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 "tgsi/tgsi_scan.h"
  73 #include "si_state.h"
  74
  75 struct radeon_shader_binary;
  76 struct radeon_shader_reloc;
  77
  78 #define SI_SGPR_RW_BUFFERS      0  /* rings (& stream-out, VS only) */
  79 #define SI_SGPR_CONST_BUFFERS   2
  80 #define SI_SGPR_SAMPLERS        4  /* images & sampler states interleaved */
  81 /* TODO: gap */
  82 #define SI_SGPR_VERTEX_BUFFERS  8  /* VS only */
  83 #define SI_SGPR_BASE_VERTEX     10 /* VS only */
  84 #define SI_SGPR_START_INSTANCE  11 /* VS only */
  85 #define SI_SGPR_VS_STATE_BITS   12 /* VS(VS) only */
  86 #define SI_SGPR_LS_OUT_LAYOUT   12 /* VS(LS) only */
  87 #define SI_SGPR_TCS_OUT_OFFSETS 8  /* TCS & TES only */
  88 #define SI_SGPR_TCS_OUT_LAYOUT  9  /* TCS & TES only */
  89 #define SI_SGPR_TCS_IN_LAYOUT   10 /* TCS only */
  90 #define SI_SGPR_ALPHA_REF       8  /* PS only */
  91
  92 #define SI_VS_NUM_USER_SGPR     13 /* API VS */
  93 #define SI_ES_NUM_USER_SGPR     12 /* API VS */
  94 #define SI_LS_NUM_USER_SGPR     13 /* API VS */
  95 #define SI_TCS_NUM_USER_SGPR    11
  96 #define SI_TES_NUM_USER_SGPR    10
  97 #define SI_GS_NUM_USER_SGPR     8
  98 #define SI_GSCOPY_NUM_USER_SGPR 4
  99 #define SI_PS_NUM_USER_SGPR     9
 100
 101 /* LLVM function parameter indices */
 102 #define SI_PARAM_RW_BUFFERS     0
 103 #define SI_PARAM_CONST_BUFFERS  1
 104 #define SI_PARAM_SAMPLERS       2
 105 #define SI_PARAM_UNUSED         3 /* TODO: use */
 106
 107 /* VS only parameters */
 108 #define SI_PARAM_VERTEX_BUFFERS 4
 109 #define SI_PARAM_BASE_VERTEX    5
 110 #define SI_PARAM_START_INSTANCE 6
 111 /* [0] = clamp vertex color */
 112 #define SI_PARAM_VS_STATE_BITS  7
 113 /* the other VS parameters are assigned dynamically */
 114
 115 /* Offsets where TCS outputs and TCS patch outputs live in LDS:
 116  *   [0:15] = TCS output patch0 offset / 16, max = NUM_PATCHES * 32 * 32
 117  *   [16:31] = TCS output patch0 offset for per-patch / 16, max = NUM_PATCHES*32*32* + 32*32
 118  */
 119 #define SI_PARAM_TCS_OUT_OFFSETS 4 /* for TCS & TES */
 120
 121 /* Layout of TCS outputs / TES inputs:
 122  *   [0:12] = stride between output patches in dwords, num_outputs * num_vertices * 4, max = 32*32*4
 123  *   [13:20] = stride between output vertices in dwords = num_inputs * 4, max = 32*4
 124  *   [26:31] = gl_PatchVerticesIn, max = 32
 125  */
 126 #define SI_PARAM_TCS_OUT_LAYOUT 5 /* for TCS & TES */
 127
 128 /* Layout of LS outputs / TCS inputs
 129  *   [0:12] = stride between patches in dwords = num_inputs * num_vertices * 4, max = 32*32*4
 130  *   [13:20] = stride between vertices in dwords = num_inputs * 4, max = 32*4
 131  */
 132 #define SI_PARAM_TCS_IN_LAYOUT  6 /* TCS only */
 133 #define SI_PARAM_LS_OUT_LAYOUT  7 /* same value as TCS_IN_LAYOUT, LS only */
 134
 135 /* TCS only parameters. */
