interface for posit8, quire8, posit16, and quire16
authorBill Zorn <bill.zorn@gmail.com>
Wed, 25 Jul 2018 01:30:10 +0000 (18:30 -0700)
committerBill Zorn <bill.zorn@gmail.com>
Wed, 25 Jul 2018 01:30:10 +0000 (18:30 -0700)
cposit.pxd
posit.pyx

index b15cc1092ea5b4595539a171ce9b66172a109de2..913a96b40fe84ee1bf2e87540af6a43312418390 100644 (file)
@@ -5,5 +5,139 @@ cdef extern from 'SoftPosit/source/include/softposit.h':
     ctypedef struct posit8_t:
         pass
 
+    ctypedef struct posit16_t:
+        pass
+
+    ctypedef struct posit32_t:
+        pass
+
+    ctypedef struct posit64_t:
+        pass
+
+    ctypedef struct posit128_t:
+        pass
+
+    ctypedef struct quire8_t:
+        pass
+
+    ctypedef struct quire16_t:
+        pass
+
+    ctypedef struct quire32_t:
+        pass
+
+    # /*----------------------------------------------------------------------------
+    # | Integer-to-posit conversion routines.
+    # *----------------------------------------------------------------------------*/
+
+    posit8_t  ui32_to_p8( uint32_t );
+    posit16_t ui32_to_p16( uint32_t );
+    posit32_t ui32_to_p32( uint32_t );
+    posit64_t ui32_to_p64( uint32_t );
+
+    posit8_t  ui64_to_p8( uint64_t );
+    posit16_t ui64_to_p16( uint64_t );
+    posit32_t ui64_to_p32( uint64_t );
+    posit64_t ui64_to_p64( uint64_t );
+
+    posit8_t  i32_to_p8( int32_t );
+    posit16_t i32_to_p16( int32_t );
+    posit32_t i32_to_p32( int32_t );
+    posit64_t i32_to_p64( int32_t );
+
     posit8_t  i64_to_p8( int64_t );
-    int_fast64_t p8_to_i64( posit8_t);
+    posit16_t i64_to_p16( int64_t );
+    posit32_t i64_to_p32( int64_t );
+    posit64_t i64_to_p64( int64_t );
+
+    # /*----------------------------------------------------------------------------
+    # | 8-bit (quad-precision) posit operations.
+    # *----------------------------------------------------------------------------*/
+
+    bint isNaRP8UI( uint8_t );
+
+    uint_fast32_t p8_to_ui32( posit8_t );
+    uint_fast64_t p8_to_ui64( posit8_t );
+    int_fast32_t p8_to_i32( posit8_t );
+    int_fast64_t p8_to_i64( posit8_t );
+
+    posit16_t p8_to_p16( posit8_t );
+    #posit32_t p8_to_p32( posit8_t );
+    #posit64_t p8_to_p64( posit8_t );
+
+    posit8_t p8_roundToInt( posit8_t );
+    posit8_t p8_add( posit8_t, posit8_t );
+    posit8_t p8_sub( posit8_t, posit8_t );
+    posit8_t p8_mul( posit8_t, posit8_t );
+    posit8_t p8_mulAdd( posit8_t, posit8_t, posit8_t );
+    posit8_t p8_div( posit8_t, posit8_t );
+    posit8_t p8_sqrt( posit8_t );
+    bint p8_eq( posit8_t, posit8_t );
+    bint p8_le( posit8_t, posit8_t );
+    bint p8_lt( posit8_t, posit8_t );
+
+    # Quire 8
+    quire8_t q8_fdp_add( quire8_t, posit8_t, posit8_t );
+    quire8_t q8_fdp_sub( quire8_t, posit8_t, posit8_t );
+    posit8_t q8_to_p8( quire8_t );
+
+    bint isNaRQ8( quire8_t );
+    bint isQ8Zero( quire8_t );
+    quire8_t q8_clr( quire8_t );
+    quire8_t castQ8( uint32_t );
+    posit8_t castP8( uint8_t );
+    uint8_t castUI8( posit8_t );
+    posit8_t negP8( posit8_t );
+
+    # Helper
+    double convertP8ToDouble( posit8_t );
+    posit8_t convertDoubleToP8( double );
+
+    # /*----------------------------------------------------------------------------
+    # | 16-bit (half-precision) posit operations.
