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    if (slot == -1) {
  71       /* This attribute does not exist in the VUE--that means that the vertex
  72        * shader did not write to it.  Behavior is undefined in this case, so
  73        * just reference the first available attribute.
  74        */
  75       return 0;
  76    }
  77
  78    /* Compute the location of the attribute relative to urb_entry_read_offset.
  79     * Each increment of urb_entry_read_offset represents a 256-bit value, so
  80     * it counts for two 128-bit VUE slots.
  81     */
  82    attr_override = slot - 2 * urb_entry_read_offset;
  83    assert (attr_override >= 0 && attr_override < 32);
  84
  85    /* If we are doing two-sided color, and the VUE slot following this one
  86     * represents a back-facing color, then we need to instruct the SF unit to
  87     * do back-facing swizzling.
  88     */
  89    if (two_side_color) {
  90       if (vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL0 &&
  91           vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC0)
  92          attr_override |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
  93       else if (vue_map->slot_to_vert_result[slot] == VERT_RESULT_COL1 &&
  94                vue_map->slot_to_vert_result[slot+1] == VERT_RESULT_BFC1)
  95          attr_override |= (ATTRIBUTE_SWIZZLE_INPUTATTR_FACING << ATTRIBUTE_SWIZZLE_SHIFT);
  96    }
  97
  98    return attr_override;
  99 }
 100
 101 static void
 102 upload_sf_state(struct brw_context *brw)
 103 {
 104    struct intel_context *intel = &brw->intel;
 105    struct gl_context *ctx = &intel->ctx;
 106    struct brw_vue_map vue_map;
 107    uint32_t urb_entry_read_length;
 108    /* CACHE_NEW_VS_PROG */
 109    GLbitfield64 vs_outputs_written = brw->vs.prog_data->outputs_written;
 110    /* BRW_NEW_FRAGMENT_PROGRAM */
 111    uint32_t num_outputs = brw_count_bits(brw->fragment_program->Base.InputsRead);
 112    uint32_t dw1, dw2, dw3, dw4, dw16, dw17;
 113    int i;
 114    /* _NEW_BUFFER */
 115    GLboolean render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0;
 116    int attr = 0, input_index = 0;
 117    int urb_entry_read_offset;
 118    float point_size;
 119    uint16_t attr_overrides[FRAG_ATTRIB_MAX];
 120    int nr_userclip;
 121
 122    /* _NEW_TRANSFORM */
 123    if (ctx->Transform.ClipPlanesEnabled)
 124       urb_entry_read_offset = 2;
 125    else
 126       urb_entry_read_offset = 1;
 127    nr_userclip = brw_count_bits(ctx->Transform.ClipPlanesEnabled);
 128
 129    brw_compute_vue_map(&vue_map, intel, nr_userclip, vs_outputs_written);
 130    urb_entry_read_length = (vue_map.num_slots + 1)/2 - urb_entry_read_offset;
 131    if (urb_entry_read_length == 0) {
 132       /* Setting the URB entry read length to 0 causes undefined behavior, so
 133        * if we have no URB data to read, set it to 1.
