src/gallium/drivers/nv50/nv50_shader_state.c

   1 /*
   2  * Copyright 2008 Ben Skeggs
   3  * Copyright 2010 Christoph Bumiller
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   9  * and/or sell copies of the Software, and to permit persons to whom the
  10  * Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice shall be included in
  13  * all copies or substantial portions 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 NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
  19  * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
  20  * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  */
  23
  24 #include "pipe/p_context.h"
  25 #include "pipe/p_defines.h"
  26 #include "pipe/p_state.h"
  27 #include "util/u_inlines.h"
  28
  29 #include "nv50_context.h"
  30
  31 void
  32 nv50_constbufs_validate(struct nv50_context *nv50)
  33 {
  34    struct nouveau_channel *chan = nv50->screen->base.channel;
  35    unsigned s;
  36
  37    for (s = 0; s < 3; ++s) {
  38       struct nv04_resource *res;
  39       int i;
  40       unsigned p, b;
  41
  42       if (s == PIPE_SHADER_FRAGMENT)
  43          p = NV50_3D_SET_PROGRAM_CB_PROGRAM_FRAGMENT;
  44       else
  45       if (s == PIPE_SHADER_GEOMETRY)
  46          p = NV50_3D_SET_PROGRAM_CB_PROGRAM_GEOMETRY;
  47       else
  48          p = NV50_3D_SET_PROGRAM_CB_PROGRAM_VERTEX;
  49
  50       while (nv50->constbuf_dirty[s]) {
  51          struct nouveau_bo *bo;
  52          unsigned start = 0;
  53          unsigned words = 0;
  54
  55          i = ffs(nv50->constbuf_dirty[s]) - 1;
  56          nv50->constbuf_dirty[s] &= ~(1 << i);
  57
  58          res = nv04_resource(nv50->constbuf[s][i]);
  59          if (!res) {
  60             if (i != 0) {
  61                BEGIN_RING(chan, RING_3D(SET_PROGRAM_CB), 1);
  62                OUT_RING  (chan, (i << 8) | p | 0);
  63             }
  64             continue;
  65          }
  66
  67          if (i == 0) {
  68             b = NV50_CB_PVP + s;
  69
  70             /* always upload GL uniforms through CB DATA */
  71             bo = nv50->screen->uniforms;
  72             words = res->base.width0 / 4;
  73          } else {
  74             b = s * 16 + i;
  75
  76             assert(0);
  77
  78             if (!nouveau_resource_mapped_by_gpu(&res->base)) {
  79                nouveau_buffer_migrate(&nv50->base, res, NOUVEAU_BO_VRAM);
  80
  81                BEGIN_RING(chan, RING_3D(CODE_CB_FLUSH), 1);
  82                OUT_RING  (chan, 0);
  83             }
  84             MARK_RING (chan, 6, 2);
  85             BEGIN_RING(chan, RING_3D(CB_DEF_ADDRESS_HIGH), 3);
  86             OUT_RESRCh(chan, res, 0, NOUVEAU_BO_RD);
  87             OUT_RESRCl(chan, res, 0, NOUVEAU_BO_RD);
  88             OUT_RING  (chan, (b << 16) | (res->base.width0 & 0xffff));
  89             BEGIN_RING(chan, RING_3D(SET_PROGRAM_CB), 1);
  90             OUT_RING  (chan, (b << 12) | (i << 8) | p | 1);
  91
  92             bo = res->bo;
  93
  94             nv50_bufctx_add_resident(nv50, NV50_BUFCTX_CONSTANT, res,
  95                                      res->domain | NOUVEAU_BO_RD);
  96          }
  97
  98          if (words) {
  99             MARK_RING(chan, 8, 1);
 100
 101             nouveau_bo_validate(chan, bo, res->domain | NOUVEAU_BO_WR);
 102          }
 103
