FP precision [ANSWERED]

Questions
1

I was working with a function that appears as follows:

Some details of the input data
– self.radius = 6371.0 – kilometers
– self.lat = math.rad(1.0)
– self.lat = math.rad(1.0)

function GeoPos:RhumbDestinationPoint(bearing,distance)
local brng = math.rad(bearing)
local dist = distance/self.radius
print(“brng”,brng)
print(“dist”,dist)

local lat1 = self.lat -- gets converted to rads in the Init() function
local lon1 = self.lon -- gets converted to rads in the Init() function
local lat2 = lat1 + dist*math.cos(brng)
local dLat = lat2-lat1
local dPhi = math.log(math.tan(lat2/2+math.pi/4)/math.tan(lat1/2+math.pi/4))
local q = 0.0

if bearing==90.0 or bearing==270.0 then
    q = math.cos(lat1)
else
    q = dLat/dPhi
end
print("q",q)

local dLon = dist*math.sin(brng)/q
local lon2 = math.fmod((lon1+dLon+3*math.pi),(2*math.pi))-math.pi

print("lat1",lat1)
print("lon1",lon1)
print("lat2",lat2)
print("lon2",lon2)
print("dPhi",dPhi)
print("dLat",dLat)
print("dLon",dLon)

return math.deg(lat2),math.deg(lon2)

end

This is a snippet from a class (GeoPos) I created. I am calling the function as follows:

function setup()
lat=1.0
lon=1.0
GP=GeoPos(lat,lon)
lat,lon=GP:RhumbDestinationPoint(45,.01) – 045 degrees at 10 meters
print(“–”,lat,lon)
end

If I run the function on a PC in LoveCodify or LuaForWindows or Python or C or C++ I get the following results:

brng 0.78539816339745
dist 1.5696123057605e-006
q 0.99984768527973
lat1 0.017453292519943
lon1 0.017453292519943
lat2 0.017454402403449
lon2 0.017454402572525
dPhi 1.1100525825855e-006
dLat 1.1098835052369e-006
dLon 1.1100525825857e-006
– 1.0000635916406 1.000063601328

When I run it on an iPad2 in Codify, I get these results:

brng 0.785398
dist 1.56961e-006
q 1.03472
lat1 0.0174533
lon1 0.0174533
lat2 0.0174544
lon2 0.0174539
dPhi 1.07288e-006
dLat 1.11014e-006
dLon 1.07264e-006
– 1.00006 1.00004

This is a nice function to simulate motion but it seems to fail because of precision issues on the iPad2. I can find a different way to simulate motion but I was surprised that the math functions broke down so quickly at small values. Any insight into iPad floating point architecture would be appreciated.

Thanks

2

Hi SonarMan

I modified Lua to use 32 bit floating point precision rather than the default 64 bit because I had performance concerns.

I’m thinking about the best way to tackle this, as Lua does not support varying FP precision. Perhaps we can look at making it a project-wide setting.

3

Simeon,

Thanks for the quick response. I was beginning to wonder if I had overlooked something. Happy Thanksgiving!!