src/gallium/drivers/radeonsi/si_shader.h

   1 /*
   2  * Copyright 2012 Advanced Micro Devices, Inc.
   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  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *      Tom Stellard <thomas.stellard@amd.com>
  25  *      Michel Dänzer <michel.daenzer@amd.com>
  26  *      Christian König <christian.koenig@amd.com>
  27  */
  28
  29 /* How linking shader inputs and outputs between vertex, tessellation, and
  30  * geometry shaders works.
  31  *
  32  * Inputs and outputs between shaders are stored in a buffer. This buffer
  33  * lives in LDS (typical case for tessellation), but it can also live
  34  * in memory (ESGS). Each input or output has a fixed location within a vertex.
  35  * The highest used input or output determines the stride between vertices.
  36  *
  37  * Since GS and tessellation are only possible in the OpenGL core profile,
  38  * only these semantics are valid for per-vertex data:
  39  *
  40  *   Name             Location
  41  *
  42  *   POSITION         0
  43  *   PSIZE            1
  44  *   CLIPDIST0..1     2..3
  45  *   CULLDIST0..1     (not implemented)
  46  *   GENERIC0..31     4..35
  47  *
  48  * For example, a shader only writing GENERIC0 has the output stride of 5.
  49  *
  50  * Only these semantics are valid for per-patch data:
  51  *
  52  *   Name             Location
  53  *
  54  *   TESSOUTER        0
  55  *   TESSINNER        1
  56  *   PATCH0..29       2..31
  57  *
  58  * That's how independent shaders agree on input and output locations.
  59  * The si_shader_io_get_unique_index function assigns the locations.
  60  *
  61  * For tessellation, other required information for calculating the input and
  62  * output addresses like the vertex stride, the patch stride, and the offsets
  63  * where per-vertex and per-patch data start, is passed to the shader via
  64  * user data SGPRs. The offsets and strides are calculated at draw time and
  65  * aren't available at compile time.
  66  */
  67
  68 #ifndef SI_SHADER_H
  69 #define SI_SHADER_H
  70
  71 #include <llvm-c/Core.h> /* LLVMModuleRef */
  72 #include <llvm-c/TargetMachine.h>
  73 #include "tgsi/tgsi_scan.h"
  74 #include "util/u_queue.h"
  75 #include "si_state.h"
  76
  77 struct radeon_shader_binary;
  78 struct radeon_shader_reloc;
  79
  80 #define SI_MAX_VS_OUTPUTS       40
  81
  82 /* SGPR user data indices */
  83 enum {
  84         SI_SGPR_RW_BUFFERS,  /* rings (& stream-out, VS only) */
  85         SI_SGPR_RW_BUFFERS_HI,
  86         SI_SGPR_CONST_BUFFERS,
  87         SI_SGPR_CONST_BUFFERS_HI,
  88         SI_SGPR_SAMPLERS,  /* images & sampler states interleaved */
  89         SI_SGPR_SAMPLERS_HI,
  90         SI_SGPR_IMAGES,
  91         SI_SGPR_IMAGES_HI,
  92         SI_SGPR_SHADER_BUFFERS,
  93         SI_SGPR_SHADER_BUFFERS_HI,
  94         SI_NUM_RESOURCE_SGPRS,
  95
  96         /* all VS variants */
  97         SI_SGPR_VERTEX_BUFFERS  = SI_NUM_RESOURCE_SGPRS,
  98         SI_SGPR_VERTEX_BUFFERS_HI,
  99         SI_SGPR_BASE_VERTEX,
 100         SI_SGPR_START_INSTANCE,
 101         SI_SGPR_DRAWID,
 102         SI_ES_NUM_USER_SGPR,
 103
 104         /* hw VS only */
 105         SI_SGPR_VS_STATE_BITS   = SI_ES_NUM_USER_SGPR,
 106         SI_VS_NUM_USER_SGPR,
 107
 108         /* hw LS only */
 109         SI_SGPR_LS_OUT_LAYOUT   = SI_ES_NUM_USER_SGPR,