 136 #define SI_PARAM_TESS_FACTOR_OFFSET 7
 137 #define SI_PARAM_PATCH_ID       8
 138 #define SI_PARAM_REL_IDS        9
 139
 140 /* GS only parameters */
 141 #define SI_PARAM_GS2VS_OFFSET   4
 142 #define SI_PARAM_GS_WAVE_ID     5
 143 #define SI_PARAM_VTX0_OFFSET    6
 144 #define SI_PARAM_VTX1_OFFSET    7
 145 #define SI_PARAM_PRIMITIVE_ID   8
 146 #define SI_PARAM_VTX2_OFFSET    9
 147 #define SI_PARAM_VTX3_OFFSET    10
 148 #define SI_PARAM_VTX4_OFFSET    11
 149 #define SI_PARAM_VTX5_OFFSET    12
 150 #define SI_PARAM_GS_INSTANCE_ID 13
 151
 152 /* PS only parameters */
 153 #define SI_PARAM_ALPHA_REF              4
 154 #define SI_PARAM_PRIM_MASK              5
 155 #define SI_PARAM_PERSP_SAMPLE           6
 156 #define SI_PARAM_PERSP_CENTER           7
 157 #define SI_PARAM_PERSP_CENTROID         8
 158 #define SI_PARAM_PERSP_PULL_MODEL       9
 159 #define SI_PARAM_LINEAR_SAMPLE          10
 160 #define SI_PARAM_LINEAR_CENTER          11
 161 #define SI_PARAM_LINEAR_CENTROID        12
 162 #define SI_PARAM_LINE_STIPPLE_TEX       13
 163 #define SI_PARAM_POS_X_FLOAT            14
 164 #define SI_PARAM_POS_Y_FLOAT            15
 165 #define SI_PARAM_POS_Z_FLOAT            16
 166 #define SI_PARAM_POS_W_FLOAT            17
 167 #define SI_PARAM_FRONT_FACE             18
 168 #define SI_PARAM_ANCILLARY              19
 169 #define SI_PARAM_SAMPLE_COVERAGE        20
 170 #define SI_PARAM_POS_FIXED_PT           21
 171
 172 #define SI_NUM_PARAMS (SI_PARAM_POS_FIXED_PT + 1)
 173
 174 struct si_shader;
 175
 176 /* A shader selector is a gallium CSO and contains shader variants and
 177  * binaries for one TGSI program. This can be shared by multiple contexts.
 178  */
 179 struct si_shader_selector {
 180         pipe_mutex              mutex;
 181         struct si_shader        *first_variant; /* immutable after the first variant */
 182         struct si_shader        *last_variant; /* mutable */
 183
 184         struct tgsi_token       *tokens;
 185         struct pipe_stream_output_info  so;
 186         struct tgsi_shader_info         info;
 187
 188         /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
 189         unsigned        type;
 190
 191         /* GS parameters. */
 192         unsigned        esgs_itemsize;
 193         unsigned        gs_input_verts_per_prim;
 194         unsigned        gs_output_prim;
 195         unsigned        gs_max_out_vertices;
 196         unsigned        gs_num_invocations;
 197         unsigned        max_gs_stream; /* count - 1 */
 198         unsigned        gsvs_vertex_size;
 199         unsigned        max_gsvs_emit_size;
 200
 201         /* PS parameters. */
 202         unsigned        db_shader_control;
 203         /* Set 0xf or 0x0 (4 bits) per each written output.
 204          * ANDed with spi_shader_col_format.
