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
  24 /* The compiler middle-end architecture: Explaining (non-)monolithic shaders
  25  * -------------------------------------------------------------------------
  26  *
  27  * Typically, there is one-to-one correspondence between API and HW shaders,
  28  * that is, for every API shader, there is exactly one shader binary in
  29  * the driver.
  30  *
  31  * The problem with that is that we also have to emulate some API states
  32  * (e.g. alpha-test, and many others) in shaders too. The two obvious ways
  33  * to deal with it are:
  34  * - each shader has multiple variants for each combination of emulated states,
  35  *   and the variants are compiled on demand, possibly relying on a shader
  36  *   cache for good performance
  37  * - patch shaders at the binary level
  38  *
  39  * This driver uses something completely different. The emulated states are
  40  * usually implemented at the beginning or end of shaders. Therefore, we can
  41  * split the shader into 3 parts:
  42  * - prolog part (shader code dependent on states)
  43  * - main part (the API shader)
  44  * - epilog part (shader code dependent on states)
  45  *
  46  * Each part is compiled as a separate shader and the final binaries are
  47  * concatenated. This type of shader is called non-monolithic, because it
  48  * consists of multiple independent binaries. Creating a new shader variant
  49  * is therefore only a concatenation of shader parts (binaries) and doesn't
  50  * involve any compilation. The main shader parts are the only parts that are
  51  * compiled when applications create shader objects. The prolog and epilog
  52  * parts are compiled on the first use and saved, so that their binaries can
  53  * be reused by many other shaders.
  54  *
  55  * One of the roles of the prolog part is to compute vertex buffer addresses
  56  * for vertex shaders. A few of the roles of the epilog part are color buffer
  57  * format conversions in pixel shaders that we have to do manually, and write
  58  * tessellation factors in tessellation control shaders. The prolog and epilog
  59  * have many other important responsibilities in various shader stages.
  60  * They don't just "emulate legacy stuff".
  61  *
  62  * Monolithic shaders are shaders where the parts are combined before LLVM
  63  * compilation, and the whole thing is compiled and optimized as one unit with
  64  * one binary on the output. The result is the same as the non-monolithic
  65  * shader, but the final code can be better, because LLVM can optimize across
  66  * all shader parts. Monolithic shaders aren't usually used except for these
  67  * special cases:
  68  *
  69  * 1) Some rarely-used states require modification of the main shader part
  70  *    itself, and in such cases, only the monolithic shader variant is
  71  *    compiled, and that's always done on the first use.
  72  *
  73  * 2) When we do cross-stage optimizations for separate shader objects and
  74  *    e.g. eliminate unused shader varyings, the resulting optimized shader
  75  *    variants are always compiled as monolithic shaders, and always
  76  *    asynchronously (i.e. not stalling ongoing rendering). We call them
  77  *    "optimized monolithic" shaders. The important property here is that
  78  *    the non-monolithic unoptimized shader variant is always available for use
  79  *    when the asynchronous compilation of the optimized shader is not done
  80  *    yet.
  81  *
  82  * Starting with GFX9 chips, some shader stages are merged, and the number of
  83  * shader parts per shader increased. The complete new list of shader parts is:
  84  * - 1st shader: prolog part
  85  * - 1st shader: main part
  86  * - 2nd shader: prolog part
  87  * - 2nd shader: main part
  88  * - 2nd shader: epilog part
  89  */
  90
  91 /* How linking shader inputs and outputs between vertex, tessellation, and
  92  * geometry shaders works.
  93  *
  94  * Inputs and outputs between shaders are stored in a buffer. This buffer
  95  * lives in LDS (typical case for tessellation), but it can also live
  96  * in memory (ESGS). Each input or output has a fixed location within a vertex.
  97  * The highest used input or output determines the stride between vertices.
