src/mesa/drivers/dri/r300/r300_fragprog_common.c

   1 /*
   2  * Copyright (C) 2009 Maciej Cencora <m.cencora@gmail.com>
   3  *
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining
   7  * a 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, sublicense, 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
  16  * portions of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  */
  27
  28 /**
  29  * \file
  30  *
  31  * Fragment program compiler. Perform transformations on the intermediate
  32  * representation until the program is in a form where we can translate
  33  * it more or less directly into machine-readable form.
  34  *
  35  * \author Ben Skeggs <darktama@iinet.net.au>
  36  * \author Jerome Glisse <j.glisse@gmail.com>
  37  */
  38
  39 #include "r300_fragprog_common.h"
  40
  41 #include "program/prog_print.h"
  42
  43 #include "compiler/radeon_compiler.h"
  44
  45 #include "radeon_mesa_to_rc.h"
  46
  47
  48 static GLuint build_dts(GLuint depthmode)
  49 {
  50         switch(depthmode) {
  51         default:
  52         case GL_LUMINANCE: return RC_SWIZZLE_XYZZ;
  53         case GL_INTENSITY: return RC_SWIZZLE_XYZW;
  54         case GL_ALPHA: return RC_SWIZZLE_WWWW;
  55         }
  56 }
  57
  58 static GLuint build_func(GLuint comparefunc)
  59 {
  60         return comparefunc - GL_NEVER;
  61 }
  62
  63 /**
  64  * Collect all external state that is relevant for compiling the given
  65  * fragment program.
  66  */
  67 static void build_state(
  68         r300ContextPtr r300,
  69         struct gl_fragment_program *fp,
  70         struct r300_fragment_program_external_state *state)
  71 {
  72         int unit;
  73
  74         memset(state, 0, sizeof(*state));
  75
  76         for(unit = 0; unit < 16; ++unit) {
  77                 if (fp->Base.ShadowSamplers & (1 << unit)) {
  78                         struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current;
  79
  80                         state->unit[unit].depth_texture_swizzle = build_dts(tex->DepthMode);
  81                         state->unit[unit].texture_compare_func = build_func(tex->CompareFunc);
  82                 }
  83         }
  84 }
  85
  86
  87 /**
  88  * Transform the program to support fragment.position.
  89  *
  90  * Introduce a small fragment at the start of the program that will be
  91  * the only code that directly reads the FRAG_ATTRIB_WPOS input.
  92  * All other code pieces that reference that input will be rewritten
  93  * to read from a newly allocated temporary.
  94  *
  95  */
  96 static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler, struct r300_fragment_program * fp)
  97 {
  98         int i;
  99
 100         fp->wpos_attr = FRAG_ATTRIB_MAX;
 101         if (!(compiler->Base.Program.InputsRead & FRAG_BIT_WPOS)) {
 102                 return;
 103         }
 104
 105         for (i = FRAG_ATTRIB_TEX0; i <= FRAG_ATTRIB_TEX7; ++i)
 106         {
 107                 if (!(compiler->Base.Program.InputsRead & (1 << i))) {
 108                         fp->wpos_attr = i;
 109                         break;
 110                 }
 111         }
 112
 113         /* No free texcoord found, fall-back to software rendering */
 114         if (fp->wpos_attr == FRAG_ATTRIB_MAX)
 115         {
 116                 compiler->Base.Error = 1;
 117                 return;
 118         }
 119
 120         rc_transform_fragment_wpos(&compiler->Base, FRAG_ATTRIB_WPOS, fp->wpos_attr, GL_FALSE);
 121 }
 122
 123 /**
 124  * Rewrite fragment.fogcoord to use a texture coordinate slot.
 125  * Note that fogcoord is forced into an X001 pattern, and this enforcement
 126  * is done here.
 127  *
 128  * See also the counterpart rewriting for vertex programs.
