src/gallium/auxiliary/draw/draw_vs_sse.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 "draw_vs.h"
  35
  36 #if defined(__i386__) || defined(__386__)
  37
  38 #include "pipe/p_util.h"
  39 #include "pipe/p_shader_tokens.h"
  40
  41 #include "draw_private.h"
  42 #include "draw_context.h"
  43
  44 #include "rtasm/rtasm_x86sse.h"
  45 #include "tgsi/exec/tgsi_sse2.h"
  46 #include "tgsi/util/tgsi_parse.h"
  47
  48 #define SSE_MAX_VERTICES 4
  49
  50 typedef void (XSTDCALL *codegen_function) (
  51    const struct tgsi_exec_vector *input,
  52    struct tgsi_exec_vector *output,
  53    float (*constant)[4],
  54    struct tgsi_exec_vector *temporary,
  55    float (*immediates)[4] );
  56
  57
  58 struct draw_sse_vertex_shader {
  59    struct draw_vertex_shader base;
  60    struct x86_function sse2_program;
  61    codegen_function func;
  62    float immediates[TGSI_EXEC_NUM_IMMEDIATES][4];
  63 };
  64
  65
  66 static void
  67 vs_sse_prepare( struct draw_vertex_shader *base,
  68                 struct draw_context *draw )
  69 {
  70    draw_update_vertex_fetch( draw );
  71 }
  72
  73 /**
  74  * Transform vertices with the current vertex program/shader
  75  * Up to four vertices can be shaded at a time.
  76  * \param vbuffer  the input vertex data
  77  * \param elts  indexes of four input vertices
  78  * \param count  number of vertices to shade [1..4]
  79  * \param vOut  array of pointers to four output vertices
  80  */
  81 static boolean
  82 vs_sse_run( struct draw_vertex_shader *base,
  83             struct draw_context *draw,
  84             const unsigned *elts,
  85             unsigned count,
  86             void *vOut )
  87 {
  88    struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
  89    struct tgsi_exec_machine *machine = &draw->machine;
  90    unsigned int i, j;
  91    unsigned int clipped = 0;
  92
  93    ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_MAX_ATTRIBS);
  94    ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_MAX_ATTRIBS);
  95    const float *scale = draw->viewport.scale;
  96    const float *trans = draw->viewport.translate;
  97
  98    assert(draw->vertex_shader->info.output_semantic_name[0]
  99           == TGSI_SEMANTIC_POSITION);
 100
 101    /* Consts does not require 16 byte alignment. */
 102    machine->Consts = (float (*)[4]) draw->user.constants;
 103    machine->Inputs = ALIGN16_ASSIGN(inputs);
 104    if (draw->rasterizer->bypass_vs) {
 105       /* outputs are just the inputs */
 106       machine->Outputs = machine->Inputs;
 107    }
 108    else {
 109       machine->Outputs = ALIGN16_ASSIGN(outputs);
 110    }
 111
 112    for (i = 0; i < count; i += SSE_MAX_VERTICES) {
 113       unsigned int max_vertices = MIN2(SSE_MAX_VERTICES, count - i);
 114       /* Fetch vertices.  This may at some point be integrated into the
 115        * compiled shader -- that would require a reorganization where
 116        * multiple versions of the compiled shader might exist,
 117        * specialized for each fetch state.
 118        */
 119       draw->vertex_fetch.fetch_func(draw, machine, &elts[i], max_vertices);
 120
 121       if (!draw->rasterizer->bypass_vs) {
 122          /* run compiled shader
 123           */
 124          shader->func(machine->Inputs,
 125                       machine->Outputs,
 126                       machine->Consts,
 127                       machine->Temps,
 128                       shader->immediates);
 129       }
 130
 131       /* XXX: Computing the clipmask and emitting results should be done
 132        *      in the vertex program as a set of instructions appended to
 133        *      the user-provided code.
