3.6.0 update (#2005)

* 3.6.0 update

* doc and swap stuff

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
Co-authored-by: Haicheng Wu <haichengw@nvidia.com>

tools/library/src/reference/conv_reference_operation.h

@@ -432,7 +432,12 @@ public:
     void const *arguments,
     void *host_workspace,
     void *device_workspace = nullptr,
-    cudaStream_t stream = nullptr) const {
+    cudaStream_t stream = nullptr,
+    bool launch_with_pdl = false) const {
+    
+    if (launch_with_pdl) {
+      return Status::kErrorNotSupported;
+    }
 
     ConvArguments const  &args = *static_cast<ConvArguments const *>(arguments);
 

tools/library/src/reference/gemm_reference_operation.h

@@ -192,7 +192,12 @@ public:
     void const *arguments,
     void *host_workspace,
     void *device_workspace = nullptr,
-    cudaStream_t stream = nullptr) const {
+    cudaStream_t stream = nullptr,
+    bool launch_with_pdl = false) const {
+
+    if (launch_with_pdl) {
+      return Status::kErrorNotSupported;
+    }
 
     GemmUniversalConfiguration const &config = *static_cast<GemmUniversalConfiguration const *>(host_workspace);
     GemmUniversalArguments const &args = *static_cast<GemmUniversalArguments const *>(arguments);

tools/library/src/reference/gemm_s8_s8_s32.cu

@@ -77,7 +77,7 @@ void initialize_gemm_reference_operations_s8_s8_s32(Manifest &manifest) {
     int8_t,                           // ElementA
     int8_t,                           // ElementB
     int32_t,                          // ElementC
-    int32_t,                          // ElementScalar / ElementCompute
+    float,                            // ElementScalar / ElementCompute
     int32_t,                          // ElementAccumulator
     int32_t                           // ElementD
   >(manifest);