From 5dd180bb66b2c5bb040ac4b59892064e59456262 Mon Sep 17 00:00:00 2001
From: guoyongzhi <guo-yong-zhi@outlook.com>
Date: Mon, 11 Nov 2024 19:26:11 +0800
Subject: [PATCH] optimize SVector4 and improve type stability

---
 src/common_datatypes.jl | 22 +++++++++++++---------
 src/qtree_functions.jl  |  4 ++--
 src/qtrees.jl           |  4 ++--
 3 files changed, 17 insertions(+), 13 deletions(-)

diff --git a/src/common_datatypes.jl b/src/common_datatypes.jl
index 875a72f..b7dc855 100644
--- a/src/common_datatypes.jl
+++ b/src/common_datatypes.jl
@@ -187,27 +187,31 @@ Base.getindex(im::IntMap, ind...) = getindex(im.map, ind...)
 Base.setindex!(im::IntMap, v, ind...) = setindex!(im.map, v, ind...)
 
 ##### SVector4
-mutable struct SVector4{T} 
+mutable struct SVector4Core{T} 
     e1::T
     e2::T
     e3::T
     e4::T
+    SVector4Core{T}() where T = new()
+end
+mutable struct SVector4{T} 
+    vec::SVector4Core{T}
     len::Int
-    SVector4{T}() where T = (v = new(); v.len = 0; v)
 end
+SVector4{T}() where T = SVector4{T}(SVector4Core{T}(), 0)
 function SVector4{T}(e1, e2, e3, e4) where T
     v = SVector4{T}()
-    v.e1 = e1
-    v.e2 = e2
-    v.e3 = e3
-    v.e4 = e4
+    v.vec.e1 = e1
+    v.vec.e2 = e2
+    v.vec.e3 = e3
+    v.vec.e4 = e4
     v.len = 4
     v
 end
 SVector4(e1::T, e2::T, e3::T, e4::T) where T = SVector4{T}(e1, e2, e3, e4)
-Base.getindex(v::SVector4, i) = getfield(v, i)
-Base.setindex!(v::SVector4, x, i) = setfield!(v, i, x)
-Base.push!(v::SVector4, x) = v[v.len += 1] = x
+Base.getindex(v::SVector4{T}, i) where T = getfield(v.vec, i)::T
+Base.setindex!(v::SVector4{T}, x::T, i) where T = setfield!(v.vec, i, x)
+Base.push!(v::SVector4{T}, x::T) where T = v[v.len += 1] = x
 Base.length(v::SVector4) = v.len
 Base.iterate(v::SVector4, i=1) = i <= length(v) ? (v[i], i+1) : nothing
 Base.eltype(::Type{SVector4{T}}) where T = T
diff --git a/src/qtree_functions.jl b/src/qtree_functions.jl
index d67a875..f173c3d 100644
--- a/src/qtree_functions.jl
+++ b/src/qtree_functions.jl
@@ -1,5 +1,5 @@
 ########## totalcollisions
-function collision_dfs(Q1::AbstractStackedQTree, Q2::AbstractStackedQTree, i=(length(Q1), 1, 1)) #faster than _collision_randbfs (6:7)
+function collision_dfs(Q1::AbstractStackedQTree, Q2::AbstractStackedQTree, i::Index=(length(Q1), 1, 1)) #faster than _collision_randbfs (6:7)
     # @assert size(Q1) == size(Q2)
     code = Q1[i] & Q2[i]
     if code == FULL # Q1[i] == FULL || Q2[i] == FULL && (Q1[i] != EMPTY && Q2[i] != EMPTY)
@@ -138,7 +138,7 @@ function linked_spacial_qtree(qts)
 end
 hash_spacial_qtree() = HashSpacialQTree()
 hash_spacial_qtree(qts) = hash_spacial_qtree()
-function positionlower(qt, ind)
+function positionlower(qt, ind::Index)
     while ind[1] > 2
         cind = ind
         flag = false
diff --git a/src/qtrees.jl b/src/qtrees.jl
index 4b2d07c..4ad8687 100644
--- a/src/qtrees.jl
+++ b/src/qtrees.jl
@@ -119,8 +119,8 @@ function PaddedMat(l::T, sz::Tuple{Int,Int}=size(l), rshift=0, cshift=0; default
     m.kernel[2:end - 2, 2:end - 2] .= l
     m
 end
-function PaddedMat{T}(kernelsz::Tuple{Int,Int}, sz::Tuple{Int,Int}=size(l), 
-    rshift=0, cshift=0; default=0x00) where {T <: AbstractMatrix{UInt8}}
+function PaddedMat{T}(kernelsz::Tuple{Int,Int}, sz::Tuple{Int,Int}, 
+    rshift=0, cshift=0; default=zero(T)) where {T <: AbstractMatrix{UInt8}}
     k = similar(T, kernelsz .+ 3) # +3 to keep top-down getindex in _collision_randbfs within the kernel bounds
     k[[1, end - 1, end], :] .= default
     k[:, [1, end - 1, end]] .= default