src/gallium/auxiliary/draw/draw_vs_exec.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  /*
  29   * Authors:
  30   *   Keith Whitwell <keith@tungstengraphics.com>
  31   *   Brian Paul
  32   */
  33
  34 #include "util/u_math.h"
  35 #include "util/u_memory.h"
  36 #include "pipe/p_shader_tokens.h"
  37
  38 #include "draw_private.h"
  39 #include "draw_context.h"
  40 #include "draw_vs.h"
  41
  42 #include "tgsi/tgsi_parse.h"
  43 #include "tgsi/tgsi_scan.h"
  44 #include "tgsi/tgsi_exec.h"
  45
  46
  47 struct exec_vertex_shader {
  48    struct draw_vertex_shader base;
  49    struct tgsi_exec_machine *machine;
  50 };
  51
  52 static struct exec_vertex_shader *exec_vertex_shader( struct draw_vertex_shader *vs )
  53 {
  54    return (struct exec_vertex_shader *)vs;
  55 }
  56
  57
  58 /* Not required for run_linear.
  59  */
  60 static void
  61 vs_exec_prepare( struct draw_vertex_shader *shader,
  62                  struct draw_context *draw )
  63 {
  64    struct exec_vertex_shader *evs = exec_vertex_shader(shader);
  65
  66    /* Specify the vertex program to interpret/execute.
  67     * Avoid rebinding when possible.
  68     */
  69    if (evs->machine->Tokens != shader->state.tokens) {
  70       tgsi_exec_machine_bind_shader(evs->machine,
  71                                     shader->state.tokens,
  72                                     draw->vs.tgsi.sampler);
  73    }
  74 }
  75
  76
  77
  78
  79 /* Simplified vertex shader interface for the pt paths.  Given the
  80  * complexity of code-generating all the above operations together,
  81  * it's time to try doing all the other stuff separately.
  82  */
  83 static void
  84 vs_exec_run_linear( struct draw_vertex_shader *shader,
  85                     const float (*input)[4],
  86                     float (*output)[4],
  87                     const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
  88                     const unsigned const_size[PIPE_MAX_CONSTANT_BUFFERS],
  89                     unsigned count,
  90                     unsigned input_stride,
  91                     unsigned output_stride )
  92 {
  93    struct exec_vertex_shader *evs = exec_vertex_shader(shader);
  94    struct tgsi_exec_machine *machine = evs->machine;
  95    unsigned int i, j;
  96    unsigned slot;
  97    boolean clamp_vertex_color = shader->draw->rasterizer->clamp_vertex_color;
  98
  99    tgsi_exec_set_constant_buffers(machine, PIPE_MAX_CONSTANT_BUFFERS,
 100                                   constants, const_size);
 101
 102    if (shader->info.uses_instanceid) {
 103       unsigned i = machine->SysSemanticToIndex[TGSI_SEMANTIC_INSTANCEID];
 104       assert(i < Elements(machine->SystemValue));
 105       for (j = 0; j < TGSI_QUAD_SIZE; j++)
 106          machine->SystemValue[i].i[j] = shader->draw->instance_id;
 107    }
 108
 109    for (i = 0; i < count; i += MAX_TGSI_VERTICES) {
 110       unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i);
 111
 112       /* Swizzle inputs.