+    # *----------------------------------------------------------------------------*/
+
+    bint isNaRP16UI( uint16_t );
+    
+    uint_fast32_t p16_to_ui32( posit16_t );
+    uint_fast64_t p16_to_ui64( posit16_t );
+    int_fast32_t p16_to_i32( posit16_t);
+    int_fast64_t p16_to_i64( posit16_t );
+    posit8_t p16_to_p8( posit16_t );
+    #posit32_t p16_to_p32( posit16_t );
+    #posit64_t p16_to_p64( posit16_t );
+
+    posit16_t p16_roundToInt( posit16_t);
+    posit16_t p16_add( posit16_t, posit16_t );
+    posit16_t p16_sub( posit16_t, posit16_t );
+    posit16_t p16_mul( posit16_t, posit16_t );
+    posit16_t p16_mulAdd( posit16_t, posit16_t, posit16_t );
+    posit16_t p16_div( posit16_t, posit16_t );
+    posit16_t p16_sqrt( posit16_t );
+    bint p16_eq( posit16_t, posit16_t );
+    bint p16_le( posit16_t, posit16_t );
+    bint p16_lt( posit16_t, posit16_t );
+
+    #Quire 16
+    quire16_t q16_fdp_add( quire16_t, posit16_t, posit16_t );
+    quire16_t q16_fdp_sub( quire16_t, posit16_t, posit16_t );
+    #posit16_t convertQ16ToP16( quire16_t );
+    posit16_t q16_to_p16( quire16_t );
+
+    bint isNaRQ16( quire16_t );
+    bint isQ16Zero( quire16_t );
+    #quire16_t q16_TwosComplement( quire16_t );
+
+    #void printBinary( uint64_t*, int );
+    #void printHex( uint64_t );
+
+    quire16_t q16_clr( quire16_t );
+    quire16_t castQ16( uint16_t, uint16_t );
+    posit16_t castP16( uint16_t );
+    uint16_t castUI16( posit16_t );
+    posit16_t negP16( posit16_t );
+
+    # Helper
+    double convertP16ToDouble( posit16_t );
+    #posit16_t convertFloatToP16( float );
+    posit16_t convertDoubleToP16( double );
index 3bbeabfb7279d15d0fb804d448d769f3e0d04f50..ce3888c2094b87684b25b78fa6c1727996bf88a3 100644 (file)
--- a/posit.pyx
+++ b/posit.pyx
 cimport cposit
 
 cdef class Posit8:
+
+    # the wrapped posit value
     cdef cposit.posit8_t _c_posit
 
+    # factory function constructors that bypass __init__
+
+    @staticmethod
+    cdef Posit8 from_c_posit(cposit.posit8_t f):
+        """Factory function to create a Posit8 object directly from
+        a C posit8_t.
+        """
+        cdef Posit8 obj = Posit8.__new__(Posit8)
+        obj._c_posit = f
+        return obj
+
+    @staticmethod
+    def from_bits(int value):
+        """Factory function to create a Posit8 object from a bit pattern
+        represented as an integer.
+        """
+        cdef Posit8 obj = Posit8.__new__(Posit8)
+        obj._c_posit = cposit.castP8(value)
+        return obj
+
+    @staticmethod
+    def from_double(float value):
+        """Factory function to create a Posit8 object from a double.
