Adapt new_gen_from_real_mpfr_element to pari 2.17 changes in precision

src/sage/libs/pari/convert_sage_real_mpfr.pyx

@@ -28,7 +28,7 @@ cpdef Gen new_gen_from_real_mpfr_element(RealNumber self):
 
     # We round up the precision to the nearest multiple of wordsize.
     cdef int rounded_prec
-    rounded_prec = (self.prec() + wordsize - 1) & ~(wordsize - 1)
+    rounded_prec = nbits2prec(self.prec())
 
     # Yes, assigning to self works fine, even in Cython.
     if rounded_prec > prec:
@@ -48,7 +48,7 @@ cpdef Gen new_gen_from_real_mpfr_element(RealNumber self):
         exponent = mpfr_get_z_exp(mantissa, self.value)
 
         # Create a PARI REAL
-        pari_float = cgetr(2 + rounded_prec / wordsize)
+        pari_float = cgetr(rounded_prec)
         pari_float[1] = evalexpo(exponent + rounded_prec - 1) + evalsigne(mpfr_sgn(self.value))
         mpz_export(&pari_float[2], NULL, 1, wordsize // 8, 0, 0, mantissa)
         mpz_clear(mantissa)