src/gallium/drivers/etnaviv/etnaviv_shader.c

   1 /*
   2  * Copyright (c) 2012-2015 Etnaviv 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  * Authors:
  24  *    Wladimir J. van der Laan <laanwj@gmail.com>
  25  */
  26
  27 #include "etnaviv_shader.h"
  28
  29 #include "etnaviv_compiler.h"
  30 #include "etnaviv_context.h"
  31 #include "etnaviv_debug.h"
  32 #include "etnaviv_screen.h"
  33 #include "etnaviv_util.h"
  34
  35 #include "tgsi/tgsi_parse.h"
  36 #include "nir/tgsi_to_nir.h"
  37 #include "util/u_math.h"
  38 #include "util/u_memory.h"
  39
  40 /* Upload shader code to bo, if not already done */
  41 static bool etna_icache_upload_shader(struct etna_context *ctx, struct etna_shader_variant *v)
  42 {
  43    if (v->bo)
  44       return true;
  45    v->bo = etna_bo_new(ctx->screen->dev, v->code_size*4, DRM_ETNA_GEM_CACHE_WC);
  46    if (!v->bo)
  47       return false;
  48
  49    void *buf = etna_bo_map(v->bo);
  50    etna_bo_cpu_prep(v->bo, DRM_ETNA_PREP_WRITE);
  51    memcpy(buf, v->code, v->code_size*4);
  52    etna_bo_cpu_fini(v->bo);
  53    DBG("Uploaded %s of %u words to bo %p", v->stage == MESA_SHADER_FRAGMENT ? "fs":"vs", v->code_size, v->bo);
  54    return true;
  55 }
  56
  57 /* Link vs and fs together: fill in shader_state from vs and fs
  58  * as this function is called every time a new fs or vs is bound, the goal is to
  59  * do little processing as possible here, and to precompute as much as possible in
  60  * the vs/fs shader_object.
  61  *
  62  * XXX we could cache the link result for a certain set of VS/PS; usually a pair
  63  * of VS and PS will be used together anyway.
  64  */
  65 static bool
  66 etna_link_shaders(struct etna_context *ctx, struct compiled_shader_state *cs,
  67                   struct etna_shader_variant *vs, struct etna_shader_variant *fs)
  68 {
  69    struct etna_shader_link_info link = { };
  70
  71    assert(vs->stage == MESA_SHADER_VERTEX);
  72    assert(fs->stage == MESA_SHADER_FRAGMENT);
  73
  74 #ifdef DEBUG
  75    if (DBG_ENABLED(ETNA_DBG_DUMP_SHADERS)) {
  76       if (DBG_ENABLED(ETNA_DBG_NIR)) {
  77          etna_dump_shader_nir(vs);
  78          etna_dump_shader_nir(fs);
  79       } else {
  80          etna_dump_shader(vs);
  81          etna_dump_shader(fs);
  82       }
  83    }
  84 #endif
  85
  86    if (etna_link_shader(&link, vs, fs)) {
  87       /* linking failed: some fs inputs do not have corresponding
  88        * vs outputs */
  89       assert(0);
  90
  91       return false;
  92    }
  93
  94    if (DBG_ENABLED(ETNA_DBG_LINKER_MSGS)) {
  95       debug_printf("link result:\n");
  96       debug_printf("  vs  -> fs  comps use     pa_attr\n");
  97
  98       for (int idx = 0; idx < link.num_varyings; ++idx)
  99          debug_printf("  t%-2u -> t%-2u %-5.*s %u,%u,%u,%u 0x%08x\n",
 100                       link.varyings[idx].reg, idx + 1,
 101                       link.varyings[idx].num_components, "xyzw",
 102                       link.varyings[idx].use[0], link.varyings[idx].use[1],
 103                       link.varyings[idx].use[2], link.varyings[idx].use[3],
