src/mesa/drivers/dri/i965/gen6_sf_state.c

   1 /*
   2  * Copyright © 2009 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * 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 OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *    Eric Anholt <eric@anholt.net>
  25  *
  26  */
  27
  28 #include "brw_context.h"
  29 #include "brw_state.h"
  30 #include "brw_defines.h"
  31 #include "brw_util.h"
  32 #include "compiler/nir/nir.h"
  33 #include "main/macros.h"
  34 #include "main/fbobject.h"
  35 #include "main/framebuffer.h"
  36 #include "intel_batchbuffer.h"
  37
  38 /**
  39  * Determine the appropriate attribute override value to store into the
  40  * 3DSTATE_SF structure for a given fragment shader attribute.  The attribute
  41  * override value contains two pieces of information: the location of the
  42  * attribute in the VUE (relative to urb_entry_read_offset, see below), and a
  43  * flag indicating whether to "swizzle" the attribute based on the direction
  44  * the triangle is facing.
  45  *
  46  * If an attribute is "swizzled", then the given VUE location is used for
  47  * front-facing triangles, and the VUE location that immediately follows is
  48  * used for back-facing triangles.  We use this to implement the mapping from
  49  * gl_FrontColor/gl_BackColor to gl_Color.
  50  *
  51  * urb_entry_read_offset is the offset into the VUE at which the SF unit is
  52  * being instructed to begin reading attribute data.  It can be set to a
  53  * nonzero value to prevent the SF unit from wasting time reading elements of
  54  * the VUE that are not needed by the fragment shader.  It is measured in
  55  * 256-bit increments.
  56  */
  57 static uint32_t
  58 get_attr_override(const struct brw_vue_map *vue_map, int urb_entry_read_offset,
  59                   int fs_attr, bool two_side_color, uint32_t *max_source_attr)
  60 {
  61    /* Find the VUE slot for this attribute. */
  62    int slot = vue_map->varying_to_slot[fs_attr];
  63
  64    /* Viewport and Layer are stored in the VUE header.  We need to override
  65     * them to zero if earlier stages didn't write them, as GL requires that
  66     * they read back as zero when not explicitly set.
  67     */
  68    if (fs_attr == VARYING_SLOT_VIEWPORT || fs_attr == VARYING_SLOT_LAYER) {
  69       unsigned override =
  70          ATTRIBUTE_0_OVERRIDE_X | ATTRIBUTE_0_OVERRIDE_W |
  71          ATTRIBUTE_CONST_0000 << ATTRIBUTE_0_CONST_SOURCE_SHIFT;
  72
  73       if (!(vue_map->slots_valid & VARYING_BIT_LAYER))
  74          override |= ATTRIBUTE_0_OVERRIDE_Y;
  75       if (!(vue_map->slots_valid & VARYING_BIT_VIEWPORT))
  76          override |= ATTRIBUTE_0_OVERRIDE_Z;
  77
  78       return override;
  79    }
  80
  81    /* If there was only a back color written but not front, use back
  82     * as the color instead of undefined
  83     */
  84    if (slot == -1 && fs_attr == VARYING_SLOT_COL0)
  85       slot = vue_map->varying_to_slot[VARYING_SLOT_BFC0];
  86    if (slot == -1 && fs_attr == VARYING_SLOT_COL1)
  87       slot = vue_map->varying_to_slot[VARYING_SLOT_BFC1];
  88
  89    if (slot == -1) {
  90       /* This attribute does not exist in the VUE--that means that the vertex
  91        * shader did not write to it.  This means that either:
  92        *
  93        * (a) This attribute is a texture coordinate, and it is going to be
  94        * replaced with point coordinates (as a consequence of a call to
  95        * glTexEnvi(GL_POINT_SPRITE, GL_COORD_REPLACE, GL_TRUE)), so the
  96        * hardware will ignore whatever attribute override we supply.
  97        *
  98        * (b) This attribute is read by the fragment shader but not written by
  99        * the vertex shader, so its value is undefined.  Therefore the
 100        * attribute override we supply doesn't matter.
