Merge pull request #15 from jw3126/functional

Add functional variants.

Author: sunoru

src/Random123.jl

@@ -21,10 +21,10 @@ using RandomNumbers
 export set_counter!
 include("common.jl")
 
-export Threefry2x, Threefry4x
+export Threefry2x, Threefry4x, threefry
 include("threefry.jl")
 
-export Philox2x, Philox4x
+export Philox2x, Philox4x, philox
 include("philox.jl")
 
 export R123_USE_AESNI
@@ -46,8 +46,8 @@ catch e
 end
 
 @static if R123_USE_AESNI
-    export AESNI1x, AESNI4x
-    export ARS1x, ARS4x
+    export AESNI1x, AESNI4x, aesni
+    export ARS1x, ARS4x, ars
     include("./aesni_common.jl")
     include("./aesni.jl")
     include("./ars.jl")

src/aesni.jl

@@ -201,26 +201,51 @@ copy(src::AESNI4x) = copyto!(AESNI4x(), src)
 ==(r1::AESNI4x, r2::AESNI4x) = unsafe_compare(r1, r2, UInt128, 2) &&
     r1.key == r2.key && r1.p == r2.p
 
-function aesni1xm128i(input::__m128i, key::AESNIKey)
-    x = key.key1 ⊻ input
-    x = _aes_enc(x, key.key2)
-    x = _aes_enc(x, key.key3)
-    x = _aes_enc(x, key.key4)
-    x = _aes_enc(x, key.key5)
-    x = _aes_enc(x, key.key6)
-    x = _aes_enc(x, key.key7)
-    x = _aes_enc(x, key.key8)
-    x = _aes_enc(x, key.key9)
-    x = _aes_enc(x, key.key10)
-    x = _aes_enc_last(x, key.key11)
+function get_key__m128i(o::Union{AESNI1x, AESNI4x})::NTuple{11, __m128i}
+    k = o.key
+    (k.key1,k.key2,k.key3,k.key4,k.key5,k.key6,k.key7,k.key8,k.key9,k.key10,k.key11)
 end
+get_ctr__m128i(o::AESNI4x)::Tuple{__m128i} = (o.ctr1,)
+get_ctr__m128i(o::AESNI1x)::Tuple{__m128i} = (o.ctr,)
+get_key(o::Union{AESNI1x, AESNI4x})::NTuple{11,UInt128} = map(UInt128, get_key__m128i(o))
+get_ctr(o::Union{AESNI1x, AESNI4x})::Tuple{UInt128} = map(UInt128, get_ctr__m128i(o))
+
+@inline function aesni(key::NTuple{11,__m128i}, ctr::Tuple{__m128i})::Tuple{__m128i}
+    key1, key2, key3, key4, key5, key6, key7, key8, key9, key10, key11 = key
+    ctr1 = only(ctr)
+    x = key1 ⊻ ctr1
+    x = _aes_enc(x, key2)
+    x = _aes_enc(x, key3)
+    x = _aes_enc(x, key4)
+    x = _aes_enc(x, key5)
+    x = _aes_enc(x, key6)
+    x = _aes_enc(x, key7)
+    x = _aes_enc(x, key8)
+    x = _aes_enc(x, key9)
+    x = _aes_enc(x, key10)
+    x = _aes_enc_last(x, key11)
+    (x,)
+end
+
+"""
+    aesni(key::NTuple{11,UInt128}, ctr::Tuple{UInt128})::Tuple{UInt128}
+
+Functional variant of [`AESNI1x`](@ref) and [`AESNI4x`](@ref). 
+This function if free of mutability and side effects.
+"""
+@inline function aesni(key::NTuple{11,UInt128}, ctr::Tuple{UInt128})::Tuple{UInt128}
+    k = map(__m128i, key)
+    c = map(__m128i, ctr)
+    map(UInt128,aesni(k,c))
+end
+
 
