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 "intel_batchbuffer.h"
  34
  35 /**
  36  * Determine the appropriate attribute override value to store into the
  37  * 3DSTATE_SF structure for a given fragment shader attribute.  The attribute
  38  * override value contains two pieces of information: the location of the
  39  * attribute in the VUE (relative to urb_entry_read_offset, see below), and a
  40  * flag indicating whether to "swizzle" the attribute based on the direction
  41  * the triangle is facing.
  42  *
  43  * If an attribute is "swizzled", then the given VUE location is used for
  44  * front-facing triangles, and the VUE location that immediately follows is
  45  * used for back-facing triangles.  We use this to implement the mapping from
  46  * gl_FrontColor/gl_BackColor to gl_Color.
  47  *
  48  * urb_entry_read_offset is the offset into the VUE at which the SF unit is
  49  * being instructed to begin reading attribute data.  It can be set to a
  50  * nonzero value to prevent the SF unit from wasting time reading elements of
  51  * the VUE that are not needed by the fragment shader.  It is measured in
  52  * 256-bit increments.
  53  */
  54 uint32_t
  55 get_attr_override(struct brw_vue_map *vue_map, int urb_entry_read_offset,
  56                   int fs_attr, bool two_side_color)
  57 {
  58    int attr_override, slot;
  59    int vs_attr = _mesa_frag_attrib_to_vert_result(fs_attr);
  60    if (vs_attr < 0 || vs_attr == VERT_RESULT_HPOS) {
  61       /* These attributes will be overwritten by the fragment shader's
  62        * interpolation code (see emit_interp() in brw_wm_fp.c), so just let
  63        * them reference the first available attribute.
  64        */
  65       return 0;
  66    }
  67
  68    /* Find the VUE slot for this attribute. */
  69    slot = vue_map->vert_result_to_slot[vs_attr];
  70
  71    /* If there was only a back color written but not front, use back
  72     * as the color instead of undefined
  73     */
  74    if (slot == -1 && vs_attr == VERT_RESULT_COL0)
  75       slot = vue_map->vert_result_to_slot[VERT_RESULT_BFC0];
  76    if (slot == -1 && vs_attr == VERT_RESULT_COL1)
  77       slot = vue_map->vert_result_to_slot[VERT_RESULT_BFC1];
  78
  79    if (slot == -1) {
  80       /* This attribute does not exist in the VUE--that means that the vertex
  81        * shader did not write to it.  Behavior is undefined in this case, so
  82        * just reference the first available attribute.
  83        */
  84       return 0;
  85    }
  86
  87    /* Compute the location of the attribute relative to urb_entry_read_offset.
  88     * Each increment of urb_entry_read_offset represents a 256-bit value, so
  89     * it counts for two 128-bit VUE slots.
  90     */
  91    attr_override = slot - 2 * urb_entry_read_offset;
  92    assert (attr_override >= 0 && attr_override < 32);
  93
  94    /* If we are doing two-sided color, and the VUE slot following this one
  95     * represents a back-facing color, then we need to instruct the SF unit to
  96     * do back-facing swizzling.
  97     */
  98    if (two_side_color) {
  99       if (vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL0 &&
 100           vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC0)
 101          attr_override |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
 102       else if (vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL1 &&
 103                vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC1)
 104          attr_override |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
 105    }
 106
 107    return attr_override;
 108 }
 109
 110 static void
 111 upload_sf_state(struct brw_context *brw)
 112 {
 113    struct intel_context *intel = &brw->intel;
 114    struct gl_context *ctx = &intel->ctx;
 115    struct brw_vue_map vue_map;
 116    uint32_t urb_entry_read_length;
 117    /* CACHE_NEW_VS_PROG */
 118    GLbitfield64 vs_outputs_written = brw->vs.prog_data->outputs_written;
 119    /* BRW_NEW_FRAGMENT_PROGRAM */
 120    uint32_t num_outputs = _mesa_bitcount_64(brw->fragment_program->Base.InputsRead);
 121    /* _NEW_LIGHT */
 122    bool shade_model_flat = ctx->Light.ShadeModel == GL_FLAT;
 123    uint32_t dw1, dw2, dw3, dw4, dw16, dw17;
 124    int i;
 125    /* _NEW_BUFFER */
 126    bool render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
 127    int attr = 0, input_index = 0;
 128    int urb_entry_read_offset = 1;
 129    float point_size;
 130    uint16_t attr_overrides[FRAG_ATTRIB_MAX];
 131    bool userclip_active;
 132
 133    /* _NEW_TRANSFORM */
 134    userclip_active = (ctx->Transform.ClipPlanesEnabled != 0);
 135
 136    brw_compute_vue_map(&vue_map, intel, userclip_active, vs_outputs_written);
 137    urb_entry_read_length = (vue_map.num_slots + 1)/2 - urb_entry_read_offset;
 138    if (urb_entry_read_length == 0) {
 139       /* Setting the URB entry read length to 0 causes undefined behavior, so
 140        * if we have no URB data to read, set it to 1.
