src/mesa/pipe/softpipe/sp_quad_fs.c

   1 /*
   2  * Mesa 3-D graphics library
   3  * Version:  6.5
   4  *
   5  * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a
   8  * copy of this software and associated documentation files (the "Software"),
   9  * to deal in the Software without restriction, including without limitation
  10  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  11  * and/or sell copies of the Software, and to permit persons to whom the
  12  * Software is furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included
  15  * in all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  18  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
  21  * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  22  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  23  */
  24
  25 /* Vertices are just an array of floats, with all the attributes
  26  * packed.  We currently assume a layout like:
  27  *
  28  * attr[0][0..3] - window position
  29  * attr[1..n][0..3] - remaining attributes.
  30  *
  31  * Attributes are assumed to be 4 floats wide but are packed so that
  32  * all the enabled attributes run contiguously.
  33  */
  34
  35 #include "glheader.h"
  36 #include "imports.h"
  37 #include "sp_context.h"
  38 #include "sp_headers.h"
  39 #include "sp_quad.h"
  40 #include "tgsi/core/tgsi_core.h"
  41
  42 #if 0
  43 #if defined __GNUC__
  44 #define ALIGNED_ATTRIBS 1
  45 #else
  46 #define ALIGNED_ATTRIBS 0
  47 #endif
  48 #else
  49 #define ALIGNED_ATTRIBS 0
  50 #endif
  51
  52 struct exec_machine {
  53    const struct setup_coefficient *coef; /**< will point to quad->coef */
  54
  55 #if ALIGNED_ATTRIBS
  56    GLfloat attr[FRAG_ATTRIB_MAX][NUM_CHANNELS][QUAD_SIZE] __attribute__(( aligned( 16 ) ));
  57 #else
  58    GLfloat attr[FRAG_ATTRIB_MAX][NUM_CHANNELS][QUAD_SIZE];
  59 #endif
  60 };
  61
  62
  63 /**
  64  * Compute quad's attributes values, as constants (GL_FLAT shading).
  65  */
  66 static INLINE void cinterp( struct exec_machine *exec,
  67                             GLuint attrib,
  68                             GLuint i )
  69 {
  70    GLuint j;
  71
  72    for (j = 0; j < QUAD_SIZE; j++) {
  73       exec->attr[attrib][i][j] = exec->coef[attrib].a0[i];
  74    }
  75 }
  76
  77
  78 /**
  79  * Compute quad's attribute values by linear interpolation.
  80  *
  81  * Push into the fp:
  82  *
  83  *   INPUT[attr] = MAD COEF_A0[attr], COEF_DADX[attr], INPUT_WPOS.xxxx
  84  *   INPUT[attr] = MAD INPUT[attr],   COEF_DADY[attr], INPUT_WPOS.yyyy
  85  */
  86 static INLINE void linterp( struct exec_machine *exec,
  87                             GLuint attrib,
  88                             GLuint i )
  89 {
  90    GLuint j;
  91
  92    for (j = 0; j < QUAD_SIZE; j++) {
  93       const GLfloat x = exec->attr[FRAG_ATTRIB_WPOS][0][j];
  94       const GLfloat y = exec->attr[FRAG_ATTRIB_WPOS][1][j];
  95       exec->attr[attrib][i][j] = (exec->coef[attrib].a0[i] +
  96                                   exec->coef[attrib].dadx[i] * x +
  97                                   exec->coef[attrib].dady[i] * y);
  98    }
  99 }
 100
 101
 102 /**
 103  * Compute quad's attribute values by linear interpolation with
 104  * perspective correction.
