I asked my C compiler (GCC 4.8.5/RH7/IA64) about this:
int32_t a = 65536;
int32_t b = 0x7ffffff;
int64_t p;
p = a*b;
and got this
32 bit init: c7 45 fc 00 00 01 00 movl $0x10000,-0x4(%rbp)
32 bit init: c7 45 f8 ff ff ff 07 movl $0x7ffffff,-0x8(%rbp)
32 bit load: 8b 45 fc mov -0x4(%rbp),%eax
32 bit mult: 0f af 45 f8 imul -0x8(%rbp),%eax
32->64 conv: 48 98 cltq
64 bit stor: 48 89 45 f0 mov %rax,-0x10(%rbp)
and then about this:
int32_t a = 65536;
int32_t b = 0x7ffffff;
int64_t p;
p = (int64_t)a*b;
32 bit init: c7 45 fc 00 00 01 00 movl $0x10000,-0x4(%rbp)
32 bit init: c7 45 f8 ff ff ff 07 movl $0x7ffffff,-0x8(%rbp)
32 bit load: 8b 45 fc mov -0x4(%rbp),%eax
64 bit cast: 48 63 d0 movslq %eax,%rdx
32 bit load: 8b 45 f8 mov -0x8(%rbp),%eax
32->64 conv: 48 98 cltq
64 bit mult: 48 0f af c2 imul %rdx,%rax
64 bit stor: 48 89 45 f0 mov %rax,-0x10(%rbp)
So yes, there may be truncation without a cast. Machine code is 6 bytes longer and 64 bit multiplication seems to require both operands in registers.