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/format/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(const struct svga_context *svga,
  67                                 const struct svga_shader_variant *variant,
  68                                 enum pipe_shader_type 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       const struct pipe_sampler_view *sv = svga->curr.sampler_views[shader][i];
  76       if (sv) {
  77          const 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++ = 1.0f / (float) tex->width0;
  86             *dest++ = 1.0f / (float) tex->height0;
  87             *dest++ = 1.0f;
  88             *dest++ = 1.0f;
  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(const 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(const struct svga_context *svga, float **dest,
 137                             const struct svga_prescale *prescale)
 138 {
 139    memcpy(*dest, prescale->scale, 4 * sizeof(float));
 140    *dest += 4;
 141
 142    memcpy(*dest, prescale->translate, 4 * sizeof(float));
 143    *dest += 4;
 144
 145    return 2;
 146 }
 147
 148 /**
 149  * Emit extra constants needed for point sprite emulation.
 150  */
 151 static unsigned
 152 svga_get_pt_sprite_constants(const struct svga_context *svga, float **dest)
 153 {
 154    const struct svga_screen *screen = svga_screen(svga->pipe.screen);
 155    float *dst = *dest;
 156
 157    dst[0] = 1.0 / (svga->curr.viewport[0].scale[0] * 2);
 158    dst[1] = 1.0 / (svga->curr.viewport[0].scale[1] * 2);
 159    dst[2] = svga->curr.rast->pointsize;
 160    dst[3] = screen->maxPointSize;
 161    *dest = *dest + 4;
 162    return 1;
 163 }
 164
 165 /**
 166  * Emit user-defined clip plane coefficients into the buffer pointed to
 167  * by '*dest'. The updated buffer pointer will be returned in 'dest'.
 168  */
 169 static unsigned
 170 svga_get_clip_plane_constants(const struct svga_context *svga,
 171                               const struct svga_shader_variant *variant,
 172                               float **dest)
 173 {
 174    unsigned count = 0;
 175
 176    /* SVGA_NEW_CLIP */
 177    if (svga_have_vgpu10(svga)) {
 178       /* append user-defined clip plane coefficients onto constant buffer */
 179       unsigned clip_planes = variant->key.clip_plane_enable;
 180       while (clip_planes) {
 181          int i = u_bit_scan(&clip_planes);
 182          COPY_4V(*dest, svga->curr.clip.ucp[i]);
 183          *dest += 4;
 184          count += 1;
 185       }
 186    }
 187    return count;
 188 }
 189
 190
 191 /**
 192  * Emit any extra vertex shader constants into the buffer pointed
 193  * to by 'dest'.
 194  * In particular, these would be the scale and bias factors computed
 195  * from the framebuffer size which are used to copy with differences in
 196  * GL vs D3D coordinate spaces.  See svga_tgsi_insn.c for more info.
 197  * \return number of float[4] constants put into the dest buffer
 198  */
 199 static unsigned
 200 svga_get_extra_vs_constants(const struct svga_context *svga, float *dest)
 201 {
 202    const struct svga_shader_variant *variant = svga->state.hw_draw.vs;
 203    unsigned count = 0;
 204
 205    /* SVGA_NEW_VS_VARIANT
 206     */
 207    if (variant->key.vs.need_prescale) {
 208       count += svga_get_prescale_constants(svga, &dest,
 209                                            &svga->state.hw_clear.prescale[0]);
 210    }
 211
 212    if (variant->key.vs.undo_viewport) {
 213       /* Used to convert window coords back to NDC coords */
 214       dest[0] = 1.0f / svga->curr.viewport[0].scale[0];
 215       dest[1] = 1.0f / svga->curr.viewport[0].scale[1];
 216       dest[2] = -svga->curr.viewport[0].translate[0];
 217       dest[3] = -svga->curr.viewport[0].translate[1];
 218       dest += 4;
 219       count += 1;
 220    }
 221
 222    /* Bias to be added to VertexID */
 223    if (variant->key.vs.need_vertex_id_bias) {
 224       uint32_t *dest_u = (uint32_t *) dest;  // uint version of dest
 225       dest_u[0] = svga->curr.vertex_id_bias;
 226       dest_u[1] = 1;
 227       dest_u[2] = 1;
 228       dest_u[3] = 1;
 229       dest+=4;
 230       count++;
 231    }
 232
 233    /* SVGA_NEW_CLIP */
 234    count += svga_get_clip_plane_constants(svga, variant, &dest);
 235
 236    /* common constants */
 237    count += svga_get_extra_constants_common(svga, variant,
 238                                             PIPE_SHADER_VERTEX, dest);
 239
 240    assert(count <= MAX_EXTRA_CONSTS);
 241
 242    return count;
 243 }
 244
 245 /**
 246  * Emit any extra geometry shader constants into the buffer pointed
 247  * to by 'dest'.
