Implement FX2DEC. Not perfect, but good enough I think.

This commit is contained in:
Kelvin Sherlock 2015-02-25 15:33:59 -05:00
parent 2a62ad32d2
commit 56fb5d9743
4 changed files with 431 additions and 19 deletions

View File

@ -5,6 +5,7 @@ MAKEFLAGS += --no-builtin-rules
.SUFFIXES:
.SECONDARY:
MPWFLAGS = -DMPWVersion=3.2
MPW ?= mpw
LIBS = \
@ -25,7 +26,7 @@ LDFLAGS = -w -c 'MPS ' -t MPST \
SCFLAGS = -p
TARGETS = test_new_handle test_new_handle_2 test_new_pointer test_volumes \
test_createresfile test_hwpriv
test_createresfile test_hwpriv test_sane
all : $(TARGETS)
@ -44,10 +45,16 @@ clean :
# test_volumes : o/test_volumes.o
# mpw Link $(LDFLAGS) -o $@ $^ $(LIBS)
test_sane: o/nan.o o/test_sane.o
$(MPW) $(MPWFLAGS) Link $(LDFLAGS) -o $@ $^ $(LIBS)
% : o/%.o
$(MPW) Link $(LDFLAGS) -o $@ $^ $(LIBS)
$(MPW) $(MPWFLAGS) Link $(LDFLAGS) -o $@ $^ $(LIBS)
o/%.o : %.c
$(MPW) SC $(SCFLAGS) $< -o $@
$(MPW) $(MPWFLAGS) SC $(SCFLAGS) $< -o $@
o/%.o : %.asm
$(MPW) $(MPWFLAGS) Asm $(ASMFLAGS) $< -o $@

35
test/nan.asm Normal file
View File

@ -0,0 +1,35 @@
case on
export nan, __inf, inf
nan proc
; The MPW 3.2 version of nan is broken. This what it's supposed to do.
; extended (80-bit) returns via d0, d1, a0
MOVEM.L data,D0/D1/A0
MOVE.W #$4000,D1
MOVE.B $0004(A7),D1
BNE.S swap
MOVE.B #$15,D1
swap
SWAP D1
RTS
entry inf
entry __inf
inf
__inf
MOVEM.L data,D0/D1/A0
RTS
data
dc.l $00007fff
dc.l $00000000
dc.l $00000000
end
endp

