t_vb_arbprogram.c

来自「mesa-6.5-minigui源码」· C语言 代码 · 共 1,571 行 · 第 1/3 页

      if (result == REG_RES) {
	 op = cvp_next_instruction(cp);
	 *op = fixup;
      }
      break;
   }

   case OPCODE_END:
      break;

   default:
      result = cvp_choose_result( cp, &inst->DstReg, &fixup );
      nr_args = _mesa_num_inst_src_regs(inst->Opcode);
      for (i = 0; i < nr_args; i++) 
	 reg[i] = cvp_emit_arg( cp, &inst->SrcReg[i], REG_ARG0 + i );

      op = cvp_next_instruction(cp);
      op->alu.opcode = inst->Opcode;
      op->alu.file0 = reg[0].file;
      op->alu.idx0 = reg[0].idx;
      op->alu.file1 = reg[1].file;
      op->alu.idx1 = reg[1].idx;
      op->alu.dst = result;

      if (result == REG_RES) {
	 op = cvp_next_instruction(cp);
	 *op = fixup;
      }      	 
      break;
   }
}

static void free_tnl_data( struct vertex_program *program  )
{
   struct tnl_compiled_program *p = (struct tnl_compiled_program *) program->TnlData;
   if (p->compiled_func)
      _mesa_free((void *)p->compiled_func);
   _mesa_free(p);
   program->TnlData = NULL;
}

static void compile_vertex_program( struct vertex_program *program,
				    GLboolean try_codegen )
{ 
   struct compilation cp;
   struct tnl_compiled_program *p = CALLOC_STRUCT(tnl_compiled_program);
   GLuint i;

   if (program->TnlData) 
      free_tnl_data( program );
   
   program->TnlData = p;

   /* Initialize cp.  Note that ctx and VB aren't used in compilation
    * so we don't have to worry about statechanges:
    */
   _mesa_memset(&cp, 0, sizeof(cp));
   cp.csr = p->instructions;

   /* Compile instructions:
    */
   for (i = 0; i < program->Base.NumInstructions; i++) {
      cvp_emit_inst(&cp, &program->Base.Instructions[i]);
   }

   /* Finish up:
    */
   p->nr_instructions = cp.csr - p->instructions;

   /* Print/disassemble:
    */
   if (DISASSEM) {
      for (i = 0; i < p->nr_instructions; i++) {
	 _tnl_disassem_vba_insn(p->instructions[i]);
      }
      _mesa_printf("\n\n");
   }
   
#ifdef USE_SSE_ASM
   if (try_codegen)
      _tnl_sse_codegen_vertex_program(p);
#endif

}




/* ----------------------------------------------------------------------
 * Execution
 */
static void userclip( GLcontext *ctx,
		      GLvector4f *clip,
		      GLubyte *clipmask,
		      GLubyte *clipormask,
		      GLubyte *clipandmask )
{
   GLuint p;

   for (p = 0; p < ctx->Const.MaxClipPlanes; p++) {
      if (ctx->Transform.ClipPlanesEnabled & (1 << p)) {
	 GLuint nr, i;
	 const GLfloat a = ctx->Transform._ClipUserPlane[p][0];
	 const GLfloat b = ctx->Transform._ClipUserPlane[p][1];
	 const GLfloat c = ctx->Transform._ClipUserPlane[p][2];
	 const GLfloat d = ctx->Transform._ClipUserPlane[p][3];
         GLfloat *coord = (GLfloat *)clip->data;
         GLuint stride = clip->stride;
         GLuint count = clip->count;

	 for (nr = 0, i = 0 ; i < count ; i++) {
	    GLfloat dp = (coord[0] * a + 
			  coord[1] * b +
			  coord[2] * c +
			  coord[3] * d);

	    if (dp < 0) {
	       nr++;
	       clipmask[i] |= CLIP_USER_BIT;
	    }

	    STRIDE_F(coord, stride);
	 }

	 if (nr > 0) {
	    *clipormask |= CLIP_USER_BIT;
	    if (nr == count) {
	       *clipandmask |= CLIP_USER_BIT;
	       return;
	    }
	 }
      }
   }
}


static GLboolean
do_ndc_cliptest(GLcontext *ctx, struct arb_vp_machine *m)
{
   TNLcontext *tnl = TNL_CONTEXT(ctx);
   struct vertex_buffer *VB = m->VB;

