MKL: Level-1 (scal) primitive support

deps/onemkl.cpp

@@ -81,6 +81,33 @@ extern "C" int onemklZgemm(syclQueue_t device_queue, onemklTranspose transA,
     return 0;
 }
 
+// Support Level-1: SCAL primitive
+extern "C" void onemklDscal(syclQueue_t device_queue, int64_t n, double alpha,
+                            double *x, int64_t incx) {
+  oneapi::mkl::blas::column_major::scal(device_queue->val, n, alpha, x, incx);
+}
+
+extern "C" void onemklSscal(syclQueue_t device_queue, int64_t n, float alpha,
+                            float *x, int64_t incx) {
+  oneapi::mkl::blas::column_major::scal(device_queue->val, n, alpha, x, incx);
+}
+
+extern "C" void onemklCscal(syclQueue_t device_queue, int64_t n,
+                            float alpha, float _Complex *x,
+                            int64_t incx) {
+  oneapi::mkl::blas::column_major::scal(
+      device_queue->val, n, alpha, reinterpret_cast<std::complex<float> *>(x),
+      incx);
+}
+
+extern "C" void onemklZscal(syclQueue_t device_queue, int64_t n,
+                            double alpha, double _Complex *x,
+                            int64_t incx) {
+  oneapi::mkl::blas::column_major::scal(
+      device_queue->val, n, alpha, reinterpret_cast<std::complex<double> *>(x),
+      incx);
+}
+
 
 // other
 

deps/onemkl.h

@@ -39,6 +39,12 @@ int onemklZgemm(syclQueue_t device_queue, onemklTranspose transA,
                 const double _Complex *B, int64_t ldb, double _Complex beta,
                 double _Complex *C, int64_t ldc);
 
+// Level-1: scal oneMKL
+void onemklDscal(syclQueue_t device_queue, int64_t n, double alpha, double *x, int64_t incx);
+void onemklSscal(syclQueue_t device_queue, int64_t n, float alpha, float *x, int64_t incx);
+void onemklCscal(syclQueue_t device_queue, int64_t n, float alpha, float _Complex *x, int64_t incx);
+void onemklZscal(syclQueue_t device_queue, int64_t n, double alpha, double _Complex *x, int64_t incx);
+
 void onemklDestroy();
 #ifdef __cplusplus
 }

lib/mkl/libonemkl.jl

@@ -41,3 +41,20 @@ function onemklZgemm(device_queue, transA, transB, m, n, k, alpha, A, lda, B, ld
                                     B::ZePtr{ComplexF64}, ldb::Int64, beta::ComplexF64,
                                     C::ZePtr{ComplexF64}, ldc::Int64)::Cint
 end
+
+function onemklDscal(device_queue, n, alpha, x, incx)
+	@ccall liboneapi_support.onemklDscal(device_queue::syclQueue_t, n::Int64, alpha::Cdouble, x::ZePtr{Cdouble}, incx::Int64)::Cvoid
+end
+
+function onemklSscal(device_queue, n, alpha, x, incx)
+	@ccall liboneapi_support.onemklSscal(device_queue::syclQueue_t, n::Int64, alpha::Cfloat, x::ZePtr{Cfloat}, incx::Int64)::Cvoid
+end
+
+function onemklZscal(device_queue, n, alpha, x, incx)
+	@ccall liboneapi_support.onemklZscal(device_queue::syclQueue_t, n::Int64, alpha::ComplexF64, x::ZePtr{ComplexF64}, incx::Int64)::Cvoid
+end
+
+function onemklCscal(device_queue, n, alpha, x, incx)
+	@ccall liboneapi_support.onemklCscal(device_queue::syclQueue_t, n::Int64, alpha::ComplexF32, x::ZePtr{ComplexF32}, incx::Int64)::Cvoid
+end
+

lib/mkl/wrappers.jl

@@ -14,7 +14,23 @@ function Base.convert(::Type{onemklTranspose}, trans::Char)
     end
 end
 
-
+# level 1
+## scal
+for (fname, elty) in
+	((:onemklDscal,:Float64),
+	 (:onemklSscal,:Float32),
+	 (:onemklZscal,:ComplexF64),
+	 (:onemklCscal,:ComplexF32))
+    @eval begin
+    	function scal!(n::Integer,
+                       alpha::Number,
+                       x::StridedArray{$elty})
+            queue = global_queue(context(x), device(x))
+            $fname(sycl_queue(queue), n, alpha, x, stride(x,1))
+            x
+        end
+    end
+end
 
 #
 # BLAS

test/onemkl.jl

@@ -0,0 +1,22 @@
+using oneAPI.oneMKL
+using LinearAlgebra
+
+m = 20
+n = 35
+k = 13
+
+
+########################
+@testset "level 1" begin
+	@testset for T in eltypes
+		A = rand(T, m)
+		gpuA = oneArray(A)
+		if T === Float32
+			oneMKL.scal!(m, 5f0, gpuA)
+		else
+			oneMKL.scal!(m, 5.0, gpuA)
+		end
+		_A = Array(gpuA)
+	    @test isapprox(A .* 5, _A)
+	end
+end