[HIPIFY][ROCm#674][rocSPARSE][feature] rocSPARSE support - Step 107 -…

… `cusparseSpVV_bufferSize` -> `rocsparse_spvv`

+ [IMP] `rocsparse_spvv` has been changed in 6.0.0, so reflected that in HIPIFY, docs, and tests
+ Updated `SPARSE` synthetic tests, the regenerated hipify-perl, and `SPARSE` `CUDA2HIP` documentation

bin/hipify-perl

@@ -2286,6 +2286,7 @@ sub rocSubstitutions {
     subst("cusparseSpMatSetValues", "rocsparse_spmat_set_values", "library");
     subst("cusparseSpSM_analysis", "rocsparse_spsm", "library");
     subst("cusparseSpSM_solve", "rocsparse_spsm", "library");
+    subst("cusparseSpVV_bufferSize", "rocsparse_spvv", "library");
     subst("cusparseSpVecGet", "rocsparse_spvec_get", "library");
     subst("cusparseSpVecGetIndexBase", "rocsparse_spvec_get_index_base", "library");
     subst("cusparseSpVecGetValues", "rocsparse_spvec_get_values", "library");

docs/tables/CUSPARSE_API_supported_by_HIP_and_ROC.md

@@ -905,7 +905,7 @@
 |`cusparseSpSV_solve`|11.3| |12.0| |`hipsparseSpSV_solve`|4.5.0| |6.0.0| | | | | | | | |
 |`cusparseSpSV_updateMatrix`|12.1| | | | | | | | | | | | | | | |
 |`cusparseSpVV`|10.2| |12.0| |`hipsparseSpVV`|4.1.0| |6.0.0| | | | | | | | |
-|`cusparseSpVV_bufferSize`|10.2| |12.0| |`hipsparseSpVV_bufferSize`|4.1.0| |6.0.0| | | | | | | | |
+|`cusparseSpVV_bufferSize`|10.2| |12.0| |`hipsparseSpVV_bufferSize`|4.1.0| |6.0.0| | |`rocsparse_spvv`|4.1.0| |6.0.0| | |
 |`cusparseSpVecGet`|10.2| | | |`hipsparseSpVecGet`|4.1.0| | | | |`rocsparse_spvec_get`|4.1.0| | | | |
 |`cusparseSpVecGetIndexBase`|10.2| |12.0| |`hipsparseSpVecGetIndexBase`|4.1.0| |6.0.0| | |`rocsparse_spvec_get_index_base`|4.1.0| |6.0.0| | |
 |`cusparseSpVecGetValues`|10.2| | | |`hipsparseSpVecGetValues`|4.1.0| | | | |`rocsparse_spvec_get_values`|4.1.0| | | | |

docs/tables/CUSPARSE_API_supported_by_ROC.md

@@ -905,7 +905,7 @@
 |`cusparseSpSV_solve`|11.3| |12.0| | | | | | | |
 |`cusparseSpSV_updateMatrix`|12.1| | | | | | | | | |
 |`cusparseSpVV`|10.2| |12.0| | | | | | | |
-|`cusparseSpVV_bufferSize`|10.2| |12.0| | | | | | | |
+|`cusparseSpVV_bufferSize`|10.2| |12.0| |`rocsparse_spvv`|4.1.0| |6.0.0| | |
 |`cusparseSpVecGet`|10.2| | | |`rocsparse_spvec_get`|4.1.0| | | | |
 |`cusparseSpVecGetIndexBase`|10.2| |12.0| |`rocsparse_spvec_get_index_base`|4.1.0| |6.0.0| | |
 |`cusparseSpVecGetValues`|10.2| | | |`rocsparse_spvec_get_values`|4.1.0| | | | |

src/CUDA2HIP_SPARSE_API_functions.cpp

@@ -855,7 +855,7 @@ const std::map<llvm::StringRef, hipCounter> CUDA_SPARSE_FUNCTION_MAP {
   // Sparse Vector * Vector Operations
   // NOTE: rocsparse_spvv has one additional argument: size_t* buffer_size, thus cusparseSpVV is still ROC_UNSUPPORTED
   {"cusparseSpVV",                                      {"hipsparseSpVV",                                      "",                                                                 CONV_LIB_FUNC, API_SPARSE, 15, ROC_UNSUPPORTED}},
-  {"cusparseSpVV_bufferSize",                           {"hipsparseSpVV_bufferSize",                           "",                                                                 CONV_LIB_FUNC, API_SPARSE, 15, ROC_UNSUPPORTED}},
+  {"cusparseSpVV_bufferSize",                           {"hipsparseSpVV_bufferSize",                           "rocsparse_spvv",                                                   CONV_LIB_FUNC, API_SPARSE, 15}},
 