 110         SI_LS_NUM_USER_SGPR,
 111
 112         /* both TCS and TES */
 113         SI_SGPR_TCS_OFFCHIP_LAYOUT = SI_NUM_RESOURCE_SGPRS,
 114         SI_TES_NUM_USER_SGPR,
 115
 116         /* TCS only */
 117         SI_SGPR_TCS_OUT_OFFSETS = SI_TES_NUM_USER_SGPR,
 118         SI_SGPR_TCS_OUT_LAYOUT,
 119         SI_SGPR_TCS_IN_LAYOUT,
 120         SI_TCS_NUM_USER_SGPR,
 121
 122         /* GS limits */
 123         SI_GS_NUM_USER_SGPR = SI_NUM_RESOURCE_SGPRS,
 124         SI_GSCOPY_NUM_USER_SGPR = SI_SGPR_RW_BUFFERS_HI + 1,
 125
 126         /* PS only */
 127         SI_SGPR_ALPHA_REF       = SI_NUM_RESOURCE_SGPRS,
 128         SI_PS_NUM_USER_SGPR,
 129
 130         /* CS only */
 131         SI_SGPR_GRID_SIZE = SI_NUM_RESOURCE_SGPRS,
 132         SI_SGPR_BLOCK_SIZE = SI_SGPR_GRID_SIZE + 3,
 133         SI_CS_NUM_USER_SGPR = SI_SGPR_BLOCK_SIZE + 3
 134 };
 135
 136 /* LLVM function parameter indices */
 137 enum {
 138         SI_PARAM_RW_BUFFERS,
 139         SI_PARAM_CONST_BUFFERS,
 140         SI_PARAM_SAMPLERS,
 141         SI_PARAM_IMAGES,
 142         SI_PARAM_SHADER_BUFFERS,
 143         SI_NUM_RESOURCE_PARAMS,
 144
 145         /* VS only parameters */
 146         SI_PARAM_VERTEX_BUFFERS = SI_NUM_RESOURCE_PARAMS,
 147         SI_PARAM_BASE_VERTEX,
 148         SI_PARAM_START_INSTANCE,
 149         SI_PARAM_DRAWID,
 150         /* [0] = clamp vertex color, VS as VS only */
 151         SI_PARAM_VS_STATE_BITS,
 152         /* same value as TCS_IN_LAYOUT, VS as LS only */
 153         SI_PARAM_LS_OUT_LAYOUT = SI_PARAM_DRAWID + 1,
 154         /* the other VS parameters are assigned dynamically */
 155
 156         /* Layout of TCS outputs in the offchip buffer
 157          *   [0:8] = the number of patches per threadgroup.
 158          *   [9:15] = the number of output vertices per patch.
 159          *   [16:31] = the offset of per patch attributes in the buffer in bytes.
 160          */
 161         SI_PARAM_TCS_OFFCHIP_LAYOUT = SI_NUM_RESOURCE_PARAMS, /* for TCS & TES */
 162
 163         /* TCS only parameters. */
 164
 165         /* Offsets where TCS outputs and TCS patch outputs live in LDS:
 166          *   [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32
 167          *   [16:31] = TCS output patch0 offset for per-patch / 16, max = NUM_PATCHES*32*32* + 32*32
 168          */
 169         SI_PARAM_TCS_OUT_OFFSETS,
 170
 171         /* Layout of TCS outputs / TES inputs:
 172          *   [0:12] = stride between output patches in dwords, num_outputs * num_vertices * 4, max = 32*32*4
 173          *   [13:20] = stride between output vertices in dwords = num_inputs * 4, max = 32*4
 174          *   [26:31] = gl_PatchVerticesIn, max = 32
 175          */
 176         SI_PARAM_TCS_OUT_LAYOUT,
 177
 178         /* Layout of LS outputs / TCS inputs
 179          *   [0:12] = stride between patches in dwords = num_inputs * num_vertices * 4, max = 32*32*4
 180          *   [13:20] = stride between vertices in dwords = num_inputs * 4, max = 32*4
 181          */
 182         SI_PARAM_TCS_IN_LAYOUT,
 183
 184         SI_PARAM_TCS_OC_LDS,
 185         SI_PARAM_TESS_FACTOR_OFFSET,
 186         SI_PARAM_PATCH_ID,
 187         SI_PARAM_REL_IDS,
 188
 189         /* GS only parameters */
 190         SI_PARAM_GS2VS_OFFSET = SI_NUM_RESOURCE_PARAMS,
 191         SI_PARAM_GS_WAVE_ID,
 192         SI_PARAM_VTX0_OFFSET,
