Sd/v05

.gitignore

@@ -1,2 +1,5 @@
 *~
-.juliahistory
+.juliahistory
+deps/deps.jl
+deps/downloads
+deps/package

.travis.yml

@@ -32,12 +32,3 @@ before_install:
       tar -xf clBLAS-${CLBLAS}.${PATCH}-Linux-x64.tar.gz;
       export LD_LIBRARY_PATH=${HOME}/build/JuliaGPU/CLBLAS.jl/clBLAS-${CLBLAS}.${PATCH}-Linux-x64/lib64:${LD_LIBRARY_PATH};
     fi;
-
-script:
-  - julia -e 'Pkg.init(); Pkg.clone(pwd())'
-  - julia -e 'Pkg.checkout("OpenCL");'
-  - julia -e 'using CLBLAS; @assert isdefined(:CLBLAS); @assert typeof(CLBLAS) === Module'
-  - julia -e 'Pkg.test("CLBLAS")'
-
-# after_success:
-#   - julia -e 'Pkg.add("Coverage"); using Coverage; Coveralls.submit(Coveralls.process_folder())';

REQUIRE

@@ -1,2 +1,3 @@
-julia 0.5-
+julia 0.5
 OpenCL
+BinDeps

deps/build.jl

@@ -0,0 +1,41 @@
+using BinDeps
+@BinDeps.setup
+libnames = ["libCLBLAS", "clBLAS", "libclBLAS"]
+libCLBLAS = library_dependency("libCLBLAS", aliases = libnames)
+archive = "package"
+libpath = "package/bin"
+baseurl = "https://github.com/clMathLibraries/clBLAS/releases/download/v2.12/clBLAS-2.12.0-"
+
+# download a pre-compiled binary (built by GLFW)
+if is_windows()
+    if Sys.ARCH == :x86_64
+        uri = URI(baseurl * "Windows-x64.zip")
+        provides(
+            Binaries, uri,
+            libCLBLAS, unpacked_dir = archive,
+            installed_libpath = libpath, os = :Windows
+        )
+    else
+        error("Only 64 bits windows supported with automatic build")
+    end
+end
+
+if is_linux()
+    provides(AptGet, "libclblas-dev", libCLBLAS)
+    if Sys.ARCH == :x86_64
+        uri = URI(baseurl * "Linux-x64.tar.gz")
+        provides(
+            Binaries, uri,
+            libCLBLAS, unpacked_dir = archive,
+            installed_libpath = libpath, os = :Linux
+        )
+    end
+end
+if is_apple()
+    error("""
+        OSX not oficially supported.
+        Find manual build instructions on: https://github.com/clMathLibraries/clBLAS/wiki/Build
+    """)
+end
+
+@BinDeps.install Dict("libCLBLAS" => "libCLBLAS")

examples/AddNumbers.jl

@@ -1,9 +1,9 @@
 import OpenCL
-const cl = OpenCL
+using OpenCL: cl
 
 const ARRAY_SIZE = 64
-sum = Array(Float32, 2)
-data = Array(Float32, ARRAY_SIZE)
+sum = Vector{Float32}(2)
+data = Vector{Float32}(ARRAY_SIZE)
 for i in 1:ARRAY_SIZE
    data[i] = i*1.0
 end

examples/DeviceInfo.jl

@@ -1,7 +1,5 @@
 import CLBLAS
-import OpenCL
-
-const cl = OpenCL
+using OpenCL: cl
 const clblas = CLBLAS
 
 for platform in cl.platforms()

examples/clblasSasum.jl

@@ -1,50 +1,49 @@
 import CLBLAS
-import OpenCL
-
-    const clblas = CLBLAS
-    clblas.setup()
-
-    const cl = OpenCL
-
-    device, ctx, queue = clblas.get_next_compute_context()
-
-    N = unsigned(5)
-    data = Array(Float32, 5)
-    data[1] = 1.1
-    data[2] = 2.22
-    data[3] = 3.333
-    data[4] = 4.4444
-    data[5] = 5.55555
-
-    bufX = cl.Buffer(Float32, ctx, (:r, :copy),hostbuf=data)
-    bufAsum = cl.Buffer(Float32, ctx, :w, length(bufX))
-    scratchBuff = cl.Buffer(Float32, ctx, :rw, length(bufX))
-
-    offx = unsigned(0)
-    incx = cl.cl_int(1)
-    ncq = cl.cl_uint(1)
-    clq = queue.id
-    newl = cl.cl_uint(0)
-
-    event = cl.UserEvent(ctx, retain=true)
-    ptrEvent = [event.id]
-
-    clblas.clblasSasum( N, bufAsum.id, 0, bufX.id, 0, incx, scratchBuff.id,
-                                    ncq, [clq], newl, C_NULL, ptrEvent);
-
-    cl.api.clWaitForEvents(cl.cl_uint(1), ptrEvent)
-
-    result = Array(Float32, length(1))
-    cl.enqueue_read_buffer(queue, bufAsum, result, unsigned(0), nothing, true)
-
-    expected = data[1:end] |> sum
-    println(result)
-    println("------------")
-    println(expected)
-
-    try
-        @assert(isapprox(norm(expected - result[1]), zero(Float32)))
-        info("success!")
-    finally
-        clblas.teardown()
-    end
+
+const clblas = CLBLAS
+clblas.setup()
+
+using OpenCL: cl
+
+device, ctx, queue = clblas.get_next_compute_context()
+
+N = Unsigned(5)
+data = Vector{Float32}(5)
+data[1] = 1.1
+data[2] = 2.22
+data[3] = 3.333
+data[4] = 4.4444
+data[5] = 5.55555
+
+bufX = cl.Buffer(Float32, ctx, (:r, :copy),hostbuf=data)
+bufAsum = cl.Buffer(Float32, ctx, :w, length(bufX))
+scratchBuff = cl.Buffer(Float32, ctx, :rw, length(bufX))
+
+offx = Unsigned(0)
+incx = cl.cl_int(1)
+ncq = cl.cl_uint(1)
+clq = queue.id
+newl = cl.cl_uint(0)
+
+event = cl.UserEvent(ctx, retain=true)
+ptrEvent = [event.id]
+
+clblas.clblasSasum( N, bufAsum.id, 0, bufX.id, 0, incx, scratchBuff.id,
+                                ncq, [clq], newl, C_NULL, ptrEvent);
+
+cl.api.clWaitForEvents(cl.cl_uint(1), ptrEvent)
+
+result = Vector{Float32}(length(1))
+cl.enqueue_read_buffer(queue, bufAsum, result, Unsigned(0), nothing, true)
+
+expected = data[1:end] |> sum
+println(result)
+println("------------")
+println(expected)
+
+try
+    @assert(isapprox(norm(expected - result[1]), zero(Float32)))
+    info("success!")
+finally
+    clblas.teardown()
+end

