80-Bus News |
September–October 1983 · Volume 2 · Issue 5 |
| Page 13 of 67 |
|---|
| Language | Time | ||||
|---|---|---|---|---|---|
| Pascal/MT+ (V5.5) normal | : | 36.7 | secs | ||
| * | Pascal/MT+ (V5.5) – Am9511 | : | 7.42 | secs | Am9511 support |
| Compas Pascal (V1.06) | : | 57.4 | secs | ||
| Hisoft Pascal (V5) | : | 6.39 | secs | Am9511 support | |
* Am9511 library routines are already available on the Pascal/MT disc
Fig 3 – PCW BM8 timings (4MHz Z80)
I decided it would be instructive to try to use the processor from another language. My current language is C in the form of the C/80 compiler from the Software Toolworks (highly recommended). The version I run does not support floating point numbers, but then as I don’t use them I find this no loss. A few moments thought showed that it would be very easy to add support for the arithmetic processor with this compiler, (in fact I first tried the board out with a simple C program), but I decided that it would be more instructive to try to add support to something like the ubiquitous Microsoft Basic80.
The benchmark above indicated that the Microsoft interpreter and its companion compiler use the same arithmetic subroutines, and as the benchmark for the compiler was only 20% faster than the interpreter I decided to use the interpreter rather than tackle the run-time routines of the compiler.
The easiest (although slightly clumsy) way to interface the card would be by the USR or CALL interface, but I decided to try to patch the interpreter directly. On turning up appendix C of the MBasic manual I encountered problem number one. The two floating formats are not identical (see below). However they are basically similar, so it is not out of the question to interface the HSA-88B to the interpreter.
The mantissa is expressed as a 24- bit (fractional) value; the exponent is expressed as an un-biased two’s complement 7-bit value having a range of −64 to +63. The most significant bit is the sign of the mantissa (0=positive, 1=negative), for a total of 32 bits. The binary point is assumed to be to the left of the most significant mantissa bit (bit 23). All floating-point data values must be normalised. Bit 23 must be equal to 1, expect for the value zero, which is represented by all zeros.
The mantissa is expressed as a 24- bit (fractional) value with the leading 1 suppressed (implied) and the binary point is to the left of the most significant bit; the exponent is expressed as a biased 8-bit number having a bias of 128. The sign of the mantissa is in bit 23 (0=positive, 1=negative). The number zero is represented by a number with an exponent of 0.
The two are similar, and re-arranging a few of the bits can get them to line up, but the only problem is that the Microsoft exponent range is +127 to −127, as opposed to the −64 to +63 of the Am9511. As this was just an exploratory hack, I took the easy way out and ignored this!
| Page 13 of 67 |
|---|