workbench/libs/mesa/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 #include "tgsi/tgsi_exec.h"
  52
  53
  54
  55 typedef void (PIPE_CDECL *codegen_function) (float (*inputs)[4][4],
  56                                              float (*outputs)[4][4],
  57                                              float (*temps)[4][4],
  58                                              float (*immeds)[4],
  59                                              float (*consts)[4],
  60                                              const float *builtins);
  61
  62
  63 struct draw_ppc_vertex_shader {
  64    struct draw_vertex_shader base;
  65    struct ppc_function ppc_program;
  66
  67    codegen_function func;
  68 };
  69
  70
  71 static void
  72 vs_ppc_prepare( struct draw_vertex_shader *base,
  73                 struct draw_context *draw )
  74 {
  75    /* nothing */
  76 }
  77
  78
  79 /**
  80  * Simplified vertex shader interface for the pt paths.  Given the
  81  * complexity of code-generating all the above operations together,
  82  * it's time to try doing all the other stuff separately.
  83  */
  84 static void
  85 vs_ppc_run_linear( struct draw_vertex_shader *base,
  86                    const float (*input)[4],
  87                    float (*output)[4],
  88                    const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
  89                    const unsigned const_size[PIPE_MAX_CONSTANT_BUFFERS],
  90                    unsigned count,
  91                    unsigned input_stride,
  92                    unsigned output_stride )
  93 {
  94    struct draw_ppc_vertex_shader *shader = (struct draw_ppc_vertex_shader *)base;
  95    unsigned int i;
  96
  97 #define MAX_VERTICES 4
  98
  99    /* loop over verts */
 100    for (i = 0; i < count; i += MAX_VERTICES) {
 101       const uint max_vertices = MIN2(MAX_VERTICES, count - i);
 102       PIPE_ALIGN_VAR(16) float inputs_soa[PIPE_MAX_SHADER_INPUTS][4][4];
 103       PIPE_ALIGN_VAR(16) float outputs_soa[PIPE_MAX_SHADER_OUTPUTS][4][4];
 104       PIPE_ALIGN_VAR(16) float temps_soa[TGSI_EXEC_NUM_TEMPS][4][4];
 105       uint attr;
 106
 107       /* convert (up to) four input verts to SoA format */
 108       for (attr = 0; attr < base->info.num_inputs; attr++) {
 109          const float *vIn = (const float *) input;
 110          uint vert;
 111          for (vert = 0; vert < max_vertices; vert++) {
 112 #if 0
 113             if (attr==0)
 114                printf("Input v%d a%d: %f %f %f %f\n",
 115                       vert, attr, vIn[0], vIn[1], vIn[2], vIn[3]);
 116 #endif
 117             inputs_soa[attr][0][vert] = vIn[attr * 4 + 0];
 118             inputs_soa[attr][1][vert] = vIn[attr * 4 + 1];
 119             inputs_soa[attr][2][vert] = vIn[attr * 4 + 2];
 120             inputs_soa[attr][3][vert] = vIn[attr * 4 + 3];
 121             vIn += input_stride / 4;
 122          }
 123       }
 124
 125       /* run compiled shader
 126        */
 127       shader->func(inputs_soa, outputs_soa, temps_soa,
 128                    (float (*)[4]) shader->base.immediates,
 129                    (float (*)[4])constants[0],
 130                    ppc_builtin_constants);
 131
 132       /* convert (up to) four output verts from SoA back to AoS format */
 133       for (attr = 0; attr < base->info.num_outputs; attr++) {
 134          float *vOut = (float *) output;
 135          uint vert;
 136          for (vert = 0; vert < max_vertices; vert++) {
 137             vOut[attr * 4 + 0] = outputs_soa[attr][0][vert];
 138             vOut[attr * 4 + 1] = outputs_soa[attr][1][vert];
 139             vOut[attr * 4 + 2] = outputs_soa[attr][2][vert];
 140             vOut[attr * 4 + 3] = outputs_soa[attr][3][vert];
 141 #if 0
 142             if (attr==0)
 143                printf("Output v%d a%d: %f %f %f %f\n",
 144                       vert, attr, vOut[0], vOut[1], vOut[2], vOut[3]);
 145 #endif
 146             vOut += output_stride / 4;
 147          }
 148       }
 149
 150       /* advance to next group of four input/output verts */
 151       input = (const float (*)[4])((const char *)input + input_stride * max_vertices);
 152       output = (float (*)[4])((char *)output + output_stride * max_vertices);
 153    }
 154 }
 155
 156
 157 static void
 158 vs_ppc_delete( struct draw_vertex_shader *base )
 159 {
 160    struct draw_ppc_vertex_shader *shader = (struct draw_ppc_vertex_shader *)base;
 161
 162    ppc_release_func( &shader->ppc_program );
 163
 164    align_free( (void *) shader->base.immediates );
 165
 166    FREE( (void*) shader->base.state.tokens );
 167    FREE( shader );
 168 }
 169
 170
 171 struct draw_vertex_shader *
 172 draw_create_vs_ppc(struct draw_context *draw,
 173                    const struct pipe_shader_state *templ)
 174 {
 175    struct draw_ppc_vertex_shader *vs;
 176
 177    vs = CALLOC_STRUCT( draw_ppc_vertex_shader );
 178    if (vs == NULL)
 179       return NULL;
 180
 181    /* we make a private copy of the tokens */
 182    vs->base.state.tokens = tgsi_dup_tokens(templ->tokens);
 183    if (!vs->base.state.tokens)
 184       goto fail;
 185
 186    tgsi_scan_shader(templ->tokens, &vs->base.info);
 187
 188    vs->base.draw = draw;
 189 #if 0
 190    if (1)
 191       vs->base.create_variant = draw_vs_variant_aos_ppc;
 192    else
 193 #endif
 194       vs->base.create_variant = draw_vs_create_variant_generic;
 195    vs->base.prepare = vs_ppc_prepare;
 196    vs->base.run_linear = vs_ppc_run_linear;
 197    vs->base.delete = vs_ppc_delete;
 198
 199    vs->base.immediates = align_malloc(TGSI_EXEC_NUM_IMMEDIATES * 4 *
 200                                       sizeof(float), 16);
 201
 202    ppc_init_func( &vs->ppc_program );
 203
 204 #if 0
 205    ppc_print_code(&vs->ppc_program, TRUE);
 206    ppc_indent(&vs->ppc_program, 8);
 207 #endif
 208
 209    if (!tgsi_emit_ppc( (struct tgsi_token *) vs->base.state.tokens,
 210                         &vs->ppc_program,
 211                        (float (*)[4]) vs->base.immediates,
 212                         TRUE ))
 213       goto fail;
 214
 215    vs->func = (codegen_function) ppc_get_func( &vs->ppc_program );
 216    if (!vs->func) {
 217       goto fail;
 218    }
 219
 220    return &vs->base;
 221
 222 fail:
 223    /*
 224    debug_error("tgsi_emit_ppc() failed, falling back to interpreter\n");
 225    */
 226
 227    ppc_release_func( &vs->ppc_program );
 228
 229    FREE(vs);
 230    return NULL;
 231 }
 232
 233
 234
 235 #else /* PIPE_ARCH_PPC */
 236
 237
 238 struct draw_vertex_shader *
 239 draw_create_vs_ppc( struct draw_context *draw,
 240                     const struct pipe_shader_state *templ )
 241 {
 242    return (void *) 0;
 243 }
 244
 245
 246 #endif /* PIPE_ARCH_PPC */