Merge pull request #1787 from aakanksha555/amd-staging

[HIPIFY][Tensor][feature] hipTensor support - Functions - Part 7
ROCm · Dec 12, 2024 · 43349da · 43349da
2 parents e63dfbf + 5b4b15b
commit 43349da
Show file tree

Hide file tree

Showing 5 changed files with 93 additions and 17 deletions.
diff --git a/bin/hipify-perl b/bin/hipify-perl
@@ -6135,6 +6135,8 @@ sub simpleSubstitutions {
     subst("cusparseZnnz", "hipsparseZnnz", "library");
     subst("cusparseZnnz_compress", "hipsparseZnnz_compress", "library");
     subst("cusparseZsctr", "hipsparseZsctr", "library");
+    subst("cutensorCreate", "hiptensorCreate", "library");
+    subst("cutensorDestroy", "hiptensorDestroy", "library");
     subst("nvrtcAddNameExpression", "hiprtcAddNameExpression", "library");
     subst("nvrtcCompileProgram", "hiprtcCompileProgram", "library");
     subst("nvrtcCreateProgram", "hiprtcCreateProgram", "library");
@@ -7367,6 +7369,7 @@ sub simpleSubstitutions {
     subst("cusparseStatus_t", "hipsparseStatus_t", "type");
     subst("cutensorAlgo_t", "hiptensorAlgo_t", "type");
     subst("cutensorDataType_t", "hiptensorComputeType_t", "type");
+    subst("cutensorHandle_t", "hiptensorHandle_t", "type");
     subst("cutensorOperator_t", "hiptensorOperator_t", "type");
     subst("cutensorStatus_t", "hiptensorStatus_t", "type");
     subst("cutensorWorksizePreference_t", "hiptensorWorksizePreference_t", "type");
@@ -9866,6 +9869,9 @@ sub warnHipOnlyUnsupportedFunctions {
         "cutensorPlanAttribute_t",
         "cutensorOperationDescriptorAttribute_t",
         "cutensorJitMode_t",
+        "cutensorHandleWritePlanCacheToFile",
+        "cutensorHandleResizePlanCache",
+        "cutensorHandleReadPlanCacheFromFile",
         "cutensorCacheMode_t",
         "cutensorAutotuneMode_t",
         "cusparseZhybsv_solve",
@@ -11186,6 +11192,9 @@ sub warnRocOnlyUnsupportedFunctions {
         "cutensorPlanAttribute_t",
         "cutensorOperationDescriptorAttribute_t",
         "cutensorJitMode_t",
+        "cutensorHandleWritePlanCacheToFile",
+        "cutensorHandleResizePlanCache",
+        "cutensorHandleReadPlanCacheFromFile",
         "cutensorCacheMode_t",
         "cutensorAutotuneMode_t",
         "cusparseZhybsv_solve",

