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 "main/macros.h"
  33 #include "main/fbobject.h"
  34 #include "intel_batchbuffer.h"
  35
  36 /**
  37  * Determine the appropriate attribute override value to store into the
  38  * 3DSTATE_SF structure for a given fragment shader attribute.  The attribute
  39  * override value contains two pieces of information: the location of the
  40  * attribute in the VUE (relative to urb_entry_read_offset, see below), and a
  41  * flag indicating whether to "swizzle" the attribute based on the direction
  42  * the triangle is facing.
  43  *
  44  * If an attribute is "swizzled", then the given VUE location is used for
  45  * front-facing triangles, and the VUE location that immediately follows is
  46  * used for back-facing triangles.  We use this to implement the mapping from
  47  * gl_FrontColor/gl_BackColor to gl_Color.
  48  *
  49  * urb_entry_read_offset is the offset into the VUE at which the SF unit is
  50  * being instructed to begin reading attribute data.  It can be set to a
  51  * nonzero value to prevent the SF unit from wasting time reading elements of
  52  * the VUE that are not needed by the fragment shader.  It is measured in
  53  * 256-bit increments.
  54  */
  55 static uint32_t
  56 get_attr_override(const struct brw_vue_map *vue_map, int urb_entry_read_offset,
  57                   int fs_attr, bool two_side_color, uint32_t *max_source_attr)
  58 {
  59    /* Find the VUE slot for this attribute. */
  60    int slot = vue_map->varying_to_slot[fs_attr];
  61
  62    /* If there was only a back color written but not front, use back
  63     * as the color instead of undefined
  64     */
  65    if (slot == -1 && fs_attr == VARYING_SLOT_COL0)
  66       slot = vue_map->varying_to_slot[VARYING_SLOT_BFC0];
  67    if (slot == -1 && fs_attr == VARYING_SLOT_COL1)
  68       slot = vue_map->varying_to_slot[VARYING_SLOT_BFC1];
  69
  70    if (slot == -1) {
  71       /* This attribute does not exist in the VUE--that means that the vertex
  72        * shader did not write to it.  This means that either:
  73        *
  74        * (a) This attribute is a texture coordinate, and it is going to be
  75        * replaced with point coordinates (as a consequence of a call to
  76        * glTexEnvi(GL_POINT_SPRITE, GL_COORD_REPLACE, GL_TRUE)), so the
  77        * hardware will ignore whatever attribute override we supply.
  78        *
  79        * (b) This attribute is read by the fragment shader but not written by
  80        * the vertex shader, so its value is undefined.  Therefore the
  81        * attribute override we supply doesn't matter.
  82        *
  83        * (c) This attribute is gl_PrimitiveID, and it wasn't written by the
  84        * previous shader stage.
  85        *
  86        * Note that we don't have to worry about the cases where the attribute
  87        * is gl_PointCoord or is undergoing point sprite coordinate
  88        * replacement, because in those cases, this function isn't called.
  89        *
  90        * In case (c), we need to program the attribute overrides so that the
  91        * primitive ID will be stored in this slot.  In every other case, the
  92        * attribute override we supply doesn't matter.  So just go ahead and
  93        * program primitive ID in every case.
  94        */
  95       return (ATTRIBUTE_0_OVERRIDE_W |
  96               ATTRIBUTE_0_OVERRIDE_Z |
  97               ATTRIBUTE_0_OVERRIDE_Y |
  98               ATTRIBUTE_0_OVERRIDE_X |
  99               (ATTRIBUTE_CONST_PRIM_ID << ATTRIBUTE_0_CONST_SOURCE_SHIFT));
 100    }
 101
 102    /* Compute the location of the attribute relative to urb_entry_read_offset.
 103     * Each increment of urb_entry_read_offset represents a 256-bit value, so
 104     * it counts for two 128-bit VUE slots.
 105     */
 106    int source_attr = slot - 2 * urb_entry_read_offset;
 107    assert(source_attr >= 0 && source_attr < 32);
 108
 109    /* If we are doing two-sided color, and the VUE slot following this one
 110     * represents a back-facing color, then we need to instruct the SF unit to
 111     * do back-facing swizzling.