229
test/test_sane.c Normal file
View File

@ -0,0 +1,229 @@
#include <Math.h>
#include <SANE.h>
#include <stdio.h>
//extern pascal extended NAN(short x);
void dump_decimal(const decimal *d)
{
fprintf(stdout, "%d : %d : %.*s\n",
d->sgn, d->exp, d->sig.length, d->sig.text);
}
void test_fxc2dec(void)
{
decimal d;
decform df;
df.style = 0;
df.digits = 100;
num2dec(&df, 1.125, &d);
dump_decimal(&d);
// 0 : -18 : 1125000000000000000
df.style = 0;
df.digits = 10;
num2dec(&df, 1.125, &d);
dump_decimal(&d);
// 0 : -9 : 1125000000
df.style = 0;
df.digits = 2;
num2dec(&df, 1.125, &d);
dump_decimal(&d);
// 0 : -1 : 11
df.style = 1;
df.digits = 100;
num2dec(&df, 1.125, &d);
dump_decimal(&d);
// 0 : -18 : 1125000000000000000
df.style = 1;
df.digits = 10;
num2dec(&df, 1.125, &d);
dump_decimal(&d);
// 0 : -10 : 11250000000
df.style = 1;
df.digits = 2;
num2dec(&df, 1.125, &d);
dump_decimal(&d);
//0 : -2 : 112
//
df.style = 0;
df.digits = 100;
num2dec(&df, 1e25, &d);
dump_decimal(&d);
// 0 : 7 : 1000000000000000000
df.style = 0;
df.digits = 10;
num2dec(&df, 1e25, &d);
dump_decimal(&d);
// 0 : 16 : 1000000000
df.style = 0;
df.digits = 2;
num2dec(&df, 1e25, &d);
dump_decimal(&d);
// 0 : 24 : 10
df.style = 1;
df.digits = 100;
num2dec(&df, 1e25, &d);
dump_decimal(&d);
// 0 : 7 : 1000000000000000000
df.style = 1;
df.digits = 10;
num2dec(&df, 1e25, &d);
dump_decimal(&d);
// 0 : 7 : 1000000000000000000
df.style = 1;
df.digits = 2;
num2dec(&df, 1e25, &d);
dump_decimal(&d);
//0 : 7 : 1000000000000000000
df.style = 1;
df.digits = 10;
num2dec(&df, 0, &d);
dump_decimal(&d);
//
df.style = 0;
df.digits = 10;
num2dec(&df, 0, &d);
dump_decimal(&d);
//
df.style = 1;
df.digits = 0;
num2dec(&df, -0.0, &d);
dump_decimal(&d);
//
df.style = 0;
df.digits = 0;
num2dec(&df, -0.0, &d);
dump_decimal(&d);
//
df.style = 1;
df.digits = 0;
num2dec(&df,nan(1), &d);
dump_decimal(&d);
//
df.style = 0;
df.digits = 0;
num2dec(&df, nan(2), &d);
dump_decimal(&d);
//
df.style = 1;
df.digits = 10;
num2dec(&df,-nan(3), &d);
dump_decimal(&d);
//
df.style = 0;
df.digits = 10;
num2dec(&df, nan(4), &d);
dump_decimal(&d);
//
df.style = 1;
df.digits = 10;
num2dec(&df,-inf(), &d);
dump_decimal(&d);
//
df.style = 0;
df.digits = 10;
num2dec(&df, -inf(), &d);
dump_decimal(&d);
df.style = 1;
df.digits = 0;
num2dec(&df, inf(), &d);
dump_decimal(&d);
//
df.style = 0;
df.digits = 0;
num2dec(&df, inf(), &d);
dump_decimal(&d);
df.style = 0;
df.digits = 10;
num2dec(&df, 0.125, &d);
dump_decimal(&d);
//
df.style = 1;
df.digits = 10;
num2dec(&df, 0.125, &d);
dump_decimal(&d);
//
df.style = 0;
df.digits = 4;
num2dec(&df, 0.00000125, &d);
dump_decimal(&d);
// s/b -9 1250
df.style = 1;
df.digits = 4;
num2dec(&df, 0.00000125, &d);
dump_decimal(&d);
// s/b -4 0
}
int main(int argc, char **argv)
{
extended x;
decimal d;
short index;
short valid;
(void)argc;
(void)argv;
index = 0;
valid = 0;
str2dec("1.125", &index, &d, &valid);
x = dec2num(&d);
test_fxc2dec();
return 0;
}