 104                       link.varyings[idx].pa_attributes);
 105    }
 106
 107    /* set last_varying_2x flag if the last varying has 1 or 2 components */
 108    bool last_varying_2x = false;
 109    if (link.num_varyings > 0 && link.varyings[link.num_varyings - 1].num_components <= 2)
 110       last_varying_2x = true;
 111
 112    cs->RA_CONTROL = VIVS_RA_CONTROL_UNK0 |
 113                     COND(last_varying_2x, VIVS_RA_CONTROL_LAST_VARYING_2X);
 114
 115    cs->PA_ATTRIBUTE_ELEMENT_COUNT = VIVS_PA_ATTRIBUTE_ELEMENT_COUNT_COUNT(link.num_varyings);
 116    for (int idx = 0; idx < link.num_varyings; ++idx)
 117       cs->PA_SHADER_ATTRIBUTES[idx] = link.varyings[idx].pa_attributes;
 118
 119    cs->VS_END_PC = vs->code_size / 4;
 120    cs->VS_OUTPUT_COUNT = 1 + link.num_varyings; /* position + varyings */
 121
 122    /* vs outputs (varyings) */
 123    DEFINE_ETNA_BITARRAY(vs_output, 16, 8) = {0};
 124    int varid = 0;
 125    etna_bitarray_set(vs_output, 8, varid++, vs->vs_pos_out_reg);
 126    for (int idx = 0; idx < link.num_varyings; ++idx)
 127       etna_bitarray_set(vs_output, 8, varid++, link.varyings[idx].reg);
 128    if (vs->vs_pointsize_out_reg >= 0)
 129       etna_bitarray_set(vs_output, 8, varid++, vs->vs_pointsize_out_reg); /* pointsize is last */
 130
 131    for (int idx = 0; idx < ARRAY_SIZE(cs->VS_OUTPUT); ++idx)
 132       cs->VS_OUTPUT[idx] = vs_output[idx];
 133
 134    if (vs->vs_pointsize_out_reg != -1) {
 135       /* vertex shader outputs point coordinate, provide extra output and make
 136        * sure PA config is
 137        * not masked */
 138       cs->PA_CONFIG = ~0;
 139       cs->VS_OUTPUT_COUNT_PSIZE = cs->VS_OUTPUT_COUNT + 1;
 140    } else {
 141       /* vertex shader does not output point coordinate, make sure thate
 142        * POINT_SIZE_ENABLE is masked
 143        * and no extra output is given */
 144       cs->PA_CONFIG = ~VIVS_PA_CONFIG_POINT_SIZE_ENABLE;
 145       cs->VS_OUTPUT_COUNT_PSIZE = cs->VS_OUTPUT_COUNT;
 146    }
 147
 148    cs->VS_LOAD_BALANCING = vs->vs_load_balancing;
 149    cs->VS_START_PC = 0;
 150
 151    cs->PS_END_PC = fs->code_size / 4;
 152    cs->PS_OUTPUT_REG = fs->ps_color_out_reg;
 153    cs->PS_INPUT_COUNT =
 154       VIVS_PS_INPUT_COUNT_COUNT(link.num_varyings + 1) | /* Number of inputs plus position */
 155       VIVS_PS_INPUT_COUNT_UNK8(fs->input_count_unk8);
 156    cs->PS_TEMP_REGISTER_CONTROL =
 157       VIVS_PS_TEMP_REGISTER_CONTROL_NUM_TEMPS(MAX2(fs->num_temps, link.num_varyings + 1));
 158    cs->PS_CONTROL = VIVS_PS_CONTROL_UNK1; /* XXX when can we set BYPASS? */
 159    cs->PS_START_PC = 0;
 160
 161    /* Precompute PS_INPUT_COUNT and TEMP_REGISTER_CONTROL in the case of MSAA
 162     * mode, avoids some fumbling in sync_context. */
 163    cs->PS_INPUT_COUNT_MSAA =
 164       VIVS_PS_INPUT_COUNT_COUNT(link.num_varyings + 2) | /* MSAA adds another input */
 165       VIVS_PS_INPUT_COUNT_UNK8(fs->input_count_unk8);
 166    cs->PS_TEMP_REGISTER_CONTROL_MSAA =
 167       VIVS_PS_TEMP_REGISTER_CONTROL_NUM_TEMPS(MAX2(fs->num_temps, link.num_varyings + 2));
 168
 169    uint32_t total_components = 0;