+        """
+        cdef Posit8 obj = Posit8.__new__(Posit8)
+        obj._c_posit = cposit.convertDoubleToP8(value)
+        return obj
+
+    # convenience interface for use inside Python
+
+    def __init__(self, value):
+        if isinstance(value, int):
+            self._c_posit = cposit.castP8(value)
+        else:
+            f = float(value)
+            self._c_posit = cposit.convertDoubleToP8(f)
+
+    def __str__(self):
+        return repr(cposit.convertP8ToDouble(self._c_posit))
+
+    def __repr__(self):
+        return 'Posit8(' + repr(cposit.convertP8ToDouble(self._c_posit)) + ')'
+
+    cpdef get_bits(self):
+        cdef int u = cposit.castUI8(self._c_posit)
+        return u
+    bits = property(get_bits)
+
+    # arithmetic
+
+    cpdef round(self):
+        cdef cposit.posit8_t f = cposit.p8_roundToInt(self._c_posit)
+        return Posit8.from_c_posit(f)
+
+    def __round__(self, Posit8 other):
+        return self.roundToInt(other)
+
+    cpdef add(self, Posit8 other):
+        cdef cposit.posit8_t f = cposit.p8_add(self._c_posit, other._c_posit)
+        return Posit8.from_c_posit(f)
+
+    def __add__(self, Posit8 other):
+        return self.add(other)
+
+    cpdef sub(self, Posit8 other):
+        cdef cposit.posit8_t f = cposit.p8_sub(self._c_posit, other._c_posit)
+        return Posit8.from_c_posit(f)
+
+    def __sub__(self, Posit8 other):
+        return self.sub(other)
+
+    cpdef mul(self, Posit8 other):
+        cdef cposit.posit8_t f = cposit.p8_mul(self._c_posit, other._c_posit)
+        return Posit8.from_c_posit(f)
+
+    def __mul__(self, Posit8 other):
+        return self.mul(other)
+
+    cpdef fma(self, Posit8 a2, Posit8 a3):
+      cdef cposit.posit8_t f = cposit.p8_mulAdd(self._c_posit, a2._c_posit, a3._c_posit)
+      return Posit8.from_c_posit(f)
+
+    cpdef div(self, Posit8 other):
+        cdef cposit.posit8_t f = cposit.p8_div(self._c_posit, other._c_posit)
+        return Posit8.from_c_posit(f)
+
+    def __truediv__(self, Posit8 other):
+        return self.div(other)
+
+    cpdef sqrt(self):
+        cdef cposit.posit8_t f = cposit.p8_sqrt(self._c_posit)
+        return Posit8.from_c_posit(f)
+
+    # in-place arithmetic
+
+    cpdef iround(self):
+        self._c_posit = cposit.p8_roundToInt(self._c_posit)
+
+    cpdef iadd(self, Posit8 other):
+        self._c_posit = cposit.p8_add(self._c_posit, other._c_posit)
+
+    def __iadd__(self, Posit8 other):
+        self.iadd(other)
+        return self
+
+    cpdef isub(self, Posit8 other):
+        self._c_posit = cposit.p8_sub(self._c_posit, other._c_posit)
+
+    def __isub__(self, Posit8 other):
+        self.isub(other)
+        return self
+
+    cpdef imul(self, Posit8 other):
+        self._c_posit = cposit.p8_mul(self._c_posit, other._c_posit)
+
+    def __imul__(self, Posit8 other):
+        self.imul(other)
+        return self
+
+    cpdef ifma(self, Posit8 a2, Posit8 a3):
+        self._c_posit = cposit.p8_mulAdd(self._c_posit, a2._c_posit, a3._c_posit)
+
+    cpdef idiv(self, Posit8 other):
+        self._c_posit = cposit.p8_div(self._c_posit, other._c_posit)
+
+    def __itruediv__(self, Posit8 other):
+        self.idiv(other)
+        return self
+
+    cpdef isqrt(self):
+        self._c_posit = cposit.p8_sqrt(self._c_posit)
+
+    # comparison
+
+    cpdef eq(self, Posit8 other):
+        cdef bint b = cposit.p8_eq(self._c_posit, other._c_posit)
+        return b
+
+    cpdef le(self, Posit8 other):
+        cdef bint b = cposit.p8_le(self._c_posit, other._c_posit)
+        return b
+
+    cpdef lt(self, Posit8 other):
+        cdef bint b = cposit.p8_lt(self._c_posit, other._c_posit)
+        return b
+
+    def __lt__(self, Posit8 other):
+        return self.lt(other)
+