 101        *
 102        * (c) This attribute is gl_PrimitiveID, and it wasn't written by the
 103        * previous shader stage.
 104        *
 105        * Note that we don't have to worry about the cases where the attribute
 106        * is gl_PointCoord or is undergoing point sprite coordinate
 107        * replacement, because in those cases, this function isn't called.
 108        *
 109        * In case (c), we need to program the attribute overrides so that the
 110        * primitive ID will be stored in this slot.  In every other case, the
 111        * attribute override we supply doesn't matter.  So just go ahead and
 112        * program primitive ID in every case.
 113        */
 114       return (ATTRIBUTE_0_OVERRIDE_W |
 115               ATTRIBUTE_0_OVERRIDE_Z |
 116               ATTRIBUTE_0_OVERRIDE_Y |
 117               ATTRIBUTE_0_OVERRIDE_X |
 118               (ATTRIBUTE_CONST_PRIM_ID << ATTRIBUTE_0_CONST_SOURCE_SHIFT));
 119    }
 120
 121    /* Compute the location of the attribute relative to urb_entry_read_offset.
 122     * Each increment of urb_entry_read_offset represents a 256-bit value, so
 123     * it counts for two 128-bit VUE slots.
 124     */
 125    int source_attr = slot - 2 * urb_entry_read_offset;
 126    assert(source_attr >= 0 && source_attr < 32);
 127
 128    /* If we are doing two-sided color, and the VUE slot following this one
 129     * represents a back-facing color, then we need to instruct the SF unit to
 130     * do back-facing swizzling.
 131     */
 132    bool swizzling = two_side_color &&
 133       ((vue_map->slot_to_varying[slot] == VARYING_SLOT_COL0 &&
 134         vue_map->slot_to_varying[slot+1] == VARYING_SLOT_BFC0) ||
 135        (vue_map->slot_to_varying[slot] == VARYING_SLOT_COL1 &&
 136         vue_map->slot_to_varying[slot+1] == VARYING_SLOT_BFC1));
 137
 138    /* Update max_source_attr.  If swizzling, the SF will read this slot + 1. */
 139    if (*max_source_attr < source_attr + swizzling)
 140       *max_source_attr = source_attr + swizzling;
 141
 142    if (swizzling) {
 143       return source_attr |
 144          (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
 145    }
 146
 147    return source_attr;
 148 }
 149
 150
 151 /**
 152  * Create the mapping from the FS inputs we produce to the previous pipeline
 153  * stage (GS or VS) outputs they source from.
 154  */
 155 void
 156 calculate_attr_overrides(const struct brw_context *brw,
 157                          uint16_t *attr_overrides,
 158                          uint32_t *point_sprite_enables,
 159                          uint32_t *urb_entry_read_length,
 160                          uint32_t *urb_entry_read_offset)
 161 {
 162    /* BRW_NEW_FS_PROG_DATA */
 163    const struct brw_wm_prog_data *wm_prog_data =
 164       brw_wm_prog_data(brw->wm.base.prog_data);
 165    uint32_t max_source_attr = 0;
 166
 167    *point_sprite_enables = 0;
 168
 169    *urb_entry_read_offset = BRW_SF_URB_ENTRY_READ_OFFSET;
 170
 171    /* BRW_NEW_FRAGMENT_PROGRAM
 172     *
 173     * If the fragment shader reads VARYING_SLOT_LAYER, then we need to pass in
 174     * the full vertex header.  Otherwise, we can program the SF to start
 175     * reading at an offset of 1 (2 varying slots) to skip unnecessary data:
 176     * - VARYING_SLOT_PSIZ and BRW_VARYING_SLOT_NDC on gen4-5
 177     * - VARYING_SLOT_{PSIZ,LAYER} and VARYING_SLOT_POS on gen6+
 178     */
 179
 180    bool fs_needs_vue_header =
 181       brw->fragment_program->Base.nir->info->inputs_read &
 182       (VARYING_BIT_LAYER | VARYING_BIT_VIEWPORT);
 183
 184    *urb_entry_read_offset = fs_needs_vue_header ? 0 : 1;
 185
 186    /* From the Ivybridge PRM, Vol 2 Part 1, 3DSTATE_SBE,
 187     * description of dw10 Point Sprite Texture Coordinate Enable:
 188     *
 189     * "This field must be programmed to zero when non-point primitives
 190     * are rendered."
