fixed a bug in the ckProfiler, and formatted the code

include/ck/utility/data_type.hpp

@@ -20,7 +20,7 @@ struct pk_i4_t
     __host__ __device__ constexpr pk_i4_t() : data{type{}} {}
     __host__ __device__ constexpr pk_i4_t(type init) : data{init} {}
 
-    __host__ __device__ constexpr operator float() const { return static_cast<float>(data); }
+    __host__ __device__ constexpr operator float() const { return static_cast<int8_t>(data); }
 };
 
 inline constexpr auto next_pow2(uint32_t x)

profiler/include/profiler/profile_gemm_b_scale_impl.hpp

@@ -342,7 +342,7 @@ bool profile_gemm_b_scale_impl(int do_verification,
                     if(do_log)
                     {
                         LogRangeAsType<float>(std::cout << "a : ", a_m_k.mData, ",") << std::endl;
-                        LogRangeAsType<float>(std::cout << "b: ", b_k_n.mData, ",") << std::endl;
+                        LogRangeAsType<int8_t>(std::cout << "b: ", b_k_n.mData, ",") << std::endl;
                         LogRangeAsType<float>(
                             std::cout << "c_host  : ", c_m_n_host_result.mData, ",")
                             << std::endl;

profiler/src/profile_gemm_b_scale.cpp

@@ -103,9 +103,9 @@ int profile_gemm_b_scale(int argc, char* argv[])
         printf("n_warmup:%d, n_iter:%d, rotating:%lu\n", n_warmup, n_iter, rotating);
     }
 
-    using F32  = float;
-    using F16  = ck::half_t;
-    using I4   = ck::pk_i4_t;
+    using F32 = float;
+    using F16 = ck::half_t;
+    using I4  = ck::pk_i4_t;
 
     using Row = ck::tensor_layout::gemm::RowMajor;
     using Col = ck::tensor_layout::gemm::ColumnMajor;