src/gallium/drivers/ilo/ilo_3d_pipeline.c

   1 /*
   2  * Mesa 3-D graphics library
   3  *
   4  * Copyright (C) 2013 LunarG, Inc.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included
  14  * in all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  21  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  22  * DEALINGS IN THE SOFTWARE.
  23  *
  24  * Authors:
  25  *    Chia-I Wu <olv@lunarg.com>
  26  */
  27
  28 #include "util/u_prim.h"
  29 #include "intel_winsys.h"
  30
  31 #include "ilo_context.h"
  32 #include "ilo_cp.h"
  33 #include "ilo_state.h"
  34 #include "ilo_3d_pipeline_gen6.h"
  35 #include "ilo_3d_pipeline_gen7.h"
  36 #include "ilo_3d_pipeline.h"
  37
  38 /* in U0.4 */
  39 struct sample_position {
  40    uint8_t x, y;
  41 };
  42
  43 /* \see gen6_get_sample_position() */
  44 static const struct sample_position sample_position_1x[1] = {
  45    {  8,  8 },
  46 };
  47
  48 static const struct sample_position sample_position_4x[4] = {
  49    {  6,  2 }, /* distance from the center is sqrt(40) */
  50    { 14,  6 }, /* distance from the center is sqrt(40) */
  51    {  2, 10 }, /* distance from the center is sqrt(40) */
  52    { 10, 14 }, /* distance from the center is sqrt(40) */
  53 };
  54
  55 static const struct sample_position sample_position_8x[8] = {
  56    {  7,  9 }, /* distance from the center is sqrt(2) */
  57    {  9, 13 }, /* distance from the center is sqrt(26) */
  58    { 11,  3 }, /* distance from the center is sqrt(34) */
  59    { 13, 11 }, /* distance from the center is sqrt(34) */
  60    {  1,  7 }, /* distance from the center is sqrt(50) */
  61    {  5,  1 }, /* distance from the center is sqrt(58) */
  62    { 15,  5 }, /* distance from the center is sqrt(58) */
  63    {  3, 15 }, /* distance from the center is sqrt(74) */
  64 };
  65
  66 struct ilo_3d_pipeline *
  67 ilo_3d_pipeline_create(struct ilo_cp *cp, const struct ilo_dev_info *dev)
  68 {
  69    struct ilo_3d_pipeline *p;
  70    int i;
  71
  72    p = CALLOC_STRUCT(ilo_3d_pipeline);
  73    if (!p)
  74       return NULL;
  75
  76    p->cp = cp;
  77    p->dev = dev;
  78
  79    switch (p->dev->gen) {
  80    case ILO_GEN(6):
  81       ilo_3d_pipeline_init_gen6(p);
  82       break;
  83    case ILO_GEN(7):
  84       ilo_3d_pipeline_init_gen7(p);
  85       break;
  86    default:
  87       assert(!"unsupported GEN");
  88       FREE(p);
  89       return NULL;
  90       break;
  91    }
  92
  93    p->invalidate_flags = ILO_3D_PIPELINE_INVALIDATE_ALL;
  94
  95    p->workaround_bo = p->cp->winsys->alloc_buffer(p->cp->winsys,
  96          "PIPE_CONTROL workaround", 4096, 0);
  97    if (!p->workaround_bo) {
  98       ilo_warn("failed to allocate PIPE_CONTROL workaround bo\n");
  99       FREE(p);
 100       return NULL;
 101    }
 102
 103    p->packed_sample_position_1x =
 104       sample_position_1x[0].x << 4 |
 105       sample_position_1x[0].y;
 106
 107    /* pack into dwords */
 108    for (i = 0; i < 4; i++) {
 109       p->packed_sample_position_4x |=
 110          sample_position_4x[i].x << (8 * i + 4) |
 111          sample_position_4x[i].y << (8 * i);
 112
 113       p->packed_sample_position_8x[0] |=
 114          sample_position_8x[i].x << (8 * i + 4) |
 115          sample_position_8x[i].y << (8 * i);
 116
 117       p->packed_sample_position_8x[1] |=
 118          sample_position_8x[4 + i].x << (8 * i + 4) |
