Generic resolution inside structure constructor

Inside a structure constructor (at least), resolution of a generic with a name that is the same as that of an intrinsic procedure does not appear to consider the specific procedures provided by the program.

If I use a temporary for the expression, the generic name is resolved correctly in the expression that sets the value of the temporary.

MODULE Direct2D
  IMPLICIT NONE

  PRIVATE

  PUBLIC :: SomeAPI
  PUBLIC :: Scale

  INTEGER, PARAMETER :: FLOAT = KIND(1.0)


  TYPE, BIND(C), PUBLIC :: D2D1_POINT_2F
    REAL(FLOAT) :: x = 0.0_FLOAT
    REAL(FLOAT) :: y = 0.0_FLOAT
  END TYPE D2D1_POINT_2F

  TYPE, BIND(C), PUBLIC :: D2D1_MATRIX_3x2_F
    REAL(FLOAT) :: data(2,3)
  END TYPE D2D1_MATRIX_3x2_F

  TYPE, BIND(C), PUBLIC :: D2D1_BRUSH_PROPERTIES
    REAL(FLOAT) :: opacity = 1.0_FLOAT
    TYPE(D2D1_MATRIX_3x2_F) :: transform
  END TYPE D2D1_BRUSH_PROPERTIES

  INTERFACE Scale
    MODULE PROCEDURE :: Scale_point
  END INTERFACE Scale
CONTAINS
  SUBROUTINE SomeAPI(brush_properties)
    TYPE(D2D1_BRUSH_PROPERTIES), INTENT(IN) :: brush_properties
    !...
  END SUBROUTINE SomeAPI

  ELEMENTAL FUNCTION Scale_point(x, y, center) RESULT(matrix)
    REAL(FLOAT), INTENT(IN) :: x
    REAL(FLOAT), INTENT(IN) :: y
    TYPE(D2D1_POINT_2F), INTENT(IN) :: center
    TYPE(D2D1_MATRIX_3X2_F) :: matrix
    !***************************************************************************
    matrix%data = RESHAPE(  &
        [ x, 0.0_FLOAT, 0.0_FLOAT, y,  &
          center.x - x * center.x, center.y - y * center.y ],  &
        [2, 3 ] )
  END FUNCTION Scale_point
END MODULE Direct2D

PROGRAM scale_bug
  USE Direct2D
  IMPLICIT NONE
  TYPE(D2D1_MATRIX_3X2_F) :: txfrm
  ! We get errors with:
  CALL SomeAPI(  &
      D2D1_BRUSH_PROPERTIES(  &
        TRANSFORM=Scale(1.0, 2.0, D2D1_POINT_2F()) ) )
  ! But not with:
  txfrm = Scale(1.0, 2.0, D2D1_POINT_2F())
  CALL SomeAPI(  &
      D2D1_BRUSH_PROPERTIES(  &
        TRANSFORM=txfrm ) )
END PROGRAM scale_bug

>ifort /check:all /warn:all /standard-semantics "2016-05-03 ScaleBug.f90"
Intel(R) Visual Fortran Intel(R) 64 Compiler for applications running on Intel(R) 64, Version 16.0.2.180 Build 20160204
Copyright (C) 1985-2016 Intel Corporation.  All rights reserved.

2016-05-03 ScaleBug.f90(30): remark #7712: This variable has not been used.   [BRUSH_PROPERTIES]
  SUBROUTINE SomeAPI(brush_properties)
---------------------^
2016-05-03 ScaleBug.f90(55): error #6506: This intrinsic procedure reference contains too many arguments.   [SCALE]
        TRANSFORM=Scale(1.0, 2.0, D2D1_POINT_2F()) ) )
------------------^
2016-05-03 ScaleBug.f90(55): error #6303: The assignment operation or the binary expression operation is invalid for the data types of the two operands.
        TRANSFORM=Scale(1.0, 2.0, D2D1_POINT_2F()) ) )
------------------^
2016-05-03 ScaleBug.f90(55): error #8001: The type of the element in a structure-constructor differs from the type of the component in the derived-type-def.
        TRANSFORM=Scale(1.0, 2.0, D2D1_POINT_2F()) ) )
------------------^
compilation aborted for 2016-05-03 ScaleBug.f90 (code 1)