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_fsqrt = true,
  47    .lower_sub = true,
  48    .lower_flrp32 = true,
  49    .lower_flrp64 = true,
  50    .lower_ftrunc = true,
  51    /* could be implemented by clamp */
  52    .lower_fsat = true,
  53 };
  54
  55 static const nir_shader_compiler_options fs_nir_options = {
  56    .lower_fpow = true,
  57    .lower_fdiv = true,
  58    .lower_sub = true,
  59    .lower_flrp32 = true,
  60    .lower_flrp64 = true,
  61    .lower_fsign = true,
  62 };
  63
  64 const void *
  65 lima_program_get_compiler_options(enum pipe_shader_type shader)
  66 {
  67    switch (shader) {
  68    case PIPE_SHADER_VERTEX:
  69       return &vs_nir_options;
  70    case PIPE_SHADER_FRAGMENT:
  71       return &fs_nir_options;
  72    default:
  73       return NULL;
  74    }
  75 }
  76
  77 static int
  78 type_size(const struct glsl_type *type, bool bindless)
  79 {
  80    return glsl_count_attribute_slots(type, false);
  81 }
  82
  83 static void
  84 lima_program_optimize_vs_nir(struct nir_shader *s)
  85 {
  86    bool progress;
  87
  88    NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size, 0);
  89    NIR_PASS_V(s, nir_lower_regs_to_ssa);
  90    NIR_PASS_V(s, nir_lower_load_const_to_scalar);
  91    NIR_PASS_V(s, lima_nir_lower_uniform_to_scalar);
  92    NIR_PASS_V(s, nir_lower_io_to_scalar,
  93               nir_var_shader_in|nir_var_shader_out);
  94    NIR_PASS_V(s, nir_lower_sincos);
  95
  96    do {
  97       progress = false;
  98
  99       NIR_PASS_V(s, nir_lower_vars_to_ssa);
 100       NIR_PASS(progress, s, nir_lower_alu_to_scalar, NULL);
 101       NIR_PASS(progress, s, nir_lower_phis_to_scalar);
 102       NIR_PASS(progress, s, nir_copy_prop);
 103       NIR_PASS(progress, s, nir_opt_remove_phis);
 104       NIR_PASS(progress, s, nir_opt_dce);
 105       NIR_PASS(progress, s, nir_opt_dead_cf);
 106       NIR_PASS(progress, s, nir_opt_cse);
 107       NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true);
 108       NIR_PASS(progress, s, nir_opt_algebraic);
 109       NIR_PASS(progress, s, nir_opt_constant_folding);
 110       NIR_PASS(progress, s, nir_opt_undef);
 111       NIR_PASS(progress, s, nir_opt_loop_unroll,
 112                nir_var_shader_in |
 113                nir_var_shader_out |
 114                nir_var_function_temp);
 115    } while (progress);
 116
 117    NIR_PASS_V(s, nir_lower_int_to_float);
 118    NIR_PASS_V(s, nir_copy_prop);
 119    NIR_PASS_V(s, nir_lower_locals_to_regs);
 120    NIR_PASS_V(s, nir_convert_from_ssa, true);
 121    NIR_PASS_V(s, nir_remove_dead_variables, nir_var_function_temp);
 122    nir_sweep(s);
 123 }
 124
 125 static void
 126 lima_program_optimize_fs_nir(struct nir_shader *s)
 127 {
 128    bool progress;
 129
 130    NIR_PASS_V(s, nir_lower_fragcoord_wtrans);
 131    NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size, 0);
 132    NIR_PASS_V(s, nir_lower_regs_to_ssa);
 133
 134    do {
 135       progress = false;
 136
 137       NIR_PASS_V(s, nir_lower_vars_to_ssa);
 138       //NIR_PASS(progress, s, nir_lower_alu_to_scalar, NULL);
 139       NIR_PASS(progress, s, nir_lower_phis_to_scalar);
 140       NIR_PASS(progress, s, nir_copy_prop);
 141       NIR_PASS(progress, s, nir_opt_remove_phis);
 142       NIR_PASS(progress, s, nir_opt_dce);
 143       NIR_PASS(progress, s, nir_opt_dead_cf);
 144       NIR_PASS(progress, s, nir_opt_cse);
 145       NIR_PASS(progress, s, nir_opt_peephole_select, 8, true, true);
 146       NIR_PASS(progress, s, nir_opt_algebraic);
 147       NIR_PASS(progress, s, nir_opt_constant_folding);
 148       NIR_PASS(progress, s, nir_opt_undef);
 149       NIR_PASS(progress, s, nir_opt_loop_unroll,
 150                nir_var_shader_in |
 151                nir_var_shader_out |
 152                nir_var_function_temp);
 153    } while (progress);
 154
 155    NIR_PASS_V(s, nir_lower_int_to_float);
 156
 157    /* Lower modifiers */
 158    NIR_PASS_V(s, nir_lower_to_source_mods, nir_lower_all_source_mods);
 159    NIR_PASS_V(s, nir_copy_prop);
 160    NIR_PASS_V(s, nir_opt_dce);
 161
 162    NIR_PASS_V(s, nir_lower_locals_to_regs);
 163    NIR_PASS_V(s, nir_convert_from_ssa, true);
 164    NIR_PASS_V(s, nir_remove_dead_variables, nir_var_function_temp);
 165
 166    NIR_PASS_V(s, nir_move_vec_src_uses_to_dest);