  98  *
  99  * Since GS and tessellation are only possible in the OpenGL core profile,
 100  * only these semantics are valid for per-vertex data:
 101  *
 102  *   Name             Location
 103  *
 104  *   POSITION         0
 105  *   PSIZE            1
 106  *   CLIPDIST0..1     2..3
 107  *   CULLDIST0..1     (not implemented)
 108  *   GENERIC0..31     4..35
 109  *
 110  * For example, a shader only writing GENERIC0 has the output stride of 5.
 111  *
 112  * Only these semantics are valid for per-patch data:
 113  *
 114  *   Name             Location
 115  *
 116  *   TESSOUTER        0
 117  *   TESSINNER        1
 118  *   PATCH0..29       2..31
 119  *
 120  * That's how independent shaders agree on input and output locations.
 121  * The si_shader_io_get_unique_index function assigns the locations.
 122  *
 123  * For tessellation, other required information for calculating the input and
 124  * output addresses like the vertex stride, the patch stride, and the offsets
 125  * where per-vertex and per-patch data start, is passed to the shader via
 126  * user data SGPRs. The offsets and strides are calculated at draw time and
 127  * aren't available at compile time.
 128  */
 129
 130 #ifndef SI_SHADER_H
 131 #define SI_SHADER_H
 132
 133 #include <llvm-c/Core.h> /* LLVMModuleRef */
 134 #include <llvm-c/TargetMachine.h>
 135 #include "tgsi/tgsi_scan.h"
 136 #include "util/u_queue.h"
 137
 138 #include "ac_binary.h"
 139 #include "si_state.h"
 140
 141 struct nir_shader;
 142
 143 #define SI_MAX_VS_OUTPUTS       40
 144
 145 /* Shader IO unique indices are supported for TGSI_SEMANTIC_GENERIC with an
 146  * index smaller than this.
 147  */
 148 #define SI_MAX_IO_GENERIC       46
 149
 150 /* SGPR user data indices */
 151 enum {
 152         SI_SGPR_RW_BUFFERS,  /* rings (& stream-out, VS only) */
 153         SI_SGPR_RW_BUFFERS_HI,
 154         SI_SGPR_BINDLESS_SAMPLERS_AND_IMAGES,
 155         SI_SGPR_BINDLESS_SAMPLERS_AND_IMAGES_HI,
 156         SI_SGPR_CONST_AND_SHADER_BUFFERS, /* or just a constant buffer 0 pointer */
 157         SI_SGPR_CONST_AND_SHADER_BUFFERS_HI,
 158         SI_SGPR_SAMPLERS_AND_IMAGES,
 159         SI_SGPR_SAMPLERS_AND_IMAGES_HI,
 160         SI_NUM_RESOURCE_SGPRS,
 161
 162         /* all VS variants */
 163         SI_SGPR_VERTEX_BUFFERS  = SI_NUM_RESOURCE_SGPRS,
 164         SI_SGPR_VERTEX_BUFFERS_HI,
 165         SI_SGPR_BASE_VERTEX,
 166         SI_SGPR_START_INSTANCE,
 167         SI_SGPR_DRAWID,
 168         SI_SGPR_VS_STATE_BITS,
 169         SI_VS_NUM_USER_SGPR,
 170
 171         SI_SGPR_VS_BLIT_DATA = SI_SGPR_CONST_AND_SHADER_BUFFERS,
 172
 173         /* TES */
 174         SI_SGPR_TES_OFFCHIP_LAYOUT = SI_NUM_RESOURCE_SGPRS,
 175         SI_SGPR_TES_OFFCHIP_ADDR_BASE64K,
 176         SI_TES_NUM_USER_SGPR,
 177
 178         /* GFX6-8: TCS only */
 179         GFX6_SGPR_TCS_OFFCHIP_LAYOUT = SI_NUM_RESOURCE_SGPRS,
