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 "util/u_math.h"
  35 #include "util/u_memory.h"
  36 #include "pipe/p_config.h"
  37
  38 #include "draw_vs.h"
  39
  40 #if defined(PIPE_ARCH_X86) && defined(PIPE_ARCH_SSE)
  41
  42 #include "pipe/p_shader_tokens.h"
  43
  44 #include "draw_private.h"
  45 #include "draw_context.h"
  46
  47 #include "rtasm/rtasm_cpu.h"
  48 #include "rtasm/rtasm_x86sse.h"
  49 #include "tgsi/tgsi_sse2.h"
  50 #include "tgsi/tgsi_parse.h"
  51
  52 #define SSE_MAX_VERTICES 4
  53
  54 typedef void (PIPE_CDECL *codegen_function) (
  55    const struct tgsi_exec_vector *input, /* 1 */
  56    struct tgsi_exec_vector *output, /* 2 */
  57    float (*constant)[4],        /* 3 */
  58    struct tgsi_exec_vector *temporary, /* 4 */
  59    float (*immediates)[4],      /* 5 */
  60    const float (*aos_input)[4], /* 6 */
  61    uint num_inputs,             /* 7 */
  62    uint input_stride,           /* 8 */
  63    float (*aos_output)[4],      /* 9 */
  64    uint num_outputs,            /* 10 */
  65    uint output_stride );        /* 11 */
  66
  67 struct draw_sse_vertex_shader {
  68    struct draw_vertex_shader base;
  69    struct x86_function sse2_program;
  70
  71    codegen_function func;
  72
  73    struct tgsi_exec_machine *machine;
  74 };
  75
  76
  77 static void
  78 vs_sse_prepare( struct draw_vertex_shader *base,
  79                 struct draw_context *draw )
  80 {
  81 }
  82
  83
  84
  85 /* Simplified vertex shader interface for the pt paths.  Given the
  86  * complexity of code-generating all the above operations together,
  87  * it's time to try doing all the other stuff separately.
  88  */
  89 static void
  90 vs_sse_run_linear( struct draw_vertex_shader *base,
  91                    const float (*input)[4],
  92                    float (*output)[4],
  93                    const float (*constants)[4],
  94                    unsigned count,
  95                    unsigned input_stride,
  96                    unsigned output_stride )
  97 {
  98    struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
  99    struct tgsi_exec_machine *machine = shader->machine;
 100    unsigned int i;
 101
 102    /* By default, execute all channels.  XXX move this inside the loop
 103     * below when we support shader conditionals/loops.
 104     */
 105    tgsi_set_exec_mask(machine, 1, 1, 1, 1);
 106
 107    for (i = 0; i < count; i += MAX_TGSI_VERTICES) {
 108       unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i);
 109
 110       if (max_vertices < 4) {
 111          /* disable the unused execution channels */
 112          tgsi_set_exec_mask(machine,
 113                             1,
 114                             max_vertices > 1,
 115                             max_vertices > 2,
 116                             0);
 117       }
 118
 119       /* run compiled shader
 120        */
 121       shader->func(machine->Inputs,
 122                    machine->Outputs,
 123                    (float (*)[4])constants,
 124                    machine->Temps,
 125                    (float (*)[4])shader->base.immediates,
 126                    input,
 127                    base->info.num_inputs,
 128                    input_stride,
 129                    output,
 130                    base->info.num_outputs,
 131                    output_stride );
 132
 133       input = (const float (*)[4])((const char *)input + input_stride * max_vertices);
 134       output = (float (*)[4])((char *)output + output_stride * max_vertices);
 135    }
 136 }
 137
 138
 139
 140
 141 static void
 142 vs_sse_delete( struct draw_vertex_shader *base )
 143 {
 144    struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
 145
 146    x86_release_func( &shader->sse2_program );
 147
 148    align_free( (void *) shader->base.immediates );
 149
 150    FREE( (void*) shader->base.state.tokens );
 151    FREE( shader );
 152 }
 153
 154
 155 struct draw_vertex_shader *
 156 draw_create_vs_sse(struct draw_context *draw,
 157                           const struct pipe_shader_state *templ)
 158 {
 159    struct draw_sse_vertex_shader *vs;
 160
 161    if (!rtasm_cpu_has_sse2())
 162       return NULL;
 163
 164    vs = CALLOC_STRUCT( draw_sse_vertex_shader );
 165    if (vs == NULL)
 166       return NULL;
 167
 168    /* we make a private copy of the tokens */
 169    vs->base.state.tokens = tgsi_dup_tokens(templ->tokens);
 170    if (!vs->base.state.tokens)
 171       goto fail;
 172
 173    tgsi_scan_shader(templ->tokens, &vs->base.info);
 174
 175    vs->base.draw = draw;
 176    if (1)
 177       vs->base.create_varient = draw_vs_varient_aos_sse;
 178    else
 179       vs->base.create_varient = draw_vs_varient_generic;
 180    vs->base.prepare = vs_sse_prepare;
 181    vs->base.run_linear = vs_sse_run_linear;
 182    vs->base.delete = vs_sse_delete;
 183
 184    vs->base.immediates = align_malloc(TGSI_EXEC_NUM_IMMEDIATES * 4 *
 185                                       sizeof(float), 16);
 186
 187    vs->machine = &draw->vs.machine;
 188
 189    x86_init_func( &vs->sse2_program );
 190
 191    if (!tgsi_emit_sse2( (struct tgsi_token *) vs->base.state.tokens,
 192                         &vs->sse2_program,
 193                         (float (*)[4])vs->base.immediates,
 194                         TRUE ))
 195       goto fail;
 196
 197    vs->func = (codegen_function) x86_get_func( &vs->sse2_program );
 198    if (!vs->func) {
 199       goto fail;
 200    }
 201
 202    return &vs->base;
 203
 204 fail:
 205    debug_error("tgsi_emit_sse2() failed, falling back to interpreter\n");
 206
 207    x86_release_func( &vs->sse2_program );
 208
 209    FREE(vs);
 210    return NULL;
 211 }
 212
 213
 214
 215 #else
 216
 217 struct draw_vertex_shader *
 218 draw_create_vs_sse( struct draw_context *draw,
 219                     const struct pipe_shader_state *templ )
 220 {
 221    return (void *) 0;
 222 }
 223
 224
 225 #endif
 226