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       prog->translated = nv50_program_translate(prog);
 135       if (!prog->translated)
 136          return FALSE;
 137    } else
 138    if (prog->res)
 139       return TRUE;
 140
 141    if (prog->type == PIPE_SHADER_FRAGMENT) heap = nv50->screen->fp_code_heap;
 142    else
 143    if (prog->type == PIPE_SHADER_GEOMETRY) heap = nv50->screen->gp_code_heap;
 144    else
 145       heap = nv50->screen->vp_code_heap;
 146
 147    size = align(prog->code_size, 0x100);
 148
 149    ret = nouveau_resource_alloc(heap, size, prog, &prog->res);
 150    if (ret) {
 151       NOUVEAU_ERR("out of code space for shader type %i\n", prog->type);
 152       return FALSE;
 153    }
 154    prog->code_base = prog->res->start;
 155
 156    nv50_relocate_program(prog, prog->code_base, 0);
 157
 158    nv50_sifc_linear_u8(&nv50->base, nv50->screen->code,
 159                        (prog->type << NV50_CODE_BO_SIZE_LOG2) + prog->code_base,
 160                        NOUVEAU_BO_VRAM, prog->code_size, prog->code);
 161
 162    BEGIN_RING(nv50->screen->base.channel, RING_3D(CODE_CB_FLUSH), 1);
 163    OUT_RING  (nv50->screen->base.channel, 0);
 164
 165    return TRUE;
 166 }
 167
 168 void
 169 nv50_vertprog_validate(struct nv50_context *nv50)
 170 {
 171    struct nouveau_channel *chan = nv50->screen->base.channel;
 172    struct nv50_program *vp = nv50->vertprog;
 173
 174    if (!nv50_program_validate(nv50, vp))
 175          return;
 176
 177    BEGIN_RING(chan, RING_3D(VP_ATTR_EN(0)), 2);
 178    OUT_RING  (chan, vp->vp.attrs[0]);
 179    OUT_RING  (chan, vp->vp.attrs[1]);
 180    BEGIN_RING(chan, RING_3D(VP_REG_ALLOC_RESULT), 1);
 181    OUT_RING  (chan, vp->max_out);
 182    BEGIN_RING(chan, RING_3D(VP_REG_ALLOC_TEMP), 1);
 183    OUT_RING  (chan, vp->max_gpr);
 184    BEGIN_RING(chan, RING_3D(VP_START_ID), 1);
 185    OUT_RING  (chan, vp->code_base);
 186 }
 187
 188 void
 189 nv50_fragprog_validate(struct nv50_context *nv50)
 190 {
 191    struct nouveau_channel *chan = nv50->screen->base.channel;
 192    struct nv50_program *fp = nv50->fragprog;
 193
 194    if (!nv50_program_validate(nv50, fp))
 195          return;
 196
 197    BEGIN_RING(chan, RING_3D(FP_REG_ALLOC_TEMP), 1);
 198    OUT_RING  (chan, fp->max_gpr);
 199    BEGIN_RING(chan, RING_3D(FP_RESULT_COUNT), 1);
 200    OUT_RING  (chan, fp->max_out);
 201    BEGIN_RING(chan, RING_3D(FP_CONTROL), 1);
 202    OUT_RING  (chan, fp->fp.flags[0]);
 203    BEGIN_RING(chan, RING_3D(FP_CTRL_UNK196C), 1);
 204    OUT_RING  (chan, fp->fp.flags[1]);
 205    BEGIN_RING(chan, RING_3D(FP_START_ID), 1);
 206    OUT_RING  (chan, fp->code_base);
 207 }
 208
 209 void
 210 nv50_gmtyprog_validate(struct nv50_context *nv50)
 211 {
 212    struct nouveau_channel *chan = nv50->screen->base.channel;
 213    struct nv50_program *gp = nv50->gmtyprog;
 214
 215    if (!gp) /* GP_ENABLE is updated in linkage validation */
 216       return;
 217    if (!nv50_program_validate(nv50, gp))
 218       return;
 219
 220    BEGIN_RING(chan, RING_3D(GP_REG_ALLOC_TEMP), 1);
