r200_vertprog.c

Mesa is an open-source implementation of the OpenGL specification - a system for rendering interacti

		(!src[0].NegateBase) ? VSF_FLAG_ALL : VSF_FLAG_NONE) | (src[0].RelAddr << 4);
	 o_inst->src2 = UNUSED_SRC_1;
	 goto next;

      case OPCODE_FLR:
      /* FRC TMP 0.X Y Z W PARAM 0{} {X Y Z W} 
         ADD RESULT 1.X Y Z W PARAM 0{} {X Y Z W} TMP 0{X Y Z W } {X Y Z W} neg Xneg Yneg Zneg W */

	 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_FRC,
	    (u_temp_i << R200_VPI_OUT_REG_INDEX_SHIFT) | R200_VSF_OUT_CLASS_TMP,
	    t_dst_mask(dst.WriteMask));

	 o_inst->src0 = t_src(vp, &src[0]);
	 o_inst->src1 = UNUSED_SRC_0;
	 o_inst->src2 = UNUSED_SRC_1;
	 o_inst++;

	 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_ADD, t_dst(&dst),
		t_dst_mask(dst.WriteMask));

	 o_inst->src0 = t_src(vp, &src[0]);
	 o_inst->src1 = MAKE_VSF_SOURCE(u_temp_i,
		VSF_IN_COMPONENT_X,
		VSF_IN_COMPONENT_Y,
		VSF_IN_COMPONENT_Z,
		VSF_IN_COMPONENT_W,
		VSF_IN_CLASS_TMP,
		/* Not 100% sure about this */
		(!src[0].NegateBase) ? VSF_FLAG_ALL : VSF_FLAG_NONE/*VSF_FLAG_ALL*/);

	 o_inst->src2 = UNUSED_SRC_0;
	 u_temp_i--;
	 goto next;

      case OPCODE_XPD:
	 /* mul r0, r1.yzxw, r2.zxyw
	    mad r0, -r2.yzxw, r1.zxyw, r0
	  */
	 hw_op=(src[0].File == PROGRAM_TEMPORARY &&
	    src[1].File == PROGRAM_TEMPORARY &&
	    (((src[0].RelAddr << 8) | src[0].Index) != ((src[1].RelAddr << 8) | src[1].Index))) ?
	    R200_VPI_OUT_OP_MAD_2 : R200_VPI_OUT_OP_MAD;

	 o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_MUL,
	    (u_temp_i << R200_VPI_OUT_REG_INDEX_SHIFT) | R200_VSF_OUT_CLASS_TMP,
	    t_dst_mask(dst.WriteMask));

	 o_inst->src0 = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]),
		t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y
		t_swizzle(GET_SWZ(src[0].Swizzle, 2)), // z
		t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x
		t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w
		t_src_class(src[0].File),
		src[0].NegateBase) | (src[0].RelAddr << 4);

	 o_inst->src1 = MAKE_VSF_SOURCE(t_src_index(vp, &src[1]),
		t_swizzle(GET_SWZ(src[1].Swizzle, 2)), // z
		t_swizzle(GET_SWZ(src[1].Swizzle, 0)), // x
		t_swizzle(GET_SWZ(src[1].Swizzle, 1)), // y
		t_swizzle(GET_SWZ(src[1].Swizzle, 3)), // w
		t_src_class(src[1].File),
		src[1].NegateBase) | (src[1].RelAddr << 4);

	 o_inst->src2 = UNUSED_SRC_1;
	 o_inst++;
	 u_temp_i--;

	 o_inst->op = MAKE_VSF_OP(hw_op, t_dst(&dst),
		t_dst_mask(dst.WriteMask));

	 o_inst->src0 = MAKE_VSF_SOURCE(t_src_index(vp, &src[1]),
		t_swizzle(GET_SWZ(src[1].Swizzle, 1)), // y
		t_swizzle(GET_SWZ(src[1].Swizzle, 2)), // z
		t_swizzle(GET_SWZ(src[1].Swizzle, 0)), // x
		t_swizzle(GET_SWZ(src[1].Swizzle, 3)), // w
		t_src_class(src[1].File),
		(!src[1].NegateBase) ? VSF_FLAG_ALL : VSF_FLAG_NONE) | (src[1].RelAddr << 4);