 170    DEFINE_ETNA_BITARRAY(num_components, ETNA_NUM_VARYINGS, 4) = {0};
 171    DEFINE_ETNA_BITARRAY(component_use, 4 * ETNA_NUM_VARYINGS, 2) = {0};
 172    for (int idx = 0; idx < link.num_varyings; ++idx) {
 173       const struct etna_varying *varying = &link.varyings[idx];
 174
 175       etna_bitarray_set(num_components, 4, idx, varying->num_components);
 176       for (int comp = 0; comp < varying->num_components; ++comp) {
 177          etna_bitarray_set(component_use, 2, total_components, varying->use[comp]);
 178          total_components += 1;
 179       }
 180    }
 181
 182    cs->GL_VARYING_TOTAL_COMPONENTS =
 183       VIVS_GL_VARYING_TOTAL_COMPONENTS_NUM(align(total_components, 2));
 184    cs->GL_VARYING_NUM_COMPONENTS = num_components[0];
 185    cs->GL_VARYING_COMPONENT_USE[0] = component_use[0];
 186    cs->GL_VARYING_COMPONENT_USE[1] = component_use[1];
 187
 188    cs->GL_HALTI5_SH_SPECIALS =
 189       0x7f7f0000 | /* unknown bits, probably other PS inputs */
 190       /* pointsize is last (see above) */
 191       VIVS_GL_HALTI5_SH_SPECIALS_VS_PSIZE_OUT((vs->vs_pointsize_out_reg != -1) ?
 192                                               cs->VS_OUTPUT_COUNT * 4 : 0x00) |
 193       VIVS_GL_HALTI5_SH_SPECIALS_PS_PCOORD_IN((link.pcoord_varying_comp_ofs != -1) ?
 194                                               link.pcoord_varying_comp_ofs : 0x7f);
 195
 196    /* reference instruction memory */
 197    cs->vs_inst_mem_size = vs->code_size;
 198    cs->VS_INST_MEM = vs->code;
 199
 200    cs->ps_inst_mem_size = fs->code_size;
 201    cs->PS_INST_MEM = fs->code;
 202
 203    if (vs->needs_icache | fs->needs_icache) {
 204       /* If either of the shaders needs ICACHE, we use it for both. It is
 205        * either switched on or off for the entire shader processor.
 206        */
 207       if (!etna_icache_upload_shader(ctx, vs) ||
 208           !etna_icache_upload_shader(ctx, fs)) {
 209          assert(0);
 210          return false;
 211       }
 212
 213       cs->VS_INST_ADDR.bo = vs->bo;
 214       cs->VS_INST_ADDR.offset = 0;
 215       cs->VS_INST_ADDR.flags = ETNA_RELOC_READ;
 216       cs->PS_INST_ADDR.bo = fs->bo;
 217       cs->PS_INST_ADDR.offset = 0;
 218       cs->PS_INST_ADDR.flags = ETNA_RELOC_READ;
 219    } else {
 220       /* clear relocs */
 221       memset(&cs->VS_INST_ADDR, 0, sizeof(cs->VS_INST_ADDR));
 222       memset(&cs->PS_INST_ADDR, 0, sizeof(cs->PS_INST_ADDR));
 223    }
 224
 225    return true;
 226 }
 227
 228 bool
 229 etna_shader_link(struct etna_context *ctx)
 230 {
 231    if (!ctx->shader.vs || !ctx->shader.fs)
 232       return false;
 233
 234    /* re-link vs and fs if needed */
 235    return etna_link_shaders(ctx, &ctx->shader_state, ctx->shader.vs, ctx->shader.fs);
 236 }
 237
 238 static bool
 239 etna_shader_update_vs_inputs(struct compiled_shader_state *cs,
 240                              const struct etna_shader_variant *vs,
 241                              const struct compiled_vertex_elements_state *ves)
 242 {
 243    unsigned num_temps, cur_temp, num_vs_inputs;
 244
 245    if (!vs)
