[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[patch] Misc fixes
- To: VSIPL++ Developers List <vsipl++@xxxxxxxxxxxxxxxx>
- Subject: [patch] Misc fixes
- From: Jules Bergmann <jules@xxxxxxxxxxxxxxxx>
- Date: Tue, 31 Jul 2007 09:11:27 -0400
This patch
- disable FFTW split-complex altogether. Out-of-place complex split
FFT appears broken on belgarath.
- Fixes ifdef logic bug that prevented optimized convolution BEs from
being used.
- Adds a 2D conv benchmark
Patch applied.
-- Jules
--
Jules Bergmann
CodeSourcery
jules@xxxxxxxxxxxxxxxx
(650) 331-3385 x705
Index: ChangeLog
===================================================================
--- ChangeLog (revision 177792)
+++ ChangeLog (working copy)
@@ -1,3 +1,17 @@
+2007-07-31 Jules Bergmann <jules@xxxxxxxxxxxxxxxx>
+
+ * src/vsip/core/signal/conv.hpp: Fix ifdef logic bug. Optimized
+ BEs were not being included.
+ * src/vsip/opt/ipp/fir.hpp: Use aligned_array for temporary storage
+ (using Vector doesn't work when dense format is split).
+ * src/vsip/opt/fftw3/fft_impl.cpp (fftw3_complex_type): Centralize
+ choice of complex_type.
+ (USE_FFTW_SPLIT): New macro, avoid use of split-complex altogether.
+ * src/vsip/opt/diag/eval.hpp (Dispatch_name): Add missing impl tags.
+ * tests/fft.cpp: Fix Wall warning.
+ * benchmarks/conv.cpp: Add diag output.
+ * benchmarks/conv2d.cpp: New file, benchmark for 2D convolution.
+
2007-07-30 Jules Bergmann <jules@xxxxxxxxxxxxxxxx>
* src/vsip/core/layout.hpp: Fix typo.
Index: src/vsip/core/signal/conv.hpp
===================================================================
--- src/vsip/core/signal/conv.hpp (revision 176624)
+++ src/vsip/core/signal/conv.hpp (working copy)
@@ -22,9 +22,8 @@
#include <vsip/core/signal/types.hpp>
#include <vsip/core/profile.hpp>
#include <vsip/core/signal/conv_common.hpp>
-#ifndef VSIP_IMPL_REF_IMPL
+#if !VSIP_IMPL_REF_IMPL
# include <vsip/opt/signal/conv_ext.hpp>
-#else
# if VSIP_IMPL_HAVE_IPP
# include <vsip/opt/ipp/conv.hpp>
# endif
Index: src/vsip/opt/ipp/fir.hpp
===================================================================
--- src/vsip/opt/ipp/fir.hpp (revision 176624)
+++ src/vsip/opt/ipp/fir.hpp (working copy)
@@ -128,11 +128,9 @@
// Make data be available with unit-stride.
typedef Layout<1, tuple<0,1,2>, Stride_unit, Cmplx_inter_fmt> layout_type;
Ext_data<sub_block_type, layout_type> ext_in
- (sub_in_block, SYNC_IN,
- Ext_data<block_type>(this->temp_in_.block()).data());
+ (sub_in_block, SYNC_IN, temp_in_.get());
Ext_data<sub_block_type, layout_type> ext_out
- (sub_out_block, SYNC_OUT,
- Ext_data<block_type>(this->temp_out_.block()).data());
+ (sub_out_block, SYNC_OUT, temp_out_.get());
Ext_data<block_type, layout_type> ext_kernel(this->kernel_.block());
Ext_data<block_type, layout_type> ext_state(this->state_.block());
length_type const d = this->decimation();
@@ -158,8 +156,8 @@
Vector<T, block_type> kernel_;
Vector<T, block_type> state_;
length_type state_saved_; // number of elements saved
- Vector<T, block_type> temp_in_;
- Vector<T, block_type> temp_out_;
+ aligned_array<T> temp_in_;
+ aligned_array<T> temp_out_;
};
} // namespace vsip::impl::ipp
Index: src/vsip/opt/fftw3/fft_impl.cpp
===================================================================
--- src/vsip/opt/fftw3/fft_impl.cpp (revision 177792)
+++ src/vsip/opt/fftw3/fft_impl.cpp (working copy)
@@ -29,18 +29,28 @@
// 070729: FFTW 3.1.2's split in-place complex-to-complex FFT is
-// broken. The plan captures the gap between the real and imaginary
-// components.