   {"cusparseAxpby",                                     {"hipsparseAxpby",                                     "rocsparse_axpby",                                                  CONV_LIB_FUNC, API_SPARSE, 15}},
   {"cusparseGather",                                    {"hipsparseGather",                                    "rocsparse_gather",                                                 CONV_LIB_FUNC, API_SPARSE, 15}},
@@ -2431,6 +2431,7 @@ const std::map<llvm::StringRef, hipAPIversions> HIP_SPARSE_FUNCTION_VER_MAP {
   {"rocsparse_dense_to_sparse",                          {HIP_4010, HIP_0,    HIP_0   }},
   {"rocsparse_spmm",                                     {HIP_4020, HIP_0,    HIP_0   }},
   {"rocsparse_spsm",                                     {HIP_4050, HIP_0,    HIP_0   }},
+  {"rocsparse_spvv",                                     {HIP_4010, HIP_0,    HIP_0   }},
 };
 
 const std::map<llvm::StringRef, cudaAPIChangedVersions> CUDA_SPARSE_FUNCTION_CHANGED_VER_MAP {
@@ -2537,6 +2538,7 @@ const std::map<llvm::StringRef, hipAPIChangedVersions> HIP_SPARSE_FUNCTION_CHANG
   {"rocsparse_dense_to_sparse",                          {HIP_6000}},
   {"rocsparse_spmm",                                     {HIP_6000}},
   {"rocsparse_spsm",                                     {HIP_6000}},
+  {"rocsparse_spvv",                                     {HIP_6000}},
 };
 
 const std::map<unsigned int, llvm::StringRef> CUDA_SPARSE_API_SECTION_MAP {

src/HipifyAction.cpp

@@ -214,6 +214,7 @@ const std::string sCusparseSpSM_solve = "cusparseSpSM_solve";
 const std::string sCusparseXcsrgeam2Nnz = "cusparseXcsrgeam2Nnz";
 const std::string sCudaMallocHost = "cudaMallocHost";
 const std::string sCusparseSpMM = "cusparseSpMM";
+const std::string sCusparseSpVV_bufferSize = "cusparseSpVV_bufferSize";
 
 // CUDA_OVERLOADED
 const std::string sCudaEventCreate = "cudaEventCreate";
@@ -2060,6 +2061,17 @@ std::map<std::string, std::vector<ArgCastStruct>> FuncArgCasts {
       }
     }
   },
+  {sCusparseSpVV_bufferSize,
+    {
+      {
+        {
+          {7, {e_add_const_argument, cw_None, "nullptr"}},
+        },
+        true,
+        false
+      }
+    }
+  },
 };
 
 void HipifyAction::RewriteString(StringRef s, clang::SourceLocation start) {
@@ -2923,7 +2935,8 @@ std::unique_ptr<clang::ASTConsumer> HipifyAction::CreateASTConsumer(clang::Compi
             sCusparseSpSM_solve,
             sCusparseXcsrgeam2Nnz,
             sCudaMallocHost,
-            sCusparseSpMM
+            sCusparseSpMM,
+            sCusparseSpVV_bufferSize
           )
         )
       )

tests/unit_tests/synthetic/libraries/cusparse2hipsparse.cu

@@ -2025,7 +2025,7 @@ int main() {
   status_t = cusparseDestroyCsrgemm2Info(csrgemm2_info);
 #endif
 
-#if (CUDA_VERSION >= 10020 && CUDA_VERSION < 11000 && !defined(_WIN32)) || CUDA_VERSION >= 11000
+#if (CUDA_VERSION >= 10010 && CUSPARSE_VERSION >= 10200 && CUDA_VERSION < 11000 && !defined(_WIN32)) || CUDA_VERSION >= 11000
   // CHECK: hipsparseSpVecDescr_t spVecDescr_t;
   cusparseSpVecDescr_t spVecDescr_t;
 
