[HIPIFY][ROCm#674][rocSPARSE][feature] rocSPARSE support - Step 104 -…

… `cusparseXcsrgeam2Nnz` -> `rocsparse_csrgeam_nnz` + Updated `SPARSE` synthetic tests, the regenerated hipify-perl, and `SPARSE` `CUDA2HIP` documentation
emankov · Feb 5, 2024 · 4749e68 · 4749e68
1 parent 1b1fb7b
commit 4749e68
Show file tree

Hide file tree

Showing 7 changed files with 29 additions and 4 deletions.
diff --git a/bin/hipify-perl b/bin/hipify-perl
@@ -2193,6 +2193,7 @@ sub rocSubstitutions {
     subst("cusparseXcsr2bsrNnz", "rocsparse_csr2bsr_nnz", "library");
     subst("cusparseXcsr2coo", "rocsparse_csr2coo", "library");
     subst("cusparseXcsr2gebsrNnz", "rocsparse_csr2gebsr_nnz", "library");
+    subst("cusparseXcsrgeam2Nnz", "rocsparse_csrgeam_nnz", "library");
     subst("cusparseXcsrgeamNnz", "rocsparse_csrgeam_nnz", "library");
     subst("cusparseXcsrgemm2Nnz", "rocsparse_csrgemm_nnz", "library");
     subst("cusparseXcsric02_zeroPivot", "rocsparse_csric0_zero_pivot", "library");

diff --git a/docs/tables/CUSPARSE_API_supported_by_HIP_and_ROC.md b/docs/tables/CUSPARSE_API_supported_by_HIP_and_ROC.md
@@ -456,7 +456,7 @@
 |`cusparseScsrgemm`| |10.2| |11.0|`hipsparseScsrgemm`|2.8.0| | | | | | | | | | |
 |`cusparseScsrgemm2`| |11.0| |12.0|`hipsparseScsrgemm2`|2.8.0| | | | |`rocsparse_scsrgemm`|2.8.0| | | | |
 |`cusparseScsrgemm2_bufferSizeExt`| |11.0| |12.0|`hipsparseScsrgemm2_bufferSizeExt`|2.8.0| | | | |`rocsparse_scsrgemm_buffer_size`|2.8.0| | | | |
-|`cusparseXcsrgeam2Nnz`|10.0| | | |`hipsparseXcsrgeam2Nnz`|3.5.0| | | | | | | | | | |
+|`cusparseXcsrgeam2Nnz`|10.0| | | |`hipsparseXcsrgeam2Nnz`|3.5.0| | | | |`rocsparse_csrgeam_nnz`|3.5.0| | | | |
 |`cusparseXcsrgeamNnz`| |10.2| |11.0|`hipsparseXcsrgeamNnz`|3.5.0| | | | |`rocsparse_csrgeam_nnz`|3.5.0| | | | |
 |`cusparseXcsrgemm2Nnz`| |11.0| |12.0|`hipsparseXcsrgemm2Nnz`|2.8.0| | | | |`rocsparse_csrgemm_nnz`|2.8.0| | | | |
 |`cusparseXcsrgemmNnz`| |10.2| |11.0|`hipsparseXcsrgemmNnz`|2.8.0| | | | | | | | | | |

diff --git a/docs/tables/CUSPARSE_API_supported_by_ROC.md b/docs/tables/CUSPARSE_API_supported_by_ROC.md
@@ -456,7 +456,7 @@
 |`cusparseScsrgemm`| |10.2| |11.0| | | | | | |
 |`cusparseScsrgemm2`| |11.0| |12.0|`rocsparse_scsrgemm`|2.8.0| | | | |
 |`cusparseScsrgemm2_bufferSizeExt`| |11.0| |12.0|`rocsparse_scsrgemm_buffer_size`|2.8.0| | | | |
-|`cusparseXcsrgeam2Nnz`|10.0| | | | | | | | | |
+|`cusparseXcsrgeam2Nnz`|10.0| | | |`rocsparse_csrgeam_nnz`|3.5.0| | | | |
 |`cusparseXcsrgeamNnz`| |10.2| |11.0|`rocsparse_csrgeam_nnz`|3.5.0| | | | |
 |`cusparseXcsrgemm2Nnz`| |11.0| |12.0|`rocsparse_csrgemm_nnz`|2.8.0| | | | |
 |`cusparseXcsrgemmNnz`| |10.2| |11.0| | | | | | |

