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

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
   4  develop this 3D driver.
   5
   6  Permission is hereby granted, free of charge, to any person obtaining
   7  a copy of this software and associated documentation files (the
   8  "Software"), to deal in the Software without restriction, including
   9  without limitation the rights to use, copy, modify, merge, publish,
  10  distribute, sublicense, and/or sell copies of the Software, and to
  11  permit persons to whom the Software is furnished to do so, subject to
  12  the following conditions:
  13
  14  The above copyright notice and this permission notice (including the
  15  next paragraph) shall be included in all copies or substantial
  16  portions of the Software.
  17
  18  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25
  26  **********************************************************************/
  27  /*
  28   * Authors:
  29   *   Keith Whitwell <keith@tungstengraphics.com>
  30   */
  31
  32
  33 #include "main/macros.h"
  34 #include "brw_context.h"
  35 #include "brw_vs.h"
  36
  37 /* Component is active if it may diverge from [0,0,0,1].  Undef values
  38  * are promoted to [0,0,0,1] for the purposes of this analysis.
  39  */
  40 struct tracker {
  41    bool twoside;
  42    GLubyte active[PROGRAM_OUTPUT+1][MAX_PROGRAM_TEMPS];
  43    GLbitfield64 size_masks[4];  /**< one bit per fragment program input attrib */
  44 };
  45
  46
  47 static void set_active_component( struct tracker *t,
  48                                   GLuint file,
  49                                   GLuint index,
  50                                   GLubyte active )
  51 {
  52    switch (file) {
  53    case PROGRAM_TEMPORARY:
  54    case PROGRAM_INPUT:
  55    case PROGRAM_OUTPUT:
  56       assert(file < PROGRAM_OUTPUT + 1);
  57       assert(index < Elements(t->active[0]));
  58       t->active[file][index] |= active;
  59       break;
  60    default:
  61       break;
  62    }
  63 }
  64
  65 static void set_active( struct tracker *t,
  66                         struct prog_dst_register dst,
  67                         GLuint active )
  68 {
  69    set_active_component( t, dst.File, dst.Index, active & dst.WriteMask );
  70 }
  71
  72
  73 static GLubyte get_active_component( struct tracker *t,
  74                                      GLuint file,
  75                                      GLuint index,
  76                                      GLuint component,
  77                                      GLubyte swz )
  78 {
  79    switch (swz) {
  80    case SWIZZLE_ZERO:
  81       return component < 3 ? 0 : (1<<component);
  82    case SWIZZLE_ONE:
  83       return component == 3 ? 0 : (1<<component);
  84    default:
  85       switch (file) {
  86       case PROGRAM_TEMPORARY:
  87       case PROGRAM_INPUT:
  88       case PROGRAM_OUTPUT:
  89          return t->active[file][index] & (1<<component);
  90       default:
  91          return 1 << component;
  92       }
  93    }
  94 }
  95
  96
  97 static GLubyte get_active( struct tracker *t,
  98                            struct prog_src_register src )
  99 {
 100    GLuint i;
 101    GLubyte active = src.Negate; /* NOTE! */
 102
 103    if (src.RelAddr)
 104       return 0xf;
 105
 106    for (i = 0; i < 4; i++)
 107       active |= get_active_component(t, src.File, src.Index, i,
 108                                      GET_SWZ(src.Swizzle, i));
 109
 110    return active;
 111 }
 112
 113 /**
 114  * Return the size (1,2,3 or 4) of the output/result for VARYING_SLOT_idx.
 115  */
 116 static GLubyte get_output_size( struct tracker *t,
 117                                 GLuint idx )
 118 {
 119    GLubyte active;
 120    assert(idx < VARYING_SLOT_MAX);
 121    active = t->active[PROGRAM_OUTPUT][idx];
 122    if (active & (1<<3)) return 4;
 123    if (active & (1<<2)) return 3;
 124    if (active & (1<<1)) return 2;
 125    if (active & (1<<0)) return 1;
 126    return 0;
 127 }
 128
 129 /* Note the potential copying that occurs in the setup program:
 130  */
 131 static void calc_sizes( struct tracker *t )
 132 {
 133    GLint vertRes;
 134
 135    if (t->twoside) {
 136       t->active[PROGRAM_OUTPUT][VARYING_SLOT_COL0] |=
 137          t->active[PROGRAM_OUTPUT][VARYING_SLOT_BFC0];
 138
 139       t->active[PROGRAM_OUTPUT][VARYING_SLOT_COL1] |=
 140          t->active[PROGRAM_OUTPUT][VARYING_SLOT_BFC1];
 141    }
 142
 143    /* Examine vertex program output sizes to set the size_masks[] info
 144     * which describes the fragment program input sizes.