 113        */
 114       for (j = 0; j < max_vertices; j++) {
 115 #if 0
 116          debug_printf("%d) Input vert:\n", i + j);
 117          for (slot = 0; slot < shader->info.num_inputs; slot++) {
 118             debug_printf("\t%d: %f %f %f %f\n", slot,
 119                          input[slot][0],
 120                          input[slot][1],
 121                          input[slot][2],
 122                          input[slot][3]);
 123          }
 124 #endif
 125
 126          if (shader->info.uses_vertexid) {
 127             unsigned vid = machine->SysSemanticToIndex[TGSI_SEMANTIC_VERTEXID];
 128             assert(vid < Elements(machine->SystemValue));
 129             machine->SystemValue[vid].i[j] = i + j;
 130          }
 131
 132          for (slot = 0; slot < shader->info.num_inputs; slot++) {
 133 #if 0
 134             assert(!util_is_inf_or_nan(input[slot][0]));
 135             assert(!util_is_inf_or_nan(input[slot][1]));
 136             assert(!util_is_inf_or_nan(input[slot][2]));
 137             assert(!util_is_inf_or_nan(input[slot][3]));
 138 #endif
 139             machine->Inputs[slot].xyzw[0].f[j] = input[slot][0];
 140             machine->Inputs[slot].xyzw[1].f[j] = input[slot][1];
 141             machine->Inputs[slot].xyzw[2].f[j] = input[slot][2];
 142             machine->Inputs[slot].xyzw[3].f[j] = input[slot][3];
 143          }
 144
 145          input = (const float (*)[4])((const char *)input + input_stride);
 146       }
 147
 148       tgsi_set_exec_mask(machine,
 149                          1,
 150                          max_vertices > 1,
 151                          max_vertices > 2,
 152                          max_vertices > 3);
 153
 154       /* run interpreter */
 155       tgsi_exec_machine_run( machine );
 156
 157       /* Unswizzle all output results.
 158        */
 159       for (j = 0; j < max_vertices; j++) {
 160          for (slot = 0; slot < shader->info.num_outputs; slot++) {
 161             unsigned name = shader->info.output_semantic_name[slot];
 162             if(clamp_vertex_color &&
 163                   (name == TGSI_SEMANTIC_COLOR || name == TGSI_SEMANTIC_BCOLOR))
 164             {
 165                output[slot][0] = CLAMP(machine->Outputs[slot].xyzw[0].f[j], 0.0f, 1.0f);
 166                output[slot][1] = CLAMP(machine->Outputs[slot].xyzw[1].f[j], 0.0f, 1.0f);
 167                output[slot][2] = CLAMP(machine->Outputs[slot].xyzw[2].f[j], 0.0f, 1.0f);
 168                output[slot][3] = CLAMP(machine->Outputs[slot].xyzw[3].f[j], 0.0f, 1.0f);
 169             }
 170             else if (name == TGSI_SEMANTIC_FOG) {
 171                output[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
 172                output[slot][1] = 0;
 173                output[slot][2] = 0;
 174                output[slot][3] = 1;
 175             } else
 176             {
 177                output[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
 178                output[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
 179                output[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
 180                output[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
 181             }
 182          }
 183
 184 #if 0
 185          debug_printf("%d) Post xform vert:\n", i + j);
 186          for (slot = 0; slot < shader->info.num_outputs; slot++) {
 187             debug_printf("\t%d: %f %f %f %f\n", slot,
 188                          output[slot][0],
 189                          output[slot][1],
 190                          output[slot][2],
 191                          output[slot][3]);
 192             assert(!util_is_inf_or_nan(output[slot][0]));
 193          }
 194 #endif
 195
 196          output = (float (*)[4])((char *)output + output_stride);
 197       }
 198
 199    }
 200 }
 201
 202
 203
 204
 205 static void
 206 vs_exec_delete( struct draw_vertex_shader *dvs )
 207 {
 208    FREE((void*) dvs->state.tokens);
 209    FREE( dvs );
 210 }
 211
 212
 213 struct draw_vertex_shader *
 214 draw_create_vs_exec(struct draw_context *draw,
 215                     const struct pipe_shader_state *state)
 216 {
 217    struct exec_vertex_shader *vs = CALLOC_STRUCT( exec_vertex_shader );
 218
 219    if (vs == NULL)
 220       return NULL;
 221
 222    /* we make a private copy of the tokens */
 223    vs->base.state.tokens = tgsi_dup_tokens(state->tokens);
 224    if (!vs->base.state.tokens) {
 225       FREE(vs);
 226       return NULL;
 227    }
 228
 229    tgsi_scan_shader(state->tokens, &vs->base.info);
 230
 231    vs->base.state.stream_output = state->stream_output;
 232    vs->base.draw = draw;
 233    vs->base.prepare = vs_exec_prepare;
 234    vs->base.run_linear = vs_exec_run_linear;
 235    vs->base.delete = vs_exec_delete;
 236    vs->base.create_variant = draw_vs_create_variant_generic;
 237    vs->machine = draw->vs.tgsi.machine;
 238
 239    return &vs->base;
 240 }