fixed a bug in the gpu gemm implementation.

src/targets/gpu/gemm.cpp

@@ -342,7 +342,7 @@ argument miopen_gemm::compute(context& ctx,
             bool transb       = args[1].get_shape().transposed();
             rocblas_int lda   = args[0].get_shape().strides()[transa ? dim_1 : dim_0];
             rocblas_int ldb   = args[1].get_shape().strides()[transb ? dim_1 : dim_0];
-            rocblas_int ldc   = args[3].get_shape().strides()[dim_0];
+            rocblas_int ldc   = args[2].get_shape().strides()[dim_0];
             auto out_lens     = output_shape.lens();
             rocblas_int m     = out_lens[dim_0];
             rocblas_int n     = out_lens[dim_1];