src/gallium/drivers/svga/svga_state_constants.c

   1
   2 /**********************************************************
   3  * Copyright 2008-2009 VMware, Inc.  All rights reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person
   6  * obtaining a copy of this software and associated documentation
   7  * files (the "Software"), to deal in the Software without
   8  * restriction, including without limitation the rights to use, copy,
   9  * modify, merge, publish, distribute, sublicense, and/or sell copies
  10  * of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be
  14  * included in all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  17  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  18  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  19  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  20  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  21  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  22  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  23  * SOFTWARE.
  24  *
  25  **********************************************************/
  26
  27 #include "util/u_format.h"
  28 #include "util/u_inlines.h"
  29 #include "util/u_memory.h"
  30 #include "pipe/p_defines.h"
  31 #include "util/u_upload_mgr.h"
  32
  33 #include "svga_screen.h"
  34 #include "svga_context.h"
  35 #include "svga_state.h"
  36 #include "svga_cmd.h"
  37 #include "svga_tgsi.h"
  38 #include "svga_debug.h"
  39 #include "svga_resource_buffer.h"
  40 #include "svga_shader.h"
  41
  42 #include "svga_hw_reg.h"
  43
  44
  45 /*
  46  * Don't try to send more than 4kb of successive constants.
  47  */
  48 #define MAX_CONST_REG_COUNT 256  /**< number of float[4] constants */
  49
  50 /**
  51  * Extra space for svga-specific VS/PS constants (such as texcoord
  52  * scale factors, vertex transformation scale/translation).
  53  */
  54 #define MAX_EXTRA_CONSTS 32
  55
  56 /** Guest-backed surface constant buffers must be this size */
  57 #define GB_CONSTBUF_SIZE (SVGA3D_CONSTREG_MAX)
  58
  59
  60 /**
  61  * Emit any extra shader-type-independent shader constants into the buffer
  62  * pointed to by 'dest'.
  63  * \return number of float[4] constants put into the 'dest' buffer
  64  */
  65 static unsigned
  66 svga_get_extra_constants_common(struct svga_context *svga,
  67                                 const struct svga_shader_variant *variant,
  68                                 unsigned shader, float *dest)
  69 {
  70    uint32_t *dest_u = (uint32_t *) dest;  // uint version of dest
  71    unsigned i;
  72    unsigned count = 0;
  73
  74    for (i = 0; i < variant->key.num_textures; i++) {
  75       struct pipe_sampler_view *sv = svga->curr.sampler_views[shader][i];
  76       if (sv) {
  77          struct pipe_resource *tex = sv->texture;
  78          /* Scaling factors needed for handling unnormalized texture coordinates
  79           * for texture rectangles.
  80           */
  81          if (variant->key.tex[i].unnormalized) {
  82             /* debug/sanity check */
  83             assert(variant->key.tex[i].width_height_idx == count);
  84
  85             *dest++ = 1.0 / (float)tex->width0;
  86             *dest++ = 1.0 / (float)tex->height0;
  87             *dest++ = 1.0;
  88             *dest++ = 1.0;
  89
  90             count++;
  91          }
  92
  93          /* Store the sizes for texture buffers.
  94          */
  95          if (tex->target == PIPE_BUFFER) {
  96             unsigned bytes_per_element = util_format_get_blocksize(sv->format);
  97             *dest_u++ = tex->width0 / bytes_per_element;
  98             *dest_u++ = 1;
  99             *dest_u++ = 1;
 100             *dest_u++ = 1;
 101
 102             count++;
 103          }
 104       }
 105    }
 106
 107    return count;
 108 }
 109
 110
 111 /**
 112  * Emit any extra fragment shader constants into the buffer pointed
 113  * to by 'dest'.
