From 4270c62812016c71db82e4d780108bb3b7c0d539 Mon Sep 17 00:00:00 2001
From: Christoph Hertzberg <chtz@informatik.uni-bremen.de>
Date: Mon, 15 Apr 2019 17:18:03 +0200
Subject: Split the implementation of i?amax/min into two. Based on PR-627 by
 Sameer Agarwal. Like the Netlib reference implementation, I*AMAX now uses the
 L1-norm instead of the L2-norm for each element. Changed I*MIN accordingly.

---
 blas/level1_cplx_impl.h | 22 ++++++++++++++++++++++
 blas/level1_impl.h      | 22 ----------------------
 blas/level1_real_impl.h | 22 ++++++++++++++++++++++
 3 files changed, 44 insertions(+), 22 deletions(-)

(limited to 'blas')

diff --git a/blas/level1_cplx_impl.h b/blas/level1_cplx_impl.h
index 4ac457175..6c7edd7eb 100644
--- a/blas/level1_cplx_impl.h
+++ b/blas/level1_cplx_impl.h
@@ -36,6 +36,28 @@ RealScalar EIGEN_CAT(REAL_SCALAR_SUFFIX, EIGEN_BLAS_FUNC(asum))(int *n, RealScal
   else          return make_vector(x,*n,std::abs(*incx)).unaryExpr<scalar_norm1_op>().sum();
 }
 
+int EIGEN_CAT(i, EIGEN_BLAS_FUNC(amax))(int *n, RealScalar *px, int *incx)
+{
+  if(*n<=0) return 0;
+  Scalar* x = reinterpret_cast<Scalar*>(px);
+
+  DenseIndex ret;
+  if(*incx==1)  make_vector(x,*n).unaryExpr<scalar_norm1_op>().maxCoeff(&ret);
+  else          make_vector(x,*n,std::abs(*incx)).unaryExpr<scalar_norm1_op>().maxCoeff(&ret);
+  return int(ret)+1;
+}
+
+int EIGEN_CAT(i, EIGEN_BLAS_FUNC(amin))(int *n, RealScalar *px, int *incx)
+{
+  if(*n<=0) return 0;
+  Scalar* x = reinterpret_cast<Scalar*>(px);
+
+  DenseIndex ret;
+  if(*incx==1)  make_vector(x,*n).unaryExpr<scalar_norm1_op>().minCoeff(&ret);
+  else          make_vector(x,*n,std::abs(*incx)).unaryExpr<scalar_norm1_op>().minCoeff(&ret);
+  return int(ret)+1;
+}
+
 // computes a dot product of a conjugated vector with another vector.
 int EIGEN_BLAS_FUNC(dotcw)(int *n, RealScalar *px, int *incx, RealScalar *py, int *incy, RealScalar* pres)
 {
diff --git a/blas/level1_impl.h b/blas/level1_impl.h
index d3ee03477..71bd534b7 100644
--- a/blas/level1_impl.h
+++ b/blas/level1_impl.h
@@ -51,28 +51,6 @@ int EIGEN_BLAS_FUNC(copy)(int *n, RealScalar *px, int *incx, RealScalar *py, int
   return 0;
 }
 
-int EIGEN_CAT(i, EIGEN_BLAS_FUNC(amax))(int *n, RealScalar *px, int *incx)
-{
-  if(*n<=0) return 0;
-  Scalar* x = reinterpret_cast<Scalar*>(px);
-
-  DenseIndex ret;
-  if(*incx==1)  make_vector(x,*n).cwiseAbs().maxCoeff(&ret);
-  else          make_vector(x,*n,std::abs(*incx)).cwiseAbs().maxCoeff(&ret);
-  return int(ret)+1;
-}
-
-int EIGEN_CAT(i, EIGEN_BLAS_FUNC(amin))(int *n, RealScalar *px, int *incx)
-{
-  if(*n<=0) return 0;
-  Scalar* x = reinterpret_cast<Scalar*>(px);
-
-  DenseIndex ret;
-  if(*incx==1)  make_vector(x,*n).cwiseAbs().minCoeff(&ret);
-  else          make_vector(x,*n,std::abs(*incx)).cwiseAbs().minCoeff(&ret);
-  return int(ret)+1;
-}
-
 int EIGEN_BLAS_FUNC(rotg)(RealScalar *pa, RealScalar *pb, RealScalar *pc, RealScalar *ps)
 {
   using std::sqrt;
diff --git a/blas/level1_real_impl.h b/blas/level1_real_impl.h
index 02586d519..c58771125 100644
--- a/blas/level1_real_impl.h
+++ b/blas/level1_real_impl.h
@@ -23,6 +23,28 @@ RealScalar EIGEN_BLAS_FUNC(asum)(int *n, RealScalar *px, int *incx)
   else          return make_vector(x,*n,std::abs(*incx)).cwiseAbs().sum();
 }
 
+int EIGEN_CAT(i, EIGEN_BLAS_FUNC(amax))(int *n, RealScalar *px, int *incx)
+{
+  if(*n<=0) return 0;
+  Scalar* x = reinterpret_cast<Scalar*>(px);
+
+  DenseIndex ret;
+  if(*incx==1)  make_vector(x,*n).cwiseAbs().maxCoeff(&ret);
+  else          make_vector(x,*n,std::abs(*incx)).cwiseAbs().maxCoeff(&ret);
+  return int(ret)+1;
+}
+
+int EIGEN_CAT(i, EIGEN_BLAS_FUNC(amin))(int *n, RealScalar *px, int *incx)
+{
+  if(*n<=0) return 0;
+  Scalar* x = reinterpret_cast<Scalar*>(px);
+
+  DenseIndex ret;
+  if(*incx==1)  make_vector(x,*n).cwiseAbs().minCoeff(&ret);
+  else          make_vector(x,*n,std::abs(*incx)).cwiseAbs().minCoeff(&ret);
+  return int(ret)+1;
+}
+
 // computes a vector-vector dot product.
 Scalar EIGEN_BLAS_FUNC(dot)(int *n, RealScalar *px, int *incx, RealScalar *py, int *incy)
 {
-- 
cgit v1.2.3