 167    NIR_PASS_V(s, nir_lower_vec_to_movs);
 168
 169    nir_sweep(s);
 170 }
 171
 172 static void *
 173 lima_create_fs_state(struct pipe_context *pctx,
 174                      const struct pipe_shader_state *cso)
 175 {
 176    struct lima_screen *screen = lima_screen(pctx->screen);
 177    struct lima_fs_shader_state *so = rzalloc(NULL, struct lima_fs_shader_state);
 178
 179    if (!so)
 180       return NULL;
 181
 182    nir_shader *nir;
 183    if (cso->type == PIPE_SHADER_IR_NIR)
 184       nir = cso->ir.nir;
 185    else {
 186       assert(cso->type == PIPE_SHADER_IR_TGSI);
 187
 188       nir = tgsi_to_nir(cso->tokens, pctx->screen);
 189    }
 190
 191    lima_program_optimize_fs_nir(nir);
 192
 193    if (lima_debug & LIMA_DEBUG_PP)
 194       nir_print_shader(nir, stdout);
 195
 196    if (!ppir_compile_nir(so, nir, screen->pp_ra)) {
 197       ralloc_free(so);
 198       return NULL;
 199    }
 200
 201    return so;
 202 }
 203
 204 static void
 205 lima_bind_fs_state(struct pipe_context *pctx, void *hwcso)
 206 {
 207    struct lima_context *ctx = lima_context(pctx);
 208
 209    ctx->fs = hwcso;
 210    ctx->dirty |= LIMA_CONTEXT_DIRTY_SHADER_FRAG;
 211 }
 212
 213 static void
 214 lima_delete_fs_state(struct pipe_context *pctx, void *hwcso)
 215 {
 216    struct lima_fs_shader_state *so = hwcso;
 217
 218    if (so->bo)
 219       lima_bo_free(so->bo);
 220
 221    ralloc_free(so);
 222 }
 223
 224 bool
 225 lima_update_vs_state(struct lima_context *ctx)
 226 {
 227    struct lima_vs_shader_state *vs = ctx->vs;
 228    if (!vs->bo) {
 229       struct lima_screen *screen = lima_screen(ctx->base.screen);
 230       vs->bo = lima_bo_create(screen, vs->shader_size, 0);
 231       if (!vs->bo) {
 232          fprintf(stderr, "lima: create vs shader bo fail\n");
 233          return false;
 234       }
 235
 236       memcpy(lima_bo_map(vs->bo), vs->shader, vs->shader_size);
 237       ralloc_free(vs->shader);
 238       vs->shader = NULL;
 239    }
 240
 241    return true;
 242 }
 243
 244 bool
 245 lima_update_fs_state(struct lima_context *ctx)
 246 {
 247    struct lima_fs_shader_state *fs = ctx->fs;
 248    if (!fs->bo) {
 249       struct lima_screen *screen = lima_screen(ctx->base.screen);
 250       fs->bo = lima_bo_create(screen, fs->shader_size, 0);
 251       if (!fs->bo) {
 252          fprintf(stderr, "lima: create fs shader bo fail\n");
 253          return false;
 254       }
 255
 256       memcpy(lima_bo_map(fs->bo), fs->shader, fs->shader_size);
 257       ralloc_free(fs->shader);
 258       fs->shader = NULL;
 259    }
 260
 261    return true;
 262 }
 263
 264 static void *
 265 lima_create_vs_state(struct pipe_context *pctx,
 266                      const struct pipe_shader_state *cso)
 267 {
 268    struct lima_vs_shader_state *so = rzalloc(NULL, struct lima_vs_shader_state);
 269
 270    if (!so)
 271       return NULL;
 272
 273    nir_shader *nir;
 274    if (cso->type == PIPE_SHADER_IR_NIR)
 275       nir = cso->ir.nir;
 276    else {
 277       assert(cso->type == PIPE_SHADER_IR_TGSI);
 278
 279       nir = tgsi_to_nir(cso->tokens, pctx->screen);
 280    }
 281
 282    lima_program_optimize_vs_nir(nir);
 283
 284    if (lima_debug & LIMA_DEBUG_GP)
 285       nir_print_shader(nir, stdout);
 286
 287    if (!gpir_compile_nir(so, nir)) {
 288       ralloc_free(so);
 289       return NULL;
 290    }
 291
 292    return so;
 293 }
 294
 295 static void
 296 lima_bind_vs_state(struct pipe_context *pctx, void *hwcso)
 297 {
 298    struct lima_context *ctx = lima_context(pctx);
 299
 300    ctx->vs = hwcso;
 301    ctx->dirty |= LIMA_CONTEXT_DIRTY_SHADER_VERT;
 302 }
 303
 304 static void
 305 lima_delete_vs_state(struct pipe_context *pctx, void *hwcso)
 306 {
 307    struct lima_vs_shader_state *so = hwcso;
 308
 309    if (so->bo)
 310       lima_bo_free(so->bo);
 311
 312    ralloc_free(so);
 313 }
 314
 315 void
 316 lima_program_init(struct lima_context *ctx)
 317 {
 318    ctx->base.create_fs_state = lima_create_fs_state;
 319    ctx->base.bind_fs_state = lima_bind_fs_state;
 320    ctx->base.delete_fs_state = lima_delete_fs_state;
 321
 322    ctx->base.create_vs_state = lima_create_vs_state;
 323    ctx->base.bind_vs_state = lima_bind_vs_state;
 324    ctx->base.delete_vs_state = lima_delete_vs_state;
 325 }