 114  * \return number of float[4] constants put into the dest buffer
 115  */
 116 static unsigned
 117 svga_get_extra_fs_constants(struct svga_context *svga, float *dest)
 118 {
 119    const struct svga_shader_variant *variant = svga->state.hw_draw.fs;
 120    unsigned count = 0;
 121
 122    count += svga_get_extra_constants_common(svga, variant,
 123                                             PIPE_SHADER_FRAGMENT, dest);
 124
 125    assert(count <= MAX_EXTRA_CONSTS);
 126
 127    return count;
 128 }
 129
 130 /**
 131  * Emit extra constants needed for prescale computation into the
 132  * the buffer pointed to by '*dest'. The updated buffer pointer
 133  * will be returned in 'dest'.
 134  */
 135 static unsigned
 136 svga_get_prescale_constants(struct svga_context *svga, float **dest)
 137 {
 138    memcpy(*dest, svga->state.hw_clear.prescale.scale, 4 * sizeof(float));
 139    *dest += 4;
 140
 141    memcpy(*dest, svga->state.hw_clear.prescale.translate, 4 * sizeof(float));
 142    *dest += 4;
 143
 144    return 2;
 145 }
 146
 147 /**
 148  * Emit extra constants needed for point sprite emulation.
 149  */
 150 static unsigned
 151 svga_get_pt_sprite_constants(struct svga_context *svga, float **dest)
 152 {
 153    struct svga_screen *screen = svga_screen(svga->pipe.screen);
 154    float *dst = *dest;
 155
 156    dst[0] = 1.0 / (svga->curr.viewport.scale[0] * 2);
 157    dst[1] = 1.0 / (svga->curr.viewport.scale[1] * 2);
 158    dst[2] = svga->curr.rast->pointsize;
 159    dst[3] = screen->maxPointSize;
 160    *dest = *dest + 4;
 161    return 1;
 162 }
 163
 164 /**
 165  * Emit user-defined clip plane coefficients into the buffer pointed to
 166  * by '*dest'. The updated buffer pointer will be returned in 'dest'.
 167  */
 168 static unsigned
 169 svga_get_clip_plane_constants(struct svga_context *svga,
 170                               const struct svga_shader_variant *variant,
 171                               float **dest)
 172 {
 173    unsigned count = 0;
 174
 175    /* SVGA_NEW_CLIP */
 176    if (svga_have_vgpu10(svga)) {
 177       /* append user-defined clip plane coefficients onto constant buffer */
 178       unsigned clip_planes = variant->key.clip_plane_enable;
 179       while (clip_planes) {
 180          int i = u_bit_scan(&clip_planes);
 181          COPY_4V(*dest, svga->curr.clip.ucp[i]);
 182          *dest += 4;
 183          count += 1;
 184       }
 185    }
 186    return count;
 187 }
 188
 189 /**
 190  * Emit any extra vertex shader constants into the buffer pointed
 191  * to by 'dest'.
 192  * In particular, these would be the scale and bias factors computed
 193  * from the framebuffer size which are used to copy with differences in
 194  * GL vs D3D coordinate spaces.  See svga_tgsi_insn.c for more info.
 195  * \return number of float[4] constants put into the dest buffer
 196  */
 197 static unsigned
 198 svga_get_extra_vs_constants(struct svga_context *svga, float *dest)
 199 {
 200    const struct svga_shader_variant *variant = svga->state.hw_draw.vs;
 201    unsigned count = 0;
 202
 203    /* SVGA_NEW_VS_VARIANT
 204     */
 205    if (variant->key.vs.need_prescale) {
 206       count += svga_get_prescale_constants(svga, &dest);
 207    }
 208
 209    if (variant->key.vs.undo_viewport) {
 210       /* Used to convert window coords back to NDC coords */
 211       dest[0] = 1.0f / svga->curr.viewport.scale[0];
 212       dest[1] = 1.0f / svga->curr.viewport.scale[1];
 213       dest[2] = -svga->curr.viewport.translate[0];
 214       dest[3] = -svga->curr.viewport.translate[1];
 215       dest += 4;
 216       count += 1;
 217    }
 218
 219    /* SVGA_NEW_CLIP */
 220    count += svga_get_clip_plane_constants(svga, variant, &dest);
 221
 222    /* common constants */
 223    count += svga_get_extra_constants_common(svga, variant,
 224                                             PIPE_SHADER_VERTEX, dest);
 225
 226    assert(count <= MAX_EXTRA_CONSTS);
 227
 228    return count;
 229 }
 230
 231 /**
 232  * Emit any extra geometry shader constants into the buffer pointed
 233  * to by 'dest'.