 221    OUT_RING  (chan, gp->max_gpr);
 222    BEGIN_RING(chan, RING_3D(GP_REG_ALLOC_RESULT), 1);
 223    OUT_RING  (chan, gp->max_out);
 224    BEGIN_RING(chan, RING_3D(GP_OUTPUT_PRIMITIVE_TYPE), 1);
 225    OUT_RING  (chan, gp->gp.prim_type);
 226    BEGIN_RING(chan, RING_3D(GP_VERTEX_OUTPUT_COUNT), 1);
 227    OUT_RING  (chan, gp->gp.vert_count);
 228    BEGIN_RING(chan, RING_3D(GP_START_ID), 1);
 229    OUT_RING  (chan, gp->code_base);
 230 }
 231
 232 static void
 233 nv50_sprite_coords_validate(struct nv50_context *nv50)
 234 {
 235    struct nouveau_channel *chan = nv50->screen->base.channel;
 236    uint32_t pntc[8], mode;
 237    struct nv50_program *fp = nv50->fragprog;
 238    unsigned i, c;
 239    unsigned m = (nv50->state.interpolant_ctrl >> 8) & 0xff;
 240
 241    if (!nv50->rast->pipe.point_quad_rasterization) {
 242       if (nv50->state.point_sprite) {
 243          BEGIN_RING(chan, RING_3D(POINT_COORD_REPLACE_MAP(0)), 8);
 244          for (i = 0; i < 8; ++i)
 245             OUT_RING(chan, 0);
 246
 247          nv50->state.point_sprite = FALSE;
 248       }
 249       return;
 250    } else {
 251       nv50->state.point_sprite = TRUE;
 252    }
 253
 254    memset(pntc, 0, sizeof(pntc));
 255
 256    for (i = 0; i < fp->in_nr; i++) {
 257       unsigned n = util_bitcount(fp->in[i].mask);
 258
 259       if (fp->in[i].sn != TGSI_SEMANTIC_GENERIC) {
 260          m += n;
 261          continue;
 262       }
 263       if (!(nv50->rast->pipe.sprite_coord_enable & (1 << fp->in[i].si))) {
 264          m += n;
 265          continue;
 266       }
 267
 268       for (c = 0; c < 4; ++c) {
 269          if (fp->in[i].mask & (1 << c)) {
 270             pntc[m / 8] |= (c + 1) << ((m % 8) * 4);
 271             ++m;
 272          }
 273       }
 274    }
 275
 276    if (nv50->rast->pipe.sprite_coord_mode == PIPE_SPRITE_COORD_LOWER_LEFT)
 277       mode = 0x00;
 278    else
 279       mode = 0x10;
 280
 281    BEGIN_RING(chan, RING_3D(POINT_SPRITE_CTRL), 1);
 282    OUT_RING  (chan, mode);
 283
 284    BEGIN_RING(chan, RING_3D(POINT_COORD_REPLACE_MAP(0)), 8);
 285    OUT_RINGp (chan, pntc, 8);
 286 }
 287
 288 /* Validate state derived from shaders and the rasterizer cso. */
 289 void
 290 nv50_validate_derived_rs(struct nv50_context *nv50)
 291 {
 292    struct nouveau_channel *chan = nv50->screen->base.channel;
 293    uint32_t color, psize;
 294
 295    nv50_sprite_coords_validate(nv50);
 296
 297    if (nv50->dirty & NV50_NEW_FRAGPROG)
 298       return;
 299    psize = nv50->state.semantic_psize & ~NV50_3D_MAP_SEMANTIC_3_PTSZ_EN__MASK;
 300    color = nv50->state.semantic_color & ~NV50_3D_MAP_SEMANTIC_0_CLMP_EN;
 301
 302    if (nv50->rast->pipe.clamp_vertex_color)
 303       color |= NV50_3D_MAP_SEMANTIC_0_CLMP_EN;
 304
 305    if (color != nv50->state.semantic_color) {
 306       nv50->state.semantic_color = color;
 307       BEGIN_RING(chan, RING_3D(MAP_SEMANTIC_0), 1);
 308       OUT_RING  (chan, color);
 309    }
 310
 311    if (nv50->rast->pipe.point_size_per_vertex)
 312       psize |= NV50_3D_MAP_SEMANTIC_3_PTSZ_EN__MASK;
 313