 134        */
 135       for (j = 0; j < max_vertices; j++) {
 136          unsigned slot;
 137          float x, y, z, w;
 138          struct vertex_header *out =
 139             draw_header_from_block(vOut, i + j);
 140
 141          x = out->clip[0] = machine->Outputs[0].xyzw[0].f[j];
 142          y = out->clip[1] = machine->Outputs[0].xyzw[1].f[j];
 143          z = out->clip[2] = machine->Outputs[0].xyzw[2].f[j];
 144          w = out->clip[3] = machine->Outputs[0].xyzw[3].f[j];
 145
 146          if (!draw->rasterizer->bypass_clipping) {
 147             out->clipmask = compute_clipmask(out->clip, draw->plane,
 148                                              draw->nr_planes);
 149             clipped += out->clipmask;
 150
 151             /* divide by w */
 152             w = 1.0f / w;
 153             x *= w;
 154             y *= w;
 155             z *= w;
 156          }
 157          else {
 158             out->clipmask = 0;
 159          }
 160          out->edgeflag = 1;
 161          out->vertex_id = UNDEFINED_VERTEX_ID;
 162
 163          if (!draw->identity_viewport) {
 164             /* Viewport mapping */
 165             out->data[0][0] = x * scale[0] + trans[0];
 166             out->data[0][1] = y * scale[1] + trans[1];
 167             out->data[0][2] = z * scale[2] + trans[2];
 168             out->data[0][3] = w;
 169          }
 170          else {
 171             out->data[0][0] = x;
 172             out->data[0][1] = y;
 173             out->data[0][2] = z;
 174             out->data[0][3] = w;
 175          }
 176
 177          /* Remaining attributes are packed into sequential post-transform
 178           * vertex attrib slots.
 179           */
 180          for (slot = 1; slot < draw->num_vs_outputs; slot++) {
 181             out->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
 182             out->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
 183             out->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
 184             out->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
 185          }
 186 #if 0 /*DEBUG*/
 187          printf("%d) Post xform vert:\n", i + j);
 188          for (slot = 0; slot < draw->num_vs_outputs; slot++) {
 189             printf("\t%d: %f %f %f %f\n", slot,
 190                    out->data[slot][0],
 191                    out->data[slot][1],
 192                    out->data[slot][2],
 193                    out->data[slot][3]);
 194          }
 195 #endif
 196       }
 197    }
 198    return clipped != 0;
 199 }
 200
 201
 202
 203 static void
 204 vs_sse_delete( struct draw_vertex_shader *base )
 205 {
 206    struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
 207
 208    x86_release_func( &shader->sse2_program );
 209
 210    FREE( (void*) shader->base.state.tokens );
 211    FREE( shader );
 212 }
 213
 214
 215 struct draw_vertex_shader *
 216 draw_create_vs_sse(struct draw_context *draw,
 217                           const struct pipe_shader_state *templ)
 218 {
 219    struct draw_sse_vertex_shader *vs;
 220    uint nt = tgsi_num_tokens(templ->tokens);
 221
 222    if (!draw->use_sse)
 223       return NULL;
 224
 225    vs = CALLOC_STRUCT( draw_sse_vertex_shader );
 226    if (vs == NULL)
 227       return NULL;
 228
 229    /* we make a private copy of the tokens */
 230    vs->base.state.tokens = mem_dup(templ->tokens, nt * sizeof(templ->tokens[0]));
 231    vs->base.prepare = vs_sse_prepare;
 232    vs->base.run = vs_sse_run;
 233    vs->base.delete = vs_sse_delete;
 234
 235    x86_init_func( &vs->sse2_program );
 236
 237    if (!tgsi_emit_sse2( (struct tgsi_token *) vs->base.state.tokens,
 238                         &vs->sse2_program, vs->immediates ))
 239       goto fail;
 240
 241    vs->func = (codegen_function) x86_get_func( &vs->sse2_program );
 242
 243    return &vs->base;
 244
 245 fail:
 246    fprintf(stderr, "tgsi_emit_sse2() failed, falling back to interpreter\n");
 247
 248    x86_release_func( &vs->sse2_program );
 249
 250    FREE(vs);
 251    return NULL;
 252 }
 253
 254
 255
 256 #else
 257
 258 struct draw_vertex_shader *
 259 draw_create_vs_sse( struct draw_context *draw,
 260                     const struct pipe_shader_state *templ )
 261 {
 262    return (void *) 0;
 263 }
 264
 265
 266 #endif
 267