texenvprogram.c

winNT技术操作系统,国外开放的原代码和LIUX一样

      struct ureg two  = register_scalar_const(p, 2);

      /* tmp0 = 2*src0 - 1
       * tmp1 = 2*src1 - 1
       *
       * dst = tmp0 dot3 tmp1 
       */
      emit_arith( p, FP_OPCODE_MAD, tmp0, WRITEMASK_XYZW, 0, 
		  two, src[0], neg1);

      if (memcmp(&src[0], &src[1], sizeof(struct ureg)) == 0)
	 tmp1 = tmp0;
      else
	 emit_arith( p, FP_OPCODE_MAD, tmp1, WRITEMASK_XYZW, 0, 
		     two, src[1], neg1);
      emit_arith( p, FP_OPCODE_DP3, dest, mask, saturate, tmp0, tmp1, undef);
      return dest;
   }
   case MODE_MODULATE_ADD_ATI:
      /* Arg0 * Arg2 + Arg1 */
      return emit_arith( p, FP_OPCODE_MAD, dest, mask, saturate,
			 src[0], src[2], src[1] );
   case MODE_MODULATE_SIGNED_ADD_ATI: {
      /* Arg0 * Arg2 + Arg1 - 0.5 */
      struct ureg tmp0 = get_temp(p);
      half = get_half(p);
      emit_arith( p, FP_OPCODE_MAD, tmp0, mask, 0, src[0], src[2], src[1] );
      emit_arith( p, FP_OPCODE_SUB, dest, mask, saturate, tmp0, half, undef );
      return dest;
   }
   case MODE_MODULATE_SUBTRACT_ATI:
      /* Arg0 * Arg2 - Arg1 */
      emit_arith( p, FP_OPCODE_MAD, dest, mask, 0, src[0], src[2], negate(src[1]) );
      return dest;
   default: 
      return src[0];
   }
}


static struct ureg emit_texenv( struct texenv_fragment_program *p, int unit )
{
   struct state_key *key = p->state;
   GLuint saturate = (unit < p->last_tex_stage);
   GLuint rgb_shift, alpha_shift;
   struct ureg out, shift;
   struct ureg dest;

   if (!key->unit[unit].enabled) {
      return get_source(p, SRC_PREVIOUS, 0);
   }
   
   switch (key->unit[unit].ModeRGB) {
   case MODE_DOT3_RGB_EXT:
      alpha_shift = key->unit[unit].ScaleShiftA;
      rgb_shift = 0;
      break;
   case MODE_DOT3_RGBA_EXT:
      alpha_shift = 0;
      rgb_shift = 0;
      break;
   default:
      rgb_shift = key->unit[unit].ScaleShiftRGB;
      alpha_shift = key->unit[unit].ScaleShiftA;
      break;
   }
   
   /* If this is the very last calculation, emit direct to output reg:
    */
   if (key->separate_specular ||
       unit != p->last_tex_stage ||
       alpha_shift ||
       rgb_shift)
      dest = get_temp( p );
   else
      dest = make_ureg(PROGRAM_OUTPUT, FRAG_OUTPUT_COLR);

   /* Emit the RGB and A combine ops
    */
   if (key->unit[unit].ModeRGB == key->unit[unit].ModeA &&
       args_match(key, unit)) {
      out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
			  unit,
			  key->unit[unit].NumArgsRGB,
			  key->unit[unit].ModeRGB,
			  key->unit[unit].OptRGB);
   }
   else if (key->unit[unit].ModeRGB == MODE_DOT3_RGBA_EXT ||
	    key->unit[unit].ModeA == MODE_DOT3_RGBA) {

      out = emit_combine( p, dest, WRITEMASK_XYZW, saturate,
			  unit,
			  key->unit[unit].NumArgsRGB,
			  key->unit[unit].ModeRGB,
			  key->unit[unit].OptRGB);
   }
   else {
      /* Need to do something to stop from re-emitting identical
       * argument calculations here:
       */
      out = emit_combine( p, dest, WRITEMASK_XYZ, saturate,
			  unit,
			  key->unit[unit].NumArgsRGB,
			  key->unit[unit].ModeRGB,
			  key->unit[unit].OptRGB);
      out = emit_combine( p, dest, WRITEMASK_W, saturate,
			  unit,
			  key->unit[unit].NumArgsA,
			  key->unit[unit].ModeA,
			  key->unit[unit].OptA);
   }