@@ -2061,7 +2061,6 @@ int main() {
   // CHECK: status_t = hipsparseDestroyDnVec(dnVecDescr_t);
   status_t = cusparseDestroyDnVec(dnVecDescr_t);
 
-  // TODO: Mark as C-Changed in 12.0.0
   // CUDA: cusparseStatus_t CUSPARSEAPI cusparseSpVV_bufferSize(cusparseHandle_t handle, cusparseOperation_t opX, const cusparseSpVecDescr_t vecX, const cusparseDnVecDescr_t vecY, const void* result, cudaDataType computeType, size_t* bufferSize);
   // HIP: HIPSPARSE_EXPORT hipsparseStatus_t hipsparseSpVV_bufferSize(hipsparseHandle_t handle, hipsparseOperation_t opX, hipsparseSpVecDescr_t vecX, hipsparseDnVecDescr_t vecY, void* result, hipDataType computeType, size_t* bufferSize);
   // CHECK: status_t = hipsparseSpVV_bufferSize(handle_t, opX, spVecDescr_t, vecY, result, dataType, &bufferSize);
@@ -2737,13 +2736,11 @@ int main() {
   cusparseSpGEMMAlg_t SPGEMM_CSR_ALG_NONDETERMINITIC = CUSPARSE_SPGEMM_CSR_ALG_NONDETERMINITIC;
 
 #if CUDA_VERSION < 12000
-  // TODO: Mark as C-Changed in 12.0.0
   // CUDA: cusparseStatus_t CUSPARSEAPI cusparseSpGEMMreuse_workEstimation(cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB, cusparseSpMatDescr_t matA, cusparseSpMatDescr_t matB, cusparseSpMatDescr_t matC, cusparseSpGEMMAlg_t alg, cusparseSpGEMMDescr_t spgemmDescr, size_t* bufferSize1, void* externalBuffer1);
   // HIP: HIPSPARSE_EXPORT hipsparseStatus_t hipsparseSpGEMMreuse_workEstimation(hipsparseHandle_t handle, hipsparseOperation_t opA, hipsparseOperation_t opB, hipsparseSpMatDescr_t matA, hipsparseSpMatDescr_t matB, hipsparseSpMatDescr_t matC, hipsparseSpGEMMAlg_t alg, hipsparseSpGEMMDescr_t spgemmDescr, size_t* bufferSize1, void* externalBuffer1);
   // CHECK: status_t = hipsparseSpGEMMreuse_workEstimation(handle_t, opA, opB, spmatA, spmatB, spmatC, spGEMMAlg_t, spGEMMDescr, &bufferSize, tempBuffer);
   status_t = cusparseSpGEMMreuse_workEstimation(handle_t, opA, opB, spmatA, spmatB, spmatC, spGEMMAlg_t, spGEMMDescr, &bufferSize, tempBuffer);
 
-  // TODO: Mark as C-Changed in 12.0.0
   // CUDA: cusparseStatus_t CUSPARSEAPI cusparseSpGEMMreuse_nnz(cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB, cusparseSpMatDescr_t matA, cusparseSpMatDescr_t matB, cusparseSpMatDescr_t matC, cusparseSpGEMMAlg_t alg, cusparseSpGEMMDescr_t spgemmDescr, size_t* bufferSize2, void* externalBuffer2, size_t* bufferSize3, void* externalBuffer3, size_t* bufferSize4, void* externalBuffer4);
   // HIP: HIPSPARSE_EXPORT hipsparseStatus_t hipsparseSpGEMMreuse_nnz(hipsparseHandle_t handle, hipsparseOperation_t opA, hipsparseOperation_t opB, hipsparseSpMatDescr_t matA, hipsparseSpMatDescr_t matB, hipsparseSpMatDescr_t matC, hipsparseSpGEMMAlg_t alg, hipsparseSpGEMMDescr_t spgemmDescr, size_t* bufferSize2, void* externalBuffer2, size_t* bufferSize3, void* externalBuffer3, size_t* bufferSize4, void* externalBuffer4);
   // CHECK: status_t = hipsparseSpGEMMreuse_nnz(handle_t, opA, opB, spmatA, spmatB, spmatC, spGEMMAlg_t, spGEMMDescr, &bufferSize2, tempBuffer2, &bufferSize3, tempBuffer3, &bufferSize4, tempBuffer4);

tests/unit_tests/synthetic/libraries/cusparse2rocsparse_10010_12000.cu

@@ -48,6 +48,7 @@ int main() {
   void *alpha = nullptr;
   void *beta = nullptr;
   void *tempBuffer = nullptr;
+  void* result = nullptr;
 