 248  */
 249 static unsigned
 250 svga_get_extra_gs_constants(const struct svga_context *svga, float *dest)
 251 {
 252    const struct svga_shader_variant *variant = svga->state.hw_draw.gs;
 253    unsigned count = 0;
 254
 255    /* SVGA_NEW_GS_VARIANT
 256     */
 257
 258    /* Constants for point sprite
 259     * These are used in the transformed gs that supports point sprite.
 260     * They need to be added before the prescale constants.
 261     */
 262    if (variant->key.gs.wide_point) {
 263       count += svga_get_pt_sprite_constants(svga, &dest);
 264    }
 265
 266    if (variant->key.gs.need_prescale) {
 267       unsigned i, num_prescale = 1;
 268
 269       /* If prescale is needed and the geometry shader writes to viewport
 270        * index, then prescale for all viewports will be added to the
 271        * constant buffer.
 272        */
 273       if (variant->key.gs.writes_viewport_index)
 274          num_prescale = svga->state.hw_clear.num_prescale;
 275
 276       for (i = 0; i < num_prescale; i++) {
 277          count +=
 278             svga_get_prescale_constants(svga, &dest,
 279                                         &svga->state.hw_clear.prescale[i]);
 280       }
 281    }
 282
 283    /* SVGA_NEW_CLIP */
 284    count += svga_get_clip_plane_constants(svga, variant, &dest);
 285
 286    /* common constants */
 287    count += svga_get_extra_constants_common(svga, variant,
 288                                             PIPE_SHADER_GEOMETRY, dest);
 289
 290    assert(count <= MAX_EXTRA_CONSTS);
 291    return count;
 292 }
 293
 294
 295 /**
 296  * Emit any extra tessellation control shader constants into the
 297  * buffer pointed to by 'dest'.
 298  */
 299 static unsigned
 300 svga_get_extra_tcs_constants(struct svga_context *svga, float *dest)
 301 {
 302    const struct svga_shader_variant *variant = svga->state.hw_draw.tcs;
 303    unsigned count = 0;
 304
 305    /* SVGA_NEW_CLIP */
 306    count += svga_get_clip_plane_constants(svga, variant, &dest);
 307
 308    /* common constants */
 309    count += svga_get_extra_constants_common(svga, variant,
 310                                             PIPE_SHADER_TESS_CTRL,
 311                                             dest);
 312
 313    assert(count <= MAX_EXTRA_CONSTS);
 314    return count;
 315 }
 316
 317
 318 /**
 319  * Emit any extra tessellation evaluation shader constants into
 320  * the buffer pointed to by 'dest'.
 321  */
 322 static unsigned
 323 svga_get_extra_tes_constants(struct svga_context *svga, float *dest)
 324 {
 325    const struct svga_shader_variant *variant = svga->state.hw_draw.tes;
 326    unsigned count = 0;
 327
 328    if (variant->key.tes.need_prescale) {
 329       count += svga_get_prescale_constants(svga, &dest,
 330                                            &svga->state.hw_clear.prescale[0]);
 331    }
 332
 333    /* SVGA_NEW_CLIP */
 334    count += svga_get_clip_plane_constants(svga, variant, &dest);
 335
 336    /* common constants */
 337    count += svga_get_extra_constants_common(svga, variant,
 338                                             PIPE_SHADER_TESS_EVAL,
 339                                             dest);
 340
 341    assert(count <= MAX_EXTRA_CONSTS);
 342    return count;
 343 }
 344
 345
 346 /**
 347  * Emit any extra compute shader constants into
 348  * the buffer pointed to by 'dest'.
 349  */
 350 static unsigned
 351 svga_get_extra_cs_constants(struct svga_context *svga, float *dest)
 352 {
 353    const struct svga_shader_variant *variant = svga->state.hw_draw.cs;
 354    unsigned count = 0;
 355
 356    /* common constants */
 357    count += svga_get_extra_constants_common(svga, variant,
 358                                             PIPE_SHADER_COMPUTE,
 359                                             dest);
 360
 361    assert(count <= MAX_EXTRA_CONSTS);
 362    return count;
 363 }
 364
 365
 366 /*
 367  * Check and emit a range of shader constant registers, trying to coalesce
 368  * successive shader constant updates in a single command in order to save
 369  * space on the command buffer.  This is a HWv8 feature.
 370  */
 371 static enum pipe_error
 372 emit_const_range(struct svga_context *svga,
 373                  enum pipe_shader_type shader,
 374                  unsigned offset,
 375                  unsigned count,
 376                  const float (*values)[4])
 377 {
 378    unsigned i, j;
 379    enum pipe_error ret;
 380
 381    assert(shader == PIPE_SHADER_VERTEX ||
 382           shader == PIPE_SHADER_FRAGMENT);
 383    assert(!svga_have_vgpu10(svga));
 384
 385 #ifdef DEBUG
 386    if (offset + count > SVGA3D_CONSTREG_MAX) {
 387       debug_printf("svga: too many constants (offset %u + count %u = %u (max = %u))\n",
 388                    offset, count, offset + count, SVGA3D_CONSTREG_MAX);
 389    }
 390 #endif
 391
 392    if (offset > SVGA3D_CONSTREG_MAX) {
 393       /* This isn't OK, but if we propagate an error all the way up we'll
 394        * just get into more trouble.