 180         GFX6_SGPR_TCS_OUT_OFFSETS,
 181         GFX6_SGPR_TCS_OUT_LAYOUT,
 182         GFX6_SGPR_TCS_IN_LAYOUT,
 183         GFX6_SGPR_TCS_OFFCHIP_ADDR_BASE64K,
 184         GFX6_SGPR_TCS_FACTOR_ADDR_BASE64K,
 185         GFX6_TCS_NUM_USER_SGPR,
 186
 187         /* GFX9: Merged LS-HS (VS-TCS) only. */
 188         GFX9_SGPR_TCS_OFFCHIP_LAYOUT = SI_VS_NUM_USER_SGPR,
 189         GFX9_SGPR_TCS_OUT_OFFSETS,
 190         GFX9_SGPR_TCS_OUT_LAYOUT,
 191         GFX9_SGPR_TCS_OFFCHIP_ADDR_BASE64K,
 192         GFX9_SGPR_TCS_FACTOR_ADDR_BASE64K,
 193         GFX9_SGPR_unused_to_align_the_next_pointer,
 194         GFX9_SGPR_TCS_CONST_AND_SHADER_BUFFERS,
 195         GFX9_SGPR_TCS_CONST_AND_SHADER_BUFFERS_HI,
 196         GFX9_SGPR_TCS_SAMPLERS_AND_IMAGES,
 197         GFX9_SGPR_TCS_SAMPLERS_AND_IMAGES_HI,
 198         GFX9_TCS_NUM_USER_SGPR,
 199
 200         /* GFX9: Merged ES-GS (VS-GS or TES-GS). */
 201         GFX9_SGPR_GS_CONST_AND_SHADER_BUFFERS = SI_VS_NUM_USER_SGPR,
 202         GFX9_SGPR_GS_CONST_AND_SHADER_BUFFERS_HI,
 203         GFX9_SGPR_GS_SAMPLERS_AND_IMAGES,
 204         GFX9_SGPR_GS_SAMPLERS_AND_IMAGES_HI,
 205         GFX9_GS_NUM_USER_SGPR,
 206
 207         /* GS limits */
 208         GFX6_GS_NUM_USER_SGPR = SI_NUM_RESOURCE_SGPRS,
 209         SI_GSCOPY_NUM_USER_SGPR = SI_SGPR_RW_BUFFERS_HI + 1,
 210
 211         /* PS only */
 212         SI_SGPR_ALPHA_REF       = SI_NUM_RESOURCE_SGPRS,
 213         SI_PS_NUM_USER_SGPR,
 214 };
 215
 216 /* LLVM function parameter indices */
 217 enum {
 218         SI_NUM_RESOURCE_PARAMS = 4,
 219
 220         /* PS only parameters */
 221         SI_PARAM_ALPHA_REF = SI_NUM_RESOURCE_PARAMS,
 222         SI_PARAM_PRIM_MASK,
 223         SI_PARAM_PERSP_SAMPLE,
 224         SI_PARAM_PERSP_CENTER,
 225         SI_PARAM_PERSP_CENTROID,
 226         SI_PARAM_PERSP_PULL_MODEL,
 227         SI_PARAM_LINEAR_SAMPLE,
 228         SI_PARAM_LINEAR_CENTER,
 229         SI_PARAM_LINEAR_CENTROID,
 230         SI_PARAM_LINE_STIPPLE_TEX,
 231         SI_PARAM_POS_X_FLOAT,
 232         SI_PARAM_POS_Y_FLOAT,
 233         SI_PARAM_POS_Z_FLOAT,
 234         SI_PARAM_POS_W_FLOAT,
 235         SI_PARAM_FRONT_FACE,
 236         SI_PARAM_ANCILLARY,
 237         SI_PARAM_SAMPLE_COVERAGE,
 238         SI_PARAM_POS_FIXED_PT,
 239
 240         SI_NUM_PARAMS = SI_PARAM_POS_FIXED_PT + 9, /* +8 for COLOR[0..1] */
 241 };
 242
 243 /* Fields of driver-defined VS state SGPR. */
 244 /* Clamp vertex color output (only used in VS as VS). */
 245 #define S_VS_STATE_CLAMP_VERTEX_COLOR(x)        (((unsigned)(x) & 0x1) << 0)
 246 #define C_VS_STATE_CLAMP_VERTEX_COLOR           0xFFFFFFFE
 247 #define S_VS_STATE_INDEXED(x)                   (((unsigned)(x) & 0x1) << 1)
 248 #define C_VS_STATE_INDEXED                      0xFFFFFFFD
 249 #define S_VS_STATE_LS_OUT_PATCH_SIZE(x)         (((unsigned)(x) & 0x1FFF) << 8)
