Formatting

src/targets/gpu/fuse_ck.cpp

@@ -131,7 +131,8 @@ struct find_ck_batched_gemm
     void apply(module_pass_manager& mpm, const match::matcher_result& r) const
     {
         auto ins = r.result;
-        mpm.get_module().replace_instruction(ins, ck_batched_gemm{ins->get_operator()}, ins->inputs());
+        mpm.get_module().replace_instruction(
+            ins, ck_batched_gemm{ins->get_operator()}, ins->inputs());
     }
 };
 

src/targets/gpu/jit/ck_batched_gemm.cpp

@@ -131,10 +131,7 @@ static std::size_t get_tuning_for(const std::vector<shape>& inputs)
     return it->second;
 }
 
-static std::size_t get_batch_stride(const shape& s)
-{
-    return s.strides()[s.strides().size() - 3];
-}
+static std::size_t get_batch_stride(const shape& s) { return s.strides()[s.strides().size() - 3]; }
 
 struct ck_batched_gemm_compiler : compiler<ck_batched_gemm_compiler>
 {
@@ -209,7 +206,8 @@ struct ck_batched_gemm_compiler : compiler<ck_batched_gemm_compiler>
         options.kernel_name    = "ck_batched_gemm_kernel";
         options.virtual_inputs = inputs;
 
-        auto src = interpolate_string(ck_batched_gemm_kernel, {{"instance", join_strings(instance, ",")}});
+        auto src =
+            interpolate_string(ck_batched_gemm_kernel, {{"instance", join_strings(instance, ",")}});
 
         return compile_hip_code_object(src, options);
     }

src/targets/gpu/kernels/include/migraphx/kernels/algorithm.hpp

@@ -111,9 +111,10 @@ constexpr F for_each(Iterator first, Iterator last, F f)
 }
 
 template <class Iterator, class T>
-constexpr void fill (Iterator first, Iterator last, const T& val)
+constexpr void fill(Iterator first, Iterator last, const T& val)
 {
-  while (first != last) {
+    while(first != last)
+    {
         *first = val;
         ++first;
     }

src/targets/gpu/kernels/include/migraphx/kernels/ck_batched_gemm.hpp

@@ -94,15 +94,17 @@ struct ComputePtrOffsetOfStridedBatch
     ck::index_t BatchStrideE_;
 };
 
-
 template <class G, class Settings, class A, class B, class E, class... Ds>
 __device__ void ck_batched_gemm(Settings s, A a, B b, E e, Ds... ds)
 {
     constexpr const G gemm{};
 
-    constexpr const auto a_grid_desc_m_k = gemm.matrix_padder.PadADescriptor_M_K(to_ck_batched_tensor<A>());
-    constexpr const auto b_grid_desc_n_k = gemm.matrix_padder.PadBDescriptor_N_K(to_ck_batched_tensor<B>());
-    constexpr const auto e_grid_desc_m_n = gemm.matrix_padder.PadCDescriptor_M_N(to_ck_batched_tensor<E>());
+    constexpr const auto a_grid_desc_m_k =
+        gemm.matrix_padder.PadADescriptor_M_K(to_ck_batched_tensor<A>());
+    constexpr const auto b_grid_desc_n_k =
+        gemm.matrix_padder.PadBDescriptor_N_K(to_ck_batched_tensor<B>());
+    constexpr const auto e_grid_desc_m_n =
+        gemm.matrix_padder.PadCDescriptor_M_N(to_ck_batched_tensor<E>());
     constexpr const auto ds_grid_desc_m_n =
         ck::make_tuple(gemm.matrix_padder.PadCDescriptor_M_N(to_ck_batched_tensor<Ds>())...);
     constexpr const auto block_2_etile_map = gemm.MakeDefaultBlock2ETileMap(e_grid_desc_m_n);
@@ -127,14 +129,15 @@ __device__ void ck_batched_gemm(Settings s, A a, B b, E e, Ds... ds)
 
     static constexpr ck::index_t NumDTensor = gemm.NumDTensor;
     std::array<ck::index_t, NumDTensor> batchStrideDs;
-    ck::static_for<0, NumDTensor, 1>{}(
-        [&](auto i) { batchStrideDs[i] = s.batchStrideC; });
-    const ComputePtrOffsetOfStridedBatch<NumDTensor> compute_ptr_offset_of_batch{s.batchStrideA, s.batchStrideB, batchStrideDs, s.batchStrideC};
+    ck::static_for<0, NumDTensor, 1>{}([&](auto i) { batchStrideDs[i] = s.batchStrideC; });
+    const ComputePtrOffsetOfStridedBatch<NumDTensor> compute_ptr_offset_of_batch{
+        s.batchStrideA, s.batchStrideB, batchStrideDs, s.batchStrideC};
 
     auto batch_count = s.batch_count;
     const ck::index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(ck::get_grid_size() / batch_count);
-    const ck::index_t g_idx = __builtin_amdgcn_readfirstlane(ck::get_block_1d_id() / num_blocks_per_batch);
+    const ck::index_t g_idx =
+        __builtin_amdgcn_readfirstlane(ck::get_block_1d_id() / num_blocks_per_batch);
 
     const ck::long_index_t a_batch_offset = __builtin_amdgcn_readfirstlane(
         static_cast<ck::long_index_t>(compute_ptr_offset_of_batch.GetAPtrOffset(g_idx)));

test/verify/0ck_batched_gemm.cpp

@@ -38,8 +38,8 @@ struct ck_batched_gemm : verify_program<ck_batched_gemm>
         std::size_t n = 3;
         std::size_t k = 3;
         migraphx::shape m1_shape{migraphx::shape::half_type, {b, m, k}};
-        std::vector<float> v1(b*m*k, 1);
-        std::vector<float> v2(b*k*n, 1);//{1, 2, 3, 4, 5, 6, 7, 8};
+        std::vector<float> v1(b * m * k, 1);
+        std::vector<float> v2(b * k * n, 1); //{1, 2, 3, 4, 5, 6, 7, 8};
         // auto l1   = mm->add_parameter("1", m1_shape);
         // auto l2   = mm->add_parameter("2", m1_shape);
         auto l1 = mm->add_literal(migraphx::literal{m1_shape, v1});