 314    if (psize != nv50->state.semantic_psize) {
 315       nv50->state.semantic_psize = psize;
 316       BEGIN_RING(chan, RING_3D(MAP_SEMANTIC_3), 1);
 317       OUT_RING  (chan, psize);
 318    }
 319 }
 320
 321 static int
 322 nv50_vec4_map(uint8_t *map, int mid, uint32_t lin[4],
 323               struct nv50_varying *in, struct nv50_varying *out)
 324 {
 325    int c;
 326    uint8_t mv = out->mask, mf = in->mask, oid = out->hw;
 327
 328    for (c = 0; c < 4; ++c) {
 329       if (mf & 1) {
 330          if (in->linear)
 331             lin[mid / 32] |= 1 << (mid % 32);
 332          if (mv & 1)
 333             map[mid] = oid;
 334          else
 335          if (c == 3)
 336             map[mid] |= 1;
 337          ++mid;
 338       }
 339
 340       oid += mv & 1;
 341       mf >>= 1;
 342       mv >>= 1;
 343    }
 344
 345    return mid;
 346 }
 347
 348 void
 349 nv50_fp_linkage_validate(struct nv50_context *nv50)
 350 {
 351    struct nouveau_channel *chan = nv50->screen->base.channel;
 352    struct nv50_program *vp = nv50->gmtyprog ? nv50->gmtyprog : nv50->vertprog;
 353    struct nv50_program *fp = nv50->fragprog;
 354    struct nv50_varying dummy;
 355    int i, n, c, m;
 356    uint32_t primid = 0;
 357    uint32_t psiz = 0x000;
 358    uint32_t interp = fp->fp.interp;
 359    uint32_t colors = fp->fp.colors;
 360    uint32_t lin[4];
 361    uint8_t map[64];
 362
 363    memset(lin, 0x00, sizeof(lin));
 364
 365    /* XXX: in buggy-endian mode, is the first element of map (u32)0x000000xx
 366     *  or is it the first byte ?
 367     */
 368    memset(map, nv50->gmtyprog ? 0x80 : 0x40, sizeof(map));
 369
 370    dummy.mask = 0xf; /* map all components of HPOS */
 371    dummy.linear = 0;
 372    m = nv50_vec4_map(map, 0, lin, &dummy, &vp->out[0]);
 373
 374    for (c = 0; c < vp->vp.clpd_nr; ++c)
 375       map[m++] = vp->vp.clpd + c;
 376
 377    colors |= m << 8; /* adjust BFC0 id */
 378
 379    /* if light_twoside is active, FFC0_ID == BFC0_ID is invalid */
 380    if (nv50->rast->pipe.light_twoside) {
 381       for (i = 0; i < 2; ++i)
 382          m = nv50_vec4_map(map, m, lin,
 383                            &fp->in[fp->vp.bfc[i]], &vp->out[vp->vp.bfc[i]]);
 384    }
 385    colors += m - 4; /* adjust FFC0 id */
 386    interp |= m << 8; /* set map id where 'normal' FP inputs start */
 387
 388    dummy.mask = 0x0;
 389    for (i = 0; i < fp->in_nr; ++i) {
 390       for (n = 0; n < vp->out_nr; ++n)
 391          if (vp->out[n].sn == fp->in[i].sn &&
 392              vp->out[n].si == fp->in[i].si)
 393             break;
 394       m = nv50_vec4_map(map, m, lin,
 395                         &fp->in[i], (n < vp->out_nr) ? &vp->out[n] : &dummy);
 396    }
 397
 398    /* PrimitiveID either is replaced by the system value, or
 399     * written by the geometry shader into an output register
 400     */
 401    if (fp->gp.primid < 0x40) {
 402       primid = m;
 403       map[m++] = vp->gp.primid;
 404    }
 405
 406    if (nv50->rast->pipe.point_size_per_vertex) {
 407       psiz = (m << 4) | 1;
 408       map[m++] = vp->vp.psiz;
 409    }
 410