 250 #define C_VS_STATE_LS_OUT_PATCH_SIZE            0xFFE000FF
 251 #define S_VS_STATE_LS_OUT_VERTEX_SIZE(x)        (((unsigned)(x) & 0xFF) << 24)
 252 #define C_VS_STATE_LS_OUT_VERTEX_SIZE           0x00FFFFFF
 253
 254 /* SI-specific system values. */
 255 enum {
 256         TGSI_SEMANTIC_DEFAULT_TESSOUTER_SI = TGSI_SEMANTIC_COUNT,
 257         TGSI_SEMANTIC_DEFAULT_TESSINNER_SI,
 258 };
 259
 260 enum {
 261         /* Use a property enum that VS wouldn't use. */
 262         TGSI_PROPERTY_VS_BLIT_SGPRS = TGSI_PROPERTY_FS_COORD_ORIGIN,
 263
 264         /* These represent the number of SGPRs the shader uses. */
 265         SI_VS_BLIT_SGPRS_POS = 3,
 266         SI_VS_BLIT_SGPRS_POS_COLOR = 7,
 267         SI_VS_BLIT_SGPRS_POS_TEXCOORD = 9,
 268 };
 269
 270 /* For VS shader key fix_fetch. */
 271 enum {
 272         SI_FIX_FETCH_NONE = 0,
 273         SI_FIX_FETCH_A2_SNORM,
 274         SI_FIX_FETCH_A2_SSCALED,
 275         SI_FIX_FETCH_A2_SINT,
 276         SI_FIX_FETCH_RGBA_32_UNORM,
 277         SI_FIX_FETCH_RGBX_32_UNORM,
 278         SI_FIX_FETCH_RGBA_32_SNORM,
 279         SI_FIX_FETCH_RGBX_32_SNORM,
 280         SI_FIX_FETCH_RGBA_32_USCALED,
 281         SI_FIX_FETCH_RGBA_32_SSCALED,
 282         SI_FIX_FETCH_RGBA_32_FIXED,
 283         SI_FIX_FETCH_RGBX_32_FIXED,
 284         SI_FIX_FETCH_RG_64_FLOAT,
 285         SI_FIX_FETCH_RGB_64_FLOAT,
 286         SI_FIX_FETCH_RGBA_64_FLOAT,
 287         SI_FIX_FETCH_RGB_8,     /* A = 1.0 */
 288         SI_FIX_FETCH_RGB_8_INT, /* A = 1 */
 289         SI_FIX_FETCH_RGB_16,
 290         SI_FIX_FETCH_RGB_16_INT,
 291 };
 292
 293 struct si_shader;
 294
 295 /* State of the context creating the shader object. */
 296 struct si_compiler_ctx_state {
 297         /* Should only be used by si_init_shader_selector_async and
 298          * si_build_shader_variant if thread_index == -1 (non-threaded). */
 299         LLVMTargetMachineRef            tm;
 300
 301         /* Used if thread_index == -1 or if debug.async is true. */
 302         struct pipe_debug_callback      debug;
 303
 304         /* Used for creating the log string for gallium/ddebug. */
 305         bool                            is_debug_context;
 306 };
 307
 308 /* A shader selector is a gallium CSO and contains shader variants and
 309  * binaries for one TGSI program. This can be shared by multiple contexts.
 310  */
 311 struct si_shader_selector {
 312         struct pipe_reference   reference;
 313         struct si_screen        *screen;
 314         struct util_queue_fence ready;
 315         struct si_compiler_ctx_state compiler_ctx_state;
 316
 317         mtx_t           mutex;
 318         struct si_shader        *first_variant; /* immutable after the first variant */
 319         struct si_shader        *last_variant; /* mutable */
 320
 321         /* The compiled TGSI shader expecting a prolog and/or epilog (not
 322          * uploaded to a buffer).
