         struct radv_es_output_info es_info;
 262                         bool as_es;
 263                         unsigned primitive_mode;
 264                         enum gl_tess_spacing spacing;
 265                         bool ccw;
 266                         bool point_mode;
 267                 } tes;
 268         };
 269 };
 270
 271 struct radv_shader_variant {
 272         uint32_t ref_count;
 273
 274         struct radeon_winsys_bo *bo;
 275         uint64_t bo_offset;
 276         struct ac_shader_config config;
 277         uint32_t code_size;
 278         struct radv_shader_variant_info info;
 279         unsigned rsrc1;
 280         unsigned rsrc2;
 281
 282         /* debug only */
 283         uint32_t *spirv;
 284         uint32_t spirv_size;
 285         struct nir_shader *nir;
 286         char *disasm_string;
 287         char *llvm_ir_string;
 288
 289         struct list_head slab_list;
 290 };
 291
 292 struct radv_shader_slab {
 293         struct list_head slabs;
 294         struct list_head shaders;
 295         struct radeon_winsys_bo *bo;
 296         uint64_t size;
 297         char *ptr;
 298 };
 299
 300 void
 301 radv_optimize_nir(struct nir_shader *shader, bool optimize_conservatively);
 302
 303 nir_shader *
 304 radv_shader_compile_to_nir(struct radv_device *device,
 305                            struct radv_shader_module *module,
 306                            const char *entrypoint_name,
 307                            gl_shader_stage stage,
 308                            const VkSpecializationInfo *spec_info,
 309                            const VkPipelineCreateFlags flags);
 310
 311 void *
 312 radv_alloc_shader_memory(struct radv_device *device,
 313                           struct radv_shader_variant *shader);
 314
 315 void
 316 radv_destroy_shader_slabs(struct radv_device *device);
 317
 318 struct radv_shader_variant *
 319 radv_shader_variant_create(struct radv_device *device,
 320                            struct radv_shader_module *module,
 321                            struct nir_shader *const *shaders,
 322                            int shader_count,
 323                            struct radv_pipeline_layout *layout,
 324                            const struct radv_shader_variant_key *key,
 325                            void **code_out,
 326                            unsigned *code_size_out);
 327
 328 struct radv_shader_variant *
 329 radv_create_gs_copy_shader(struct radv_device *device, struct nir_shader *nir,
 330                            void **code_out, unsigned *code_size_out,
 331                            bool multiview);
 332
 333 void
 334 radv_shader_variant_destroy(struct radv_device *device,
 335                             struct radv_shader_variant *variant);
 336
 337 const char *
 338 radv_get_shader_name(struct radv_shader_variant *var, gl_shader_stage stage);
 339
 340 void
 341 radv_shader_dump_stats(struct radv_device *device,
 342                        struct radv_shader_variant *variant,
 343                        gl_shader_stage stage,
 344                        FILE *file);
 345
 346 static inline bool
 347 radv_can_dump_shader(struct radv_device *device,
 348                      struct radv_shader_module *module,
 349                      bool is_gs_copy_shader)
 350 {
 351         if (!(device->instance->debug_flags & RADV_DEBUG_DUMP_SHADERS))
 352                 return false;
 353
 354         /* Only dump non-meta shaders, useful for debugging purposes. */
 355         return (module && !module->nir) || is_gs_copy_shader;
 356 }
 357
 358 static inline bool
 359 radv_can_dump_shader_stats(struct radv_device *device,
 360                            struct radv_shader_module *module)
 361 {
 362         /* Only dump non-meta shader stats. */
 363         return device->instance->debug_flags & RADV_DEBUG_DUMP_SHADER_STATS &&
 364                module && !module->nir;
 365 }
 366
 367 static inline unsigned shader_io_get_unique_index(gl_varying_slot slot)
 368 {
 369         /* handle patch indices separate */
 370         if (slot == VARYING_SLOT_TESS_LEVEL_OUTER)
 371                 return 0;
 372         if (slot == VARYING_SLOT_TESS_LEVEL_INNER)
 373                 return 1;
 374         if (slot >= VARYING_SLOT_PATCH0 && slot <= VARYING_SLOT_TESS_MAX)
 375                 return 2 + (slot - VARYING_SLOT_PATCH0);
 376         if (slot == VARYING_SLOT_POS)
 377                 return 0;
 378         if (slot == VARYING_SLOT_PSIZ)
 379                 return 1;
 380         if (slot == VARYING_SLOT_CLIP_DIST0)
 381                 return 2;
 382         /* 3 is reserved for clip dist as well */
 383         if (slot >= VARYING_SLOT_VAR0 && slot <= VARYING_SLOT_VAR31)
 384                 return 4 + (slot - VARYING_SLOT_VAR0);
 385         unreachable("illegal slot in get unique index\n");
 386 }
 387
 388 static inline uint32_t
 389 radv_get_num_physical_sgprs(struct radv_physical_device *physical_device)
 390 {
 391         return physical_device->rad_info.chip_class >= VI ? 800 : 512;
 392 }
 393
 394 #endif