From 938ca2707fed28cff5ad35d2472c09828ef08ca7 Mon Sep 17 00:00:00 2001 From: CNugteren Date: Fri, 31 Jul 2015 17:35:42 +0200 Subject: Added HEMV routine --- src/clblast.cc | 39 +++++++++++++++++ src/routines/level2/xgemv.cc | 2 +- src/routines/level2/xhemv.cc | 100 +++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 140 insertions(+), 1 deletion(-) create mode 100644 src/routines/level2/xhemv.cc (limited to 'src') diff --git a/src/clblast.cc b/src/clblast.cc index c04462f3..12c7b880 100644 --- a/src/clblast.cc +++ b/src/clblast.cc @@ -22,6 +22,7 @@ // BLAS level-2 includes #include "internal/routines/level2/xgemv.h" +#include "internal/routines/level2/xhemv.h" #include "internal/routines/level2/xsymv.h" // BLAS level-3 includes @@ -129,6 +130,44 @@ template StatusCode Gemv(const Layout, const Transpose, // ================================================================================================= +// HEMV +template +StatusCode Hemv(const Layout layout, const Triangle triangle, + const size_t n, const T alpha, + const cl_mem a_buffer, const size_t a_offset, const size_t a_ld, + const cl_mem x_buffer, const size_t x_offset, const size_t x_inc, const T beta, + cl_mem y_buffer, const size_t y_offset, const size_t y_inc, + cl_command_queue* queue, cl_event* event) { + + auto queue_cpp = Queue(*queue); + auto event_cpp = Event(*event); + auto routine = Xhemv(queue_cpp, event_cpp); + + // Compiles the routine's device kernels + auto status = routine.SetUp(); + if (status != StatusCode::kSuccess) { return status; } + + // Runs the routine + return routine.DoHemv(layout, triangle, n, alpha, + Buffer(a_buffer), a_offset, a_ld, + Buffer(x_buffer), x_offset, x_inc, beta, + Buffer(y_buffer), y_offset, y_inc); +} +template StatusCode Hemv(const Layout, const Triangle, + const size_t, const float2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, const float2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); +template StatusCode Hemv(const Layout, const Triangle, + const size_t, const double2, + const cl_mem, const size_t, const size_t, + const cl_mem, const size_t, const size_t, const double2, + cl_mem, const size_t, const size_t, + cl_command_queue*, cl_event*); + +// ================================================================================================= + // SYMV template StatusCode Symv(const Layout layout, const Triangle triangle, diff --git a/src/routines/level2/xgemv.cc b/src/routines/level2/xgemv.cc index 7505e9f7..99b282b0 100644 --- a/src/routines/level2/xgemv.cc +++ b/src/routines/level2/xgemv.cc @@ -32,7 +32,7 @@ template Xgemv::Xgemv(Queue &queue, Event &event): Routine(queue, event, "GEMV", {"Pad", "Xgemv"}, precision_) { source_string_ = - #include "../../kernels/pad.opencl" // For SymmUpperToSquared and SymmLowerToSquared (for SYMV) + #include "../../kernels/pad.opencl" // For {Herm,Symm}{Upper,Lower}ToSquared (for HEMV/SYMV) #include "../../kernels/xgemv.opencl" ; } diff --git a/src/routines/level2/xhemv.cc b/src/routines/level2/xhemv.cc new file mode 100644 index 00000000..7883b7b9 --- /dev/null +++ b/src/routines/level2/xhemv.cc @@ -0,0 +1,100 @@ + +// ================================================================================================= +// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This +// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max- +// width of 100 characters per line. +// +// Author(s): +// Cedric Nugteren +// +// This file implements the Xhemv class (see the header for information about the class). +// +// ================================================================================================= + +#include "internal/routines/level2/xhemv.h" + +#include +#include + +namespace clblast { +// ================================================================================================= + +// Constructor: forwards to base class constructor +template +Xhemv::Xhemv(Queue &queue, Event &event): + Xgemv(queue, event) { +} + +// ================================================================================================= + +// The main routine +template +StatusCode Xhemv::DoHemv(const Layout layout, const Triangle triangle, + const size_t n, + const T alpha, + const Buffer &a_buffer, const size_t a_offset, const size_t a_ld, + const Buffer &x_buffer, const size_t x_offset, const size_t x_inc, + const T beta, + const Buffer &y_buffer, const size_t y_offset, const size_t y_inc) { + + // Makes sure all dimensions are larger than zero + if (n == 0) { return StatusCode::kInvalidDimension; } + + // Checks for validity of the squared A matrix + auto status = TestMatrixA(n, n, a_buffer, a_offset, a_ld, sizeof(T)); + if (ErrorIn(status)) { return status; } + + // Determines which kernel to run based on the layout (the Xgemv kernel assumes column-major as + // default) and on whether we are dealing with an upper or lower triangle of the hermitian matrix + bool is_upper = ((triangle == Triangle::kUpper && layout != Layout::kRowMajor) || + (triangle == Triangle::kLower && layout == Layout::kRowMajor)); + auto kernel_name = (is_upper) ? "HermUpperToSquared" : "HermLowerToSquared"; + + // Temporary buffer for a copy of the hermitian matrix + try { + auto temp_herm = Buffer(context_, n*n); + + // Creates a general matrix from the hermitian matrix to be able to run the regular Xgemv + // routine afterwards + try { + auto& program = GetProgramFromCache(); + auto kernel = Kernel(program, kernel_name); + + // Sets the arguments for the hermitian-to-squared kernel + kernel.SetArgument(0, static_cast(n)); + kernel.SetArgument(1, static_cast(a_ld)); + kernel.SetArgument(2, static_cast(a_offset)); + kernel.SetArgument(3, a_buffer()); + kernel.SetArgument(4, static_cast(n)); + kernel.SetArgument(5, static_cast(n)); + kernel.SetArgument(6, static_cast(0)); + kernel.SetArgument(7, temp_herm()); + + // Uses the common padding kernel's thread configuration. This is allowed, since the + // hermitian-to-squared kernel uses the same parameters. + auto global = std::vector{Ceil(CeilDiv(n, db_["PAD_WPTX"]), db_["PAD_DIMX"]), + Ceil(CeilDiv(n, db_["PAD_WPTY"]), db_["PAD_DIMY"])}; + auto local = std::vector{db_["PAD_DIMX"], db_["PAD_DIMY"]}; + status = RunKernel(kernel, global, local); + if (ErrorIn(status)) { return status; } + + // Runs the regular Xgemv code + status = DoGemv(layout, Transpose::kNo, n, n, alpha, + temp_herm, 0, n, + x_buffer, x_offset, x_inc, beta, + y_buffer, y_offset, y_inc); + + // Return the status of the Xgemv routine + return status; + } catch (...) { return StatusCode::kInvalidKernel; } + } catch (...) { return StatusCode::kTempBufferAllocFailure; } +} + +// ================================================================================================= + +// Compiles the templated class +template class Xhemv; +template class Xhemv; + +// ================================================================================================= +} // namespace clblast -- cgit v1.2.3