 323          */
 324         struct si_shader        *main_shader_part;
 325         struct si_shader        *main_shader_part_ls; /* as_ls is set in the key */
 326         struct si_shader        *main_shader_part_es; /* as_es is set in the key */
 327
 328         struct si_shader        *gs_copy_shader;
 329
 330         struct tgsi_token       *tokens;
 331         struct nir_shader       *nir;
 332         struct pipe_stream_output_info  so;
 333         struct tgsi_shader_info         info;
 334         struct tgsi_tessctrl_info       tcs_info;
 335
 336         /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
 337         unsigned        type;
 338         bool            vs_needs_prolog;
 339         bool            force_correct_derivs_after_kill;
 340         unsigned        pa_cl_vs_out_cntl;
 341         ubyte           clipdist_mask;
 342         ubyte           culldist_mask;
 343
 344         /* GS parameters. */
 345         unsigned        esgs_itemsize;
 346         unsigned        gs_input_verts_per_prim;
 347         unsigned        gs_output_prim;
 348         unsigned        gs_max_out_vertices;
 349         unsigned        gs_num_invocations;
 350         unsigned        max_gs_stream; /* count - 1 */
 351         unsigned        gsvs_vertex_size;
 352         unsigned        max_gsvs_emit_size;
 353         unsigned        enabled_streamout_buffer_mask;
 354
 355         /* PS parameters. */
 356         unsigned        color_attr_index[2];
 357         unsigned        db_shader_control;
 358         /* Set 0xf or 0x0 (4 bits) per each written output.
 359          * ANDed with spi_shader_col_format.
 360          */
 361         unsigned        colors_written_4bit;
 362
 363         /* CS parameters */
 364         unsigned local_size;
 365
 366         uint64_t        outputs_written;        /* "get_unique_index" bits */
 367         uint32_t        patch_outputs_written;  /* "get_unique_index_patch" bits */
 368
 369         uint64_t        inputs_read;            /* "get_unique_index" bits */
 370
 371         /* bitmasks of used descriptor slots */
 372         uint32_t        active_const_and_shader_buffers;
 373         uint64_t        active_samplers_and_images;
 374 };
 375
 376 /* Valid shader configurations:
 377  *
 378  * API shaders       VS | TCS | TES | GS |pass| PS
 379  * are compiled as:     |     |     |    |thru|
 380  *                      |     |     |    |    |
 381  * Only VS & PS:     VS |     |     |    |    | PS
 382  * GFX6 - with GS:   ES |     |     | GS | VS | PS
 383  *      - with tess: LS | HS  | VS  |    |    | PS
 384  *      - with both: LS | HS  | ES  | GS | VS | PS
 385  * GFX9 - with GS:   -> |     |     | GS | VS | PS
 386  *      - with tess: -> | HS  | VS  |    |    | PS
 387  *      - with both: -> | HS  | ->  | GS | VS | PS
 388  *
 389  * -> = merged with the next stage
 390  */
 391
 392 /* Use the byte alignment for all following structure members for optimal
 393  * shader key memory footprint.
 394  */
 395 #pragma pack(push, 1)
 396
 397 /* Common VS bits between the shader key and the prolog key. */
 398 struct si_vs_prolog_bits {
 399         /* - If neither "is_one" nor "is_fetched" has a bit set, the instance
 400          *   divisor is 0.
 401          * - If "is_one" has a bit set, the instance divisor is 1.
 402          * - If "is_fetched" has a bit set, the instance divisor will be loaded
 403          *   from the constant buffer.