 145     */
 146    for (vertRes = 0; vertRes < VARYING_SLOT_MAX; vertRes++) {
 147       if (!_mesa_varying_slot_in_fs(vertRes))
 148          continue;
 149
 150       switch (get_output_size(t, vertRes)) {
 151       case 4: t->size_masks[4-1] |= BITFIELD64_BIT(vertRes);
 152       case 3: t->size_masks[3-1] |= BITFIELD64_BIT(vertRes);
 153       case 2: t->size_masks[2-1] |= BITFIELD64_BIT(vertRes);
 154       case 1: t->size_masks[1-1] |= BITFIELD64_BIT(vertRes);
 155          break;
 156       }
 157    }
 158 }
 159
 160 static GLubyte szflag[4+1] = {
 161    0,
 162    0x1,
 163    0x3,
 164    0x7,
 165    0xf
 166 };
 167
 168 /* Pull a size out of the packed array:
 169  */
 170 static GLuint get_input_size(struct brw_context *brw,
 171                              GLuint attr)
 172 {
 173    GLuint sizes_dword = brw->vb.info.sizes[attr/16];
 174    GLuint sizes_bits = (sizes_dword>>((attr%16)*2)) & 0x3;
 175    return sizes_bits + 1;
 176 /*    return brw->vb.inputs[attr].glarray->Size; */
 177 }
 178
 179 /* Calculate sizes of vertex program outputs.  Size is the largest
 180  * component index which might vary from [0,0,0,1]
 181  */
 182 static void calc_wm_input_sizes( struct brw_context *brw )
 183 {
 184    struct gl_context *ctx = &brw->intel.ctx;
 185    /* BRW_NEW_VERTEX_PROGRAM */
 186    const struct brw_vertex_program *vp =
 187       brw_vertex_program_const(brw->vertex_program);
 188    /* BRW_NEW_INPUT_DIMENSIONS */
 189    struct tracker t;
 190    GLuint insn;
 191    GLuint i;
 192
 193    /* Mesa IR is not generated for GLSL vertex shaders.  If there's no Mesa
 194     * IR, the code below cannot determine which output components are
 195     * written.  So, skip it and assume everything is written.  This
 196     * circumvents some optimizations in the fragment shader, but it guarantees
 197     * that correct code is generated.
 198     */
 199    if (vp->program.Base.NumInstructions == 0) {
 200       brw->wm.input_size_masks[0] = ~(GLbitfield64) 0;
 201       brw->wm.input_size_masks[1] = ~(GLbitfield64) 0;
 202       brw->wm.input_size_masks[2] = ~(GLbitfield64) 0;
 203       brw->wm.input_size_masks[3] = ~(GLbitfield64) 0;
 204       return;
 205    }
 206
 207
 208    memset(&t, 0, sizeof(t));
 209
 210    /* _NEW_LIGHT | _NEW_PROGRAM */
 211    if (ctx->VertexProgram._TwoSideEnabled)
 212       t.twoside = 1;
 213
 214    for (i = 0; i < VERT_ATTRIB_MAX; i++)
 215       if (vp->program.Base.InputsRead & BITFIELD64_BIT(i))
 216          set_active_component(&t, PROGRAM_INPUT, i,
 217                               szflag[get_input_size(brw, i)]);
 218
 219    for (insn = 0; insn < vp->program.Base.NumInstructions; insn++) {
 220       struct prog_instruction *inst = &vp->program.Base.Instructions[insn];
 221
 222       switch (inst->Opcode) {
 223       case OPCODE_ARL:
 224          break;
 225
 226       case OPCODE_MOV:
 227          set_active(&t, inst->DstReg, get_active(&t, inst->SrcReg[0]));
 228          break;
 229
 230       default:
 231          set_active(&t, inst->DstReg, 0xf);
 232          break;
 233       }
 234    }
 235
 236    calc_sizes(&t);
 237
 238    /* _NEW_POINT
 239     *
 240     * If the SF will be replacing the vertex output with a reference to
 241     * gl_PointCoord, then tell the fragment shader that the value actually
 242     * does vary.
 243     */
 244    if (ctx->Point.PointSprite) {
 245       for (int i = 0; i < 8; i++) {
 246          if (ctx->Point.CoordReplace[i]) {
 247             t.size_masks[4-1] |= VARYING_BIT_TEX(i);
 248             t.size_masks[3-1] |= VARYING_BIT_TEX(i);
 249             t.size_masks[2-1] |= VARYING_BIT_TEX(i);
 250             t.size_masks[1-1] |= VARYING_BIT_TEX(i);
 251          }
 252       }
 253    }
 254
 255    if (memcmp(brw->wm.input_size_masks, t.size_masks, sizeof(t.size_masks)) != 0) {
 256       memcpy(brw->wm.input_size_masks, t.size_masks, sizeof(t.size_masks));
 257       brw->state.dirty.brw |= BRW_NEW_WM_INPUT_DIMENSIONS;
 258    }
 259 }
 260
 261 const struct brw_tracked_state brw_wm_input_sizes = {
 262    .dirty = {
 263       .mesa  = _NEW_LIGHT | _NEW_PROGRAM | _NEW_POINT,
 264       .brw   = BRW_NEW_VERTEX_PROGRAM | BRW_NEW_INPUT_DIMENSIONS,
 265       .cache = 0
 266    },
 267    .emit = calc_wm_input_sizes
 268 };
 269