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    if (base->info.uses_instanceid) {
  76       unsigned i = machine->SysSemanticToIndex[TGSI_SEMANTIC_INSTANCEID];
  77       assert(i < Elements(machine->SystemValue));
  78       machine->SystemValue[i][0] = base->draw->instance_id;
  79    }
  80 }
  81
  82
  83
  84 /* Simplified vertex shader interface for the pt paths.  Given the
  85  * complexity of code-generating all the above operations together,
  86  * it's time to try doing all the other stuff separately.
  87  */
  88 static void
  89 vs_sse_run_linear( struct draw_vertex_shader *base,
  90                    const float (*input)[4],
  91                    float (*output)[4],
  92                   const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
  93                   const unsigned const_size[PIPE_MAX_CONSTANT_BUFFERS],
  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,
 122                    (const float (*)[4])constants[0],
 123                    shader->base.immediates,
 124                    input,
 125                    base->info.num_inputs,
 126                    input_stride,
 127                    output,
 128                    base->info.num_outputs,
 129                    output_stride );
 130
 131       input = (const float (*)[4])((const char *)input + input_stride * max_vertices);
 132       output = (float (*)[4])((char *)output + output_stride * max_vertices);
 133    }
 134 }
 135
 136
 137
 138
 139 static void
 140 vs_sse_delete( struct draw_vertex_shader *base )
 141 {
 142    struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
 143
 144    x86_release_func( &shader->sse2_program );
 145
 146    align_free( (void *) shader->base.immediates );
 147
 148    FREE( (void*) shader->base.state.tokens );
 149    FREE( shader );
 150 }
 151
 152
 153 struct draw_vertex_shader *
 154 draw_create_vs_sse(struct draw_context *draw,
 155                           const struct pipe_shader_state *templ)
 156 {
 157    struct draw_sse_vertex_shader *vs;
 158
 159    if (!rtasm_cpu_has_sse2())
 160       return NULL;
 161
 162    vs = CALLOC_STRUCT( draw_sse_vertex_shader );
 163    if (vs == NULL)
 164       return NULL;
 165
 166    /* we make a private copy of the tokens */
 167    vs->base.state.tokens = tgsi_dup_tokens(templ->tokens);
 168    if (!vs->base.state.tokens)
 169       goto fail;
 170
 171    tgsi_scan_shader(templ->tokens, &vs->base.info);
 172
 173    vs->base.draw = draw;
 174    if (1)
 175       vs->base.create_variant = draw_vs_create_variant_aos_sse;
 176    else
 177       vs->base.create_variant = draw_vs_create_variant_generic;
 178    vs->base.prepare = vs_sse_prepare;
 179    vs->base.run_linear = vs_sse_run_linear;
 180    vs->base.delete = vs_sse_delete;
 181
 182    vs->base.immediates = align_malloc(TGSI_EXEC_NUM_IMMEDIATES * 4 *
 183                                       sizeof(float), 16);
 184
 185    vs->machine = draw->vs.machine;
 186
 187    x86_init_func( &vs->sse2_program );
 188
 189    if (!tgsi_emit_sse2( (struct tgsi_token *) vs->base.state.tokens,
 190                         &vs->sse2_program,
 191                         (float (*)[4])vs->base.immediates,
 192                         TRUE ))
 193       goto fail;
 194
 195    vs->func = (tgsi_sse2_vs_func) x86_get_func( &vs->sse2_program );
 196    if (!vs->func) {
 197       goto fail;
 198    }
 199
 200    return &vs->base;
 201
 202 fail:
 203    if (0)
 204       debug_warning("tgsi_emit_sse2() failed, falling back to interpreter\n");
 205
 206    x86_release_func( &vs->sse2_program );
 207
 208    FREE(vs);
 209    return NULL;
 210 }
 211
 212
 213
 214 #else
 215
 216 struct draw_vertex_shader *
 217 draw_create_vs_sse( struct draw_context *draw,
 218                     const struct pipe_shader_state *templ )
 219 {
 220    return (void *) 0;
 221 }
 222
 223
 224 #endif
 225