 246       return false;
 247
 248    /* Number of vertex elements determines number of VS inputs. Otherwise,
 249     * the GPU crashes. Allocate any unused vertex elements to VS temporary
 250     * registers. */
 251    num_vs_inputs = MAX2(ves->num_elements, vs->infile.num_reg);
 252    if (num_vs_inputs != ves->num_elements) {
 253       BUG("Number of elements %u does not match the number of VS inputs %zu",
 254           ves->num_elements, vs->infile.num_reg);
 255       return false;
 256    }
 257
 258    cur_temp = vs->num_temps;
 259    num_temps = num_vs_inputs - vs->infile.num_reg + cur_temp;
 260
 261    cs->VS_INPUT_COUNT = VIVS_VS_INPUT_COUNT_COUNT(num_vs_inputs) |
 262                         VIVS_VS_INPUT_COUNT_UNK8(vs->input_count_unk8);
 263    cs->VS_TEMP_REGISTER_CONTROL =
 264       VIVS_VS_TEMP_REGISTER_CONTROL_NUM_TEMPS(num_temps);
 265
 266    /* vs inputs (attributes) */
 267    DEFINE_ETNA_BITARRAY(vs_input, 16, 8) = {0};
 268    for (int idx = 0; idx < num_vs_inputs; ++idx) {
 269       if (idx < vs->infile.num_reg)
 270          etna_bitarray_set(vs_input, 8, idx, vs->infile.reg[idx].reg);
 271       else
 272          etna_bitarray_set(vs_input, 8, idx, cur_temp++);
 273    }
 274
 275    for (int idx = 0; idx < ARRAY_SIZE(cs->VS_INPUT); ++idx)
 276       cs->VS_INPUT[idx] = vs_input[idx];
 277
 278    return true;
 279 }
 280
 281 static inline const char *
 282 etna_shader_stage(struct etna_shader_variant *shader)
 283 {
 284    switch (shader->stage) {
 285    case MESA_SHADER_VERTEX:     return "VERT";
 286    case MESA_SHADER_FRAGMENT:   return "FRAG";
 287    case MESA_SHADER_COMPUTE:    return "CL";
 288    default:
 289       unreachable("invalid type");
 290       return NULL;
 291    }
 292 }
 293
 294 static void
 295 dump_shader_info(struct etna_shader_variant *v, struct pipe_debug_callback *debug)
 296 {
 297    if (!unlikely(etna_mesa_debug & ETNA_DBG_SHADERDB))
 298       return;
 299
 300    pipe_debug_message(debug, SHADER_INFO, "\n"
 301          "SHADER-DB: %s prog %d/%d: %u instructions %u temps\n"
 302          "SHADER-DB: %s prog %d/%d: %u immediates %u loops\n",
 303          etna_shader_stage(v),
 304          v->shader->id, v->id,
 305          v->code_size,
 306          v->num_temps,
 307          etna_shader_stage(v),
 308          v->shader->id, v->id,
 309          v->uniforms.imm_count,
 310          v->num_loops);
 311 }
 312
 313 bool
 314 etna_shader_update_vertex(struct etna_context *ctx)
 315 {
 316    return etna_shader_update_vs_inputs(&ctx->shader_state, ctx->shader.vs,
 317                                        ctx->vertex_elements);
 318 }
 319
 320 static struct etna_shader_variant *
 321 create_variant(struct etna_shader *shader, struct etna_shader_key key)
 322 {
 323    struct etna_shader_variant *v = CALLOC_STRUCT(etna_shader_variant);
 324    int ret;
 325
 326    if (!v)
 327       return NULL;
 328
 329    v->shader = shader;
 330    v->key = key;
 331
 332    ret = etna_compile_shader(v);
 333    if (!ret) {
 334       debug_error("compile failed!");
 335       goto fail;
 336    }
 337
 338    v->id = ++shader->variant_count;
