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
  30 #include "svga_context.h"
  31 #include "svga_state.h"
  32 #include "svga_cmd.h"
  33 #include "svga_tgsi.h"
  34
  35 #include "svga_hw_reg.h"
  36
  37
  38
  39 static INLINE int compare_fs_keys( const struct svga_fs_compile_key *a,
  40                                    const struct svga_fs_compile_key *b )
  41 {
  42    unsigned keysize = svga_fs_key_size( a );
  43    return memcmp( a, b, keysize );
  44 }
  45
  46
  47 static struct svga_shader_result *search_fs_key( struct svga_fragment_shader *fs,
  48                                                  const struct svga_fs_compile_key *key )
  49 {
  50    struct svga_shader_result *result = fs->base.results;
  51
  52    assert(key);
  53
  54    for ( ; result; result = result->next) {
  55       if (compare_fs_keys( key, &result->key.fkey ) == 0)
  56          return result;
  57    }
  58
  59    return NULL;
  60 }
  61
  62
  63 static enum pipe_error compile_fs( struct svga_context *svga,
  64                                    struct svga_fragment_shader *fs,
  65                                    const struct svga_fs_compile_key *key,
  66                                    struct svga_shader_result **out_result )
  67 {
  68    struct svga_shader_result *result;
  69    enum pipe_error ret;
  70
  71    result = svga_translate_fragment_program( fs, key );
  72    if (result == NULL) {
  73       ret = PIPE_ERROR_OUT_OF_MEMORY;
  74       goto fail;
  75    }
  76
  77
  78    ret = SVGA3D_DefineShader(svga->swc,
  79                              svga->state.next_fs_id,
  80                              SVGA3D_SHADERTYPE_PS,
  81                              result->tokens,
  82                              result->nr_tokens * sizeof result->tokens[0]);
  83    if (ret)
  84       goto fail;
  85
  86    *out_result = result;
  87    result->id = svga->state.next_fs_id++;
  88    result->next = fs->base.results;
  89    fs->base.results = result;
  90    return PIPE_OK;
  91
  92 fail:
  93    if (result)
  94       svga_destroy_shader_result( result );
  95    return ret;
  96 }
  97
  98 /* The blend workaround for simulating logicop xor behaviour requires
  99  * that the incoming fragment color be white.  This change achieves
 100  * that by hooking up a hard-wired fragment shader that just emits
 101  * color 1,1,1,1
 102  *
 103  * This is a slightly incomplete solution as it assumes that the
 104  * actual bound shader has no other effects beyond generating a
 105  * fragment color.  In particular shaders containing TEXKIL and/or
 106  * depth-write will not have the correct behaviour, nor will those
 107  * expecting to use alphatest.
 108  *
 109  * These are avoidable issues, but they are not much worse than the
 110  * unavoidable ones associated with this technique, so it's not clear
 111  * how much effort should be expended trying to resolve them - the
 112  * ultimate result will still not be correct in most cases.
 113  *
 114  * Shader below was generated with:
 115  *   SVGA_DEBUG=tgsi ./mesa/progs/fp/fp-tri white.txt
 116  */
 117 static int emit_white_fs( struct svga_context *svga )
 118 {
 119    int ret;
 120
 121    /* ps_3_0
 122     * def c0, 1.000000, 0.000000, 0.000000, 1.000000
 123     * mov oC0, c0.x
 124     * end
 125     */
 126    static const unsigned white_tokens[] = {
 127       0xffff0300,
 128       0x05000051,
 129       0xa00f0000,
 130       0x3f800000,
 131       0x00000000,
 132       0x00000000,
 133       0x3f800000,
 134       0x02000001,
 135       0x800f0800,
 136       0xa0000000,
 137       0x0000ffff,
 138    };
 139
 140    ret = SVGA3D_DefineShader(svga->swc,
 141                              svga->state.next_fs_id,
 142                              SVGA3D_SHADERTYPE_PS,
 143                              white_tokens,
 144                              sizeof(white_tokens));
 145    if (ret)
 146       return ret;
 147
 148    svga->state.white_fs_id = svga->state.next_fs_id++;
 149    return 0;
 150 }
 151
 152
 153 /* SVGA_NEW_TEXTURE_BINDING
 154  * SVGA_NEW_RAST
 155  * SVGA_NEW_NEED_SWTNL
 156  * SVGA_NEW_SAMPLER
 157  */
 158 static int make_fs_key( const struct svga_context *svga,
 159                         struct svga_fs_compile_key *key )
 160 {
 161    int i;
 162    int idx = 0;
 163
 164    memset(key, 0, sizeof *key);
 165
 166    /* Only need fragment shader fixup for twoside lighting if doing
 167     * hwtnl.  Otherwise the draw module does the whole job for us.
