Merge pull request #1055 from emankov/HIPIFY

[HIPIFY][#674][rocSPARSE][feature] rocSPARSE support - Step 51 - functions `rocsparse_(s|d|c|z)srgemm_buffer_size`
ROCm · Oct 8, 2023 · 518dc6a · 518dc6a
2 parents 291878e + 11f145f
commit 518dc6a
Show file tree

Hide file tree

Showing 6 changed files with 66 additions and 16 deletions.
diff --git a/bin/hipify-perl b/bin/hipify-perl
@@ -1689,6 +1689,7 @@ sub rocSubstitutions {
     subst("cusparseCcsr2gebsr_bufferSize", "rocsparse_ccsr2gebsr_buffer_size", "library");
     subst("cusparseCcsr2hyb", "rocsparse_ccsr2hyb", "library");
     subst("cusparseCcsrcolor", "rocsparse_ccsrcolor", "library");
+    subst("cusparseCcsrgemm2_bufferSizeExt", "rocsparse_ccsrgemm_buffer_size", "library");
     subst("cusparseCcsric02", "rocsparse_ccsric0", "library");
     subst("cusparseCcsric02_analysis", "rocsparse_ccsric0_analysis", "library");
     subst("cusparseCcsric02_bufferSize", "rocsparse_ccsric0_buffer_size", "library");
@@ -1753,6 +1754,7 @@ sub rocSubstitutions {
     subst("cusparseDcsr2gebsr_bufferSize", "rocsparse_dcsr2gebsr_buffer_size", "library");
     subst("cusparseDcsr2hyb", "rocsparse_dcsr2hyb", "library");
     subst("cusparseDcsrcolor", "rocsparse_dcsrcolor", "library");
+    subst("cusparseDcsrgemm2_bufferSizeExt", "rocsparse_dcsrgemm_buffer_size", "library");
     subst("cusparseDcsric02", "rocsparse_dcsric0", "library");
     subst("cusparseDcsric02_analysis", "rocsparse_dcsric0_analysis", "library");
     subst("cusparseDcsric02_bufferSize", "rocsparse_dcsric0_buffer_size", "library");
@@ -1836,6 +1838,7 @@ sub rocSubstitutions {
     subst("cusparseScsr2gebsr_bufferSize", "rocsparse_scsr2gebsr_buffer_size", "library");
     subst("cusparseScsr2hyb", "rocsparse_scsr2hyb", "library");
     subst("cusparseScsrcolor", "rocsparse_scsrcolor", "library");
+    subst("cusparseScsrgemm2_bufferSizeExt", "rocsparse_scsrgemm_buffer_size", "library");
     subst("cusparseScsric02", "rocsparse_scsric0", "library");
     subst("cusparseScsric02_analysis", "rocsparse_scsric0_analysis", "library");
     subst("cusparseScsric02_bufferSize", "rocsparse_scsric0_buffer_size", "library");
@@ -1927,6 +1930,7 @@ sub rocSubstitutions {
     subst("cusparseZcsr2gebsr_bufferSize", "rocsparse_zcsr2gebsr_buffer_size", "library");
     subst("cusparseZcsr2hyb", "rocsparse_zcsr2hyb", "library");
     subst("cusparseZcsrcolor", "rocsparse_zcsrcolor", "library");
+    subst("cusparseZcsrgemm2_bufferSizeExt", "rocsparse_zcsrgemm_buffer_size", "library");
     subst("cusparseZcsric02", "rocsparse_zcsric0", "library");
     subst("cusparseZcsric02_analysis", "rocsparse_zcsric0_analysis", "library");
