[arrayfire] 111/248: Added CPU fallback for CUDA QR when CUDA older than 7
Ghislain Vaillant
ghisvail-guest at moszumanska.debian.org
Tue Nov 17 15:54:11 UTC 2015
This is an automated email from the git hooks/post-receive script.
ghisvail-guest pushed a commit to branch dfsg-clean
in repository arrayfire.
commit 590da11c4a509417eed308f6e3e8cf02840538d2
Author: Shehzan Mohammed <shehzan at arrayfire.com>
Date: Thu Oct 8 17:36:35 2015 -0400
Added CPU fallback for CUDA QR when CUDA older than 7
---
src/backend/cuda/cpu_lapack/cpu_qr.cpp | 193 +++++++++++++++++++++++++++++++++
src/backend/cuda/cpu_lapack/cpu_qr.hpp | 22 ++++
src/backend/cuda/qr.cu | 29 +++++
3 files changed, 244 insertions(+)
diff --git a/src/backend/cuda/cpu_lapack/cpu_qr.cpp b/src/backend/cuda/cpu_lapack/cpu_qr.cpp
new file mode 100644
index 0000000..0fd04b1
--- /dev/null
+++ b/src/backend/cuda/cpu_lapack/cpu_qr.cpp
@@ -0,0 +1,193 @@
+/*******************************************************
+ * Copyright (c) 2014, ArrayFire
+ * All rights reserved.
+ *
+ * This file is distributed under 3-clause BSD license.
+ * The complete license agreement can be obtained at:
+ * http://arrayfire.com/licenses/BSD-3-Clause
+ ********************************************************/
+
+#include <qr.hpp>
+#include <err_common.hpp>
+
+#if defined(WITH_CPU_LINEAR_ALGEBRA)
+
+#include <af/dim4.hpp>
+#include <handle.hpp>
+#include <copy.hpp>
+#include <iostream>
+#include <cassert>
+#include <triangle.hpp>
+
+#include "lapack_helper.hpp"
+
+namespace cuda
+{
+namespace cpu
+{
+
+template<typename T>
+using geqrf_func_def = int (*)(ORDER_TYPE, int, int,
+ T*, int,
+ T*);
+
+template<typename T>
+using gqr_func_def = int (*)(ORDER_TYPE, int, int, int,
+ T*, int,
+ const T*);
+
+#define QR_FUNC_DEF( FUNC ) \
+template<typename T> FUNC##_func_def<T> FUNC##_func();
+
+
+#define QR_FUNC( FUNC, TYPE, PREFIX ) \
+template<> FUNC##_func_def<TYPE> FUNC##_func<TYPE>() \
+{ return & LAPACK_NAME(PREFIX##FUNC); }
+
+QR_FUNC_DEF( geqrf )
+QR_FUNC(geqrf , float , s)
+QR_FUNC(geqrf , double , d)
+QR_FUNC(geqrf , cfloat , c)
+QR_FUNC(geqrf , cdouble, z)
+
+#define GQR_FUNC_DEF( FUNC ) \
+template<typename T> FUNC##_func_def<T> FUNC##_func();
+
+#define GQR_FUNC( FUNC, TYPE, PREFIX ) \
+template<> FUNC##_func_def<TYPE> FUNC##_func<TYPE>() \
+{ return & LAPACK_NAME(PREFIX); }
+
+GQR_FUNC_DEF( gqr )
+GQR_FUNC(gqr , float , sorgqr)
+GQR_FUNC(gqr , double , dorgqr)
+GQR_FUNC(gqr , cfloat , cungqr)
+GQR_FUNC(gqr , cdouble, zungqr)
+
+template<typename T, bool is_upper, bool is_unit_diag>
+void triangle(T *o, const T *i, const dim4 odm, const dim4 ost, const dim4 ist)
+{
+ for(dim_t ow = 0; ow < odm[3]; ow++) {
+ const dim_t oW = ow * ost[3];
+ const dim_t iW = ow * ist[3];
+
+ for(dim_t oz = 0; oz < odm[2]; oz++) {
+ const dim_t oZW = oW + oz * ost[2];
+ const dim_t iZW = iW + oz * ist[2];
+
+ for(dim_t oy = 0; oy < odm[1]; oy++) {
+ const dim_t oYZW = oZW + oy * ost[1];
+ const dim_t iYZW = iZW + oy * ist[1];
+
+ for(dim_t ox = 0; ox < odm[0]; ox++) {
+ const dim_t oMem = oYZW + ox;
+ const dim_t iMem = iYZW + ox;
+
+ bool cond = is_upper ? (oy >= ox) : (oy <= ox);
+ bool do_unit_diag = (is_unit_diag && ox == oy);
+ if(cond) {
+ o[oMem] = do_unit_diag ? scalar<T>(1) : i[iMem];
+ } else {
+ o[oMem] = scalar<T>(0);
+ }
+ }
+ }
+ }
+ }
+}
+
+template<typename T>
+void qr(Array<T> &q, Array<T> &r, Array<T> &t, const Array<T> &in)
+{
+ dim4 iDims = in.dims();
+ int M = iDims[0];
+ int N = iDims[1];
+
+ dim4 padDims(M, max(M, N));
+ q = padArray<T, T>(in, padDims, scalar<T>(0));
+ q.resetDims(iDims);
+
+ dim4 qdims = q.dims();
