From 12637dbae5f930c60b4850db64b298873b13095b Mon Sep 17 00:00:00 2001
From: Jarrett Revels <jarrettrevels@gmail.com>
Date: Mon, 25 Jan 2016 19:07:36 -0500
Subject: [PATCH] add SIMD benchmarks from JuliaLang/julia#13777 and
 JuliaLang/julia#13686

---
 src/simd/SIMDBenchmarks.jl | 110 +++++++++++++++++++++++++++++++++----
 1 file changed, 98 insertions(+), 12 deletions(-)

diff --git a/src/simd/SIMDBenchmarks.jl b/src/simd/SIMDBenchmarks.jl
index a0b4b465..fd6cbfa1 100644
--- a/src/simd/SIMDBenchmarks.jl
+++ b/src/simd/SIMDBenchmarks.jl
@@ -2,12 +2,13 @@ module SIMDBenchmarks
 
 import ..BaseBenchmarks
 using ..BenchmarkTrackers
+using ..BaseBenchmarks: samerand
 
 ###########
 # Methods #
 ###########
 
-function perf_simd_axpy!(a, X, Y)
+function perf_axpy!(a, X, Y)
     # LLVM's auto-vectorizer typically vectorizes this loop even without @simd
     @simd for i in eachindex(X)
         @inbounds Y[i] += a*X[i]
@@ -15,32 +16,117 @@ function perf_simd_axpy!(a, X, Y)
     return Y
 end
 
-function perf_simd_inner(x, y)
-    s = zero(eltype(x))
-    @simd for i in eachindex(x)
-        @inbounds s += x[i]*y[i]
+function perf_inner(X, Y)
+    s = zero(eltype(X))
+    @simd for i in eachindex(X)
+        @inbounds s += X[i]*Y[i]
     end
     return s
 end
 
-function perf_simd_sumreduce(x, i, j)
-    s = zero(eltype(x))
+function perf_sum_reduce(X)
+    s = zero(eltype(X))
+    i = div(length(X), 4)
+    j = div(length(X), 2)
     @simd for k in i:j
-        @inbounds s += x[k]
+        @inbounds s += X[k]
     end
     return s
 end
 
+function perf_manual_example!(X, Y, Z)
+    s = zero(eltype(Z))
+    n = min(length(X),length(Y),length(Z))
+    @simd for i in 1:n
+        @inbounds begin
+            Z[i] = X[i]-Y[i]
+            s += Z[i]*Z[i]
+        end
+    end
+    return s
+end
+
+function perf_two_reductions(X, Y, Z)
+    # Use non-zero initial value to make sure reduction values include it.
+    (s,t) = (one(eltype(X)), one(eltype(Y)))
+    @simd for i in 1:length(Z)
+        @inbounds begin
+            s += X[i]
+            t += 2*Y[i]
+            s += Z[i]   # Two reductions go into s
+        end
+    end
+    return (s,t)
+end
+
+function perf_conditional_loop!(X, Y, Z)
+    # SIMD loop with a long conditional expression
+    @simd for i=1:length(X)
+        @inbounds begin
+            X[i] = Y[i] * (Z[i] > Y[i]) * (Z[i] < Y[i]) * (Z[i] >= Y[i]) * (Z[i] <= Y[i])
+        end
+    end
+    return X
+end
+
+function perf_local_arrays(V)
+    # SIMD loop on local arrays declared without type annotations
+    T, n = eltype(V), length(V)
+    X = samerand(T, n)
+    Y = samerand(T, n)
+    Z = samerand(T, n)
+    @simd for i in eachindex(X)
+        @inbounds X[i] = Y[i] * Z[i]
+    end
+    return X
+end
+
+immutable ImmutableFields{V<:AbstractVector}
+    X::V
+    Y::V
+    Z::V
+end
+
+ImmutableFields{V}(X::V) = ImmutableFields{V}(X, X, X)
+
+type MutableFields{V<:AbstractVector}
+    X::V
+    Y::V
+    Z::V
+end
+
+MutableFields{V}(X::V) = ImmutableFields{V}(X, X, X)
+
+function perf_loop_fields!(obj)
+    # SIMD loop with field access
+    @simd for i = 1:length(obj.X)
+        @inbounds obj.X[i] = obj.Y[i] * obj.Z[i]
+    end
+    return obj
+end
+
 ##############
 # Benchmarks #
 ##############
 
 @track BaseBenchmarks.TRACKER "simd" begin
-    @setup vectors = map(T -> BaseBenchmarks.samerand(T, 1000), (Float32, Float64))
+    @setup begin
+        lens = (9, 10, 255, 256, 999, 1000)
+        int32_vecs = map(n -> samerand(Int32, n), lens)
+        int64_vecs = map(n -> samerand(Int64, n), lens)
+        float32_vecs = map(n -> samerand(Float32, n), lens)
+        float64_vecs = map(n -> samerand(Float64, n), lens)
+        vectors = (int32_vecs..., int64_vecs..., float32_vecs..., float64_vecs...)
+    end
     @benchmarks begin
-        [(:axpy!, string(eltype(v))) => perf_simd_axpy!(first(v), v, copy(v)) for v in vectors]
-        [(:inner, string(eltype(v))) => perf_simd_inner(v, v) for v in vectors]
-        [(:sumreduce, string(eltype(v))) => perf_simd_sumreduce(v, 500, 700) for v in vectors]
+        [(:axpy!, string(eltype(v)), length(v)) => perf_axpy!(first(v), v, copy(v)) for v in vectors]
+        [(:inner, string(eltype(v)), length(v)) => perf_inner(v, v) for v in vectors]
+        [(:sum_reduce, string(eltype(v)), length(v)) => perf_sum_reduce(v) for v in vectors]
+        [(:manual_example!, string(eltype(v)), length(v)) => perf_manual_example!(v, v, v) for v in vectors]
+        [(:two_reductions, string(eltype(v)), length(v)) => perf_two_reductions(v, v, v) for v in vectors]
+        [(:conditional_loop!, string(eltype(v)), length(v)) => perf_conditional_loop!(v, v, v) for v in vectors]
+        [(:local_arrays, string(eltype(v)), length(v)) => perf_local_arrays(v) for v in vectors]
+        [(:loop_fields!, T, string(eltype(v)), length(v)) => perf_loop_fields!(T(v)) for v in vectors, T in (MutableFields, ImmutableFields)]
     end
     @tags "array" "inbounds" "mul" "axpy!" "inner" "sum" "reduce"
 end