 395        * XXX note that offset is always zero at this time so this is moot.
 396        */
 397       return PIPE_OK;
 398    }
 399
 400    if (offset + count > SVGA3D_CONSTREG_MAX) {
 401       /* Just drop the extra constants for now.
 402        * Ideally we should not have allowed the app to create a shader
 403        * that exceeds our constant buffer size but there's no way to
 404        * express that in gallium at this time.
 405        */
 406       count = SVGA3D_CONSTREG_MAX - offset;
 407    }
 408
 409    i = 0;
 410    while (i < count) {
 411       if (memcmp(svga->state.hw_draw.cb[shader][offset + i],
 412                  values[i],
 413                  4 * sizeof(float)) != 0) {
 414          /* Found one dirty constant
 415           */
 416          if (SVGA_DEBUG & DEBUG_CONSTS)
 417             debug_printf("%s %s %d: %f %f %f %f\n",
 418                          __FUNCTION__,
 419                          shader == PIPE_SHADER_VERTEX ? "VERT" : "FRAG",
 420                          offset + i,
 421                          values[i][0],
 422                          values[i][1],
 423                          values[i][2],
 424                          values[i][3]);
 425
 426          /* Look for more consecutive dirty constants.
 427           */
 428          j = i + 1;
 429          while (j < count &&
 430                 j < i + MAX_CONST_REG_COUNT &&
 431                 memcmp(svga->state.hw_draw.cb[shader][offset + j],
 432                        values[j],
 433                        4 * sizeof(float)) != 0) {
 434
 435             if (SVGA_DEBUG & DEBUG_CONSTS)
 436                debug_printf("%s %s %d: %f %f %f %f\n",
 437                             __FUNCTION__,
 438                             shader == PIPE_SHADER_VERTEX ? "VERT" : "FRAG",
 439                             offset + j,
 440                             values[j][0],
 441                             values[j][1],
 442                             values[j][2],
 443                             values[j][3]);
 444
 445             ++j;
 446          }
 447
 448          assert(j >= i + 1);
 449
 450          /* Send them all together.
 451           */
 452          if (svga_have_gb_objects(svga)) {
 453             ret = SVGA3D_SetGBShaderConstsInline(svga->swc,
 454                                                  offset + i, /* start */
 455                                                  j - i,  /* count */
 456                                                  svga_shader_type(shader),
 457                                                  SVGA3D_CONST_TYPE_FLOAT,
 458                                                  values + i);
 459          }
 460          else {
 461             ret = SVGA3D_SetShaderConsts(svga->swc,
 462                                          offset + i, j - i,
 463                                          svga_shader_type(shader),
 464                                          SVGA3D_CONST_TYPE_FLOAT,
 465                                          values + i);
 466          }
 467          if (ret != PIPE_OK) {
 468             return ret;
 469          }
 470
 471          /*
 472           * Local copy of the hardware state.
 473           */
 474          memcpy(svga->state.hw_draw.cb[shader][offset + i],
 475                 values[i],
 476                 (j - i) * 4 * sizeof(float));
 477
 478          i = j + 1;
 479
 480          svga->hud.num_const_updates++;
 481
 482       } else {
 483          ++i;
 484       }
 485    }
 486
 487    return PIPE_OK;
 488 }
 489
 490
 491 /**
 492  * Emit all the constants in a constant buffer for a shader stage.
 493  * On VGPU10, emit_consts_vgpu10 is used instead.
 494  */
 495 static enum pipe_error
 496 emit_consts_vgpu9(struct svga_context *svga, enum pipe_shader_type shader)
 497 {
 498    const struct pipe_constant_buffer *cbuf;
 499    struct pipe_transfer *transfer = NULL;
 500    unsigned count;
 501    const float (*data)[4] = NULL;
 502    enum pipe_error ret = PIPE_OK;
 503    const unsigned offset = 0;
 504
 505    assert(shader < PIPE_SHADER_TYPES);
 506    assert(!svga_have_vgpu10(svga));
 507    /* Only one constant buffer per shader is supported before VGPU10.
 508     * This is only an approximate check against that.
