Update compat and versions, add intel compiler to benchmarks.

Author: chriselrod

Manifest.toml

@@ -11,9 +11,9 @@ version = "0.2.2"
 
 [[DataStructures]]
 deps = ["InteractiveUtils", "OrderedCollections"]
-git-tree-sha1 = "f784254f428fb8fd7ac15982e5862a38a44523d3"
+git-tree-sha1 = "b7720de347734f4716d1815b00ce5664ed6bbfd4"
 uuid = "864edb3b-99cc-5e75-8d2d-829cb0a9cfe8"
-version = "0.17.7"
+version = "0.17.9"
 
 [[Distributed]]
 deps = ["Random", "Serialization", "Sockets"]
@@ -61,15 +61,15 @@ uuid = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 
 [[SIMDPirates]]
 deps = ["MacroTools", "VectorizationBase"]
-git-tree-sha1 = "c0f42ddb2645c54b8620979df5dc979c4742db59"
+git-tree-sha1 = "910193d289b41e570118c4e444f0c05cc700a2f7"
 uuid = "21efa798-c60a-11e8-04d3-e1a92915a26a"
-version = "0.1.4"
+version = "0.1.5"
 
 [[SLEEFPirates]]
 deps = ["SIMDPirates", "VectorizationBase"]
-git-tree-sha1 = "547bcf7d30967d87d4c62b3fe5efdb0e57a6e436"
+git-tree-sha1 = "4733445246d3d5536c7aee1bffb55ab37b88347b"
 uuid = "476501e8-09a2-5ece-8869-fb82de89a1fa"
-version = "0.1.2"
+version = "0.1.3"
 
 [[Serialization]]
 uuid = "9e88b42a-f829-5b0c-bbe9-9e923198166b"
@@ -83,6 +83,6 @@ uuid = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [[VectorizationBase]]
 deps = ["CpuId", "LinearAlgebra"]
-git-tree-sha1 = "81c1b3171d93e64345d75a9f08d190a155e9f009"
+git-tree-sha1 = "a2576763aa20968ffb5668e2e15d45ae8e364d05"
 uuid = "3d5dd08c-fd9d-11e8-17fa-ed2836048c2f"
-version = "0.1.7"
+version = "0.1.9"

Project.toml

@@ -14,9 +14,9 @@ VectorizationBase = "3d5dd08c-fd9d-11e8-17fa-ed2836048c2f"
 [compat]
 MacroTools = "0.5"
 Parameters = "0.12.0"
-SIMDPirates = "0.1.4"
-SLEEFPirates = "0.1.2"
-VectorizationBase = "0.1.7"
+SIMDPirates = "0.1.5"
+SLEEFPirates = "0.1.3"
+VectorizationBase = "0.1.9"
 julia = "1.3.0"
 
 [extras]