+    def __le__(self, Posit8 other):
+        return self.le(other)
+
+    def __eq__(self, Posit8 other):
+        return self.eq(other)
+
+    def __ne__(self, Posit8 other):
+        return not self.eq(other)
+
+    def __ge__(self, Posit8 other):
+        return other.le(self)
+
+    def __gt__(self, Posit8 other):
+        return other.lt(self)
+
+    # conversion to other posit types
+
+    cpdef to_p16(self):
+        cdef cposit.posit16_t f = cposit.p8_to_p16(self._c_posit)
+        return Posit16.from_c_posit(f)
+
+
+cdef class Quire8:
+
+    # the wrapped quire value
+    cdef cposit.quire8_t _c_quire
+
+    # limited interface for now
+
     def __cinit__(self):
-        self._c_posit = cposit.i64_to_p8(1)
+        cposit.q8_clr(self._c_quire)
+
+    def __str__(self):
+        return repr(cposit.convertP8ToDouble(cposit.q8_to_p8(self._c_quire)))
+
+    def __repr__(self):
+        return 'Quire8(' + repr(cposit.convertP8ToDouble(cposit.q8_to_p8(self._c_quire))) + ')'
+
+    cpdef fdp_add(self, Posit8 a2, Posit8 a3):
+        self._c_quire = cposit.q8_fdp_add(self._c_quire, a2._c_posit, a3._c_posit)
+
+    cpdef fdp_sub(self, Posit8 a2, Posit8 a3):
+        self._c_quire = cposit.q8_fdp_sub(self._c_quire, a2._c_posit, a3._c_posit)
+
+    cpdef get_p8(self):
+        cpdef cposit.posit8_t f = cposit.q8_to_p8(self._c_quire)
+        return Posit8.from_c_posit(f)
+    p8 = property(get_p8)
+
+
+cdef class Posit16:
+
+    # the wrapped posit value
+    cdef cposit.posit16_t _c_posit
+
+    # factory function constructors that bypass __init__
+
+    @staticmethod
+    cdef Posit16 from_c_posit(cposit.posit16_t f):
+        """Factory function to create a Posit16 object directly from
+        a C posit16_t.
+        """
+        cdef Posit16 obj = Posit16.__new__(Posit16)
+        obj._c_posit = f
+        return obj
+
+    @staticmethod
+    def from_bits(int value):
+        """Factory function to create a Posit16 object from a bit pattern
+        represented as an integer.
+        """
+        cdef Posit16 obj = Posit16.__new__(Posit16)
+        obj._c_posit = cposit.castP16(value)
+        return obj
+
+    @staticmethod
+    def from_double(float value):
+        """Factory function to create a Posit16 object from a double.
+        """
+        cdef Posit16 obj = Posit16.__new__(Posit16)
+        obj._c_posit = cposit.convertDoubleToP16(value)
+        return obj
+
+    # convenience interface for use inside Python
+
+    def __init__(self, value):
+        if isinstance(value, int):
+            self._c_posit = cposit.castP16(value)
+        else:
+            f = float(value)
+            self._c_posit = cposit.convertDoubleToP16(f)
+
+    def __str__(self):
+        return repr(cposit.convertP16ToDouble(self._c_posit))
+
+    def __repr__(self):
+        return 'Posit16(' + repr(cposit.convertP16ToDouble(self._c_posit)) + ')'
+
+    cpdef get_bits(self):
+        cdef int u = cposit.castUI16(self._c_posit)
+        return u
+    bits = property(get_bits)
+
+    # arithmetic
+
+    cpdef round(self):
+        cdef cposit.posit16_t f = cposit.p16_roundToInt(self._c_posit)
+        return Posit16.from_c_posit(f)
+
+    def __round__(self, Posit16 other):
+        return self.roundToInt(other)
+
+    cpdef add(self, Posit16 other):
+        cdef cposit.posit16_t f = cposit.p16_add(self._c_posit, other._c_posit)
+        return Posit16.from_c_posit(f)
+
+    def __add__(self, Posit16 other):
+        return self.add(other)
+
+    cpdef sub(self, Posit16 other):
+        cdef cposit.posit16_t f = cposit.p16_sub(self._c_posit, other._c_posit)
+        return Posit16.from_c_posit(f)
+
+    def __sub__(self, Posit16 other):
+        return self.sub(other)
+
+    cpdef mul(self, Posit16 other):
