Format

example/20_cgemm/cgemm_xdl_bf16.cpp

@@ -278,8 +278,16 @@ int main(int argc, char* argv[])
 
         ref_invoker.Run(ref_argument);
 
-        ck::utils::check_err(c_m_n_real_device_f32_result.mData, c_m_n_real_host_result.mData);
-        ck::utils::check_err(c_m_n_imag_device_f32_result.mData, c_m_n_imag_host_result.mData);
+        ck::utils::check_err(c_m_n_real_device_f32_result.mData,
+                             c_m_n_real_host_result.mData,
+                             "Verification error: incorrect results in real part!",
+                             1e-2f,
+                             1e-3f);
+        ck::utils::check_err(c_m_n_imag_device_f32_result.mData,
+                             c_m_n_imag_host_result.mData,
+                             "Verification error: incorrect results in imaginary part!",
+                             1e-2f,
+                             1e-3f);
     }
 
     return 0;

include/ck/tensor_operation/gpu/device/device_cgemm_4gemm_xdl_cshuffle.hpp

@@ -100,7 +100,7 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
             make_tuple(generate_sequence_v2([&](auto I) { return I; }, Number<2>{})),
             make_tuple(Sequence<0>{}));
 
-        return PadDescriptor_M0_1d(desc_m0, gridSize,blockSize);
+        return PadDescriptor_M0_1d(desc_m0, gridSize, blockSize);
     }
 
     static auto MakeAGridDescriptor_AK0_M_AK1(index_t MRaw, index_t KRaw, index_t StrideA)

include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp

@@ -42,8 +42,8 @@ struct Substract
     {
         dst = src1 - src2;
     }
-    
-       __host__ __device__ constexpr void
+
+    __host__ __device__ constexpr void
     operator()(float& dst, const float& src1, const float& src2) const
     {
         dst = src1 - src2;