src/gallium/auxiliary/draw/draw_vs_ppc.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2008 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_PPC)
  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_ppc.h"
  49 #include "tgsi/tgsi_ppc.h"
  50 #include "tgsi/tgsi_parse.h"
  51
  52
  53
  54 typedef void (PIPE_CDECL *codegen_function) (float (*inputs)[4][4],
  55                                              float (*outputs)[4][4],
  56                                              float (*temps)[4][4],
  57                                              float (*immeds)[4][4],
  58                                              float (*consts)[4]);
  59
  60 #if 0
  61    const struct tgsi_exec_vector *input,
  62    struct tgsi_exec_vector *output,
  63    float (*constant)[4],        /* 3 */
  64    struct tgsi_exec_vector *temporary, /* 4 */
  65    float (*immediates)[4],      /* 5 */
  66    const float (*aos_input)[4], /* 6 */
  67    uint num_inputs,             /* 7 */
  68    uint input_stride,           /* 8 */
  69    float (*aos_output)[4],      /* 9 */
  70    uint num_outputs,            /* 10 */
  71    uint output_stride );        /* 11 */
  72 #endif
  73
  74 struct draw_ppc_vertex_shader {
  75    struct draw_vertex_shader base;
  76    struct ppc_function ppc_program;
  77
  78    codegen_function func;
  79
  80    struct tgsi_exec_machine *machine;
  81 };
  82
  83
  84 static void
  85 vs_ppc_prepare( struct draw_vertex_shader *base,
  86                 struct draw_context *draw )
  87 {
  88 }
  89
  90
  91
  92 /* Simplified vertex shader interface for the pt paths.  Given the
  93  * complexity of code-generating all the above operations together,
  94  * it's time to try doing all the other stuff separately.
  95  */
  96 static void
  97 vs_ppc_run_linear( struct draw_vertex_shader *base,
  98                    const float (*input)[4],
  99                    float (*output)[4],
 100                    const float (*constants)[4],
 101                    unsigned count,
 102                    unsigned input_stride,
 103                    unsigned output_stride )
 104 {
 105    struct draw_ppc_vertex_shader *shader = (struct draw_ppc_vertex_shader *)base;
 106    struct tgsi_exec_machine *machine = shader->machine;
 107    unsigned int i;
 108
 109 #define MAX_VERTICES 4
 110
 111    /* loop over verts */
 112    for (i = 0; i < count; i += MAX_VERTICES) {
 113       const uint max_vertices = MIN2(MAX_VERTICES, count - i);
 114       float inputs_soa[PIPE_MAX_SHADER_INPUTS][4][4] ALIGN16_ATTRIB;
 115       float outputs_soa[PIPE_MAX_SHADER_OUTPUTS][4][4] ALIGN16_ATTRIB;
 116       float temps_soa[TGSI_EXEC_NUM_TEMPS][4][4] ALIGN16_ATTRIB;
 117       uint attr;
 118
 119       /* convert (up to) four input verts to SoA format */
 120       for (attr = 0; attr < base->info.num_inputs; attr++) {
 121          const float *vIn = (const float *) input;
 122          uint vert;
 123          for (vert = 0; vert < max_vertices; vert++) {
 124 #if 0
 125             if (attr==0)
 126                printf("Input v%d a%d: %f %f %f %f\n",
 127                       vert, attr, vIn[0], vIn[1], vIn[2], vIn[3]);
 128 #endif
 129             inputs_soa[attr][0][vert] = vIn[attr * 4 + 0];
 130             inputs_soa[attr][1][vert] = vIn[attr * 4 + 1];
 131             inputs_soa[attr][2][vert] = vIn[attr * 4 + 2];
 132             inputs_soa[attr][3][vert] = vIn[attr * 4 + 3];
 133             vIn += input_stride / 4;
 134          }
 135       }
 136
 137       /* run compiled shader
 138        */
 139 #if 0
 140       shader->func(machine->Inputs,
 141                    machine->Outputs,
 142                    (float (*)[4])constants,
 143                    machine->Temps,
 144                    (float (*)[4])shader->base.immediates,
 145                    input,
 146                    base->info.num_inputs,