 104          while (words) {
 105             unsigned nr = AVAIL_RING(chan);
 106
 107             if (nr < 16) {
 108                FIRE_RING(chan);
 109                nouveau_bo_validate(chan, bo, res->domain | NOUVEAU_BO_WR);
 110                continue;
 111             }
 112             nr = MIN2(MIN2(nr - 3, words), NV04_PFIFO_MAX_PACKET_LEN);
 113
 114             BEGIN_RING(chan, RING_3D(CB_ADDR), 1);
 115             OUT_RING  (chan, (start << 8) | b);
 116             BEGIN_RING_NI(chan, RING_3D(CB_DATA(0)), nr);
 117             OUT_RINGp (chan, &res->data[start * 4], nr);
 118
 119             start += nr;
 120             words -= nr;
 121          }
 122       }
 123    }
 124 }
 125
 126 static boolean
 127 nv50_program_validate(struct nv50_context *nv50, struct nv50_program *prog)
 128 {
 129    struct nouveau_resource *heap;
 130    int ret;
 131    unsigned size;
 132
 133    if (prog->translated)
 134       return TRUE;
 135
 136    prog->translated = nv50_program_translate(prog);
 137    if (!prog->translated)
 138       return FALSE;
 139
 140    if (prog->type == PIPE_SHADER_FRAGMENT) heap = nv50->screen->fp_code_heap;
 141    else
 142    if (prog->type == PIPE_SHADER_GEOMETRY) heap = nv50->screen->gp_code_heap;
 143    else
 144       heap = nv50->screen->vp_code_heap;
 145
 146    size = align(prog->code_size, 0x100);
 147
 148    ret = nouveau_resource_alloc(heap, size, prog, &prog->res);
 149    if (ret) {
 150       NOUVEAU_ERR("out of code space for shader type %i\n", prog->type);
 151       return FALSE;
 152    }
 153    prog->code_base = prog->res->start;
 154
 155    nv50_relocate_program(prog, prog->code_base, 0);
 156
 157    nv50_sifc_linear_u8(&nv50->base, nv50->screen->code,
 158                        (prog->type << NV50_CODE_BO_SIZE_LOG2) + prog->code_base,
 159                        NOUVEAU_BO_VRAM, prog->code_size, prog->code);
 160
 161    BEGIN_RING(nv50->screen->base.channel, RING_3D(CODE_CB_FLUSH), 1);
 162    OUT_RING  (nv50->screen->base.channel, 0);
 163
 164    return TRUE;
 165 }
 166
 167 void
 168 nv50_vertprog_validate(struct nv50_context *nv50)
 169 {
 170    struct nouveau_channel *chan = nv50->screen->base.channel;
 171    struct nv50_program *vp = nv50->vertprog;
 172
 173    if (nv50->clip.nr > vp->vp.clpd_nr) {
 174       if (vp->translated)
 175          nv50_program_destroy(nv50, vp);
 176       vp->vp.clpd_nr = nv50->clip.nr;
 177    }
 178
 179    if (!nv50_program_validate(nv50, vp))
 180          return;
 181
 182    BEGIN_RING(chan, RING_3D(VP_ATTR_EN(0)), 2);
 183    OUT_RING  (chan, vp->vp.attrs[0]);
 184    OUT_RING  (chan, vp->vp.attrs[1]);
 185    BEGIN_RING(chan, RING_3D(VP_REG_ALLOC_RESULT), 1);
 186    OUT_RING  (chan, vp->max_out);
 187    BEGIN_RING(chan, RING_3D(VP_REG_ALLOC_TEMP), 1);
 188    OUT_RING  (chan, vp->max_gpr);
 189    BEGIN_RING(chan, RING_3D(VP_START_ID), 1);
 190    OUT_RING  (chan, vp->code_base);
 191 }
 192
 193 void
 194 nv50_fragprog_validate(struct nv50_context *nv50)
 195 {
 196    struct nouveau_channel *chan = nv50->screen->base.channel;
 197    struct nv50_program *fp = nv50->fragprog;
 198
 199    if (!nv50_program_validate(nv50, fp))
 200          return;
 201
 202    BEGIN_RING(chan, RING_3D(FP_REG_ALLOC_TEMP), 1);
 203    OUT_RING  (chan, fp->max_gpr);
 204    BEGIN_RING(chan, RING_3D(FP_RESULT_COUNT), 1);