	 o_inst->src1 = MAKE_VSF_SOURCE(t_src_index(vp, &src[0]),
		t_swizzle(GET_SWZ(src[0].Swizzle, 2)), // z
		t_swizzle(GET_SWZ(src[0].Swizzle, 0)), // x
		t_swizzle(GET_SWZ(src[0].Swizzle, 1)), // y
		t_swizzle(GET_SWZ(src[0].Swizzle, 3)), // w
		t_src_class(src[0].File),
		src[0].NegateBase) | (src[0].RelAddr << 4);

	 o_inst->src2 = MAKE_VSF_SOURCE(u_temp_i+1,
		VSF_IN_COMPONENT_X,
		VSF_IN_COMPONENT_Y,
		VSF_IN_COMPONENT_Z,
		VSF_IN_COMPONENT_W,
		VSF_IN_CLASS_TMP,
		VSF_FLAG_NONE);
	 goto next;

      case OPCODE_END:
	 assert(0);
      default:
	 break;
      }

      o_inst->op = MAKE_VSF_OP(t_opcode(vpi->Opcode), t_dst(&dst),
	    t_dst_mask(dst.WriteMask));

      if(are_srcs_scalar){
	 switch(operands){
	    case 1:
		o_inst->src0 = t_src_scalar(vp, &src[0]);
		o_inst->src1 = UNUSED_SRC_0;
		o_inst->src2 = UNUSED_SRC_1;
	    break;

	    case 2:
		o_inst->src0 = t_src_scalar(vp, &src[0]);
		o_inst->src1 = t_src_scalar(vp, &src[1]);
		o_inst->src2 = UNUSED_SRC_1;
	    break;

	    case 3:
		o_inst->src0 = t_src_scalar(vp, &src[0]);
		o_inst->src1 = t_src_scalar(vp, &src[1]);
		o_inst->src2 = t_src_scalar(vp, &src[2]);
	    break;

	    default:
		fprintf(stderr, "illegal number of operands %lu\n", operands);
		exit(-1);
	    break;
	 }
      } else {
	 switch(operands){
	    case 1:
		o_inst->src0 = t_src(vp, &src[0]);
		o_inst->src1 = UNUSED_SRC_0;
		o_inst->src2 = UNUSED_SRC_1;
	    break;

	    case 2:
		o_inst->src0 = t_src(vp, &src[0]);
		o_inst->src1 = t_src(vp, &src[1]);
		o_inst->src2 = UNUSED_SRC_1;
	    break;

	    case 3:
		o_inst->src0 = t_src(vp, &src[0]);
		o_inst->src1 = t_src(vp, &src[1]);
		o_inst->src2 = t_src(vp, &src[2]);
	    break;

	    default:
		fprintf(stderr, "illegal number of operands %lu\n", operands);
		exit(-1);
	    break;
	 }
      }
      next:

