From 2eeef2e231a55cac770543b6dd673e349adfd797 Mon Sep 17 00:00:00 2001
From: Oscar Smith <oscardssmith@gmail.com>
Date: Sun, 14 Feb 2021 12:01:49 -0600
Subject: [PATCH] faster, more accurate log2, log10 (#39556)

* faster, more accurate log2, log10

* fix error message for negative numbers

* fix rebase

* Maybe fix doctest for domainerror

* Fix accidental change

* don't use metaprogramming

* fix typo
---
 base/math.jl        | 14 ++++---------
 base/special/log.jl | 51 +++++++++++++++++++++++++++++++--------------
 2 files changed, 39 insertions(+), 26 deletions(-)

diff --git a/base/math.jl b/base/math.jl
index 39f7367329708..42f10760ed4fb 100644
--- a/base/math.jl
+++ b/base/math.jl
@@ -475,9 +475,9 @@ julia> log2(10)
 
 julia> log2(-2)
 ERROR: DomainError with -2.0:
-NaN result for non-NaN input.
+log2 will only return a complex result if called with a complex argument. Try log2(Complex(x)).
 Stacktrace:
- [1] nan_dom_err at ./math.jl:325 [inlined]
+ [1] throw_complex_domainerror(f::Symbol, x::Float64) at ./math.jl:31
 [...]
 ```
 """
@@ -499,9 +499,9 @@ julia> log10(2)
 
 julia> log10(-2)
 ERROR: DomainError with -2.0:
-NaN result for non-NaN input.
+log10 will only return a complex result if called with a complex argument. Try log10(Complex(x)).
 Stacktrace:
- [1] nan_dom_err at ./math.jl:325 [inlined]
+ [1] throw_complex_domainerror(f::Symbol, x::Float64) at ./math.jl:31
 [...]
 ```
 """
@@ -530,12 +530,6 @@ Stacktrace:
 ```
 """
 log1p(x)
-for f in (:log2, :log10)
-    @eval begin
-        @inline ($f)(x::Float64) = nan_dom_err(ccall(($(string(f)), libm), Float64, (Float64,), x), x)
-        @inline ($f)(x::Float32) = nan_dom_err(ccall(($(string(f, "f")), libm), Float32, (Float32,), x), x)
-    end
-end
 
 @inline function sqrt(x::Union{Float32,Float64})
     x < zero(x) && throw_complex_domainerror(:sqrt, x)
diff --git a/base/special/log.jl b/base/special/log.jl
index caa41e7ec43ac..95bc9e32b5719 100644
--- a/base/special/log.jl
+++ b/base/special/log.jl
@@ -149,9 +149,18 @@ const FMA_NATIVE = muladd(nextfloat(1.0),nextfloat(1.0),-nextfloat(1.0,2)) != 0
     reinterpret(Float64, reinterpret(UInt64,x) & 0xffff_ffff_f800_0000)
 end
 
+logb(::Type{Float32},::Val{2})  = 1.4426950408889634
+logb(::Type{Float32},::Val{:ℯ}) = 1.0
+logb(::Type{Float32},::Val{10}) = 0.4342944819032518
+logbU(::Type{Float64},::Val{2})  = 1.4426950408889634
+logbL(::Type{Float64},::Val{2})  = 2.0355273740931033e-17
+logbU(::Type{Float64},::Val{:ℯ}) = 1.0
+logbL(::Type{Float64},::Val{:ℯ}) = 0.0
+logbU(::Type{Float64},::Val{10}) = 0.4342944819032518
+logbL(::Type{Float64},::Val{10}) = 1.098319650216765e-17
 
 # Procedure 1
-@inline function log_proc1(y::Float64,mf::Float64,F::Float64,f::Float64,jp::Int)
+@inline function log_proc1(y::Float64,mf::Float64,F::Float64,f::Float64,jp::Int,base=Val(:ℯ))
     ## Steps 1 and 2
     @inbounds hi,lo = t_log_Float64[jp]
     l_hi = mf* 0.6931471805601177 + hi
@@ -175,11 +184,13 @@ end
                     0.012500053168098584)
 