diff --git a/docs/tables/CUTENSOR_API_supported_by_HIP.md b/docs/tables/CUTENSOR_API_supported_by_HIP.md
@@ -90,21 +90,21 @@
 |`CUTENSOR_R_64U`|2.0.0.0| | | | | | | | | |
 |`CUTENSOR_R_8I`|2.0.0.0| | | |`HIPTENSOR_COMPUTE_8I`|5.7.0| | | | |
 |`CUTENSOR_R_8U`|2.0.0.0| | | |`HIPTENSOR_COMPUTE_8U`|5.7.0| | | | |
-|`CUTENSOR_STATUS_ALLOC_FAILED`| | | | |`HIPTENSOR_STATUS_ALLOC_FAILED`| | | | | |
-|`CUTENSOR_STATUS_ARCH_MISMATCH`| | | | |`HIPTENSOR_STATUS_ARCH_MISMATCH`| | | | | |
-|`CUTENSOR_STATUS_CUBLAS_ERROR`| | | | | | | | | | |
-|`CUTENSOR_STATUS_CUDA_ERROR`| | | | | | | | | | |
-|`CUTENSOR_STATUS_EXECUTION_FAILED`| | | | |`HIPTENSOR_STATUS_EXECUTION_FAILED`| | | | | |
-|`CUTENSOR_STATUS_INSUFFICIENT_DRIVER`| | | | |`HIPTENSOR_STATUS_INSUFFICIENT_DRIVER`| | | | | |
-|`CUTENSOR_STATUS_INSUFFICIENT_WORKSPACE`| | | | |`HIPTENSOR_STATUS_INSUFFICIENT_WORKSPACE`| | | | | |
-|`CUTENSOR_STATUS_INTERNAL_ERROR`| | | | |`HIPTENSOR_STATUS_INTERNAL_ERROR`| | | | | |
-|`CUTENSOR_STATUS_INVALID_VALUE`| | | | |`HIPTENSOR_STATUS_INVALID_VALUE`| | | | | |
-|`CUTENSOR_STATUS_IO_ERROR`| | | | |`HIPTENSOR_STATUS_IO_ERROR`| | | | | |
-|`CUTENSOR_STATUS_LICENSE_ERROR`| | | | | | | | | | |
-|`CUTENSOR_STATUS_MAPPING_ERROR`| | | | | | | | | | |
-|`CUTENSOR_STATUS_NOT_INITIALIZED`| | | | |`HIPTENSOR_STATUS_NOT_INITIALIZED`| | | | | |
-|`CUTENSOR_STATUS_NOT_SUPPORTED`| | | | |`HIPTENSOR_STATUS_NOT_SUPPORTED`| | | | | |
-|`CUTENSOR_STATUS_SUCCESS`| | | | |`HIPTENSOR_STATUS_SUCCESS`| | | | | |
+|`CUTENSOR_STATUS_ALLOC_FAILED`|1.0.1.0| | | |`HIPTENSOR_STATUS_ALLOC_FAILED`|5.7.0| | | | |
+|`CUTENSOR_STATUS_ARCH_MISMATCH`|1.0.1.0| | | |`HIPTENSOR_STATUS_ARCH_MISMATCH`|5.7.0| | | | |
+|`CUTENSOR_STATUS_CUBLAS_ERROR`|1.0.1.0| | | | | | | | | |
+|`CUTENSOR_STATUS_CUDA_ERROR`|1.0.1.0| | | | | | | | | |
+|`CUTENSOR_STATUS_EXECUTION_FAILED`|1.0.1.0| | | |`HIPTENSOR_STATUS_EXECUTION_FAILED`|5.7.0| | | | |
+|`CUTENSOR_STATUS_INSUFFICIENT_DRIVER`|1.0.1.0| | | |`HIPTENSOR_STATUS_INSUFFICIENT_DRIVER`|5.7.0| | | | |
+|`CUTENSOR_STATUS_INSUFFICIENT_WORKSPACE`|1.0.1.0| | | |`HIPTENSOR_STATUS_INSUFFICIENT_WORKSPACE`|5.7.0| | | | |
+|`CUTENSOR_STATUS_INTERNAL_ERROR`|1.0.1.0| | | |`HIPTENSOR_STATUS_INTERNAL_ERROR`|5.7.0| | | | |
+|`CUTENSOR_STATUS_INVALID_VALUE`|1.0.1.0| | | |`HIPTENSOR_STATUS_INVALID_VALUE`|5.7.0| | | | |
+|`CUTENSOR_STATUS_IO_ERROR`|1.2.0.0| | | |`HIPTENSOR_STATUS_IO_ERROR`|5.7.0| | | | |
+|`CUTENSOR_STATUS_LICENSE_ERROR`|1.0.1.0| | | | | | | | | |
+|`CUTENSOR_STATUS_MAPPING_ERROR`|1.0.1.0| | | | | | | | | |
+|`CUTENSOR_STATUS_NOT_INITIALIZED`|1.0.1.0| | | |`HIPTENSOR_STATUS_NOT_INITIALIZED`|5.7.0| | | | |
+|`CUTENSOR_STATUS_NOT_SUPPORTED`|1.0.1.0| | | |`HIPTENSOR_STATUS_NOT_SUPPORTED`|5.7.0| | | | |
+|`CUTENSOR_STATUS_SUCCESS`|1.0.1.0| | | |`HIPTENSOR_STATUS_SUCCESS`|5.7.0| | | | |
 |`CUTENSOR_WORKSPACE_DEFAULT`|2.0.0.0| | | | | | | | | |
 |`CUTENSOR_WORKSPACE_MAX`|1.0.1.0| | | |`HIPTENSOR_WORKSPACE_MAX`|5.7.0| | | | |
 |`CUTENSOR_WORKSPACE_MIN`|1.0.1.0| | | |`HIPTENSOR_WORKSPACE_MIN`|5.7.0| | | | |
@@ -113,17 +113,24 @@
 |`cutensorAutotuneMode_t`|1.2.0.0| | | | | | | | | |
 |`cutensorCacheMode_t`|1.2.0.0| | | | | | | | | |
 |`cutensorDataType_t`|2.0.0.0| | | |`hiptensorComputeType_t`|5.7.0| | | | |
+|`cutensorHandle_t`| | | | |`hiptensorHandle_t`| | | | | |
 |`cutensorJitMode_t`|2.0.0.0| | | | | | | | | |
 |`cutensorOperationDescriptorAttribute_t`|2.0.0.0| | | | | | | | | |
 |`cutensorOperator_t`|1.0.1.0| | | |`hiptensorOperator_t`|5.7.0| | | | |
 |`cutensorPlanAttribute_t`|2.0.0.0| | | | | | | | | |
 |`cutensorPlanPreferenceAttribute_t`|2.0.0.0| | | | | | | | | |
-|`cutensorStatus_t`| | | | |`hiptensorStatus_t`| | | | | |
+|`cutensorStatus_t`|1.0.1.0| | | |`hiptensorStatus_t`|5.7.0| | | | |
 |`cutensorWorksizePreference_t`|1.0.1.0| | | |`hiptensorWorksizePreference_t`|5.7.0| | | | |
 