 134        */
 135       urb_entry_read_length = 1;
 136    }
 137
 138    dw1 =
 139       GEN6_SF_SWIZZLE_ENABLE |
 140       num_outputs << GEN6_SF_NUM_OUTPUTS_SHIFT |
 141       urb_entry_read_length << GEN6_SF_URB_ENTRY_READ_LENGTH_SHIFT |
 142       urb_entry_read_offset << GEN6_SF_URB_ENTRY_READ_OFFSET_SHIFT;
 143    dw2 = GEN6_SF_VIEWPORT_TRANSFORM_ENABLE |
 144       GEN6_SF_STATISTICS_ENABLE;
 145    dw3 = 0;
 146    dw4 = 0;
 147    dw16 = 0;
 148    dw17 = 0;
 149
 150    /* _NEW_POLYGON */
 151    if ((ctx->Polygon.FrontFace == GL_CCW) ^ render_to_fbo)
 152       dw2 |= GEN6_SF_WINDING_CCW;
 153
 154    if (ctx->Polygon.OffsetFill)
 155        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_SOLID;
 156
 157    if (ctx->Polygon.OffsetLine)
 158        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_WIREFRAME;
 159
 160    if (ctx->Polygon.OffsetPoint)
 161        dw2 |= GEN6_SF_GLOBAL_DEPTH_OFFSET_POINT;
 162
 163    switch (ctx->Polygon.FrontMode) {
 164    case GL_FILL:
 165        dw2 |= GEN6_SF_FRONT_SOLID;
 166        break;
 167
 168    case GL_LINE:
 169        dw2 |= GEN6_SF_FRONT_WIREFRAME;
 170        break;
 171
 172    case GL_POINT:
 173        dw2 |= GEN6_SF_FRONT_POINT;
 174        break;
 175
 176    default:
 177        assert(0);
 178        break;
 179    }
 180
 181    switch (ctx->Polygon.BackMode) {
 182    case GL_FILL:
 183        dw2 |= GEN6_SF_BACK_SOLID;
 184        break;
 185
 186    case GL_LINE:
 187        dw2 |= GEN6_SF_BACK_WIREFRAME;
 188        break;
 189
 190    case GL_POINT:
 191        dw2 |= GEN6_SF_BACK_POINT;
 192        break;
 193
 194    default:
 195        assert(0);
 196        break;
 197    }
 198
 199    /* _NEW_SCISSOR */
 200    if (ctx->Scissor.Enabled)
 201       dw3 |= GEN6_SF_SCISSOR_ENABLE;
 202
 203    /* _NEW_POLYGON */
 204    if (ctx->Polygon.CullFlag) {
 205       switch (ctx->Polygon.CullFaceMode) {
 206       case GL_FRONT:
 207          dw3 |= GEN6_SF_CULL_FRONT;
 208          break;
 209       case GL_BACK:
 210          dw3 |= GEN6_SF_CULL_BACK;
 211          break;
 212       case GL_FRONT_AND_BACK:
 213          dw3 |= GEN6_SF_CULL_BOTH;
 214          break;
 215       default:
 216          assert(0);
 217          break;
 218       }
 219    } else {
 220       dw3 |= GEN6_SF_CULL_NONE;
 221    }
 222
 223    /* _NEW_LINE */
 224    dw3 |= U_FIXED(CLAMP(ctx->Line.Width, 0.0, 7.99), 7) <<
 225       GEN6_SF_LINE_WIDTH_SHIFT;
 226    if (ctx->Line.SmoothFlag) {
 227       dw3 |= GEN6_SF_LINE_AA_ENABLE;
 228       dw3 |= GEN6_SF_LINE_AA_MODE_TRUE;
 229       dw3 |= GEN6_SF_LINE_END_CAP_WIDTH_1_0;
 230    }
 231
 232    /* _NEW_POINT */
 233    if (!(ctx->VertexProgram.PointSizeEnabled ||
 234          ctx->Point._Attenuated))
 235       dw4 |= GEN6_SF_USE_STATE_POINT_WIDTH;
 236
 237    /* Clamp to ARB_point_parameters user limits */
 238    point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize);
 239
 240    /* Clamp to the hardware limits and convert to fixed point */
 241    dw4 |= U_FIXED(CLAMP(point_size, 0.125, 255.875), 3);
 242
 243    if (ctx->Point.SpriteOrigin == GL_LOWER_LEFT)
 244       dw1 |= GEN6_SF_POINT_SPRITE_LOWERLEFT;
 245
 246    /* _NEW_LIGHT */
 247    if (ctx->Light.ProvokingVertex != GL_FIRST_VERTEX_CONVENTION) {
 248       dw4 |=
 249          (2 << GEN6_SF_TRI_PROVOKE_SHIFT) |
 250          (2 << GEN6_SF_TRIFAN_PROVOKE_SHIFT) |
 251          (1 << GEN6_SF_LINE_PROVOKE_SHIFT);
 252    } else {
 253       dw4 |=
 254          (1 << GEN6_SF_TRIFAN_PROVOKE_SHIFT);
 255    }
 256
 257    /* flat shading */
 258    if (ctx->Light.ShadeModel == GL_FLAT) {
 259        dw17 |= ((brw->fragment_program->Base.InputsRead & (FRAG_BIT_COL0 | FRAG_BIT_COL1)) >>
 260                 ((brw->fragment_program->Base.InputsRead & FRAG_BIT_WPOS) ? 0 : 1));
 261    }
 262
 263    /* Create the mapping from the FS inputs we produce to the VS outputs
 264     * they source from.