examples/clblasSaxpy.jl

@@ -1,61 +1,60 @@
 import CLBLAS
-import OpenCL
-
-    const clblas = CLBLAS
-    clblas.setup()
-
-    const cl = OpenCL
-
-    device, ctx, queue = clblas.get_next_compute_context()
-
-    N = unsigned(5)
-    alpha = cl.cl_float(12)
-    data = Array(Float32, 5)
-    data[1] = 1.1
-    data[2] = 2.22
-    data[3] = 3.333
-    data[4] = 4.4444
-    data[5] = 5.55555
-
-    dat = Array(Float32, 5)
-    dat[1] = 6.1
-    dat[2] = 7.22
-    dat[3] = 8.333
-    dat[4] = 9.4444
-    dat[5] = 10.55555
-
-    bufX = cl.Buffer(Float32, ctx, (:rw, :copy),hostbuf=data)
-    bufY = cl.Buffer(Float32, ctx, (:rw, :copy),hostbuf=dat)
-
-    offx = unsigned(0)
-    incx = cl.cl_int(1)
-    incy = cl.cl_int(1)
-    ncq = cl.cl_uint(1)
-    clq = queue.id
-    newl = cl.cl_uint(0)
-
-    event = cl.UserEvent(ctx, retain=true)
-    ptrEvent = [event.id]
-
-    clblas.clblasSaxpy( N, alpha, bufX.id, 0, incx, bufY.id, 0, incy,
-                                    ncq, [clq], newl, C_NULL, ptrEvent);
-
-    cl.api.clWaitForEvents(cl.cl_uint(1), ptrEvent)
-
-    result = Array(Float32, length(bufY))
-    cl.enqueue_read_buffer(queue, bufY, result, unsigned(0), nothing, true)
-
-    #Y←αX+Y
-    expected = alpha * data + dat 
-    println(result)
-    println("------------")
-    println(expected)
-
-    try
-        for i in 1:length(result) 
-           @assert(isapprox(norm(expected[i] - result[i]), zero(Float32), atol=0.00001))   
-        end
-        info("success!")
-    finally
-        clblas.teardown()
+
+const clblas = CLBLAS
+clblas.setup()
+
+using OpenCL: cl
+
+device, ctx, queue = clblas.get_next_compute_context()
+
+N = unsigned(5)
+alpha = cl.cl_float(12)
+data = Array(Float32, 5)
+data[1] = 1.1
+data[2] = 2.22
+data[3] = 3.333
+data[4] = 4.4444
+data[5] = 5.55555
+
+dat = Array(Float32, 5)
+dat[1] = 6.1
+dat[2] = 7.22
+dat[3] = 8.333
+dat[4] = 9.4444
+dat[5] = 10.55555
+
+bufX = cl.Buffer(Float32, ctx, (:rw, :copy),hostbuf=data)
+bufY = cl.Buffer(Float32, ctx, (:rw, :copy),hostbuf=dat)
+
+offx = unsigned(0)
+incx = cl.cl_int(1)
+incy = cl.cl_int(1)
+ncq = cl.cl_uint(1)
+clq = queue.id
+newl = cl.cl_uint(0)
+
+event = cl.UserEvent(ctx, retain=true)
+ptrEvent = [event.id]
+
+clblas.clblasSaxpy( N, alpha, bufX.id, 0, incx, bufY.id, 0, incy,
+                                ncq, [clq], newl, C_NULL, ptrEvent);
+
+cl.api.clWaitForEvents(cl.cl_uint(1), ptrEvent)
+
+result = Array(Float32, length(bufY))
+cl.enqueue_read_buffer(queue, bufY, result, unsigned(0), nothing, true)
+
+#Y←αX+Y
+expected = alpha * data + dat
+println(result)
+println("------------")
+println(expected)
+
+try
+    for i in 1:length(result)
+       @assert(isapprox(norm(expected[i] - result[i]), zero(Float32), atol=0.00001))
     end
+    info("success!")
+finally
+    clblas.teardown()
+end