      if (dofogfix) {
	 o_inst++;
	 if (vp->fogmode == GL_EXP) {
	    o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_MUL,
		(fog_temp_i << R200_VPI_OUT_REG_INDEX_SHIFT) | R200_VSF_OUT_CLASS_TMP,
		VSF_FLAG_X);
	    o_inst->src0 = EASY_VSF_SOURCE(fog_temp_i, X, X, X, X, TMP, NONE);
	    o_inst->src1 = EASY_VSF_SOURCE(vp->fogpidx, X, X, X, X, PARAM, NONE);
	    o_inst->src2 = UNUSED_SRC_1;
	    o_inst++;
	    o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_EXP_E,
		R200_VSF_OUT_CLASS_RESULT_FOGC,
		VSF_FLAG_X);
	    o_inst->src0 = EASY_VSF_SOURCE(fog_temp_i, X, X, X, X, TMP, ALL);
	    o_inst->src1 = UNUSED_SRC_0;
	    o_inst->src2 = UNUSED_SRC_1;
	 }
	 else if (vp->fogmode == GL_EXP2) {
	    o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_MUL,
		(fog_temp_i << R200_VPI_OUT_REG_INDEX_SHIFT) | R200_VSF_OUT_CLASS_TMP,
		VSF_FLAG_X);
	    o_inst->src0 = EASY_VSF_SOURCE(fog_temp_i, X, X, X, X, TMP, NONE);
	    o_inst->src1 = EASY_VSF_SOURCE(vp->fogpidx, X, X, X, X, PARAM, NONE);
	    o_inst->src2 = UNUSED_SRC_1;
	    o_inst++;
	    o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_MUL,
		(fog_temp_i << R200_VPI_OUT_REG_INDEX_SHIFT) | R200_VSF_OUT_CLASS_TMP,
		VSF_FLAG_X);
	    o_inst->src0 = EASY_VSF_SOURCE(fog_temp_i, X, X, X, X, TMP, NONE);
	    o_inst->src1 = EASY_VSF_SOURCE(fog_temp_i, X, X, X, X, TMP, NONE);
	    o_inst->src2 = UNUSED_SRC_1;
	    o_inst++;
	    o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_EXP_E,
		R200_VSF_OUT_CLASS_RESULT_FOGC,
		VSF_FLAG_X);
	    o_inst->src0 = EASY_VSF_SOURCE(fog_temp_i, X, X, X, X, TMP, ALL);
	    o_inst->src1 = UNUSED_SRC_0;
	    o_inst->src2 = UNUSED_SRC_1;
	 }
	 else { /* fogmode == GL_LINEAR */
		/* could do that with single op (dot) if using params like
		   with fixed function pipeline fog */
	    o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_ADD,
		(fog_temp_i << R200_VPI_OUT_REG_INDEX_SHIFT) | R200_VSF_OUT_CLASS_TMP,
		VSF_FLAG_X);
	    o_inst->src0 = EASY_VSF_SOURCE(fog_temp_i, X, X, X, X, TMP, ALL);
	    o_inst->src1 = EASY_VSF_SOURCE(vp->fogpidx, Z, Z, Z, Z, PARAM, NONE);
	    o_inst->src2 = UNUSED_SRC_1;
	    o_inst++;
	    o_inst->op = MAKE_VSF_OP(R200_VPI_OUT_OP_MUL,
		R200_VSF_OUT_CLASS_RESULT_FOGC,
		VSF_FLAG_X);
	    o_inst->src0 = EASY_VSF_SOURCE(fog_temp_i, X, X, X, X, TMP, NONE);
	    o_inst->src1 = EASY_VSF_SOURCE(vp->fogpidx, W, W, W, W, PARAM, NONE);
	    o_inst->src2 = UNUSED_SRC_1;

	 }
         dofogfix = 0;
      }

      u_temp_used = (R200_VSF_MAX_TEMPS - 1) - u_temp_i;
      if (mesa_vp->Base.NumNativeTemporaries <
	 (mesa_vp->Base.NumTemporaries + u_temp_used)) {
	 mesa_vp->Base.NumNativeTemporaries =
	    mesa_vp->Base.NumTemporaries + u_temp_used;
      }
      if ((mesa_vp->Base.NumTemporaries + u_temp_used) > R200_VSF_MAX_TEMPS) {
	 if (R200_DEBUG & DEBUG_FALLBACKS) {
	    fprintf(stderr, "Ran out of temps, num temps %d, us %d\n", mesa_vp->Base.NumTemporaries, u_temp_used);
	 }
	 return GL_FALSE;
      }
      u_temp_i = R200_VSF_MAX_TEMPS - 1;
      if(o_inst - vp->instr >= R200_VSF_MAX_INST) {
	 mesa_vp->Base.NumNativeInstructions = 129;
	 if (R200_DEBUG & DEBUG_FALLBACKS) {
	    fprintf(stderr, "more than 128 native instructions\n");
	 }
	 return GL_FALSE;
      }
      if ((o_inst->op & R200_VSF_OUT_CLASS_MASK) == R200_VSF_OUT_CLASS_RESULT_POS) {
	 vp->pos_end = (o_inst - vp->instr);
      }
   }

   vp->native = GL_TRUE;
   mesa_vp->Base.NumNativeInstructions = (o_inst - vp->instr);
#if 0
   fprintf(stderr, "hw program:\n");
   for(i=0; i < vp->program.length; i++)
      fprintf(stderr, "%08x\n", vp->instr[i]);
#endif
   return GL_TRUE;
}