 509     */
 510    assert(svga->curr.constbufs[shader][1].buffer == NULL);
 511
 512    cbuf = &svga->curr.constbufs[shader][0];
 513
 514    if (svga->curr.constbufs[shader][0].buffer) {
 515       /* emit user-provided constants */
 516       data = (const float (*)[4])
 517          pipe_buffer_map(&svga->pipe, svga->curr.constbufs[shader][0].buffer,
 518                          PIPE_TRANSFER_READ, &transfer);
 519       if (!data) {
 520          return PIPE_ERROR_OUT_OF_MEMORY;
 521       }
 522
 523       /* sanity check */
 524       assert(cbuf->buffer->width0 >= cbuf->buffer_size);
 525
 526       /* Use/apply the constant buffer size and offsets here */
 527       count = cbuf->buffer_size / (4 * sizeof(float));
 528       data += cbuf->buffer_offset / (4 * sizeof(float));
 529
 530       ret = emit_const_range( svga, shader, offset, count, data );
 531
 532       pipe_buffer_unmap(&svga->pipe, transfer);
 533
 534       if (ret != PIPE_OK) {
 535          return ret;
 536       }
 537    }
 538
 539    /* emit extra shader constants */
 540    {
 541       const struct svga_shader_variant *variant = NULL;
 542       unsigned offset;
 543       float extras[MAX_EXTRA_CONSTS][4];
 544       unsigned count;
 545
 546       switch (shader) {
 547       case PIPE_SHADER_VERTEX:
 548          variant = svga->state.hw_draw.vs;
 549          count = svga_get_extra_vs_constants(svga, (float *) extras);
 550          break;
 551       case PIPE_SHADER_FRAGMENT:
 552          variant = svga->state.hw_draw.fs;
 553          count = svga_get_extra_fs_constants(svga, (float *) extras);
 554          break;
 555       default:
 556          assert(!"Unexpected shader type");
 557          count = 0;
 558       }
 559
 560       assert(variant);
 561       offset = variant->shader->info.file_max[TGSI_FILE_CONSTANT] + 1;
 562       assert(count <= ARRAY_SIZE(extras));
 563
 564       if (count > 0) {
 565          ret = emit_const_range(svga, shader, offset, count,
 566                                 (const float (*) [4])extras);
 567       }
 568    }
 569
 570    return ret;
 571 }
 572
 573
 574 /**
 575  * A helper function to emit a constant buffer binding at the
 576  * specified slot for the specified shader type
 577  */
 578 static enum pipe_error
 579 emit_constbuf(struct svga_context *svga,
 580               unsigned slot,
 581               enum pipe_shader_type shader,
 582               unsigned buffer_offset,
 583               unsigned buffer_size,
 584               const void *buffer,
 585               unsigned extra_buffer_offset,
 586               unsigned extra_buffer_size,
 587               const void *extra_buffer)
 588 {
 589    struct svga_buffer *sbuf = svga_buffer((struct pipe_resource *)buffer);
 590    struct pipe_resource *dst_buffer = NULL;
 591    enum pipe_error ret = PIPE_OK;
 592    struct pipe_transfer *src_transfer;
 593    struct svga_winsys_surface *dst_handle = NULL;
 594    unsigned new_buf_size = 0;
 595    unsigned alloc_buf_size;
 596    unsigned offset = 0;;
 597    void *src_map = NULL, *dst_map;
 598
 599    if ((sbuf && sbuf->swbuf) || extra_buffer) {
 600
 601       /* buffer here is a user-space buffer so mapping it is really cheap. */
 602       if (buffer_size > 0) {
 603          src_map = pipe_buffer_map_range(&svga->pipe,
 604                                          (struct pipe_resource *)buffer,
 605                                          buffer_offset, buffer_size,
 606                                          PIPE_TRANSFER_READ, &src_transfer);
 607          assert(src_map);
 608          if (!src_map) {
 609             return PIPE_ERROR_OUT_OF_MEMORY;
 610          }
 611       }
 612
 613       new_buf_size = MAX2(buffer_size, extra_buffer_offset) + extra_buffer_size;
 614
 615       /* According to the DX10 spec, the constant buffer size must be
 616        * in multiples of 16.
 617        */
 618       new_buf_size = align(new_buf_size, 16);
 619
 620       /* Constant buffer size in the upload buffer must be in multiples of 256.
 621        * In order to maximize the chance of merging the upload buffer chunks
 622        * when svga_buffer_add_range() is called,
 623        * the allocate buffer size needs to be in multiples of 256 as well.
 624        * Otherwise, since there is gap between each dirty range of the upload buffer,
 625        * each dirty range will end up in its own UPDATE_GB_IMAGE command.
 626        */
 627       alloc_buf_size = align(new_buf_size, CONST0_UPLOAD_ALIGNMENT);
 628
 629       u_upload_alloc(svga->const0_upload, 0, alloc_buf_size,
 630                      CONST0_UPLOAD_ALIGNMENT, &offset,
 631                      &dst_buffer, &dst_map);
 632
 633       if (!dst_map) {
 634          if (src_map)
 635             pipe_buffer_unmap(&svga->pipe, src_transfer);
 636          return PIPE_ERROR_OUT_OF_MEMORY;
 637       }
 638
 639       /* Initialize the allocated buffer slot to 0 to ensure the padding is
 640        * filled with 0.
