src/gallium/drivers/softpipe/sp_quad_fs.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * 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, sub license, 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 portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 /* Vertices are just an array of floats, with all the attributes
  29  * packed.  We currently assume a layout like:
  30  *
  31  * attr[0][0..3] - window position
  32  * attr[1..n][0..3] - remaining attributes.
  33  *
  34  * Attributes are assumed to be 4 floats wide but are packed so that
  35  * all the enabled attributes run contiguously.
  36  */
  37
  38 #include "util/u_math.h"
  39 #include "util/u_memory.h"
  40 #include "pipe/p_defines.h"
  41 #include "pipe/p_shader_tokens.h"
  42
  43 #include "sp_context.h"
  44 #include "sp_state.h"
  45 #include "sp_headers.h"
  46 #include "sp_quad.h"
  47 #include "sp_texture.h"
  48 #include "sp_tex_sample.h"
  49
  50
  51 struct quad_shade_stage
  52 {
  53    struct quad_stage stage;  /**< base class */
  54    struct sp_shader_sampler samplers[PIPE_MAX_SAMPLERS];
  55    struct sp_shader_sampler *samplers_list[PIPE_MAX_SAMPLERS];
  56    struct tgsi_exec_machine machine;
  57    struct tgsi_exec_vector *inputs, *outputs;
  58 };
  59
  60
  61 /** cast wrapper */
  62 static INLINE struct quad_shade_stage *
  63 quad_shade_stage(struct quad_stage *qs)
  64 {
  65    return (struct quad_shade_stage *) qs;
  66 }
  67
  68
  69
  70 /**
  71  * Execute fragment shader for the four fragments in the quad.
  72  */
  73 static void
  74 shade_quad(
  75    struct quad_stage *qs,
  76    struct quad_header *quad )
  77 {
  78    struct quad_shade_stage *qss = quad_shade_stage( qs );
  79    struct softpipe_context *softpipe = qs->softpipe;
  80    struct tgsi_exec_machine *machine = &qss->machine;
  81    boolean z_written;
  82
  83    /* Consts do not require 16 byte alignment. */
  84    machine->Consts = softpipe->mapped_constants[PIPE_SHADER_FRAGMENT];
  85
  86    machine->InterpCoefs = quad->coef;
  87
  88    /* run shader */
  89    quad->inout.mask &= softpipe->fs->run( softpipe->fs,
  90                                     &qss->machine,
  91                                     quad );
  92
  93    /* store outputs */
  94    z_written = FALSE;
  95    {
  96       const ubyte *sem_name = softpipe->fs->info.output_semantic_name;
  97       const ubyte *sem_index = softpipe->fs->info.output_semantic_index;
  98       const uint n = qss->stage.softpipe->fs->info.num_outputs;
  99       uint i;
 100       for (i = 0; i < n; i++) {
 101          switch (sem_name[i]) {
 102          case TGSI_SEMANTIC_COLOR:
 103             {
 104                uint cbuf = sem_index[i];
 105                memcpy(quad->output.color[cbuf],
 106                       &machine->Outputs[i].xyzw[0].f[0],
 107                       sizeof(quad->output.color[0]) );
 108             }
 109             break;
 110          case TGSI_SEMANTIC_POSITION:
 111             {
 112                uint j;
 113                for (j = 0; j < 4; j++) {
 114                   quad->output.depth[j] = machine->Outputs[0].xyzw[2].f[j];
 115                }
 116                z_written = TRUE;
 117             }
 118             break;
 119          }
 120       }
 121    }
 122
 123    if (!z_written) {
 124       /* compute Z values now, as in the quad earlyz stage */
 125       /* XXX we should really only do this if the earlyz stage is not used */
 126       const float fx = (float) quad->input.x0;
 127       const float fy = (float) quad->input.y0;
 128       const float dzdx = quad->posCoef->dadx[2];
 129       const float dzdy = quad->posCoef->dady[2];
 130       const float z0 = quad->posCoef->a0[2] + dzdx * fx + dzdy * fy;
 131
 132       quad->output.depth[0] = z0;
 133       quad->output.depth[1] = z0 + dzdx;
 134       quad->output.depth[2] = z0 + dzdy;
 135       quad->output.depth[3] = z0 + dzdx + dzdy;
 136    }
 137
 138    /* shader may cull fragments */
 139    if( quad->inout.mask ) {
 140       qs->next->run( qs->next, quad );
 141    }
 142 }
 143
 144 /**
 145  * Per-primitive (or per-begin?) setup
 146  */
 147 static void shade_begin(struct quad_stage *qs)
 148 {
 149    struct quad_shade_stage *qss = quad_shade_stage(qs);
 150    struct softpipe_context *softpipe = qs->softpipe;
 151
 152    softpipe->fs->prepare( softpipe->fs,
 153                           &qss->machine,
 154                           (struct tgsi_sampler **) qss->samplers_list );
 155
 156    qs->next->begin(qs->next);
 157 }
 158
 159
 160 static void shade_destroy(struct quad_stage *qs)
 161 {
 162    struct quad_shade_stage *qss = (struct quad_shade_stage *) qs;
 163
 164    tgsi_exec_machine_free_data(&qss->machine);
 165    FREE( qss->inputs );
 166    FREE( qss->outputs );
 167    FREE( qs );
 168 }
 169
 170
 171 struct quad_stage *sp_quad_shade_stage( struct softpipe_context *softpipe )
 172 {
 173    struct quad_shade_stage *qss = CALLOC_STRUCT(quad_shade_stage);
 174    uint i;
 175
 176    /* allocate storage for program inputs/outputs, aligned to 16 bytes */
 177    qss->inputs = MALLOC(PIPE_MAX_ATTRIBS * sizeof(*qss->inputs) + 16);
 178    qss->outputs = MALLOC(PIPE_MAX_ATTRIBS * sizeof(*qss->outputs) + 16);
 179    qss->machine.Inputs = align16(qss->inputs);
 180    qss->machine.Outputs = align16(qss->outputs);
 181
 182    qss->stage.softpipe = softpipe;
 183    qss->stage.begin = shade_begin;
 184    qss->stage.run = shade_quad;
 185    qss->stage.destroy = shade_destroy;
 186
 187    /* setup TGSI sampler state */
 188    for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
 189       assert(softpipe->tex_cache[i]);
 190       qss->samplers[i].base.get_samples = sp_get_samples;
 191       qss->samplers[i].unit = i;
 192       qss->samplers[i].sp = softpipe;
 193       qss->samplers[i].cache = softpipe->tex_cache[i];
 194       qss->samplers_list[i] = &qss->samplers[i];
 195    }
 196
 197    tgsi_exec_machine_init( &qss->machine );
 198
 199    return &qss->stage;
 200 }