     subst("cusparseZcsric02_bufferSize", "rocsparse_zcsric0_buffer_size", "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
@@ -441,19 +441,19 @@
 |`cusparseCcsrgeam2_bufferSizeExt`|10.0| | |`hipsparseCcsrgeam2_bufferSizeExt`|3.5.0| | | | | | | | |
 |`cusparseCcsrgemm`| |10.2|11.0|`hipsparseCcsrgemm`|3.1.0| | | | | | | | |
 |`cusparseCcsrgemm2`| |11.0|12.0|`hipsparseCcsrgemm2`|3.1.0| | | | | | | | |
-|`cusparseCcsrgemm2_bufferSizeExt`| |11.0|12.0|`hipsparseCcsrgemm2_bufferSizeExt`|3.1.0| | | | | | | | |
+|`cusparseCcsrgemm2_bufferSizeExt`| |11.0|12.0|`hipsparseCcsrgemm2_bufferSizeExt`|3.1.0| | | |`rocsparse_ccsrgemm_buffer_size`|2.8.0| | | |
 |`cusparseDcsrgeam`| |10.2|11.0|`hipsparseDcsrgeam`|3.5.0| | | | | | | | |
 |`cusparseDcsrgeam2`|10.0| | |`hipsparseDcsrgeam2`|3.5.0| | | | | | | | |
 |`cusparseDcsrgeam2_bufferSizeExt`|10.0| | |`hipsparseDcsrgeam2_bufferSizeExt`|3.5.0| | | | | | | | |
 |`cusparseDcsrgemm`| |10.2|11.0|`hipsparseDcsrgemm`|2.8.0| | | | | | | | |
 |`cusparseDcsrgemm2`| |11.0|12.0|`hipsparseDcsrgemm2`|2.8.0| | | | | | | | |
-|`cusparseDcsrgemm2_bufferSizeExt`| |11.0|12.0|`hipsparseDcsrgemm2_bufferSizeExt`|2.8.0| | | | | | | | |
+|`cusparseDcsrgemm2_bufferSizeExt`| |11.0|12.0|`hipsparseDcsrgemm2_bufferSizeExt`|2.8.0| | | |`rocsparse_dcsrgemm_buffer_size`|2.8.0| | | |
 |`cusparseScsrgeam`| |10.2|11.0|`hipsparseScsrgeam`|3.5.0| | | | | | | | |
 |`cusparseScsrgeam2`|10.0| | |`hipsparseScsrgeam2`|3.5.0| | | | | | | | |
 |`cusparseScsrgeam2_bufferSizeExt`|10.0| | |`hipsparseScsrgeam2_bufferSizeExt`|3.5.0| | | | | | | | |
 |`cusparseScsrgemm`| |10.2|11.0|`hipsparseScsrgemm`|2.8.0| | | | | | | | |
 |`cusparseScsrgemm2`| |11.0|12.0|`hipsparseScsrgemm2`|2.8.0| | | | | | | | |
-|`cusparseScsrgemm2_bufferSizeExt`| |11.0|12.0|`hipsparseScsrgemm2_bufferSizeExt`|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| | | | | | | | |
 |`cusparseXcsrgeamNnz`| |10.2|11.0|`hipsparseXcsrgeamNnz`|3.5.0| | | | | | | | |
 |`cusparseXcsrgemm2Nnz`| |11.0|12.0|`hipsparseXcsrgemm2Nnz`|2.8.0| | | |`rocsparse_csrgemm_nnz`|2.8.0| | | |
@@ -463,7 +463,7 @@
 |`cusparseZcsrgeam2_bufferSizeExt`|10.0| | |`hipsparseZcsrgeam2_bufferSizeExt`|3.5.0| | | | | | | | |
 |`cusparseZcsrgemm`| |10.2|11.0|`hipsparseZcsrgemm`|3.1.0| | | | | | | | |
 |`cusparseZcsrgemm2`| |11.0|12.0|`hipsparseZcsrgemm2`|3.1.0| | | | | | | | |
-|`cusparseZcsrgemm2_bufferSizeExt`| |11.0|12.0|`hipsparseZcsrgemm2_bufferSizeExt`|3.1.0| | | | | | | | |
+|`cusparseZcsrgemm2_bufferSizeExt`| |11.0|12.0|`hipsparseZcsrgemm2_bufferSizeExt`|3.1.0| | | |`rocsparse_zcsrgemm_buffer_size`|2.8.0| | | |
 