 641        */
 642       memset(dst_map, 0, alloc_buf_size);
 643
 644       if (src_map) {
 645          memcpy(dst_map, src_map, buffer_size);
 646          pipe_buffer_unmap(&svga->pipe, src_transfer);
 647       }
 648
 649       if (extra_buffer_size) {
 650          assert(extra_buffer_offset + extra_buffer_size <= new_buf_size);
 651          memcpy((char *) dst_map + extra_buffer_offset, extra_buffer,
 652                 extra_buffer_size);
 653       }
 654
 655       /* Get winsys handle for the constant buffer */
 656       if (svga->state.hw_draw.const0_buffer == dst_buffer &&
 657           svga->state.hw_draw.const0_handle) {
 658          /* re-reference already mapped buffer */
 659          dst_handle = svga->state.hw_draw.const0_handle;
 660       }
 661       else {
 662          /* we must unmap the buffer before getting the winsys handle */
 663          u_upload_unmap(svga->const0_upload);
 664
 665          dst_handle = svga_buffer_handle(svga, dst_buffer,
 666                                          PIPE_BIND_CONSTANT_BUFFER);
 667          if (!dst_handle) {
 668             pipe_resource_reference(&dst_buffer, NULL);
 669             return PIPE_ERROR_OUT_OF_MEMORY;
 670          }
 671       }
 672    }
 673    else if (sbuf) {
 674       dst_handle = svga_buffer_handle(svga, &sbuf->b.b, PIPE_BIND_CONSTANT_BUFFER);
 675       new_buf_size = align(buffer_size, 16);
 676       offset = buffer_offset;
 677    }
 678
 679    assert(new_buf_size % 16 == 0);
 680
 681    const struct svga_screen *screen = svga_screen(svga->pipe.screen);
 682    const struct svga_winsys_screen *sws = screen->sws;
 683
 684    /* Issue the SetSingleConstantBuffer command */
 685    if (!sws->have_constant_buffer_offset_cmd ||
 686        svga->state.hw_draw.constbufoffsets[shader][slot].handle != dst_handle ||
 687        svga->state.hw_draw.constbufoffsets[shader][slot].size != new_buf_size) {
 688       ret = SVGA3D_vgpu10_SetSingleConstantBuffer(svga->swc,
 689                                                   slot, /* index */
 690                                                   svga_shader_type(shader),
 691                                                   dst_handle,
 692                                                   offset,
 693                                                   new_buf_size);
 694    }
 695    else if (dst_handle){
 696       unsigned command = SVGA_3D_CMD_DX_SET_VS_CONSTANT_BUFFER_OFFSET + shader;
 697       ret = SVGA3D_vgpu10_SetConstantBufferOffset(svga->swc,
 698                                                   command,
 699                                                   slot, /* index */
 700                                                   offset);
 701    }
 702
 703    if (ret != PIPE_OK) {
 704       pipe_resource_reference(&dst_buffer, NULL);
 705       return ret;
 706    }
 707
 708    /* save the upload buffer / handle for next time */
 709    if (dst_buffer != buffer && dst_buffer) {
 710       pipe_resource_reference(&svga->state.hw_draw.const0_buffer, dst_buffer);
 711       svga->state.hw_draw.const0_handle = dst_handle;
 712    }
 713
 714    /* Save this const buffer until it's replaced in the future.
 715     * Otherwise, all references to the buffer will go away after the
 716     * command buffer is submitted, it'll get recycled and we will have
 717     * incorrect constant buffer bindings.
 718     */
 719    pipe_resource_reference(&svga->state.hw_draw.constbuf[shader][slot], dst_buffer);
 720    svga->state.hw_draw.constbufoffsets[shader][slot].handle = dst_handle;
 721    svga->state.hw_draw.constbufoffsets[shader][slot].size = new_buf_size;
 722
 723    pipe_resource_reference(&dst_buffer, NULL);
 724
 725    return PIPE_OK;
 726 }
 727
 728
 729 /* For constbuf 0 */
 730 static enum pipe_error
 731 emit_consts_vgpu10(struct svga_context *svga, enum pipe_shader_type shader)
 732 {
 733    const struct pipe_constant_buffer *cbuf;
 734    enum pipe_error ret = PIPE_OK;
 735    float extras[MAX_EXTRA_CONSTS][4];
 736    unsigned extra_count, extra_size, extra_offset;
 737    const struct svga_shader_variant *variant;
 738
 739    assert(shader == PIPE_SHADER_VERTEX ||
 740           shader == PIPE_SHADER_GEOMETRY ||
 741           shader == PIPE_SHADER_FRAGMENT ||
 742           shader == PIPE_SHADER_TESS_CTRL ||
 743           shader == PIPE_SHADER_TESS_EVAL ||
 744           shader == PIPE_SHADER_COMPUTE);
 745
 746    cbuf = &svga->curr.constbufs[shader][0];
 747
 748    switch (shader) {
 749    case PIPE_SHADER_VERTEX:
 750       variant = svga->state.hw_draw.vs;
 751       extra_count = svga_get_extra_vs_constants(svga, (float *) extras);
 752       break;
 753    case PIPE_SHADER_FRAGMENT:
 754       variant = svga->state.hw_draw.fs;
 755       extra_count = svga_get_extra_fs_constants(svga, (float *) extras);
 756       break;
 757    case PIPE_SHADER_GEOMETRY:
 758       variant = svga->state.hw_draw.gs;
 759       extra_count = svga_get_extra_gs_constants(svga, (float *) extras);
 760       break;
 761    case PIPE_SHADER_TESS_CTRL:
 762       variant = svga->state.hw_draw.tcs;
 763       extra_count = svga_get_extra_tcs_constants(svga, (float *) extras);
 764       break;
 765    case PIPE_SHADER_TESS_EVAL:
 766       variant = svga->state.hw_draw.tes;
 767       extra_count = svga_get_extra_tes_constants(svga, (float *) extras);
 768       break;
 769    case PIPE_SHADER_COMPUTE:
 770       variant = svga->state.hw_draw.cs;
 771       extra_count = svga_get_extra_cs_constants(svga, (float *) extras);
 772       break;
 773    default:
 774       assert(!"Unexpected shader type");
 775       /* Don't return an error code since we don't want to keep re-trying
 776        * this function and getting stuck in an infinite loop.