 ## **2. CUTENSOR Function Reference**
 
-Unsupported
+|**CUDA**|**A**|**D**|**C**|**R**|**HIP**|**A**|**D**|**C**|**R**|**E**|
+|:--|:-:|:-:|:-:|:-:|:--|:-:|:-:|:-:|:-:|:-:|
+|`cutensorCreate`|1.7.0.0| | | |`hiptensorCreate`|5.7.0| | | | |
+|`cutensorDestroy`|1.7.0.0| | | |`hiptensorDestroy`|5.7.0| | | | |
+|`cutensorHandleReadPlanCacheFromFile`|2.0.0.0| | | | | | | | | |
+|`cutensorHandleResizePlanCache`|2.0.0.0| | | | | | | | | |
+|`cutensorHandleWritePlanCacheToFile`|2.0.0.0| | | | | | | | | |
 
 
 \*A - Added; D - Deprecated; C - Changed; R - Removed; E - Experimental
diff --git a/src/CUDA2HIP_TENSOR_API_functions.cpp b/src/CUDA2HIP_TENSOR_API_functions.cpp
@@ -23,13 +23,25 @@ THE SOFTWARE.
 #include "CUDA2HIP.h"
 
 const std::map<llvm::StringRef, hipCounter> CUDA_TENSOR_FUNCTION_MAP {
+  {"cutensorCreate",                                              {"hiptensorCreate",                                     "",                                     CONV_LIB_FUNC, API_TENSOR, 2}},
+  {"cutensorDestroy",                                             {"hiptensorDestroy",                                    "",                                     CONV_LIB_FUNC, API_TENSOR, 2}},
+  {"cutensorHandleResizePlanCache",                               {"",                                                    "",                                     CONV_LIB_FUNC, API_TENSOR, 2, UNSUPPORTED}},
+  {"cutensorHandleWritePlanCacheToFile",                          {"",                                                    "",                                     CONV_LIB_FUNC, API_TENSOR, 2, UNSUPPORTED}},
+  {"cutensorHandleReadPlanCacheFromFile",                         {"",                                                    "",                                     CONV_LIB_FUNC, API_TENSOR, 2, UNSUPPORTED}},
 };
 