 112     */
 113    bool swizzling = two_side_color &&
 114       ((vue_map->slot_to_varying[slot] == VARYING_SLOT_COL0 &&
 115         vue_map->slot_to_varying[slot+1] == VARYING_SLOT_BFC0) ||
 116        (vue_map->slot_to_varying[slot] == VARYING_SLOT_COL1 &&
 117         vue_map->slot_to_varying[slot+1] == VARYING_SLOT_BFC1));
 118
 119    /* Update max_source_attr.  If swizzling, the SF will read this slot + 1. */
 120    if (*max_source_attr < source_attr + swizzling)
 121       *max_source_attr = source_attr + swizzling;
 122
 123    if (swizzling) {
 124       return source_attr |
 125          (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
 126    }
 127
 128    return source_attr;
 129 }
 130
 131
 132 static bool
 133 is_drawing_points(const struct brw_context *brw)
 134 {
 135    /* Determine if the primitives *reaching the SF* are points */
 136    /* _NEW_POLYGON */
 137    if (brw->ctx.Polygon.FrontMode == GL_POINT ||
 138        brw->ctx.Polygon.BackMode == GL_POINT) {
 139       return true;
 140    }
 141
 142    if (brw->geometry_program) {
 143       /* BRW_NEW_GEOMETRY_PROGRAM */
 144       return brw->geometry_program->OutputType == GL_POINTS;
 145    } else {
 146       /* BRW_NEW_PRIMITIVE */
 147       return brw->primitive == _3DPRIM_POINTLIST;
 148    }
 149 }
 150
 151
 152 /**
 153  * Create the mapping from the FS inputs we produce to the previous pipeline
 154  * stage (GS or VS) outputs they source from.
 155  */
 156 void
 157 calculate_attr_overrides(const struct brw_context *brw,
 158                          uint16_t *attr_overrides,
 159                          uint32_t *point_sprite_enables,
 160                          uint32_t *flat_enables,
 161                          uint32_t *urb_entry_read_length)
 162 {
 163    const int urb_entry_read_offset = BRW_SF_URB_ENTRY_READ_OFFSET;
 164    uint32_t max_source_attr = 0;
 165
 166    *point_sprite_enables = 0;
 167    *flat_enables = 0;
 168
 169    /* _NEW_LIGHT */
 170    bool shade_model_flat = brw->ctx.Light.ShadeModel == GL_FLAT;
 171
 172    /* From the Ivybridge PRM, Vol 2 Part 1, 3DSTATE_SBE,
 173     * description of dw10 Point Sprite Texture Coordinate Enable:
 174     *
 175     * "This field must be programmed to zero when non-point primitives
 176     * are rendered."
 177     *
 178     * The SandyBridge PRM doesn't explicitly say that point sprite enables
 179     * must be programmed to zero when rendering non-point primitives, but
 180     * the IvyBridge PRM does, and if we don't, we get garbage.
 181     *
 182     * This is not required on Haswell, as the hardware ignores this state
 183     * when drawing non-points -- although we do still need to be careful to
 184     * correctly set the attr overrides.
 185     */
 186    /* BRW_NEW_PRIMITIVE | BRW_NEW_GEOMETRY_PROGRAM */
 187    bool drawing_points = is_drawing_points(brw);
 188
 189    /* Initialize all the attr_overrides to 0.  In the loop below we'll modify
 190     * just the ones that correspond to inputs used by the fs.
