To predict the outcome one have to know the representation of float (IEEE 754) and the big/little endian issues which could be different between floats and ints.
printf("%d\n", a) does promote the value a to a double first (promotion rules) and then either the first or last part i(assuming sizeof(double) == 2*sizeof(int)) is shown depending on little or big endian architecture.
Reinterpreting an object as an incompatible object is undefined. int and float are incompatible.
C Standard also specifically states that using an incorrect printf format specifier will result in undefined behavior.
On x86-64, it displays a random value each time, which actually happens to be some pointer (randomised by ASLR). This is because register passing is used on x86-64, and the the double is stored into a floating point register, but %d reads from a different register.
printf("%d\n", * (int*)&a); produces the number 1095237632 for both x86 and x86-64.
Also these are C puzzles and casting pointers to incompatible types and dereferencing them is undefined behavior. Using memcpy would indeed fix this according to e.g. [1].
[1] http://blog.llvm.org/2011/05/what-every-c-programmer-should-... (section "Violating Type Rules")