+// broken on PowerPC and x86. The plan captures the gap between the
+// real and imaginary components.
//
+// 070730: FFTW 3.1.2 split out-of-place complex FFT is broken on
+// PowerPC.
+//
// 070730: FFTW 3.1.2's split real-complex and complex-real FFT
-// also appear broken.
+// also appear broken on x86.
//
-// As a work-around: copy, then perform transform out-of-place.
+// Brave souls may set
+// USE_FFTW_SPLIT 1
+// USE_BROKEN_FFTW_SPLIT 0
+// to attempt a work-around: copy, then perform transform out-of-place.
+// Control whether FFTW split-complex transforms are performed at all.
+#define USE_FFTW_SPLIT 0
+
+// Control whether a subset broken FFTW split-complex transforms are
+// worked around.
#define USE_BROKEN_FFTW_SPLIT 0
-
/***********************************************************************
Declarations
***********************************************************************/
@@ -52,6 +62,12 @@
namespace fftw3
{
+#if USE_FFTW_SPLIT
+typedef vsip::impl::dense_complex_type fftw3_complex_type;
+#else
+typedef Cmplx_inter_fmt fftw3_complex_type;
+#endif
+
template <dimension_type D>
struct Fft_base<D, std::complex<SCALAR_TYPE>, std::complex<SCALAR_TYPE> >
{
@@ -68,13 +84,13 @@
for(index_type i=0;i<D;i++) size_[i] = dom[i].size();
plan_in_place_ =
- Create_plan<vsip::impl::dense_complex_type>
+ Create_plan<fftw3_complex_type>
::create<FFTW(plan), FFTW(iodim)>
(in_buffer_.ptr(), in_buffer_.ptr(), exp, flags, dom);
if (!plan_in_place_) VSIP_IMPL_THROW(std::bad_alloc());
- plan_by_reference_ = Create_plan<vsip::impl::dense_complex_type>
+ plan_by_reference_ = Create_plan<fftw3_complex_type>
::create<FFTW(plan), FFTW(iodim)>
(in_buffer_.ptr(), out_buffer_.ptr(), exp, flags, dom);
@@ -90,8 +106,8 @@
if (plan_by_reference_) FFTW(destroy_plan)(plan_by_reference_);
}
- Cmplx_buffer<dense_complex_type, SCALAR_TYPE> in_buffer_;
- Cmplx_buffer<dense_complex_type, SCALAR_TYPE> out_buffer_;
+ Cmplx_buffer<fftw3_complex_type, SCALAR_TYPE> in_buffer_;
+ Cmplx_buffer<fftw3_complex_type, SCALAR_TYPE> out_buffer_;
FFTW(plan) plan_in_place_;
FFTW(plan) plan_by_reference_;
int size_[D];
@@ -112,7 +128,7 @@
// FFTW3 assumes A == D - 1.
// See also query_layout().
if (A != D - 1) std::swap(size_[A], size_[D - 1]);
- plan_by_reference_ = Create_plan<dense_complex_type>::
+ plan_by_reference_ = Create_plan<fftw3_complex_type>::
create<FFTW(plan), FFTW(iodim)>
(in_buffer_.get(), out_buffer_.ptr(), A, flags, dom);
if (!plan_by_reference_) VSIP_IMPL_THROW(std::bad_alloc());
@@ -123,7 +139,7 @@
}
aligned_array<SCALAR_TYPE> in_buffer_;
- Cmplx_buffer<dense_complex_type, SCALAR_TYPE> out_buffer_;
+ Cmplx_buffer<fftw3_complex_type, SCALAR_TYPE> out_buffer_;
FFTW(plan) plan_by_reference_;
int size_[D];
bool aligned_;
@@ -143,7 +159,7 @@
// FFTW3 assumes A == D - 1.
// See also query_layout().