 205    OUT_RING  (chan, fp->max_out);
 206    BEGIN_RING(chan, RING_3D(FP_CONTROL), 1);
 207    OUT_RING  (chan, fp->fp.flags[0]);
 208    BEGIN_RING(chan, RING_3D(FP_CTRL_UNK196C), 1);
 209    OUT_RING  (chan, fp->fp.flags[1]);
 210    BEGIN_RING(chan, RING_3D(FP_START_ID), 1);
 211    OUT_RING  (chan, fp->code_base);
 212 }
 213
 214 void
 215 nv50_gmtyprog_validate(struct nv50_context *nv50)
 216 {
 217    struct nouveau_channel *chan = nv50->screen->base.channel;
 218    struct nv50_program *gp = nv50->gmtyprog;
 219
 220    if (!gp) /* GP_ENABLE is updated in linkage validation */
 221       return;
 222    if (!nv50_program_validate(nv50, gp))
 223       return;
 224
 225    BEGIN_RING(chan, RING_3D(GP_REG_ALLOC_TEMP), 1);
 226    OUT_RING  (chan, gp->max_gpr);
 227    BEGIN_RING(chan, RING_3D(GP_REG_ALLOC_RESULT), 1);
 228    OUT_RING  (chan, gp->max_out);
 229    BEGIN_RING(chan, RING_3D(GP_OUTPUT_PRIMITIVE_TYPE), 1);
 230    OUT_RING  (chan, gp->gp.prim_type);
 231    BEGIN_RING(chan, RING_3D(GP_VERTEX_OUTPUT_COUNT), 1);
 232    OUT_RING  (chan, gp->gp.vert_count);
 233    BEGIN_RING(chan, RING_3D(GP_START_ID), 1);
 234    OUT_RING  (chan, gp->code_base);
 235 }
 236
 237 static void
 238 nv50_sprite_coords_validate(struct nv50_context *nv50)
 239 {
 240    struct nouveau_channel *chan = nv50->screen->base.channel;
 241    uint32_t pntc[8], mode;
 242    struct nv50_program *fp = nv50->fragprog;
 243    unsigned i, c;
 244    unsigned m = (nv50->state.interpolant_ctrl >> 8) & 0xff;
 245
 246    if (!nv50->rast->pipe.point_quad_rasterization) {
 247       if (nv50->state.point_sprite) {
 248          BEGIN_RING(chan, RING_3D(POINT_COORD_REPLACE_MAP(0)), 8);
 249          for (i = 0; i < 8; ++i)
 250             OUT_RING(chan, 0);
 251
 252          nv50->state.point_sprite = FALSE;
 253       }
 254       return;
 255    } else {
 256       nv50->state.point_sprite = TRUE;
 257    }
 258
 259    memset(pntc, 0, sizeof(pntc));
 260
 261    for (i = 0; i < fp->in_nr; i++) {
 262       unsigned n = util_bitcount(fp->in[i].mask);
 263
 264       if (fp->in[i].sn != TGSI_SEMANTIC_GENERIC) {
 265          m += n;
 266          continue;
 267       }
 268       if (!(nv50->rast->pipe.sprite_coord_enable & (1 << fp->in[i].si))) {
 269          m += n;
 270          continue;
 271       }
 272
 273       for (c = 0; c < 4; ++c) {
 274          if (fp->in[i].mask & (1 << c)) {
 275             pntc[m / 8] |= (c + 1) << ((m % 8) * 4);
 276             ++m;
 277          }
 278       }
 279    }
 280
 281    if (nv50->rast->pipe.sprite_coord_mode == PIPE_SPRITE_COORD_LOWER_LEFT)
 282       mode = 0x00;
 283    else
 284       mode = 0x10;
 285
 286    BEGIN_RING(chan, RING_3D(POINT_SPRITE_CTRL), 1);
 287    OUT_RING  (chan, mode);
 288
 289    BEGIN_RING(chan, RING_3D(POINT_COORD_REPLACE_MAP(0)), 8);
 290    OUT_RINGp (chan, pntc, 8);
 291 }
 292
 293 /* Validate state derived from shaders and the rasterizer cso. */
 294 void
 295 nv50_validate_derived_rs(struct nv50_context *nv50)
 296 {
 297    struct nouveau_channel *chan = nv50->screen->base.channel;
 298    uint32_t color, psize;
 299
 300    nv50_sprite_coords_validate(nv50);
 301
 302    if (nv50->dirty & NV50_NEW_FRAGPROG)
 303       return;
 304    psize = nv50->state.semantic_psize & ~NV50_3D_MAP_SEMANTIC_3_PTSZ_EN__MASK;
