src/gallium/drivers/svga/svga_state_fs.c

   1 /**********************************************************
   2  * Copyright 2008-2009 VMware, Inc.  All rights reserved.
   3  *
   4  * Permission is hereby granted, free of charge, to any person
   5  * obtaining a copy of this software and associated documentation
   6  * files (the "Software"), to deal in the Software without
   7  * restriction, including without limitation the rights to use, copy,
   8  * modify, merge, publish, distribute, sublicense, and/or sell copies
   9  * of the Software, and to permit persons to whom the Software is
  10  * furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice shall be
  13  * included in all copies or substantial portions of the Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  16  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  17  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  18  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  19  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  20  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  21  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  *
  24  **********************************************************/
  25
  26 #include "pipe/p_inlines.h"
  27 #include "pipe/p_defines.h"
  28 #include "util/u_math.h"
  29 #include "util/u_bitmask.h"
  30
  31 #include "svga_context.h"
  32 #include "svga_state.h"
  33 #include "svga_cmd.h"
  34 #include "svga_tgsi.h"
  35
  36 #include "svga_hw_reg.h"
  37
  38
  39
  40 static INLINE int compare_fs_keys( const struct svga_fs_compile_key *a,
  41                                    const struct svga_fs_compile_key *b )
  42 {
  43    unsigned keysize = svga_fs_key_size( a );
  44    return memcmp( a, b, keysize );
  45 }
  46
  47
  48 static struct svga_shader_result *search_fs_key( struct svga_fragment_shader *fs,
  49                                                  const struct svga_fs_compile_key *key )
  50 {
  51    struct svga_shader_result *result = fs->base.results;
  52
  53    assert(key);
  54
  55    for ( ; result; result = result->next) {
  56       if (compare_fs_keys( key, &result->key.fkey ) == 0)
  57          return result;
  58    }
  59
  60    return NULL;
  61 }
  62
  63
  64 static enum pipe_error compile_fs( struct svga_context *svga,
  65                                    struct svga_fragment_shader *fs,
  66                                    const struct svga_fs_compile_key *key,
  67                                    struct svga_shader_result **out_result )
  68 {
  69    struct svga_shader_result *result;
  70    enum pipe_error ret = PIPE_ERROR;
  71
  72    result = svga_translate_fragment_program( fs, key );
  73    if (result == NULL) {
  74       ret = PIPE_ERROR_OUT_OF_MEMORY;
  75       goto fail;
  76    }
  77
  78    result->id = util_bitmask_add(svga->fs_bm);
  79    if(result->id == UTIL_BITMASK_INVALID_INDEX) {
  80       ret = PIPE_ERROR_OUT_OF_MEMORY;
  81       goto fail;
  82    }
  83
  84    ret = SVGA3D_DefineShader(svga->swc,
  85                              result->id,
  86                              SVGA3D_SHADERTYPE_PS,
  87                              result->tokens,
  88                              result->nr_tokens * sizeof result->tokens[0]);
  89    if (ret)
  90       goto fail;
  91
  92    *out_result = result;
  93    result->next = fs->base.results;
  94    fs->base.results = result;
  95    return PIPE_OK;
  96
  97 fail:
  98    if (result) {
  99       if (result->id != UTIL_BITMASK_INVALID_INDEX)
 100          util_bitmask_clear( svga->fs_bm, result->id );
 101       svga_destroy_shader_result( result );
 102    }
 103    return ret;
 104 }
 105
 106 /* The blend workaround for simulating logicop xor behaviour requires
 107  * that the incoming fragment color be white.  This change achieves
 108  * that by hooking up a hard-wired fragment shader that just emits
 109  * color 1,1,1,1
 110  *
 111  * This is a slightly incomplete solution as it assumes that the
 112  * actual bound shader has no other effects beyond generating a
 113  * fragment color.  In particular shaders containing TEXKIL and/or
 114  * depth-write will not have the correct behaviour, nor will those
 115  * expecting to use alphatest.
 116  *
 117  * These are avoidable issues, but they are not much worse than the
 118  * unavoidable ones associated with this technique, so it's not clear
 119  * how much effort should be expended trying to resolve them - the
 120  * ultimate result will still not be correct in most cases.
