This page is an accumulation of (we hope) helpful notes, hints, etc., that might be useful to you in doing Project #4.
There is no direct way to output floating-point numbers in assembler. Instead, output them as two 32-bit integers. For the ia32 architecture, the least-significant 32 bits come first, followed in the next four bytes by the most-significant 21 bits (of which the most significant is always 1 and there not included), an 11-bit exponent, and a sign bit (IEEE floating-point numbers are represented in sign-magnitude form rather than 2's complement).
In Java, you can get at these bits as a long value from a floating-point value X, with the call
Double.doubleToRawLongBits (X)
The low-order 32 bits are then (int) (X & 0xffffffffL) and the high-order bits are (int) (X >> 32).
Using GNU C++, there are a number of tricks for doing this. Again, if X is a variable of type double, you can use the (illicit) expression
*(long long int*)&X
to get the equivalent of doubleToRawLongBits.
Caution: When pushing a floating-point value (as opposed to laying out a floating-point literal in static storage with .long< pseudo-instructions), remember to push the most-significant part first, since the stack grows toward lower addresses.
Grr.
There was a glitch in what we orginally said about returning the value
of a function. When GCC passes a pointer to the return value space as its
first argument, it also pops this pointer from the
stack
ret $4
rather than
ret
and should f(42), write
subl $16, %esp # Reserve space for result
movl $0, 4(%esp)
pushl $0 # Push 42
pushl $42
pushl $__obj_Int
pushl $0
pushl $0 # Static link
leal 20(%esp), %eax # Return value address
pushl %eax
call f
addl $20, %esp # NOT $24
...
and return from f (and all your functions) with
leave
ret $4
The only exception is that your _pyth_main routine probably should use just
plain "ret", but I'll make sure it's harmless if it doesn't.
[CS164 Homework] [CS164 Home Page]
Page was last modified on Tue Dec 5 19:48:53 2006.