void r200SetupVertexProg( GLcontext *ctx ) {
   r200ContextPtr rmesa = R200_CONTEXT(ctx);
   struct r200_vertex_program *vp = (struct r200_vertex_program *)ctx->VertexProgram.Current;
   GLboolean fallback;
   GLint i;

   if (!vp->translated || (ctx->Fog.Enabled && ctx->Fog.Mode != vp->fogmode)) {
      rmesa->curr_vp_hw = NULL;
      r200_translate_vertex_program(ctx, vp);
   }
   /* could optimize setting up vertex progs away for non-tcl hw */
   fallback = !(vp->native && r200VertexProgUpdateParams(ctx, vp) &&
      rmesa->r200Screen->drmSupportsVertexProgram);
   TCL_FALLBACK(ctx, R200_TCL_FALLBACK_VERTEX_PROGRAM, fallback);
   if (rmesa->TclFallback) return;

   R200_STATECHANGE( rmesa, vap );
   /* FIXME: fglrx sets R200_VAP_SINGLE_BUF_STATE_ENABLE too. Do we need it?
             maybe only when using more than 64 inst / 96 param? */
   rmesa->hw.vap.cmd[VAP_SE_VAP_CNTL] |= R200_VAP_PROG_VTX_SHADER_ENABLE /*| R200_VAP_SINGLE_BUF_STATE_ENABLE*/;

   R200_STATECHANGE( rmesa, pvs );

   rmesa->hw.pvs.cmd[PVS_CNTL_1] = (0 << R200_PVS_CNTL_1_PROGRAM_START_SHIFT) |
      ((vp->mesa_program.Base.NumNativeInstructions - 1) << R200_PVS_CNTL_1_PROGRAM_END_SHIFT) |
      (vp->pos_end << R200_PVS_CNTL_1_POS_END_SHIFT);
   rmesa->hw.pvs.cmd[PVS_CNTL_2] = (0 << R200_PVS_CNTL_2_PARAM_OFFSET_SHIFT) |
      (vp->mesa_program.Base.NumNativeParameters << R200_PVS_CNTL_2_PARAM_COUNT_SHIFT);

   /* maybe user clip planes just work with vertex progs... untested */
   if (ctx->Transform.ClipPlanesEnabled) {
      R200_STATECHANGE( rmesa, tcl );
      if (vp->mesa_program.IsPositionInvariant) {
	 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] |= (ctx->Transform.ClipPlanesEnabled << 2);
      }
      else {
	 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] &= ~(0xfc);
      }
   }

   if (vp != rmesa->curr_vp_hw) {
      GLuint count = vp->mesa_program.Base.NumNativeInstructions;
      drm_radeon_cmd_header_t tmp;

      R200_STATECHANGE( rmesa, vpi[0] );
      R200_STATECHANGE( rmesa, vpi[1] );

      /* FIXME: what about using a memcopy... */
      for (i = 0; (i < 64) && i < count; i++) {
	 rmesa->hw.vpi[0].cmd[VPI_OPDST_0 + 4 * i] = vp->instr[i].op;
	 rmesa->hw.vpi[0].cmd[VPI_SRC0_0 + 4 * i] = vp->instr[i].src0;
	 rmesa->hw.vpi[0].cmd[VPI_SRC1_0 + 4 * i] = vp->instr[i].src1;
	 rmesa->hw.vpi[0].cmd[VPI_SRC2_0 + 4 * i] = vp->instr[i].src2;
      }
      /* hack up the cmd_size so not the whole state atom is emitted always.
         This may require some more thought, we may emit half progs on lost state, but
         hopefully it won't matter?
         WARNING: must not use R200_DB_STATECHANGE, this will produce bogus (and rejected)
         packet emits (due to the mismatched cmd_size and count in cmd/last_cmd) */
      rmesa->hw.vpi[0].cmd_size = 1 + 4 * ((count > 64) ? 64 : count);
      tmp.i = rmesa->hw.vpi[0].cmd[VPI_CMD_0];
      tmp.veclinear.count = (count > 64) ? 64 : count;
      rmesa->hw.vpi[0].cmd[VPI_CMD_0] = tmp.i;
      if (count > 64) {
	 for (i = 0; i < (count - 64); i++) {
	    rmesa->hw.vpi[1].cmd[VPI_OPDST_0 + 4 * i] = vp->instr[i + 64].op;
	    rmesa->hw.vpi[1].cmd[VPI_SRC0_0 + 4 * i] = vp->instr[i + 64].src0;
	    rmesa->hw.vpi[1].cmd[VPI_SRC1_0 + 4 * i] = vp->instr[i + 64].src1;
	    rmesa->hw.vpi[1].cmd[VPI_SRC2_0 + 4 * i] = vp->instr[i + 64].src2;
	 }
	 rmesa->hw.vpi[1].cmd_size = 1 + 4 * (count - 64);
	 tmp.i = rmesa->hw.vpi[1].cmd[VPI_CMD_0];
	 tmp.veclinear.count = count - 64;
	 rmesa->hw.vpi[1].cmd[VPI_CMD_0] = tmp.i;
      }
      rmesa->curr_vp_hw = vp;
   }
}