if (A != D - 1) std::swap(size_[A], size_[D - 1]);
- plan_by_reference_ = Create_plan<dense_complex_type>::
+ plan_by_reference_ = Create_plan<fftw3_complex_type>::
create<FFTW(plan), FFTW(iodim)>
(in_buffer_.ptr(), out_buffer_.get(), A, flags, dom);
@@ -154,7 +170,7 @@
if (plan_by_reference_) FFTW(destroy_plan)(plan_by_reference_);
}
- Cmplx_buffer<dense_complex_type, SCALAR_TYPE> in_buffer_;
+ Cmplx_buffer<fftw3_complex_type, SCALAR_TYPE> in_buffer_;
aligned_array<SCALAR_TYPE> out_buffer_;
FFTW(plan) plan_by_reference_;
int size_[D];
@@ -186,14 +202,14 @@
rtl_inout.align = VSIP_IMPL_ALLOC_ALIGNMENT;
rtl_inout.order = tuple<0, 1, 2>();
// make default based on library
- rtl_inout.complex = Create_plan<dense_complex_type>::format;
+ rtl_inout.complex = Create_plan<fftw3_complex_type>::format;
}
virtual void query_layout(Rt_layout<1> &rtl_in, Rt_layout<1> &rtl_out)
{
rtl_in.pack = this->aligned_ ? stride_unit_align : stride_unit_dense;
rtl_in.align = VSIP_IMPL_ALLOC_ALIGNMENT;
rtl_in.order = tuple<0, 1, 2>();
- rtl_in.complex = Create_plan<dense_complex_type>::format;
+ rtl_in.complex = Create_plan<fftw3_complex_type>::format;
rtl_out = rtl_in;
}
virtual void in_place(ctype *inout, stride_type s, length_type l)
@@ -217,7 +233,7 @@
inout.second, inout.first,
inout.second, inout.first);
#else
- typedef Storage<dense_complex_type, ctype> storage_type;
+ typedef Storage<fftw3_complex_type, ctype> storage_type;
rtype* real = storage_type::get_real_ptr(in_buffer_.ptr());
rtype* imag = storage_type::get_imag_ptr(in_buffer_.ptr());
memcpy(real, inout.first, l*sizeof(rtype));
@@ -282,7 +298,7 @@
rtl_in.pack = this->aligned_ ? stride_unit_align : stride_unit_dense;
rtl_in.align = VSIP_IMPL_ALLOC_ALIGNMENT;
rtl_in.order = tuple<0, 1, 2>();
- rtl_in.complex = Create_plan<dense_complex_type>::format;
+ rtl_in.complex = Create_plan<fftw3_complex_type>::format;
rtl_out = rtl_in;
}
virtual void by_reference(rtype *in, stride_type,
@@ -302,7 +318,7 @@
FFTW(execute_split_dft_r2c)(plan_by_reference_,
in, out.first, out.second);
#else
- typedef Storage<dense_complex_type, ctype> storage_type;
+ typedef Storage<fftw3_complex_type, ctype> storage_type;
rtype* out_r = storage_type::get_real_ptr(out_buffer_.ptr());
rtype* out_i = storage_type::get_imag_ptr(out_buffer_.ptr());
FFTW(execute_split_dft_r2c)(plan_by_reference_,
@@ -336,7 +352,7 @@
rtl_in.pack = this->aligned_ ? stride_unit_align : stride_unit_dense;
rtl_in.align = VSIP_IMPL_ALLOC_ALIGNMENT;
rtl_in.order = tuple<0, 1, 2>();
- rtl_in.complex = Create_plan<dense_complex_type>::format;
+ rtl_in.complex = Create_plan<fftw3_complex_type>::format;
rtl_out = rtl_in;
}
@@ -359,7 +375,7 @@
FFTW(execute_split_dft_c2r)(plan_by_reference_,
in.first, in.second, out);
#else
- typedef Storage<dense_complex_type, ctype> storage_type;
+ typedef Storage<fftw3_complex_type, ctype> storage_type;
rtype* in_r = storage_type::get_real_ptr(in_buffer_.ptr());
rtype* in_i = storage_type::get_imag_ptr(in_buffer_.ptr());
memcpy(in_r, in.first, (length/2+1)*sizeof(rtype));
@@ -392,7 +408,7 @@
{
rtl_in.pack = stride_unit_dense;
rtl_in.order = row2_type();
- rtl_in.complex = Create_plan<dense_complex_type>::format;
+ rtl_in.complex = Create_plan<fftw3_complex_type>::format;
rtl_out = rtl_in;
}
virtual void in_place(ctype *inout,
@@ -477,7 +493,7 @@
// FFTW3 assumes A is the last dimension.
if (A == 0) rtl_in.order = tuple<1, 0, 2>();
else rtl_in.order = tuple<0, 1, 2>();
- rtl_in.complex = Create_plan<dense_complex_type>::format;
+ rtl_in.complex = Create_plan<fftw3_complex_type>::format;
rtl_out = rtl_in;
}
virtual bool requires_copy(Rt_layout<2> &) { return true;}
@@ -527,7 +543,7 @@
// FFTW3 assumes A is the last dimension.
