src/mesa/pipe/draw/draw_vertex_shader.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 "pipe/p_util.h"
  35 #include "pipe/p_shader_tokens.h"
  36 #include "draw_private.h"
  37 #include "draw_context.h"
  38 #include "draw_vertex.h"
  39
  40 #include "x86/rtasm/x86sse.h"
  41
  42 #include "pipe/llvm/gallivm.h"
  43
  44
  45 #define DBG_VS 0
  46
  47
  48 static INLINE unsigned
  49 compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
  50 {
  51    unsigned mask = 0;
  52    unsigned i;
  53
  54    for (i = 0; i < nr; i++) {
  55       if (dot4(clip, plane[i]) < 0)
  56          mask |= (1<<i);
  57    }
  58
  59    return mask;
  60 }
  61
  62
  63 typedef void (XSTDCALL *codegen_function) (
  64    const struct tgsi_exec_vector *input,
  65    struct tgsi_exec_vector *output,
  66    float (*constant)[4],
  67    struct tgsi_exec_vector *temporary );
  68
  69
  70 /**
  71  * Transform vertices with the current vertex program/shader
  72  * Up to four vertices can be shaded at a time.
  73  * \param vbuffer  the input vertex data
  74  * \param elts  indexes of four input vertices
  75  * \param count  number of vertices to shade [1..4]
  76  * \param vOut  array of pointers to four output vertices
  77  */
  78 static void
  79 run_vertex_program(struct draw_context *draw,
  80                    unsigned elts[4], unsigned count,
  81                    struct vertex_header *vOut[])
  82 {
  83    struct tgsi_exec_machine *machine = &draw->machine;
  84    unsigned int j;
  85
  86    ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
  87    ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_ATTRIB_MAX);
  88    const float *scale = draw->viewport.scale;
  89    const float *trans = draw->viewport.translate;
  90
  91    assert(count <= 4);
  92    assert(draw->vertex_shader->state->output_semantic_name[0]
  93           == TGSI_SEMANTIC_POSITION);
  94
  95    /* Consts does not require 16 byte alignment. */
  96    machine->Consts = (float (*)[4]) draw->user.constants;
  97
  98    machine->Inputs = ALIGN16_ASSIGN(inputs);
  99    machine->Outputs = ALIGN16_ASSIGN(outputs);
 100
 101    draw_vertex_fetch( draw, machine, elts, count );
 102
 103    /* run shader */
 104 #if defined(__i386__) || defined(__386__)
 105    if (draw->use_sse) {
 106       /* SSE */
 107       codegen_function func = (codegen_function) x86_get_func( &draw->vertex_shader->sse2_program );
 108       func(
 109          machine->Inputs,
 110          machine->Outputs,
 111          machine->Consts,
 112          machine->Temps );
 113    }
 114    else
 115 #endif
 116    {
 117       /* interpreter */
 118       tgsi_exec_machine_run( machine );
 119    }
 120
 121    /* store machine results */
 122    for (j = 0; j < count; j++) {
 123       unsigned slot;
 124       float x, y, z, w;
 125
 126       /* Handle attr[0] (position) specially:
 127        *
 128        * XXX: Computing the clipmask should be done in the vertex
 129        * program as a set of DP4 instructions appended to the
 130        * user-provided code.
 131        */
 132       x = vOut[j]->clip[0] = machine->Outputs[0].xyzw[0].f[j];
 133       y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j];
 134       z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j];
 135       w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j];
 136
 137       vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes);
 138       vOut[j]->edgeflag = 1;
 139
 140       /* divide by w */
 141       w = 1.0f / w;
 142       x *= w;
 143       y *= w;
 144       z *= w;
 145
 146       /* Viewport mapping */
 147       vOut[j]->data[0][0] = x * scale[0] + trans[0];
 148       vOut[j]->data[0][1] = y * scale[1] + trans[1];
 149       vOut[j]->data[0][2] = z * scale[2] + trans[2];
 150       vOut[j]->data[0][3] = w;
 151
 152 #if DBG_VS
 153       printf("output[%d]win: %f %f %f %f\n", j,
 154              vOut[j]->data[0][0],
 155              vOut[j]->data[0][1],
 156              vOut[j]->data[0][2],
 157              vOut[j]->data[0][3]);
 158 #endif
 159       /* Remaining attributes are packed into sequential post-transform
 160        * vertex attrib slots.