 191     *
 192     * The SandyBridge PRM doesn't explicitly say that point sprite enables
 193     * must be programmed to zero when rendering non-point primitives, but
 194     * the IvyBridge PRM does, and if we don't, we get garbage.
 195     *
 196     * This is not required on Haswell, as the hardware ignores this state
 197     * when drawing non-points -- although we do still need to be careful to
 198     * correctly set the attr overrides.
 199     *
 200     * _NEW_POLYGON
 201     * BRW_NEW_PRIMITIVE | BRW_NEW_GS_PROG_DATA | BRW_NEW_TES_PROG_DATA
 202     */
 203    bool drawing_points = brw_is_drawing_points(brw);
 204
 205    /* Initialize all the attr_overrides to 0.  In the loop below we'll modify
 206     * just the ones that correspond to inputs used by the fs.
 207     */
 208    memset(attr_overrides, 0, 16*sizeof(*attr_overrides));
 209
 210    for (int attr = 0; attr < VARYING_SLOT_MAX; attr++) {
 211       int input_index = wm_prog_data->urb_setup[attr];
 212
 213       if (input_index < 0)
 214          continue;
 215
 216       /* _NEW_POINT */
 217       bool point_sprite = false;
 218       if (drawing_points) {
 219          if (brw->ctx.Point.PointSprite &&
 220              (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7) &&
 221              (brw->ctx.Point.CoordReplace & (1u << (attr - VARYING_SLOT_TEX0)))) {
 222             point_sprite = true;
 223          }
 224
 225          if (attr == VARYING_SLOT_PNTC)
 226             point_sprite = true;
 227
 228          if (point_sprite)
 229             *point_sprite_enables |= (1 << input_index);
 230       }
 231
 232       /* BRW_NEW_VUE_MAP_GEOM_OUT | _NEW_LIGHT | _NEW_PROGRAM */
 233       uint16_t attr_override = point_sprite ? 0 :
 234          get_attr_override(&brw->vue_map_geom_out,
 235                            *urb_entry_read_offset, attr,
 236                            brw->ctx.VertexProgram._TwoSideEnabled,
 237                            &max_source_attr);
 238
 239       /* The hardware can only do the overrides on 16 overrides at a
 240        * time, and the other up to 16 have to be lined up so that the
 241        * input index = the output index.  We'll need to do some
 242        * tweaking to make sure that's the case.
 243        */
 244       if (input_index < 16)
 245          attr_overrides[input_index] = attr_override;
 246       else
 247          assert(attr_override == input_index);
 248    }
 249
 250    /* From the Sandy Bridge PRM, Volume 2, Part 1, documentation for
 251     * 3DSTATE_SF DWord 1 bits 15:11, "Vertex URB Entry Read Length":
 252     *
 253     * "This field should be set to the minimum length required to read the
 254     *  maximum source attribute.  The maximum source attribute is indicated
 255     *  by the maximum value of the enabled Attribute # Source Attribute if
 256     *  Attribute Swizzle Enable is set, Number of Output Attributes-1 if
 257     *  enable is not set.
 258     *  read_length = ceiling((max_source_attr + 1) / 2)
 259     *
 260     *  [errata] Corruption/Hang possible if length programmed larger than
 261     *  recommended"
 262     *
 263     * Similar text exists for Ivy Bridge.
