Merge branch '750' into 'master'

Resolve "直接法の線型方程式ソルバーのバグ修正"

Closes #750

See merge request ricos/monolish!510

CHANGELOG.md

@@ -31,6 +31,7 @@ Unreleased
 ### Fixed
 - Fix benchmark result token <https://gitlab.ritc.jp/ricos/monolish/-/merge_requests/502> <https://github.com/ricosjp/monolish/issues/733>
 - Fix set_ptr bug <https://gitlab.ritc.jp/ricos/monolish/-/merge_requests/508> <https://github.com/ricosjp/monolish/issues/741>
+- Fix linear solver bug for non-symmetric matrix <https://gitlab.ritc.jp/ricos/monolish/-/merge_requests/510> <https://github.com/ricosjp/monolish/issues/750>
 
 ### Changed
 - Update cuda version of allgebra <https://gitlab.ritc.jp/ricos/monolish/-/merge_requests/506> <https://github.com/ricosjp/monolish/issues/739>

src/internal/lapack/getrs/dense_double_getrs.cpp

@@ -26,7 +26,7 @@ int internal::lapack::getrs(const matrix::Dense<double> &A, vector<double> &B,
   const double *Ad = A.data();
   double *Bd = B.data();
   const int *ipivd = ipiv.data();
-  const char trans = 'N';
+  const char trans = 'T';
 
   if (A.get_device_mem_stat() == true && B.get_device_mem_stat() == true) {
 #if MONOLISH_USE_NVIDIA_GPU
@@ -42,7 +42,11 @@ int internal::lapack::getrs(const matrix::Dense<double> &A, vector<double> &B,
 
 #pragma omp target data use_device_ptr(Ad, ipivd, Bd, devinfod)
     {
-      internal::check_CUDA(cusolverDnDgetrs(h, CUBLAS_OP_N, M, K, Ad, N, ipivd,
+      auto cublas_trans = CUBLAS_OP_N;
+      if (trans == 'T') {
+        cublas_trans = CUBLAS_OP_T;
+      }
+      internal::check_CUDA(cusolverDnDgetrs(h, cublas_trans, M, K, Ad, N, ipivd,
                                             Bd, M, devinfod));
     }
 

src/internal/lapack/getrs/dense_float_getrs.cpp

@@ -26,7 +26,7 @@ int internal::lapack::getrs(const matrix::Dense<float> &A, vector<float> &B,
   const float *Ad = A.data();
   float *Bd = B.data();
   const int *ipivd = ipiv.data();
-  const char trans = 'N';
+  const char trans = 'T';
 
   if (A.get_device_mem_stat() == true && B.get_device_mem_stat() == true) {
 #if MONOLISH_USE_NVIDIA_GPU
@@ -42,7 +42,11 @@ int internal::lapack::getrs(const matrix::Dense<float> &A, vector<float> &B,
 
 #pragma omp target data use_device_ptr(Ad, ipivd, Bd, devinfod)
     {
-      internal::check_CUDA(cusolverDnSgetrs(h, CUBLAS_OP_N, M, K, Ad, N, ipivd,
+      auto cublas_trans = CUBLAS_OP_N;
+      if (trans == 'T') {
+        cublas_trans = CUBLAS_OP_T;
+      }
+      internal::check_CUDA(cusolverDnSgetrs(h, cublas_trans, M, K, Ad, N, ipivd,
                                             Bd, M, devinfod));
     }
 

test/equation/dense_lu/Makefile

@@ -16,15 +16,19 @@ sxat:
 
 run_cpu:
 	./$(FUNC)_cpu.out ../../test.mtx 1
+	./$(FUNC)_cpu.out ../../test2.mtx 1
 
 run_gpu:
 	$(PROFILER)./$(FUNC)_gpu.out ../../test.mtx 1
+	$(PROFILER)./$(FUNC)_gpu.out ../../test2.mtx 1
 
 run_a64fx:
 	$(PROFILER)./$(FUNC)_a64fx.out ../../test.mtx 1
+	$(PROFILER)./$(FUNC)_a64fx.out ../../test2.mtx 1
 
 run_sxat:
 	$(PROFILER)./$(FUNC)_sxat.out ../../test.mtx 1
+	$(PROFILER)./$(FUNC)_sxat.out ../../test2.mtx 1
 
 clean:
 	- rm *.out

test/equation/sparse_ic/Makefile

@@ -16,15 +16,19 @@ sxat:
 