+        cdef cposit.posit16_t f = cposit.p16_mul(self._c_posit, other._c_posit)
+        return Posit16.from_c_posit(f)
+
+    def __mul__(self, Posit16 other):
+        return self.mul(other)
+
+    cpdef fma(self, Posit16 a2, Posit16 a3):
+      cdef cposit.posit16_t f = cposit.p16_mulAdd(self._c_posit, a2._c_posit, a3._c_posit)
+      return Posit16.from_c_posit(f)
+
+    cpdef div(self, Posit16 other):
+        cdef cposit.posit16_t f = cposit.p16_div(self._c_posit, other._c_posit)
+        return Posit16.from_c_posit(f)
+
+    def __truediv__(self, Posit16 other):
+        return self.div(other)
+
+    cpdef sqrt(self):
+        cdef cposit.posit16_t f = cposit.p16_sqrt(self._c_posit)
+        return Posit16.from_c_posit(f)
+
+    # in-place arithmetic
+
+    cpdef iround(self):
+        self._c_posit = cposit.p16_roundToInt(self._c_posit)
+
+    cpdef iadd(self, Posit16 other):
+        self._c_posit = cposit.p16_add(self._c_posit, other._c_posit)
+
+    def __iadd__(self, Posit16 other):
+        self.iadd(other)
+        return self
+
+    cpdef isub(self, Posit16 other):
+        self._c_posit = cposit.p16_sub(self._c_posit, other._c_posit)
+
+    def __isub__(self, Posit16 other):
+        self.isub(other)
+        return self
+
+    cpdef imul(self, Posit16 other):
+        self._c_posit = cposit.p16_mul(self._c_posit, other._c_posit)
+
+    def __imul__(self, Posit16 other):
+        self.imul(other)
+        return self
+
+    cpdef ifma(self, Posit16 a2, Posit16 a3):
+        self._c_posit = cposit.p16_mulAdd(self._c_posit, a2._c_posit, a3._c_posit)
+
+    cpdef idiv(self, Posit16 other):
+        self._c_posit = cposit.p16_div(self._c_posit, other._c_posit)
+
+    def __itruediv__(self, Posit16 other):
+        self.idiv(other)
+        return self
+
+    cpdef isqrt(self):
+        self._c_posit = cposit.p16_sqrt(self._c_posit)
+
+    # comparison
+
+    cpdef eq(self, Posit16 other):
+        cdef bint b = cposit.p16_eq(self._c_posit, other._c_posit)
+        return b
+
+    cpdef le(self, Posit16 other):
+        cdef bint b = cposit.p16_le(self._c_posit, other._c_posit)
+        return b
+
+    cpdef lt(self, Posit16 other):
+        cdef bint b = cposit.p16_lt(self._c_posit, other._c_posit)
+        return b
+
+    def __lt__(self, Posit16 other):
+        return self.lt(other)
+
+    def __le__(self, Posit16 other):
+        return self.le(other)
+
+    def __eq__(self, Posit16 other):
+        return self.eq(other)
+
+    def __ne__(self, Posit16 other):
+        return not self.eq(other)
+
+    def __ge__(self, Posit16 other):
+        return other.le(self)
+
+    def __gt__(self, Posit16 other):
+        return other.lt(self)
+
+    # conversion to other posit types
+
+    cpdef to_p8(self):
+        cdef cposit.posit8_t f = cposit.p16_to_p8(self._c_posit)
+        return Posit8.from_c_posit(f)
+
+
+cdef class Quire16:
+
+    # the wrapped quire value
+    cdef cposit.quire16_t _c_quire
+
+    # limited interface for now
+
+    def __cinit__(self):
+        cposit.q16_clr(self._c_quire)
+
+    def __str__(self):
+        return repr(cposit.convertP16ToDouble(cposit.q16_to_p16(self._c_quire)))
+
+    def __repr__(self):
+        return 'Quire16(' + repr(cposit.convertP16ToDouble(cposit.q16_to_p16(self._c_quire))) + ')'
+
+    cpdef fdp_add(self, Posit16 a2, Posit16 a3):
+        self._c_quire = cposit.q16_fdp_add(self._c_quire, a2._c_posit, a3._c_posit)
+
+    cpdef fdp_sub(self, Posit16 a2, Posit16 a3):
+        self._c_quire = cposit.q16_fdp_sub(self._c_quire, a2._c_posit, a3._c_posit)
 
-    cpdef show(self):
-        print(cposit.p8_to_i64(self._c_posit))
-    
+    cpdef get_p16(self):
+        cpdef cposit.posit16_t f = cposit.q16_to_p16(self._c_quire)
+        return Posit16.from_c_posit(f)
+    p16 = property(get_p16)