 193         SI_PARAM_VTX1_OFFSET,
 194         SI_PARAM_PRIMITIVE_ID,
 195         SI_PARAM_VTX2_OFFSET,
 196         SI_PARAM_VTX3_OFFSET,
 197         SI_PARAM_VTX4_OFFSET,
 198         SI_PARAM_VTX5_OFFSET,
 199         SI_PARAM_GS_INSTANCE_ID,
 200
 201         /* PS only parameters */
 202         SI_PARAM_ALPHA_REF = SI_NUM_RESOURCE_PARAMS,
 203         SI_PARAM_PRIM_MASK,
 204         SI_PARAM_PERSP_SAMPLE,
 205         SI_PARAM_PERSP_CENTER,
 206         SI_PARAM_PERSP_CENTROID,
 207         SI_PARAM_PERSP_PULL_MODEL,
 208         SI_PARAM_LINEAR_SAMPLE,
 209         SI_PARAM_LINEAR_CENTER,
 210         SI_PARAM_LINEAR_CENTROID,
 211         SI_PARAM_LINE_STIPPLE_TEX,
 212         SI_PARAM_POS_X_FLOAT,
 213         SI_PARAM_POS_Y_FLOAT,
 214         SI_PARAM_POS_Z_FLOAT,
 215         SI_PARAM_POS_W_FLOAT,
 216         SI_PARAM_FRONT_FACE,
 217         SI_PARAM_ANCILLARY,
 218         SI_PARAM_SAMPLE_COVERAGE,
 219         SI_PARAM_POS_FIXED_PT,
 220
 221         /* CS only parameters */
 222         SI_PARAM_GRID_SIZE = SI_NUM_RESOURCE_PARAMS,
 223         SI_PARAM_BLOCK_SIZE,
 224         SI_PARAM_BLOCK_ID,
 225         SI_PARAM_THREAD_ID,
 226
 227         SI_NUM_PARAMS = SI_PARAM_POS_FIXED_PT + 9, /* +8 for COLOR[0..1] */
 228 };
 229
 230 /* SI-specific system values. */
 231 enum {
 232         TGSI_SEMANTIC_DEFAULT_TESSOUTER_SI = TGSI_SEMANTIC_COUNT,
 233         TGSI_SEMANTIC_DEFAULT_TESSINNER_SI,
 234 };
 235
 236 struct si_shader;
 237
 238 /* A shader selector is a gallium CSO and contains shader variants and
 239  * binaries for one TGSI program. This can be shared by multiple contexts.
 240  */
 241 struct si_shader_selector {
 242         struct si_screen        *screen;
 243         struct util_queue_fence ready;
 244
 245         /* Should only be used by si_init_shader_selector_async
 246          * if thread_index == -1 (non-threaded). */
 247         LLVMTargetMachineRef    tm;
 248         struct pipe_debug_callback debug;
 249         bool                    is_debug_context;
 250
 251         pipe_mutex              mutex;
 252         struct si_shader        *first_variant; /* immutable after the first variant */
 253         struct si_shader        *last_variant; /* mutable */
 254
 255         /* The compiled TGSI shader expecting a prolog and/or epilog (not
 256          * uploaded to a buffer).
 257          */
 258         struct si_shader        *main_shader_part;
 259
 260         struct tgsi_token       *tokens;
 261         struct pipe_stream_output_info  so;
 262         struct tgsi_shader_info         info;
 263
 264         /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
 265         unsigned        type;
 266
 267         /* GS parameters. */
 268         unsigned        esgs_itemsize;
 269         unsigned        gs_input_verts_per_prim;
 270         unsigned        gs_output_prim;
 271         unsigned        gs_max_out_vertices;
 272         unsigned        gs_num_invocations;
 273         unsigned        max_gs_stream; /* count - 1 */
 274         unsigned        gsvs_vertex_size;
 275         unsigned        max_gsvs_emit_size;
 276
 277         /* PS parameters. */
 278         unsigned        color_attr_index[2];
 279         unsigned        db_shader_control;
 280         /* Set 0xf or 0x0 (4 bits) per each written output.
 281          * ANDed with spi_shader_col_format.