 @inline function random123_r(r::AESNI1x)
-    r.x = aesni1xm128i(r.ctr, r.key)
+    r.x = only(aesni(get_key__m128i(r), get_ctr__m128i(r)))
     (UInt128(r.x),)
 end
 
 @inline function random123_r(r::AESNI4x)
-    r.x = aesni1xm128i(r.ctr1, r.key)
+    r.x = only(aesni(get_key__m128i(r), get_ctr__m128i(r)))
     split_uint(UInt128(r.x), UInt32)
 end

src/ars.jl

@@ -98,19 +98,15 @@ copy(src::ARS4x{R}) where R = ARS4x{R}(src.x, src.ctr1, src.key, src.p)
 
 ==(r1::ARS4x{R}, r2::ARS4x{R}) where R = unsafe_compare(r1, r2, UInt128, 3) && r1.p ≡ r2.p
 
-@generated function ars1xm128i(r::Union{ARS1x{R}, ARS4x{R}}) where R
+function expr_ars1xm128i(expr_key, expr_ctr, R)
     @assert R isa Int && 1 ≤ R ≤ 10
     rounds = [quote
         kk += kweyl
         v = _aes_enc(v, kk)
     end for _ in 2:R]
-    ctr = :(r.ctr)
-    if r <: ARS4x
-        ctr.args[2] = :(:ctr1)
-    end
     quote
-        ctr = $ctr
-        key = r.key
+        ctr = $(expr_ctr)
+        key = $(expr_key)
         kweyl = __m128i(0xbb67ae8584caa73b, 0x9e3779b97f4a7c15)
         kk = key
         v = ctr ⊻ kk
@@ -122,12 +118,43 @@ copy(src::ARS4x{R}) where R = ARS4x{R}(src.x, src.ctr1, src.key, src.p)
     end
 end
 
+@generated function ars1xm128i(r::Union{ARS1x{R}, ARS4x{R}}) where R
+    expr_ctr = if r <: ARS1x
+        :(r.ctr)
+    elseif r <: ARS4x
+        :(r.ctr1)
+    else
+        :(error("Unreachable"))
+    end
+    expr_key = :(r.key)
+    expr_ars1xm128i(expr_key, expr_ctr, R)
+end
+
+@generated function ars(key::Tuple{__m128i}, ctr::Tuple{__m128i}, ::Val{R})::Tuple{__m128i} where {R}
+    :(($(expr_ars1xm128i(:(only(key)), :(only(ctr)), R)),))
+end
+
+"""
+    ars(key::Tuple{UInt128}, ctr::Tuple{UInt128}, rounds::Val{R})::Tuple{UInt128} where {R}
+
+Functional variant of [`ARS1x`](@ref) and [`ARS4x`](@ref). 
+This function if free of mutability and side effects.
+"""
+function ars(key::Tuple{UInt128}, ctr::Tuple{UInt128}, rounds::Val{R})::Tuple{UInt128} where {R}
+    k = map(__m128i, key)
+    c = map(__m128i, ctr)
+    map(UInt128,ars(k,c,rounds))
+end
+
+get_key(r::Union{ARS1x, ARS4x}) = (UInt128(r.key),)
+get_ctr(r::ARS1x) = (UInt128(r.ctr),)
+get_ctr(r::ARS4x) = (UInt128(r.ctr1),)
+
 @inline function random123_r(r::ARS1x{R}) where R
     r.x = ars1xm128i(r)
     (UInt128(r.x),)
 end
 
