Update pure gpu splin, spcol, sp proxs tests (prog. arguments and output).

misc/test/src/C++/test_prox_sp_gpu.cpp.in

@@ -2,6 +2,8 @@
 #include "faust_MatDense_gpu.h"
 #include "faust_prox_gpu.h"
 #include <iostream>
+//#include "faust_init_from_matio_params.h"
+//#include "faust_init_from_matio_core.h"
 
 /** \brief unitary test for MatDense conjugate
 */
@@ -29,10 +31,20 @@ int main(int argc, char* argv[])
 	faust_unsigned_int dim1 = 204;
 	faust_unsigned_int dim2 = 204;
 
+//	auto MatrixFilename = "D.mat";
+//	Faust::MatDense<FPP,Cpu> D;
+//    init_faust_mat_from_matio(D,MatrixFilename,"D");
 
-	auto num_its = 30000;
+	auto num_its = 60;
 
-	auto k = dim1/10;
+	auto k = 1632;
+
+	char* str_k = getenv("K");
+
+	if(str_k)
+	{
+		k = std::atoi(str_k);
+	}
 
 	if(argc > 1)
 	{
@@ -47,10 +59,19 @@ int main(int argc, char* argv[])
 		}
 	}
 
+	std::cout << "M nrows: " << dim1 << std::endl;
+	std::cout << "M ncols: " << dim2 << std::endl;
+	cout << "k:" << k << endl;
+	std::cout << "benchmark num_its: " << num_its << std::endl;
+
+
 	MatDense<FPP,Cpu>* M;
 	MatDense<FPP,GPU2> M1, M2;
+
 	M = MatDense<FPP,Cpu>::randMat(dim1,dim2);
-	*M *= FPP(100);
+//	M = &D;
+//	M->Display();
+//	*M *= FPP(100);
 	M1 = *M;
 	M2 = *M;
 
@@ -61,17 +82,28 @@ int main(int argc, char* argv[])
 	for(int i=0; i < num_its; i++)
 	{
 		pure_gpu_prox_start = std::chrono::steady_clock::now();
-		prox_sp(M1, k, /* normalized */ false, /* pos*/ false, /* pure_gpu */ true);
+		prox_sp(M1, k, /* normalized */ true, /* pos*/ false, /* pure_gpu */ true);
 		pure_gpu_prox_end = std::chrono::steady_clock::now();
 		pure_gpu_prox_dur += pure_gpu_prox_end-pure_gpu_prox_start;
 
 		gpu_cpu_rt_prox_start = std::chrono::steady_clock::now();
-		prox_sp(M2, k, /* normalized */ false, /* pos*/ false, /* pure_gpu */ false);
+		prox_sp(M2, k, /* normalized */ true, /* pos*/ false, /* pure_gpu */ false);
 		gpu_cpu_rt_prox_end = std::chrono::steady_clock::now();
 		gpu_cpu_rt_prox_dur += gpu_cpu_rt_prox_end-gpu_cpu_rt_prox_start;
+		assert(M1.getNonZeros() == M2.getNonZeros());
+
 		M1 -= M2;
 		auto err = M1.norm();
-//		cout << "err:" << err << endl;
+#ifdef DEBUG
+		if(i == 0)
+		{
+			cout << "M1 norm:" << M1.norm() << endl;
+			M1.Display();
+			cout << "M2 norm:" << M2.norm() << endl;
+			M2.Display();
+			cout << "err:" << err << endl;
+		}
+#endif
 		assert(err < 1e-6);
 		M1 = *M;
 		M2 = *M;
@@ -79,6 +111,7 @@ int main(int argc, char* argv[])
 
 	cout << "pure gpu prox time:" << pure_gpu_prox_dur.count() << endl;
 	cout << "gpu2cpu round-trip prox time:" << gpu_cpu_rt_prox_dur.count() << endl;
+	cout << "pure gpu speedup: " << gpu_cpu_rt_prox_dur.count()/pure_gpu_prox_dur.count() << endl;
 
 	return 0;
 }

misc/test/src/C++/test_prox_splin_spcol_gpu.cpp.in

@@ -17,13 +17,13 @@ int main(int argc, char* argv[])
 {
 	Faust::enable_gpu_mod();
 	faust_unsigned_int dim1 = 204;
-	faust_unsigned_int dim2 = 204;
+	faust_unsigned_int dim2 = 8193;
 
-	auto num_its = 30000;
+	auto num_its = 270;
 
-	int32_t k = dim1/10;
+	int32_t k = 10;
 
-	int row_or_col = 0; // row by deft
+	int row_or_col = 1; // col by deft
 
 	if(argc > 1)
 	{
@@ -46,6 +46,11 @@ int main(int argc, char* argv[])
 		}
 	}
 
+	std::cout << "M nrows: " << dim1 << std::endl;
+	std::cout << "M ncols: " << dim2 << std::endl;
+	std::cout << (row_or_col==1?"prox_spcol":"prox_splin") << " k: " << k << std::endl;
+	std::cout << "benchmark num_its: " << num_its << std::endl;
+
 	MatDense<FPP,Cpu>* M;
 	MatDense<FPP,GPU2> M1, M2;
 	M = MatDense<FPP,Cpu>::randMat(dim1,dim2);
@@ -61,17 +66,17 @@ int main(int argc, char* argv[])
 	{
 		pure_gpu_prox_start = std::chrono::steady_clock::now();
 		if(row_or_col)
-			prox_spcol(M1, k, /* normalized */ false, /* pos*/ false, /* pure_gpu */ true);
+			prox_spcol(M1, k, /* normalized */ true, /* pos*/ false, /* pure_gpu */ true);
 		else
-			prox_splin(M1, k, /* normalized */ false, /* pos*/ false, /* pure_gpu */ true);
+			prox_splin(M1, k, /* normalized */ true, /* pos*/ false, /* pure_gpu */ true);
 		pure_gpu_prox_end = std::chrono::steady_clock::now();
 		pure_gpu_prox_dur += pure_gpu_prox_end-pure_gpu_prox_start;
 
 		gpu_cpu_rt_prox_start = std::chrono::steady_clock::now();
 		if(row_or_col)
-			prox_spcol(M2, k, /* normalized */ false, /* pos*/ false, /* pure_gpu */ false);
+			prox_spcol(M2, k, /* normalized */ true, /* pos*/ false, /* pure_gpu */ false);
 		else
-			prox_splin(M2, k, /* normalized */ false, /* pos*/ false, /* pure_gpu */ false);
+			prox_splin(M2, k, /* normalized */ true, /* pos*/ false, /* pure_gpu */ false);
 		gpu_cpu_rt_prox_end = std::chrono::steady_clock::now();
 		gpu_cpu_rt_prox_dur += gpu_cpu_rt_prox_end-gpu_cpu_rt_prox_start;
 		M1 -= M2;
@@ -84,6 +89,7 @@ int main(int argc, char* argv[])
 
 	cout << "pure gpu prox time:" << pure_gpu_prox_dur.count() << endl;
 	cout << "gpu2cpu round-trip prox time:" << gpu_cpu_rt_prox_dur.count() << endl;
+	cout << "pure gpu speedup: " << gpu_cpu_rt_prox_dur.count()/pure_gpu_prox_dur.count() << endl;
 
 	return 0;
 }