 168     *
 169     * SVGA_NEW_SWTNL
 170     */
 171    if (!svga->state.sw.need_swtnl) {
 172       /* SVGA_NEW_RAST
 173        */
 174       key->light_twoside = svga->curr.rast->templ.light_twoside;
 175       key->front_cw = (svga->curr.rast->templ.front_winding ==
 176                        PIPE_WINDING_CW);
 177    }
 178
 179
 180    /* XXX: want to limit this to the textures that the shader actually
 181     * refers to.
 182     *
 183     * SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER
 184     */
 185    for (i = 0; i < svga->curr.num_textures; i++) {
 186       if (svga->curr.texture[i]) {
 187          assert(svga->curr.sampler[i]);
 188          key->tex[i].texture_target = svga->curr.texture[i]->target;
 189          if (!svga->curr.sampler[i]->normalized_coords) {
 190             key->tex[i].width_height_idx = idx++;
 191             key->tex[i].unnormalized = TRUE;
 192             ++key->num_unnormalized_coords;
 193          }
 194       }
 195    }
 196    key->num_textures = svga->curr.num_textures;
 197
 198    idx = 0;
 199    for (i = 0; i < svga->curr.num_samplers; ++i) {
 200       if (svga->curr.sampler[i]) {
 201          key->tex[i].compare_mode = svga->curr.sampler[i]->compare_mode;
 202          key->tex[i].compare_func = svga->curr.sampler[i]->compare_func;
 203       }
 204    }
 205
 206    return 0;
 207 }
 208
 209
 210
 211 static int emit_hw_fs( struct svga_context *svga,
 212                        unsigned dirty )
 213 {
 214    struct svga_shader_result *result = NULL;
 215    unsigned id = SVGA3D_INVALID_ID;
 216    int ret = 0;
 217
 218    /* SVGA_NEW_BLEND
 219     */
 220    if (svga->curr.blend->need_white_fragments) {
 221       if (svga->state.white_fs_id == SVGA3D_INVALID_ID) {
 222          ret = emit_white_fs( svga );
 223          if (ret)
 224             return ret;
 225       }
 226       id = svga->state.white_fs_id;
 227    }
 228    else {
 229       struct svga_fragment_shader *fs = svga->curr.fs;
 230       struct svga_fs_compile_key key;
 231
 232       /* SVGA_NEW_TEXTURE_BINDING
 233        * SVGA_NEW_RAST
 234        * SVGA_NEW_NEED_SWTNL
 235        * SVGA_NEW_SAMPLER
 236        */
 237       ret = make_fs_key( svga, &key );
 238       if (ret)
 239          return ret;
 240
 241       result = search_fs_key( fs, &key );
 242       if (!result) {
 243          ret = compile_fs( svga, fs, &key, &result );
 244          if (ret)
 245             return ret;
 246       }
 247
 248       assert (result);
 249       id = result->id;
 250    }
 251
 252    assert(id != SVGA3D_INVALID_ID);
 253
 254    if (id != svga->state.hw_draw.shader_id[PIPE_SHADER_FRAGMENT]) {
 255       ret = SVGA3D_SetShader(svga->swc,
 256                              SVGA3D_SHADERTYPE_PS,
 257                              id );
 258       if (ret)
 259          return ret;
 260
 261       svga->dirty |= SVGA_NEW_FS_RESULT;
 262       svga->state.hw_draw.shader_id[PIPE_SHADER_FRAGMENT] = id;
 263       svga->state.hw_draw.fs = result;
 264    }
 265
 266    return 0;
 267 }
 268
 269 struct svga_tracked_state svga_hw_fs =
 270 {
 271    "fragment shader (hwtnl)",
 272    (SVGA_NEW_FS |
 273     SVGA_NEW_TEXTURE_BINDING |
 274     SVGA_NEW_NEED_SWTNL |
 275     SVGA_NEW_RAST |
 276     SVGA_NEW_SAMPLER |
 277     SVGA_NEW_BLEND),
 278    emit_hw_fs
 279 };
 280
 281
 282