   /* Deal with the final shift:
    */
   if (alpha_shift || rgb_shift) {
      if (rgb_shift == alpha_shift) {
	 shift = register_scalar_const(p, 1<<rgb_shift);
      }
      else {
	 shift = register_const4f(p, 
				  1<<rgb_shift,
				  1<<rgb_shift,
				  1<<rgb_shift,
				  1<<alpha_shift);
      }
      return emit_arith( p, FP_OPCODE_MUL, dest, WRITEMASK_XYZW, 
			 saturate, out, shift, undef );
   }
   else
      return out;
}



static void load_texture( struct texenv_fragment_program *p, GLuint unit )
{
   if (is_undef(p->src_texture[unit])) {
      GLuint dim = p->state->unit[unit].source_index;
      struct ureg texcoord = register_input(p, FRAG_ATTRIB_TEX0+unit);
      struct ureg tmp = get_tex_temp( p );

      if (dim == TEXTURE_UNKNOWN_INDEX)
         program_error(p, "TexSrcBit");
			  
      /* TODO: Use D0_MASK_XY where possible.
       */
      p->src_texture[unit] = emit_texld( p, FP_OPCODE_TXP,
					 tmp, WRITEMASK_XYZW, 
					 unit, dim, texcoord );
   }
}

static GLboolean load_texenv_source( struct texenv_fragment_program *p, 
				     GLuint src, GLuint unit )
{
   switch (src) {
   case SRC_TEXTURE: 
      load_texture(p, unit);
      break;

   case SRC_TEXTURE0:
   case SRC_TEXTURE1:
   case SRC_TEXTURE2:
   case SRC_TEXTURE3:
   case SRC_TEXTURE4:
   case SRC_TEXTURE5:
   case SRC_TEXTURE6:
   case SRC_TEXTURE7:       
      if (!p->state->unit[src - SRC_TEXTURE0].enabled) 
	 return GL_FALSE;
      load_texture(p, src - SRC_TEXTURE0);
      break;
      
   default:
      break;
   }
 
   return GL_TRUE;
}

static GLboolean load_texunit_sources( struct texenv_fragment_program *p, int unit )
{
   struct state_key *key = p->state;
   int i, nr = key->unit[unit].NumArgsRGB;
   for (i = 0; i < nr; i++) {
      if (!load_texenv_source( p, key->unit[unit].OptRGB[i].Source, unit) ||
	  !load_texenv_source( p, key->unit[unit].OptA[i].Source, unit ))
	 return GL_FALSE;
   }
   return GL_TRUE;
}

static void create_new_program(struct state_key *key, GLcontext *ctx,
			       struct fragment_program *program)
{
   struct texenv_fragment_program p;
   GLuint unit;
   struct ureg cf, out;

   _mesa_memset(&p, 0, sizeof(p));
   p.ctx = ctx;
   p.state = key;
   p.program = program;

   p.program->Instructions = MALLOC(sizeof(struct fp_instruction) * 100);
   p.program->Base.NumInstructions = 0;
   p.program->Base.Target = GL_FRAGMENT_PROGRAM_ARB;
   p.program->NumTexIndirections = 1;	/* correct? */
   p.program->NumTexInstructions = 0;
   p.program->NumAluInstructions = 0;
   p.program->Base.String = 0;
   p.program->Base.NumInstructions =
      p.program->Base.NumTemporaries =
      p.program->Base.NumParameters =
      p.program->Base.NumAttributes = p.program->Base.NumAddressRegs = 0;
   p.program->Parameters = _mesa_new_parameter_list();

   p.program->InputsRead = 0;
   p.program->OutputsWritten = 1 << FRAG_OUTPUT_COLR;

   for (unit = 0; unit < MAX_TEXTURE_UNITS; unit++)
      p.src_texture[unit] = undef;

   p.src_previous = undef;
   p.last_tex_stage = 0;
   release_temps(&p);

   if (key->enabled_units) {
      /* First pass - to support texture_env_crossbar, first identify
       * all referenced texture sources and emit texld instructions
       * for each:
       */
      for (unit = 0 ; unit < ctx->Const.MaxTextureUnits ; unit++)
	 if (key->unit[unit].enabled) {
	    if (load_texunit_sources( &p, unit ))
	       p.last_tex_stage = unit;
	 }

