double-long cast on Intel 8080

[sblendorio]

Hi, I produced this little program:

#include <stdio.h>
#include <stdlib.h>
#include <math.h>

void main(int argc, char *argv[]) {
    long a = 6;
    long b = 3;
    double c;

    c = a/(2.0*b);

    printf("value=%lf\n",c);
}

It should print "1.0000". It works if compiled for z80 cpu:

$ zcc +cpm -lm test.c

But, compiling it for Intel 8080, it produces a compile-time error:

$ zcc +cpm -lm -clib=8080 test.c
Error at file 'test.c::main' line 5: symbol 'dpush3' not defined
Errors in source file /usr/local/share/z88dk//lib/target/cpm/classic/cpm_crt0.asm:
Error at file 'test.c::main' line 5: symbol 'dpush3' not defined
                   ^ ----     void main(int argc, char *argv[]) {

[suborb]

Please see this page: https://github.com/z88dk/z88dk/wiki/Classic--Maths-Libraries for the maths libraries supported on the 8080.

Effectively you'll need to use --math-dai32_8080 or if you want to buildit from source --math-mbf32_8080

[sblendorio]

tried to use these, I got this:

$ zcc +cpm --math-dai32_8080 -clib=8080 test.c
Error at file 'c/sccz80/l_f32_long2f.asm' line 27: symbol '___dai32_FPSIGN' not defined
Error at file 'c/sccz80/l_f32_long2f.asm' line 31: symbol '___dai32_FPEXP' not defined
Error at file 'c/sccz80/l_f32_long2f.asm' line 33: symbol '___dai32_BNORM' not defined
Errors in source file /usr/local/share/z88dk//lib/target/cpm/classic/cpm_crt0.asm:
Error at file 'c/sccz80/l_f32_long2f.asm' line 27: symbol '___dai32_FPSIGN' not defined
Error at file 'c/sccz80/l_f32_long2f.asm' line 31: symbol '___dai32_FPEXP' not defined
Error at file 'c/sccz80/l_f32_long2f.asm' line 33: symbol '___dai32_BNORM' not defined
$ zcc +cpm --math-mbf32_8080 -clib=8080 test.c
Error at file 'c/asm/___mbf32_DVBCDE.asm' line 4: symbol 'DVBCDE' not defined
Error at file 'c/asm/___mbf32_BNORM.asm' line 4: symbol 'BNORM' not defined
Error at file 'c/asm/___mbf32_FPADD.asm' line 4: symbol 'FPADD' not defined
Error at file 'c/asm/___mbf32_CMPNUM.asm' line 4: symbol 'CMPNUM' not defined
Error at file 'c/asm/___mbf32_FPINT.asm' line 4: symbol 'FPINT' not defined
Error at file 'c/asm/___mbf32_FPMULT.asm' line 4: symbol 'FPMULT' not defined
Error at file 'c/asm/___mbf32_SUBCDE.asm' line 4: symbol 'SUBCDE' not defined
Errors in source file /usr/local/share/z88dk//lib/target/cpm/classic/cpm_crt0.asm:
Error at file 'c/asm/___mbf32_DVBCDE.asm' line 4: symbol 'DVBCDE' not defined
Error at file 'c/asm/___mbf32_BNORM.asm' line 4: symbol 'BNORM' not defined
Error at file 'c/asm/___mbf32_FPADD.asm' line 4: symbol 'FPADD' not defined
Error at file 'c/asm/___mbf32_CMPNUM.asm' line 4: symbol 'CMPNUM' not defined
Error at file 'c/asm/___mbf32_FPINT.asm' line 4: symbol 'FPINT' not defined
Error at file 'c/asm/___mbf32_FPMULT.asm' line 4: symbol 'FPMULT' not defined
Error at file 'c/asm/___mbf32_SUBCDE.asm' line 4: symbol 'SUBCDE' not defined

[suborb]

I should have fixed up the dai problem (there was some errant code leftover from when I copied mbf32 as the base) - please try again tomorrow!

For MBF32 the actual maths code isn't supplied with z88dk - you need to download an additional file and then build the library yourself.

[sblendorio]

What is the download URL for the additional file?

[sblendorio]

Just did a git pull but I got this:

$ zcc +cpm --math-dai32_8080 -clib=8080 test.c -o test.com
Error at file 'c/sccz80/l_f32_long2f.asm' line 27: symbol '___dai32_FPSIGN' not defined
Error at file 'c/sccz80/l_f32_long2f.asm' line 31: symbol '___dai32_FPEXP' not defined
Error at file 'c/sccz80/l_f32_long2f.asm' line 33: symbol '___dai32_BNORM' not defined
Errors in source file /usr/local/share/z88dk//lib/target/cpm/classic/cpm_crt0.asm:
Error at file 'c/sccz80/l_f32_long2f.asm' line 27: symbol '___dai32_FPSIGN' not defined
Error at file 'c/sccz80/l_f32_long2f.asm' line 31: symbol '___dai32_FPEXP' not defined
Error at file 'c/sccz80/l_f32_long2f.asm' line 33: symbol '___dai32_BNORM' not defined

[suborb]

Sorry, okay. So:

Download URL is in this file: https://github.com/z88dk/z88dk/blob/master/libsrc/math/mbf32/z80/math_mbf32.asm

Regarding daimath you'll need to rebuild:

cd libsrc/math/daimath
make clean
rm ../../daimath*.lib
make
cp ../../daimath* ../../../lib/clibs/

[sblendorio]

Just did a complete removal of z88dk and total rebuild with:

git clone --recursive https://github.com/z88dk/z88dk.git
cd z88dk
sudo ./build.sh
sudo make install

then recompiled my code with

zcc +cpm --math-dai32_8080 -clib=8080 test.c

it worked! Thank you

[sblendorio]

a question: are there reasons for preferring mbf32 instead of dai32 ?

[suborb]

On CP/M I'm not sure it makes a difference - the dai32 fp format is the same as the AM9511 FPU so in theory the library could be extended to offload certain operations - see #1543 for the benefits (or not)

If I remember correctly, the dai32math mantissa is 1 bit shorter than mbf32 but I don't think that has a serious impact.

The mbf32 implementation is certainly a lot more tested than dai32math! - It's been embedded in the ROMs of various machines and nascom basic which has been ported pretty much everywhere.

My gut says go with mbf32, but it's a personal choice - I think they perform roughly the same.

[sblendorio]

is it on a different URL for license reasons?

[feilipu]

If I remember correctly, the dai32math mantissa is 1 bit shorter than mbf32 but I don't think that has a serious impact.

Actually it is a one bit shorter 7-bit exponent, which makes the range +63 to -64. The mantissa is identical.
This means the results are identically accurate (assuming one can rely on the calculations from both being accurate), but the DAI math format has a smaller range of +/-(2.7x10^-20 to 9.2x10^18.

My gut says go with mbf32, but it's a personal choice - I think they perform roughly the same.

👍