diff --git a/src/CUDA2HIP_SPARSE_API_functions.cpp b/src/CUDA2HIP_SPARSE_API_functions.cpp
@@ -300,7 +300,7 @@ const std::map<llvm::StringRef, hipCounter> CUDA_SPARSE_FUNCTION_MAP {
   {"cusparseDcsrgeam2",                                 {"hipsparseDcsrgeam2",                                 "rocsparse_dcsrgeam",                                               CONV_LIB_FUNC, API_SPARSE, 11}},
   {"cusparseCcsrgeam2",                                 {"hipsparseCcsrgeam2",                                 "rocsparse_ccsrgeam",                                               CONV_LIB_FUNC, API_SPARSE, 11}},
   {"cusparseZcsrgeam2",                                 {"hipsparseZcsrgeam2",                                 "rocsparse_zcsrgeam",                                               CONV_LIB_FUNC, API_SPARSE, 11}},
-  {"cusparseXcsrgeam2Nnz",                              {"hipsparseXcsrgeam2Nnz",                              "",                                                                 CONV_LIB_FUNC, API_SPARSE, 11, ROC_UNSUPPORTED}},
+  {"cusparseXcsrgeam2Nnz",                              {"hipsparseXcsrgeam2Nnz",                              "rocsparse_csrgeam_nnz",                                            CONV_LIB_FUNC, API_SPARSE, 11}},
 
   {"cusparseScsrgeam2_bufferSizeExt",                   {"hipsparseScsrgeam2_bufferSizeExt",                   "",                                                                 CONV_LIB_FUNC, API_SPARSE, 11, ROC_UNSUPPORTED}},
   {"cusparseDcsrgeam2_bufferSizeExt",                   {"hipsparseDcsrgeam2_bufferSizeExt",                   "",                                                                 CONV_LIB_FUNC, API_SPARSE, 11, ROC_UNSUPPORTED}},

diff --git a/src/HipifyAction.cpp b/src/HipifyAction.cpp
@@ -210,6 +210,7 @@ const std::string sCusparseDenseToSparse_bufferSize = "cusparseDenseToSparse_buf
 const std::string sCusparseDenseToSparse_analysis = "cusparseDenseToSparse_analysis";
 const std::string sCusparseSpMM_bufferSize = "cusparseSpMM_bufferSize";
 const std::string sCusparseSpSM_analysis = "cusparseSpSM_analysis";
+const std::string sCusparseXcsrgeam2Nnz = "cusparseXcsrgeam2Nnz";
 
 // CUDA_OVERLOADED
 const std::string sCudaEventCreate = "cudaEventCreate";
@@ -1693,6 +1694,15 @@ std::map<std::string, ArgCastStruct> FuncArgCasts {
       false
     }
   },
+  {sCusparseXcsrgeam2Nnz,
+    {
+      {
+        {14, {e_remove_argument, cw_None}}
+      },
+      true,
+      false
+    }
+  },
 };
 
 void HipifyAction::RewriteString(StringRef s, clang::SourceLocation start) {
@@ -2550,7 +2560,8 @@ std::unique_ptr<clang::ASTConsumer> HipifyAction::CreateASTConsumer(clang::Compi
             sCusparseDenseToSparse_bufferSize,
             sCusparseDenseToSparse_analysis,
             sCusparseSpMM_bufferSize,
-            sCusparseSpSM_analysis
+            sCusparseSpSM_analysis,
+            sCusparseXcsrgeam2Nnz
           )
         )
       )