if (A == 0) rtl_in.order = tuple<1, 0, 2>();
else rtl_in.order = tuple<0, 1, 2>();
- rtl_in.complex = Create_plan<dense_complex_type>::format;
+ rtl_in.complex = Create_plan<fftw3_complex_type>::format;
rtl_out = rtl_in;
}
virtual bool requires_copy(Rt_layout<2> &) { return true;}
@@ -575,7 +591,7 @@
{
rtl_in.pack = stride_unit_dense;
rtl_in.order = row3_type();
- rtl_in.complex = Create_plan<dense_complex_type>::format;
+ rtl_in.complex = Create_plan<fftw3_complex_type>::format;
rtl_out = rtl_in;
}
virtual void in_place(ctype *inout,
@@ -706,7 +722,7 @@
case 1: rtl_in.order = tuple<0, 2, 1>(); break;
default: rtl_in.order = tuple<0, 1, 2>(); break;
}
- rtl_in.complex = Create_plan<dense_complex_type>::format;
+ rtl_in.complex = Create_plan<fftw3_complex_type>::format;
rtl_out = rtl_in;
}
virtual bool requires_copy(Rt_layout<3> &) { return true;}
@@ -772,7 +788,7 @@
case 1: rtl_in.order = tuple<0, 2, 1>(); break;
default: rtl_in.order = tuple<0, 1, 2>(); break;
}
- rtl_in.complex = Create_plan<dense_complex_type>::format;
+ rtl_in.complex = Create_plan<fftw3_complex_type>::format;
rtl_out = rtl_in;
}
virtual bool requires_copy(Rt_layout<3> &) { return true;}
@@ -837,7 +853,7 @@
if (A == 0) rtl_in.order = tuple<1, 0, 2>();
else rtl_in.order = tuple<0, 1, 2>();
// make default based on library
- rtl_in.complex = Create_plan<dense_complex_type>::format;
+ rtl_in.complex = Create_plan<fftw3_complex_type>::format;
rtl_out = rtl_in;
}
virtual void by_reference(rtype *in,
@@ -882,7 +898,7 @@
rtype* out_i = out.second;
#if !USE_BROKEN_FFTW_SPLIT
- typedef Storage<dense_complex_type, ctype> storage_type;
+ typedef Storage<fftw3_complex_type, ctype> storage_type;
rtype* tmp_out_r = storage_type::get_real_ptr(out_buffer_.ptr());
rtype* tmp_out_i = storage_type::get_imag_ptr(out_buffer_.ptr());
#endif
@@ -937,7 +953,7 @@
if (A == 0) rtl_in.order = tuple<1, 0, 2>();
else rtl_in.order = tuple<0, 1, 2>();
// make default based on library
- rtl_in.complex = Create_plan<dense_complex_type>::format;
+ rtl_in.complex = Create_plan<fftw3_complex_type>::format;
rtl_out = rtl_in;
}
virtual bool requires_copy(Rt_layout<2> &) { return true;}
@@ -984,7 +1000,7 @@
rtype* in_i = in.second;
#if !USE_BROKEN_FFTW_SPLIT
- typedef Storage<dense_complex_type, ctype> storage_type;
+ typedef Storage<fftw3_complex_type, ctype> storage_type;
rtype* tmp_in_r = storage_type::get_real_ptr(in_buffer_.ptr());
rtype* tmp_in_i = storage_type::get_imag_ptr(in_buffer_.ptr());
#endif
@@ -1040,7 +1056,7 @@
if (A == 0) rtl_inout.order = tuple<1, 0, 2>();
else rtl_inout.order = tuple<0, 1, 2>();
// make default based on library
- rtl_inout.complex = Create_plan<dense_complex_type>::format;
+ rtl_inout.complex = Create_plan<fftw3_complex_type>::format;
}
virtual void query_layout(Rt_layout<2> &rtl_in, Rt_layout<2> &rtl_out)
@@ -1050,7 +1066,7 @@
if (A == 0) rtl_in.order = tuple<1, 0, 2>();
else rtl_in.order = tuple<0, 1, 2>();
// make default based on library
- rtl_in.complex = Create_plan<dense_complex_type>::format;
+ rtl_in.complex = Create_plan<fftw3_complex_type>::format;
rtl_out = rtl_in;
}
@@ -1058,7 +1074,7 @@
stride_type str_0, stride_type str_1,
length_type rows, length_type cols)
{