-
 @inline function random123_r(r::ARS4x{R}) where R
     r.x = ars1xm128i(r)
     split_uint(UInt128(r.x), UInt32)

src/philox.jl

@@ -89,8 +89,24 @@ end
     key + PHILOX_W_0(T)
 end
 
+@inline get_key(r::Philox2x) = (r.key,)
+@inline get_ctr(r::Philox2x) = (r.ctr1, r.ctr2)
+
 @inline function random123_r(r::Philox2x{T, R}) where {T <: Union{UInt32, UInt64}, R}
-    ctr1, ctr2, key = r.ctr1, r.ctr2, r.key
+    r.x1, r.x2 = philox(get_key(r), get_ctr(r), Val(R))
+end
+
+"""
+    philox(key::NTuple{1,T}, ctr::NTuple{2,T}, ::Val{R})::NTuple{2,T}
+    philox(key::NTuple{2,T}, ctr::NTuple{4,T}, ::Val{R})::NTuple{4,T}
+
+Functional variant of [`Philox2x`](@ref) and [`Philox4x`](@ref). 
+Produces a pseudorandom output of type `T = UInt64` or `T = UInt32` from the inputs.
+This function if free of mutability and side effects.
+"""
+@inline function philox(key_::Tuple{T}, ctr::NTuple{2,T}, ::Val{R}) where {T,R}
+    key = first(key_)
+    ctr1, ctr2 = ctr
     if R > 0                               ctr1, ctr2 = philox2x_round(ctr1, ctr2, key); end
     if R > 1  key = philox2x_bumpkey(key); ctr1, ctr2 = philox2x_round(ctr1, ctr2, key); end
     if R > 2  key = philox2x_bumpkey(key); ctr1, ctr2 = philox2x_round(ctr1, ctr2, key); end
@@ -107,7 +123,7 @@ end
     if R > 13 key = philox2x_bumpkey(key); ctr1, ctr2 = philox2x_round(ctr1, ctr2, key); end
     if R > 14 key = philox2x_bumpkey(key); ctr1, ctr2 = philox2x_round(ctr1, ctr2, key); end
     if R > 15 key = philox2x_bumpkey(key); ctr1, ctr2 = philox2x_round(ctr1, ctr2, key); end
-    r.x1, r.x2 = ctr1, ctr2
+    ctr1, ctr2
 end
 
 """
@@ -183,9 +199,15 @@ end
     key1 + PHILOX_W_0(T), key2 + PHILOX_W_1(T)
 end
 
+@inline get_ctr(r::Philox4x) = (r.ctr1, r.ctr2, r.ctr3, r.ctr4)
+@inline get_key(r::Philox4x) = (r.key1, r.key2)
 @inline function random123_r(r::Philox4x{T, R}) where {T <: Union{UInt32, UInt64}, R}
-    ctr1, ctr2, ctr3, ctr4 = r.ctr1, r.ctr2, r.ctr3, r.ctr4
-    key1, key2 = r.key1, r.key2
+    r.x1, r.x2, r.x3, r.x4 = philox(get_key(r), get_ctr(r), Val(R))
+end
+
+@inline function philox(key::NTuple{2,T}, ctr::NTuple{4,T}, ::Val{R}) where {T <:Union{UInt32, UInt64}, R}
+    ctr1, ctr2, ctr3, ctr4 = ctr
+    key1, key2 = key
     if R > 0
         ctr1, ctr2, ctr3, ctr4 = philox4x_round(ctr1, ctr2, ctr3, ctr4, key1, key2);
     end
@@ -249,5 +271,5 @@ end
         key1, key2 = philox4x_bumpkey(key1, key2);
         ctr1, ctr2, ctr3, ctr4 = philox4x_round(ctr1, ctr2, ctr3, ctr4, key1, key2);
     end
-    r.x1, r.x2, r.x3, r.x4 = ctr1, ctr2, ctr3, ctr4
+    ctr1, ctr2, ctr3, ctr4
 end

src/threefry.jl

@@ -116,13 +116,25 @@ copy(src::Threefry2x{T, R}) where {T, R} = Threefry2x{T, R}(src.x1, src.x2, src.
 