 777        */
 778       return PIPE_OK;
 779    }
 780
 781    assert(variant);
 782
 783    cbuf = &svga->curr.constbufs[shader][0];
 784
 785    /* Compute extra constants size and offset in bytes */
 786    extra_size = extra_count * 4 * sizeof(float);
 787    extra_offset = 4 * sizeof(float) * variant->extra_const_start;
 788
 789    if (cbuf->buffer_size + extra_size == 0)
 790       return PIPE_OK;  /* nothing to do */
 791
 792    ret = emit_constbuf(svga, 0, shader,
 793                        cbuf->buffer_offset, cbuf->buffer_size, cbuf->buffer,
 794                        extra_offset, extra_size, extras);
 795    if (ret != PIPE_OK)
 796       return ret;
 797
 798    svga->state.hw_draw.default_constbuf_size[shader] =
 799       svga->state.hw_draw.constbufoffsets[shader][0].size;
 800
 801    svga->hud.num_const_updates++;
 802
 803    return ret;
 804 }
 805
 806
 807 static enum pipe_error
 808 emit_constbuf_vgpu10(struct svga_context *svga, enum pipe_shader_type shader)
 809 {
 810    enum pipe_error ret = PIPE_OK;
 811    unsigned dirty_constbufs;
 812    unsigned enabled_constbufs;
 813
 814    enabled_constbufs = svga->state.hw_draw.enabled_constbufs[shader] | 1u;
 815    dirty_constbufs = (svga->state.dirty_constbufs[shader]|enabled_constbufs) & ~1u;
 816
 817    while (dirty_constbufs) {
 818       unsigned index = u_bit_scan(&dirty_constbufs);
 819       unsigned offset = svga->curr.constbufs[shader][index].buffer_offset;
 820       unsigned size = svga->curr.constbufs[shader][index].buffer_size;
 821       struct svga_buffer *buffer =
 822          svga_buffer(svga->curr.constbufs[shader][index].buffer);
 823
 824       if (buffer) {
 825          enabled_constbufs |= 1 << index;
 826       }
 827       else {
 828          enabled_constbufs &= ~(1 << index);
 829          assert(offset == 0);
 830          assert(size == 0);
 831       }
 832
 833       if (size % 16 != 0) {
 834          /* GL's buffer range sizes can be any number of bytes but the
 835           * SVGA3D device requires a multiple of 16 bytes.
 836           */
 837          const unsigned total_size = buffer->b.b.width0;
 838
 839          if (offset + align(size, 16) <= total_size) {
 840             /* round up size to multiple of 16 */
 841             size = align(size, 16);
 842          }
 843          else {
 844             /* round down to mulitple of 16 (this may cause rendering problems
 845              * but should avoid a device error).