      /* Second pass - emit combine instructions to build final color:
       */
      for (unit = 0 ; unit < ctx->Const.MaxTextureUnits; unit++)
	 if (key->enabled_units & (1<<unit)) {
	    p.src_previous = emit_texenv( &p, unit );
	    release_temps(&p);	/* release all temps */
	 }
   }

   cf = get_source( &p, SRC_PREVIOUS, 0 );
   out = make_ureg( PROGRAM_OUTPUT, FRAG_OUTPUT_COLR );

   if (key->separate_specular) {
      /* Emit specular add.
       */
      struct ureg s = register_input(&p, FRAG_ATTRIB_COL1);
      emit_arith( &p, FP_OPCODE_ADD, out, WRITEMASK_XYZ, 0, cf, s, undef );
   }
   else if (memcmp(&cf, &out, sizeof(cf)) != 0) {
      /* Will wind up in here if no texture enabled or a couple of
       * other scenarios (GL_REPLACE for instance).
       */
      emit_arith( &p, FP_OPCODE_MOV, out, WRITEMASK_XYZW, 0, cf, undef, undef );
   }

   /* Finish up:
    */
   emit_arith( &p, FP_OPCODE_END, undef, WRITEMASK_XYZW, 0, undef, undef, undef);

   if (key->fog_enabled) {
      /* Pull fog mode from GLcontext, the value in the state key is
       * a reduced value and not what is expected in FogOption
       */
      p.program->FogOption = ctx->Fog.Mode;
   } else
      p.program->FogOption = GL_NONE;

   if (p.program->NumTexIndirections > ctx->Const.MaxFragmentProgramTexIndirections) 
      program_error(&p, "Exceeded max nr indirect texture lookups");

   if (p.program->NumTexInstructions > ctx->Const.MaxFragmentProgramTexInstructions)
      program_error(&p, "Exceeded max TEX instructions");

   if (p.program->NumAluInstructions > ctx->Const.MaxFragmentProgramAluInstructions)
      program_error(&p, "Exceeded max ALU instructions");


   /* Notify driver the fragment program has (actually) changed.
    */
   if (ctx->Driver.ProgramStringNotify || DISASSEM) {
      if (ctx->Driver.ProgramStringNotify)
	 ctx->Driver.ProgramStringNotify( ctx, GL_FRAGMENT_PROGRAM_ARB, 
					  &p.program->Base );

      if (DISASSEM) {
	 _mesa_debug_fp_inst(p.program->NumTexInstructions + p.program->NumAluInstructions,
			     p.program->Instructions);
	 _mesa_printf("\n");
      }
      
   }

}

static void *search_cache( struct texenvprog_cache *cache,
			   GLuint hash,
			   const void *key,
			   GLuint keysize)
{
   struct texenvprog_cache *c;

   for (c = cache; c; c = c->next) {
      if (c->hash == hash && memcmp(c->key, key, keysize) == 0)
	 return c->data;
   }

   return NULL;
}

static void cache_item( struct texenvprog_cache **cache,
			GLuint hash,
			void *key,
			void *data )
{
   struct texenvprog_cache *c = MALLOC(sizeof(*c));
   c->hash = hash;
   c->key = key;
   c->data = data;
   c->next = *cache;
   *cache = c;
}

static GLuint hash_key( struct state_key *key )
{
   GLuint *ikey = (GLuint *)key;
   GLuint hash = 0, i;

   /* I'm sure this can be improved on, but speed is important:
    */
   for (i = 0; i < sizeof(*key)/sizeof(GLuint); i++)
      hash ^= ikey[i];

   return hash;
}

void _mesa_UpdateTexEnvProgram( GLcontext *ctx )
{
   struct state_key *key;
   GLuint hash;
	
   if (ctx->FragmentProgram._Enabled)
      return;
	
   key = make_state_key(ctx);
   hash = hash_key(key);

   ctx->FragmentProgram._Current = ctx->_TexEnvProgram =
      (struct fragment_program *)
      search_cache(ctx->Texture.env_fp_cache, hash, key, sizeof(*key));
	
   if (!ctx->_TexEnvProgram) {
      if (0) _mesa_printf("Building new texenv proggy for key %x\n", hash);
		
      ctx->FragmentProgram._Current = ctx->_TexEnvProgram = 
	 (struct fragment_program *) 
	 ctx->Driver.NewProgram(ctx, GL_FRAGMENT_PROGRAM_ARB, 0);
		
      create_new_program(key, ctx, ctx->_TexEnvProgram);

      cache_item(&ctx->Texture.env_fp_cache, hash, key, ctx->_TexEnvProgram);
   } else {
      FREE(key);
      if (0) _mesa_printf("Found existing texenv program for key %x\n", hash);
   }
	
}

void _mesa_TexEnvProgramCacheDestroy( GLcontext *ctx )
{
   struct texenvprog_cache *a, *tmp;

   for (a = ctx->Texture.env_fp_cache; a; a = tmp) {
      tmp = a->next;
      FREE(a->key);
      FREE(a->data);
      FREE(a);
   }
}