benchmark/benchmarkflops.jl

@@ -39,7 +39,7 @@ end
 tothreetuple(i::Int) = (i,i,i)
 tothreetuple(i::NTuple{3,Int}) = i
 function benchmark_gemm(sizes)
-    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFort-loops", "GFort-intrinsic", "LoopVectorization"]
+    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFortran", "GFort-intrinsic", "icc", "ifort", "ifort-intrinsic", "LoopVectorization"]
     br = BenchmarkResult(tests, sizes)
     for (i,s) ∈ enumerate(sizes)
         M, K, N = tothreetuple(s)
@@ -57,7 +57,13 @@ function benchmark_gemm(sizes)
         @assert C ≈ Cblas "Fort gemm wrong?"
         br[5,i] = n_gflop / @belapsed fgemm_builtin!($C, $A, $B)
         @assert C ≈ Cblas "Fort intrinsic gemm wrong?"
-        br[6,i] = n_gflop / @belapsed gemmavx!($C, $A, $B)
+        br[6,i] = n_gflop / @belapsed icgemm_nkm!($C, $A, $B)
+        @assert C ≈ Cblas "icc gemm wrong?"
+        br[7,i] = n_gflop / @belapsed ifgemm_nkm!($C, $A, $B)
+        @assert C ≈ Cblas "ifort gemm wrong?"
+        br[8,i] = n_gflop / @belapsed ifgemm_builtin!($C, $A, $B)
+        @assert C ≈ Cblas "ifort intrinsic gemm wrong?"
+        br[9,i] = n_gflop / @belapsed gemmavx!($C, $A, $B)
         @assert C ≈ Cblas "LoopVec gemm wrong?"
         # if i % 10 == 0
             # percent_complete = round(100i/ length(sizes), sigdigits = 4)
@@ -67,7 +73,7 @@ function benchmark_gemm(sizes)
     br
 end
 function benchmark_AtmulB(sizes)
-    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFort-loops", "GFort-intrinsic", "LoopVectorization"]
+    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFortran", "GFort-intrinsic", "icc", "ifort", "ifort-intrinsic", "LoopVectorization"]
     br = BenchmarkResult(tests, sizes)
     for (i,s) ∈ enumerate(sizes)
         M, K, N = tothreetuple(s)
@@ -85,7 +91,13 @@ function benchmark_AtmulB(sizes)
         @assert C ≈ Cblas "Fort gemm wrong?"
         br[5,i] = n_gflop / @belapsed fAtmulB_builtin!($C, $At, $B)
         @assert C ≈ Cblas "Fort intrinsic gemm wrong?"
-        br[6,i] = n_gflop / @belapsed jAtmulBavx!($C, $At, $B)
+        br[6,i] = n_gflop / @belapsed cAtmulB!($C, $At, $B)
+        @assert C ≈ Cblas "icc gemm wrong?"
+        br[7,i] = n_gflop / @belapsed ifAtmulB!($C, $At, $B)
+        @assert C ≈ Cblas "iort gemm wrong?"
+        br[8,i] = n_gflop / @belapsed ifAtmulB_builtin!($C, $At, $B)
+        @assert C ≈ Cblas "ifort intrinsic gemm wrong?"
+        br[9,i] = n_gflop / @belapsed jAtmulBavx!($C, $At, $B)
         @assert C ≈ Cblas "LoopVec gemm wrong?"
         # if i % 10 == 0
             # percent_complete = round(100i/ length(sizes), sigdigits = 4)
@@ -96,7 +108,7 @@ function benchmark_AtmulB(sizes)
 end
 