- assert((Type_equal<dense_complex_type, Cmplx_inter_fmt>::value));
+ assert((Type_equal<fftw3_complex_type, Cmplx_inter_fmt>::value));
length_type const n_fft = (A == 1) ? rows : cols;
stride_type const fft_stride = (A == 1) ? str_0 : str_1;
@@ -1083,7 +1099,7 @@
length_type rows,
length_type cols)
{
- assert((Type_equal<dense_complex_type, Cmplx_split_fmt>::value));
+ assert((Type_equal<fftw3_complex_type, Cmplx_split_fmt>::value));
length_type const n_fft = (A == 1) ? rows : cols;
stride_type const fft_stride = (A == 1) ? str_0 : str_1;
@@ -1107,7 +1123,7 @@
imag_ptr, real_ptr,
imag_ptr, real_ptr);
#else
- typedef Storage<dense_complex_type, ctype> storage_type;
+ typedef Storage<fftw3_complex_type, ctype> storage_type;
rtype* tmp_in_r = storage_type::get_real_ptr(in_buffer_.ptr());
rtype* tmp_in_i = storage_type::get_imag_ptr(in_buffer_.ptr());
memcpy(tmp_in_r, real_ptr, size_[0]*sizeof(rtype));
Index: src/vsip/opt/diag/eval.hpp
===================================================================
--- src/vsip/opt/diag/eval.hpp (revision 176624)
+++ src/vsip/opt/diag/eval.hpp (working copy)
@@ -66,6 +66,8 @@
VSIP_IMPL_DISPATCH_NAME(Fc_expr_tag)
VSIP_IMPL_DISPATCH_NAME(Rbo_expr_tag)
VSIP_IMPL_DISPATCH_NAME(Loop_fusion_tag)
+VSIP_IMPL_DISPATCH_NAME(Cvsip_tag)
+VSIP_IMPL_DISPATCH_NAME(Generic_tag)
VSIP_IMPL_DISPATCH_NAME(Tag_illegal_mix_of_local_and_global_in_assign)
VSIP_IMPL_DISPATCH_NAME(Tag_serial_expr)
Index: tests/fft.cpp
===================================================================
--- tests/fft.cpp (revision 177792)
+++ tests/fft.cpp (working copy)
@@ -72,6 +72,8 @@
std::cout << " v1 =\n" << v1;
std::cout << " v2 =\n" << v2;
}
+#else
+ (void)where;
#endif
test_assert(error < epsilon);
}
Index: benchmarks/conv.cpp
===================================================================
--- benchmarks/conv.cpp (revision 173215)
+++ benchmarks/conv.cpp (working copy)
@@ -7,7 +7,7 @@
/** @file benchmarks/conv.cpp
@author Jules Bergmann
@date 2005-07-11
- @brief VSIPL++ Library: Benchmark for Convolution.
+ @brief VSIPL++ Library: Benchmark for 1D Convolution.
*/
@@ -22,7 +22,7 @@
#include <vsip/math.hpp>
#include <vsip/signal.hpp>
-#include <vsip/core/profile.hpp>
+#include <vsip/opt/diag/eval.hpp>
#include <vsip_csl/test.hpp>
#include "loop.hpp"
@@ -99,6 +99,13 @@
t_conv1(length_type coeff_size) : coeff_size_(coeff_size) {}
+ void diag()
+ {
+ using impl::diag_detail::Dispatch_name;
+ typedef typename impl::Choose_conv_impl<1, T>::type impl_tag;
+ std::cout << "BE: " << Dispatch_name<impl_tag>::name() << std::endl;
+ }
+
length_type coeff_size_;
};
Index: benchmarks/conv2d.cpp
===================================================================
--- benchmarks/conv2d.cpp (revision 0)
+++ benchmarks/conv2d.cpp (revision 0)
@@ -0,0 +1,172 @@
+/* Copyright (c) 2005, 2006 by CodeSourcery. All rights reserved.
+
+ This file is available for license from CodeSourcery, Inc. under the terms
+ of a commercial license and under the GPL. It is not part of the VSIPL++
+ reference implementation and is not available under the BSD license.
+*/
+/** @file benchmarks/conv.cpp
+ @author Jules Bergmann
+ @date 2005-07-11
+ @brief VSIPL++ Library: Benchmark for Convolution.