 282          */
 283         unsigned        colors_written_4bit;
 284
 285         /* CS parameters */
 286         unsigned local_size;
 287
 288         /* masks of "get_unique_index" bits */
 289         uint64_t        outputs_written;
 290         uint32_t        patch_outputs_written;
 291 };
 292
 293 /* Valid shader configurations:
 294  *
 295  * API shaders       VS | TCS | TES | GS |pass| PS
 296  * are compiled as:     |     |     |    |thru|
 297  *                      |     |     |    |    |
 298  * Only VS & PS:     VS | --  | --  | -- | -- | PS
 299  * With GS:          ES | --  | --  | GS | VS | PS
 300  * With Tessel.:     LS | HS  | VS  | -- | -- | PS
 301  * With both:        LS | HS  | ES  | GS | VS | PS
 302  */
 303
 304 /* Common VS bits between the shader key and the prolog key. */
 305 struct si_vs_prolog_bits {
 306         unsigned        instance_divisors[SI_NUM_VERTEX_BUFFERS];
 307 };
 308
 309 /* Common VS bits between the shader key and the epilog key. */
 310 struct si_vs_epilog_bits {
 311         unsigned        export_prim_id:1; /* when PS needs it and GS is disabled */
 312         /* TODO:
 313          * - skip clipdist, culldist (including clipvertex code) exports based
 314          *   on which clip_plane_enable bits are set
 315          * - skip layer, viewport, clipdist, and culldist parameter exports
 316          *   if PS doesn't read them
 317          */
 318 };
 319
 320 /* Common TCS bits between the shader key and the epilog key. */
 321 struct si_tcs_epilog_bits {
 322         unsigned        prim_mode:3;
 323         uint64_t        inputs_to_copy;
 324 };
 325
 326 /* Common PS bits between the shader key and the prolog key. */
 327 struct si_ps_prolog_bits {
 328         unsigned        color_two_side:1;
 329         unsigned        flatshade_colors:1;
 330         unsigned        poly_stipple:1;
 331         unsigned        force_persp_sample_interp:1;
 332         unsigned        force_linear_sample_interp:1;
 333         unsigned        force_persp_center_interp:1;
 334         unsigned        force_linear_center_interp:1;
 335         unsigned        bc_optimize_for_persp:1;
 336         unsigned        bc_optimize_for_linear:1;
 337 };
 338
 339 /* Common PS bits between the shader key and the epilog key. */
 340 struct si_ps_epilog_bits {
 341         unsigned        spi_shader_col_format;
 342         unsigned        color_is_int8:8;
 343         unsigned        last_cbuf:3;
 344         unsigned        alpha_func:3;
 345         unsigned        alpha_to_one:1;
 346         unsigned        poly_line_smoothing:1;
 347         unsigned        clamp_color:1;
 348 };
 349
 350 union si_shader_part_key {
 351         struct {
 352                 struct si_vs_prolog_bits states;
 353                 unsigned        num_input_sgprs:5;
 354                 unsigned        last_input:4;
 355         } vs_prolog;
 356         struct {
 357                 struct si_vs_epilog_bits states;
 358                 unsigned        prim_id_param_offset:5;
 359         } vs_epilog;
 360         struct {
 361                 struct si_tcs_epilog_bits states;
 362         } tcs_epilog;
 363         struct {
 364                 struct si_ps_prolog_bits states;
 365                 unsigned        num_input_sgprs:5;
 366                 unsigned        num_input_vgprs:5;
 367                 /* Color interpolation and two-side color selection. */
 368                 unsigned        colors_read:8; /* color input components read */
 369                 unsigned        num_interp_inputs:5; /* BCOLOR is at this location */
 370                 unsigned        face_vgpr_index:5;
 371                 unsigned        wqm:1;
 372                 char            color_attr_index[2];
 373                 char            color_interp_vgpr_index[2]; /* -1 == constant */
 374         } ps_prolog;
 375         struct {
 376                 struct si_ps_epilog_bits states;
 377                 unsigned        colors_written:8;
 378                 unsigned        writes_z:1;
 379                 unsigned        writes_stencil:1;
 380                 unsigned        writes_samplemask:1;
 381         } ps_epilog;
 382 };
 383
 384 union si_shader_key {
 385         struct {
 386                 struct si_ps_prolog_bits prolog;
 387                 struct si_ps_epilog_bits epilog;
 388         } ps;
 389         struct {
 390                 struct si_vs_prolog_bits prolog;
 391                 struct si_vs_epilog_bits epilog;
 392                 unsigned        as_es:1; /* export shader */
 393                 unsigned        as_ls:1; /* local shader */
 394         } vs;
 395         struct {
 396                 struct si_tcs_epilog_bits epilog;
 397         } tcs; /* tessellation control shader */
 398         struct {
 399                 struct si_vs_epilog_bits epilog; /* same as VS */