 265     */
 266    for (; attr < FRAG_ATTRIB_MAX; attr++) {
 267       if (!(brw->fragment_program->Base.InputsRead & BITFIELD64_BIT(attr)))
 268          continue;
 269
 270       /* _NEW_POINT */
 271       if (ctx->Point.PointSprite &&
 272           (attr >= FRAG_ATTRIB_TEX0 && attr <= FRAG_ATTRIB_TEX7) &&
 273           ctx->Point.CoordReplace[attr - FRAG_ATTRIB_TEX0]) {
 274          dw16 |= (1 << input_index);
 275       }
 276
 277       if (attr == FRAG_ATTRIB_PNTC)
 278          dw16 |= (1 << input_index);
 279
 280       /* The hardware can only do the overrides on 16 overrides at a
 281        * time, and the other up to 16 have to be lined up so that the
 282        * input index = the output index.  We'll need to do some
 283        * tweaking to make sure that's the case.
 284        */
 285       assert(input_index < 16 || attr == input_index);
 286
 287       /* _NEW_LIGHT | _NEW_PROGRAM */
 288       attr_overrides[input_index++] =
 289          get_attr_override(&vue_map, urb_entry_read_offset, attr,
 290                            ctx->VertexProgram._TwoSideEnabled);
 291    }
 292
 293    for (; input_index < FRAG_ATTRIB_MAX; input_index++)
 294       attr_overrides[input_index] = 0;
 295
 296    BEGIN_BATCH(20);
 297    OUT_BATCH(_3DSTATE_SF << 16 | (20 - 2));
 298    OUT_BATCH(dw1);
 299    OUT_BATCH(dw2);
 300    OUT_BATCH(dw3);
 301    OUT_BATCH(dw4);
 302    OUT_BATCH_F(ctx->Polygon.OffsetUnits * 2); /* constant.  copied from gen4 */
 303    OUT_BATCH_F(ctx->Polygon.OffsetFactor); /* scale */
 304    OUT_BATCH_F(0.0); /* XXX: global depth offset clamp */
 305    for (i = 0; i < 8; i++) {
 306       OUT_BATCH(attr_overrides[i * 2] | attr_overrides[i * 2 + 1] << 16);
 307    }
 308    OUT_BATCH(dw16); /* point sprite texcoord bitmask */
 309    OUT_BATCH(dw17); /* constant interp bitmask */
 310    OUT_BATCH(0); /* wrapshortest enables 0-7 */
 311    OUT_BATCH(0); /* wrapshortest enables 8-15 */
 312    ADVANCE_BATCH();
 313 }
 314
 315 const struct brw_tracked_state gen6_sf_state = {
 316    .dirty = {
 317       .mesa  = (_NEW_LIGHT |
 318                 _NEW_PROGRAM |
 319                 _NEW_POLYGON |
 320                 _NEW_LINE |
 321                 _NEW_SCISSOR |
 322                 _NEW_BUFFERS |
 323                 _NEW_POINT |
 324                 _NEW_TRANSFORM),
 325       .brw   = (BRW_NEW_CONTEXT |
 326                 BRW_NEW_FRAGMENT_PROGRAM),
 327       .cache = CACHE_NEW_VS_PROG
 328    },
 329    .emit = upload_sf_state,
 330 };