 234  */
 235 static unsigned
 236 svga_get_extra_gs_constants(struct svga_context *svga, float *dest)
 237 {
 238    const struct svga_shader_variant *variant = svga->state.hw_draw.gs;
 239    unsigned count = 0;
 240
 241    /* SVGA_NEW_GS_VARIANT
 242     */
 243
 244    /* Constants for point sprite
 245     * These are used in the transformed gs that supports point sprite.
 246     * They need to be added before the prescale constants.
 247     */
 248    if (variant->key.gs.wide_point) {
 249       count += svga_get_pt_sprite_constants(svga, &dest);
 250    }
 251
 252    if (variant->key.gs.need_prescale) {
 253       count += svga_get_prescale_constants(svga, &dest);
 254    }
 255
 256    /* SVGA_NEW_CLIP */
 257    count += svga_get_clip_plane_constants(svga, variant, &dest);
 258
 259    /* common constants */
 260    count += svga_get_extra_constants_common(svga, variant,
 261                                             PIPE_SHADER_GEOMETRY, dest);
 262
 263    assert(count <= MAX_EXTRA_CONSTS);
 264    return count;
 265 }
 266
 267 /**
 268  * Check and emit one shader constant register.
 269  * \param shader  PIPE_SHADER_FRAGMENT or PIPE_SHADER_VERTEX
 270  * \param i  which float[4] constant to change
 271  * \param value  the new float[4] value
 272  */
 273 static enum pipe_error
 274 emit_const(struct svga_context *svga, unsigned shader, unsigned i,
 275            const float *value)
 276 {
 277    enum pipe_error ret = PIPE_OK;
 278
 279    assert(shader < PIPE_SHADER_TYPES);
 280    assert(i < SVGA3D_CONSTREG_MAX);
 281    assert(!svga_have_vgpu10(svga));
 282
 283    if (memcmp(svga->state.hw_draw.cb[shader][i], value,
 284               4 * sizeof(float)) != 0) {
 285       if (SVGA_DEBUG & DEBUG_CONSTS)
 286          debug_printf("%s %s %u: %f %f %f %f\n",
 287                       __FUNCTION__,
 288                       shader == PIPE_SHADER_VERTEX ? "VERT" : "FRAG",
 289                       i,
 290                       value[0],
 291                       value[1],
 292                       value[2],
 293                       value[3]);
 294
 295       ret = SVGA3D_SetShaderConst( svga->swc,
 296                                    i,
 297                                    svga_shader_type(shader),
 298                                    SVGA3D_CONST_TYPE_FLOAT,
 299                                    value );
 300       if (ret != PIPE_OK)
 301          return ret;
 302
 303       memcpy(svga->state.hw_draw.cb[shader][i], value, 4 * sizeof(float));
 304
 305       svga->hud.num_const_updates++;
 306    }
 307
 308    return ret;
 309 }
 310
 311
 312 /*
 313  * Check and emit a range of shader constant registers, trying to coalesce
 314  * successive shader constant updates in a single command in order to save
 315  * space on the command buffer.  This is a HWv8 feature.
 316  */
 317 static enum pipe_error
 318 emit_const_range(struct svga_context *svga,
 319                  unsigned shader,
 320                  unsigned offset,
 321                  unsigned count,
 322                  const float (*values)[4])
 323 {
 324    unsigned i, j;
 325    enum pipe_error ret;
 326
 327    assert(shader == PIPE_SHADER_VERTEX ||
 328           shader == PIPE_SHADER_FRAGMENT);
 329    assert(!svga_have_vgpu10(svga));
 330
 331 #ifdef DEBUG
 332    if (offset + count > SVGA3D_CONSTREG_MAX) {
 333       debug_printf("svga: too many constants (offset %u + count %u = %u (max = %u))\n",
 334                    offset, count, offset + count, SVGA3D_CONSTREG_MAX);
 335    }
 336 #endif
 337
 338    if (offset > SVGA3D_CONSTREG_MAX) {
 339       /* This isn't OK, but if we propagate an error all the way up we'll
 340        * just get into more trouble.