   /* Cliptest and perspective divide.  Clip functions must clear
    * the clipmask.
    */
   m->ormask = 0;
   m->andmask = CLIP_ALL_BITS;

   if (tnl->NeedNdcCoords) {
      VB->NdcPtr =
         _mesa_clip_tab[VB->ClipPtr->size]( VB->ClipPtr,
                                            &m->ndcCoords,
                                            m->clipmask,
                                            &m->ormask,
                                            &m->andmask );
   }
   else {
      VB->NdcPtr = NULL;
      _mesa_clip_np_tab[VB->ClipPtr->size]( VB->ClipPtr,
                                            NULL,
                                            m->clipmask,
                                            &m->ormask,
                                            &m->andmask );
   }

   if (m->andmask) {
      /* All vertices are outside the frustum */
      return GL_FALSE;
   }

   /* Test userclip planes.  This contributes to VB->ClipMask.
    */
   if (ctx->Transform.ClipPlanesEnabled && !ctx->VertexProgram._Enabled) {
      userclip( ctx,
		VB->ClipPtr,
		m->clipmask,
		&m->ormask,
		&m->andmask );

      if (m->andmask) {
	 return GL_FALSE;
      }
   }

   VB->ClipAndMask = m->andmask;
   VB->ClipOrMask = m->ormask;
   VB->ClipMask = m->clipmask;

   return GL_TRUE;
}


static INLINE void call_func( struct tnl_compiled_program *p,
			      struct arb_vp_machine *m )
{
   p->compiled_func(m);
}

/**
 * Execute the given vertex program.  
 * 
 * TODO: Integrate the t_vertex.c code here, to build machine vertices
 * directly at this point.
 *
 * TODO: Eliminate the VB struct entirely and just use
 * struct arb_vertex_machine.
 */
static GLboolean
run_arb_vertex_program(GLcontext *ctx, struct tnl_pipeline_stage *stage)
{
   struct vertex_program *program;
   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
   struct arb_vp_machine *m = ARB_VP_MACHINE(stage);
   struct tnl_compiled_program *p;
   GLuint i, j;
   GLbitfield outputs;

   if (ctx->ShaderObjects._VertexShaderPresent)
      return GL_TRUE;

   program = (ctx->VertexProgram._Enabled ? ctx->VertexProgram.Current : ctx->_TnlProgram);
   if (!program || program->IsNVProgram)
      return GL_TRUE;   

   if (program->Base.Parameters) {
      _mesa_load_state_parameters(ctx, program->Base.Parameters);
   }   
   
   p = (struct tnl_compiled_program *)program->TnlData;
   assert(p);


   m->nr_inputs = m->nr_outputs = 0;

   for (i = 0; i < _TNL_ATTRIB_MAX; i++) {
      if (program->Base.InputsRead & (1<<i) ||
	  (i == VERT_ATTRIB_POS && program->IsPositionInvariant)) {
	 GLuint j = m->nr_inputs++;
	 m->input[j].idx = i;
	 m->input[j].data = (GLfloat *)m->VB->AttribPtr[i]->data;
	 m->input[j].stride = m->VB->AttribPtr[i]->stride;
	 m->input[j].size = m->VB->AttribPtr[i]->size;
	 ASSIGN_4V(m->File[0][REG_IN0 + i], 0, 0, 0, 1);
      }
   }     

   for (i = 0; i < VERT_RESULT_MAX; i++) {
      if (program->Base.OutputsWritten & (1 << i) ||
	  (i == VERT_RESULT_HPOS && program->IsPositionInvariant)) {
	 GLuint j = m->nr_outputs++;
	 m->output[j].idx = i;
	 m->output[j].data = (GLfloat *)m->attribs[i].data;
      }
   }     


   /* Run the actual program:
    */
   for (m->vtx_nr = 0; m->vtx_nr < VB->Count; m->vtx_nr++) {
      for (j = 0; j < m->nr_inputs; j++) {
	 GLuint idx = REG_IN0 + m->input[j].idx;
	 switch (m->input[j].size) {
	 case 4: m->File[0][idx][3] = m->input[j].data[3];
	 case 3: m->File[0][idx][2] = m->input[j].data[2];
	 case 2: m->File[0][idx][1] = m->input[j].data[1];
	 case 1: m->File[0][idx][0] = m->input[j].data[0];
	 }

	 STRIDE_F(m->input[j].data, m->input[j].stride);
      }


      if (p->compiled_func) {
	 call_func( p, m );
      }
      else {
	 for (j = 0; j < p->nr_instructions; j++) {
	    union instruction inst = p->instructions[j];	 
	    opcode_func[inst.alu.opcode]( m, inst );
	 }
      }