 129  */
 130 static void rewriteFog(struct r300_fragment_program_compiler *compiler, struct r300_fragment_program * fp)
 131 {
 132         struct rc_src_register src;
 133         int i;
 134
 135         fp->fog_attr = FRAG_ATTRIB_MAX;
 136         if (!(compiler->Base.Program.InputsRead & FRAG_BIT_FOGC)) {
 137                 return;
 138         }
 139
 140         for (i = FRAG_ATTRIB_TEX0; i <= FRAG_ATTRIB_TEX7; ++i)
 141         {
 142                 if (!(compiler->Base.Program.InputsRead & (1 << i))) {
 143                         fp->fog_attr = i;
 144                         break;
 145                 }
 146         }
 147
 148         /* No free texcoord found, fall-back to software rendering */
 149         if (fp->fog_attr == FRAG_ATTRIB_MAX)
 150         {
 151                 compiler->Base.Error = 1;
 152                 return;
 153         }
 154
 155         memset(&src, 0, sizeof(src));
 156         src.File = RC_FILE_INPUT;
 157         src.Index = fp->fog_attr;
 158         src.Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ONE);
 159         rc_move_input(&compiler->Base, FRAG_ATTRIB_FOGC, src);
 160 }
 161
 162
 163 /**
 164  * Reserve hardware temporary registers for the program inputs.
 165  *
 166  * @note This allocation is performed explicitly, because the order of inputs
 167  * is determined by the RS hardware.
 168  */
 169 static void allocate_hw_inputs(
 170         struct r300_fragment_program_compiler * c,
 171         void (*allocate)(void * data, unsigned input, unsigned hwreg),
 172         void * mydata)
 173 {
 174         GLuint InputsRead = c->Base.Program.InputsRead;
 175         int i;
 176         GLuint hwindex = 0;
 177
 178         /* Primary colour */
 179         if (InputsRead & FRAG_BIT_COL0)
 180                 allocate(mydata, FRAG_ATTRIB_COL0, hwindex++);
 181         InputsRead &= ~FRAG_BIT_COL0;
 182
 183         /* Secondary color */
 184         if (InputsRead & FRAG_BIT_COL1)
 185                 allocate(mydata, FRAG_ATTRIB_COL1, hwindex++);
 186         InputsRead &= ~FRAG_BIT_COL1;
 187
 188         /* Texcoords */
 189         for (i = 0; i < 8; i++) {
 190                 if (InputsRead & (FRAG_BIT_TEX0 << i))
 191                         allocate(mydata, FRAG_ATTRIB_TEX0+i, hwindex++);
 192         }
 193         InputsRead &= ~FRAG_BITS_TEX_ANY;
 194
 195         /* Fogcoords treated as a texcoord */
 196         if (InputsRead & FRAG_BIT_FOGC)
 197                 allocate(mydata, FRAG_ATTRIB_FOGC, hwindex++);
 198         InputsRead &= ~FRAG_BIT_FOGC;
 199
 200         /* fragment position treated as a texcoord */
 201         if (InputsRead & FRAG_BIT_WPOS)
 202                 allocate(mydata, FRAG_ATTRIB_WPOS, hwindex++);
 203         InputsRead &= ~FRAG_BIT_WPOS;
 204
 205         /* Anything else */
 206         if (InputsRead)
 207                 rc_error(&c->Base, "Don't know how to handle inputs 0x%x\n", InputsRead);
 208 }
 209
 210
 211 static void translate_fragment_program(struct gl_context *ctx, struct r300_fragment_program_cont *cont, struct r300_fragment_program *fp)
 212 {
 213         r300ContextPtr r300 = R300_CONTEXT(ctx);
 214         struct r300_fragment_program_compiler compiler;
 215
 216         memset(&compiler, 0, sizeof(compiler));
 217         rc_init(&compiler.Base);
 218         compiler.Base.Debug = (RADEON_DEBUG & RADEON_PIXEL) ? GL_TRUE : GL_FALSE;
 219
 220         compiler.code = &fp->code;
 221         compiler.state = fp->state;
 222         compiler.enable_shadow_ambient = GL_TRUE;