 341        * XXX note that offset is always zero at this time so this is moot.
 342        */
 343       return PIPE_OK;
 344    }
 345
 346    if (offset + count > SVGA3D_CONSTREG_MAX) {
 347       /* Just drop the extra constants for now.
 348        * Ideally we should not have allowed the app to create a shader
 349        * that exceeds our constant buffer size but there's no way to
 350        * express that in gallium at this time.
 351        */
 352       count = SVGA3D_CONSTREG_MAX - offset;
 353    }
 354
 355    i = 0;
 356    while (i < count) {
 357       if (memcmp(svga->state.hw_draw.cb[shader][offset + i],
 358                  values[i],
 359                  4 * sizeof(float)) != 0) {
 360          /* Found one dirty constant
 361           */
 362          if (SVGA_DEBUG & DEBUG_CONSTS)
 363             debug_printf("%s %s %d: %f %f %f %f\n",
 364                          __FUNCTION__,
 365                          shader == PIPE_SHADER_VERTEX ? "VERT" : "FRAG",
 366                          offset + i,
 367                          values[i][0],
 368                          values[i][1],
 369                          values[i][2],
 370                          values[i][3]);
 371
 372          /* Look for more consecutive dirty constants.
 373           */
 374          j = i + 1;
 375          while (j < count &&
 376                 j < i + MAX_CONST_REG_COUNT &&
 377                 memcmp(svga->state.hw_draw.cb[shader][offset + j],
 378                        values[j],
 379                        4 * sizeof(float)) != 0) {
 380
 381             if (SVGA_DEBUG & DEBUG_CONSTS)
 382                debug_printf("%s %s %d: %f %f %f %f\n",
 383                             __FUNCTION__,
 384                             shader == PIPE_SHADER_VERTEX ? "VERT" : "FRAG",
 385                             offset + j,
 386                             values[j][0],
 387                             values[j][1],
 388                             values[j][2],
 389                             values[j][3]);
 390
 391             ++j;
 392          }
 393
 394          assert(j >= i + 1);
 395
 396          /* Send them all together.
 397           */
 398          if (svga_have_gb_objects(svga)) {
 399             ret = SVGA3D_SetGBShaderConstsInline(svga->swc,
 400                                                  offset + i, /* start */
 401                                                  j - i,  /* count */
 402                                                  svga_shader_type(shader),
 403                                                  SVGA3D_CONST_TYPE_FLOAT,
 404                                                  values + i);
 405          }
 406          else {
 407             ret = SVGA3D_SetShaderConsts(svga->swc,
 408                                          offset + i, j - i,
 409                                          svga_shader_type(shader),
 410                                          SVGA3D_CONST_TYPE_FLOAT,
 411                                          values + i);
 412          }
 413          if (ret != PIPE_OK) {
 414             return ret;
 415          }
 416
 417          /*
 418           * Local copy of the hardware state.
 419           */
 420          memcpy(svga->state.hw_draw.cb[shader][offset + i],
 421                 values[i],
 422                 (j - i) * 4 * sizeof(float));
 423
 424          i = j + 1;
 425
 426          svga->hud.num_const_updates++;
 427
 428       } else {
 429          ++i;
 430       }
 431    }
 432
 433    return PIPE_OK;
 434 }
 435
 436
 437 /**
 438  * Emit all the constants in a constant buffer for a shader stage.
 439  * On VGPU10, emit_consts_vgpu10 is used instead.
 440  */
 441 static enum pipe_error
 442 emit_consts_vgpu9(struct svga_context *svga, unsigned shader)
 443 {
 444    const struct pipe_constant_buffer *cbuf;
 445    struct svga_screen *ss = svga_screen(svga->pipe.screen);
 446    struct pipe_transfer *transfer = NULL;
 447    unsigned count;
 448    const float (*data)[4] = NULL;
 449    unsigned i;
 450    enum pipe_error ret = PIPE_OK;
 451    const unsigned offset = 0;
 452
 453    assert(shader < PIPE_SHADER_TYPES);
 454    assert(!svga_have_vgpu10(svga));
 455    /* Only one constant buffer per shader is supported before VGPU10.