 function benchmark_dot(sizes)
-    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFort-loops", "LoopVectorization"]
+    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFortran", "icc", "ifort", "LoopVectorization"]
     br = BenchmarkResult(tests, sizes)
     for (i,s) ∈ enumerate(sizes)
         a = rand(s); b = rand(s);
@@ -109,7 +121,11 @@ function benchmark_dot(sizes)
         @assert cdot(a,b) ≈ dotblas "Polly dot wrong?"
         br[4,i] = n_gflop / @belapsed fdot($a, $b)
         @assert fdot(a,b) ≈ dotblas "Fort dot wrong?"
-        br[5,i] = n_gflop / @belapsed jdotavx($a, $b)
+        br[5,i] = n_gflop / @belapsed icdot($a, $b)
+        @assert cdot(a,b) ≈ dotblas "icc dot wrong?"
+        br[6,i] = n_gflop / @belapsed ifdot($a, $b)
+        @assert fdot(a,b) ≈ dotblas "ifort dot wrong?"
+        br[7,i] = n_gflop / @belapsed jdotavx($a, $b)
         @assert jdotavx(a,b) ≈ dotblas "LoopVec dot wrong?"
         # if i % 10 == 0
             # percent_complete = round(100i/ length(sizes), sigdigits = 4)
@@ -119,7 +135,7 @@ function benchmark_dot(sizes)
     br
 end
 function benchmark_selfdot(sizes)
-    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFort-loops", "LoopVectorization"]
+    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFortran", "icc", "ifort", "LoopVectorization"]
     br = BenchmarkResult(tests, sizes)
     for (i,s) ∈ enumerate(sizes)
         a = rand(s);
@@ -132,7 +148,11 @@ function benchmark_selfdot(sizes)
         @assert cselfdot(a) ≈ dotblas "Polly dot wrong?"
         br[4,i] = n_gflop / @belapsed fselfdot($a)
         @assert fselfdot(a) ≈ dotblas "Fort dot wrong?"
-        br[5,i] = n_gflop / @belapsed jselfdotavx($a)
+        br[5,i] = n_gflop / @belapsed icselfdot($a)
+        @assert cselfdot(a) ≈ dotblas "icc dot wrong?"
+        br[6,i] = n_gflop / @belapsed ifselfdot($a)
+        @assert fselfdot(a) ≈ dotblas "ifort dot wrong?"
+        br[7,i] = n_gflop / @belapsed jselfdotavx($a)
         @assert jselfdotavx(a) ≈ dotblas "LoopVec dot wrong?"
         # if i % 10 == 0
             # percent_complete = round(100i/ length(sizes), sigdigits = 4)
@@ -144,7 +164,7 @@ end
 totwotuple(i::Int) = (i,i)
 totwotuple(i::Tuple{Int,Int}) = i
 function benchmark_gemv(sizes)
-    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFort-loops", "LoopVectorization"]
+    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFortran", "icc", "ifort", "LoopVectorization"]
     br = BenchmarkResult(tests, sizes)
     for (i,s) ∈ enumerate(sizes)
         M, N = totwotuple(s)
@@ -158,7 +178,11 @@ function benchmark_gemv(sizes)
         @assert x ≈ xblas "Polly wrong?"
         br[4,i] = n_gflop / @belapsed fgemv!($x, $A, $y)
         @assert x ≈ xblas "Fort wrong?"
-        br[5,i] = n_gflop / @belapsed jgemvavx!($x, $A, $y)
+        br[5,i] = n_gflop / @belapsed icgemv!($x, $A, $y)
+        @assert x ≈ xblas "icc wrong?"
+        br[6,i] = n_gflop / @belapsed ifgemv!($x, $A, $y)
+        @assert x ≈ xblas "ifort wrong?"
+        br[7,i] = n_gflop / @belapsed jgemvavx!($x, $A, $y)
         @assert x ≈ xblas "LoopVec wrong?"
         # if i % 10 == 0
             # percent_complete = round(100i/ length(sizes), sigdigits = 4)
@@ -168,7 +192,7 @@ function benchmark_gemv(sizes)
     br
 end
 function benchmark_dot3(sizes)
-    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFort-loops", "LoopVectorization"]
+    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFortran", "icc", "ifort", "LoopVectorization"]
     br = BenchmarkResult(tests, sizes)
     for (i,s) ∈ enumerate(sizes)
         M, N = totwotuple(s)
@@ -182,7 +206,11 @@ function benchmark_dot3(sizes)
         @assert cdot3(x, A, y) ≈ dotblas "Polly dot wrong?"
         br[4,i] = n_gflop / @belapsed fdot3($x, $A, $y)
         @assert fdot3(x, A, y) ≈ dotblas "Fort dot wrong?"
-        br[5,i] = n_gflop / @belapsed jdot3avx($x, $A, $y)
+        br[5,i] = n_gflop / @belapsed icdot3($x, $A, $y)
+        @assert cdot3(x, A, y) ≈ dotblas "icc dot wrong?"
+        br[6,i] = n_gflop / @belapsed ifdot3($x, $A, $y)
+        @assert fdot3(x, A, y) ≈ dotblas "ifort dot wrong?"
+        br[7,i] = n_gflop / @belapsed jdot3avx($x, $A, $y)
         @assert jdot3avx(x, A, y) ≈ dotblas "LoopVec dot wrong?"
         # if i % 10 == 0
             # percent_complete = round(100i/ length(sizes), sigdigits = 4)
@@ -196,7 +224,7 @@ function sse!(Xβ, y, X, β)
     dot(Xβ, Xβ)
 end
 function benchmark_sse(sizes)
-    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFort-loops", "LoopVectorization"]
+    tests = [BLAS.vendor() === :mkl ? "IntelMKL" : "OpenBLAS", "Julia", "Clang-Polly", "GFortran", "icc", "ifort", "LoopVectorization"]
     br = BenchmarkResult(tests, sizes)
     for (i,s) ∈ enumerate(sizes)
         N, P = totwotuple(s)
@@ -212,7 +240,11 @@ function benchmark_sse(sizes)
         @assert cOLSlp(y, X, β) ≈ lpblas "Polly wrong?"
         br[4,i] = n_gflop / @belapsed fOLSlp($y, $X, $β)
         @assert fOLSlp(y, X, β) ≈ lpblas "Fort wrong?"
-        br[5,i] = n_gflop / @belapsed jOLSlp_avx($y, $X, $β)
+        br[5,i] = n_gflop / @belapsed icOLSlp($y, $X, $β)
+        @assert cOLSlp(y, X, β) ≈ lpblas "icc wrong?"
+        br[6,i] = n_gflop / @belapsed ifOLSlp($y, $X, $β)
+        @assert fOLSlp(y, X, β) ≈ lpblas "ifort wrong?"
+        br[7,i] = n_gflop / @belapsed jOLSlp_avx($y, $X, $β)
         @assert jOLSlp_avx(y, X, β) ≈ lpblas "LoopVec wrong?"
         # if i % 10 == 0
             # percent_complete = round(100i/ length(sizes), sigdigits = 4)
@@ -223,7 +255,7 @@ function benchmark_sse(sizes)
 end
 
