use format-11.1 (#38)

* format-11.1

* md-link-config

src/turbomind/utils/IA3.h

@@ -18,7 +18,8 @@
 
 namespace turbomind {
 
-enum IA3_config {
+enum IA3_config
+{
     KEY_ADAPTER   = 1 << 0,
     VALUE_ADAPTER = 1 << 1,
     MLP_ADAPTER   = 1 << 2,

src/turbomind/utils/Tensor.cc

@@ -59,9 +59,9 @@ Tensor::Tensor(const MemoryType          _where,
 
 void Tensor::parseNpyIntro(FILE*& f_ptr, uint32_t& header_len, uint32_t& start_data)
 {
-    const char magic[]                   = "\x93"
-                                           "NUMPY";
-    char       magic_test[sizeof(magic)] = "\0";
+    const char magic[] = "\x93"
+                         "NUMPY";
+    char magic_test[sizeof(magic)] = "\0";
 
     size_t n_elems = fread((void*)magic_test, sizeof(char), sizeof(magic) - 1, f_ptr);
     if (n_elems != sizeof(magic) - 1 || std::string(magic) != std::string(magic_test)) {
@@ -292,8 +292,8 @@ void Tensor::saveNpy(const std::string& filename) const
         cudaMemcpy(cpu_data, data, tensor_size * Tensor::getTypeSize(type), cudaMemcpyDeviceToHost);
     }
 
-    const char    magic[]   = "\x93"
-                              "NUMPY";
+    const char magic[] = "\x93"
+                         "NUMPY";
     const uint8_t npy_major = 1;
     const uint8_t npy_minor = 0;
 

src/turbomind/utils/Tensor.h

@@ -35,7 +35,8 @@
 
 namespace turbomind {
 
-typedef enum datatype_enum {
+typedef enum datatype_enum
+{
     TYPE_INVALID,
     TYPE_BOOL,
     TYPE_UINT8,
@@ -98,7 +99,8 @@ DataType getTensorType()
     }
 }
 
-typedef enum memorytype_enum {
+typedef enum memorytype_enum
+{
     MEMORY_CPU,
     MEMORY_CPU_PINNED,
     MEMORY_GPU

src/turbomind/utils/activation_types.h

@@ -20,7 +20,8 @@
 
 namespace turbomind {
 
-enum class ActivationType {
+enum class ActivationType
+{
     Gelu,
     Relu,
     Silu,

src/turbomind/utils/allocator.h

@@ -49,13 +49,15 @@
 
 namespace turbomind {
 
-enum class AllocatorType {
+enum class AllocatorType
+{
     CUDA,
     TF,
     TH
 };
 
-enum class ReallocType {
+enum class ReallocType
+{
     INCREASE,
     REUSE,
     DECREASE,

src/turbomind/utils/cublasFP8MMWrapper.h

@@ -169,9 +169,9 @@ public:
                        cudaStream_t         stream);
 
 private:
-    int                                 version_major_, version_minor_, version_patch_;
+    int                         version_major_, version_minor_, version_patch_;
     turbomind::qgmma1x1Launcher qgmmaLauncher;
-    void*                               cublas_workspace_qgemm_ = nullptr;
+    void*                       cublas_workspace_qgemm_ = nullptr;
 };
 
 }  // namespace turbomind

src/turbomind/utils/cuda_fp8_utils.h

@@ -35,7 +35,8 @@ namespace turbomind {
 
 const float FP8_E4M3_MAX = 480.0f;
 
-enum QUANTIZE_MODE {
+enum QUANTIZE_MODE
+{
     PER_CHANNEL,
     PER_TENSOR,
     PER_CHANNEL_WEIGHT_PER_TENSOR_ACT

src/turbomind/utils/cuda_utils.h

@@ -46,7 +46,8 @@ half4;
 
 /* **************************** type definition ***************************** */
 
-enum CublasDataType {
+enum CublasDataType
+{
     FLOAT_DATATYPE    = 0,
     HALF_DATATYPE     = 1,
     BFLOAT16_DATATYPE = 2,
@@ -54,7 +55,8 @@ enum CublasDataType {
     FP8_DATATYPE      = 4
 };
 
-enum FtCudaDataType {
+enum FtCudaDataType
+{
     FP32 = 0,
     FP16 = 1,
     BF16 = 2,
@@ -62,7 +64,8 @@ enum FtCudaDataType {
     FP8  = 4
 };
 
-enum class OperationType {
+enum class OperationType
+{
     FP32,
     FP16,
     BF16,
@@ -212,7 +215,7 @@ inline void myAssert(bool result, const char* const file, int const line, std::s
     do {                                                                                                               \
         bool is_valid_val = (val);                                                                                     \
         if (!is_valid_val) {                                                                                           \
-            turbomind::myAssert(is_valid_val, __FILE__, __LINE__, (info));                                     \
+            turbomind::myAssert(is_valid_val, __FILE__, __LINE__, (info));                                             \
         }                                                                                                              \
     } while (0)
 

src/turbomind/utils/gemm.h

@@ -47,7 +47,8 @@ namespace turbomind {
 // A wrapper of cublas or cusparse matrix operator.
 //  - GEMM_OP_N = CUBLAS_OP_N or CUSPARSE_OP_N
 //  - GEMM_OP_T = CUBLAS_OP_T or CUSPARSE_OP_T
-enum GemmOp {
+enum GemmOp
+{
     GEMM_OP_N,
     GEMM_OP_T
 };