 404          */
 405         uint16_t        instance_divisor_is_one;     /* bitmask of inputs */
 406         uint16_t        instance_divisor_is_fetched; /* bitmask of inputs */
 407         unsigned        ls_vgpr_fix:1;
 408 };
 409
 410 /* Common TCS bits between the shader key and the epilog key. */
 411 struct si_tcs_epilog_bits {
 412         unsigned        prim_mode:3;
 413         unsigned        invoc0_tess_factors_are_def:1;
 414         unsigned        tes_reads_tess_factors:1;
 415 };
 416
 417 struct si_gs_prolog_bits {
 418         unsigned        tri_strip_adj_fix:1;
 419 };
 420
 421 /* Common PS bits between the shader key and the prolog key. */
 422 struct si_ps_prolog_bits {
 423         unsigned        color_two_side:1;
 424         unsigned        flatshade_colors:1;
 425         unsigned        poly_stipple:1;
 426         unsigned        force_persp_sample_interp:1;
 427         unsigned        force_linear_sample_interp:1;
 428         unsigned        force_persp_center_interp:1;
 429         unsigned        force_linear_center_interp:1;
 430         unsigned        bc_optimize_for_persp:1;
 431         unsigned        bc_optimize_for_linear:1;
 432         unsigned        samplemask_log_ps_iter:3;
 433 };
 434
 435 /* Common PS bits between the shader key and the epilog key. */
 436 struct si_ps_epilog_bits {
 437         unsigned        spi_shader_col_format;
 438         unsigned        color_is_int8:8;
 439         unsigned        color_is_int10:8;
 440         unsigned        last_cbuf:3;
 441         unsigned        alpha_func:3;
 442         unsigned        alpha_to_one:1;
 443         unsigned        poly_line_smoothing:1;
 444         unsigned        clamp_color:1;
 445 };
 446
 447 union si_shader_part_key {
 448         struct {
 449                 struct si_vs_prolog_bits states;
 450                 unsigned        num_input_sgprs:6;
 451                 /* For merged stages such as LS-HS, HS input VGPRs are first. */
 452                 unsigned        num_merged_next_stage_vgprs:3;
 453                 unsigned        last_input:4;
 454                 unsigned        as_ls:1;
 455                 unsigned        as_es:1;
 456                 /* Prologs for monolithic shaders shouldn't set EXEC. */
 457                 unsigned        is_monolithic:1;
 458         } vs_prolog;
 459         struct {
 460                 struct si_tcs_epilog_bits states;
 461         } tcs_epilog;
 462         struct {
 463                 struct si_gs_prolog_bits states;
 464                 /* Prologs of monolithic shaders shouldn't set EXEC. */
 465                 unsigned        is_monolithic:1;
 466         } gs_prolog;
 467         struct {
 468                 struct si_ps_prolog_bits states;
 469                 unsigned        num_input_sgprs:6;
 470                 unsigned        num_input_vgprs:5;
 471                 /* Color interpolation and two-side color selection. */
 472                 unsigned        colors_read:8; /* color input components read */
 473                 unsigned        num_interp_inputs:5; /* BCOLOR is at this location */
 474                 unsigned        face_vgpr_index:5;
 475                 unsigned        ancillary_vgpr_index:5;
 476                 unsigned        wqm:1;
 477                 char            color_attr_index[2];
 478                 char            color_interp_vgpr_index[2]; /* -1 == constant */
 479         } ps_prolog;
 480         struct {
 481                 struct si_ps_epilog_bits states;
 482                 unsigned        colors_written:8;
 483                 unsigned        writes_z:1;
 484                 unsigned        writes_stencil:1;
 485                 unsigned        writes_samplemask:1;
 486         } ps_epilog;
 487 };
 488
 489 struct si_shader_key {
 490         /* Prolog and epilog flags. */
 491         union {
 492                 struct {
 493                         struct si_vs_prolog_bits prolog;
 494                 } vs;
 495                 struct {
 496                         struct si_vs_prolog_bits ls_prolog; /* for merged LS-HS */
 497                         struct si_shader_selector *ls;   /* for merged LS-HS */
 498                         struct si_tcs_epilog_bits epilog;
 499                 } tcs; /* tessellation control shader */
 500                 struct {
 501                         struct si_vs_prolog_bits vs_prolog; /* for merged ES-GS */
 502                         struct si_shader_selector *es;   /* for merged ES-GS */
 503                         struct si_gs_prolog_bits prolog;
 504                 } gs;
 505                 struct {
 506                         struct si_ps_prolog_bits prolog;
 507                         struct si_ps_epilog_bits epilog;
 508                 } ps;
 509         } part;
 510
 511         /* These two are initially set according to the NEXT_SHADER property,
 512          * or guessed if the property doesn't seem correct.