 205          */
 206         unsigned        colors_written_4bit;
 207
 208         /* masks of "get_unique_index" bits */
 209         uint64_t        outputs_written;
 210         uint32_t        patch_outputs_written;
 211 };
 212
 213 /* Valid shader configurations:
 214  *
 215  * API shaders       VS | TCS | TES | GS |pass| PS
 216  * are compiled as:     |     |     |    |thru|
 217  *                      |     |     |    |    |
 218  * Only VS & PS:     VS | --  | --  | -- | -- | PS
 219  * With GS:          ES | --  | --  | GS | VS | PS
 220  * With Tessel.:     LS | HS  | VS  | -- | -- | PS
 221  * With both:        LS | HS  | ES  | GS | VS | PS
 222  */
 223
 224 /* Common VS bits between the shader key and the prolog key. */
 225 struct si_vs_prolog_bits {
 226         unsigned        instance_divisors[SI_NUM_VERTEX_BUFFERS];
 227 };
 228
 229 /* Common VS bits between the shader key and the epilog key. */
 230 struct si_vs_epilog_bits {
 231         unsigned        export_prim_id:1; /* when PS needs it and GS is disabled */
 232         /* TODO:
 233          * - skip clipdist, culldist (including clipvertex code) exports based
 234          *   on which clip_plane_enable bits are set
 235          * - skip layer, viewport, clipdist, and culldist parameter exports
 236          *   if PS doesn't read them
 237          */
 238 };
 239
 240 /* Common TCS bits between the shader key and the epilog key. */
 241 struct si_tcs_epilog_bits {
 242         unsigned        prim_mode:3;
 243 };
 244
 245 /* Common PS bits between the shader key and the prolog key. */
 246 struct si_ps_prolog_bits {
 247         unsigned        color_two_side:1;
 248         /* TODO: add a flatshade bit that skips interpolation for colors */
 249         unsigned        poly_stipple:1;
 250         unsigned        force_persample_interp:1;
 251         /* TODO:
 252          * - add force_center_interp if MSAA is disabled and centroid or
 253          *   sample are present
 254          * - add force_center_interp_bc_optimize to force center interpolation
 255          *   based on the bc_optimize SGPR bit if MSAA is enabled, centroid is
 256          *   present and sample isn't present.
 257          */
 258 };
 259
 260 /* Common PS bits between the shader key and the epilog key. */
 261 struct si_ps_epilog_bits {
 262         unsigned        spi_shader_col_format;
 263         unsigned        color_is_int8:8;
 264         unsigned        last_cbuf:3;
 265         unsigned        alpha_func:3;
 266         unsigned        alpha_to_one:1;
 267         unsigned        poly_line_smoothing:1;
 268         unsigned        clamp_color:1;
 269 };
 270
 271 union si_shader_part_key {
 272         struct {
 273                 struct si_vs_prolog_bits states;
 274                 unsigned        num_input_sgprs:5;
 275                 unsigned        last_input:4;
 276         } vs_prolog;
 277         struct {
 278                 struct si_vs_epilog_bits states;
 279                 unsigned        prim_id_param_offset:5;
 280         } vs_epilog;
 281 };
 282
 283 union si_shader_key {
 284         struct {
 285                 struct si_ps_prolog_bits prolog;
 286                 struct si_ps_epilog_bits epilog;
 287         } ps;
 288         struct {
 289                 struct si_vs_prolog_bits prolog;
 290                 struct si_vs_epilog_bits epilog;
 291                 unsigned        as_es:1; /* export shader */
 292                 unsigned        as_ls:1; /* local shader */
 293         } vs;
 294         struct {
 295                 struct si_tcs_epilog_bits epilog;
 296         } tcs; /* tessellation control shader */
 297         struct {
 298                 struct si_vs_epilog_bits epilog; /* same as VS */
 299                 unsigned        as_es:1; /* export shader */
 300         } tes; /* tessellation evaluation shader */
 301 };
 302
 303 struct si_shader_config {
 304         unsigned                        num_sgprs;
 305         unsigned                        num_vgprs;
 306         unsigned                        lds_size;
 307         unsigned                        spi_ps_input_ena;
 308         unsigned                        spi_ps_input_addr;