     ## Step 4
-    l_hi + (u + (q + l_lo))
+    m_hi = logbU(Float64, base)
+    m_lo = logbL(Float64, base)
+    return fma(m_hi, l_hi, fma(m_hi, (u + (q + l_lo)), m_lo*l_hi))
 end
 
 # Procedure 2
-@inline function log_proc2(f::Float64)
+@inline function log_proc2(f::Float64,base=Val(:ℯ))
     ## Step 1
     g = 1.0/(2.0+f)
     u = 2.0*f*g
@@ -206,12 +217,14 @@ end
         f2 = f-f1
         u2 = ((2.0*(f-u1)-u1*f1)-u1*f2)*g
         ## Step 4
-        return u1 + (u2 + q)
+        m_hi = logbU(Float64, base)
+        m_lo = logbL(Float64, base)
+        return fma(m_hi, u1, fma(m_hi, (u2 + q), m_lo*u1))
     end
 end
 
 
-@inline function log_proc1(y::Float32,mf::Float32,F::Float32,f::Float32,jp::Int)
+@inline function log_proc1(y::Float32,mf::Float32,F::Float32,f::Float32,jp::Int,base=Val(:ℯ))
     ## Steps 1 and 2
     @inbounds hi = t_log_Float32[jp]
     l = mf*0.6931471805599453 + hi
@@ -228,10 +241,10 @@ end
     q = u*v*0.08333351f0
 
     ## Step 4
-    Float32(l + (u + q))
+    Float32(logb(Float32, base)*(l + (u + q)))
 end
 
-@inline function log_proc2(f::Float32)
+@inline function log_proc2(f::Float32,base=Val(:ℯ))
     ## Step 1
     # compute in higher precision
     u64 = Float64(2f0*f)/(2.0+f)
@@ -246,18 +259,24 @@ end
     ## Step 3: not required
 
     ## Step 4
-    Float32(u64 + q)
+    Float32(logb(Float32, base)*(u64 + q))
 end
 
+log2(x::Float32)  = _log(x, Val(2),  :log2)
+log(x::Float32)   = _log(x, Val(:ℯ), :log)
+log10(x::Float32) = _log(x, Val(10), :log10)
+log2(x::Float64)  = _log(x, Val(2),  :log2)
+log(x::Float64)   = _log(x, Val(:ℯ), :log)
+log10(x::Float64) = _log(x, Val(10), :log10)
 
-function log(x::Float64)
+function _log(x::Float64, base, func)
     if x > 0.0
         x == Inf && return x
 
         # Step 2
         if 0.9394130628134757 < x < 1.0644944589178595
             f = x-1.0
-            return log_proc2(f)
+            return log_proc2(f, base)
         end
 
         # Step 3
@@ -276,24 +295,24 @@ function log(x::Float64)
         f = y-F
         jp = unsafe_trunc(Int,128.0*F)-127
 
-        return log_proc1(y,mf,F,f,jp)
+        return log_proc1(y,mf,F,f,jp,base)
     elseif x == 0.0
         -Inf
     elseif isnan(x)
         NaN
     else
-        throw_complex_domainerror(:log, x)
+        throw_complex_domainerror(func, x)
     end
 end
 
-function log(x::Float32)
+function _log(x::Float32, base, func)
     if x > 0f0
         x == Inf32 && return x
 
         # Step 2
         if 0.939413f0 < x < 1.0644945f0
             f = x-1f0
-            return log_proc2(f)
+            return log_proc2(f, base)
         end
 
         # Step 3
@@ -312,13 +331,13 @@ function log(x::Float32)
         f = y-F
         jp = unsafe_trunc(Int,128.0f0*F)-127
 
-        log_proc1(y,mf,F,f,jp)
+        log_proc1(y,mf,F,f,jp,base)
     elseif x == 0f0
         -Inf32
     elseif isnan(x)
         NaN32
     else
-        throw_complex_domainerror(:log, x)
+        throw_complex_domainerror(func, x)
     end
 end