 141        */
 142       urb_entry_read_length = 1;
 143    }
 144
 145    dw1 =
 146       GEN6_SF_SWIZZLE_ENABLE |
 147       num_outputs << GEN6_SF_NUM_OUTPUTS_SHIFT |
 148       urb_entry_read_length << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
 149       urb_entry_read_offset << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT;
 150
 151    dw2 = GEN6_SF_STATISTICS_ENABLE;
 152
 153    /* Enable viewport transform only if no HiZ operation is progress
 154     *
 155     * From page 11 of the SandyBridge PRM, Volume 2, Part 1, Section 1.3, "3D
 156     * Primitives Overview":
 157     *     RECTLIST: Viewport Mapping must be DISABLED (as is typical with the
 158     *     use of screen- space coordinates).
 159     */
 160    if (!brw->hiz.op)
 161       dw2 |= GEN6_SF_VIEWPORT_TRANSFORM_ENABLE;
 162
 163    dw3 = 0;
 164    dw4 = 0;
 165    dw16 = 0;
 166    dw17 = 0;
 167
 168    /* _NEW_POLYGON */
 169    if ((ctx->Polygon.FrontFace == GL_CCW) ^ render_to_fbo)
 170       dw2 |= GEN6_SF_WINDING_CCW;
 171
 172    if (ctx->Polygon.OffsetFill)
 173        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID;
 174
 175    if (ctx->Polygon.OffsetLine)
 176        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME;
 177
 178    if (ctx->Polygon.OffsetPoint)
 179        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT;
 180
 181    switch (ctx->Polygon.FrontMode) {
 182    case GL_FILL:
 183        dw2 |= GEN6_SF_FRONT_SOLID;
 184        break;
 185
 186    case GL_LINE:
 187        dw2 |= GEN6_SF_FRONT_WIREFRAME;
 188        break;
 189
 190    case GL_POINT:
 191        dw2 |= GEN6_SF_FRONT_POINT;
 192        break;
 193
 194    default:
 195        assert(0);
 196        break;
 197    }
 198
 199    switch (ctx->Polygon.BackMode) {
 200    case GL_FILL:
 201        dw2 |= GEN6_SF_BACK_SOLID;
 202        break;
 203
 204    case GL_LINE:
 205        dw2 |= GEN6_SF_BACK_WIREFRAME;
 206        break;
 207
 208    case GL_POINT:
 209        dw2 |= GEN6_SF_BACK_POINT;
 210        break;
 211
 212    default:
 213        assert(0);
 214        break;
 215    }
 216
 217    /* _NEW_SCISSOR */
 218    if (ctx->Scissor.Enabled)
 219       dw3 |= GEN6_SF_SCISSOR_ENABLE;
 220
 221    /* _NEW_POLYGON */
 222    if (ctx->Polygon.CullFlag) {
 223       switch (ctx->Polygon.CullFaceMode) {
 224       case GL_FRONT:
 225          dw3 |= GEN6_SF_CULL_FRONT;
 226          break;
 227       case GL_BACK:
 228          dw3 |= GEN6_SF_CULL_BACK;
 229          break;
 230       case GL_FRONT_AND_BACK:
 231          dw3 |= GEN6_SF_CULL_BOTH;
 232          break;
 233       default:
 234          assert(0);
 235          break;
 236       }
 237    } else {
 238       dw3 |= GEN6_SF_CULL_NONE;
 239    }
 240
 241    /* _NEW_LINE */
 242    dw3 |= U_FIXED(CLAMP(ctx->Line.Width, 0.0, 7.99), 7) <<
 243       GEN6_SF_LINE_WIDTH_SHIFT;
 244    if (ctx->Line.SmoothFlag) {
 245       dw3 |= GEN6_SF_LINE_AA_ENABLE;
 246       dw3 |= GEN6_SF_LINE_AA_MODE_TRUE;
 247       dw3 |= GEN6_SF_LINE_END_CAP_WIDTH_1_0;
 248    }
 249
 250    /* _NEW_POINT */
 251    if (!(ctx->VertexProgram.PointSizeEnabled ||
 252          ctx->Point._Attenuated))
 253       dw4 |= GEN6_SF_USE_STATE_POINT_WIDTH;
 254
 255    /* Clamp to ARB_point_parameters user limits */
 256    point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
 257
 258    /* Clamp to the hardware limits and convert to fixed point */
 259    dw4 |= U_FIXED(CLAMP(point_size, 0.125, 255.875), 3);
 260
 261    if (ctx->Point.SpriteOrigin == GL_LOWER_LEFT)
 262       dw1 |= GEN6_SF_POINT_SPRITE_LOWERLEFT;
 263
 264    /* _NEW_LIGHT */
 265    if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
 266       dw4 |=
 267          (2 << GEN6_SF_TRI_PROVOKE_SHIFT) |
 268          (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT) |
 269          (1 << GEN6_SF_LINE_PROVOKE_SHIFT);
 270    } else {
 271       dw4 |=
 272          (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT);
 273    }
 274
 275    /* Create the mapping from the FS inputs we produce to the VS outputs
 276     * they source from.