 305    color = nv50->state.semantic_color & ~NV50_3D_MAP_SEMANTIC_0_CLMP_EN;
 306
 307    if (nv50->rast->pipe.clamp_vertex_color)
 308       color |= NV50_3D_MAP_SEMANTIC_0_CLMP_EN;
 309
 310    if (color != nv50->state.semantic_color) {
 311       nv50->state.semantic_color = color;
 312       BEGIN_RING(chan, RING_3D(MAP_SEMANTIC_0), 1);
 313       OUT_RING  (chan, color);
 314    }
 315
 316    if (nv50->rast->pipe.point_size_per_vertex)
 317       psize |= NV50_3D_MAP_SEMANTIC_3_PTSZ_EN__MASK;
 318
 319    if (psize != nv50->state.semantic_psize) {
 320       nv50->state.semantic_psize = psize;
 321       BEGIN_RING(chan, RING_3D(MAP_SEMANTIC_3), 1);
 322       OUT_RING  (chan, psize);
 323    }
 324 }
 325
 326 static int
 327 nv50_vec4_map(uint8_t *map, int mid, uint32_t lin[4],
 328               struct nv50_varying *in, struct nv50_varying *out)
 329 {
 330    int c;
 331    uint8_t mv = out->mask, mf = in->mask, oid = out->hw;
 332
 333    for (c = 0; c < 4; ++c) {
 334       if (mf & 1) {
 335          if (in->linear)
 336             lin[mid / 32] |= 1 << (mid % 32);
 337          if (mv & 1)
 338             map[mid] = oid;
 339          else
 340          if (c == 3)
 341             map[mid] |= 1;
 342          ++mid;
 343       }
 344
 345       oid += mv & 1;
 346       mf >>= 1;
 347       mv >>= 1;
 348    }
 349
 350    return mid;
 351 }
 352
 353 void
 354 nv50_fp_linkage_validate(struct nv50_context *nv50)
 355 {
 356    struct nouveau_channel *chan = nv50->screen->base.channel;
 357    struct nv50_program *vp = nv50->gmtyprog ? nv50->gmtyprog : nv50->vertprog;
 358    struct nv50_program *fp = nv50->fragprog;
 359    struct nv50_varying dummy;
 360    int i, n, c, m;
 361    uint32_t primid = 0;
 362    uint32_t psiz = 0x000;
 363    uint32_t interp = fp->fp.interp;
 364    uint32_t colors = fp->fp.colors;
 365    uint32_t lin[4];
 366    uint8_t map[64];
 367
 368    memset(lin, 0x00, sizeof(lin));
 369
 370    /* XXX: in buggy-endian mode, is the first element of map (u32)0x000000xx
 371     *  or is it the first byte ?
 372     */
 373    memset(map, nv50->gmtyprog ? 0x80 : 0x40, sizeof(map));
 374
 375    dummy.mask = 0xf; /* map all components of HPOS */
 376    dummy.linear = 0;
 377    m = nv50_vec4_map(map, 0, lin, &dummy, &vp->out[0]);
 378
 379    for (c = 0; c < vp->vp.clpd_nr; ++c)
 380       map[m++] = vp->vp.clpd + c;
 381
 382    colors |= m << 8; /* adjust BFC0 id */
 383
 384    /* if light_twoside is active, FFC0_ID == BFC0_ID is invalid */
 385    if (nv50->rast->pipe.light_twoside) {
 386       for (i = 0; i < 2; ++i)
 387          m = nv50_vec4_map(map, m, lin,
 388                            &fp->in[fp->vp.bfc[i]], &vp->out[vp->vp.bfc[i]]);
 389    }
 390    colors += m - 4; /* adjust FFC0 id */
 391    interp |= m << 8; /* set map id where 'normal' FP inputs start */
 392
 393    dummy.mask = 0x0;
 394    for (i = 0; i < fp->in_nr; ++i) {
 395       for (n = 0; n < vp->out_nr; ++n)
 396          if (vp->out[n].sn == fp->in[i].sn &&
 397              vp->out[n].si == fp->in[i].si)
 398             break;
 399       m = nv50_vec4_map(map, m, lin,
 400                         &fp->in[i], (n < vp->out_nr) ? &vp->out[n] : &dummy);
 401    }
 402
 403    /* PrimitiveID either is replaced by the system value, or
 404     * written by the geometry shader into an output register
 405     */
