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    if (brw->geometry_program) {
 137       /* BRW_NEW_GEOMETRY_PROGRAM */
 138       return brw->geometry_program->OutputType == GL_POINTS;
 139    } else {
 140       /* BRW_NEW_PRIMITIVE */
 141       return brw->primitive == _3DPRIM_POINTLIST;
 142    }
 143 }
 144
 145
 146 /**
 147  * Create the mapping from the FS inputs we produce to the previous pipeline
 148  * stage (GS or VS) outputs they source from.
 149  */
 150 void
 151 calculate_attr_overrides(const struct brw_context *brw,
 152                          uint16_t *attr_overrides,
 153                          uint32_t *point_sprite_enables,
 154                          uint32_t *flat_enables,
 155                          uint32_t *urb_entry_read_length)
 156 {
 157    const int urb_entry_read_offset = BRW_SF_URB_ENTRY_READ_OFFSET;
 158    uint32_t max_source_attr = 0;
 159
 160    *point_sprite_enables = 0;
 161    *flat_enables = 0;
 162
 163    /* _NEW_LIGHT */
 164    bool shade_model_flat = brw->ctx.Light.ShadeModel == GL_FLAT;
 165
 166    /* From the Ivybridge PRM, Vol 2 Part 1, 3DSTATE_SBE,
 167     * description of dw10 Point Sprite Texture Coordinate Enable:
 168     *
 169     * "This field must be programmed to zero when non-point primitives
 170     * are rendered."
 171     *
 172     * The SandyBridge PRM doesn't explicitly say that point sprite enables
 173     * must be programmed to zero when rendering non-point primitives, but
 174     * the IvyBridge PRM does, and if we don't, we get garbage.
 175     *
 176     * This is not required on Haswell, as the hardware ignores this state
 177     * when drawing non-points -- although we do still need to be careful to
 178     * correctly set the attr overrides.
 179     */
 180    /* BRW_NEW_PRIMITIVE | BRW_NEW_GEOMETRY_PROGRAM */
 181    bool drawing_points = is_drawing_points(brw);
 182
 183    /* Initialize all the attr_overrides to 0.  In the loop below we'll modify
 184     * just the ones that correspond to inputs used by the fs.
 185     */
 186    memset(attr_overrides, 0, 16*sizeof(*attr_overrides));
 187
 188    for (int attr = 0; attr < VARYING_SLOT_MAX; attr++) {
 189       /* BRW_NEW_FRAGMENT_PROGRAM */
 190       enum glsl_interp_qualifier interp_qualifier =
 191          brw->fragment_program->InterpQualifier[attr];
 192       bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1;
 193       /* BRW_NEW_FS_PROG_DATA */
 194       int input_index = brw->wm.prog_data->urb_setup[attr];
 195
 196       if (input_index < 0)
 197          continue;
 198
 199       /* _NEW_POINT */
 200       bool point_sprite = false;
 201       if (drawing_points) {
 202          if (brw->ctx.Point.PointSprite &&
 203              (attr >= VARYING_SLOT_TEX0 && attr <= VARYING_SLOT_TEX7) &&
 204              brw->ctx.Point.CoordReplace[attr - VARYING_SLOT_TEX0]) {
 205             point_sprite = true;
 206          }
 207
 208          if (attr == VARYING_SLOT_PNTC)
 209             point_sprite = true;
 210
 211          if (point_sprite)
 212             *point_sprite_enables |= (1 << input_index);
 213       }
 214
 215       /* flat shading */
 216       if (interp_qualifier == INTERP_QUALIFIER_FLAT ||
 217           (shade_model_flat && is_gl_Color &&
 218            interp_qualifier == INTERP_QUALIFIER_NONE))
 219          *flat_enables |= (1 << input_index);
 220
 221       /* BRW_NEW_VUE_MAP_GEOM_OUT | _NEW_LIGHT | _NEW_PROGRAM */
 222       uint16_t attr_override = point_sprite ? 0 :
 223          get_attr_override(&brw->vue_map_geom_out,
 224                            urb_entry_read_offset, attr,
 225                            brw->ctx.VertexProgram._TwoSideEnabled,
 226                            &max_source_attr);
 227
 228       /* The hardware can only do the overrides on 16 overrides at a
 229        * time, and the other up to 16 have to be lined up so that the
 230        * input index = the output index.  We'll need to do some
 231        * tweaking to make sure that's the case.
