| /* |
| Licensed to the Apache Software Foundation (ASF) under one |
| or more contributor license agreements. See the NOTICE file |
| distributed with this work for additional information |
| regarding copyright ownership. The ASF licenses this file |
| to you under the Apache License, Version 2.0 (the |
| "License"); you may not use this file except in compliance |
| with the License. You may obtain a copy of the License at |
| |
| http://www.apache.org/licenses/LICENSE-2.0 |
| |
| Unless required by applicable law or agreed to in writing, |
| software distributed under the License is distributed on an |
| "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY |
| KIND, either express or implied. See the License for the |
| specific language governing permissions and limitations |
| under the License. |
| */ |
| |
| // |
| // fp16.swift |
| // |
| // Created by Michael Scott on 07/07/2015. |
| // Copyright (c) 2015 Michael Scott. All rights reserved. |
| // |
| |
| /* Finite Field arithmetic Fp^16 functions */ |
| |
| /* FP16 elements are of the form a+ib, where i is sqrt(sqrt(sqrt(-1+sqrt(-1))) */ |
| |
| public struct FP16 { |
| private var a:FP8 |
| private var b:FP8 |
| |
| /* constructors */ |
| init() |
| { |
| a=FP8() |
| b=FP8() |
| } |
| |
| init(_ c:Int) |
| { |
| a=FP8(c) |
| b=FP8() |
| } |
| |
| init(_ x:FP16) |
| { |
| a=FP8(x.a) |
| b=FP8(x.b) |
| } |
| |
| init(_ c:FP8,_ d:FP8) |
| { |
| a=FP8(c) |
| b=FP8(d) |
| } |
| |
| init(_ c:FP8) |
| { |
| a=FP8(c) |
| b=FP8() |
| } |
| /* reduce all components of this mod Modulus */ |
| mutating func reduce() |
| { |
| a.reduce() |
| b.reduce() |
| } |
| /* normalise all components of this mod Modulus */ |
| mutating func norm() |
| { |
| a.norm() |
| b.norm() |
| } |
| /* test this==0 ? */ |
| func iszilch() -> Bool |
| { |
| return a.iszilch() && b.iszilch() |
| } |
| |
| mutating func cmove(_ g:FP16,_ d:Int) |
| { |
| a.cmove(g.a,d) |
| b.cmove(g.b,d) |
| } |
| |
| |
| /* test this==1 ? */ |
| func isunity() -> Bool |
| { |
| let one=FP8(1); |
| return a.equals(one) && b.iszilch() |
| } |
| |
| /* test is w real? That is in a+ib test b is zero */ |
| func isreal() -> Bool |
| { |
| return b.iszilch(); |
| } |
| /* extract real part a */ |
| func real() -> FP8 |
| { |
| return a; |
| } |
| |
| func geta() -> FP8 |
| { |
| return a; |
| } |
| /* extract imaginary part b */ |
| func getb() -> FP8 |
| { |
| return b; |
| } |
| mutating func set_fp8s(_ c: FP8,_ d: FP8) |
| { |
| a.copy(c) |
| b.copy(d) |
| } |
| mutating func set_fp8(_ c: FP8) |
| { |
| a.copy(c) |
| b.zero() |
| } |
| mutating func set_fp8h(_ c: FP8) |
| { |
| b.copy(c) |
| a.zero() |
| } |
| /* test self=x? */ |
| func equals(_ x:FP16) -> Bool |
| { |
| return a.equals(x.a) && b.equals(x.b) |
| } |
| /* copy self=x */ |
| mutating func copy(_ x:FP16) |
| { |
| a.copy(x.a) |
| b.copy(x.b) |
| } |
| |
| /* set this=0 */ |
| mutating func zero() |
| { |
| a.zero() |
| b.zero() |
| } |
| /* set this=1 */ |
| mutating func one() |
| { |
| a.one() |
| b.zero() |
| } |
| /* set self=-self */ |
| mutating func neg() |
| { |
| norm() |
| var m=FP8(a) |
| var t=FP8() |
| m.add(b) |
| m.neg() |
| t.copy(m); t.add(b) |
| b.copy(m) |
| b.add(a) |
| a.copy(t) |
| norm() |
| } |
| /* self=conjugate(self) */ |
| mutating func conj() |
| { |
| b.neg(); norm() |