 513          */
 514         unsigned as_es:1; /* export shader, which precedes GS */
 515         unsigned as_ls:1; /* local shader, which precedes TCS */
 516
 517         /* Flags for monolithic compilation only. */
 518         struct {
 519                 /* One byte for every input: SI_FIX_FETCH_* enums. */
 520                 uint8_t         vs_fix_fetch[SI_MAX_ATTRIBS];
 521
 522                 union {
 523                         uint64_t        ff_tcs_inputs_to_copy; /* for fixed-func TCS */
 524                         /* When PS needs PrimID and GS is disabled. */
 525                         unsigned        vs_export_prim_id:1;
 526                         struct {
 527                                 unsigned interpolate_at_sample_force_center:1;
 528                         } ps;
 529                 } u;
 530         } mono;
 531
 532         /* Optimization flags for asynchronous compilation only. */
 533         struct {
 534                 /* For HW VS (it can be VS, TES, GS) */
 535                 uint64_t        kill_outputs; /* "get_unique_index" bits */
 536                 unsigned        clip_disable:1;
 537
 538                 /* For shaders where monolithic variants have better code.
 539                  *
 540                  * This is a flag that has no effect on code generation,
 541                  * but forces monolithic shaders to be used as soon as
 542                  * possible, because it's in the "opt" group.
 543                  */
 544                 unsigned        prefer_mono:1;
 545         } opt;
 546 };
 547
 548 /* Restore the pack alignment to default. */
 549 #pragma pack(pop)
 550
 551 struct si_shader_config {
 552         unsigned                        num_sgprs;
 553         unsigned                        num_vgprs;
 554         unsigned                        spilled_sgprs;
 555         unsigned                        spilled_vgprs;
 556         unsigned                        private_mem_vgprs;
 557         unsigned                        lds_size;
 558         unsigned                        spi_ps_input_ena;
 559         unsigned                        spi_ps_input_addr;
 560         unsigned                        float_mode;
 561         unsigned                        scratch_bytes_per_wave;
 562         unsigned                        rsrc1;
 563         unsigned                        rsrc2;
 564 };
 565
 566 /* GCN-specific shader info. */
 567 struct si_shader_info {
 568         ubyte                   vs_output_param_offset[SI_MAX_VS_OUTPUTS];
 569         ubyte                   num_input_sgprs;
 570         ubyte                   num_input_vgprs;
 571         signed char             face_vgpr_index;
 572         signed char             ancillary_vgpr_index;
 573         bool                    uses_instanceid;
 574         ubyte                   nr_pos_exports;
 575         ubyte                   nr_param_exports;
 576 };
 577
 578 struct si_shader {
 579         struct si_compiler_ctx_state    compiler_ctx_state;
 580
 581         struct si_shader_selector       *selector;
 582         struct si_shader_selector       *previous_stage_sel; /* for refcounting */
 583         struct si_shader                *next_variant;
 584
 585         struct si_shader_part           *prolog;
 586         struct si_shader                *previous_stage; /* for GFX9 */
 587         struct si_shader_part           *prolog2;
 588         struct si_shader_part           *epilog;
 589
 590         struct si_pm4_state             *pm4;
 591         struct r600_resource            *bo;
 592         struct r600_resource            *scratch_bo;
 593         struct si_shader_key            key;
 594         struct util_queue_fence         ready;
 595         bool                            compilation_failed;
 596         bool                            is_monolithic;
 597         bool                            is_optimized;
 598         bool                            is_binary_shared;
 599         bool                            is_gs_copy_shader;
 600
 601         /* The following data is all that's needed for binary shaders. */
 602         struct ac_shader_binary binary;
 603         struct si_shader_config         config;
 604         struct si_shader_info           info;
 605
 606         /* Shader key + LLVM IR + disassembly + statistics.