 406    if (fp->gp.primid < 0x40) {
 407       primid = m;
 408       map[m++] = vp->gp.primid;
 409    }
 410
 411    if (nv50->rast->pipe.point_size_per_vertex) {
 412       psiz = (m << 4) | 1;
 413       map[m++] = vp->vp.psiz;
 414    }
 415
 416    if (nv50->rast->pipe.clamp_vertex_color)
 417       colors |= NV50_3D_MAP_SEMANTIC_0_CLMP_EN;
 418
 419    n = (m + 3) / 4;
 420    assert(m <= 64);
 421
 422    if (unlikely(nv50->gmtyprog)) {
 423       BEGIN_RING(chan, RING_3D(GP_RESULT_MAP_SIZE), 1);
 424       OUT_RING  (chan, m);
 425       BEGIN_RING(chan, RING_3D(GP_RESULT_MAP(0)), n);
 426       OUT_RINGp (chan, map, n);
 427    } else {
 428       BEGIN_RING(chan, RING_3D(VP_GP_BUILTIN_ATTR_EN), 1);
 429       OUT_RING  (chan, vp->vp.attrs[2]);
 430
 431       BEGIN_RING(chan, RING_3D(MAP_SEMANTIC_4), 1);
 432       OUT_RING  (chan, primid);
 433
 434       BEGIN_RING(chan, RING_3D(VP_RESULT_MAP_SIZE), 1);
 435       OUT_RING  (chan, m);
 436       BEGIN_RING(chan, RING_3D(VP_RESULT_MAP(0)), n);
 437       OUT_RINGp (chan, map, n);
 438    }
 439
 440    BEGIN_RING(chan, RING_3D(MAP_SEMANTIC_0), 4);
 441    OUT_RING  (chan, colors);
 442    OUT_RING  (chan, (vp->vp.clpd_nr << 8) | 4);
 443    OUT_RING  (chan, 0);
 444    OUT_RING  (chan, psiz);
 445
 446    BEGIN_RING(chan, RING_3D(FP_INTERPOLANT_CTRL), 1);
 447    OUT_RING  (chan, interp);
 448
 449    nv50->state.interpolant_ctrl = interp;
 450
 451    nv50->state.semantic_color = colors;
 452    nv50->state.semantic_psize = psiz;
 453
 454    BEGIN_RING(chan, RING_3D(NOPERSPECTIVE_BITMAP(0)), 4);
 455    OUT_RINGp (chan, lin, 4);
 456
 457    BEGIN_RING(chan, RING_3D(GP_ENABLE), 1);
 458    OUT_RING  (chan, nv50->gmtyprog ? 1 : 0);
 459 }
 460
 461 static int
 462 nv50_vp_gp_mapping(uint8_t *map, int m,
 463                    struct nv50_program *vp, struct nv50_program *gp)
 464 {
 465    int i, j, c;
 466
 467    for (i = 0; i < gp->in_nr; ++i) {
 468       uint8_t oid = 0, mv = 0, mg = gp->in[i].mask;
 469
 470       for (j = 0; j < vp->out_nr; ++j) {
 471          if (vp->out[j].sn == gp->in[i].sn &&
 472              vp->out[j].si == gp->in[i].si) {
 473             mv = vp->out[j].mask;
 474             oid = vp->out[j].hw;
 475             break;
 476          }
 477       }
 478
 479       for (c = 0; c < 4; ++c, mv >>= 1, mg >>= 1) {
 480          if (mg & mv & 1)
 481             map[m++] = oid;
 482          else
 483          if (mg & 1)
 484             map[m++] = (c == 3) ? 0x41 : 0x40;
 485          oid += mv & 1;
 486       }
 487    }
 488    return m;
 489 }
 490
 491 void
 492 nv50_gp_linkage_validate(struct nv50_context *nv50)
 493 {
 494    struct nouveau_channel *chan = nv50->screen->base.channel;
 495    struct nv50_program *vp = nv50->vertprog;
 496    struct nv50_program *gp = nv50->gmtyprog;
 497    int m = 0;
 498    int n;
 499    uint8_t map[64];
 500
 501    if (!gp)
 502       return;
 503    memset(map, 0, sizeof(map));
 504
 505    m = nv50_vp_gp_mapping(map, m, vp, gp);
 506
 507    n = (m + 3) / 4;
 508
 509    BEGIN_RING(chan, RING_3D(VP_GP_BUILTIN_ATTR_EN), 1);
 510    OUT_RING  (chan, vp->vp.attrs[2] | gp->vp.attrs[2]);
 511
 512    BEGIN_RING(chan, RING_3D(VP_RESULT_MAP_SIZE), 1);
 513    OUT_RING  (chan, m);
 514    BEGIN_RING(chan, RING_3D(VP_RESULT_MAP(0)), n);
 515    OUT_RINGp (chan, map, n);
 516 }