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_a = svga_fs_key_size( a );
  44    unsigned keysize_b = svga_fs_key_size( b );
  45
  46    if (keysize_a != keysize_b) {
  47       return (int)(keysize_a - keysize_b);
  48    }
  49    return memcmp( a, b, keysize_a );
  50 }
  51
  52
  53 static struct svga_shader_result *search_fs_key( struct svga_fragment_shader *fs,
  54                                                  const struct svga_fs_compile_key *key )
  55 {
  56    struct svga_shader_result *result = fs->base.results;
  57
  58    assert(key);
  59
  60    for ( ; result; result = result->next) {
  61       if (compare_fs_keys( key, &result->key.fkey ) == 0)
  62          return result;
  63    }
  64
  65    return NULL;
  66 }
  67
  68
  69 static enum pipe_error compile_fs( struct svga_context *svga,
  70                                    struct svga_fragment_shader *fs,
  71                                    const struct svga_fs_compile_key *key,
  72                                    struct svga_shader_result **out_result )
  73 {
  74    struct svga_shader_result *result;
  75    enum pipe_error ret = PIPE_ERROR;
  76
  77    result = svga_translate_fragment_program( fs, key );
  78    if (result == NULL) {
  79       ret = PIPE_ERROR_OUT_OF_MEMORY;
  80       goto fail;
  81    }
  82
  83    result->id = util_bitmask_add(svga->fs_bm);
  84    if(result->id == UTIL_BITMASK_INVALID_INDEX) {
  85       ret = PIPE_ERROR_OUT_OF_MEMORY;
  86       goto fail;
  87    }
  88
  89    ret = SVGA3D_DefineShader(svga->swc,
  90                              result->id,
  91                              SVGA3D_SHADERTYPE_PS,
  92                              result->tokens,
  93                              result->nr_tokens * sizeof result->tokens[0]);
  94    if (ret)
  95       goto fail;
  96
  97    *out_result = result;
  98    result->next = fs->base.results;
  99    fs->base.results = result;
 100    return PIPE_OK;
 101
 102 fail:
 103    if (result) {
 104       if (result->id != UTIL_BITMASK_INVALID_INDEX)
 105          util_bitmask_clear( svga->fs_bm, result->id );
 106       svga_destroy_shader_result( result );
 107    }
 108    return ret;
 109 }
 110
 111 /* The blend workaround for simulating logicop xor behaviour requires
 112  * that the incoming fragment color be white.  This change achieves
 113  * that by hooking up a hard-wired fragment shader that just emits
 114  * color 1,1,1,1
 115  *
 116  * This is a slightly incomplete solution as it assumes that the
 117  * actual bound shader has no other effects beyond generating a
 118  * fragment color.  In particular shaders containing TEXKIL and/or
 119  * depth-write will not have the correct behaviour, nor will those
 120  * expecting to use alphatest.
 121  *
 122  * These are avoidable issues, but they are not much worse than the
 123  * unavoidable ones associated with this technique, so it's not clear
 124  * how much effort should be expended trying to resolve them - the
 125  * ultimate result will still not be correct in most cases.
 126  *
 127  * Shader below was generated with:
 128  *   SVGA_DEBUG=tgsi ./mesa/progs/fp/fp-tri white.txt
 129  */
 130 static int emit_white_fs( struct svga_context *svga )
 131 {
 132    int ret = PIPE_ERROR;
 133
 134    /* ps_3_0
 135     * def c0, 1.000000, 0.000000, 0.000000, 1.000000
 136     * mov oC0, c0.x
 137     * end
 138     */
 139    static const unsigned white_tokens[] = {
 140       0xffff0300,
 141       0x05000051,
 142       0xa00f0000,
 143       0x3f800000,
 144       0x00000000,
 145       0x00000000,
 146       0x3f800000,
 147       0x02000001,
 148       0x800f0800,
 149       0xa0000000,
 150       0x0000ffff,
 151    };
 152
 153    assert(SVGA3D_INVALID_ID == UTIL_BITMASK_INVALID_INDEX);
 154    svga->state.white_fs_id = util_bitmask_add(svga->fs_bm);
 155    if(svga->state.white_fs_id == SVGA3D_INVALID_ID)
 156       goto no_fs_id;
 157
 158    ret = SVGA3D_DefineShader(svga->swc,
 159                              svga->state.white_fs_id,
 160                              SVGA3D_SHADERTYPE_PS,
 161                              white_tokens,
 162                              sizeof(white_tokens));
 163    if (ret)
 164       goto no_definition;
 165
 166    return 0;
 167
 168 no_definition:
 169    util_bitmask_clear(svga->fs_bm, svga->state.white_fs_id);
 170    svga->state.white_fs_id = SVGA3D_INVALID_ID;
 171 no_fs_id:
 172    return ret;
 173 }
 174
 175
 176 /* SVGA_NEW_TEXTURE_BINDING
 177  * SVGA_NEW_RAST
 178  * SVGA_NEW_NEED_SWTNL
 179  * SVGA_NEW_SAMPLER
 180  */
 181 static int make_fs_key( const struct svga_context *svga,
 182                         struct svga_fs_compile_key *key )
 183 {
 184    int i;
 185    int idx = 0;
 186
 187    memset(key, 0, sizeof *key);
 188
 189    /* Only need fragment shader fixup for twoside lighting if doing
 190     * hwtnl.  Otherwise the draw module does the whole job for us.