 456     * This is only an approximate check against that.
 457     */
 458    assert(svga->curr.constbufs[shader][1].buffer == NULL);
 459
 460    cbuf = &svga->curr.constbufs[shader][0];
 461
 462    if (svga->curr.constbufs[shader][0].buffer) {
 463       /* emit user-provided constants */
 464       data = (const float (*)[4])
 465          pipe_buffer_map(&svga->pipe, svga->curr.constbufs[shader][0].buffer,
 466                          PIPE_TRANSFER_READ, &transfer);
 467       if (!data) {
 468          return PIPE_ERROR_OUT_OF_MEMORY;
 469       }
 470
 471       /* sanity check */
 472       assert(cbuf->buffer->width0 >=
 473              cbuf->buffer_size);
 474
 475       /* Use/apply the constant buffer size and offsets here */
 476       count = cbuf->buffer_size / (4 * sizeof(float));
 477       data += cbuf->buffer_offset / (4 * sizeof(float));
 478
 479       if (ss->hw_version >= SVGA3D_HWVERSION_WS8_B1) {
 480          ret = emit_const_range( svga, shader, offset, count, data );
 481       }
 482       else {
 483          for (i = 0; i < count; i++) {
 484             ret = emit_const( svga, shader, offset + i, data[i] );
 485             if (ret != PIPE_OK) {
 486                break;
 487             }
 488          }
 489       }
 490
 491       pipe_buffer_unmap(&svga->pipe, transfer);
 492
 493       if (ret != PIPE_OK) {
 494          return ret;
 495       }
 496    }
 497
 498    /* emit extra shader constants */
 499    {
 500       const struct svga_shader_variant *variant = NULL;
 501       unsigned offset;
 502       float extras[MAX_EXTRA_CONSTS][4];
 503       unsigned count, i;
 504
 505       switch (shader) {
 506       case PIPE_SHADER_VERTEX:
 507          variant = svga->state.hw_draw.vs;
 508          count = svga_get_extra_vs_constants(svga, (float *) extras);
 509          break;
 510       case PIPE_SHADER_FRAGMENT:
 511          variant = svga->state.hw_draw.fs;
 512          count = svga_get_extra_fs_constants(svga, (float *) extras);
 513          break;
 514       default:
 515          assert(!"Unexpected shader type");
 516          count = 0;
 517       }
 518
 519       assert(variant);
 520       offset = variant->shader->info.file_max[TGSI_FILE_CONSTANT] + 1;
 521       assert(count <= Elements(extras));
 522
 523       if (count > 0) {
 524          if (ss->hw_version >= SVGA3D_HWVERSION_WS8_B1) {
 525             ret = emit_const_range(svga, shader, offset, count,
 526                                    (const float (*) [4])extras);
 527          }
 528          else {
 529             for (i = 0; i < count; i++) {
 530                ret = emit_const(svga, shader, offset + i, extras[i]);
 531                if (ret != PIPE_OK)
 532                   return ret;
 533             }
 534          }
 535       }
 536    }
 537
 538    return ret;
 539 }
 540
 541
 542
 543 static enum pipe_error
 544 emit_constbuf_vgpu10(struct svga_context *svga, unsigned shader)
 545 {
 546    const struct pipe_constant_buffer *cbuf;
 547    struct pipe_resource *dst_buffer = NULL;
 548    enum pipe_error ret = PIPE_OK;
 549    struct pipe_transfer *src_transfer;
 550    struct svga_winsys_surface *dst_handle;
 551    float extras[MAX_EXTRA_CONSTS][4];
 552    unsigned extra_count, extra_size, extra_offset;
 553    unsigned new_buf_size;
 554    void *src_map = NULL, *dst_map;
 555    unsigned offset;
 556    const struct svga_shader_variant *variant;
 557    unsigned alloc_buf_size;
 558
 559    assert(shader == PIPE_SHADER_VERTEX ||
 560           shader == PIPE_SHADER_GEOMETRY ||
 561           shader == PIPE_SHADER_FRAGMENT);
 562
 563    cbuf = &svga->curr.constbufs[shader][0];
 564
 565    switch (shader) {
 566    case PIPE_SHADER_VERTEX:
 567       variant = svga->state.hw_draw.vs;
 568       extra_count = svga_get_extra_vs_constants(svga, (float *) extras);
 569       break;
 570    case PIPE_SHADER_FRAGMENT:
 571       variant = svga->state.hw_draw.fs;
 572       extra_count = svga_get_extra_fs_constants(svga, (float *) extras);
 573       break;
 574    case PIPE_SHADER_GEOMETRY:
 575       variant = svga->state.hw_draw.gs;
 576       extra_count = svga_get_extra_gs_constants(svga, (float *) extras);
 577       break;
 578    default:
 579       assert(!"Unexpected shader type");
 580       /* Don't return an error code since we don't want to keep re-trying
 581        * this function and getting stuck in an infinite loop.
