added benchmark examples

examples/matrix_matrix_multiplication.jl

@@ -0,0 +1,54 @@
+using OpenCL, CLBLAS, CLBlast, BenchmarkTools
+
+CLBLAS.setup()
+
+function run_gemm(m::Integer, n::Integer, k::Integer, T=Float32)
+    srand(12345)
+    A = rand(T, m, k)
+    B = rand(T, k, n)
+    C = rand(T, m, n)
+    α = rand(T)
+    β = rand(T)
+
+    @printf("\nm = %d, n = %d, k = %d, eltype = %s\n", m, n, k, T)
+
+    println("BLAS:")
+    C_true = copy(C)
+    LinAlg.BLAS.gemm!('N', 'N', α, A, B, β, C_true)
+    CC = copy(C)
+    display(@benchmark LinAlg.BLAS.gemm!($('N'), $('N'), $α, $A, $B, $β, $CC))
+    println()
+
+    for device in cl.devices()
+        println("-"^70)
+        @printf("Platform name   : %s\n", device[:platform][:name])
+        @printf("Platform version: %s\n", device[:platform][:version])
+        @printf("Device name     : %s\n", device[:name])
+        @printf("Device type     : %s\n", device[:device_type])
+        println()
+
+        ctx = cl.Context(device)
+        queue = cl.CmdQueue(ctx)
+
+        println("CLBLAS:")
+        A_cl = cl.CLArray(queue, A)
+        B_cl = cl.CLArray(queue, B)
+        C_cl = cl.CLArray(queue, C)
+        CLBLAS.gemm!('N', 'N', α, A_cl, B_cl, β, C_cl)
+        @assert cl.to_host(C_cl) ≈ C_true
+        display(@benchmark CLBLAS.gemm!($('N'), $('N'), $α, $A_cl, $B_cl, $β, $C_cl))
+        println()
+
+
+        println("CLBlast:")
+        A_cl = cl.CLArray(queue, A)
+        B_cl = cl.CLArray(queue, B)
+        C_cl = cl.CLArray(queue, C)
+        CLBlast.gemm!('N', 'N', α, A_cl, B_cl, β, C_cl)
+        @assert cl.to_host(C_cl) ≈ C_true
+        display(@benchmark CLBlast.gemm!($('N'), $('N'), $α, $A_cl, $B_cl, $β, $C_cl))
+        println()
+    end
+end
+
+run_gemm(2^10, 2^10, 2^10, Float32)

examples/matrix_vector_multiplication.jl

@@ -0,0 +1,53 @@
+using OpenCL, CLBLAS, CLBlast, BenchmarkTools
+
+CLBLAS.setup()
+
+function run_gemv(m::Integer, n::Integer, T=Float32)
+    srand(12345)
+    A = rand(T, m, n)
+    y = rand(T, m)
+    x = rand(T, n)
+    α = rand(T)
+    β = rand(T)
+
+    @printf("\nm = %d, n = %d, eltype = %s\n", m, n, T)
+
+    println("BLAS:")
+    y_true = copy(y)
+    LinAlg.BLAS.gemv!('N', α, A, x, β, y_true)
+    yy = copy(y)
+    display(@benchmark LinAlg.BLAS.gemv!($('N'), $α, $A, $x, $β, $yy))
+    println()
+
+    for device in cl.devices()
+        println("-"^70)
+        @printf("Platform name   : %s\n", device[:platform][:name])
+        @printf("Platform version: %s\n", device[:platform][:version])
+        @printf("Device name     : %s\n", device[:name])
+        @printf("Device type     : %s\n", device[:device_type])
+        println()
+
+        ctx = cl.Context(device)
+        queue = cl.CmdQueue(ctx)
+
+        println("CLBLAS:")
+        A_cl = cl.CLArray(queue, A)
+        y_cl = cl.CLArray(queue, y)
+        x_cl = cl.CLArray(queue, x)
+        CLBLAS.gemv!('N', α, A_cl, x_cl, β, y_cl)
+        @assert cl.to_host(y_cl) ≈ y_true
+        display(@benchmark CLBLAS.gemv!($('N'), $α, $A_cl, $x_cl, $β, $y_cl))
+        println()
+
+        println("CLBlast:")
+        A_cl = cl.CLArray(queue, A)
+        y_cl = cl.CLArray(queue, y)
+        x_cl = cl.CLArray(queue, x)
+        CLBlast.gemv!('N', α, A_cl, x_cl, β, y_cl)
+        @assert cl.to_host(y_cl) ≈ y_true
+        display(@benchmark CLBlast.gemv!($('N'), $α, $A_cl, $x_cl, $β, $y_cl))
+        println()
+    end
+end
+
+run_gemv(2^10, 2^10, Float32)

src/L3/gemm.jl

@@ -61,7 +61,7 @@ for (func, elty) in [(:CLBlastSgemm, Float32), (:CLBlastDgemm, Float64),
         layout = CLBlastLayoutColMajor
 
         # output event
-        event::cl.event::cl.Event = cl.Event(C_NULL)
+        event::cl.Event = cl.Event(C_NULL)
 
         $func(layout, a_transpose, b_transpose,
               m, n, ka,