src/turbomind/utils/gemm_test/gpt_gemm_func.cc

@@ -639,7 +639,8 @@ void generate_gpt_gemm_config(int   batch_size,
                     cudaStream_t              streams[1]  = {stream};
                     CHECK_CUSPARSE(cusparseLtStructuredDescriptorInit(
                         &handle, &mat_A, m, k, m, alignment, CUDA_R_16F, order, CUSPARSELT_SPARSITY_50_PERCENT))
-                    CHECK_CUSPARSE(cusparseLtDenseDescriptorInit(&handle, &mat_B, k, n, k, alignment, CUDA_R_16F, order))
+                    CHECK_CUSPARSE(
+                        cusparseLtDenseDescriptorInit(&handle, &mat_B, k, n, k, alignment, CUDA_R_16F, order))
                     CHECK_CUSPARSE(
                         cusparseLtDenseDescriptorInit(&handle, &mat_C, m, n, m, alignment, CUDA_R_16F, order))
                     cudaDeviceSynchronize();

src/turbomind/utils/gemm_test/t5_gemm_func.cc

@@ -638,7 +638,8 @@ void generate_t5_gemm_config(int   batch_size,
                     cudaStream_t              streams[1]  = {stream};
                     CHECK_CUSPARSE(cusparseLtStructuredDescriptorInit(
                         &handle, &mat_A, m, k, m, alignment, CUDA_R_16F, order, CUSPARSELT_SPARSITY_50_PERCENT))
-                    CHECK_CUSPARSE(cusparseLtDenseDescriptorInit(&handle, &mat_B, k, n, k, alignment, CUDA_R_16F, order))
+                    CHECK_CUSPARSE(
+                        cusparseLtDenseDescriptorInit(&handle, &mat_B, k, n, k, alignment, CUDA_R_16F, order))
                     CHECK_CUSPARSE(
                         cusparseLtDenseDescriptorInit(&handle, &mat_C, m, n, m, alignment, CUDA_R_16F, order))
                     cudaDeviceSynchronize();

src/turbomind/utils/gemm_test/xlnet_gemm_func.cc

@@ -91,15 +91,15 @@ void generate_xlnet_gemm_config(int   batch_size,
     int               ldc[gemm_num];
     int               strideC[gemm_num];
     cublasOperation_t transa[gemm_num]     = {CUBLAS_OP_N,
-                                              CUBLAS_OP_N,
-                                              CUBLAS_OP_T,
-                                              CUBLAS_OP_T,
-                                              CUBLAS_OP_T,
-                                              CUBLAS_OP_T,
-                                              CUBLAS_OP_N,
-                                              CUBLAS_OP_T,
-                                              CUBLAS_OP_N,
-                                              CUBLAS_OP_N};
+                                          CUBLAS_OP_N,
+                                          CUBLAS_OP_T,
+                                          CUBLAS_OP_T,
+                                          CUBLAS_OP_T,
+                                          CUBLAS_OP_T,
+                                          CUBLAS_OP_N,
+                                          CUBLAS_OP_T,
+                                          CUBLAS_OP_N,
+                                          CUBLAS_OP_N};
     cublasOperation_t transb[gemm_num]     = {CUBLAS_OP_N};
     int               batchCount[gemm_num] = {1};
     char              mess[gemm_num][256];

src/turbomind/utils/logger.h

@@ -27,7 +27,8 @@ namespace turbomind {
 class Logger {
 
 public:
-    enum Level {
+    enum Level
+    {
         TRACE   = 0,
         DEBUG   = 10,
         INFO    = 20,
@@ -40,7 +41,7 @@ public:
         thread_local Logger instance;
         return instance;
     }
-    Logger(Logger const&)         = delete;
+    Logger(Logger const&) = delete;
     void operator=(Logger const&) = delete;
 
     template<typename... Args>
@@ -108,8 +109,8 @@ private:
 
 #define TM_LOG(level, ...)                                                                                             \
     do {                                                                                                               \
-        if (turbomind::Logger::getLogger().getLevel() <= level) {                                              \
-            turbomind::Logger::getLogger().log(level, __VA_ARGS__);                                            \
+        if (turbomind::Logger::getLogger().getLevel() <= level) {                                                      \
+            turbomind::Logger::getLogger().log(level, __VA_ARGS__);                                                    \
         }                                                                                                              \
     } while (0)
 

src/turbomind/utils/mpi_utils.h

@@ -43,7 +43,8 @@ namespace turbomind {
 namespace mpi {
 
 // A wrapper of MPI data type. MPI_TYPE_{data_type}
-enum MpiType {
+enum MpiType
+{
     MPI_TYPE_BYTE,
     MPI_TYPE_CHAR,
     MPI_TYPE_INT,
@@ -53,7 +54,8 @@ enum MpiType {
 };
 
 // A wrapper of the level of MPI thread support
-enum MpiThreadSupport {
+enum MpiThreadSupport
+{
     THREAD_SINGLE,
     THREAD_FUNNELED,
     THREAD_SERIALIZED,

src/turbomind/utils/prompt_learning.h

@@ -19,7 +19,8 @@
 
 namespace turbomind {
 
-enum class PromptLearningType {
+enum class PromptLearningType
+{
     no_prompt,
     soft_prompt,
     prefix_prompt,