      /* If the program is position invariant, multiply the input position
       * by the MVP matrix and store in the vertex position result register.
       */
      if (program->IsPositionInvariant) {
	 TRANSFORM_POINT( m->File[0][REG_OUT0+0], 
			  ctx->_ModelProjectMatrix.m, 
			  m->File[0][REG_IN0+0]);
      }

      for (j = 0; j < m->nr_outputs; j++) {
	 GLuint idx = REG_OUT0 + m->output[j].idx;
	 m->output[j].data[0] = m->File[0][idx][0];
	 m->output[j].data[1] = m->File[0][idx][1];
	 m->output[j].data[2] = m->File[0][idx][2];
	 m->output[j].data[3] = m->File[0][idx][3];
	 m->output[j].data += 4;
      }

   }

   /* Setup the VB pointers so that the next pipeline stages get
    * their data from the right place (the program output arrays).
    *
    * TODO: 1) Have tnl use these RESULT values for outputs rather
    * than trying to shoe-horn inputs and outputs into one set of
    * values.
    *
    * TODO: 2) Integrate t_vertex.c so that we just go straight ahead
    * and build machine vertices here.
    */
   VB->ClipPtr = &m->attribs[VERT_RESULT_HPOS];
   VB->ClipPtr->count = VB->Count;

   outputs = program->Base.OutputsWritten;
   if (program->IsPositionInvariant) 
      outputs |= (1<<VERT_RESULT_HPOS);

   if (outputs & (1<<VERT_RESULT_COL0)) {
      VB->ColorPtr[0] = &m->attribs[VERT_RESULT_COL0];
      VB->AttribPtr[VERT_ATTRIB_COLOR0] = VB->ColorPtr[0];
   }

   if (outputs & (1<<VERT_RESULT_BFC0)) {
      VB->ColorPtr[1] = &m->attribs[VERT_RESULT_BFC0];
   }

   if (outputs & (1<<VERT_RESULT_COL1)) {
      VB->SecondaryColorPtr[0] = &m->attribs[VERT_RESULT_COL1];
      VB->AttribPtr[VERT_ATTRIB_COLOR1] = VB->SecondaryColorPtr[0];
   }

   if (outputs & (1<<VERT_RESULT_BFC1)) {
      VB->SecondaryColorPtr[1] = &m->attribs[VERT_RESULT_BFC1];
   }

   if (outputs & (1<<VERT_RESULT_FOGC)) {
      VB->FogCoordPtr = &m->attribs[VERT_RESULT_FOGC];
      VB->AttribPtr[VERT_ATTRIB_FOG] = VB->FogCoordPtr;
   }

   if (outputs & (1<<VERT_RESULT_PSIZ)) {
      VB->PointSizePtr = &m->attribs[VERT_RESULT_PSIZ];
      VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &m->attribs[VERT_RESULT_PSIZ];
   }

   for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
      if (outputs & (1<<(VERT_RESULT_TEX0+i))) {
	 VB->TexCoordPtr[i] = &m->attribs[VERT_RESULT_TEX0 + i];
	 VB->AttribPtr[VERT_ATTRIB_TEX0+i] = VB->TexCoordPtr[i];
      }
   }

#if 0
   for (i = 0; i < VB->Count; i++) {
      printf("Out %d: %f %f %f %f %f %f %f %f\n", i,
	     VEC_ELT(VB->ClipPtr, GLfloat, i)[0],
	     VEC_ELT(VB->ClipPtr, GLfloat, i)[1],
	     VEC_ELT(VB->ClipPtr, GLfloat, i)[2],
	     VEC_ELT(VB->ClipPtr, GLfloat, i)[3],
	     VEC_ELT(VB->TexCoordPtr[0], GLfloat, i)[0],
	     VEC_ELT(VB->TexCoordPtr[0], GLfloat, i)[1],
	     VEC_ELT(VB->TexCoordPtr[0], GLfloat, i)[2],
	     VEC_ELT(VB->TexCoordPtr[0], GLfloat, i)[3]);
   }
#endif

   /* Perform NDC and cliptest operations:
    */
   return do_ndc_cliptest(ctx, m);
}


static void
validate_vertex_program( GLcontext *ctx, struct tnl_pipeline_stage *stage )
{
   struct arb_vp_machine *m = ARB_VP_MACHINE(stage);
   struct vertex_program *program;

   if (ctx->ShaderObjects._VertexShaderPresent)
      return;

   program = (ctx->VertexProgram._Enabled ? ctx->VertexProgram.Current : 0);
   if (!program && ctx->_MaintainTnlProgram) {
      program = ctx->_TnlProgram;
   }

   if (program) {
      if (!program->TnlData)
	 compile_vertex_program( program, m->try_codegen );
      