 191     */
 192    memset(attr_overrides, 0, 16*sizeof(*attr_overrides));
 193
 194    for (int attr = 0; attr < VARYING_SLOT_MAX; attr++) {
 195       /* BRW_NEW_FRAGMENT_PROGRAM */
 196       enum glsl_interp_qualifier interp_qualifier =
 197          brw->fragment_program->InterpQualifier[attr];
 198       bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1;
 199       /* BRW_NEW_FS_PROG_DATA */
 200       int input_index = brw->wm.prog_data->urb_setup[attr];
 201
 202       if (input_index < 0)
 203          continue;
 204
 205       /* _NEW_POINT */
 206       bool point_sprite = false;
 207       if (drawing_points) {
 208          if (brw->ctx.Point.PointSprite &&
 209              (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7) &&
 210              brw->ctx.Point.CoordReplace[attr - VARYING_SLOT_TEX0]) {
 211             point_sprite = true;
 212          }
 213
 214          if (attr == VARYING_SLOT_PNTC)
 215             point_sprite = true;
 216
 217          if (point_sprite)
 218             *point_sprite_enables |= (1 << input_index);
 219       }
 220
 221       /* flat shading */
 222       if (interp_qualifier == INTERP_QUALIFIER_FLAT ||
 223           (shade_model_flat && is_gl_Color &&
 224            interp_qualifier == INTERP_QUALIFIER_NONE))
 225          *flat_enables |= (1 << input_index);
 226
 227       /* BRW_NEW_VUE_MAP_GEOM_OUT | _NEW_LIGHT | _NEW_PROGRAM */
 228       uint16_t attr_override = point_sprite ? 0 :
 229          get_attr_override(&brw->vue_map_geom_out,
 230                            urb_entry_read_offset, attr,
 231                            brw->ctx.VertexProgram._TwoSideEnabled,
 232                            &max_source_attr);
 233
 234       /* The hardware can only do the overrides on 16 overrides at a
 235        * time, and the other up to 16 have to be lined up so that the
 236        * input index = the output index.  We'll need to do some
 237        * tweaking to make sure that's the case.
 238        */
 239       if (input_index < 16)
 240          attr_overrides[input_index] = attr_override;
 241       else
 242          assert(attr_override == input_index);
 243    }
 244
 245    /* From the Sandy Bridge PRM, Volume 2, Part 1, documentation for
 246     * 3DSTATE_SF DWord 1 bits 15:11, "Vertex URB Entry Read Length":
 247     *
 248     * "This field should be set to the minimum length required to read the
 249     *  maximum source attribute.  The maximum source attribute is indicated
 250     *  by the maximum value of the enabled Attribute # Source Attribute if
 251     *  Attribute Swizzle Enable is set, Number of Output Attributes-1 if
 252     *  enable is not set.
 253     *  read_length = ceiling((max_source_attr + 1) / 2)
 254     *
 255     *  [errata] Corruption/Hang possible if length programmed larger than
 256     *  recommended"
 257     *
 258     * Similar text exists for Ivy Bridge.
 259     */
 260    *urb_entry_read_length = ALIGN(max_source_attr + 1, 2) / 2;
 261 }
 262
 263
 264 static void
 265 upload_sf_state(struct brw_context *brw)
 266 {
 267    struct gl_context *ctx = &brw->ctx;
 268    /* BRW_NEW_FS_PROG_DATA */
 269    uint32_t num_outputs = brw->wm.prog_data->num_varying_inputs;
 270    uint32_t dw1, dw2, dw3, dw4;
 271    uint32_t point_sprite_enables;
 272    uint32_t flat_enables;
 273    int i;
 274    /* _NEW_BUFFER */
 275    bool render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 276    bool multisampled_fbo = ctx->DrawBuffer->Visual.samples > 1;