 582        */
 583       return PIPE_OK;
 584    }
 585
 586    assert(variant);
 587
 588    /* Compute extra constants size and offset in bytes */
 589    extra_size = extra_count * 4 * sizeof(float);
 590    extra_offset = 4 * sizeof(float) * variant->extra_const_start;
 591
 592    if (cbuf->buffer_size + extra_size == 0)
 593       return PIPE_OK;  /* nothing to do */
 594
 595    /* Typically, the cbuf->buffer here is a user-space buffer so mapping
 596     * it is really cheap.  If we ever get real HW buffers for constants
 597     * we should void mapping and instead use a ResourceCopy command.
 598     */
 599    if (cbuf->buffer_size > 0) {
 600       src_map = pipe_buffer_map_range(&svga->pipe, cbuf->buffer,
 601                                       cbuf->buffer_offset, cbuf->buffer_size,
 602                                       PIPE_TRANSFER_READ, &src_transfer);
 603       assert(src_map);
 604       if (!src_map) {
 605          return PIPE_ERROR_OUT_OF_MEMORY;
 606       }
 607    }
 608
 609    /* The new/dest buffer's size must be large enough to hold the original,
 610     * user-specified constants, plus the extra constants.
 611     * The size of the original constant buffer _should_ agree with what the
 612     * shader is expecting, but it might not (it's not enforced anywhere by
 613     * gallium).
 614     */
 615    new_buf_size = MAX2(cbuf->buffer_size, extra_offset) + extra_size;
 616
 617    /* According to the DX10 spec, the constant buffer size must be
 618     * in multiples of 16.
 619     */
 620    new_buf_size = align(new_buf_size, 16);
 621
 622    /* Constant buffer size in the upload buffer must be in multiples of 256.
 623     * In order to maximize the chance of merging the upload buffer chunks
 624     * when svga_buffer_add_range() is called,
 625     * the allocate buffer size needs to be in multiples of 256 as well.
 626     * Otherwise, since there is gap between each dirty range of the upload buffer,
 627     * each dirty range will end up in its own UPDATE_GB_IMAGE command.