 339
 340    return v;
 341
 342 fail:
 343    FREE(v);
 344    return NULL;
 345 }
 346
 347 struct etna_shader_variant *
 348 etna_shader_variant(struct etna_shader *shader, struct etna_shader_key key,
 349                    struct pipe_debug_callback *debug)
 350 {
 351    struct etna_shader_variant *v;
 352
 353    for (v = shader->variants; v; v = v->next)
 354       if (etna_shader_key_equal(&key, &v->key))
 355          return v;
 356
 357    /* compile new variant if it doesn't exist already */
 358    v = create_variant(shader, key);
 359    if (v) {
 360       v->next = shader->variants;
 361       shader->variants = v;
 362       dump_shader_info(v, debug);
 363    }
 364
 365    return v;
 366 }
 367
 368 static void *
 369 etna_create_shader_state(struct pipe_context *pctx,
 370                          const struct pipe_shader_state *pss)
 371 {
 372    struct etna_context *ctx = etna_context(pctx);
 373    struct etna_shader *shader = CALLOC_STRUCT(etna_shader);
 374
 375    if (!shader)
 376       return NULL;
 377
 378    static uint32_t id;
 379    shader->id = id++;
 380    shader->specs = &ctx->specs;
 381
 382    if (DBG_ENABLED(ETNA_DBG_NIR))
 383       shader->nir = (pss->type == PIPE_SHADER_IR_NIR) ? pss->ir.nir :
 384                      tgsi_to_nir(pss->tokens, pctx->screen);
 385    else
 386       shader->tokens = tgsi_dup_tokens(pss->tokens);
 387
 388
 389
 390    if (etna_mesa_debug & ETNA_DBG_SHADERDB) {
 391       /* if shader-db run, create a standard variant immediately
 392        * (as otherwise nothing will trigger the shader to be
 393        * actually compiled).
 394        */
 395       struct etna_shader_key key = {};
 396       etna_shader_variant(shader, key, &ctx->debug);
 397    }
 398
 399    return shader;
 400 }
 401
 402 static void
 403 etna_delete_shader_state(struct pipe_context *pctx, void *ss)
 404 {
 405    struct etna_shader *shader = ss;
 406    struct etna_shader_variant *v, *t;
 407
 408    v = shader->variants;
 409    while (v) {
 410       t = v;
 411       v = v->next;
 412       if (t->bo)
 413          etna_bo_del(t->bo);
 414       etna_destroy_shader(t);
 415    }
 416
 417    ralloc_free(shader->nir);
 418    FREE(shader);
 419 }
 420
 421 static void
 422 etna_bind_fs_state(struct pipe_context *pctx, void *hwcso)
 423 {
 424    struct etna_context *ctx = etna_context(pctx);
 425
 426    ctx->shader.bind_fs = hwcso;
 427    ctx->dirty |= ETNA_DIRTY_SHADER;
 428 }
 429
 430 static void
 431 etna_bind_vs_state(struct pipe_context *pctx, void *hwcso)
 432 {
 433    struct etna_context *ctx = etna_context(pctx);
 434
 435    ctx->shader.bind_vs = hwcso;
 436    ctx->dirty |= ETNA_DIRTY_SHADER;
 437 }
 438
 439 void
 440 etna_shader_init(struct pipe_context *pctx)
 441 {
 442    pctx->create_fs_state = etna_create_shader_state;
 443    pctx->bind_fs_state = etna_bind_fs_state;
 444    pctx->delete_fs_state = etna_delete_shader_state;
 445    pctx->create_vs_state = etna_create_shader_state;
 446    pctx->bind_vs_state = etna_bind_vs_state;
 447    pctx->delete_vs_state = etna_delete_shader_state;
 448 }