 277
 278    const int urb_entry_read_offset = BRW_SF_URB_ENTRY_READ_OFFSET;
 279    float point_size;
 280    uint16_t attr_overrides[16];
 281    uint32_t point_sprite_origin;
 282
 283    dw1 = GEN6_SF_SWIZZLE_ENABLE | num_outputs << GEN6_SF_NUM_OUTPUTS_SHIFT;
 284    dw2 = GEN6_SF_STATISTICS_ENABLE;
 285
 286    if (brw->sf.viewport_transform_enable)
 287        dw2 |= GEN6_SF_VIEWPORT_TRANSFORM_ENABLE;
 288
 289    dw3 = 0;
 290    dw4 = 0;
 291
 292    /* _NEW_POLYGON */
 293    if (ctx->Polygon._FrontBit == render_to_fbo)
 294       dw2 |= GEN6_SF_WINDING_CCW;
 295
 296    if (ctx->Polygon.OffsetFill)
 297        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID;
 298
 299    if (ctx->Polygon.OffsetLine)
 300        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME;
 301
 302    if (ctx->Polygon.OffsetPoint)
 303        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT;
 304
 305    switch (ctx->Polygon.FrontMode) {
 306    case GL_FILL:
 307        dw2 |= GEN6_SF_FRONT_SOLID;
 308        break;
 309
 310    case GL_LINE:
 311        dw2 |= GEN6_SF_FRONT_WIREFRAME;
 312        break;
 313
 314    case GL_POINT:
 315        dw2 |= GEN6_SF_FRONT_POINT;
 316        break;
 317
 318    default:
 319        unreachable("not reached");
 320    }
 321
 322    switch (ctx->Polygon.BackMode) {
 323    case GL_FILL:
 324        dw2 |= GEN6_SF_BACK_SOLID;
 325        break;
 326
 327    case GL_LINE:
 328        dw2 |= GEN6_SF_BACK_WIREFRAME;
 329        break;
 330
 331    case GL_POINT:
 332        dw2 |= GEN6_SF_BACK_POINT;
 333        break;
 334
 335    default:
 336        unreachable("not reached");
 337    }
 338
 339    /* _NEW_SCISSOR */
 340    if (ctx->Scissor.EnableFlags)
 341       dw3 |= GEN6_SF_SCISSOR_ENABLE;
 342
 343    /* _NEW_POLYGON */
 344    if (ctx->Polygon.CullFlag) {
 345       switch (ctx->Polygon.CullFaceMode) {
 346       case GL_FRONT:
 347          dw3 |= GEN6_SF_CULL_FRONT;
 348          break;
 349       case GL_BACK:
 350          dw3 |= GEN6_SF_CULL_BACK;
 351          break;
 352       case GL_FRONT_AND_BACK:
 353          dw3 |= GEN6_SF_CULL_BOTH;
 354          break;
 355       default:
 356          unreachable("not reached");
 357       }
 358    } else {
 359       dw3 |= GEN6_SF_CULL_NONE;
 360    }
 361
 362    /* _NEW_LINE */
 363    {
 364       /* OpenGL dictates that line width should be rounded to the nearest
 365        * integer
 366        */
 367       float line_width =
 368          roundf(CLAMP(ctx->Line.Width, 0.0, ctx->Const.MaxLineWidth));
 369       uint32_t line_width_u3_7 = U_FIXED(line_width, 7);
 370
 371       /* Line width of 0 is not allowed when MSAA enabled */
 372       if (ctx->Multisample._Enabled) {
 373          if (line_width_u3_7 == 0)
 374              line_width_u3_7 = 1;
 375       } else if (ctx->Line.SmoothFlag && ctx->Line.Width < 1.5) {
 376          /* For 1 pixel line thickness or less, the general
 377           * anti-aliasing algorithm gives up, and a garbage line is
 378           * generated.  Setting a Line Width of 0.0 specifies the
 379           * rasterization of the "thinnest" (one-pixel-wide),
 380           * non-antialiased lines.
 381           *
 382           * Lines rendered with zero Line Width are rasterized using
 383           * Grid Intersection Quantization rules as specified by
 384           * bspec section 6.3.12.1 Zero-Width (Cosmetic) Line
 385           * Rasterization.