      /* Grab the state GL state and put into registers:
       */
      m->File[FILE_LOCAL_PARAM] = program->Base.LocalParams;
      m->File[FILE_ENV_PARAM] = ctx->VertexProgram.Parameters;
      /* GL_NV_vertex_programs can't reference GL state */
      if (program->Base.Parameters)
         m->File[FILE_STATE_PARAM] = program->Base.Parameters->ParameterValues;
      else
         m->File[FILE_STATE_PARAM] = NULL;
   }
}







/**
 * Called the first time stage->run is called.  In effect, don't
 * allocate data until the first time the stage is run.
 */
static GLboolean init_vertex_program( GLcontext *ctx,
				      struct tnl_pipeline_stage *stage )
{
   TNLcontext *tnl = TNL_CONTEXT(ctx);
   struct vertex_buffer *VB = &(tnl->vb);
   struct arb_vp_machine *m;
   const GLuint size = VB->Size;
   GLuint i;

   stage->privatePtr = _mesa_calloc(sizeof(*m));
   m = ARB_VP_MACHINE(stage);
   if (!m)
      return GL_FALSE;

   /* arb_vertex_machine struct should subsume the VB:
    */
   m->VB = VB;

   m->File[0] = (GLfloat(*)[4])ALIGN_MALLOC(REG_MAX * sizeof(GLfloat) * 4, 16);

   /* Initialize regs where necessary:
    */
   ASSIGN_4V(m->File[0][REG_ID], 0, 0, 0, 1);
   ASSIGN_4V(m->File[0][REG_ONES], 1, 1, 1, 1);
   ASSIGN_4V(m->File[0][REG_SWZ], -1, 1, 0, 0);
   ASSIGN_4V(m->File[0][REG_NEG], -1, -1, -1, -1);
   ASSIGN_4V(m->File[0][REG_LIT], 1, 0, 0, 1);
   ASSIGN_4V(m->File[0][REG_LIT2], 1, .5, .2, 1); /* debug value */

   if (_mesa_getenv("MESA_EXPERIMENTAL"))
      m->try_codegen = GL_TRUE;

   /* Allocate arrays of vertex output values */
   for (i = 0; i < VERT_RESULT_MAX; i++) {
      _mesa_vector4f_alloc( &m->attribs[i], 0, size, 32 );
      m->attribs[i].size = 4;
   }

   /* a few other misc allocations */
   _mesa_vector4f_alloc( &m->ndcCoords, 0, size, 32 );
   m->clipmask = (GLubyte *) ALIGN_MALLOC(sizeof(GLubyte)*size, 32 );

   if (ctx->_MaintainTnlProgram)
      _mesa_allow_light_in_model( ctx, GL_FALSE );

   m->fpucntl_rnd_neg = RND_NEG_FPU; /* const value */
   m->fpucntl_restore = RESTORE_FPU; /* const value */

   return GL_TRUE;
}




/**
 * Destructor for this pipeline stage.
 */
static void dtr( struct tnl_pipeline_stage *stage )
{
   struct arb_vp_machine *m = ARB_VP_MACHINE(stage);

   if (m) {
      GLuint i;

      /* free the vertex program result arrays */
      for (i = 0; i < VERT_RESULT_MAX; i++)
         _mesa_vector4f_free( &m->attribs[i] );

      /* free misc arrays */
      _mesa_vector4f_free( &m->ndcCoords );
      ALIGN_FREE( m->clipmask );
      ALIGN_FREE( m->File[0] );

      _mesa_free( m );
      stage->privatePtr = NULL;
   }
}

/**
 * Public description of this pipeline stage.
 */
const struct tnl_pipeline_stage _tnl_arb_vertex_program_stage =
{
   "vertex-program",
   NULL,			/* private_data */
   init_vertex_program,		/* create */
   dtr,				/* destroy */
   validate_vertex_program,	/* validate */
   run_arb_vertex_program	/* run */
};


/**
 * Called via ctx->Driver.ProgramStringNotify() after a new vertex program
 * string has been parsed.
 */
void
_tnl_program_string(GLcontext *ctx, GLenum target, struct program *program)
{
   if (program->Target == GL_VERTEX_PROGRAM_ARB) {
      /* free any existing tnl data hanging off the program */
      struct vertex_program *vprog = (struct vertex_program *) program;
      if (vprog->TnlData) {
         free_tnl_data(vprog);
      }
   }
}