src/gallium/drivers/lima/lima_program.c

   1 /*
   2  * Copyright (c) 2017-2019 Lima Project
   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, sub license,
   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
  12  * next paragraph) shall be included in all copies or substantial portions
  13  * of the 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 AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  21  * DEALINGS IN THE SOFTWARE.
  22  *
  23  */
  24
  25 #include "util/u_memory.h"
  26 #include "util/ralloc.h"
  27 #include "util/u_debug.h"
  28
  29 #include "tgsi/tgsi_dump.h"
  30 #include "compiler/nir/nir.h"
  31 #include "nir/tgsi_to_nir.h"
  32
  33 #include "pipe/p_state.h"
  34
  35 #include "lima_screen.h"
  36 #include "lima_context.h"
  37 #include "lima_program.h"
  38 #include "lima_bo.h"
  39 #include "ir/lima_ir.h"
  40
  41 static const nir_shader_compiler_options vs_nir_options = {
  42    .lower_ffma = true,
  43    .lower_fpow = true,
  44    .lower_ffract = true,
  45    .lower_fdiv = true,
  46    .lower_fmod = true,
  47    .lower_fsqrt = true,
  48    .lower_sub = true,
  49    .lower_flrp32 = true,
  50    .lower_flrp64 = true,
  51    .lower_ftrunc = true,
  52    /* could be implemented by clamp */
  53    .lower_fsat = true,
  54    .lower_bitops = true,
  55    .lower_rotate = true,
  56    .lower_sincos = true,
  57 };
  58
  59 static const nir_shader_compiler_options fs_nir_options = {
  60    .lower_ffma = true,
  61    .lower_fpow = true,
  62    .lower_fdiv = true,
  63    .lower_fmod = true,
  64    .lower_sub = true,
  65    .lower_flrp32 = true,
  66    .lower_flrp64 = true,
  67    .lower_fsign = true,
  68    .lower_rotate = true,
  69    .lower_fdot = true,
  70    .lower_bitops = true,
  71 };
  72
  73 static const struct nir_lower_tex_options tex_options = {
  74    .lower_txp = ~0u,
  75 };
  76
  77 const void *
  78 lima_program_get_compiler_options(enum pipe_shader_type shader)
  79 {
  80    switch (shader) {
  81    case PIPE_SHADER_VERTEX:
  82       return &vs_nir_options;
  83    case PIPE_SHADER_FRAGMENT:
  84       return &fs_nir_options;
  85    default:
  86       return NULL;
  87    }
  88 }
  89
  90 static int
  91 type_size(const struct glsl_type *type, bool bindless)
  92 {
  93    return glsl_count_attribute_slots(type, false);
  94 }
  95
  96 void
  97 lima_program_optimize_vs_nir(struct nir_shader *s)
  98 {
  99    bool progress;
 100
 101    NIR_PASS_V(s, nir_lower_viewport_transform);
 102    NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size, 0);
 103    NIR_PASS_V(s, nir_lower_load_const_to_scalar);
 104    NIR_PASS_V(s, lima_nir_lower_uniform_to_scalar);
 105    NIR_PASS_V(s, nir_lower_io_to_scalar,
 106               nir_var_shader_in|nir_var_shader_out);
 107
 108    do {
 109       progress = false;
 110
 111       NIR_PASS_V(s, nir_lower_vars_to_ssa);
 112       NIR_PASS(progress, s, nir_lower_alu_to_scalar, NULL);
 113       NIR_PASS(progress, s, nir_lower_phis_to_scalar);
 114       NIR_PASS(progress, s, nir_copy_prop);
 115       NIR_PASS(progress, s, nir_opt_remove_phis);
 116       NIR_PASS(progress, s, nir_opt_dce);
 117       NIR_PASS(progress, s, nir_opt_dead_cf);
 118       NIR_PASS(progress, s, nir_opt_cse);
 119       NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true);
 120       NIR_PASS(progress, s, nir_opt_algebraic);
 121       NIR_PASS(progress, s, nir_opt_constant_folding);
 122       NIR_PASS(progress, s, nir_opt_undef);
 123       NIR_PASS(progress, s, nir_opt_loop_unroll,
 124                nir_var_shader_in |
 125                nir_var_shader_out |
 126                nir_var_function_temp);