 628     */
 629    alloc_buf_size = align(new_buf_size, CONST0_UPLOAD_ALIGNMENT);
 630
 631    u_upload_alloc(svga->const0_upload, 0, alloc_buf_size,
 632                   CONST0_UPLOAD_ALIGNMENT, &offset,
 633                   &dst_buffer, &dst_map);
 634    if (!dst_map) {
 635       if (src_map)
 636          pipe_buffer_unmap(&svga->pipe, src_transfer);
 637       return PIPE_ERROR_OUT_OF_MEMORY;
 638    }
 639
 640    if (src_map) {
 641       memcpy(dst_map, src_map, cbuf->buffer_size);
 642       pipe_buffer_unmap(&svga->pipe, src_transfer);
 643    }
 644
 645    if (extra_size) {
 646       assert(extra_offset + extra_size <= new_buf_size);
 647       memcpy((char *) dst_map + extra_offset, extras, extra_size);
 648    }
 649    u_upload_unmap(svga->const0_upload);
 650
 651    /* Issue the SetSingleConstantBuffer command */
 652    dst_handle = svga_buffer_handle(svga, dst_buffer);
 653    if (!dst_handle) {
 654       pipe_resource_reference(&dst_buffer, NULL);
 655       return PIPE_ERROR_OUT_OF_MEMORY;
 656    }
 657
 658    assert(new_buf_size % 16 == 0);
 659    ret = SVGA3D_vgpu10_SetSingleConstantBuffer(svga->swc,
 660                                                0, /* index */
 661                                                svga_shader_type(shader),
 662                                                dst_handle,
 663                                                offset,
 664                                                new_buf_size);
 665
 666    if (ret != PIPE_OK) {
 667       pipe_resource_reference(&dst_buffer, NULL);
 668       return ret;
 669    }
 670
 671    /* Save this const buffer until it's replaced in the future.
 672     * Otherwise, all references to the buffer will go away after the
 673     * command buffer is submitted, it'll get recycled and we will have
 674     * incorrect constant buffer bindings.
 675     */
 676    pipe_resource_reference(&svga->state.hw_draw.constbuf[shader], dst_buffer);
 677
 678    svga->state.hw_draw.default_constbuf_size[shader] = new_buf_size;
 679
 680    pipe_resource_reference(&dst_buffer, NULL);
 681
 682    svga->hud.num_const_buf_updates++;
 683
 684    return ret;
 685 }
 686
 687
 688 static enum pipe_error
 689 emit_consts_vgpu10(struct svga_context *svga, unsigned shader)
 690 {
 691    enum pipe_error ret;
 692    unsigned dirty_constbufs;
 693    unsigned enabled_constbufs;
 694
 695    /* Emit 0th constant buffer (with extra constants) */
 696    ret = emit_constbuf_vgpu10(svga, shader);
 697    if (ret != PIPE_OK) {
 698       return ret;
 699    }
 700
 701    enabled_constbufs = svga->state.hw_draw.enabled_constbufs[shader] | 1u;
 702
 703    /* Emit other constant buffers (UBOs) */
 704    dirty_constbufs = svga->state.dirty_constbufs[shader] & ~1u;
 705
 706    while (dirty_constbufs) {
 707       unsigned index = u_bit_scan(&dirty_constbufs);
 708       unsigned offset = svga->curr.constbufs[shader][index].buffer_offset;
 709       unsigned size = svga->curr.constbufs[shader][index].buffer_size;
 710       struct svga_buffer *buffer =
 711          svga_buffer(svga->curr.constbufs[shader][index].buffer);
 712       struct svga_winsys_surface *handle;
 713
 714       if (buffer) {
 715          handle = svga_buffer_handle(svga, &buffer->b.b);
 716          enabled_constbufs |= 1 << index;
 717       }
 718       else {
 719          handle = NULL;
 720          enabled_constbufs &= ~(1 << index);
 721          assert(offset == 0);
 722          assert(size == 0);
 723       }
 724
 725       if (size % 16 != 0) {
 726          /* GL's buffer range sizes can be any number of bytes but the
 727           * SVGA3D device requires a multiple of 16 bytes.
 728           */
 729          const unsigned total_size = buffer->b.b.width0;
 730
 731          if (offset + align(size, 16) <= total_size) {
 732             /* round up size to multiple of 16 */
 733             size = align(size, 16);
 734          }
 735          else {
 736             /* round down to mulitple of 16 (this may cause rendering problems
 737              * but should avoid a device error).