 105  *
 106  * Push into the fp:
 107  *
 108  *   INPUT[attr] = MAD COEF_DADX[attr], INPUT_WPOS.xxxx, COEF_A0[attr]
 109  *   INPUT[attr] = MAD COEF_DADY[attr], INPUT_WPOS.yyyy, INPUT[attr]
 110  *   TMP         = RCP INPUT_WPOS.w
 111  *   INPUT[attr] = MUL INPUT[attr], TMP.xxxx
 112  *
 113  */
 114 static INLINE void pinterp( struct exec_machine *exec,
 115                             GLuint attrib,
 116                             GLuint i )
 117 {
 118    GLuint j;
 119
 120    for (j = 0; j < QUAD_SIZE; j++) {
 121       const GLfloat x = exec->attr[FRAG_ATTRIB_WPOS][0][j];
 122       const GLfloat y = exec->attr[FRAG_ATTRIB_WPOS][1][j];
 123       /* FRAG_ATTRIB_WPOS.w here is really 1/w */
 124       const GLfloat w = 1.0 / exec->attr[FRAG_ATTRIB_WPOS][3][j];
 125       exec->attr[attrib][i][j] = ((exec->coef[attrib].a0[i] +
 126                                    exec->coef[attrib].dadx[i] * x +
 127                                    exec->coef[attrib].dady[i] * y) * w);
 128    }
 129 }
 130
 131
 132
 133 /* This should be done by the fragment shader execution unit (code
 134  * generated from the decl instructions).  Do it here for now.
 135  */
 136 static void
 137 shade_quad( struct quad_stage *qs, struct quad_header *quad )
 138 {
 139    const struct softpipe_context *softpipe = qs->softpipe;
 140    struct exec_machine exec;
 141    const GLfloat fx = quad->x0;
 142    const GLfloat fy = quad->y0;
 143    GLuint attr, i;
 144
 145    exec.coef = quad->coef;
 146
 147    /* Position:
 148     */
 149    exec.attr[FRAG_ATTRIB_WPOS][0][0] = fx;
 150    exec.attr[FRAG_ATTRIB_WPOS][0][1] = fx + 1.0;
 151    exec.attr[FRAG_ATTRIB_WPOS][0][2] = fx;
 152    exec.attr[FRAG_ATTRIB_WPOS][0][3] = fx + 1.0;
 153
 154    exec.attr[FRAG_ATTRIB_WPOS][1][0] = fy;
 155    exec.attr[FRAG_ATTRIB_WPOS][1][1] = fy;
 156    exec.attr[FRAG_ATTRIB_WPOS][1][2] = fy + 1.0;
 157    exec.attr[FRAG_ATTRIB_WPOS][1][3] = fy + 1.0;
 158
 159    /* Z and W are done by linear interpolation */
 160    if (softpipe->need_z) {
 161       linterp(&exec, 0, 2);   /* attr[0].z */
 162    }
 163
 164    if (softpipe->need_w) {
 165       linterp(&exec, 0, 3);  /* attr[0].w */
 166       /*invert(&exec, 0, 3);*/
 167    }
 168
 169    /* Interpolate all the remaining attributes.  This will get pushed
 170     * into the fragment program's responsibilities at some point.
 171     * Start at 1 to skip fragment position attribute (computed above).