 386           */
 387          line_width_u3_7 = 0;
 388       }
 389       dw3 |= line_width_u3_7 << GEN6_SF_LINE_WIDTH_SHIFT;
 390    }
 391    if (ctx->Line.SmoothFlag) {
 392       dw3 |= GEN6_SF_LINE_AA_ENABLE;
 393       dw3 |= GEN6_SF_LINE_AA_MODE_TRUE;
 394       dw3 |= GEN6_SF_LINE_END_CAP_WIDTH_1_0;
 395    }
 396    /* _NEW_MULTISAMPLE */
 397    if (multisampled_fbo && ctx->Multisample.Enabled)
 398       dw3 |= GEN6_SF_MSRAST_ON_PATTERN;
 399
 400    /* _NEW_PROGRAM | _NEW_POINT */
 401    if (!(ctx->VertexProgram.PointSizeEnabled ||
 402          ctx->Point._Attenuated))
 403       dw4 |= GEN6_SF_USE_STATE_POINT_WIDTH;
 404
 405    /* Clamp to ARB_point_parameters user limits */
 406    point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
 407
 408    /* Clamp to the hardware limits and convert to fixed point */
 409    dw4 |= U_FIXED(CLAMP(point_size, 0.125, 255.875), 3);
 410
 411    /*
 412     * Window coordinates in an FBO are inverted, which means point
 413     * sprite origin must be inverted, too.
 414     */
 415    if ((ctx->Point.SpriteOrigin == GL_LOWER_LEFT) != render_to_fbo) {
 416       point_sprite_origin = GEN6_SF_POINT_SPRITE_LOWERLEFT;
 417    } else {
 418       point_sprite_origin = GEN6_SF_POINT_SPRITE_UPPERLEFT;
 419    }
 420    dw1 |= point_sprite_origin;
 421
 422    /* _NEW_LIGHT */
 423    if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
 424       dw4 |=
 425          (2 << GEN6_SF_TRI_PROVOKE_SHIFT) |
 426          (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT) |
 427          (1 << GEN6_SF_LINE_PROVOKE_SHIFT);
 428    } else {
 429       dw4 |=
 430          (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT);
 431    }
 432
 433    /* BRW_NEW_VUE_MAP_GEOM_OUT | BRW_NEW_FRAGMENT_PROGRAM |
 434     * _NEW_POINT | _NEW_LIGHT | _NEW_PROGRAM | BRW_NEW_FS_PROG_DATA
 435     */
 436    uint32_t urb_entry_read_length;
 437    calculate_attr_overrides(brw, attr_overrides, &point_sprite_enables,
 438                             &flat_enables, &urb_entry_read_length);
 439    dw1 |= (urb_entry_read_length << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
 440            urb_entry_read_offset << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT);
 441
 442    BEGIN_BATCH(20);
 443    OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
 444    OUT_BATCH(dw1);
 445    OUT_BATCH(dw2);
 446    OUT_BATCH(dw3);
 447    OUT_BATCH(dw4);
 448    OUT_BATCH_F(ctx->Polygon.OffsetUnits * 2); /* constant.  copied from gen4 */
 449    OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
 450    OUT_BATCH_F(ctx->Polygon.OffsetClamp); /* global depth offset clamp */
 451    for (i = 0; i < 8; i++) {
 452       OUT_BATCH(attr_overrides[i * 2] | attr_overrides[i * 2 + 1] << 16);
 453    }
 454    OUT_BATCH(point_sprite_enables); /* dw16 */
 455    OUT_BATCH(flat_enables);
 456    OUT_BATCH(0); /* wrapshortest enables 0-7 */
 457    OUT_BATCH(0); /* wrapshortest enables 8-15 */
 458    ADVANCE_BATCH();
 459 }
 460
 461 const struct brw_tracked_state gen6_sf_state = {
 462    .dirty = {
 463       .mesa  = _NEW_BUFFERS |
 464                _NEW_LIGHT |
 465                _NEW_LINE |
 466                _NEW_MULTISAMPLE |
 467                _NEW_POINT |
 468                _NEW_POLYGON |
 469                _NEW_PROGRAM |
 470                _NEW_SCISSOR,
 471       .brw   = BRW_NEW_CONTEXT |
 472                BRW_NEW_FRAGMENT_PROGRAM |
 473                BRW_NEW_FS_PROG_DATA |
 474                BRW_NEW_GEOMETRY_PROGRAM |
 475                BRW_NEW_PRIMITIVE |
 476                BRW_NEW_VUE_MAP_GEOM_OUT,
 477    },
 478    .emit = upload_sf_state,
 479 };