Why ML always use LEAVE but not ENTER?

[EduardoS]

enter is much more complex than leave, it don't just push ebp/mov ebp, esp:

temp_RBP = RSP // This value of RSP will eventually be loaded
// into RBP.
IF (temp_LEVEL>0) // Push "temp_LEVEL" parameters to the stack.
{
ENTER_START:
temp_ALLOC_SPACE = word-sized immediate specified in the instruction
(first operand), zero-extended to 64 bits
temp_LEVEL = byte-sized immediate specified in the instruction
(second operand), zero-extended to 64 bits
temp_LEVEL = temp_LEVEL AND 0x1f
// only keep 5 bits of level count
PUSH.v old_RBP
FOR (I=1; I<temp_LEVEL; I++)
// All but one of the parameters are copied
// from higher up on the stack.
{
temp_DATA = READ_MEM.v [SS:old_RBP-I*V]
PUSH.v temp_DATA
}
PUSH.v temp_RBP // The last parameter is the offset of the old
// value of RSP on the stack.
}
RSP.s = RSP - temp_ALLOC_SPACE // Leave "temp_ALLOC_SPACE" free bytes on
// the stack
WRITE_MEM.v [SS:RSP.s] = temp_unused // ENTER finishes with a memory write
// check on the final stack pointer,
// but no write actually occurs.
RBP.v = temp_RBP
EXIT

I never needed the second parameter, and i guess most of you neither, but it exists and the processor must intepret it,

Both leave and pop generates 2 uops, but pop can't be used alone to exit a stack frame.

[Ossa]

Yes, the second parameter is for nested procedures - almost no language supports them (not even Java IIRC) I think ML and Haskell might, but I'm rusty on those languages.

Ossa

[savage]

Gotcha.  So it basically expands the complex instructions into simlpe instructions, which takes time.  So we may as well expand it in the first place.

[Mark Jones]

...and omit the complex nesting. Cool.