 147                    input_stride,
 148                    output,
 149                    base->info.num_outputs,
 150                    output_stride );
 151 #else
 152       shader->func(inputs_soa, outputs_soa, temps_soa,
 153                    (float (*)[4][4]) shader->base.immediates,
 154                    (float (*)[4]) constants);
 155
 156       /*output[0][0] = input[0][0] * 0.5;*/
 157 #endif
 158
 159       /* convert (up to) four output verts from SoA back to AoS format */
 160       for (attr = 0; attr < base->info.num_outputs; attr++) {
 161          float *vOut = (float *) output;
 162          uint vert;
 163          for (vert = 0; vert < max_vertices; vert++) {
 164             vOut[attr * 4 + 0] = outputs_soa[attr][0][vert];
 165             vOut[attr * 4 + 1] = outputs_soa[attr][1][vert];
 166             vOut[attr * 4 + 2] = outputs_soa[attr][2][vert];
 167             vOut[attr * 4 + 3] = outputs_soa[attr][3][vert];
 168 #if 0
 169             if (attr==0)
 170                printf("Output v%d a%d: %f %f %f %f\n",
 171                       vert, attr, vOut[0], vOut[1], vOut[2], vOut[3]);
 172 #endif
 173             vOut += output_stride / 4;
 174          }
 175       }
 176
 177       /* advance to next group of four input/output verts */
 178       input = (const float (*)[4])((const char *)input + input_stride * max_vertices);
 179       output = (float (*)[4])((char *)output + output_stride * max_vertices);
 180    }
 181 }
 182
 183
 184
 185
 186 static void
 187 vs_ppc_delete( struct draw_vertex_shader *base )
 188 {
 189    struct draw_ppc_vertex_shader *shader = (struct draw_ppc_vertex_shader *)base;
 190
 191    ppc_release_func( &shader->ppc_program );
 192
 193    align_free( (void *) shader->base.immediates );
 194
 195    FREE( (void*) shader->base.state.tokens );
 196    FREE( shader );
 197 }
 198
 199
 200 struct draw_vertex_shader *
 201 draw_create_vs_ppc(struct draw_context *draw,
 202                           const struct pipe_shader_state *templ)
 203 {
 204    struct draw_ppc_vertex_shader *vs;
 205
 206    vs = CALLOC_STRUCT( draw_ppc_vertex_shader );
 207    if (vs == NULL)
 208       return NULL;
 209
 210    /* we make a private copy of the tokens */
 211    vs->base.state.tokens = tgsi_dup_tokens(templ->tokens);
 212    if (!vs->base.state.tokens)
 213       goto fail;
 214
 215    tgsi_scan_shader(templ->tokens, &vs->base.info);
 216
 217    vs->base.draw = draw;
 218 #if 0
 219    if (1)
 220       vs->base.create_varient = draw_vs_varient_aos_ppc;
 221    else
 222 #endif
 223       vs->base.create_varient = draw_vs_varient_generic;
 224    vs->base.prepare = vs_ppc_prepare;
 225    vs->base.run_linear = vs_ppc_run_linear;
 226    vs->base.delete = vs_ppc_delete;
 227
 228    vs->base.immediates = align_malloc(TGSI_EXEC_NUM_IMMEDIATES * 4 * 4 *
 229                                       sizeof(float), 16);
 230
 231    vs->machine = &draw->vs.machine;
 232
 233    ppc_init_func( &vs->ppc_program, 2000 ); /* XXX fix limit */
 234
 235    if (!tgsi_emit_ppc( (struct tgsi_token *) vs->base.state.tokens,
 236                         &vs->ppc_program,
 237                         (float (*)[4])vs->base.immediates,
 238                         TRUE ))
 239       goto fail;
 240
 241    vs->func = (codegen_function) ppc_get_func( &vs->ppc_program );
 242    if (!vs->func) {
 243       goto fail;
 244    }
 245
 246    return &vs->base;
 247
 248 fail:
 249    debug_error("tgsi_emit_ppc() failed, falling back to interpreter\n");
 250
 251    ppc_release_func( &vs->ppc_program );
 252
 253    FREE(vs);
 254    return NULL;
 255 }
 256
 257
 258
 259 #else /* PIPE_ARCH_PPC */
 260
 261
 262 struct draw_vertex_shader *
 263 draw_create_vs_ppc( struct draw_context *draw,
 264                     const struct pipe_shader_state *templ )
 265 {
 266    return (void *) 0;
 267 }
 268
 269
 270 #endif /* PIPE_ARCH_PPC */