 232        */
 233       if (input_index < 16)
 234          attr_overrides[input_index] = attr_override;
 235       else
 236          assert(attr_override == input_index);
 237    }
 238
 239    /* From the Sandy Bridge PRM, Volume 2, Part 1, documentation for
 240     * 3DSTATE_SF DWord 1 bits 15:11, "Vertex URB Entry Read Length":
 241     *
 242     * "This field should be set to the minimum length required to read the
 243     *  maximum source attribute.  The maximum source attribute is indicated
 244     *  by the maximum value of the enabled Attribute # Source Attribute if
 245     *  Attribute Swizzle Enable is set, Number of Output Attributes-1 if
 246     *  enable is not set.
 247     *  read_length = ceiling((max_source_attr + 1) / 2)
 248     *
 249     *  [errata] Corruption/Hang possible if length programmed larger than
 250     *  recommended"
 251     *
 252     * Similar text exists for Ivy Bridge.
 253     */
 254    *urb_entry_read_length = ALIGN(max_source_attr + 1, 2) / 2;
 255 }
 256
 257
 258 static void
 259 upload_sf_state(struct brw_context *brw)
 260 {
 261    struct gl_context *ctx = &brw->ctx;
 262    /* BRW_NEW_FS_PROG_DATA */
 263    uint32_t num_outputs = brw->wm.prog_data->num_varying_inputs;
 264    uint32_t dw1, dw2, dw3, dw4;
 265    uint32_t point_sprite_enables;
 266    uint32_t flat_enables;
 267    int i;
 268    /* _NEW_BUFFER */
 269    bool render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 270    bool multisampled_fbo = ctx->DrawBuffer->Visual.samples > 1;
 271
 272    const int urb_entry_read_offset = BRW_SF_URB_ENTRY_READ_OFFSET;
 273    float point_size;
 274    uint16_t attr_overrides[16];
 275    uint32_t point_sprite_origin;
 276
 277    dw1 = GEN6_SF_SWIZZLE_ENABLE | num_outputs << GEN6_SF_NUM_OUTPUTS_SHIFT;
 278    dw2 = GEN6_SF_STATISTICS_ENABLE;
 279
 280    if (brw->sf.viewport_transform_enable)
 281        dw2 |= GEN6_SF_VIEWPORT_TRANSFORM_ENABLE;
 282
 283    dw3 = 0;
 284    dw4 = 0;
 285
 286    /* _NEW_POLYGON */
 287    if ((ctx->Polygon.FrontFace == GL_CCW) ^ render_to_fbo)
 288       dw2 |= GEN6_SF_WINDING_CCW;
 289
 290    if (ctx->Polygon.OffsetFill)
 291        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID;
 292
 293    if (ctx->Polygon.OffsetLine)
 294        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME;
 295
 296    if (ctx->Polygon.OffsetPoint)
 297        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT;
 298
 299    switch (ctx->Polygon.FrontMode) {
 300    case GL_FILL:
 301        dw2 |= GEN6_SF_FRONT_SOLID;
 302        break;
 303
 304    case GL_LINE:
 305        dw2 |= GEN6_SF_FRONT_WIREFRAME;
 306        break;
 307
 308    case GL_POINT:
 309        dw2 |= GEN6_SF_FRONT_POINT;
 310        break;
 311
 312    default:
 313        unreachable("not reached");
 314    }
 315
 316    switch (ctx->Polygon.BackMode) {
 317    case GL_FILL:
 318        dw2 |= GEN6_SF_BACK_SOLID;
 319        break;
 320
 321    case GL_LINE:
 322        dw2 |= GEN6_SF_BACK_WIREFRAME;
 323        break;
 324
 325    case GL_POINT:
 326        dw2 |= GEN6_SF_BACK_POINT;
 327        break;
 328
 329    default:
 330        unreachable("not reached");
 331    }
 332
 333    /* _NEW_SCISSOR */
 334    if (ctx->Scissor.EnableFlags)
 335       dw3 |= GEN6_SF_SCISSOR_ENABLE;
 336
 337    /* _NEW_POLYGON */
 338    if (ctx->Polygon.CullFlag) {
 339       switch (ctx->Polygon.CullFaceMode) {
 340       case GL_FRONT:
 341          dw3 |= GEN6_SF_CULL_FRONT;
 342          break;
 343       case GL_BACK:
 344          dw3 |= GEN6_SF_CULL_BACK;
 345          break;
 346       case GL_FRONT_AND_BACK:
 347          dw3 |= GEN6_SF_CULL_BOTH;
 348          break;
 349       default:
 350          unreachable("not reached");
 351       }
 352    } else {
 353       dw3 |= GEN6_SF_CULL_NONE;
 354    }
 355
 356    /* _NEW_LINE */