 119          sample_position_8x[4 + i].y << (8 * i);
 120    }
 121
 122    return p;
 123 }
 124
 125 void
 126 ilo_3d_pipeline_destroy(struct ilo_3d_pipeline *p)
 127 {
 128    if (p->workaround_bo)
 129       p->workaround_bo->unreference(p->workaround_bo);
 130
 131    FREE(p);
 132 }
 133
 134 static void
 135 handle_invalid_batch_bo(struct ilo_3d_pipeline *p, bool unset)
 136 {
 137    if (p->invalidate_flags & ILO_3D_PIPELINE_INVALIDATE_BATCH_BO) {
 138       if (p->dev->gen == ILO_GEN(6))
 139          p->state.has_gen6_wa_pipe_control = false;
 140
 141       if (unset)
 142          p->invalidate_flags &= ~ILO_3D_PIPELINE_INVALIDATE_BATCH_BO;
 143    }
 144 }
 145
 146 /* XXX move to u_prim.h */
 147 static unsigned
 148 prim_count(unsigned prim, unsigned num_verts)
 149 {
 150    unsigned num_prims;
 151
 152    u_trim_pipe_prim(prim, &num_verts);
 153
 154    switch (prim) {
 155    case PIPE_PRIM_POINTS:
 156       num_prims = num_verts;
 157       break;
 158    case PIPE_PRIM_LINES:
 159       num_prims = num_verts / 2;
 160       break;
 161    case PIPE_PRIM_LINE_LOOP:
 162       num_prims = num_verts;
 163       break;
 164    case PIPE_PRIM_LINE_STRIP:
 165       num_prims = num_verts - 1;
 166       break;
 167    case PIPE_PRIM_TRIANGLES:
 168       num_prims = num_verts / 3;
 169       break;
 170    case PIPE_PRIM_TRIANGLE_STRIP:
 171    case PIPE_PRIM_TRIANGLE_FAN:
 172       num_prims = num_verts - 2;
 173       break;
 174    case PIPE_PRIM_QUADS:
 175       num_prims = (num_verts / 4) * 2;
 176       break;
 177    case PIPE_PRIM_QUAD_STRIP:
 178       num_prims = (num_verts / 2 - 1) * 2;
 179       break;
 180    case PIPE_PRIM_POLYGON:
 181       num_prims = num_verts - 2;
 182       break;
 183    case PIPE_PRIM_LINES_ADJACENCY:
 184       num_prims = num_verts / 4;
 185       break;
 186    case PIPE_PRIM_LINE_STRIP_ADJACENCY:
 187       num_prims = num_verts - 3;
 188       break;
 189    case PIPE_PRIM_TRIANGLES_ADJACENCY:
 190       /* u_trim_pipe_prim is wrong? */
 191       num_verts += 1;
 192
 193       num_prims = num_verts / 6;
 194       break;
 195    case PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY:
 196       /* u_trim_pipe_prim is wrong? */
 197       if (num_verts >= 6)
 198          num_verts -= (num_verts % 2);
 199       else
 200          num_verts = 0;
 201
 202       num_prims = (num_verts / 2 - 2);
 203       break;
 204    default:
 205       assert(!"unknown pipe prim");
 206       num_prims = 0;
 207       break;
 208    }
 209
 210    return num_prims;
 211 }
 212
 213 /**
 214  * Emit context states and 3DPRIMITIVE.
 215  */
 216 bool
 217 ilo_3d_pipeline_emit_draw(struct ilo_3d_pipeline *p,
 218                           const struct ilo_context *ilo,
 219                           const struct pipe_draw_info *info,
 220                           int *prim_generated, int *prim_emitted)
 221 {
 222    const bool so_enabled = (ilo->stream_output_targets.num_targets > 0);
 223    bool success;
 224
 225    if (ilo->dirty & ILO_DIRTY_STREAM_OUTPUT_TARGETS &&
 226        so_enabled && !ilo->stream_output_targets.append_bitmask) {
 227       /*
 228        * We keep track of the SVBI in the driver, so that we can restore it
 229        * when the HW context is invalidated (by another process).  The value
 230        * needs to be reset when stream output is enabled and the targets are
 231        * changed.