 ## **12. CUSPARSE Preconditioners Reference**
 

diff --git a/docs/tables/CUSPARSE_API_supported_by_ROC.md b/docs/tables/CUSPARSE_API_supported_by_ROC.md
@@ -441,19 +441,19 @@
 |`cusparseCcsrgeam2_bufferSizeExt`|10.0| | | | | | | |
 |`cusparseCcsrgemm`| |10.2|11.0| | | | | |
 |`cusparseCcsrgemm2`| |11.0|12.0| | | | | |
-|`cusparseCcsrgemm2_bufferSizeExt`| |11.0|12.0| | | | | |
+|`cusparseCcsrgemm2_bufferSizeExt`| |11.0|12.0|`rocsparse_ccsrgemm_buffer_size`|2.8.0| | | |
 |`cusparseDcsrgeam`| |10.2|11.0| | | | | |
 |`cusparseDcsrgeam2`|10.0| | | | | | | |
 |`cusparseDcsrgeam2_bufferSizeExt`|10.0| | | | | | | |
 |`cusparseDcsrgemm`| |10.2|11.0| | | | | |
 |`cusparseDcsrgemm2`| |11.0|12.0| | | | | |
-|`cusparseDcsrgemm2_bufferSizeExt`| |11.0|12.0| | | | | |
+|`cusparseDcsrgemm2_bufferSizeExt`| |11.0|12.0|`rocsparse_dcsrgemm_buffer_size`|2.8.0| | | |
 |`cusparseScsrgeam`| |10.2|11.0| | | | | |
 |`cusparseScsrgeam2`|10.0| | | | | | | |
 |`cusparseScsrgeam2_bufferSizeExt`|10.0| | | | | | | |
 |`cusparseScsrgemm`| |10.2|11.0| | | | | |
 |`cusparseScsrgemm2`| |11.0|12.0| | | | | |
-|`cusparseScsrgemm2_bufferSizeExt`| |11.0|12.0| | | | | |
+|`cusparseScsrgemm2_bufferSizeExt`| |11.0|12.0|`rocsparse_scsrgemm_buffer_size`|2.8.0| | | |
 |`cusparseXcsrgeam2Nnz`|10.0| | | | | | | |
 |`cusparseXcsrgeamNnz`| |10.2|11.0| | | | | |
 |`cusparseXcsrgemm2Nnz`| |11.0|12.0|`rocsparse_csrgemm_nnz`|2.8.0| | | |
@@ -463,7 +463,7 @@
 |`cusparseZcsrgeam2_bufferSizeExt`|10.0| | | | | | | |
 |`cusparseZcsrgemm`| |10.2|11.0| | | | | |
 |`cusparseZcsrgemm2`| |11.0|12.0| | | | | |
-|`cusparseZcsrgemm2_bufferSizeExt`| |11.0|12.0| | | | | |
+|`cusparseZcsrgemm2_bufferSizeExt`| |11.0|12.0|`rocsparse_zcsrgemm_buffer_size`|2.8.0| | | |
 
 ## **12. CUSPARSE Preconditioners Reference**
 

diff --git a/src/CUDA2HIP_SPARSE_API_functions.cpp b/src/CUDA2HIP_SPARSE_API_functions.cpp
@@ -314,10 +314,10 @@ const std::map<llvm::StringRef, hipCounter> CUDA_SPARSE_FUNCTION_MAP {
   {"cusparseZcsrgemm2",                                 {"hipsparseZcsrgemm2",                                 "",                                                                 CONV_LIB_FUNC, API_SPARSE, 11, ROC_UNSUPPORTED | CUDA_DEPRECATED | CUDA_REMOVED}},
   {"cusparseXcsrgemm2Nnz",                              {"hipsparseXcsrgemm2Nnz",                              "rocsparse_csrgemm_nnz",                                            CONV_LIB_FUNC, API_SPARSE, 11, CUDA_DEPRECATED | CUDA_REMOVED}},
 