 ==(r1::Threefry2x{T, R}, r2::Threefry2x{T, R}) where {T, R} = unsafe_compare(r1, r2, T, 6) && r1.p == r2.p
 
+@inline get_key(r::Threefry2x) = (r.key1, r.key2)
+@inline get_ctr(r::Threefry2x) = (r.ctr1, r.ctr2)
 @inline function random123_r(r::Threefry2x{T, R}) where {T <: Union{UInt32, UInt64}, R}
+    r.x1, r.x2 = threefry(get_key(r), get_ctr(r), Val(R))
+end
+
+"""
+    threefry(key::NTuple{2,T}, ctr::NTuple{2,T}, ::Val{R})::NTuple{2,T}
+    threefry(key::NTuple{4,T}, ctr::NTuple{4,T}, ::Val{R})::NTuple{4,T}
+
+Functional variant of [`Threefry2x`](@ref) and [`Threefry4x`](@ref). 
+Produces a pseudorandom output of type `T = UInt64` or `T = UInt32` from the inputs.
+This function if free of mutability and side effects.
+"""
+@inline function threefry(key::NTuple{2,T}, ctr::NTuple{2,T}, ::Val{R})::NTuple{2,T} where {T <: Union{UInt32, UInt64}, R}
     ks2 = SKEIN_KS_PARITY(T)
-    ks0 = r.key1
-    x0 = r.ctr1
+    x0,x1 = ctr
+    ks0,ks1 = key
     ks2 ⊻= ks0
-    ks1 = r.key2
-    x1 = r.ctr2
     ks2 ⊻= ks1
     x0 += ks0
     x1 += ks1
@@ -191,7 +203,7 @@ copy(src::Threefry2x{T, R}) where {T, R} = Threefry2x{T, R}(src.x1, src.x2, src.
         x0 += ks2; x1 += ks0;
         x1 += 8 % T;
     end
-    r.x1, r.x2 = x0, x1
+    x0, x1
 end
 
 """
@@ -257,19 +269,20 @@ copy(src::Threefry4x{T, R}) where {T, R} = Threefry4x{T, R}(src.x1, src.x2, src.
 
 ==(r1::Threefry4x{T, R}, r2::Threefry4x{T, R}) where {T, R} = unsafe_compare(r1, r2, T, 12) && r1.p == r2.p
 
+@inline get_key(r::Threefry4x) = (r.key1, r.key2, r.key3, r.key4)
+@inline get_ctr(r::Threefry4x) = (r.ctr1, r.ctr2, r.ctr3, r.ctr4)
+
 @inline function random123_r(r::Threefry4x{T, R}) where {T <: Union{UInt32, UInt64}, R}
+    r.x1, r.x2, r.x3, r.x4 = threefry(get_key(r), get_ctr(r), Val(R))
+end
+
+@inline function threefry(key::NTuple{4,T},ctr::NTuple{4,T}, rounds::Val{R})::NTuple{4,T} where {T <: Union{UInt32, UInt64}, R}
     ks4 = SKEIN_KS_PARITY(T)
-    ks0 = r.key1
-    x0 = r.ctr1
+    ks0,ks1,ks2,ks3 = key
+    x0,x1,x2,x3 = ctr
     ks4 ⊻= ks0
-    ks1 = r.key2
-    x1 = r.ctr2
     ks4 ⊻= ks1
-    ks2 = r.key3
-    x2 = r.ctr3
     ks4 ⊻= ks2
-    ks3 = r.key4
-    x3 = r.ctr4
     ks4 ⊻= ks3
     x0 += ks0; x1 += ks1; x2 += ks2; x3 += ks3;
 
@@ -633,5 +646,5 @@ copy(src::Threefry4x{T, R}) where {T, R} = Threefry4x{T, R}(src.x1, src.x2, src.
         x0 += ks3; x1 += ks4; x2 += ks0; x3 += ks1;
         x3 += 18 % T;
     end
-    r.x1, r.x2, r.x3, r.x4 = x0, x1, x2, x3
+    x0, x1, x2, x3
 end