[Code] Cuda flags fixes to pass mxm tests

Author: EgorOrachyov

cubool/CMakeLists.txt

@@ -140,17 +140,13 @@ if (CUBOOL_WITH_CUDA)
 
     # Settings: https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
     target_compile_options(cubool PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:
-            -arch=sm_30
-             -gencode=arch=compute_50,code=sm_50
-             -gencode=arch=compute_52,code=sm_52
-             -gencode=arch=compute_60,code=sm_60
-             -gencode=arch=compute_61,code=sm_61
-             -gencode=arch=compute_70,code=sm_70
-             -gencode=arch=compute_75,code=sm_75
-             -gencode=arch=compute_75,code=compute_75>)
-
-    target_compile_options(cubool PRIVATE $<$<AND:$<CONFIG:Debug>,$<COMPILE_LANGUAGE:CUDA>>: -Xptxas -O2>)
-    target_compile_options(cubool PRIVATE $<$<AND:$<CONFIG:Release>,$<COMPILE_LANGUAGE:CUDA>>: -Xptxas -O0>)
+        -gencode=arch=compute_30,code=sm_30
+        -gencode=arch=compute_35,code=sm_35
+        -gencode=arch=compute_50,code=sm_50
+        -gencode=arch=compute_52,code=sm_52
+        -gencode=arch=compute_52,code=compute_52>)
+
+    target_compile_options(cubool PRIVATE $<$<COMPILE_LANGUAGE:CUDA>: -use_fast_math -Xptxas -O2>)
 
     target_compile_definitions(cubool PRIVATE CUBOOL_WITH_CUDA)
     target_link_libraries(cubool PRIVATE nsparse_um)

cubool/utils/testing/matrix.hpp

@@ -143,8 +143,8 @@ namespace testing {
             if (cuBool_Matrix_Nvals(matrix, &extNvals) != CUBOOL_STATUS_SUCCESS)
                 return false;
 
-//            if (extNvals != nvals)
-//                return false;
+            if (extNvals != nvals)
+                return false;
 
             std::vector<cuBool_Index> extRows(extNvals);
             std::vector<cuBool_Index> extCols(extNvals);
@@ -157,25 +157,12 @@ namespace testing {
 
             for (cuBool_Index idx = 0; idx < extNvals; idx++) {
                 extracted[idx] = Pair{extRows[idx], extCols[idx]};
-            }
-
-            for (cuBool_Index idx = 0; idx < nvals; idx++) {
-                 reference[idx] = Pair{rowsIndex[idx], colsIndex[idx]};
+                reference[idx] = Pair{rowsIndex[idx], colsIndex[idx]};
             }
 
             std::sort(extracted.begin(), extracted.end(), PairCmp());
             std::sort(reference.begin(), reference.end(), PairCmp());
 
-            for (auto i = 0; i < nvals; i++ ) {
-                std::cout << extracted[i].i << " " << extracted[i].j << " "
-                             << reference[i].i << " " << reference[i].j << std::endl;
-
-                if (! (extracted[i] == reference[i])) {
-                    std::cout << "!!!!!" << std::endl;
-                }
-            }
-
-
             return extracted == reference;
         }
 

cubool/utils/testing/matrox_printing.hpp renamed to cubool/utils/testing/matrix_printing.hpp

@@ -72,8 +72,13 @@ namespace testing {
         }
     }
 
+    struct Print {
+        cuBool_Matrix matrix;
+    };
+
     template <typename Stream>
-    Stream& operator <<(Stream& stream, cuBool_Matrix matrix) {
+    Stream& operator <<(Stream& stream, Print p) {
+        cuBool_Matrix matrix = p.matrix;
         assert(matrix);
 
         cuBool_Index nrows;
@@ -83,13 +88,13 @@ namespace testing {
         // Query matrix data
         cuBool_Matrix_Nrows(matrix, &nrows);
         cuBool_Matrix_Ncols(matrix, &ncols);
-        cuBool_Matrix_Ncols(matrix, &nvals);
+        cuBool_Matrix_Nvals(matrix, &nvals);
 
         std::vector<cuBool_Index> rowIndex(nvals);
         std::vector<cuBool_Index> colIndex(nvals);
 
         cuBool_Matrix_ExtractPairs(matrix, rowIndex.data(), colIndex.data(), &nvals);
-        printMatrix(stream, rowIndex, colIndex, nrows, ncols, nvals);
+        printMatrix(stream, rowIndex.data(), colIndex.data(), nrows, ncols, nvals);
 
         return stream;
     }

cubool/utils/testing/testing.hpp

@@ -29,7 +29,7 @@
 #include <testing/pair.hpp>
 #include <testing/timer.hpp>
 #include <testing/matrix.hpp>
-#include <testing/matrox_printing.hpp>
+#include <testing/matrix_printing.hpp>
 #include <testing/matrix_generator.hpp>
 #include <testing/matrix_ewiseadd.hpp>
 #include <testing/matrix_mxm.hpp>