zondag 28 februari 2016

SPO600 Lab 7

In this lab we are going to look at some software packages that use inline assembler, we are going to look at why it uses it, whether that use is correct and what platforms it is present for.

I will be looking at libmad or MAD, which is an MPEG Audio Decoding package. If you wish to follow along, the source code package is available on sourceforge at https://sourceforge.net/projects/mad/files/libmad/, as in demonstrated in lab 1, we can just extract the tar ball by using the tar -xvf [filename] command. We don't have to make the program to look at the source code. First off I will be using egrep -R "asm | __(asm)__*\(" to recursively search downwards in the map for which programs inside the package contain assembler.
This gives us 3 results, 2 of which are actual inline use of asssembler.
intl/libgnuintl.h.in and audio_jaguar.c

First I will be looking at the intl/libgnuintl.h.in to see if I can figure out what the instruction is there for, the instruction we are looking at is:
# define _INTL_ASM(cname) __asm__ (_INTL_ASMNAME (__USER_LABEL_PREFIX__, #cname))
According to the comments this sets up an alternative prefix to the usual command, to be used if the libintl_ library does not work. The reason this is here, is that due to duplicate names in different libraries glibc 2.0 and newer and Solaris 2.4 overwrite a function used in the libintl_ library. I can only assume that this is a valid reason to rename it.

On to the audio_jaguar.c where it doesn't concern just the renaming of a function, here we find the following function:
 * NAME:        float32()
 * DESCRIPTION: store 32-bit IEEE Standard 754 floating point representation
void float32(void **ptr, mad_fixed_t sample)
  unsigned long **mem = (void *) ptr, ieee = 0;

  if (sample) {
    unsigned int zc;
    unsigned long abs, s, e, f;

    enum {
      BIAS = 0x7f

    /* |seeeeeee|efffffff|ffffffff|ffffffff| */

    s = sample & 0x80000000UL;
    abs = s ? -sample : sample;

    /* PPC count leading zeros */
    asm ("cntlzw %0,%1" : "=r" (zc) : "r" (abs));

    /* calculate exponent */

    e = (((32 - MAD_F_FRACBITS - 1) - zc + BIAS) << 23) & 0x7f800000UL;

    /* normalize 1.f - round? */

    f = ((abs << zc) >> 8) & 0x007fffffUL;

    ieee = s | e | f;
Here the assembly instruction is used to count the leading zero's in a register, used to convert a floating point value to the IEEE Standard 754(a fixed point representation). Note that the code offers no alternatives and is only valid on PowerPC. I think this is because most modern processors support this IEEE standard, therefore they do not need a conversion function. In my opinion it is a little bit sloppy to allow the function to break if ran on any other processor, interestingly enough this code was written to increase the portability of the library. The standard was absent on this specific platform so they wrote an alternative implementation.

That was it for this library, there was not that much assembly to be found in here. The assembly that was there, was written to increase portability to systems that have naming issues or lack the IEEE 754 standard. I would have liked to have seen an alternative in C that makes sure the latter function does not break on any other platform than PowerPC, but it is probably assumed that any modern system supports the IEEE 754 standard.

Geen opmerkingen:

Een reactie posten