 607          * Generated for debug contexts only.
 608          */
 609         char                            *shader_log;
 610         size_t                          shader_log_size;
 611 };
 612
 613 struct si_shader_part {
 614         struct si_shader_part *next;
 615         union si_shader_part_key key;
 616         struct ac_shader_binary binary;
 617         struct si_shader_config config;
 618 };
 619
 620 /* si_shader.c */
 621 struct si_shader *
 622 si_generate_gs_copy_shader(struct si_screen *sscreen,
 623                            LLVMTargetMachineRef tm,
 624                            struct si_shader_selector *gs_selector,
 625                            struct pipe_debug_callback *debug);
 626 int si_compile_tgsi_shader(struct si_screen *sscreen,
 627                            LLVMTargetMachineRef tm,
 628                            struct si_shader *shader,
 629                            bool is_monolithic,
 630                            struct pipe_debug_callback *debug);
 631 int si_shader_create(struct si_screen *sscreen, LLVMTargetMachineRef tm,
 632                      struct si_shader *shader,
 633                      struct pipe_debug_callback *debug);
 634 void si_shader_destroy(struct si_shader *shader);
 635 unsigned si_shader_io_get_unique_index_patch(unsigned semantic_name, unsigned index);
 636 unsigned si_shader_io_get_unique_index(unsigned semantic_name, unsigned index);
 637 int si_shader_binary_upload(struct si_screen *sscreen, struct si_shader *shader);
 638 void si_shader_dump(struct si_screen *sscreen, const struct si_shader *shader,
 639                     struct pipe_debug_callback *debug, unsigned processor,
 640                     FILE *f, bool check_debug_option);
 641 void si_multiwave_lds_size_workaround(struct si_screen *sscreen,
 642                                       unsigned *lds_size);
 643 void si_shader_apply_scratch_relocs(struct si_shader *shader,
 644                                     uint64_t scratch_va);
 645 void si_shader_binary_read_config(struct ac_shader_binary *binary,
 646                                   struct si_shader_config *conf,
 647                                   unsigned symbol_offset);
 648 unsigned si_get_spi_shader_z_format(bool writes_z, bool writes_stencil,
 649                                     bool writes_samplemask);
 650 const char *si_get_shader_name(const struct si_shader *shader, unsigned processor);
 651
 652 /* si_shader_nir.c */
 653 void si_nir_scan_shader(const struct nir_shader *nir,
 654                         struct tgsi_shader_info *info);
 655 void si_lower_nir(struct si_shader_selector *sel);
 656
 657 /* Inline helpers. */
 658
 659 /* Return the pointer to the main shader part's pointer. */
 660 static inline struct si_shader **
 661 si_get_main_shader_part(struct si_shader_selector *sel,
 662                         struct si_shader_key *key)
 663 {
 664         if (key->as_ls)
 665                 return &sel->main_shader_part_ls;
 666         if (key->as_es)
 667                 return &sel->main_shader_part_es;
 668         return &sel->main_shader_part;
 669 }
 670
 671 static inline bool
 672 si_shader_uses_bindless_samplers(struct si_shader_selector *selector)
 673 {
 674         return selector ? selector->info.uses_bindless_samplers : false;
 675 }
 676
 677 static inline bool
 678 si_shader_uses_bindless_images(struct si_shader_selector *selector)
 679 {
 680         return selector ? selector->info.uses_bindless_images : false;
 681 }
 682
 683 void si_destroy_shader_selector(struct si_context *sctx,
 684                                 struct si_shader_selector *sel);
 685
 686 static inline void
 687 si_shader_selector_reference(struct si_context *sctx,
 688                              struct si_shader_selector **dst,
 689                              struct si_shader_selector *src)
 690 {
 691         if (pipe_reference(&(*dst)->reference, &src->reference))
 692                 si_destroy_shader_selector(sctx, *dst);
 693
 694         *dst = src;
 695 }
 696
 697 #endif