去掉blaslt的依赖

src/turbomind/utils/cublasMMWrapper.cc

@@ -802,10 +802,10 @@ std::pair<bool, cublasLtMatmulAlgo_t> cublasMMWrapper::findBestAlgo(cublasLtHand
                                                                     cublasLtMatrixLayout_t Ddesc,
                                                                     cudaStream_t           stream)
 {
-#if (CUBLAS_VERSION) <= 11601
+//#if (CUBLAS_VERSION) <= 11601
     FT_CHECK_WITH_INFO(false, "CUBLAS version too low.");
     return {false, cublasLtMatmulAlgo_t{}};
-#else
+/*#else
     size_t returnSize;
     int32_t pointer_mode;
     cublasLtMatmulDescGetAttribute(
@@ -893,7 +893,7 @@ std::pair<bool, cublasLtMatmulAlgo_t> cublasMMWrapper::findBestAlgo(cublasLtHand
     }
 
     return {best_time != INFINITY, result.algo};
-#endif
+#endif*/
 }
 
 cublasMMWrapper::MatrixLayout cublasMMWrapper::createMatrixLayout(cublasLtMatrixLayout_t Mdesc)
@@ -901,6 +901,8 @@ cublasMMWrapper::MatrixLayout cublasMMWrapper::createMatrixLayout(cublasLtMatrix
     size_t       returnSize;
     MatrixLayout m_layout;
 
+    FT_CHECK_WITH_INFO(false, "cublasLtMatrixLayoutGetAttribute is not support.");
+/*
     cublasLtMatrixLayoutGetAttribute(
         Mdesc, CUBLASLT_MATRIX_LAYOUT_TYPE, &std::get<0>(m_layout), sizeof(std::get<0>(m_layout)), &returnSize);
     cublasLtMatrixLayoutGetAttribute(
@@ -909,7 +911,7 @@ cublasMMWrapper::MatrixLayout cublasMMWrapper::createMatrixLayout(cublasLtMatrix
         Mdesc, CUBLASLT_MATRIX_LAYOUT_ROWS, &std::get<2>(m_layout), sizeof(std::get<2>(m_layout)), &returnSize);
     cublasLtMatrixLayoutGetAttribute(
         Mdesc, CUBLASLT_MATRIX_LAYOUT_COLS, &std::get<3>(m_layout), sizeof(std::get<3>(m_layout)), &returnSize);
-
+*/
     return m_layout;
 }