 400                 unsigned        as_es:1; /* export shader */
 401         } tes; /* tessellation evaluation shader */
 402 };
 403
 404 struct si_shader_config {
 405         unsigned                        num_sgprs;
 406         unsigned                        num_vgprs;
 407         unsigned                        spilled_sgprs;
 408         unsigned                        spilled_vgprs;
 409         unsigned                        lds_size;
 410         unsigned                        spi_ps_input_ena;
 411         unsigned                        spi_ps_input_addr;
 412         unsigned                        float_mode;
 413         unsigned                        scratch_bytes_per_wave;
 414         unsigned                        rsrc1;
 415         unsigned                        rsrc2;
 416 };
 417
 418 enum {
 419         /* SPI_PS_INPUT_CNTL_i.OFFSET[0:4] */
 420         EXP_PARAM_OFFSET_0 = 0,
 421         EXP_PARAM_OFFSET_31 = 31,
 422         /* SPI_PS_INPUT_CNTL_i.DEFAULT_VAL[0:1] */
 423         EXP_PARAM_DEFAULT_VAL_0000 = 64,
 424         EXP_PARAM_DEFAULT_VAL_0001,
 425         EXP_PARAM_DEFAULT_VAL_1110,
 426         EXP_PARAM_DEFAULT_VAL_1111,
 427 };
 428
 429 /* GCN-specific shader info. */
 430 struct si_shader_info {
 431         ubyte                   vs_output_param_offset[SI_MAX_VS_OUTPUTS];
 432         ubyte                   num_input_sgprs;
 433         ubyte                   num_input_vgprs;
 434         char                    face_vgpr_index;
 435         bool                    uses_instanceid;
 436         ubyte                   nr_pos_exports;
 437         ubyte                   nr_param_exports;
 438 };
 439
 440 struct si_shader {
 441         struct si_shader_selector       *selector;
 442         struct si_shader                *next_variant;
 443
 444         struct si_shader_part           *prolog;
 445         struct si_shader_part           *epilog;
 446
 447         struct si_shader                *gs_copy_shader;
 448         struct si_pm4_state             *pm4;
 449         struct r600_resource            *bo;
 450         struct r600_resource            *scratch_bo;
 451         union si_shader_key             key;
 452         bool                            is_binary_shared;
 453
 454         /* The following data is all that's needed for binary shaders. */
 455         struct radeon_shader_binary     binary;
 456         struct si_shader_config         config;
 457         struct si_shader_info           info;
 458
 459         /* Shader key + LLVM IR + disassembly + statistics.
 460          * Generated for debug contexts only.
 461          */
 462         char                            *shader_log;
 463         size_t                          shader_log_size;
 464 };
 465
 466 struct si_shader_part {
 467         struct si_shader_part *next;
 468         union si_shader_part_key key;
 469         struct radeon_shader_binary binary;
 470         struct si_shader_config config;
 471 };
 472
 473 /* si_shader.c */
 474 int si_compile_tgsi_shader(struct si_screen *sscreen,
 475                            LLVMTargetMachineRef tm,
 476                            struct si_shader *shader,
 477                            bool is_monolithic,
 478                            struct pipe_debug_callback *debug);
 479 int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm,
 480                      struct si_shader *shader,
 481                      struct pipe_debug_callback *debug);
 482 int si_compile_llvm(struct si_screen *sscreen,
 483                     struct radeon_shader_binary *binary,
 484                     struct si_shader_config *conf,
 485                     LLVMTargetMachineRef tm,
 486                     LLVMModuleRef mod,
 487                     struct pipe_debug_callback *debug,
 488                     unsigned processor,
 489                     const char *name);
 490 void si_shader_destroy(struct si_shader *shader);
 491 unsigned si_shader_io_get_unique_index(unsigned semantic_name, unsigned index);
 492 int si_shader_binary_upload(struct si_screen *sscreen, struct si_shader *shader);
 493 void si_shader_dump(struct si_screen *sscreen, struct si_shader *shader,
 494                     struct pipe_debug_callback *debug, unsigned processor,
 495                     FILE *f);
 496 void si_shader_apply_scratch_relocs(struct si_context *sctx,
 497                         struct si_shader *shader,
 498                         struct si_shader_config *config,
 499                         uint64_t scratch_va);
 500 void si_shader_binary_read_config(struct radeon_shader_binary *binary,
 501                                   struct si_shader_config *conf,
 502                                   unsigned symbol_offset);
 503 unsigned si_get_spi_shader_z_format(bool writes_z, bool writes_stencil,
 504                                     bool writes_samplemask);
 505
 506 #endif