static void
r200BindProgram(GLcontext *ctx, GLenum target, struct gl_program *prog)
{
   r200ContextPtr rmesa = R200_CONTEXT(ctx);

   switch(target){
   case GL_VERTEX_PROGRAM_ARB:
      rmesa->curr_vp_hw = NULL;
      break;
   default:
      _mesa_problem(ctx, "Target not supported yet!");
      break;
   }
}

static struct gl_program *
r200NewProgram(GLcontext *ctx, GLenum target, GLuint id)
{
   struct r200_vertex_program *vp;

   switch(target){
   case GL_VERTEX_PROGRAM_ARB:
      vp = CALLOC_STRUCT(r200_vertex_program);
      return _mesa_init_vertex_program(ctx, &vp->mesa_program, target, id);
   case GL_FRAGMENT_PROGRAM_ARB:
   case GL_FRAGMENT_PROGRAM_NV:
      return _mesa_init_fragment_program( ctx, CALLOC_STRUCT(gl_fragment_program), target, id );
   default:
      _mesa_problem(ctx, "Bad target in r200NewProgram");
   }
   return NULL;	
}


static void
r200DeleteProgram(GLcontext *ctx, struct gl_program *prog)
{
   _mesa_delete_program(ctx, prog);
}

static void
r200ProgramStringNotify(GLcontext *ctx, GLenum target, struct gl_program *prog)
{
   struct r200_vertex_program *vp = (void *)prog;
   r200ContextPtr rmesa = R200_CONTEXT(ctx);

   switch(target) {
   case GL_VERTEX_PROGRAM_ARB:
      vp->translated = GL_FALSE;
      vp->fogpidx = 0;
/*      memset(&vp->translated, 0, sizeof(struct r200_vertex_program) - sizeof(struct gl_vertex_program));*/
      r200_translate_vertex_program(ctx, vp);
      rmesa->curr_vp_hw = NULL;
      break;
   case GL_FRAGMENT_SHADER_ATI:
      rmesa->afs_loaded = NULL;
      break;
   }
   /* need this for tcl fallbacks */
   _tnl_program_string(ctx, target, prog);
}

static GLboolean
r200IsProgramNative(GLcontext *ctx, GLenum target, struct gl_program *prog)
{
   struct r200_vertex_program *vp = (void *)prog;

   switch(target){
   case GL_VERTEX_STATE_PROGRAM_NV:
   case GL_VERTEX_PROGRAM_ARB:
      if (!vp->translated) {
	 r200_translate_vertex_program(ctx, vp);
      }
     /* does not take parameters etc. into account */
      return vp->native;
   default:
      _mesa_problem(ctx, "Bad target in r200NewProgram");
   }
   return 0;
}

void r200InitShaderFuncs(struct dd_function_table *functions)
{
   functions->NewProgram = r200NewProgram;
   functions->BindProgram = r200BindProgram;
   functions->DeleteProgram = r200DeleteProgram;
   functions->ProgramStringNotify = r200ProgramStringNotify;
   functions->IsProgramNative = r200IsProgramNative;
}