 function benchmark_exp(sizes)
-    tests = ["Julia", "Clang-Polly", "GFort-loops", "LoopVectorization"]
+    tests = ["Julia", "Clang-Polly", "GFortran", "icc", "ifort", "LoopVectorization"]
     br = BenchmarkResult(tests, sizes)
     for (i,s) ∈ enumerate(sizes)
         a = rand(s); b = similar(a)
@@ -234,7 +266,11 @@ function benchmark_exp(sizes)
         @assert b ≈ baseb "Clang wrong?"
         br[3,i] = n_gflop / @belapsed fvexp!($b, $a)
         @assert b ≈ baseb "Fort wrong?"
-        br[4,i] = n_gflop / @belapsed @avx @. $b = exp($a)
+        br[4,i] = n_gflop / @belapsed icvexp!($b, $a)
+        @assert b ≈ baseb "icc wrong?"
+        br[5,i] = n_gflop / @belapsed ifvexp!($b, $a)
+        @assert b ≈ baseb "ifort wrong?"
+        br[6,i] = n_gflop / @belapsed @avx @. $b = exp($a)
         @assert b ≈ baseb "LoopVec wrong?"
         # if i % 10 == 0
             # percent_complete = round(100i/ length(sizes), sigdigits = 4)
@@ -245,7 +281,7 @@ function benchmark_exp(sizes)
 end
 
 function benchmark_aplusBc(sizes)
-    tests = ["Julia", "Clang-Polly", "GFort-loops", "LoopVectorization"]
+    tests = ["Julia", "Clang-Polly", "GFortran", "icc", "ifort", "LoopVectorization"]
     br = BenchmarkResult(tests, sizes)
     for (i,s) ∈ enumerate(sizes)
         M, N = totwotuple(s)
@@ -258,7 +294,11 @@ function benchmark_aplusBc(sizes)
         @assert D ≈ Dcopy "Polly wrong?"
         br[3,i] = n_gflop / @belapsed faplusBc!($D, $a, $B, $c)
         @assert D ≈ Dcopy "Fort wrong?"
-        br[4,i] = n_gflop / @belapsed @avx @. $D = $a + $B * $c′
+        br[4,i] = n_gflop / @belapsed icaplusBc!($D, $a, $B, $c)
+        @assert D ≈ Dcopy "icc wrong?"
+        br[5,i] = n_gflop / @belapsed ifaplusBc!($D, $a, $B, $c)
+        @assert D ≈ Dcopy "ifort wrong?"
+        br[6,i] = n_gflop / @belapsed @avx @. $D = $a + $B * $c′
         @assert D ≈ Dcopy "LoopVec wrong?"
         # if i % 10 == 0
             # percent_complete = round(100i/ length(sizes), sigdigits = 4)

benchmark/driver.jl

@@ -3,7 +3,7 @@
 # includet(joinpath(LOOPVECBENCHDIR, "driver.jl"))
 
 pkgdir(pkg::String) = abspath(joinpath(dirname(Base.find_package(pkg)), ".."))
-const LOOPVECBENCHDIR = joinpath(pkgdir("LoopVectorization"), "benchmarks")
+const LOOPVECBENCHDIR = joinpath(pkgdir("LoopVectorization"), "benchmark")
 include(joinpath(LOOPVECBENCHDIR, "benchmarkflops.jl"))
 include(joinpath(LOOPVECBENCHDIR, "plotbenchmarks.jl"))
 
@@ -14,8 +14,9 @@ addprocs(9);
 
 @everywhere begin
     pkgdir(pkg::String) = abspath(joinpath(dirname(Base.find_package(pkg)), ".."))
-    const LOOPVECBENCHDIR = joinpath(pkgdir("LoopVectorization"), "benchmarks")
+    const LOOPVECBENCHDIR = joinpath(pkgdir("LoopVectorization"), "benchmark")
     include(joinpath(LOOPVECBENCHDIR, "benchmarkflops.jl"))
+    BenchmarkTools.DEFAULT_PARAMETERS.seconds = 1
 end
 
 gemm_future = @spawnat 2 benchmark_gemm(2:256);