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)
  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 #include "tgsi/tgsi_exec.h"
  52
  53 #define SSE_MAX_VERTICES 4
  54
  55
  56 struct draw_sse_vertex_shader {
  57    struct draw_vertex_shader base;
  58    struct x86_function sse2_program;
  59
  60    tgsi_sse2_vs_func func;
  61
  62    struct tgsi_exec_machine *machine;
  63 };
  64
  65
  66 static void
  67 vs_sse_prepare( struct draw_vertex_shader *base,
  68                 struct draw_context *draw )
  69 {
  70    struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
  71    struct tgsi_exec_machine *machine = shader->machine;
  72
  73    machine->Samplers = draw->vs.samplers;
  74 }
  75
  76
  77
  78 /* Simplified vertex shader interface for the pt paths.  Given the
  79  * complexity of code-generating all the above operations together,
  80  * it's time to try doing all the other stuff separately.
  81  */
  82 static void
  83 vs_sse_run_linear( struct draw_vertex_shader *base,
  84                    const float (*input)[4],
  85                    float (*output)[4],
  86                   const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
  87                    unsigned count,
  88                    unsigned input_stride,
  89                    unsigned output_stride )
  90 {
  91    struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
  92    struct tgsi_exec_machine *machine = shader->machine;
  93    unsigned int i;
  94
  95    /* By default, execute all channels.  XXX move this inside the loop
  96     * below when we support shader conditionals/loops.
  97     */
  98    tgsi_set_exec_mask(machine, 1, 1, 1, 1);
  99
 100    for (i = 0; i < count; i += MAX_TGSI_VERTICES) {
 101       unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i);
 102
 103       if (max_vertices < 4) {
 104          /* disable the unused execution channels */
 105          tgsi_set_exec_mask(machine,
 106                             1,
 107                             max_vertices > 1,
 108                             max_vertices > 2,
 109                             0);
 110       }
 111
 112       /* run compiled shader
 113        */
 114       shader->func(machine,
 115                    (const float (*)[4])constants[0],
 116                    shader->base.immediates,
 117                    input,
 118                    base->info.num_inputs,
 119                    input_stride,
 120                    output,
 121                    base->info.num_outputs,
 122                    output_stride );
 123
 124       input = (const float (*)[4])((const char *)input + input_stride * max_vertices);
 125       output = (float (*)[4])((char *)output + output_stride * max_vertices);
 126    }
 127 }
 128
 129
 130
 131
 132 static void
 133 vs_sse_delete( struct draw_vertex_shader *base )
 134 {
 135    struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
 136
 137    x86_release_func( &shader->sse2_program );
 138
 139    align_free( (void *) shader->base.immediates );
 140
 141    FREE( (void*) shader->base.state.tokens );
 142    FREE( shader );
 143 }
 144
 145
 146 struct draw_vertex_shader *
 147 draw_create_vs_sse(struct draw_context *draw,
 148                           const struct pipe_shader_state *templ)
 149 {
 150    struct draw_sse_vertex_shader *vs;
 151
 152    if (!rtasm_cpu_has_sse2())
 153       return NULL;
 154
 155    vs = CALLOC_STRUCT( draw_sse_vertex_shader );
 156    if (vs == NULL)
 157       return NULL;
 158
 159    /* we make a private copy of the tokens */
 160    vs->base.state.tokens = tgsi_dup_tokens(templ->tokens);
 161    if (!vs->base.state.tokens)
 162       goto fail;
 163
 164    tgsi_scan_shader(templ->tokens, &vs->base.info);
 165
 166    vs->base.draw = draw;
 167    if (1)
 168       vs->base.create_varient = draw_vs_varient_aos_sse;
 169    else
 170       vs->base.create_varient = draw_vs_varient_generic;
 171    vs->base.prepare = vs_sse_prepare;
 172    vs->base.run_linear = vs_sse_run_linear;
 173    vs->base.delete = vs_sse_delete;
 174
 175    vs->base.immediates = align_malloc(TGSI_EXEC_NUM_IMMEDIATES * 4 *
 176                                       sizeof(float), 16);
 177
 178    vs->machine = draw->vs.machine;
 179
 180    x86_init_func( &vs->sse2_program );
 181
 182    if (!tgsi_emit_sse2( (struct tgsi_token *) vs->base.state.tokens,
 183                         &vs->sse2_program,
 184                         (float (*)[4])vs->base.immediates,
 185                         TRUE ))
 186       goto fail;
 187
 188    vs->func = (tgsi_sse2_vs_func) x86_get_func( &vs->sse2_program );
 189    if (!vs->func) {
 190       goto fail;
 191    }
 192
 193    return &vs->base;
 194
 195 fail:
 196    debug_error("tgsi_emit_sse2() failed, falling back to interpreter\n");
 197
 198    x86_release_func( &vs->sse2_program );
 199
 200    FREE(vs);
 201    return NULL;
 202 }
 203
 204
 205
 206 #else
 207
 208 struct draw_vertex_shader *
 209 draw_create_vs_sse( struct draw_context *draw,
 210                     const struct pipe_shader_state *templ )
 211 {
 212    return (void *) 0;
 213 }
 214
 215
 216 #endif
 217