[src,build] support for cuda 11.1 and rtx ampere (#4295)

Author: qmmn

src/configure

@@ -368,11 +368,16 @@ Either your CUDA is too new or too old."
           MIN_UNSUPPORTED_GCC_VER_NUM=80000;
           CUSOLVER=true
         ;;
-        10_1 | 10_* | 11_*)
+        10_1 | 10_*)
           MIN_UNSUPPORTED_GCC_VER="9.0"
           MIN_UNSUPPORTED_GCC_VER_NUM=90000;
           CUSOLVER=true
         ;;
+        11_*)
+          MIN_UNSUPPORTED_GCC_VER="10.0"
+          MIN_UNSUPPORTED_GCC_VER_NUM=100000;
+          CUSOLVER=true
+        ;;
         *)
           echo "Unsupported CUDA_VERSION (CUDA_VERSION=$CUDA_VERSION), please report it to Kaldi mailing list, together with 'nvcc -h' or 'ptxas -h' which lists allowed -gencode values..."; exit 1;
         ;;
@@ -392,7 +397,8 @@ Either your CUDA is too new or too old."
             #8_*) CUDA_ARCH="-gencode arch=compute_30,code=sm_30 -gencode arch=compute_35,code=sm_35 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61" ;;
             9_*) CUDA_ARCH="-gencode arch=compute_30,code=sm_30 -gencode arch=compute_35,code=sm_35 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70" ;;
             10_*) CUDA_ARCH="-gencode arch=compute_30,code=sm_30 -gencode arch=compute_35,code=sm_35 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75" ;;
-            11_*) CUDA_ARCH="-gencode arch=compute_35,code=sm_35 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80" ;;
+            11_0) CUDA_ARCH="-gencode arch=compute_35,code=sm_35 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80" ;;
+			11_1 | 11_*) CUDA_ARCH="-gencode arch=compute_35,code=sm_35 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86" ;;
             *) echo "Unsupported CUDA_VERSION (CUDA_VERSION=$CUDA_VERSION), please report it to Kaldi mailing list, together with 'nvcc -h' or 'ptxas -h' which lists allowed -gencode values..."; exit 1 ;;
           esac
         ;;

src/cudamatrix/cublas-wrappers.h

@@ -265,13 +265,13 @@ inline cusparseStatus_t cusparse_csrmm2(cusparseHandle_t handle,
 
   size_t buffer_size;
   status = cusparseSpMM_bufferSize(handle, transA, transB, alpha, matA, matB,
-                                   beta, matC, valType, CUSPARSE_MM_ALG_DEFAULT,
+                                   beta, matC, valType, CUSPARSE_SPMM_CSR_ALG2,
                                    &buffer_size);
   if (status != CUSPARSE_STATUS_SUCCESS) return status;
 
   void *buffer = (buffer_size > 0) ? CuDevice::Instantiate().Malloc(buffer_size) : NULL; 
   status = cusparseSpMM(handle, transA, transB, alpha, matA, matB, beta, matC,
-                        valType, CUSPARSE_MM_ALG_DEFAULT, buffer);
+                        valType, CUSPARSE_SPMM_CSR_ALG2, buffer);
 
   if (status != CUSPARSE_STATUS_SUCCESS) return status;
   if(buffer)