 127    } while (progress);
 128
 129    NIR_PASS_V(s, nir_lower_int_to_float);
 130    NIR_PASS_V(s, nir_lower_bool_to_float);
 131
 132    /* Some ops must be lowered after being converted from int ops,
 133     * so re-run nir_opt_algebraic after int lowering. */
 134    do {
 135       progress = false;
 136       NIR_PASS(progress, s, nir_opt_algebraic);
 137    } while (progress);
 138
 139    NIR_PASS_V(s, nir_copy_prop);
 140    NIR_PASS_V(s, nir_opt_dce);
 141    NIR_PASS_V(s, nir_lower_locals_to_regs);
 142    NIR_PASS_V(s, nir_convert_from_ssa, true);
 143    NIR_PASS_V(s, nir_remove_dead_variables, nir_var_function_temp);
 144    nir_sweep(s);
 145 }
 146
 147 void
 148 lima_program_optimize_fs_nir(struct nir_shader *s)
 149 {
 150    BITSET_DECLARE(alu_lower, nir_num_opcodes) = {0};
 151    bool progress;
 152
 153    BITSET_SET(alu_lower, nir_op_frcp);
 154    BITSET_SET(alu_lower, nir_op_frsq);
 155    BITSET_SET(alu_lower, nir_op_flog2);
 156    BITSET_SET(alu_lower, nir_op_fexp2);
 157    BITSET_SET(alu_lower, nir_op_fsqrt);
 158    BITSET_SET(alu_lower, nir_op_fsin);
 159    BITSET_SET(alu_lower, nir_op_fcos);
 160
 161    NIR_PASS_V(s, nir_lower_fragcoord_wtrans);
 162    NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size, 0);
 163    NIR_PASS_V(s, nir_lower_regs_to_ssa);
 164    NIR_PASS_V(s, nir_lower_tex, &tex_options);
 165
 166    do {
 167       progress = false;
 168
 169       NIR_PASS_V(s, nir_lower_vars_to_ssa);
 170       NIR_PASS(progress, s, nir_lower_alu_to_scalar, alu_lower);
 171       NIR_PASS(progress, s, nir_lower_phis_to_scalar);
 172       NIR_PASS(progress, s, nir_copy_prop);
 173       NIR_PASS(progress, s, nir_opt_remove_phis);
 174       NIR_PASS(progress, s, nir_opt_dce);
 175       NIR_PASS(progress, s, nir_opt_dead_cf);
 176       NIR_PASS(progress, s, nir_opt_cse);
 177       NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true);
 178       NIR_PASS(progress, s, nir_opt_algebraic);
 179       NIR_PASS(progress, s, nir_opt_constant_folding);
 180       NIR_PASS(progress, s, nir_opt_undef);
 181       NIR_PASS(progress, s, nir_opt_loop_unroll,
 182                nir_var_shader_in |
 183                nir_var_shader_out |
 184                nir_var_function_temp);
 185    } while (progress);
 186
 187    NIR_PASS_V(s, nir_lower_int_to_float);
 188    NIR_PASS_V(s, nir_lower_bool_to_float);
 189
 190    /* Some ops must be lowered after being converted from int ops,
 191     * so re-run nir_opt_algebraic after int lowering. */
 192    do {
 193       progress = false;
 194       NIR_PASS(progress, s, nir_opt_algebraic);
 195    } while (progress);
 196
 197    /* Lower modifiers */
 198    NIR_PASS_V(s, nir_lower_to_source_mods, nir_lower_all_source_mods);
 199    NIR_PASS_V(s, nir_copy_prop);
 200    NIR_PASS_V(s, nir_opt_dce);
 201
 202    NIR_PASS_V(s, nir_lower_locals_to_regs);
 203    NIR_PASS_V(s, nir_convert_from_ssa, true);
 204    NIR_PASS_V(s, nir_remove_dead_variables, nir_var_function_temp);
 205
 206    NIR_PASS_V(s, nir_move_vec_src_uses_to_dest);
 207    NIR_PASS_V(s, nir_lower_vec_to_movs);
 208
 209    nir_sweep(s);
 210 }
 211
 212 static void *
 213 lima_create_fs_state(struct pipe_context *pctx,
 214                      const struct pipe_shader_state *cso)
 215 {
 216    struct lima_screen *screen = lima_screen(pctx->screen);
 217    struct lima_fs_shader_state *so = rzalloc(NULL, struct lima_fs_shader_state);
 218
 219    if (!so)
 220       return NULL;
 221
 222    nir_shader *nir;
 223    if (cso->type == PIPE_SHADER_IR_NIR)
 224       nir = cso->ir.nir;
 225    else {
 226       assert(cso->type == PIPE_SHADER_IR_TGSI);
 227
 228       nir = tgsi_to_nir(cso->tokens, pctx->screen);