 191     *
 192     * SVGA_NEW_SWTNL
 193     */
 194    if (!svga->state.sw.need_swtnl) {
 195       /* SVGA_NEW_RAST
 196        */
 197       key->light_twoside = svga->curr.rast->templ.light_twoside;
 198       key->front_cw = (svga->curr.rast->templ.front_winding ==
 199                        PIPE_WINDING_CW);
 200    }
 201
 202
 203    /* XXX: want to limit this to the textures that the shader actually
 204     * refers to.
 205     *
 206     * SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER
 207     */
 208    for (i = 0; i < svga->curr.num_textures; i++) {
 209       if (svga->curr.texture[i]) {
 210          assert(svga->curr.sampler[i]);
 211          key->tex[i].texture_target = svga->curr.texture[i]->target;
 212          if (!svga->curr.sampler[i]->normalized_coords) {
 213             key->tex[i].width_height_idx = idx++;
 214             key->tex[i].unnormalized = TRUE;
 215             ++key->num_unnormalized_coords;
 216          }
 217       }
 218    }
 219    key->num_textures = svga->curr.num_textures;
 220
 221    idx = 0;
 222    for (i = 0; i < svga->curr.num_samplers; ++i) {
 223       if (svga->curr.sampler[i]) {
 224          key->tex[i].compare_mode = svga->curr.sampler[i]->compare_mode;
 225          key->tex[i].compare_func = svga->curr.sampler[i]->compare_func;
 226       }
 227    }
 228
 229    return 0;
 230 }
 231
 232
 233
 234 static int emit_hw_fs( struct svga_context *svga,
 235                        unsigned dirty )
 236 {
 237    struct svga_shader_result *result = NULL;
 238    unsigned id = SVGA3D_INVALID_ID;
 239    int ret = 0;
 240
 241    /* SVGA_NEW_BLEND
 242     */
 243    if (svga->curr.blend->need_white_fragments) {
 244       if (svga->state.white_fs_id == SVGA3D_INVALID_ID) {
 245          ret = emit_white_fs( svga );
 246          if (ret)
 247             return ret;
 248       }
 249       id = svga->state.white_fs_id;
 250    }
 251    else {
 252       struct svga_fragment_shader *fs = svga->curr.fs;
 253       struct svga_fs_compile_key key;
 254
 255       /* SVGA_NEW_TEXTURE_BINDING
 256        * SVGA_NEW_RAST
 257        * SVGA_NEW_NEED_SWTNL
 258        * SVGA_NEW_SAMPLER
 259        */
 260       ret = make_fs_key( svga, &key );
 261       if (ret)
 262          return ret;
 263
 264       result = search_fs_key( fs, &key );
 265       if (!result) {
 266          ret = compile_fs( svga, fs, &key, &result );
 267          if (ret)
 268             return ret;
 269       }
 270
 271       assert (result);
 272       id = result->id;
 273    }
 274
 275    assert(id != SVGA3D_INVALID_ID);
 276
 277    if (result != svga->state.hw_draw.fs) {
 278       ret = SVGA3D_SetShader(svga->swc,
 279                              SVGA3D_SHADERTYPE_PS,
 280                              id );
 281       if (ret)
 282          return ret;
 283
 284       svga->dirty |= SVGA_NEW_FS_RESULT;
 285       svga->state.hw_draw.fs = result;
 286    }
 287
 288    return 0;
 289 }
 290
 291 struct svga_tracked_state svga_hw_fs =
 292 {
 293    "fragment shader (hwtnl)",
 294    (SVGA_NEW_FS |
 295     SVGA_NEW_TEXTURE_BINDING |
 296     SVGA_NEW_NEED_SWTNL |
 297     SVGA_NEW_RAST |
 298     SVGA_NEW_SAMPLER |
 299     SVGA_NEW_BLEND),
 300    emit_hw_fs
 301 };
 302
 303
 304