clang-format changes for pr881

example/53_gemv_splitk/gemv_splitk_fp16.cpp

@@ -19,12 +19,10 @@ using CElementOp = PassThrough;
 
 static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
 
-
-
-#define K1 8 //K1PerThread:2,4,8
-#define K0 4 //K0PerBlock:1,2,3,4...32 
-#define N1 2 //Nperthread:2,4,8
-#define B 64 //block-size:64
+#define K1 8 // K1PerThread:2,4,8
+#define K0 4 // K0PerBlock:1,2,3,4...32
+#define N1 2 // Nperthread:2,4,8
+#define B 64 // block-size:64
 
 // clang-format off
 using DeviceGemvInstance = ck::tensor_operation::device::deviceGemvDl/*

include/ck/tensor_operation/gpu/device/device_gemv.hpp

@@ -20,8 +20,7 @@ template <typename ALayout,
           typename CElementwiseOperation>
 struct DeviceGemv : public BaseOperator
 {
-    virtual std::unique_ptr<BaseArgument>
-    MakeArgumentPointer(const void* p_a,
+    virtual std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_a,
                                                               const void* p_b,
                                                               void* p_c,
                                                               ck::index_t M,
@@ -33,7 +32,7 @@ struct DeviceGemv : public BaseOperator
                                                               AElementwiseOperation a_element_op,
                                                               BElementwiseOperation b_element_op,
                                                               CElementwiseOperation c_element_op,
-                        ck::index_t KBatch=1) = 0;
+                                                              ck::index_t KBatch = 1) = 0;
 
     virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
 };

include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp

@@ -658,11 +658,11 @@ struct BlockToCTileMap_3DGrid_KSplit
         return make_tuple(blockIdx.z, blockIdx.y, blockIdx.x);
     }
 
-        //HS: Map 1D block-id to 3D tuple (M,N,K)
+    // HS: Map 1D block-id to 3D tuple (M,N,K)
     __host__ __device__ inline constexpr auto convert_1D_block_idx_to_3D_tuple(
         const index_t& block_1d_id, const index_t& N, const index_t& k_batch) const
     {
-        const auto Ndim= math::integer_divide_ceil(N, NPerBlock);
+        const auto Ndim = math::integer_divide_ceil(N, NPerBlock);
         return make_tuple(((block_1d_id) / (k_batch * Ndim)),
                           (((block_1d_id) / k_batch) % Ndim),
                           (block_1d_id) % k_batch); // returns 3D tuple as (Mid,Nid,Kid)

include/ck/tensor_operation/gpu/grid/gridwise_gemv_splitk.hpp

@@ -27,7 +27,7 @@ template <typename GridwiseGemv,
           typename Block2CTileMap>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
-__launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
         kernel_gemv_dl_v1r3(
             typename GridwiseGemv::Argument karg,