diff --git a/tests/unit_tests/synthetic/libraries/cusparse2hipsparse.cu b/tests/unit_tests/synthetic/libraries/cusparse2hipsparse.cu
@@ -237,6 +237,7 @@ int main() {
   void *tempBuffer5 = nullptr;
   void *c_coeff = nullptr;
   void *s_coeff = nullptr;
+  void *workspace = nullptr;
   size_t dataSize = 0;
   size_t bufferSize = 0;
   size_t bufferSize1 = 0;
@@ -1852,6 +1853,11 @@ int main() {
   // HIP: HIPSPARSE_EXPORT hipsparseStatus_t hipsparseScsrgeam2(hipsparseHandle_t handle, int m, int n, const float* alpha, const hipsparseMatDescr_t descrA, int nnzA, const float* csrSortedValA, const int* csrSortedRowPtrA, const int* csrSortedColIndA, const float* beta, const hipsparseMatDescr_t descrB, int nnzB, const float* csrSortedValB, const int* csrSortedRowPtrB, const int* csrSortedColIndB, const hipsparseMatDescr_t descrC, float* csrSortedValC, int* csrSortedRowPtrC, int* csrSortedColIndC, void* pBuffer);
   // CHECK: status_t = hipsparseScsrgeam2(handle_t, m, n, &fA, matDescr_A, nnza, &csrSortedValA, &csrRowPtrA, &csrColIndA, &fB, matDescr_B, nnzb, &csrSortedValB, &csrRowPtrB, &csrColIndB, matDescr_C, &csrSortedValC, &csrRowPtrC, &csrColIndC, pBuffer);
   status_t = cusparseScsrgeam2(handle_t, m, n, &fA, matDescr_A, nnza, &csrSortedValA, &csrRowPtrA, &csrColIndA, &fB, matDescr_B, nnzb, &csrSortedValB, &csrRowPtrB, &csrColIndB, matDescr_C, &csrSortedValC, &csrRowPtrC, &csrColIndC, pBuffer);
+
+  // CUDA: cusparseStatus_t CUSPARSEAPI cusparseXcsrgeam2Nnz(cusparseHandle_t handle, int m, int n, const cusparseMatDescr_t descrA, int nnzA, const int* csrSortedRowPtrA, const int* csrSortedColIndA, const cusparseMatDescr_t descrB, int nnzB, const int* csrSortedRowPtrB, const int* csrSortedColIndB, const cusparseMatDescr_t descrC, int* csrSortedRowPtrC, int* nnzTotalDevHostPtr, void* workspace);
+  // HIP: HIPSPARSE_EXPORT hipsparseStatus_t hipsparseXcsrgeam2Nnz(hipsparseHandle_t handle, int m, int n, const hipsparseMatDescr_t descrA, int nnzA, const int* csrSortedRowPtrA, const int* csrSortedColIndA, const hipsparseMatDescr_t descrB, int nnzB, const int* csrSortedRowPtrB, const int* csrSortedColIndB, const hipsparseMatDescr_t descrC, int* csrSortedRowPtrC, int* nnzTotalDevHostPtr, void* workspace);
+  // CHECK: status_t = hipsparseXcsrgeam2Nnz(handle_t, m, n, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, matDescr_C, &csrRowPtrC, &nnzTotalDevHostPtr, workspace);
+  status_t = cusparseXcsrgeam2Nnz(handle_t, m, n, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, matDescr_C, &csrRowPtrC, &nnzTotalDevHostPtr, workspace);
 #endif
 
