prog8/compiler/res/prog8lib/virtual/floats.p8

208 lines
4.2 KiB
Lua

; Prog8 definitions for floating point handling on the VirtualMachine
%option enable_floats, ignore_unused
floats {
const float π = 3.141592653589793
const float PI = π
const float TWOPI = 2*π
sub print(float value) {
; ---- prints the floating point value (without a newline and no leading spaces).
%ir {{
loadm.f fr65535,floats.print.value
syscall 25 (fr65535.f)
return
}}
}
sub tostr(float value) -> str {
; ---- converts the floating point value to a string (no leading spaces)
str @shared buffer=" "*20
%ir {{
load.w r65535,floats.tostr.buffer
loadm.f fr65535,floats.tostr.value
syscall 47 (r65535.w, fr65535.f)
load.w r0,floats.tostr.buffer
returnr.w r0
}}
}
sub parse(str value) -> float {
; -- parse a string value of a number to float
%ir {{
loadm.w r65535,floats.parse.value
syscall 45 (r65535.w): fr0.f
returnr.f fr0
}}
}
sub pow(float value, float power) -> float {
%ir {{
loadm.f fr0,floats.pow.value
loadm.f fr1,floats.pow.power
fpow.f fr0,fr1
returnr.f fr0
}}
}
sub sin(float angle) -> float {
%ir {{
loadm.f fr0,floats.sin.angle
fsin.f fr0,fr0
returnr.f fr0
}}
}
sub cos(float angle) -> float {
%ir {{
loadm.f fr0,floats.cos.angle
fcos.f fr0,fr0
returnr.f fr0
}}
}
sub tan(float value) -> float {
%ir {{
loadm.f fr0,floats.tan.value
ftan.f fr0,fr0
returnr.f fr0
}}
}
sub atan(float value) -> float {
%ir {{
loadm.f fr0,floats.atan.value
fatan.f fr0,fr0
returnr.f fr0
}}
}
; two-argument arctangent that returns an angle in the correct quadrant
; for the signs of x and y, normalized to the range [0, 2π]
sub atan2(float y, float x) -> float {
float atn = atan(y / x)
if x < 0 atn += π
if atn < 0 atn += 2*π
return atn
}
; reciprocal functions
sub secant(float value) -> float { return 1.0 / cos(value) }
sub csc(float value) -> float { return 1.0 / sin(value) }
sub cot(float value) -> float { return 1.0 / tan(value) }
sub ln(float value) -> float {
%ir {{
loadm.f fr0,floats.ln.value
fln.f fr0,fr0
returnr.f fr0
}}
}
sub log2(float value) -> float {
%ir {{
loadm.f fr0,floats.log2.value
flog.f fr0,fr0
returnr.f fr0
}}
}
sub rad(float angle) -> float {
; -- convert degrees to radians (d * pi / 180)
return angle * PI / 180.0
}
sub deg(float angle) -> float {
; -- convert radians to degrees (d * (1/ pi * 180))
return angle * 180.0 / PI
}
sub round(float value) -> float {
%ir {{
loadm.f fr0,floats.round.value
fround.f fr0,fr0
returnr.f fr0
}}
}
sub floor(float value) -> float {
%ir {{
loadm.f fr0,floats.floor.value
ffloor.f fr0,fr0
returnr.f fr0
}}
}
sub ceil(float value) -> float {
; -- ceil: tr = int(f); if tr==f -> return else return tr+1
%ir {{
loadm.f fr0,floats.ceil.value
fceil.f fr0,fr0
returnr.f fr0
}}
}
sub rnd() -> float {
%ir {{
syscall 35 () : fr0.f
returnr.f fr0
}}
}
sub rndseed(float seed) {
%ir {{
loadm.f fr65535,floats.rndseed.seed
syscall 32 (fr65535.f)
return
}}
}
sub minf(float f1, float f2) -> float {
if f1<f2
return f1
return f2
}
sub maxf(float f1, float f2) -> float {
if f1>f2
return f1
return f2
}
sub clampf(float value, float minimum, float maximum) -> float {
if value<minimum
value=minimum
if value<maximum
return value
return maximum
}
sub normalize(float value) -> float {
return value
}
sub push(float value) {
; note: this *should* be inlined, however since the VM has separate program counter and value stacks, this also works
%ir {{
loadm.f fr65535,floats.push.value
push.f fr65535
return
}}
}
sub pop() -> float {
; note: this *should* be inlined, however since the VM has separate program counter and value stacks, this also works
%ir {{
pop.f fr65535
returnr.f fr65535
}}
}
}