X-Git-Url: https://thelambdalab.xyz/gitweb/index.cgi?a=blobdiff_plain;f=src%2Ffloat.4th;h=51ba1585f268468981c40b3cb1c9b0fb59069bed;hb=6c019df6b459c86bf5de4fd817db2316038935da;hp=bd429ce23215ee766d1373504e7c80bec80e99ea;hpb=18d5eef9c97da9c27bf9bb5cdaba6baf4a85f966;p=scheme.forth.jl.git diff --git a/src/float.4th b/src/float.4th index bd429ce..51ba158 100644 --- a/src/float.4th +++ b/src/float.4th @@ -2,9 +2,9 @@ ( Cheating for now by using forth.jl CODE/END-CODE to access Julia's floating point support. This isn't - at all portable. That said, the year is 2016 and most - CPUs implement these operations - even the trig functions, - so I don't feel too bad! ) + at all portable. That said, the year is 2016 and + I've only cheated for words that have corresponding + x87 FPU instructions, so I don't feel too bad! ) CODE f+ b = reinterpret(Float64, popPS()) @@ -21,19 +21,24 @@ END-CODE CODE f* b = reinterpret(Float64, popPS()) a = reinterpret(Float64, popPS()) - pushPS(reinterpret(Int64, a*b)) + pushPS(reinterpret(Int64, a*b + 0.0)) END-CODE CODE f/ b = reinterpret(Float64, popPS()) a = reinterpret(Float64, popPS()) - pushPS(reinterpret(Int64, a/b)) + pushPS(reinterpret(Int64, a/b + 0.0)) END-CODE CODE f^ b = reinterpret(Float64, popPS()) a = reinterpret(Float64, popPS()) - pushPS(reinterpret(Int64, a^b)) + pushPS(reinterpret(Int64, a^b + 0.0)) +END-CODE + +CODE fsqrt + a = reinterpret(Float64, popPS()) + pushPS(reinterpret(Int64, sqrt(a))) END-CODE CODE f> @@ -75,7 +80,7 @@ END-CODE CODE fmod b = reinterpret(Float64, popPS()) a = reinterpret(Float64, popPS()) - pushPS(reinterpret(Int64, a%b)) + pushPS(reinterpret(Int64, a%b + 0.0)) END-CODE CODE flog @@ -88,6 +93,26 @@ CODE fexp pushPS(reinterpret(Int64, exp(a))) END-CODE +CODE fsin + a = reinterpret(Float64, popPS()) + pushPS(reinterpret(Int64, sin(a))) +END-CODE + +CODE fcos + a = reinterpret(Float64, popPS()) + pushPS(reinterpret(Int64, cos(a))) +END-CODE + +CODE ftan + a = reinterpret(Float64, popPS()) + pushPS(reinterpret(Int64, tan(a))) +END-CODE + +CODE fatan + a = reinterpret(Float64, popPS()) + pushPS(reinterpret(Int64, atan(a))) +END-CODE + CODE fnan? a = reinterpret(Float64, popPS()) if isnan(a) @@ -115,6 +140,16 @@ CODE f->i pushPS(Int64(round(a))) END-CODE +CODE fabs + a = reinterpret(Float64, popPS()) + pushPS(reinterpret(Int64, abs(a))) +END-CODE + +CODE fround + a = reinterpret(Float64, popPS()) + pushPS(reinterpret(Int64, round(a))) +END-CODE + : f/mod 2dup fmod -rot f/ ; @@ -133,15 +168,12 @@ END-CODE : flog10 flog [ 10 i->f flog ] literal f/ ; -: fabs - dup 0.0 f< if - -1.0 f* - then -; +: truncate + dup 1.0 fmod f- ; : floor dup 0.0 f>= if - dup 1.0 fmod f- + truncate else dup 1.0 fmod dup 0.0 <> if f- 1.0 f- @@ -151,6 +183,31 @@ END-CODE then ; +: ceiling + dup 0.0 f>= if + dup 1.0 fmod dup 0.0 <> if + f- 1.0 f+ + else + drop + then + else + truncate + then +; + +: fasin ( float -- res ) + dup + dup f* 1.0 swap f- fsqrt + f/ + + fatan +; + +: facos ( float -- res ) + dup f* 1.0 swap f/ 1.0 f- fsqrt + fatan +; + : fhead ( float -- ) floor f->i 0 .R ;