 #if CUDA_VERSION >= 10010

diff --git a/tests/unit_tests/synthetic/libraries/cusparse2rocsparse_10000.cu b/tests/unit_tests/synthetic/libraries/cusparse2rocsparse_10000.cu
@@ -40,6 +40,7 @@ int main() {
   int csrColIndA = 0;
   int csrColIndB = 0;
   int csrColIndC = 0;
+  int nnzTotalDevHostPtr = 0;
   double dAlpha = 0.f;
   double dBeta = 0.f;
   double dA = 0.f;
@@ -55,6 +56,7 @@ int main() {
   float csrSortedValB = 0.f;
   float csrSortedValC = 0.f;
   void *pBuffer = nullptr;
+  void *workspace = nullptr;
 
   // TODO: should be rocsparse_double_complex
   // TODO: add to TypeOverloads cuDoubleComplex -> rocsparse_double_complex under a new option --sparse
@@ -86,6 +88,11 @@ int main() {
   // ROC: ROCSPARSE_EXPORT rocsparse_status rocsparse_scsrgeam(rocsparse_handle handle, rocsparse_int m, rocsparse_int n, const float* alpha, const rocsparse_mat_descr descr_A, rocsparse_int nnz_A, const float* csr_val_A, const rocsparse_int* csr_row_ptr_A, const rocsparse_int* csr_col_ind_A, const float* beta, const rocsparse_mat_descr descr_B, rocsparse_int nnz_B, const float* csr_val_B, const rocsparse_int* csr_row_ptr_B, const rocsparse_int* csr_col_ind_B, const rocsparse_mat_descr descr_C, float* csr_val_C, const rocsparse_int* csr_row_ptr_C, rocsparse_int* csr_col_ind_C);
   // CHECK: status_t = rocsparse_scsrgeam(handle_t, m, n, &fA, matDescr_A, nnza, &csrSortedValA, &csrRowPtrA, &csrColIndA, &fB, matDescr_B, nnzb, &csrSortedValB, &csrRowPtrB, &csrColIndB, matDescr_C, &csrSortedValC, &csrRowPtrC, &csrColIndC);
   status_t = cusparseScsrgeam2(handle_t, m, n, &fA, matDescr_A, nnza, &csrSortedValA, &csrRowPtrA, &csrColIndA, &fB, matDescr_B, nnzb, &csrSortedValB, &csrRowPtrB, &csrColIndB, matDescr_C, &csrSortedValC, &csrRowPtrC, &csrColIndC, pBuffer);
+
+  // CUDA: cusparseStatus_t CUSPARSEAPI cusparseXcsrgeam2Nnz(cusparseHandle_t handle, int m, int n, const cusparseMatDescr_t descrA, int nnzA, const int* csrSortedRowPtrA, const int* csrSortedColIndA, const cusparseMatDescr_t descrB, int nnzB, const int* csrSortedRowPtrB, const int* csrSortedColIndB, const cusparseMatDescr_t descrC, int* csrSortedRowPtrC, int* nnzTotalDevHostPtr, void* workspace);
+  // ROC: ROCSPARSE_EXPORT rocsparse_status rocsparse_csrgeam_nnz(rocsparse_handle handle, rocsparse_int m, rocsparse_int n, const rocsparse_mat_descr descr_A, rocsparse_int nnz_A, const rocsparse_int* csr_row_ptr_A, const rocsparse_int* csr_col_ind_A, const rocsparse_mat_descr descr_B, rocsparse_int nnz_B, const rocsparse_int* csr_row_ptr_B, const rocsparse_int* csr_col_ind_B, const rocsparse_mat_descr descr_C, rocsparse_int* csr_row_ptr_C, rocsparse_int* nnz_C);
+  // CHECK: status_t = rocsparse_csrgeam_nnz(handle_t, m, n, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, matDescr_C, &csrRowPtrC, &nnzTotalDevHostPtr);
+  status_t = cusparseXcsrgeam2Nnz(handle_t, m, n, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, matDescr_C, &csrRowPtrC, &nnzTotalDevHostPtr, workspace);
 #endif
 
   return 0;