 121  *
 122  * Shader below was generated with:
 123  *   SVGA_DEBUG=tgsi ./mesa/progs/fp/fp-tri white.txt
 124  */
 125 static int emit_white_fs( struct svga_context *svga )
 126 {
 127    int ret = PIPE_ERROR;
 128
 129    /* ps_3_0
 130     * def c0, 1.000000, 0.000000, 0.000000, 1.000000
 131     * mov oC0, c0.x
 132     * end
 133     */
 134    static const unsigned white_tokens[] = {
 135       0xffff0300,
 136       0x05000051,
 137       0xa00f0000,
 138       0x3f800000,
 139       0x00000000,
 140       0x00000000,
 141       0x3f800000,
 142       0x02000001,
 143       0x800f0800,
 144       0xa0000000,
 145       0x0000ffff,
 146    };
 147
 148    assert(SVGA3D_INVALID_ID == UTIL_BITMASK_INVALID_INDEX);
 149    svga->state.white_fs_id = util_bitmask_add(svga->fs_bm);
 150    if(svga->state.white_fs_id == SVGA3D_INVALID_ID)
 151       goto no_fs_id;
 152
 153    ret = SVGA3D_DefineShader(svga->swc,
 154                              svga->state.white_fs_id,
 155                              SVGA3D_SHADERTYPE_PS,
 156                              white_tokens,
 157                              sizeof(white_tokens));
 158    if (ret)
 159       goto no_definition;
 160
 161    return 0;
 162
 163 no_definition:
 164    util_bitmask_clear(svga->fs_bm, svga->state.white_fs_id);
 165    svga->state.white_fs_id = SVGA3D_INVALID_ID;
 166 no_fs_id:
 167    return ret;
 168 }
 169
 170
 171 /* SVGA_NEW_TEXTURE_BINDING
 172  * SVGA_NEW_RAST
 173  * SVGA_NEW_NEED_SWTNL
 174  * SVGA_NEW_SAMPLER
 175  */
 176 static int make_fs_key( const struct svga_context *svga,
 177                         struct svga_fs_compile_key *key )
 178 {
 179    int i;
 180    int idx = 0;
 181
 182    memset(key, 0, sizeof *key);
 183
 184    /* Only need fragment shader fixup for twoside lighting if doing
 185     * hwtnl.  Otherwise the draw module does the whole job for us.
 186     *
 187     * SVGA_NEW_SWTNL
 188     */
 189    if (!svga->state.sw.need_swtnl) {
 190       /* SVGA_NEW_RAST
 191        */
 192       key->light_twoside = svga->curr.rast->templ.light_twoside;
 193       key->front_cw = (svga->curr.rast->templ.front_winding ==
 194                        PIPE_WINDING_CW);
 195    }
 196
 197
 198    /* XXX: want to limit this to the textures that the shader actually
 199     * refers to.
 200     *
 201     * SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER
 202     */
 203    for (i = 0; i < svga->curr.num_textures; i++) {
 204       if (svga->curr.texture[i]) {
 205          assert(svga->curr.sampler[i]);
 206          key->tex[i].texture_target = svga->curr.texture[i]->target;
 207          if (!svga->curr.sampler[i]->normalized_coords) {
 208             key->tex[i].width_height_idx = idx++;
 209             key->tex[i].unnormalized = TRUE;
 210             ++key->num_unnormalized_coords;
 211          }
 212       }
 213    }
 214    key->num_textures = svga->curr.num_textures;
 215
 216    idx = 0;
 217    for (i = 0; i < svga->curr.num_samplers; ++i) {
 218       if (svga->curr.sampler[i]) {
 219          key->tex[i].compare_mode = svga->curr.sampler[i]->compare_mode;
 220          key->tex[i].compare_func = svga->curr.sampler[i]->compare_func;
 221       }
 222    }
 223
 224    return 0;
 225 }
 226
 227
 228
 229 static int emit_hw_fs( struct svga_context *svga,
 230                        unsigned dirty )
 231 {
 232    struct svga_shader_result *result = NULL;
 233    unsigned id = SVGA3D_INVALID_ID;
 234    int ret = 0;
 235
 236    /* SVGA_NEW_BLEND
 237     */
 238    if (svga->curr.blend->need_white_fragments) {
 239       if (svga->state.white_fs_id == SVGA3D_INVALID_ID) {
 240          ret = emit_white_fs( svga );
 241          if (ret)
 242             return ret;
 243       }
 244       id = svga->state.white_fs_id;
 245    }
 246    else {
 247       struct svga_fragment_shader *fs = svga->curr.fs;
 248       struct svga_fs_compile_key key;
 249
 250       /* SVGA_NEW_TEXTURE_BINDING
 251        * SVGA_NEW_RAST
 252        * SVGA_NEW_NEED_SWTNL
 253        * SVGA_NEW_SAMPLER
 254        */
 255       ret = make_fs_key( svga, &key );
 256       if (ret)
 257          return ret;
 258
 259       result = search_fs_key( fs, &key );
 260       if (!result) {
 261          ret = compile_fs( svga, fs, &key, &result );
 262          if (ret)
 263             return ret;
 264       }
 265
 266       assert (result);
 267       id = result->id;
 268    }
 269
 270    assert(id != SVGA3D_INVALID_ID);
 271
 272    if (result != svga->state.hw_draw.fs) {
 273       ret = SVGA3D_SetShader(svga->swc,
 274                              SVGA3D_SHADERTYPE_PS,
 275                              id );
 276       if (ret)
 277          return ret;
 278
 279       svga->dirty |= SVGA_NEW_FS_RESULT;
 280       svga->state.hw_draw.fs = result;
 281    }
 282
 283    return 0;
 284 }
 285
 286 struct svga_tracked_state svga_hw_fs =
 287 {
 288    "fragment shader (hwtnl)",
 289    (SVGA_NEW_FS |
 290     SVGA_NEW_TEXTURE_BINDING |
 291     SVGA_NEW_NEED_SWTNL |
 292     SVGA_NEW_RAST |
 293     SVGA_NEW_SAMPLER |
 294     SVGA_NEW_BLEND),
 295    emit_hw_fs
 296 };
 297
 298
 299