 229    }
 230
 231    lima_program_optimize_fs_nir(nir);
 232
 233    if (lima_debug & LIMA_DEBUG_PP)
 234       nir_print_shader(nir, stdout);
 235
 236    if (!ppir_compile_nir(so, nir, screen->pp_ra)) {
 237       ralloc_free(so);
 238       return NULL;
 239    }
 240
 241    return so;
 242 }
 243
 244 static void
 245 lima_bind_fs_state(struct pipe_context *pctx, void *hwcso)
 246 {
 247    struct lima_context *ctx = lima_context(pctx);
 248
 249    ctx->fs = hwcso;
 250    ctx->dirty |= LIMA_CONTEXT_DIRTY_SHADER_FRAG;
 251 }
 252
 253 static void
 254 lima_delete_fs_state(struct pipe_context *pctx, void *hwcso)
 255 {
 256    struct lima_fs_shader_state *so = hwcso;
 257
 258    if (so->bo)
 259       lima_bo_free(so->bo);
 260
 261    ralloc_free(so);
 262 }
 263
 264 bool
 265 lima_update_vs_state(struct lima_context *ctx)
 266 {
 267    struct lima_vs_shader_state *vs = ctx->vs;
 268    if (!vs->bo) {
 269       struct lima_screen *screen = lima_screen(ctx->base.screen);
 270       vs->bo = lima_bo_create(screen, vs->shader_size, 0);
 271       if (!vs->bo) {
 272          fprintf(stderr, "lima: create vs shader bo fail\n");
 273          return false;
 274       }
 275
 276       memcpy(lima_bo_map(vs->bo), vs->shader, vs->shader_size);
 277       ralloc_free(vs->shader);
 278       vs->shader = NULL;
 279    }
 280
 281    return true;
 282 }
 283
 284 bool
 285 lima_update_fs_state(struct lima_context *ctx)
 286 {
 287    struct lima_fs_shader_state *fs = ctx->fs;
 288    if (!fs->bo) {
 289       struct lima_screen *screen = lima_screen(ctx->base.screen);
 290       fs->bo = lima_bo_create(screen, fs->shader_size, 0);
 291       if (!fs->bo) {
 292          fprintf(stderr, "lima: create fs shader bo fail\n");
 293          return false;
 294       }
 295
 296       memcpy(lima_bo_map(fs->bo), fs->shader, fs->shader_size);
 297       ralloc_free(fs->shader);
 298       fs->shader = NULL;
 299    }
 300
 301    return true;
 302 }
 303
 304 static void *
 305 lima_create_vs_state(struct pipe_context *pctx,
 306                      const struct pipe_shader_state *cso)
 307 {
 308    struct lima_vs_shader_state *so = rzalloc(NULL, struct lima_vs_shader_state);
 309
 310    if (!so)
 311       return NULL;
 312
 313    nir_shader *nir;
 314    if (cso->type == PIPE_SHADER_IR_NIR)
 315       nir = cso->ir.nir;
 316    else {
 317       assert(cso->type == PIPE_SHADER_IR_TGSI);
 318
 319       nir = tgsi_to_nir(cso->tokens, pctx->screen);
 320    }
 321
 322    lima_program_optimize_vs_nir(nir);
 323
 324    if (lima_debug & LIMA_DEBUG_GP)
 325       nir_print_shader(nir, stdout);
 326
 327    if (!gpir_compile_nir(so, nir)) {
 328       ralloc_free(so);
 329       return NULL;
 330    }
 331
 332    return so;
 333 }
 334
 335 static void
 336 lima_bind_vs_state(struct pipe_context *pctx, void *hwcso)
 337 {
 338    struct lima_context *ctx = lima_context(pctx);
 339
 340    ctx->vs = hwcso;
 341    ctx->dirty |= LIMA_CONTEXT_DIRTY_SHADER_VERT;
 342 }
 343
 344 static void
 345 lima_delete_vs_state(struct pipe_context *pctx, void *hwcso)
 346 {
 347    struct lima_vs_shader_state *so = hwcso;
 348
 349    if (so->bo)
 350       lima_bo_free(so->bo);
 351
 352    ralloc_free(so);
 353 }
 354
 355 void
 356 lima_program_init(struct lima_context *ctx)
 357 {
 358    ctx->base.create_fs_state = lima_create_fs_state;
 359    ctx->base.bind_fs_state = lima_bind_fs_state;
 360    ctx->base.delete_fs_state = lima_delete_fs_state;
 361
 362    ctx->base.create_vs_state = lima_create_vs_state;
 363    ctx->base.bind_vs_state = lima_bind_vs_state;
 364    ctx->base.delete_vs_state = lima_delete_vs_state;
 365 }