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_cpu.h"
  45 #include "rtasm/rtasm_x86sse.h"
  46 #include "tgsi/exec/tgsi_sse2.h"
  47 #include "tgsi/util/tgsi_parse.h"
  48
  49 #define SSE_MAX_VERTICES 4
  50
  51 typedef void (XSTDCALL *codegen_function) (
  52    const struct tgsi_exec_vector *input, /* 1 */
  53    struct tgsi_exec_vector *output, /* 2 */
  54    float (*constant)[4],        /* 3 */
  55    struct tgsi_exec_vector *temporary, /* 4 */
  56    float (*immediates)[4],      /* 5 */
  57    const float (*aos_input)[4], /* 6 */
  58    uint num_inputs,             /* 7 */
  59    uint input_stride,           /* 8 */
  60    float (*aos_output)[4],      /* 9 */
  61    uint num_outputs,            /* 10 */
  62    uint output_stride );        /* 11 */
  63
  64 struct draw_sse_vertex_shader {
  65    struct draw_vertex_shader base;
  66    struct x86_function sse2_program;
  67
  68    codegen_function func;
  69
  70    struct tgsi_exec_machine *machine;
  71 };
  72
  73
  74 static void
  75 vs_sse_prepare( struct draw_vertex_shader *base,
  76                 struct draw_context *draw )
  77 {
  78 }
  79
  80
  81
  82 /* Simplified vertex shader interface for the pt paths.  Given the
  83  * complexity of code-generating all the above operations together,
  84  * it's time to try doing all the other stuff separately.
  85  */
  86 static void
  87 vs_sse_run_linear( struct draw_vertex_shader *base,
  88                    const float (*input)[4],
  89                    float (*output)[4],
  90                    const float (*constants)[4],
  91                    unsigned count,
  92                    unsigned input_stride,
  93                    unsigned output_stride )
  94 {
  95    struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
  96    struct tgsi_exec_machine *machine = shader->machine;
  97    unsigned int i;
  98
  99    for (i = 0; i < count; i += MAX_TGSI_VERTICES) {
 100       unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i);
 101
 102       /* run compiled shader
 103        */
 104       shader->func(machine->Inputs,
 105                    machine->Outputs,
 106                    (float (*)[4])constants,
 107                    machine->Temps,
 108                    (float (*)[4])shader->base.immediates,
 109                    input,
 110                    base->info.num_inputs,
 111                    input_stride,
 112                    output,
 113                    base->info.num_outputs,
 114                    output_stride );
 115
 116       input = (const float (*)[4])((const char *)input + input_stride * max_vertices);
 117       output = (float (*)[4])((char *)output + output_stride * max_vertices);
 118    }
 119 }
 120
 121
 122
 123
 124 static void
 125 vs_sse_delete( struct draw_vertex_shader *base )
 126 {
 127    struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
 128
 129    x86_release_func( &shader->sse2_program );
 130
 131    align_free(shader->base.immediates);
 132
 133    FREE( (void*) shader->base.state.tokens );
 134    FREE( shader );
 135 }
 136
 137
 138 struct draw_vertex_shader *
 139 draw_create_vs_sse(struct draw_context *draw,
 140                           const struct pipe_shader_state *templ)
 141 {
 142    struct draw_sse_vertex_shader *vs;
 143
 144    if (!rtasm_cpu_has_sse2())
 145       return NULL;
 146
 147    vs = CALLOC_STRUCT( draw_sse_vertex_shader );
 148    if (vs == NULL)
 149       return NULL;
 150
 151    /* we make a private copy of the tokens */
 152    vs->base.state.tokens = tgsi_dup_tokens(templ->tokens);
 153    if (!vs->base.state.tokens)
 154       goto fail;
 155
 156    tgsi_scan_shader(templ->tokens, &vs->base.info);
 157
 158    vs->base.draw = draw;
 159    vs->base.create_varient = draw_vs_varient_aos_sse;
 160 //   vs->base.create_varient = draw_vs_varient_generic;
 161    vs->base.prepare = vs_sse_prepare;
 162    vs->base.run_linear = vs_sse_run_linear;
 163    vs->base.delete = vs_sse_delete;
 164
 165    vs->base.immediates = align_malloc(TGSI_EXEC_NUM_IMMEDIATES * 4 *
 166                                       sizeof(float), 16);
 167
 168    vs->machine = &draw->vs.machine;
 169
 170    x86_init_func( &vs->sse2_program );
 171
 172    if (!tgsi_emit_sse2( (struct tgsi_token *) vs->base.state.tokens,
 173                         &vs->sse2_program,
 174                         (float (*)[4])vs->base.immediates,
 175                         TRUE ))
 176       goto fail;
 177
 178    vs->func = (codegen_function) x86_get_func( &vs->sse2_program );
 179    if (!vs->func) {
 180       goto fail;
 181    }
 182
 183    return &vs->base;
 184
 185 fail:
 186    fprintf(stderr, "tgsi_emit_sse2() failed, falling back to interpreter\n");
 187
 188    x86_release_func( &vs->sse2_program );
 189
 190    FREE(vs);
 191    return NULL;
 192 }
 193
 194
 195
 196 #else
 197
 198 struct draw_vertex_shader *
 199 draw_create_vs_sse( struct draw_context *draw,
 200                     const struct pipe_shader_state *templ )
 201 {
 202    return (void *) 0;
 203 }
 204
 205
 206 #endif
 207