-  {"cusparseScsrgemm2_bufferSizeExt",                   {"hipsparseScsrgemm2_bufferSizeExt",                   "",                                                                 CONV_LIB_FUNC, API_SPARSE, 11, ROC_UNSUPPORTED | CUDA_DEPRECATED | CUDA_REMOVED}},
-  {"cusparseDcsrgemm2_bufferSizeExt",                   {"hipsparseDcsrgemm2_bufferSizeExt",                   "",                                                                 CONV_LIB_FUNC, API_SPARSE, 11, ROC_UNSUPPORTED | CUDA_DEPRECATED | CUDA_REMOVED}},
-  {"cusparseCcsrgemm2_bufferSizeExt",                   {"hipsparseCcsrgemm2_bufferSizeExt",                   "",                                                                 CONV_LIB_FUNC, API_SPARSE, 11, ROC_UNSUPPORTED | CUDA_DEPRECATED | CUDA_REMOVED}},
-  {"cusparseZcsrgemm2_bufferSizeExt",                   {"hipsparseZcsrgemm2_bufferSizeExt",                   "",                                                                 CONV_LIB_FUNC, API_SPARSE, 11, ROC_UNSUPPORTED | CUDA_DEPRECATED | CUDA_REMOVED}},
+  {"cusparseScsrgemm2_bufferSizeExt",                   {"hipsparseScsrgemm2_bufferSizeExt",                   "rocsparse_scsrgemm_buffer_size",                                   CONV_LIB_FUNC, API_SPARSE, 11, CUDA_DEPRECATED | CUDA_REMOVED}},
+  {"cusparseDcsrgemm2_bufferSizeExt",                   {"hipsparseDcsrgemm2_bufferSizeExt",                   "rocsparse_dcsrgemm_buffer_size",                                   CONV_LIB_FUNC, API_SPARSE, 11, CUDA_DEPRECATED | CUDA_REMOVED}},
+  {"cusparseCcsrgemm2_bufferSizeExt",                   {"hipsparseCcsrgemm2_bufferSizeExt",                   "rocsparse_ccsrgemm_buffer_size",                                   CONV_LIB_FUNC, API_SPARSE, 11, CUDA_DEPRECATED | CUDA_REMOVED}},
+  {"cusparseZcsrgemm2_bufferSizeExt",                   {"hipsparseZcsrgemm2_bufferSizeExt",                   "rocsparse_zcsrgemm_buffer_size",                                   CONV_LIB_FUNC, API_SPARSE, 11, CUDA_DEPRECATED | CUDA_REMOVED}},
 
   // 12. cuSPARSE Preconditioners Reference
   // 12.1. Incomplete Cholesky Factorization : level 0
@@ -2237,6 +2237,10 @@ const std::map<llvm::StringRef, hipAPIversions> HIP_SPARSE_FUNCTION_VER_MAP {
   {"rocsparse_sbsric0_buffer_size",                      {HIP_3080, HIP_0,    HIP_0   }},
   {"rocsparse_bsric0_zero_pivot",                        {HIP_3080, HIP_0,    HIP_0   }},
   {"rocsparse_csrgemm_nnz",                              {HIP_2080, HIP_0,    HIP_0   }},
+  {"rocsparse_zcsrgemm_buffer_size",                     {HIP_2080, HIP_0,    HIP_0   }},
+  {"rocsparse_ccsrgemm_buffer_size",                     {HIP_2080, HIP_0,    HIP_0   }},
+  {"rocsparse_dcsrgemm_buffer_size",                     {HIP_2080, HIP_0,    HIP_0   }},
+  {"rocsparse_scsrgemm_buffer_size",                     {HIP_2080, HIP_0,    HIP_0   }},
 };
 
 const std::map<unsigned int, llvm::StringRef> CUDA_SPARSE_API_SECTION_MAP {

diff --git a/tests/unit_tests/synthetic/libraries/cusparse2hipsparse.cu b/tests/unit_tests/synthetic/libraries/cusparse2hipsparse.cu
@@ -120,6 +120,7 @@ int main() {
   int batchCount = 0;
   int m = 0;
   int n = 0;
+  int k = 0;
   int mb = 0;
   int nb = 0;
   int nnza = 0;
@@ -1942,8 +1943,28 @@ int main() {
 