 161        */
 162       for (slot = 1; slot < draw->vertex_info.num_attribs; slot++) {
 163          vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
 164          vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
 165          vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
 166          vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
 167 #if DBG_VS
 168          printf("output[%d][%d]: %f %f %f %f\n", j, slot,
 169                 vOut[j]->data[slot][0],
 170                 vOut[j]->data[slot][1],
 171                 vOut[j]->data[slot][2],
 172                 vOut[j]->data[slot][3]);
 173 #endif
 174       }
 175    } /* loop over vertices */
 176 }
 177
 178
 179 /**
 180  * Run the vertex shader on all vertices in the vertex queue.
 181  * Called by the draw module when the vertx cache needs to be flushed.
 182  */
 183 void draw_vertex_shader_queue_flush( struct draw_context *draw )
 184 {
 185    unsigned i, j;
 186
 187 //   fprintf(stderr, " q(%d) ", draw->vs.queue_nr );
 188 #ifdef MESA_LLVM
 189    if (draw->vertex_shader->llvm_prog) {
 190       draw_vertex_shader_queue_flush_llvm(draw);
 191       return;
 192    }
 193 #endif
 194
 195    /* run vertex shader on vertex cache entries, four per invokation */
 196    for (i = 0; i < draw->vs.queue_nr; i += 4) {
 197       struct vertex_header *dests[4];
 198       unsigned elts[4];
 199       int n;
 200
 201       for (j = 0; j < 4; j++) {
 202          elts[j] = draw->vs.queue[i + j].elt;
 203          dests[j] = draw->vs.queue[i + j].dest;
 204       }
 205
 206       n = MIN2(4, draw->vs.queue_nr - i);
 207       assert(n > 0);
 208       assert(n <= 4);
 209
 210       run_vertex_program(draw, elts, n, dests);
 211    }
 212
 213    draw->vs.queue_nr = 0;
 214 }
 215
 216
 217 void *
 218 draw_create_vertex_shader(struct draw_context *draw,
 219                           const struct pipe_shader_state *shader)
 220 {
 221    struct draw_vertex_shader *vs;
 222
 223    vs = CALLOC_STRUCT( draw_vertex_shader );
 224    if (vs == NULL) {
 225       return NULL;
 226    }
 227
 228    vs->state = shader;
 229
 230 #if defined(__i386__) || defined(__386__)
 231    if (draw->use_sse) {
 232       /* cast-away const */
 233       struct pipe_shader_state *sh = (struct pipe_shader_state *) shader;
 234
 235       x86_init_func( &vs->sse2_program );
 236       tgsi_emit_sse2( sh->tokens, &vs->sse2_program );
 237    }
 238 #endif
 239 #ifdef MESA_LLVM
 240    vs->llvm_prog = gallivm_from_tgsi(shader->tokens, GALLIVM_VS);
 241    draw->engine = gallivm_global_cpu_engine();
 242    if (!draw->engine) {
 243       draw->engine = gallivm_cpu_engine_create(vs->llvm_prog);
 244    }
 245    else
 246       gallivm_cpu_jit_compile(draw->engine, vs->llvm_prog);
 247 #endif
 248
 249    return vs;
 250 }
 251
 252 void draw_bind_vertex_shader(struct draw_context *draw,
 253                              void *vcso)
 254 {
 255    draw_flush(draw);
 256    draw->vertex_shader = (struct draw_vertex_shader*)(vcso);
 257
 258    /* specify the fragment program to interpret/execute */
 259    tgsi_exec_machine_init(&draw->machine,
 260                           draw->vertex_shader->state->tokens,
 261                           PIPE_MAX_SAMPLERS,
 262                           NULL /*samplers*/ );
 263 }
 264
 265 void draw_delete_vertex_shader(struct draw_context *draw,
 266                                void *vcso)
 267 {
 268    struct draw_vertex_shader *vs;
 269
 270    vs = (struct draw_vertex_shader *) vcso;
 271
 272 #if defined(__i386__) || defined(__386__)
 273    x86_release_func( (struct x86_function *) &vs->sse2_program );
 274 #endif
 275
 276    FREE( vs );
 277 }