+
+*/
+
+/***********************************************************************
+ Included Files
+***********************************************************************/
+
+#include <iostream>
+
+#include <vsip/initfin.hpp>
+#include <vsip/support.hpp>
+#include <vsip/math.hpp>
+#include <vsip/signal.hpp>
+
+#include <vsip/opt/diag/eval.hpp>
+
+#include <vsip_csl/test.hpp>
+#include "loop.hpp"
+
+using namespace vsip;
+
+
+
+/***********************************************************************
+ Definitions
+***********************************************************************/
+
+template <support_region_type Supp,
+ typename T>
+struct t_conv2d : Benchmark_base
+{
+ static length_type const rdec = 1;
+ static length_type const cdec = 1;
+
+ char* what() { return "t_conv2d"; }
+
+ void output_size(
+ length_type rows, length_type cols,
+ length_type& o_rows, length_type& o_cols)
+ {
+ if (Supp == support_full)
+ {
+ o_rows = ((rows + m_ - 2)/rdec) + 1;
+ o_cols = ((cols + n_ - 2)/cdec) + 1;
+ }
+ else if (Supp == support_same)
+ {
+ o_rows = ((rows - 1)/rdec) + 1;
+ o_cols = ((cols - 1)/cdec) + 1;
+ }
+ else /* (Supp == support_min) */
+ {
+ o_rows = ((rows-1)/rdec) - ((m_-1)/rdec) + 1;
+ o_cols = ((cols-1)/cdec) - ((n_-1)/rdec) + 1;
+ }
+ }
+
+ float ops_per_point(length_type cols)
+ {
+ length_type o_rows, o_cols;
+
+ output_size(rows_, cols, o_rows, o_cols);
+
+ float ops = m_ * n_ * o_rows * o_cols *
+ (vsip::impl::Ops_info<T>::mul + vsip::impl::Ops_info<T>::add);
+
+ return ops / cols;
+ }
+
+ int riob_per_point(length_type) { return -1; }
+ int wiob_per_point(length_type) { return -1; }
+ int mem_per_point(length_type) { return 2*sizeof(T); }
+
+ void operator()(length_type cols, length_type loop, float& time)
+ {
+ length_type o_rows, o_cols;
+
+ output_size(rows_, cols, o_rows, o_cols);
+
+ Matrix<T> in (rows_, cols, T());
+ Matrix<T> out(o_rows, o_cols);
+ Matrix<T> coeff(m_, n_, T());
+
+ coeff = T(1);
+
+ symmetry_type const symmetry = nonsym;
+
+ typedef Convolution<const_Matrix, symmetry, Supp, T> conv_type;
+
+ conv_type conv(coeff, Domain<2>(rows_, cols), rdec);
+
+ vsip::impl::profile::Timer t1;
+
+ t1.start();
+ for (index_type l=0; l<loop; ++l)
+ conv(in, out);
+ t1.stop();
+
+ time = t1.delta();
+ }
+
+ t_conv2d(length_type rows, length_type m, length_type n)
+ : rows_(rows), m_(m), n_(n)
+ {}
+
+ void diag()
+ {
+ using impl::diag_detail::Dispatch_name;
+ typedef typename impl::Choose_conv_impl<2, T>::type impl_tag;
+ std::cout << "BE: " << Dispatch_name<impl_tag>::name() << std::endl;
+ }
+
+ length_type rows_;
+ length_type m_;
+ length_type n_;
+};
+
+
+
+void
+defaults(Loop1P& loop)
+{
+ loop.loop_start_ = 5000;
+ loop.start_ = 4;
+
+ loop.param_["rows"] = "16";
+ loop.param_["mn"] = "0";
+ loop.param_["m"] = "3";
+ loop.param_["n"] = "3";
+}
+
+
+
+int
+test(Loop1P& loop, int what)
+{
+ typedef complex<float> cf_type;
+
+ length_type rows = atoi(loop.param_["rows"].c_str());
+ length_type MN = atoi(loop.param_["mn"].c_str());
+ length_type M = atoi(loop.param_["m"].c_str());
+ length_type N = atoi(loop.param_["n"].c_str());
+
+ std::cout << "MN: " << MN << std::endl;
+
+ if (MN != 0)
+ M = N = MN;
+
+ switch (what)
+ {
+ case 1: loop(t_conv2d<support_full, float>(rows, M, N)); break;
+ case 2: loop(t_conv2d<support_same, float>(rows, M, N)); break;
+ case 3: loop(t_conv2d<support_min, float> (rows, M, N)); break;
+
+ // case 4: loop(t_conv1<support_full, cf_type>(loop.user_param_)); break;
+ // case 5: loop(t_conv1<support_same, cf_type>(loop.user_param_)); break;
+ // case 6: loop(t_conv1<support_min, cf_type>(loop.user_param_)); break;
+
+ default: return 0;
+ }
+ return 1;
+}