 223         compiler.Base.is_r500 = (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) ? GL_TRUE : GL_FALSE;
 224         compiler.Base.disable_optimizations = 0;
 225         compiler.Base.has_half_swizzles = 1;
 226         compiler.Base.max_temp_regs = (compiler.Base.is_r500) ? 128 : 32;
 227         compiler.Base.max_constants = compiler.Base.is_r500 ? 256 : 32;
 228         compiler.Base.max_alu_insts = compiler.Base.is_r500 ? 512 : 64;
 229         compiler.OutputDepth = FRAG_RESULT_DEPTH;
 230         memset(compiler.OutputColor, 0, 4 * sizeof(unsigned));
 231         compiler.OutputColor[0] = FRAG_RESULT_COLOR;
 232         compiler.AllocateHwInputs = &allocate_hw_inputs;
 233
 234         if (compiler.Base.Debug) {
 235                 fflush(stderr);
 236                 printf("Fragment Program: Initial program:\n");
 237                 _mesa_print_program(&cont->Base.Base);
 238                 fflush(stderr);
 239         }
 240
 241         radeon_mesa_to_rc_program(&compiler.Base, &cont->Base.Base);
 242
 243         insert_WPOS_trailer(&compiler, fp);
 244
 245         rewriteFog(&compiler, fp);
 246
 247         r3xx_compile_fragment_program(&compiler);
 248
 249         if (compiler.Base.is_r500) {
 250                 /* We need to support the non-KMS DRM interface, which
 251                  * artificially limits the number of instructions and
 252                  * constants which are available to us.
 253                  *
 254                  * See also the comment in r300_context.c where we
 255                  * set the MAX_NATIVE_xxx values.
 256                  */
 257                 if (fp->code.code.r500.inst_end >= 255 || fp->code.constants.Count > 255)
 258                         rc_error(&compiler.Base, "Program is too big (upgrade to r300g to avoid this limitation).\n");
 259         }
 260
 261         fp->error = compiler.Base.Error;
 262
 263         fp->InputsRead = compiler.Base.Program.InputsRead;
 264
 265         /* Clear the fog/wpos_attr if code accessing these
 266          * attributes has been removed during compilation
 267          */
 268         if (fp->fog_attr != FRAG_ATTRIB_MAX) {
 269                 if (!(fp->InputsRead & (1 << fp->fog_attr)))
 270                         fp->fog_attr = FRAG_ATTRIB_MAX;
 271         }
 272
 273         if (fp->wpos_attr != FRAG_ATTRIB_MAX) {
 274                 if (!(fp->InputsRead & (1 << fp->wpos_attr)))
 275                         fp->wpos_attr = FRAG_ATTRIB_MAX;
 276         }
 277
 278         rc_destroy(&compiler.Base);
 279 }
 280
 281 struct r300_fragment_program *r300SelectAndTranslateFragmentShader(struct gl_context *ctx)
 282 {
 283         r300ContextPtr r300 = R300_CONTEXT(ctx);
 284         struct r300_fragment_program_cont *fp_list;
 285         struct r300_fragment_program *fp;
 286         struct r300_fragment_program_external_state state;
 287
 288         fp_list = (struct r300_fragment_program_cont *)ctx->FragmentProgram._Current;
 289         build_state(r300, ctx->FragmentProgram._Current, &state);
 290
 291         fp = fp_list->progs;
 292         while (fp) {
 293                 if (memcmp(&fp->state, &state, sizeof(state)) == 0) {
 294                         return r300->selected_fp = fp;
 295                 }
 296                 fp = fp->next;
 297         }
 298
 299         fp = calloc(1, sizeof(struct r300_fragment_program));
 300
 301         fp->state = state;
 302
 303         fp->next = fp_list->progs;
 304         fp_list->progs = fp;
 305
 306         translate_fragment_program(ctx, fp_list, fp);
 307
 308         return r300->selected_fp = fp;
 309 }