   // CUDA: CUSPARSE_DEPRECATED(cusparseSpGEMM) cusparseStatus_t CUSPARSEAPI cusparseXcsrgemm2Nnz(cusparseHandle_t handle, int m, int n, int k, 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 descrD, int nnzD, const int* csrSortedRowPtrD, const int* csrSortedColIndD, const cusparseMatDescr_t descrC, int* csrSortedRowPtrC, int* nnzTotalDevHostPtr, const csrgemm2Info_t info, void* pBuffer);
   // HIP: DEPRECATED_CUDA_11000("The routine will be removed in CUDA 12") HIPSPARSE_EXPORT hipsparseStatus_t hipsparseXcsrgemm2Nnz(hipsparseHandle_t handle, int m, int n, int k, const hipsparseMatDescr_t descrA, int nnzA, const int* csrRowPtrA, const int* csrColIndA, const hipsparseMatDescr_t descrB, int nnzB, const int* csrRowPtrB, const int* csrColIndB, const hipsparseMatDescr_t descrD, int nnzD, const int* csrRowPtrD, const int* csrColIndD, const hipsparseMatDescr_t descrC, int* csrRowPtrC, int* nnzTotalDevHostPtr, const csrgemm2Info_t info, void* pBuffer);
-  // CHECK: status_t = hipsparseXcsrgemm2Nnz(handle_t, m, n, k, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzB, &csrRowPtrB, &csrColIndB, matDescr_D, nnzD, &csrRowPtrD, &csrColIndD, matDescr_C, &csrRowPtrC, nnzTotalDevHostPtr, csrgemm2_info, pBuffer);
-  status_t = cusparseXcsrgemm2Nnz(handle_t, m, n, k, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzB, &csrRowPtrB, &csrColIndB, matDescr_D, nnzD, &csrRowPtrD, &csrColIndD, matDescr_C, &csrRowPtrC, nnzTotalDevHostPtr, csrgemm2_info, pBuffer);
+  // CHECK: status_t = hipsparseXcsrgemm2Nnz(handle_t, m, n, k, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, matDescr_D, nnzd, &csrRowPtrD, &csrColIndD, matDescr_C, &csrRowPtrC, &nnzTotalDevHostPtr, csrgemm2_info, pBuffer);
+  status_t = cusparseXcsrgemm2Nnz(handle_t, m, n, k, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, matDescr_D, nnzd, &csrRowPtrD, &csrColIndD, matDescr_C, &csrRowPtrC, &nnzTotalDevHostPtr, csrgemm2_info, pBuffer);
+
+  // CUDA: CUSPARSE_DEPRECATED(cusparseSpGEMM) cusparseStatus_t CUSPARSEAPI cusparseZcsrgemm2_bufferSizeExt(cusparseHandle_t handle, int m, int n, int k, const cuDoubleComplex* alpha, 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 cuDoubleComplex* beta, const cusparseMatDescr_t descrD, int nnzD, const int* csrSortedRowPtrD, const int* csrSortedColIndD, csrgemm2Info_t info, size_t* pBufferSizeInBytes);
+  // HIP: DEPRECATED_CUDA_11000("The routine will be removed in CUDA 12") HIPSPARSE_EXPORT hipsparseStatus_t hipsparseZcsrgemm2_bufferSizeExt(hipsparseHandle_t handle, int m, int n, int k, const hipDoubleComplex* alpha, const hipsparseMatDescr_t descrA, int nnzA, const int* csrRowPtrA, const int* csrColIndA, const hipsparseMatDescr_t descrB, int nnzB, const int* csrRowPtrB, const int* csrColIndB, const hipDoubleComplex* beta, const hipsparseMatDescr_t descrD, int nnzD, const int* csrRowPtrD, const int* csrColIndD, csrgemm2Info_t info, size_t* pBufferSizeInBytes);
+  // CHECK: status_t = hipsparseZcsrgemm2_bufferSizeExt(handle_t, m, n, k, &dcomplexA, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, &dcomplexB, matDescr_D, nnzd, &csrRowPtrD, &csrColIndD, csrgemm2_info, &bufferSize);
+  status_t = cusparseZcsrgemm2_bufferSizeExt(handle_t, m, n, k, &dcomplexA, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, &dcomplexB, matDescr_D, nnzd, &csrRowPtrD, &csrColIndD, csrgemm2_info, &bufferSize);
+
+  // CUDA: CUSPARSE_DEPRECATED(cusparseSpGEMM) cusparseStatus_t CUSPARSEAPI cusparseCcsrgemm2_bufferSizeExt(cusparseHandle_t handle, int m, int n, int k, const cuComplex* alpha, 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 cuComplex* beta, const cusparseMatDescr_t descrD, int nnzD, const int* csrSortedRowPtrD, const int* csrSortedColIndD, csrgemm2Info_t info, size_t* pBufferSizeInBytes);