| } |
| /* this=-conjugate(this) */ |
| mutating func nconj() |
| { |
| a.neg(); norm() |
| } |
| /* self+=x */ |
| |
| mutating func adds(_ x: FP8) |
| { |
| a.add(x) |
| } |
| |
| mutating func add(_ x:FP16) |
| { |
| a.add(x.a) |
| b.add(x.b) |
| } |
| /* self-=x */ |
| mutating func sub(_ x:FP16) |
| { |
| var m=FP16(x) |
| m.neg() |
| add(m) |
| } |
| |
| /* self-=x */ |
| mutating func rsub(_ x: FP16) { |
| neg() |
| add(x) |
| } |
| |
| /* self*=s where s is FP8 */ |
| mutating func pmul(_ s:FP8) |
| { |
| a.mul(s) |
| b.mul(s) |
| } |
| |
| /* self*=s where s is FP2 */ |
| mutating func qmul(_ s:FP2) { |
| a.qmul(s) |
| b.qmul(s) |
| } |
| |
| /* self*=c where c is int */ |
| mutating func imul(_ c:Int) |
| { |
| a.imul(c) |
| b.imul(c) |
| } |
| |
| /* self*=self */ |
| mutating func sqr() |
| { |
| var t1=FP8(a) |
| var t2=FP8(b) |
| var t3=FP8(a) |
| |
| t3.mul(b) |
| t1.add(b) |
| t2.times_i() |
| |
| t2.add(a) |
| |
| t1.norm(); t2.norm() |
| a.copy(t1) |
| |
| a.mul(t2) |
| |
| t2.copy(t3) |
| t2.times_i() |
| t2.add(t3); t2.norm() |
| |
| t2.neg() |
| a.add(t2) |
| |
| b.copy(t3) |
| b.add(t3) |
| |
| norm() |
| } |
| |
| /* self*=y */ |
| mutating func mul(_ y:FP16) |
| { |
| var t1=FP8(a) |
| var t2=FP8(b) |
| var t3=FP8() |
| var t4=FP8(b) |
| |
| t1.mul(y.a) |
| t2.mul(y.b) |
| t3.copy(y.b) |
| t3.add(y.a) |
| t4.add(a) |
| |
| t3.norm(); t4.norm() |
| |
| t4.mul(t3) |
| |
| t3.copy(t1) |
| t3.neg() |
| t4.add(t3) |
| t4.norm() |
| |
| t3.copy(t2) |
| t3.neg() |
| b.copy(t4) |
| b.add(t3) |
| |
| t2.times_i() |
| a.copy(t2) |
| a.add(t1) |
| |
| norm(); |
| } |
| /* convert this to hex string */ |
| func toString() -> String |
| { |
| return ("["+a.toString()+","+b.toString()+"]") |
| } |
| |
| func toRawString() -> String |
| { |
| return ("["+a.toRawString()+","+b.toRawString()+"]") |
| } |
| /* self=1/self */ |
| mutating func inverse() |
| { |
| var t1=FP8(a) |
| var t2=FP8(b) |
| |
| t1.sqr() |
| t2.sqr() |
| t2.times_i(); t2.norm() |
| t1.sub(t2); t1.norm() |
| t1.inverse() |
| a.mul(t1) |
| t1.neg(); t1.norm() |
| b.mul(t1) |
| } |
| |
| /* self*=i where i = sqrt(sqrt(-1+sqrt(-1))) */ |
| mutating func times_i() |
| { |
| var s=FP8(b) |
| let t=FP8(a) |
| s.times_i() |
| a.copy(s) |
| b.copy(t) |
| norm() |
| } |
| |
| mutating func times_i2() { |
| a.times_i() |
| b.times_i() |
| } |
| |
| mutating func times_i4() { |
| a.times_i2() |
| b.times_i2() |
| } |
| |
| /* self=self^p using Frobenius */ |
| mutating func frob(_ f:FP2) |
| { |
| var ff=FP2(f); ff.sqr(); ff.norm() |
| |
| a.frob(ff) |
| b.frob(ff) |
| b.qmul(f) |
| b.times_i() |
| } |
| |
| /* self=self^e */ |
| func pow(_ e:BIG) -> FP16 |
| { |
| var w=FP16(self) |
| w.norm() |
| var z=BIG(e) |
| var r=FP16(1) |
| z.norm() |
| while (true) |
| { |
| let bt=z.parity() |
| z.fshr(1) |
| if bt==1 {r.mul(w)} |
| if z.iszilch() {break} |
| w.sqr() |
| } |
| r.reduce() |
| return r |
| } |
| |
| /* XTR xtr_a function */ |
| mutating func xtr_A(_ w:FP16,_ y:FP16,_ z:FP16) |
| { |
| var r=FP16(w) |
| var t=FP16(w) |
| r.sub(y); r.norm() |
| r.pmul(a) |
| t.add(y); t.norm() |
| t.pmul(b) |
| t.times_i() |
| |
| copy(r) |
| add(t) |
| add(z) |
| |
| norm() |
| } |
| /* XTR xtr_d function */ |
| mutating func xtr_D() |
| { |
| var w=FP16(self) |
| sqr(); w.conj() |
| w.add(w); w.norm(); |
