Bump deps to support some BitMatrix ops, and add tests.

Manifest.toml

@@ -42,9 +42,9 @@ uuid = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 
 [[SIMDPirates]]
 deps = ["VectorizationBase"]
-git-tree-sha1 = "53c43af0172c24b0783bd93650bd8b78afb3e57b"
+git-tree-sha1 = "adbd979a7bc7e83efabd60c34ff118788653955a"
 uuid = "21efa798-c60a-11e8-04d3-e1a92915a26a"
-version = "0.7.5"
+version = "0.7.6"
 
 [[SLEEFPirates]]
 deps = ["Libdl", "SIMDPirates", "VectorizationBase"]
@@ -69,6 +69,6 @@ version = "0.1.0"
 
 [[VectorizationBase]]
 deps = ["CpuId", "LinearAlgebra"]
-git-tree-sha1 = "2a83ab02d3fb6ad5de0c6d04103b0ca403d9a7d8"
+git-tree-sha1 = "d0edd2aec08b18d39929f088c44873e2138e5be2"
 uuid = "3d5dd08c-fd9d-11e8-17fa-ed2836048c2f"
-version = "0.9.5"
+version = "0.9.6"

Project.toml

@@ -1,7 +1,7 @@
 name = "LoopVectorization"
 uuid = "bdcacae8-1622-11e9-2a5c-532679323890"
 authors = ["Chris Elrod <[email protected]>"]
-version = "0.6.25"
+version = "0.6.26"
 
 [deps]
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
@@ -13,10 +13,10 @@ VectorizationBase = "3d5dd08c-fd9d-11e8-17fa-ed2836048c2f"
 
 [compat]
 OffsetArrays = "1"
-SIMDPirates = "0.7.5"
+SIMDPirates = "0.7.6"
 SLEEFPirates = "0.4"
 UnPack = "0"
-VectorizationBase = "0.9.5"
+VectorizationBase = "0.9.6"
 julia = "1.1"
 
 [extras]

src/determinestrategy.jl

@@ -272,13 +272,13 @@ function solve_tilesize(X, R, UL, TL)
     c = -RR*R[1]*X[3]
     discriminant = b^2 - 4a*c
     discriminant < 0 && return -1,-1,Inf
-    Ufloat = (sqrt(discriminant) - b) / (2a)
-    Tfloat = (RR - max(1.0,Ufloat)*R[2])/(max(1.0,Ufloat)*R[1])
+    Ufloat = max(1.0, (sqrt(discriminant) - b) / (2a)) # must be at least 1
+    Tfloat = (RR - Ufloat*R[2])/(Ufloat*R[1])
     if !(isfinite(Tfloat) && isfinite(Ufloat))
         return 4, 4, tile_cost(X, 4, 4, UL, TL)
         # return itertilesize(X, UL, TL)
     end
-    Ulow = max(1, floor(Int, Ufloat)) # must be at least 1
+    Ulow = floor(Int, Ufloat)
     Tlow = max(1, floor(Int, Tfloat)) # must be at least 1
     Uhigh = Ulow + 1 #ceil(Int, Ufloat)
     Thigh = Tlow + 1 #ceil(Int, Tfloat)

test/gemm.jl

@@ -626,6 +626,12 @@
             @test C ≈ C2
             fill!(C, 9999.999); gemm_accurate!(C, At', Bt');
             @test C ≈ C2
+            Abit = A .> 0.5
+            fill!(C, 9999.999); AmulBavx1!(C, Abit, B)
+            @test C ≈ Abit * B
+            Bbit = B .> 0.5
+            fill!(C, 9999.999); AmulBavx1!(C, A, Bbit)
+            @test C ≈ A * Bbit
         end
         @time @testset "_avx $T dynamic gemm" begin
             AmulB_avx1!(C, A, B)

test/gemv.jl

@@ -150,7 +150,7 @@ using Test
     for T ∈ (Float32, Float64, Int32, Int64)
         @show T, @__LINE__
         TC = sizeof(T) == 4 ? Float32 : Float64
-        R = T <: Integer ? (T(1):T(1000)) : T
+        R = T <: Integer ? (T(-1000):T(1000)) : T
 
         A = rand(R, M, K);
         x = rand(R, K);
@@ -164,6 +164,13 @@ using Test
         mygemvavx_range!(y2, A, x)
         @test y1 ≈ y2
 
+        Abit = A .> 0.5
+        fill!(y2, -999.9); mygemv_avx!(y2, Abit, x)
+        @test y2 ≈ Abit * x
+        xbit = x .> 0.5
+        fill!(y2, -999.9); mygemv_avx!(y2, A, xbit)
+        @test y2 ≈ A * xbit
+
         B = rand(R, N, N);
         G1 = Matrix{TC}(undef, N, 1);
         G2 = similar(G1);