 264     */
 265    *urb_entry_read_length = ALIGN(max_source_attr + 1, 2) / 2;
 266 }
 267
 268
 269 static void
 270 upload_sf_state(struct brw_context *brw)
 271 {
 272    struct gl_context *ctx = &brw->ctx;
 273    /* BRW_NEW_FS_PROG_DATA */
 274    const struct brw_wm_prog_data *wm_prog_data =
 275       brw_wm_prog_data(brw->wm.base.prog_data);
 276    uint32_t num_outputs = wm_prog_data->num_varying_inputs;
 277    uint32_t dw1, dw2, dw3, dw4;
 278    uint32_t point_sprite_enables;
 279    int i;
 280    /* _NEW_BUFFER */
 281    bool render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 282    const bool multisampled_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 283
 284    float point_size;
 285    uint16_t attr_overrides[16];
 286    uint32_t point_sprite_origin;
 287
 288    dw1 = GEN6_SF_SWIZZLE_ENABLE | num_outputs << GEN6_SF_NUM_OUTPUTS_SHIFT;
 289    dw2 = GEN6_SF_STATISTICS_ENABLE;
 290
 291    if (brw->sf.viewport_transform_enable)
 292        dw2 |= GEN6_SF_VIEWPORT_TRANSFORM_ENABLE;
 293
 294    dw3 = 0;
 295    dw4 = 0;
 296
 297    /* _NEW_POLYGON */
 298    if (ctx->Polygon._FrontBit == render_to_fbo)
 299       dw2 |= GEN6_SF_WINDING_CCW;
 300
 301    if (ctx->Polygon.OffsetFill)
 302        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID;
 303
 304    if (ctx->Polygon.OffsetLine)
 305        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME;
 306
 307    if (ctx->Polygon.OffsetPoint)
 308        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT;
 309
 310    switch (ctx->Polygon.FrontMode) {
 311    case GL_FILL:
 312        dw2 |= GEN6_SF_FRONT_SOLID;
 313        break;
 314
 315    case GL_LINE:
 316        dw2 |= GEN6_SF_FRONT_WIREFRAME;
 317        break;
 318
 319    case GL_POINT:
 320        dw2 |= GEN6_SF_FRONT_POINT;
 321        break;
 322
 323    default:
 324        unreachable("not reached");
 325    }
 326
 327    switch (ctx->Polygon.BackMode) {
 328    case GL_FILL:
 329        dw2 |= GEN6_SF_BACK_SOLID;
 330        break;
 331
 332    case GL_LINE:
 333        dw2 |= GEN6_SF_BACK_WIREFRAME;
 334        break;
 335
 336    case GL_POINT:
 337        dw2 |= GEN6_SF_BACK_POINT;
 338        break;
 339
 340    default:
 341        unreachable("not reached");
 342    }
 343
 344    /* _NEW_SCISSOR | _NEW_POLYGON,
 345     * BRW_NEW_GS_PROG_DATA | BRW_NEW_TES_PROG_DATA | BRW_NEW_PRIMITIVE
 346     */
 347    if (ctx->Scissor.EnableFlags ||
 348        brw_is_drawing_points(brw) || brw_is_drawing_lines(brw))
 349       dw3 |= GEN6_SF_SCISSOR_ENABLE;
 350
 351    /* _NEW_POLYGON */
 352    if (ctx->Polygon.CullFlag) {
 353       switch (ctx->Polygon.CullFaceMode) {
 354       case GL_FRONT:
 355          dw3 |= GEN6_SF_CULL_FRONT;
 356          break;
 357       case GL_BACK:
 358          dw3 |= GEN6_SF_CULL_BACK;
 359          break;
 360       case GL_FRONT_AND_BACK:
 361          dw3 |= GEN6_SF_CULL_BOTH;
 362          break;
 363       default:
 364          unreachable("not reached");
 365       }
 366    } else {
 367       dw3 |= GEN6_SF_CULL_NONE;
 368    }
 369
 370    /* _NEW_LINE */
 371    {
 372       uint32_t line_width_u3_7 = brw_get_line_width(brw);
 373       dw3 |= line_width_u3_7 << GEN6_SF_LINE_WIDTH_SHIFT;
 374    }
 375    if (ctx->Line.SmoothFlag) {
 376       dw3 |= GEN6_SF_LINE_AA_ENABLE;
 377       dw3 |= GEN6_SF_LINE_AA_MODE_TRUE;
 378       dw3 |= GEN6_SF_LINE_END_CAP_WIDTH_1_0;
 379    }
 380    /* _NEW_MULTISAMPLE */
 381    if (multisampled_fbo && ctx->Multisample.Enabled)
 382       dw3 |= GEN6_SF_MSRAST_ON_PATTERN;
 383
 384    /* _NEW_PROGRAM | _NEW_POINT, BRW_NEW_VUE_MAP_GEOM_OUT */
 385    if (use_state_point_size(brw))
 386       dw4 |= GEN6_SF_USE_STATE_POINT_WIDTH;
 387
 388    /* _NEW_POINT - Clamp to ARB_point_parameters user limits */
 389    point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
 390
 391    /* Clamp to the hardware limits and convert to fixed point */
 392    dw4 |= U_FIXED(CLAMP(point_size, 0.125f, 255.875f), 3);
 393
 394    /*
 395     * Window coordinates in an FBO are inverted, which means point
 396     * sprite origin must be inverted, too.