 846              */
 847             size &= ~15;
 848          }
 849       }
 850
 851       assert(size % 16 == 0);
 852
 853       ret = emit_constbuf(svga, index, shader, offset, size, buffer,
 854                           0, 0, NULL);
 855       if (ret != PIPE_OK)
 856          return ret;
 857
 858       svga->hud.num_const_buf_updates++;
 859    }
 860
 861    svga->state.hw_draw.enabled_constbufs[shader] = enabled_constbufs;
 862    svga->state.dirty_constbufs[shader] = 0;
 863
 864    return ret;
 865 }
 866
 867 static enum pipe_error
 868 emit_fs_consts(struct svga_context *svga, uint64_t dirty)
 869 {
 870    const struct svga_shader_variant *variant = svga->state.hw_draw.fs;
 871    enum pipe_error ret = PIPE_OK;
 872
 873    /* SVGA_NEW_FS_VARIANT
 874     */
 875    if (!variant)
 876       return PIPE_OK;
 877
 878    /* SVGA_NEW_FS_CONSTS
 879     */
 880    if (svga_have_vgpu10(svga)) {
 881       ret = emit_consts_vgpu10(svga, PIPE_SHADER_FRAGMENT);
 882    }
 883    else {
 884       ret = emit_consts_vgpu9(svga, PIPE_SHADER_FRAGMENT);
 885    }
 886
 887    return ret;
 888 }
 889
 890 static enum pipe_error
 891 emit_fs_constbuf(struct svga_context *svga, uint64_t dirty)
 892 {
 893    const struct svga_shader_variant *variant = svga->state.hw_draw.fs;
 894    enum pipe_error ret = PIPE_OK;
 895
 896    /* SVGA_NEW_FS_VARIANT
 897     */
 898    if (!variant)
 899       return PIPE_OK;
 900
 901    /* SVGA_NEW_FS_CONSTBUF
 902     */
 903    assert(svga_have_vgpu10(svga));
 904    ret = emit_constbuf_vgpu10(svga, PIPE_SHADER_FRAGMENT);
 905
 906    return ret;
 907 }
 908
 909 struct svga_tracked_state svga_hw_fs_constants =
 910 {
 911    "hw fs params",
 912    (SVGA_NEW_FS_CONSTS |
 913     SVGA_NEW_FS_VARIANT |
 914     SVGA_NEW_TEXTURE_CONSTS),
 915    emit_fs_consts
 916 };
 917
 918
 919 struct svga_tracked_state svga_hw_fs_constbufs =
 920 {
 921    "hw fs params",
 922    SVGA_NEW_FS_CONST_BUFFER,
 923    emit_fs_constbuf
 924 };
 925
 926
 927 static enum pipe_error
 928 emit_vs_consts(struct svga_context *svga, uint64_t dirty)
 929 {
 930    const struct svga_shader_variant *variant = svga->state.hw_draw.vs;
 931    enum pipe_error ret = PIPE_OK;
 932
 933    /* SVGA_NEW_VS_VARIANT
 934     */
 935    if (!variant)
 936       return PIPE_OK;
 937
 938    /* SVGA_NEW_VS_CONST_BUFFER
 939     */
 940    if (svga_have_vgpu10(svga)) {
 941       ret = emit_consts_vgpu10(svga, PIPE_SHADER_VERTEX);
 942    }
 943    else {
 944       ret = emit_consts_vgpu9(svga, PIPE_SHADER_VERTEX);
 945    }
 946
 947    return ret;
 948 }
 949
 950
 951 static enum pipe_error
 952 emit_vs_constbuf(struct svga_context *svga, uint64_t dirty)
 953 {
 954    const struct svga_shader_variant *variant = svga->state.hw_draw.vs;
 955    enum pipe_error ret = PIPE_OK;
 956
 957    /* SVGA_NEW_FS_VARIANT
 958     */
 959    if (!variant)
 960       return PIPE_OK;
 961
 962    /* SVGA_NEW_FS_CONSTBUF
 963     */
 964    assert(svga_have_vgpu10(svga));
 965    ret = emit_constbuf_vgpu10(svga, PIPE_SHADER_VERTEX);
 966
 967    return ret;
 968 }
 969
 970
 971 struct svga_tracked_state svga_hw_vs_constants =
 972 {
 973    "hw vs params",
 974    (SVGA_NEW_PRESCALE |
 975     SVGA_NEW_VS_CONSTS |
 976     SVGA_NEW_VS_VARIANT |
 977     SVGA_NEW_TEXTURE_CONSTS),
 978    emit_vs_consts
 979 };
 980
 981
 982 struct svga_tracked_state svga_hw_vs_constbufs =
 983 {
 984    "hw vs params",
 985    SVGA_NEW_VS_CONST_BUFFER,
 986    emit_vs_constbuf
 987 };
 988
 989
 990 static enum pipe_error
 991 emit_gs_consts(struct svga_context *svga, uint64_t dirty)
 992 {
 993    const struct svga_shader_variant *variant = svga->state.hw_draw.gs;
 994    enum pipe_error ret = PIPE_OK;
 995
 996    /* SVGA_NEW_GS_VARIANT
 997     */
 998    if (!variant)
 999       return PIPE_OK;
1000
1001    /* SVGA_NEW_GS_CONST_BUFFER
1002     */
1003    assert(svga_have_vgpu10(svga));
1004
1005    /**
1006     * If only the rasterizer state has changed and the current geometry
1007     * shader does not emit wide points, then there is no reason to
1008     * re-emit the GS constants, so skip it.