 411    if (nv50->rast->pipe.clamp_vertex_color)
 412       colors |= NV50_3D_MAP_SEMANTIC_0_CLMP_EN;
 413
 414    n = (m + 3) / 4;
 415    assert(m <= 64);
 416
 417    if (unlikely(nv50->gmtyprog)) {
 418       BEGIN_RING(chan, RING_3D(GP_RESULT_MAP_SIZE), 1);
 419       OUT_RING  (chan, m);
 420       BEGIN_RING(chan, RING_3D(GP_RESULT_MAP(0)), n);
 421       OUT_RINGp (chan, map, n);
 422    } else {
 423       BEGIN_RING(chan, RING_3D(VP_GP_BUILTIN_ATTR_EN), 1);
 424       OUT_RING  (chan, vp->vp.attrs[2]);
 425
 426       BEGIN_RING(chan, RING_3D(MAP_SEMANTIC_4), 1);
 427       OUT_RING  (chan, primid);
 428
 429       BEGIN_RING(chan, RING_3D(VP_RESULT_MAP_SIZE), 1);
 430       OUT_RING  (chan, m);
 431       BEGIN_RING(chan, RING_3D(VP_RESULT_MAP(0)), n);
 432       OUT_RINGp (chan, map, n);
 433    }
 434
 435    BEGIN_RING(chan, RING_3D(MAP_SEMANTIC_0), 4);
 436    OUT_RING  (chan, colors);
 437    OUT_RING  (chan, (vp->vp.clpd_nr << 8) | 4);
 438    OUT_RING  (chan, 0);
 439    OUT_RING  (chan, psiz);
 440
 441    BEGIN_RING(chan, RING_3D(FP_INTERPOLANT_CTRL), 1);
 442    OUT_RING  (chan, interp);
 443
 444    nv50->state.interpolant_ctrl = interp;
 445
 446    nv50->state.semantic_color = colors;
 447    nv50->state.semantic_psize = psiz;
 448
 449    BEGIN_RING(chan, RING_3D(NOPERSPECTIVE_BITMAP(0)), 4);
 450    OUT_RINGp (chan, lin, 4);
 451
 452    BEGIN_RING(chan, RING_3D(GP_ENABLE), 1);
 453    OUT_RING  (chan, nv50->gmtyprog ? 1 : 0);
 454 }
 455
 456 static int
 457 nv50_vp_gp_mapping(uint8_t *map, int m,
 458                    struct nv50_program *vp, struct nv50_program *gp)
 459 {
 460    int i, j, c;
 461
 462    for (i = 0; i < gp->in_nr; ++i) {
 463       uint8_t oid = 0, mv = 0, mg = gp->in[i].mask;
 464
 465       for (j = 0; j < vp->out_nr; ++j) {
 466          if (vp->out[j].sn == gp->in[i].sn &&
 467              vp->out[j].si == gp->in[i].si) {
 468             mv = vp->out[j].mask;
 469             oid = vp->out[j].hw;
 470             break;
 471          }
 472       }
 473
 474       for (c = 0; c < 4; ++c, mv >>= 1, mg >>= 1) {
 475          if (mg & mv & 1)
 476             map[m++] = oid;
 477          else
 478          if (mg & 1)
 479             map[m++] = (c == 3) ? 0x41 : 0x40;
 480          oid += mv & 1;
 481       }
 482    }
 483    return m;
 484 }
 485
 486 void
 487 nv50_gp_linkage_validate(struct nv50_context *nv50)
 488 {
 489    struct nouveau_channel *chan = nv50->screen->base.channel;
 490    struct nv50_program *vp = nv50->vertprog;
 491    struct nv50_program *gp = nv50->gmtyprog;
 492    int m = 0;
 493    int n;
 494    uint8_t map[64];
 495
 496    if (!gp)
 497       return;
 498    memset(map, 0, sizeof(map));
 499
 500    m = nv50_vp_gp_mapping(map, m, vp, gp);
 501
 502    n = (m + 3) / 4;
 503
 504    BEGIN_RING(chan, RING_3D(VP_GP_BUILTIN_ATTR_EN), 1);
 505    OUT_RING  (chan, vp->vp.attrs[2] | gp->vp.attrs[2]);
 506
 507    BEGIN_RING(chan, RING_3D(VP_RESULT_MAP_SIZE), 1);
 508    OUT_RING  (chan, m);
 509    BEGIN_RING(chan, RING_3D(VP_RESULT_MAP(0)), n);
 510    OUT_RINGp (chan, map, n);
 511 }