 
 const std::map<llvm::StringRef, cudaAPIversions> CUDA_TENSOR_FUNCTION_VER_MAP {
+  {"cutensorCreate",                                 {CUTENSOR_1700,  CUDA_0,    CUDA_0   }},
+  {"cutensorDestroy",                                {CUTENSOR_1700,  CUDA_0,    CUDA_0   }},
+  {"cutensorHandleResizePlanCache",                  {CUTENSOR_2000,  CUDA_0,    CUDA_0   }},
+  {"cutensorHandleWritePlanCacheToFile",             {CUTENSOR_2000,  CUDA_0,    CUDA_0   }},
+  {"cutensorHandleReadPlanCacheFromFile",            {CUTENSOR_2000,  CUDA_0,    CUDA_0   }},
 };
 
 const std::map<llvm::StringRef, hipAPIversions> HIP_TENSOR_FUNCTION_VER_MAP {
+  {"hiptensorCreate",                                {HIP_5070,       HIP_0,         HIP_0,       }},
+  {"hiptensorDestroy",                               {HIP_5070,       HIP_0,         HIP_0,       }},
 };
 
 const std::map<unsigned int, llvm::StringRef> CUDA_TENSOR_API_SECTION_MAP {

diff --git a/src/CUDA2HIP_TENSOR_API_types.cpp b/src/CUDA2HIP_TENSOR_API_types.cpp
@@ -152,6 +152,8 @@ const std::map<llvm::StringRef, hipCounter> CUDA_TENSOR_TYPE_NAME_MAP {
   {"cutensorPlanAttribute_t",                          {"",                                                         "",           CONV_TYPE, API_TENSOR, 1, UNSUPPORTED}},
   {"CUTENSOR_PLAN_REQUIRED_WORKSPACE",                 {"",                                                         "",           CONV_NUMERIC_LITERAL, API_TENSOR, 1, UNSUPPORTED}},
 
+  {"cutensorHandle_t",                                 {"hiptensorHandle_t",                                        "",           CONV_TYPE, API_TENSOR, 1}},
+
 };
 