View File

@ -35,6 +35,7 @@
#include <stdlib.h>
#include <string>
#include <cstring>
#include <algorithm>
#include <cmath>
#include "stackframe.h"
@ -50,6 +51,7 @@ namespace SANE
using std::to_string;
using std::fpclassify;
using std::signbit;
using std::abs;
using its_complicated::to_string;
using its_complicated::fpclassify;
@ -310,6 +312,72 @@ using its_complicated::signbit;
ToolBox::WritePString(address + _sig, sig);
}
void format_f(extended x, int precision, std::string &mm, std::string &nn)
{
char *buffer = nullptr;
mm.clear();
nn.clear();
if (precision < 0) precision = 0;
int length = asprintf(&buffer, "%.*Lf", precision, x);
std::string tmp(buffer, length);
free(buffer);
auto dot = tmp.find('.');
if (dot == tmp.npos)
{
mm = std::move(tmp);
}
else
{
mm = tmp.substr(0, dot);
nn = tmp.substr(dot + 1);
}
// skip mm if it's 0
if (mm.length() == 1 && mm.front() == '0') mm.clear();
// skip nn if it's 0000...
if (std::all_of(nn.begin(), nn.end(), [](char c){ return c == '0'; }))
nn.clear();
}
void format_e(extended x, int precision, std::string &mm, std::string &nn, int &exp)
{
char *buffer = nullptr;
exp = 0;
mm.clear();
nn.clear();
if (precision < 0) precision = 0;
if (precision > 19) precision = 19;
int length = asprintf(&buffer, "%.*Le", precision, x);
// output mm . nn e[+-]exp
// mm e[+-]exp
std::string tmp(buffer, length);
free(buffer);
auto dot = tmp.find('.');
auto e = tmp.find('e');
if (dot == tmp.npos)
{
mm = tmp.substr(0, e);
}
else
{
mm = tmp.substr(0, dot);
nn = tmp.substr(dot + 1, e - dot - 1);
}
char sign = tmp[e+1];
tmp = tmp.substr(e + 2);
exp = std::stoi(tmp);
if (sign == '-') exp = -exp;
}
uint16_t fx2dec()
{
@ -331,8 +399,6 @@ using its_complicated::signbit;
Log(" FX2DEC(%08x, %08x, %08x, %04x)\n", f_adr, a_adr, d_adr, op);
fprintf(stderr, "warning: FX2DEC not yet implemented\n");
extended s = readnum<extended>(a_adr);
df = decform::read(f_adr);
@ -353,18 +419,20 @@ using its_complicated::signbit;
* [-]mmm[.nnn]
*/
// this doesn't entirely apply for fx2dec, though.
if (df.digits < 0) df.digits = 0;
if (df.digits > 19) df.digits = 19;
fpinfo fpi(s);
//fprintf(stderr, "%02x %02x %08x %016llx\n", fpi.sign, fpi.one, fpi.exp, fpi.sig);
//fprintf(stderr, "%02x %02x %d %016llx\n", fpi.sign, fpi.one, fpi.exp, fpi.sig);
d.sgn = signbit(s);
// handle infinity, nan as a special case.
switch (fpclassify(s))
{
case FP_ZERO:
d.sgn = signbit(s);
d.sig = "0";
d.write(d_adr);
return 0;
@ -374,7 +442,6 @@ using its_complicated::signbit;
{
char buffer[20]; // 16 + 2 needed
snprintf(buffer, 20, "N%016llx", fpi.sig);
d.sgn = signbit(s);
d.sig = buffer;
d.write(d_adr);
}
@ -382,7 +449,6 @@ using its_complicated::signbit;
case FP_INFINITE:
d.sgn = signbit(s);
d.sig = "I";
d.write(d_adr);
return 0;
@ -394,21 +460,96 @@ using its_complicated::signbit;
// normal and subnormal handled here....
#if 0
if (df.style == decform::FIXEDDECIMAL)
// float decimal: df.digits refers to the total length
// fixed decimal: df.digits refers to the fractional part only.
s = abs(s);
if (s < 1.0 && df.style == decform::FLOATDECIMAL)
{
char buffer[decimal::SIGDIGLEN];
snprintf(buffer, sizeof(buffer), "%.*Lg", df.digits, s);
std::string mm;
std::string nn;
format_e(s, df.digits - 1, mm, nn, d.exp);
}
#endif
d.sig = mm + nn;
// ugh, really don't want to write this code right now.
// better be < 0...
if (d.exp < 0)
d.exp -= nn.length();
d.write(d_adr);
return 0;
}
else // s > 1
{
std::string mm;
std::string nn;
format_f(s, df.digits, mm, nn);
if (mm.empty() && nn.empty())
{
// very large 0.
d.sig = "0";
d.write(d_adr);
return 0;
}
// if nn is empty (or 0s), this is a large number,
// and we don't have to worry about the fraction.
if (nn.empty())
{
d.exp = 0;
if (df.style == decform::FIXEDDECIMAL) df.digits = 19;
// limit the length.
if (mm.length() > df.digits)
{
d.exp = mm.length() - df.digits;
mm.resize(df.digits);
}
d.sig = std::move(mm);
}
else
{
if (df.style == decform::FIXEDDECIMAL)
{
// digits is the total size, mm + nn
// re-format with new precision.
// this is getting repetitive...
if (mm.length())
{
int precision = df.digits - mm.length();
if (precision < 0) precision = 1;
format_f(s, precision, mm, nn);
}
}
// todo -- if mm is empty and nn has leading 0s,
// drop the leading 0s and adjust the exponent
// accordingly.
d.sig = mm + nn;
d.exp = -nn.length();
if (d.sig.length() > 19)
{
d.exp += (d.sig.length() - 19);
d.sig.resize(19);
}
}
d.write(d_adr);
return 0;
}
}
template<class SrcType, class DestType = extended>