 397     */
 398    if ((ctx->Point.SpriteOrigin == GL_LOWER_LEFT) != render_to_fbo) {
 399       point_sprite_origin = GEN6_SF_POINT_SPRITE_LOWERLEFT;
 400    } else {
 401       point_sprite_origin = GEN6_SF_POINT_SPRITE_UPPERLEFT;
 402    }
 403    dw1 |= point_sprite_origin;
 404
 405    /* _NEW_LIGHT */
 406    if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
 407       dw4 |=
 408          (2 << GEN6_SF_TRI_PROVOKE_SHIFT) |
 409          (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT) |
 410          (1 << GEN6_SF_LINE_PROVOKE_SHIFT);
 411    } else {
 412       dw4 |=
 413          (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT);
 414    }
 415
 416    /* BRW_NEW_VUE_MAP_GEOM_OUT | BRW_NEW_FRAGMENT_PROGRAM |
 417     * _NEW_POINT | _NEW_LIGHT | _NEW_PROGRAM | BRW_NEW_FS_PROG_DATA
 418     */
 419    uint32_t urb_entry_read_length;
 420    uint32_t urb_entry_read_offset;
 421    calculate_attr_overrides(brw, attr_overrides, &point_sprite_enables,
 422                             &urb_entry_read_length, &urb_entry_read_offset);
 423    dw1 |= (urb_entry_read_length << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
 424            urb_entry_read_offset << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT);
 425
 426    BEGIN_BATCH(20);
 427    OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
 428    OUT_BATCH(dw1);
 429    OUT_BATCH(dw2);
 430    OUT_BATCH(dw3);
 431    OUT_BATCH(dw4);
 432    OUT_BATCH_F(ctx->Polygon.OffsetUnits * 2); /* constant.  copied from gen4 */
 433    OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
 434    OUT_BATCH_F(ctx->Polygon.OffsetClamp); /* global depth offset clamp */
 435    for (i = 0; i < 8; i++) {
 436       OUT_BATCH(attr_overrides[i * 2] | attr_overrides[i * 2 + 1] << 16);
 437    }
 438    OUT_BATCH(point_sprite_enables); /* dw16 */
 439    OUT_BATCH(wm_prog_data->flat_inputs);
 440    OUT_BATCH(0); /* wrapshortest enables 0-7 */
 441    OUT_BATCH(0); /* wrapshortest enables 8-15 */
 442    ADVANCE_BATCH();
 443 }
 444
 445 const struct brw_tracked_state gen6_sf_state = {
 446    .dirty = {
 447       .mesa  = _NEW_BUFFERS |
 448                _NEW_LIGHT |
 449                _NEW_LINE |
 450                _NEW_MULTISAMPLE |
 451                _NEW_POINT |
 452                _NEW_POLYGON |
 453                _NEW_PROGRAM |
 454                _NEW_SCISSOR,
 455       .brw   = BRW_NEW_BLORP |
 456                BRW_NEW_CONTEXT |
 457                BRW_NEW_FRAGMENT_PROGRAM |
 458                BRW_NEW_FS_PROG_DATA |
 459                BRW_NEW_GS_PROG_DATA |
 460                BRW_NEW_PRIMITIVE |
 461                BRW_NEW_TES_PROG_DATA |
 462                BRW_NEW_VUE_MAP_GEOM_OUT,
 463    },
 464    .emit = upload_sf_state,
 465 };