   // CHECK: rocsparse_operation opA, opB, opX;
   cusparseOperation_t opA, opB, opX;
@@ -58,7 +59,7 @@ int main() {
   cudaDataType dataType;
 #endif
 
-#if CUDA_VERSION >= 10010
+#if CUDA_VERSION >= 10010 && CUSPARSE_VERSION >= 10200
   // TODO: cusparseCsr2CscAlg_t has no analogue in rocSPARSE. The deletion of declaration and usage is needed to be implemented
   cusparseCsr2CscAlg_t Csr2CscAlg_t;
 
@@ -72,6 +73,16 @@ int main() {
   // CHECK: rocsparse_spmm_alg spMMAlg_t;
   cusparseSpMMAlg_t spMMAlg_t;
 
+  // CHECK: _rocsparse_spvec_descr *spVecDescr = nullptr;
+  // CHECK-NEXT: rocsparse_spvec_descr spVecDescr_t;
+  cusparseSpVecDescr *spVecDescr = nullptr;
+  cusparseSpVecDescr_t spVecDescr_t;
+
+  // CHECK: _rocsparse_dnvec_descr *dnVecDescr = nullptr;
+  // CHECK-NEXT: rocsparse_dnvec_descr dnVecDescr_t, vecX, vecY;
+  cusparseDnVecDescr *dnVecDescr = nullptr;
+  cusparseDnVecDescr_t dnVecDescr_t, vecX, vecY;
+
 #if CUDA_VERSION < 12000
   // CUDA: cusparseStatus_t CUSPARSEAPI cusparseSpMM_bufferSize(cusparseHandle_t handle, cusparseOperation_t opA, cusparseOperation_t opB, const void* alpha, const cusparseSpMatDescr_t matA, const cusparseDnMatDescr_t matB, const void* beta, cusparseDnMatDescr_t matC, cudaDataType computeType, cusparseSpMMAlg_t alg, size_t* bufferSize);
   // ROC: ROCSPARSE_EXPORT rocsparse_status rocsparse_spmm(rocsparse_handle handle, rocsparse_operation trans_A, rocsparse_operation trans_B, const void* alpha, const rocsparse_spmat_descr mat_A, const rocsparse_dnmat_descr mat_B, const void* beta, const rocsparse_dnmat_descr mat_C, rocsparse_datatype compute_type, rocsparse_spmm_alg alg, rocsparse_spmm_stage stage, size_t* buffer_size, void* temp_buffer);
@@ -82,6 +93,11 @@ int main() {
   // ROC: ROCSPARSE_EXPORT rocsparse_status rocsparse_spmm(rocsparse_handle handle, rocsparse_operation trans_A, rocsparse_operation trans_B, const void* alpha, const rocsparse_spmat_descr mat_A, const rocsparse_dnmat_descr mat_B, const void* beta, const rocsparse_dnmat_descr mat_C, rocsparse_datatype compute_type, rocsparse_spmm_alg alg, rocsparse_spmm_stage stage, size_t* buffer_size, void* temp_buffer);
   // CHECK: status_t = rocsparse_spmm(handle_t, opA, opB, alpha, spmatA, dnmatB, beta, dnmatC, dataType, spMMAlg_t, rocsparse_spmm_stage_compute, nullptr, tempBuffer);
   status_t = cusparseSpMM(handle_t, opA, opB, alpha, spmatA, dnmatB, beta, dnmatC, dataType, spMMAlg_t, tempBuffer);
+
+  // CUDA: cusparseStatus_t CUSPARSEAPI cusparseSpVV_bufferSize(cusparseHandle_t handle, cusparseOperation_t opX, const cusparseSpVecDescr_t vecX, const cusparseDnVecDescr_t vecY, const void* result, cudaDataType computeType, size_t* bufferSize);
+  // ROC: ROCSPARSE_EXPORT rocsparse_status rocsparse_spvv(rocsparse_handle handle, rocsparse_operation trans, const rocsparse_spvec_descr x, const rocsparse_dnvec_descr y, void* result, rocsparse_datatype compute_type, size_t* buffer_size, void* temp_buffer);
+  // CHECK: status_t = rocsparse_spvv(handle_t, opX, spVecDescr_t, vecY, result, dataType, &bufferSize, nullptr);
+  status_t = cusparseSpVV_bufferSize(handle_t, opX, spVecDescr_t, vecY, result, dataType, &bufferSize);
 #endif
 #endif
 #endif