 const std::map<llvm::StringRef, cudaAPIversions> CUDA_TENSOR_TYPE_NAME_VER_MAP {
@@ -218,6 +220,22 @@ const std::map<llvm::StringRef, cudaAPIversions> CUDA_TENSOR_TYPE_NAME_VER_MAP {
   {"CUTENSOR_OP_MAX",                                  {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
   {"CUTENSOR_OP_MIN",                                  {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
   {"CUTENSOR_OP_UNKNOWN",                              {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"cutensorStatus_t",                                 {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_SUCCESS",                          {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_NOT_INITIALIZED",                  {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_ALLOC_FAILED",                     {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_INVALID_VALUE",                    {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_ARCH_MISMATCH",                    {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_MAPPING_ERROR",                    {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_EXECUTION_FAILED",                 {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_INTERNAL_ERROR",                   {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_NOT_SUPPORTED",                    {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_LICENSE_ERROR",                    {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_CUBLAS_ERROR",                     {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_CUDA_ERROR",                       {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_INSUFFICIENT_WORKSPACE",           {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_INSUFFICIENT_DRIVER",              {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
+  {"CUTENSOR_STATUS_IO_ERROR",                         {CUTENSOR_1200,  CUDA_0,        CUDA_0,      }},
   {"cutensorAlgo_t",                                   {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
   {"CUTENSOR_ALGO_DEFAULT_PATIENT",                    {CUTENSOR_1400,  CUDA_0,        CUDA_0,      }},
   {"CUTENSOR_ALGO_GETT",                               {CUTENSOR_1010,  CUDA_0,        CUDA_0,      }},
@@ -279,6 +297,18 @@ const std::map<llvm::StringRef, hipAPIversions> HIP_TENSOR_TYPE_NAME_VER_MAP {
   {"HIPTENSOR_OP_MAX",                                 {HIP_6030,       HIP_0,         HIP_0,       }},
   {"HIPTENSOR_OP_MIN",                                 {HIP_6030,       HIP_0,         HIP_0,       }},
   {"HIPTENSOR_OP_UNKNOWN",                             {HIP_5070,       HIP_0,         HIP_0,       }},
+  {"hiptensorStatus_t",                                {HIP_5070,       HIP_0,         HIP_0,       }},
+  {"HIPTENSOR_STATUS_SUCCESS",                         {HIP_5070,       HIP_0,         HIP_0,       }},
+  {"HIPTENSOR_STATUS_NOT_INITIALIZED",                 {HIP_5070,       HIP_0,         HIP_0,       }},
+  {"HIPTENSOR_STATUS_ALLOC_FAILED",                    {HIP_5070,       HIP_0,         HIP_0,       }},
+  {"HIPTENSOR_STATUS_INVALID_VALUE",                   {HIP_5070,       HIP_0,         HIP_0,       }},
+  {"HIPTENSOR_STATUS_ARCH_MISMATCH",                   {HIP_5070,       HIP_0,         HIP_0,       }},
+  {"HIPTENSOR_STATUS_EXECUTION_FAILED",                {HIP_5070,       HIP_0,         HIP_0,       }},
+  {"HIPTENSOR_STATUS_INTERNAL_ERROR",                  {HIP_5070,       HIP_0,         HIP_0,       }},
+  {"HIPTENSOR_STATUS_NOT_SUPPORTED",                   {HIP_5070,       HIP_0,         HIP_0,       }},
+  {"HIPTENSOR_STATUS_INSUFFICIENT_WORKSPACE",          {HIP_5070,       HIP_0,         HIP_0,       }},
+  {"HIPTENSOR_STATUS_INSUFFICIENT_DRIVER",             {HIP_5070,       HIP_0,         HIP_0,       }},
+  {"HIPTENSOR_STATUS_IO_ERROR",                        {HIP_5070,       HIP_0,         HIP_0,       }},
   {"hiptensorAlgo_t",                                  {HIP_5070,       HIP_0,         HIP_0,       }},
   {"HIPTENSOR_ALGO_DEFAULT",                           {HIP_5070,       HIP_0,         HIP_0,       }},
   {"HIPTENSOR_ALGO_DEFAULT_PATIENT",                   {HIP_5070,       HIP_0,         HIP_0,       }},

diff --git a/tests/unit_tests/synthetic/libraries/cutensor2hiptensor.cu b/tests/unit_tests/synthetic/libraries/cutensor2hiptensor.cu
@@ -100,5 +100,23 @@ int main() {
   cutensorWorksizePreference_t TENSOR_WORKSPACE_RECOMMENDED = CUTENSOR_WORKSPACE_RECOMMENDED;
 #endif
 
+  // CHECK: hiptensorHandle_t handle;
+  cutensorHandle_t handle;
+
+  //CHECK: hiptensorStatus_t status;
+  cutensorStatus_t status;
+
+#if (CUTENSOR_MAJOR == 1 && CUTENSOR_MINOR >= 7) || CUTENSOR_MAJOR >= 2
+  // CUDA: cutensorStatus_t cutensorCreate(cutensorHandle_t* handle);
+  // HIP: hiptensorStatus_t hiptensorCreate(hiptensorHandle_t** handle);
+  // CHECK: status = hiptensorCreate(&handle);
+  status = cutensorCreate(&handle);
+
+  // CUDA: cutensorStatus_t cutensorDestroy(cutensorHandle_t handle);
+  // HIP: hiptensorStatus_t hiptensorDestroy(hiptensorHandle_t* handle);
+  // CHECK: status = hiptensorDestroy(handle);
+  status = cutensorDestroy(handle);
+#endif
+
   return 0;
 }