+  // HIP: DEPRECATED_CUDA_11000("The routine will be removed in CUDA 12") HIPSPARSE_EXPORT hipsparseStatus_t hipsparseCcsrgemm2_bufferSizeExt(hipsparseHandle_t handle, int m, int n, int k, const hipComplex* alpha, const hipsparseMatDescr_t descrA, int nnzA, const int* csrRowPtrA, const int* csrColIndA, const hipsparseMatDescr_t descrB, int nnzB, const int* csrRowPtrB, const int* csrColIndB, const hipComplex* beta, const hipsparseMatDescr_t descrD, int nnzD, const int* csrRowPtrD, const int* csrColIndD, csrgemm2Info_t info, size_t* pBufferSizeInBytes);
+  // CHECK: status_t = hipsparseCcsrgemm2_bufferSizeExt(handle_t, m, n, k, &complexA, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, &complexB, matDescr_D, nnzd, &csrRowPtrD, &csrColIndD, csrgemm2_info, &bufferSize);
+  status_t = cusparseCcsrgemm2_bufferSizeExt(handle_t, m, n, k, &complexA, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, &complexB, matDescr_D, nnzd, &csrRowPtrD, &csrColIndD, csrgemm2_info, &bufferSize);
+
+  // CUDA: CUSPARSE_DEPRECATED(cusparseSpGEMM) cusparseStatus_t CUSPARSEAPI cusparseDcsrgemm2_bufferSizeExt(cusparseHandle_t handle, int m, int n, int k, const double* alpha, 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 double* beta, const cusparseMatDescr_t descrD, int nnzD, const int* csrSortedRowPtrD, const int* csrSortedColIndD, csrgemm2Info_t info, size_t* pBufferSizeInBytes);
+  // HIP: DEPRECATED_CUDA_11000("The routine will be removed in CUDA 12") HIPSPARSE_EXPORT hipsparseStatus_t hipsparseDcsrgemm2_bufferSizeExt(hipsparseHandle_t handle, int m, int n, int k, const double* alpha, const hipsparseMatDescr_t descrA, int nnzA, const int* csrRowPtrA, const int* csrColIndA, const hipsparseMatDescr_t descrB, int nnzB, const int* csrRowPtrB, const int* csrColIndB, const double* beta, const hipsparseMatDescr_t descrD, int nnzD, const int* csrRowPtrD, const int* csrColIndD, csrgemm2Info_t info, size_t* pBufferSizeInBytes);
+  // CHECK: status_t = hipsparseDcsrgemm2_bufferSizeExt(handle_t, m, n, k, &dA, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, &dB, matDescr_D, nnzd, &csrRowPtrD, &csrColIndD, csrgemm2_info, &bufferSize);
+  status_t = cusparseDcsrgemm2_bufferSizeExt(handle_t, m, n, k, &dA, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, &dB, matDescr_D, nnzd, &csrRowPtrD, &csrColIndD, csrgemm2_info, &bufferSize);
+
+  // CUDA: CUSPARSE_DEPRECATED(cusparseSpGEMM) cusparseStatus_t CUSPARSEAPI cusparseScsrgemm2_bufferSizeExt(cusparseHandle_t handle, int m, int n, int k, const float* alpha, 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 float* beta, const cusparseMatDescr_t descrD, int nnzD, const int* csrSortedRowPtrD, const int* csrSortedColIndD, csrgemm2Info_t info, size_t* pBufferSizeInBytes);
+  // HIP: DEPRECATED_CUDA_11000("The routine will be removed in CUDA 12") HIPSPARSE_EXPORT hipsparseStatus_t hipsparseScsrgemm2_bufferSizeExt(hipsparseHandle_t handle, int m, int n, int k, const float* alpha, const hipsparseMatDescr_t descrA, int nnzA, const int* csrRowPtrA, const int* csrColIndA, const hipsparseMatDescr_t descrB, int nnzB, const int* csrRowPtrB, const int* csrColIndB, const float* beta, const hipsparseMatDescr_t descrD, int nnzD, const int* csrRowPtrD, const int* csrColIndD, csrgemm2Info_t info, size_t* pBufferSizeInBytes);
+  // CHECK: status_t = hipsparseScsrgemm2_bufferSizeExt(handle_t, m, n, k, &fA, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, &fB, matDescr_D, nnzd, &csrRowPtrD, &csrColIndD, csrgemm2_info, &bufferSize);
+  status_t = cusparseScsrgemm2_bufferSizeExt(handle_t, m, n, k, &fA, matDescr_A, nnza, &csrRowPtrA, &csrColIndA, matDescr_B, nnzb, &csrRowPtrB, &csrColIndB, &fB, matDescr_D, nnzd, &csrRowPtrD, &csrColIndD, csrgemm2_info, &bufferSize);
 #endif
 
 #if CUDA_VERSION >= 12000