 309         unsigned                        float_mode;
 310         unsigned                        scratch_bytes_per_wave;
 311         unsigned                        rsrc1;
 312         unsigned                        rsrc2;
 313 };
 314
 315 struct si_shader {
 316         struct si_shader_selector       *selector;
 317         struct si_shader                *next_variant;
 318
 319         struct si_shader_part           *prolog;
 320         struct si_shader_part           *epilog;
 321
 322         struct si_shader                *gs_copy_shader;
 323         struct si_pm4_state             *pm4;
 324         struct r600_resource            *bo;
 325         struct r600_resource            *scratch_bo;
 326         union si_shader_key             key;
 327         struct radeon_shader_binary     binary;
 328         struct si_shader_config         config;
 329
 330         ubyte                   num_input_sgprs;
 331         ubyte                   num_input_vgprs;
 332         char                    face_vgpr_index;
 333
 334         unsigned                vs_output_param_offset[PIPE_MAX_SHADER_OUTPUTS];
 335         bool                    uses_instanceid;
 336         unsigned                nr_pos_exports;
 337         unsigned                nr_param_exports;
 338 };
 339
 340 struct si_shader_part {
 341         struct si_shader_part *next;
 342         union si_shader_part_key key;
 343         struct radeon_shader_binary binary;
 344         struct si_shader_config config;
 345 };
 346
 347 static inline struct tgsi_shader_info *si_get_vs_info(struct si_context *sctx)
 348 {
 349         if (sctx->gs_shader.cso)
 350                 return &sctx->gs_shader.cso->info;
 351         else if (sctx->tes_shader.cso)
 352                 return &sctx->tes_shader.cso->info;
 353         else if (sctx->vs_shader.cso)
 354                 return &sctx->vs_shader.cso->info;
 355         else
 356                 return NULL;
 357 }
 358
 359 static inline struct si_shader* si_get_vs_state(struct si_context *sctx)
 360 {
 361         if (sctx->gs_shader.current)
 362                 return sctx->gs_shader.current->gs_copy_shader;
 363         else if (sctx->tes_shader.current)
 364                 return sctx->tes_shader.current;
 365         else
 366                 return sctx->vs_shader.current;
 367 }
 368
 369 static inline bool si_vs_exports_prim_id(struct si_shader *shader)
 370 {
 371         if (shader->selector->type == PIPE_SHADER_VERTEX)
 372                 return shader->key.vs.epilog.export_prim_id;
 373         else if (shader->selector->type == PIPE_SHADER_TESS_EVAL)
 374                 return shader->key.tes.epilog.export_prim_id;
 375         else
 376                 return false;
 377 }
 378
 379 /* si_shader.c */
 380 int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm,
 381                      struct si_shader *shader,
 382                      struct pipe_debug_callback *debug);
 383 void si_dump_shader_key(unsigned shader, union si_shader_key *key, FILE *f);
 384 int si_compile_llvm(struct si_screen *sscreen,
 385                     struct radeon_shader_binary *binary,
 386                     struct si_shader_config *conf,
 387                     LLVMTargetMachineRef tm,
 388                     LLVMModuleRef mod,
 389                     struct pipe_debug_callback *debug,
 390                     unsigned processor,
 391                     const char *name);
 392 void si_shader_destroy(struct si_shader *shader);
 393 unsigned si_shader_io_get_unique_index(unsigned semantic_name, unsigned index);
 394 int si_shader_binary_upload(struct si_screen *sscreen, struct si_shader *shader);
 395 void si_shader_dump(struct si_screen *sscreen, struct si_shader *shader,
 396                     struct pipe_debug_callback *debug, unsigned processor);
 397 void si_shader_apply_scratch_relocs(struct si_context *sctx,
 398                         struct si_shader *shader,
 399                         uint64_t scratch_va);
 400 void si_shader_binary_read_config(struct radeon_shader_binary *binary,
 401                                   struct si_shader_config *conf,
 402                                   unsigned symbol_offset);
 403
 404 #endif