| sub(w) |
| reduce() |
| } |
| /* r=x^n using XTR method on traces of FP48s */ |
| func xtr_pow(_ n:BIG) -> FP16 |
| { |
| var sf=FP16(self) |
| sf.norm() |
| var a=FP16(3) |
| var b=FP16(sf) |
| var c=FP16(b) |
| c.xtr_D() |
| var t=FP16() |
| var r=FP16() |
| |
| let par=n.parity() |
| var v=BIG(n); v.norm(); v.fshr(1) |
| if par==0 {v.dec(1); v.norm()} |
| |
| let nb=v.nbits() |
| var i=nb-1 |
| while i>=0 |
| { |
| if (v.bit(UInt(i)) != 1) |
| { |
| t.copy(b) |
| sf.conj() |
| c.conj() |
| b.xtr_A(a,sf,c) |
| sf.conj() |
| c.copy(t) |
| c.xtr_D() |
| a.xtr_D() |
| } |
| else |
| { |
| t.copy(a); t.conj() |
| a.copy(b) |
| a.xtr_D() |
| b.xtr_A(c,sf,t) |
| c.xtr_D() |
| } |
| i-=1 |
| } |
| if par==0 {r.copy(c)} |
| else {r.copy(b)} |
| r.reduce() |
| return r |
| } |
| |
| /* r=ck^a.cl^n using XTR double exponentiation method on traces of FP48s. See Stam thesis. */ |
| func xtr_pow2(_ ck:FP16,_ ckml:FP16,_ ckm2l:FP16,_ a:BIG,_ b:BIG) -> FP16 |
| { |
| |
| var e=BIG(a) |
| var d=BIG(b) |
| var w=BIG(0) |
| e.norm(); d.norm() |
| |
| var cu=FP16(ck) // can probably be passed in w/o copying |
| var cv=FP16(self) |
| var cumv=FP16(ckml) |
| var cum2v=FP16(ckm2l) |
| var r=FP16() |
| var t=FP16() |
| |
| var f2:Int=0 |
| while d.parity()==0 && e.parity()==0 |
| { |
| d.fshr(1); |
| e.fshr(1); |
| f2 += 1; |
| } |
| |
| while (BIG.comp(d,e) != 0) |
| { |
| if BIG.comp(d,e)>0 |
| { |
| w.copy(e); w.imul(4); w.norm() |
| if BIG.comp(d,w)<=0 |
| { |
| w.copy(d); d.copy(e) |
| e.rsub(w); e.norm() |
| |
| t.copy(cv) |
| t.xtr_A(cu,cumv,cum2v) |
| cum2v.copy(cumv) |
| cum2v.conj() |
| cumv.copy(cv) |
| cv.copy(cu) |
| cu.copy(t) |
| |
| } |
| else if d.parity()==0 |
| { |
| d.fshr(1) |
| r.copy(cum2v); r.conj() |
| t.copy(cumv) |
| t.xtr_A(cu,cv,r) |
| cum2v.copy(cumv) |
| cum2v.xtr_D() |
| cumv.copy(t) |
| cu.xtr_D() |
| } |
| else if e.parity()==1 |
| { |
| d.sub(e); d.norm() |
| d.fshr(1) |
| t.copy(cv) |
| t.xtr_A(cu,cumv,cum2v) |
| cu.xtr_D() |
| cum2v.copy(cv) |
| cum2v.xtr_D() |
| cum2v.conj() |
| cv.copy(t) |
| } |
| else |
| { |
| w.copy(d) |
| d.copy(e); d.fshr(1) |
| e.copy(w) |
| t.copy(cumv) |
| t.xtr_D() |
| cumv.copy(cum2v); cumv.conj() |
| cum2v.copy(t); cum2v.conj() |
| t.copy(cv) |
| t.xtr_D() |
| cv.copy(cu) |
| cu.copy(t) |
| } |
| } |
| if BIG.comp(d,e)<0 |
| { |
| w.copy(d); w.imul(4); w.norm() |
| if BIG.comp(e,w)<=0 |
| { |
| e.sub(d); e.norm() |
| t.copy(cv) |
| t.xtr_A(cu,cumv,cum2v) |
| cum2v.copy(cumv) |
| cumv.copy(cu) |
| cu.copy(t) |
| } |
| else if e.parity()==0 |
| { |
| w.copy(d) |
| d.copy(e); d.fshr(1) |
| e.copy(w) |
| t.copy(cumv) |
| t.xtr_D() |
| cumv.copy(cum2v); cumv.conj() |
| cum2v.copy(t); cum2v.conj() |
| t.copy(cv) |
| t.xtr_D() |
| cv.copy(cu) |
| cu.copy(t) |
| } |
| else if d.parity()==1 |
| { |
| w.copy(e) |
| e.copy(d) |
| w.sub(d); w.norm() |
| d.copy(w); d.fshr(1) |
| t.copy(cv) |
| t.xtr_A(cu,cumv,cum2v) |
| cumv.conj() |
| cum2v.copy(cu) |
| cum2v.xtr_D() |
| cum2v.conj() |
| cu.copy(cv) |
| cu.xtr_D() |
| cv.copy(t) |
| } |
| else |
| { |
| d.fshr(1) |
| r.copy(cum2v); r.conj() |
| t.copy(cumv) |
| t.xtr_A(cu,cv,r) |
| cum2v.copy(cumv) |
| cum2v.xtr_D() |
| cumv.copy(t) |
| cu.xtr_D() |
| } |
| } |
| } |
| r.copy(cv) |
| r.xtr_A(cu,cumv,cum2v) |
| for _ in 0 ..< f2 |
| {r.xtr_D()} |
| r=r.xtr_pow(d) |
| return r |
| } |
| } |