 277     */
 278    for (; attr < FRAG_ATTRIB_MAX; attr++) {
 279       enum glsl_interp_qualifier interp_qualifier =
 280          brw->fragment_program->InterpQualifier[attr];
 281       bool is_gl_Color = attr == FRAG_ATTRIB_COL0 || attr == FRAG_ATTRIB_COL1;
 282
 283       if (!(brw->fragment_program->Base.InputsRead & BITFIELD64_BIT(attr)))
 284          continue;
 285
 286       /* _NEW_POINT */
 287       if (ctx->Point.PointSprite &&
 288           (attr >= FRAG_ATTRIB_TEX0 && attr <= FRAG_ATTRIB_TEX7) &&
 289           ctx->Point.CoordReplace[attr - FRAG_ATTRIB_TEX0]) {
 290          dw16 |= (1 << input_index);
 291       }
 292
 293       if (attr == FRAG_ATTRIB_PNTC)
 294          dw16 |= (1 << input_index);
 295
 296       /* flat shading */
 297       if (interp_qualifier == INTERP_QUALIFIER_FLAT ||
 298           (shade_model_flat && is_gl_Color &&
 299            interp_qualifier == INTERP_QUALIFIER_NONE))
 300          dw17 |= (1 << input_index);
 301
 302       /* The hardware can only do the overrides on 16 overrides at a
 303        * time, and the other up to 16 have to be lined up so that the
 304        * input index = the output index.  We'll need to do some
 305        * tweaking to make sure that's the case.
 306        */
 307       assert(input_index < 16 || attr == input_index);
 308
 309       /* _NEW_LIGHT | _NEW_PROGRAM */
 310       attr_overrides[input_index++] =
 311          get_attr_override(&vue_map, urb_entry_read_offset, attr,
 312                            ctx->VertexProgram._TwoSideEnabled);
 313    }
 314
 315    for (; input_index < FRAG_ATTRIB_MAX; input_index++)
 316       attr_overrides[input_index] = 0;
 317
 318    BEGIN_BATCH(20);
 319    OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
 320    OUT_BATCH(dw1);
 321    OUT_BATCH(dw2);
 322    OUT_BATCH(dw3);
 323    OUT_BATCH(dw4);
 324    OUT_BATCH_F(ctx->Polygon.OffsetUnits * 2); /* constant.  copied from gen4 */
 325    OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
 326    OUT_BATCH_F(0.0); /* XXX: global depth offset clamp */
 327    for (i = 0; i < 8; i++) {
 328       OUT_BATCH(attr_overrides[i * 2] | attr_overrides[i * 2 + 1] << 16);
 329    }
 330    OUT_BATCH(dw16); /* point sprite texcoord bitmask */
 331    OUT_BATCH(dw17); /* constant interp bitmask */
 332    OUT_BATCH(0); /* wrapshortest enables 0-7 */
 333    OUT_BATCH(0); /* wrapshortest enables 8-15 */
 334    ADVANCE_BATCH();
 335 }
 336
 337 const struct brw_tracked_state gen6_sf_state = {
 338    .dirty = {
 339       .mesa  = (_NEW_LIGHT |
 340                 _NEW_PROGRAM |
 341                 _NEW_POLYGON |
 342                 _NEW_LINE |
 343                 _NEW_SCISSOR |
 344                 _NEW_BUFFERS |
 345                 _NEW_POINT |
 346                 _NEW_TRANSFORM),
 347       .brw   = (BRW_NEW_CONTEXT |
 348                 BRW_NEW_FRAGMENT_PROGRAM |
 349                 BRW_NEW_HIZ),
 350       .cache = CACHE_NEW_VS_PROG
 351    },
 352    .emit = upload_sf_state,
 353 };