 738              */
 739             size &= ~15;
 740          }
 741       }
 742
 743       assert(size % 16 == 0);
 744       ret = SVGA3D_vgpu10_SetSingleConstantBuffer(svga->swc,
 745                                                   index,
 746                                                   svga_shader_type(shader),
 747                                                   handle,
 748                                                   offset,
 749                                                   size);
 750       if (ret != PIPE_OK)
 751          return ret;
 752
 753       svga->hud.num_const_buf_updates++;
 754    }
 755
 756    svga->state.hw_draw.enabled_constbufs[shader] = enabled_constbufs;
 757    svga->state.dirty_constbufs[shader] = 0;
 758
 759    return ret;
 760 }
 761
 762 static enum pipe_error
 763 emit_fs_consts(struct svga_context *svga, unsigned dirty)
 764 {
 765    const struct svga_shader_variant *variant = svga->state.hw_draw.fs;
 766    enum pipe_error ret = PIPE_OK;
 767
 768    /* SVGA_NEW_FS_VARIANT
 769     */
 770    if (!variant)
 771       return PIPE_OK;
 772
 773    /* SVGA_NEW_FS_CONST_BUFFER
 774     */
 775    if (svga_have_vgpu10(svga)) {
 776       ret = emit_consts_vgpu10(svga, PIPE_SHADER_FRAGMENT);
 777    }
 778    else {
 779       ret = emit_consts_vgpu9(svga, PIPE_SHADER_FRAGMENT);
 780    }
 781
 782    return ret;
 783 }
 784
 785
 786 struct svga_tracked_state svga_hw_fs_constants =
 787 {
 788    "hw fs params",
 789    (SVGA_NEW_FS_CONST_BUFFER |
 790     SVGA_NEW_FS_VARIANT |
 791     SVGA_NEW_TEXTURE_BINDING),
 792    emit_fs_consts
 793 };
 794
 795
 796
 797 static enum pipe_error
 798 emit_vs_consts(struct svga_context *svga, unsigned dirty)
 799 {
 800    const struct svga_shader_variant *variant = svga->state.hw_draw.vs;
 801    enum pipe_error ret = PIPE_OK;
 802
 803    /* SVGA_NEW_VS_VARIANT
 804     */
 805    if (!variant)
 806       return PIPE_OK;
 807
 808    /* SVGA_NEW_VS_CONST_BUFFER
 809     */
 810    if (svga_have_vgpu10(svga)) {
 811       ret = emit_consts_vgpu10(svga, PIPE_SHADER_VERTEX);
 812    }
 813    else {
 814       ret = emit_consts_vgpu9(svga, PIPE_SHADER_VERTEX);
 815    }
 816
 817    return ret;
 818 }
 819
 820
 821 struct svga_tracked_state svga_hw_vs_constants =
 822 {
 823    "hw vs params",
 824    (SVGA_NEW_PRESCALE |
 825     SVGA_NEW_VS_CONST_BUFFER |
 826     SVGA_NEW_VS_VARIANT),
 827    emit_vs_consts
 828 };
 829
 830
 831 static enum pipe_error
 832 emit_gs_consts(struct svga_context *svga, unsigned dirty)
 833 {
 834    const struct svga_shader_variant *variant = svga->state.hw_draw.gs;
 835    enum pipe_error ret = PIPE_OK;
 836
 837    /* SVGA_NEW_GS_VARIANT
 838     */
 839    if (!variant)
 840       return PIPE_OK;
 841
 842    /* SVGA_NEW_GS_CONST_BUFFER
 843     */
 844    if (svga_have_vgpu10(svga)) {
 845       /**
 846        * If only the rasterizer state has changed and the current geometry
 847        * shader does not emit wide points, then there is no reason to
 848        * re-emit the GS constants, so skip it.
 849        */
 850       if (dirty == SVGA_NEW_RAST && !variant->key.gs.wide_point)
 851          return PIPE_OK;
 852
 853       ret = emit_consts_vgpu10(svga, PIPE_SHADER_GEOMETRY);
 854    }
 855
 856    return ret;
 857 }
 858
 859
 860 struct svga_tracked_state svga_hw_gs_constants =
 861 {
 862    "hw gs params",
 863    (SVGA_NEW_GS_CONST_BUFFER |
 864     SVGA_NEW_RAST |
 865     SVGA_NEW_GS_VARIANT),
 866    emit_gs_consts
 867 };