tests/unit_tests/synthetic/libraries/cusparse2rocsparse_12000.cu

@@ -71,6 +71,7 @@ int main() {
   void *tempBuffer = nullptr;
   void *alpha = nullptr;
   void *beta = nullptr;
+  void* result = nullptr;
 
   // TODO: should be rocsparse_double_complex
   // TODO: add to TypeOverloads cuDoubleComplex -> rocsparse_double_complex under a new option --sparse
@@ -128,6 +129,12 @@ int main() {
   cusparseConstDnMatDescr_t constDnMatDescr = nullptr;
   cusparseConstDnMatDescr_t constDnMatDescrB = nullptr;
 
+  // CHECK: rocsparse_const_spvec_descr constSpVecDescr = nullptr;
+  cusparseConstSpVecDescr_t constSpVecDescr = nullptr;
+
+  // CHECK: rocsparse_const_dnvec_descr constDnVecDescr = nullptr;
+  cusparseConstDnVecDescr_t constDnVecDescr = nullptr;
+
   // CUDA: cusparseStatus_t CUSPARSEAPI cusparseSparseToDense(cusparseHandle_t handle, cusparseConstSpMatDescr_t matA, cusparseDnMatDescr_t matB, cusparseSparseToDenseAlg_t alg, void* externalBuffer);
   // ROC: ROCSPARSE_EXPORT rocsparse_status rocsparse_sparse_to_dense(rocsparse_handle handle, rocsparse_const_spmat_descr mat_A, rocsparse_dnmat_descr mat_B, rocsparse_sparse_to_dense_alg alg, size_t* buffer_size, void* temp_buffer);
   // CHECK: status_t = rocsparse_sparse_to_dense(handle_t, constSpMatDescr, dnmatB, sparseToDenseAlg_t, nullptr, tempBuffer);
@@ -167,6 +174,11 @@ int main() {
   // ROC: ROCSPARSE_EXPORT rocsparse_status rocsparse_spmm(rocsparse_handle handle, rocsparse_operation trans_A, rocsparse_operation trans_B, const void* alpha, rocsparse_const_spmat_descr mat_A, rocsparse_const_dnmat_descr mat_B, const void* beta, const rocsparse_dnmat_descr mat_C, rocsparse_datatype compute_type, rocsparse_spmm_alg alg, rocsparse_spmm_stage stage, size_t* buffer_size, void* temp_buffer);
   // CHECK: status_t = rocsparse_spmm(handle_t, opA, opB, alpha, constSpMatDescr, constDnMatDescr, beta, dnmatC, dataType, spMMAlg_t, rocsparse_spmm_stage_compute, nullptr, tempBuffer);
   status_t = cusparseSpMM(handle_t, opA, opB, alpha, constSpMatDescr, constDnMatDescr, beta, dnmatC, dataType, spMMAlg_t, tempBuffer);
+
+  // CUDA: cusparseStatus_t CUSPARSEAPI cusparseSpVV_bufferSize(cusparseHandle_t handle, cusparseOperation_t opX, cusparseConstSpVecDescr_t vecX, cusparseConstDnVecDescr_t vecY, const void* result, cudaDataType computeType, size_t* bufferSize);
+  // ROC: ROCSPARSE_EXPORT rocsparse_status rocsparse_spvv(rocsparse_handle handle, rocsparse_operation trans, rocsparse_const_spvec_descr x, rocsparse_const_dnvec_descr y, void* result, rocsparse_datatype compute_type, size_t* buffer_size, void* temp_buffer);
+  // CHECK: status_t = rocsparse_spvv(handle_t, opX, constSpVecDescr, constDnVecDescr, result, dataType, &bufferSize, nullptr);
+  status_t = cusparseSpVV_bufferSize(handle_t, opX, constSpVecDescr, constDnVecDescr, result, dataType, &bufferSize);
 #endif
 
   return 0;