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