+
+ T *tPtr = NULL;
+ T *qPtr = NULL;
+ int nT = 0;
+ {
+ ///////////////////////////////////////////////
+ // QR Inplace on q
+ int M_ = qdims[0];
+ int N_ = qdims[1];
+ nT = min(M_, N_);
+
+ tPtr = pinnedAlloc<T>(nT);
+ qPtr = pinnedAlloc<T>(padDims.elements());
+ q.resetDims(padDims);
+ copyData(qPtr, q);
+ q.resetDims(iDims);
+
+ geqrf_func<T>()(AF_LAPACK_COL_MAJOR, M, N,
+ qPtr, M,
+ tPtr);
+ ///////////////////////////////////////////////
+ }
+
+ // SPLIT into q and r
+ dim4 rdims(M, N);
+ T *rPtr = pinnedAlloc<T>(rdims.elements());
+
+ dim4 rst(1, rdims[0], rdims[0] * rdims[1], rdims[0] * rdims[1] * rdims[2]);
+
+ triangle<T, true, false>(rPtr, qPtr, rdims, rst, q.strides());
+
+ gqr_func<T>()(AF_LAPACK_COL_MAJOR,
+ M, M, min(M, N),
+ qPtr, q.strides()[1],
+ tPtr);
+
+ q.resetDims(dim4(M, M));
+
+ t = createHostDataArray<T>(af::dim4(nT), tPtr);
+ r = createHostDataArray<T>(rdims, rPtr);
+ writeHostDataArray<T>(q, qPtr, q.elements() * sizeof(T));
+
+ pinnedFree(tPtr);
+ pinnedFree(rPtr);
+ pinnedFree(qPtr);
+}
+
+template<typename T>
+Array<T> qr_inplace(Array<T> &in)
+{
+ dim4 iDims = in.dims();
+ int M = iDims[0];
+ int N = iDims[1];
+
+ T *tPtr = pinnedAlloc<T>(min(M, N));
+ T *inPtr = pinnedAlloc<T>(in.elements());
+ copyData(inPtr, in);
+
+ geqrf_func<T>()(AF_LAPACK_COL_MAJOR, M, N,
+ inPtr, in.strides()[1],
+ tPtr);
+
+ writeHostDataArray<T>(in, inPtr, in.elements() * sizeof(T));
+ Array<T> t = createHostDataArray<T>(af::dim4(min(M, N)), tPtr);
+
+ pinnedFree(inPtr);
+ pinnedFree(tPtr);
+
+ return t;
+}
+
+#define INSTANTIATE_QR(T) \
+ template Array<T> qr_inplace<T>(Array<T> &in); \
+ template void qr<T>(Array<T> &q, Array<T> &r, Array<T> &t, const Array<T> &in);
+
+INSTANTIATE_QR(float)
+INSTANTIATE_QR(cfloat)
+INSTANTIATE_QR(double)
+INSTANTIATE_QR(cdouble)
+
+}
+}
+
+#endif
diff --git a/src/backend/cuda/cpu_lapack/cpu_qr.hpp b/src/backend/cuda/cpu_lapack/cpu_qr.hpp
new file mode 100644
index 0000000..a7a6284
--- /dev/null
+++ b/src/backend/cuda/cpu_lapack/cpu_qr.hpp
@@ -0,0 +1,22 @@
+/*******************************************************
+ * Copyright (c) 2014, ArrayFire
+ * All rights reserved.
+ *
+ * This file is distributed under 3-clause BSD license.
+ * The complete license agreement can be obtained at:
+ * http://arrayfire.com/licenses/BSD-3-Clause
+ ********************************************************/
+
+#include <Array.hpp>
+
+namespace cuda
+{
+namespace cpu
+{
+ template<typename T>
+ void qr(Array<T> &q, Array<T> &r, Array<T> &t, const Array<T> &in);
+
+ template<typename T>
+ Array<T> qr_inplace(Array<T> &in);
+}
+}
diff --git a/src/backend/cuda/qr.cu b/src/backend/cuda/qr.cu
index 4654ee6..41ad1c2 100644
--- a/src/backend/cuda/qr.cu
+++ b/src/backend/cuda/qr.cu
@@ -219,6 +219,35 @@ INSTANTIATE_QR(double)
INSTANTIATE_QR(cdouble)
}
+#elif defined(WITH_CPU_LINEAR_ALGEBRA)
+#include <cpu_lapack/cpu_qr.hpp>
+
+namespace cuda
+{
+
+template<typename T>
+void qr(Array<T> &q, Array<T> &r, Array<T> &t, const Array<T> &in)
+{
+ return cpu::qr(q, r, t, in);
+}
+
+template<typename T>
+Array<T> qr_inplace(Array<T> &in)
+{
+ return cpu::qr_inplace(in);
+}
+
+#define INSTANTIATE_QR(T) \
+ template Array<T> qr_inplace<T>(Array<T> &in); \
+ template void qr<T>(Array<T> &q, Array<T> &r, Array<T> &t, const Array<T> &in);
+
+INSTANTIATE_QR(float)
+INSTANTIATE_QR(cfloat)
+INSTANTIATE_QR(double)
+INSTANTIATE_QR(cdouble)
+
+}
+
#else
namespace cuda
{
--
Alioth's /usr/local/bin/git-commit-notice on /srv/git.debian.org/git/debian-science/packages/arrayfire.git
More information about the debian-science-commits
mailing list