 run_cpu:
 	./$(FUNC)_cpu.out ../../test.mtx 1
+	./$(FUNC)_cpu.out ../../test2.mtx 1
 
 run_gpu:
 	$(PROFILER)./$(FUNC)_gpu.out ../../test.mtx 1
+	$(PROFILER)./$(FUNC)_gpu.out ../../test2.mtx 1
 
 run_a64fx:
 	$(PROFILER)./$(FUNC)_a64fx.out ../../test.mtx 1
+	$(PROFILER)./$(FUNC)_a64fx.out ../../test2.mtx 1
 
 run_sxat:
 	$(PROFILER)./$(FUNC)_sxat.out ../../test.mtx 1
+	$(PROFILER)./$(FUNC)_sxat.out ../../test2.mtx 1
 
 clean:
 	- rm *.out

test/equation/sparse_ilu/Makefile

@@ -16,15 +16,19 @@ sxat:
 
 run_cpu:
 	./$(FUNC)_cpu.out ../../test.mtx 1
+	./$(FUNC)_cpu.out ../../test2.mtx 1
 
 run_gpu:
 	$(PROFILER)./$(FUNC)_gpu.out ../../test.mtx 1
+	$(PROFILER)./$(FUNC)_gpu.out ../../test2.mtx 1
 
 run_a64fx:
 	$(PROFILER)./$(FUNC)_a64fx.out ../../test.mtx 1
+	$(PROFILER)./$(FUNC)_a64fx.out ../../test2.mtx 1
 
 run_sxat:
 	$(PROFILER)./$(FUNC)_sxat.out ../../test.mtx 1
+	$(PROFILER)./$(FUNC)_sxat.out ../../test2.mtx 1
 
 clean:
 	- rm *.out

test/equation/sparse_qr/Makefile

@@ -16,15 +16,19 @@ sxat:
 
 run_cpu:
 	./$(FUNC)_cpu.out ../../test.mtx 1
+	./$(FUNC)_cpu.out ../../test2.mtx 1
 
 run_gpu:
 	$(PROFILER)./$(FUNC)_gpu.out ../../test.mtx 1
+	$(PROFILER)./$(FUNC)_gpu.out ../../test2.mtx 1
 
 run_a64fx:
 	$(PROFILER)./$(FUNC)_a64fx.out ../../test.mtx 1
+	$(PROFILER)./$(FUNC)_a64fx.out ../../test2.mtx 1
 
 run_sxat:
 	$(PROFILER)./$(FUNC)_sxat.out ../../test.mtx 1
+	$(PROFILER)./$(FUNC)_sxat.out ../../test2.mtx 1
 
 clean:
 	- rm *.out

test/test2.mtx

@@ -0,0 +1,9 @@
+%%MatrixMarket matrix coordinate real general
+3 3 7
+1 1 2
+1 2 1
+2 1 -1
+2 2 2
+2 3 -1
+3 2 1
+3 3 2

test/test_utils.hpp

@@ -31,8 +31,8 @@ bool ans_check(const std::string &func, double result, double ans, double tol) {
   }
 
   if (err < tol) {
-    std::cout << func << "(" << get_type<T>() << ")" << std::flush;
-    std::cout << ": pass" << std::endl;
+    // std::cout << func << "(" << get_type<T>() << ")" << std::flush;
+    // std::cout << ": pass" << std::endl;
     return true;
   } else {
     std::cout << "Error!!" << std::endl;
@@ -104,8 +104,8 @@ bool ans_check(const std::string &func, const T *result, const T *ans, int size,
   }
 
   if (check) {
-    std::cout << func << "(" << get_type<T>() << ")" << std::flush;
-    std::cout << ": pass" << std::endl;
+    // std::cout << func << "(" << get_type<T>() << ")" << std::flush;
+    // std::cout << ": pass" << std::endl;
     return check;
   } else {
     std::cout << "Error!!" << std::endl;
@@ -144,9 +144,9 @@ bool ans_check(const std::string &func, const std::string &type,
   }
 
   if (check) {
-    std::cout << func << "(" << get_type<T>() << "," << type << ")"
-              << std::flush;
-    std::cout << ": pass" << std::endl;
+    // std::cout << func << "(" << get_type<T>() << "," << type << ")"
+    //           << std::flush;
+    // std::cout << ": pass" << std::endl;
     return check;
   } else {
     std::cout << "Error!!" << std::endl;