 172     */
 173    for (attr = 1; attr < quad->nr_attrs; attr++) {
 174       switch (softpipe->interp[attr]) {
 175       case INTERP_CONSTANT:
 176          for (i = 0; i < NUM_CHANNELS; i++)
 177             cinterp(&exec, attr, i);
 178          break;
 179
 180       case INTERP_LINEAR:
 181          for (i = 0; i < NUM_CHANNELS; i++)
 182             linterp(&exec, attr, i);
 183          break;
 184
 185       case INTERP_PERSPECTIVE:
 186          for (i = 0; i < NUM_CHANNELS; i++)
 187             pinterp(&exec, attr, i);
 188          break;
 189       }
 190    }
 191
 192 #if 1
 193    /*softpipe->run_fs( tri->fp, quad, &tri->outputs );*/
 194
 195    {
 196       struct tgsi_exec_machine machine;
 197       struct tgsi_exec_vector outputs[FRAG_ATTRIB_MAX + 1];
 198       struct tgsi_exec_vector *aoutputs;
 199       GLuint i;
 200
 201 #if !ALIGNED_ATTRIBS
 202       struct tgsi_exec_vector inputs[FRAG_ATTRIB_MAX + 1];
 203       struct tgsi_exec_vector *ainputs;
 204 #endif
 205
 206 #ifdef DEBUG
 207       memset(&machine, 0, sizeof(machine));
 208 #endif
 209
 210       /* init machine state */
 211       tgsi_exec_machine_init(
 212          &machine,
 213          softpipe->fs.tokens );
 214
 215       /* Consts does not require 16 byte alignment. */
 216       machine.Consts = softpipe->fs.constants->constant;
 217
 218       aoutputs = (struct tgsi_exec_vector *) tgsi_align_128bit( outputs );
 219       machine.Outputs = aoutputs;
 220
 221       assert( sizeof( struct tgsi_exec_vector ) == sizeof( exec.attr[0] ) );
 222
 223 #if ALIGNED_ATTRIBS
 224       machine.Inputs = (struct tgsi_exec_vector *) exec.attr;
 225
 226       for (i = 0; i < softpipe->nr_attrs; i++) {
 227          /* Make sure fp_attr_to_slot[] is an identity transform. */
 228          assert( softpipe->fp_attr_to_slot[i] == i );
 229       }
 230 #else
 231       ainputs = (struct tgsi_exec_vector *) tgsi_align_128bit( inputs );
 232       machine.Inputs = ainputs;
 233
 234       /* load input registers */
 235       for (i = 0; i < softpipe->nr_attrs; i++) {
 236 #if 0
 237          /* Make sure fp_attr_to_slot[] is an identity transform. */
 238          assert( softpipe->fp_attr_to_slot[i] == i );
 239
 240          memcpy(
 241             &ainputs[i],
 242             exec.attr[i],
 243             sizeof( ainputs[0] ) );
 244 #else
 245          memcpy(
 246             &ainputs[i],
 247             exec.attr[softpipe->fp_attr_to_slot[i]],
 248             sizeof( ainputs[0] ) );
 249 #endif
 250       }
 251 #endif
 252
 253       /* run shader */
 254       tgsi_exec_machine_run( &machine );
 255
 256       /* store result color */
 257       memcpy(quad->outputs.color,
 258              &aoutputs[FRAG_ATTRIB_COL0].xyzw[0].f[0],
 259              sizeof(quad->outputs.color));
 260       if (softpipe->need_z) {
 261          /* XXX temporary */
 262          quad->outputs.depth[0] = exec.attr[0][2][0];
 263          quad->outputs.depth[1] = exec.attr[0][2][1];
 264          quad->outputs.depth[2] = exec.attr[0][2][2];
 265          quad->outputs.depth[3] = exec.attr[0][2][3];
 266       }
 267    }
 268 #else
 269    {
 270       GLuint attr = softpipe->fp_attr_to_slot[FRAG_ATTRIB_COL0];
 271       assert(attr);
 272
 273       memcpy(quad->outputs.color,
 274              exec.attr[attr],
 275              sizeof(quad->outputs.color));
 276
 277       if (softpipe->need_z) {
 278          quad->outputs.depth[0] = exec.attr[0][2][0];
 279          quad->outputs.depth[1] = exec.attr[0][2][1];
 280          quad->outputs.depth[2] = exec.attr[0][2][2];
 281          quad->outputs.depth[3] = exec.attr[0][2][3];
 282       }
 283    }
 284 #endif
 285
 286    /* shader may cull fragments */
 287    if (quad->mask)
 288       qs->next->run(qs->next, quad);
 289 }
 290
 291
 292
 293 struct quad_stage *sp_quad_shade_stage( struct softpipe_context *softpipe )
 294 {
 295    struct quad_stage *stage = CALLOC_STRUCT(quad_stage);
 296
 297    stage->softpipe = softpipe;
 298    stage->run = shade_quad;
 299
 300    return stage;
 301 }