1009     */
1010    if (dirty == SVGA_NEW_RAST && !variant->key.gs.wide_point)
1011       return PIPE_OK;
1012
1013    ret = emit_consts_vgpu10(svga, PIPE_SHADER_GEOMETRY);
1014
1015    return ret;
1016 }
1017
1018
1019 static enum pipe_error
1020 emit_gs_constbuf(struct svga_context *svga, uint64_t dirty)
1021 {
1022    const struct svga_shader_variant *variant = svga->state.hw_draw.gs;
1023    enum pipe_error ret = PIPE_OK;
1024
1025    /* SVGA_NEW_GS_VARIANT
1026     */
1027    if (!variant)
1028       return PIPE_OK;
1029
1030    /* SVGA_NEW_GS_CONSTBUF
1031     */
1032    assert(svga_have_vgpu10(svga));
1033    ret = emit_constbuf_vgpu10(svga, PIPE_SHADER_GEOMETRY);
1034
1035    return ret;
1036 }
1037
1038
1039 struct svga_tracked_state svga_hw_gs_constants =
1040 {
1041    "hw gs params",
1042    (SVGA_NEW_PRESCALE |
1043     SVGA_NEW_GS_CONSTS |
1044     SVGA_NEW_RAST |
1045     SVGA_NEW_GS_VARIANT |
1046     SVGA_NEW_TEXTURE_CONSTS),
1047    emit_gs_consts
1048 };
1049
1050
1051 struct svga_tracked_state svga_hw_gs_constbufs =
1052 {
1053    "hw gs params",
1054    SVGA_NEW_GS_CONST_BUFFER,
1055    emit_gs_constbuf
1056 };
1057
1058
1059 /**
1060  * Emit constant buffer for tessellation control shader
1061  */
1062 static enum pipe_error
1063 emit_tcs_consts(struct svga_context *svga, uint64_t dirty)
1064 {
1065    const struct svga_shader_variant *variant = svga->state.hw_draw.tcs;
1066    enum pipe_error ret = PIPE_OK;
1067
1068    assert(svga_have_sm5(svga));
1069
1070    /* SVGA_NEW_TCS_VARIANT */
1071    if (!variant)
1072       return PIPE_OK;
1073
1074    /* SVGA_NEW_TCS_CONST_BUFFER */
1075
1076    ret = emit_consts_vgpu10(svga, PIPE_SHADER_TESS_CTRL);
1077
1078    return ret;
1079 }
1080
1081
1082 static enum pipe_error
1083 emit_tcs_constbuf(struct svga_context *svga, uint64_t dirty)
1084 {
1085    const struct svga_shader_variant *variant = svga->state.hw_draw.tcs;
1086    enum pipe_error ret = PIPE_OK;
1087
1088    /* SVGA_NEW_TCS_VARIANT
1089     */
1090    if (!variant)
1091       return PIPE_OK;
1092
1093    /* SVGA_NEW_TCS_CONSTBUF
1094     */
1095    assert(svga_have_vgpu10(svga));
1096    ret = emit_constbuf_vgpu10(svga, PIPE_SHADER_TESS_CTRL);
1097
1098    return ret;
1099 }
1100
1101
1102 struct svga_tracked_state svga_hw_tcs_constants =
1103 {
1104    "hw tcs params",
1105    (SVGA_NEW_TCS_CONSTS |
1106     SVGA_NEW_TCS_VARIANT),
1107    emit_tcs_consts
1108 };
1109
1110
1111 struct svga_tracked_state svga_hw_tcs_constbufs =
1112 {
1113    "hw tcs params",
1114    SVGA_NEW_TCS_CONST_BUFFER,
1115    emit_tcs_constbuf
1116 };
1117
1118
1119 /**
1120  * Emit constant buffer for tessellation evaluation shader
1121  */
1122 static enum pipe_error
1123 emit_tes_consts(struct svga_context *svga, uint64_t dirty)
1124 {
1125    const struct svga_shader_variant *variant = svga->state.hw_draw.tes;
1126    enum pipe_error ret = PIPE_OK;
1127
1128    assert(svga_have_sm5(svga));
1129
1130    /* SVGA_NEW_TES_VARIANT */
1131    if (!variant)
1132       return PIPE_OK;
1133
1134    ret = emit_consts_vgpu10(svga, PIPE_SHADER_TESS_EVAL);
1135
1136    return ret;
1137 }
1138
1139
1140 static enum pipe_error
1141 emit_tes_constbuf(struct svga_context *svga, uint64_t dirty)
1142 {
1143    const struct svga_shader_variant *variant = svga->state.hw_draw.tes;
1144    enum pipe_error ret = PIPE_OK;
1145
1146    /* SVGA_NEW_TES_VARIANT
1147     */
1148    if (!variant)
1149       return PIPE_OK;
1150
1151    /* SVGA_NEW_TES_CONSTBUF
1152     */
1153    assert(svga_have_vgpu10(svga));
1154    ret = emit_constbuf_vgpu10(svga, PIPE_SHADER_TESS_EVAL);
1155
1156    return ret;
1157 }
1158
1159
1160 struct svga_tracked_state svga_hw_tes_constants =
1161 {
1162    "hw tes params",
1163    (SVGA_NEW_PRESCALE |
1164     SVGA_NEW_TES_CONSTS |
1165     SVGA_NEW_TES_VARIANT),
1166    emit_tes_consts
1167 };
1168
1169
1170 struct svga_tracked_state svga_hw_tes_constbufs =
1171 {
1172    "hw gs params",
1173    SVGA_NEW_TES_CONST_BUFFER,
1174    emit_tes_constbuf
1175 };