 232        */
 233       p->state.so_num_vertices = 0;
 234
 235       /* on GEN7+, we need SOL_RESET to reset the SO write offsets */
 236       if (p->dev->gen >= ILO_GEN(7))
 237          ilo_cp_set_one_off_flags(p->cp, INTEL_EXEC_GEN7_SOL_RESET);
 238    }
 239
 240
 241    while (true) {
 242       struct ilo_cp_jmp_buf jmp;
 243       int err;
 244
 245       /* we will rewind if aperture check below fails */
 246       ilo_cp_setjmp(p->cp, &jmp);
 247
 248       handle_invalid_batch_bo(p, false);
 249
 250       /* draw! */
 251       ilo_cp_assert_no_implicit_flush(p->cp, true);
 252       p->emit_draw(p, ilo, info);
 253       ilo_cp_assert_no_implicit_flush(p->cp, false);
 254
 255       err = ilo->winsys->check_aperture_space(ilo->winsys, &p->cp->bo, 1);
 256       if (!err) {
 257          success = true;
 258          break;
 259       }
 260
 261       /* rewind */
 262       ilo_cp_longjmp(p->cp, &jmp);
 263
 264       if (ilo_cp_empty(p->cp)) {
 265          success = false;
 266          break;
 267       }
 268       else {
 269          /* flush and try again */
 270          ilo_cp_flush(p->cp);
 271       }
 272    }
 273
 274    if (success) {
 275       const int num_verts = u_vertices_per_prim(u_reduced_prim(info->mode));
 276       const int max_emit =
 277          (p->state.so_max_vertices - p->state.so_num_vertices) / num_verts;
 278       const int generated = prim_count(info->mode, info->count);
 279       const int emitted = MIN2(generated, max_emit);
 280
 281       p->state.so_num_vertices += emitted * num_verts;
 282
 283       if (prim_generated)
 284          *prim_generated = generated;
 285
 286       if (prim_emitted)
 287          *prim_emitted = emitted;
 288    }
 289
 290    p->invalidate_flags = 0x0;
 291
 292    return success;
 293 }
 294
 295 /**
 296  * Emit PIPE_CONTROL to flush all caches.
 297  */
 298 void
 299 ilo_3d_pipeline_emit_flush(struct ilo_3d_pipeline *p)
 300 {
 301    handle_invalid_batch_bo(p, true);
 302    p->emit_flush(p);
 303 }
 304
 305 /**
 306  * Emit PIPE_CONTROL with PIPE_CONTROL_WRITE_TIMESTAMP post-sync op.
 307  */
 308 void
 309 ilo_3d_pipeline_emit_write_timestamp(struct ilo_3d_pipeline *p,
 310                                      struct intel_bo *bo, int index)
 311 {
 312    handle_invalid_batch_bo(p, true);
 313    p->emit_write_timestamp(p, bo, index);
 314 }
 315
 316 /**
 317  * Emit PIPE_CONTROL with PIPE_CONTROL_WRITE_DEPTH_COUNT post-sync op.
 318  */
 319 void
 320 ilo_3d_pipeline_emit_write_depth_count(struct ilo_3d_pipeline *p,
 321                                        struct intel_bo *bo, int index)
 322 {
 323    handle_invalid_batch_bo(p, true);
 324    p->emit_write_depth_count(p, bo, index);
 325 }
 326
 327 void
 328 ilo_3d_pipeline_get_sample_position(struct ilo_3d_pipeline *p,
 329                                     unsigned sample_count,
 330                                     unsigned sample_index,
 331                                     float *x, float *y)
 332 {
 333    const struct sample_position *pos;
 334
 335    switch (sample_count) {
 336    case 1:
 337       assert(sample_index < Elements(sample_position_1x));
 338       pos = sample_position_1x;
 339       break;
 340    case 4:
 341       assert(sample_index < Elements(sample_position_4x));
 342       pos = sample_position_4x;
 343       break;
 344    case 8:
 345       assert(sample_index < Elements(sample_position_8x));
 346       pos = sample_position_8x;
 347       break;
 348    default:
 349       assert(!"unknown sample count");
 350       *x = 0.5f;
 351       *y = 0.5f;
 352       return;
 353       break;
 354    }
 355
 356    *x = (float) pos[sample_index].x / 16.0f;
 357    *y = (float) pos[sample_index].y / 16.0f;
 358 }