 357    {
 358       uint32_t line_width_u3_7 =
 359          U_FIXED(CLAMP(ctx->Line.Width, 0.0, ctx->Const.MaxLineWidth), 7);
 360       /* TODO: line width of 0 is not allowed when MSAA enabled */
 361       if (line_width_u3_7 == 0)
 362          line_width_u3_7 = 1;
 363       dw3 |= line_width_u3_7 << GEN6_SF_LINE_WIDTH_SHIFT;
 364    }
 365    if (ctx->Line.SmoothFlag) {
 366       dw3 |= GEN6_SF_LINE_AA_ENABLE;
 367       dw3 |= GEN6_SF_LINE_AA_MODE_TRUE;
 368       dw3 |= GEN6_SF_LINE_END_CAP_WIDTH_1_0;
 369    }
 370    /* _NEW_MULTISAMPLE */
 371    if (multisampled_fbo && ctx->Multisample.Enabled)
 372       dw3 |= GEN6_SF_MSRAST_ON_PATTERN;
 373
 374    /* _NEW_PROGRAM | _NEW_POINT */
 375    if (!(ctx->VertexProgram.PointSizeEnabled ||
 376          ctx->Point._Attenuated))
 377       dw4 |= GEN6_SF_USE_STATE_POINT_WIDTH;
 378
 379    /* Clamp to ARB_point_parameters user limits */
 380    point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
 381
 382    /* Clamp to the hardware limits and convert to fixed point */
 383    dw4 |= U_FIXED(CLAMP(point_size, 0.125, 255.875), 3);
 384
 385    /*
 386     * Window coordinates in an FBO are inverted, which means point
 387     * sprite origin must be inverted, too.
 388     */
 389    if ((ctx->Point.SpriteOrigin == GL_LOWER_LEFT) != render_to_fbo) {
 390       point_sprite_origin = GEN6_SF_POINT_SPRITE_LOWERLEFT;
 391    } else {
 392       point_sprite_origin = GEN6_SF_POINT_SPRITE_UPPERLEFT;
 393    }
 394    dw1 |= point_sprite_origin;
 395
 396    /* _NEW_LIGHT */
 397    if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
 398       dw4 |=
 399          (2 << GEN6_SF_TRI_PROVOKE_SHIFT) |
 400          (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT) |
 401          (1 << GEN6_SF_LINE_PROVOKE_SHIFT);
 402    } else {
 403       dw4 |=
 404          (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT);
 405    }
 406
 407    /* BRW_NEW_VUE_MAP_GEOM_OUT | BRW_NEW_FRAGMENT_PROGRAM |
 408     * _NEW_POINT | _NEW_LIGHT | _NEW_PROGRAM | BRW_NEW_FS_PROG_DATA
 409     */
 410    uint32_t urb_entry_read_length;
 411    calculate_attr_overrides(brw, attr_overrides, &point_sprite_enables,
 412                             &flat_enables, &urb_entry_read_length);
 413    dw1 |= (urb_entry_read_length << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
 414            urb_entry_read_offset << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT);
 415
 416    BEGIN_BATCH(20);
 417    OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
 418    OUT_BATCH(dw1);
 419    OUT_BATCH(dw2);
 420    OUT_BATCH(dw3);
 421    OUT_BATCH(dw4);
 422    OUT_BATCH_F(ctx->Polygon.OffsetUnits * 2); /* constant.  copied from gen4 */
 423    OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
 424    OUT_BATCH_F(0.0); /* XXX: global depth offset clamp */
 425    for (i = 0; i < 8; i++) {
 426       OUT_BATCH(attr_overrides[i * 2] | attr_overrides[i * 2 + 1] << 16);
 427    }
 428    OUT_BATCH(point_sprite_enables); /* dw16 */
 429    OUT_BATCH(flat_enables);
 430    OUT_BATCH(0); /* wrapshortest enables 0-7 */
 431    OUT_BATCH(0); /* wrapshortest enables 8-15 */
 432    ADVANCE_BATCH();
 433 }
 434
 435 const struct brw_tracked_state gen6_sf_state = {
 436    .dirty = {
 437       .mesa  = _NEW_BUFFERS |
 438                _NEW_LIGHT |
 439                _NEW_LINE |
 440                _NEW_MULTISAMPLE |
 441                _NEW_POINT |
 442                _NEW_POLYGON |
 443                _NEW_PROGRAM |
 444                _NEW_SCISSOR,
 445       .brw   = BRW_NEW_CONTEXT |
 446                BRW_NEW_FRAGMENT_PROGRAM |
 447                BRW_NEW_GEOMETRY_PROGRAM |
 448                BRW_NEW_PRIMITIVE |
 449                BRW_NEW_VUE_MAP_GEOM_OUT,
 450       .cache = BRW_NEW_FS_PROG_DATA
 451    },
 452    .emit = upload_sf_state,
 453 };