Myocardial perfusion MR provides a useful biomarker for diagnosis of cardiac diseases and treatment monitoring. Multiple methods have been proposed, mainly based on first-pass DCE-MRI or FAIR ASL. These methods suffer from low reliability due to low SNR of the MR data, need for ECG triggering and breath-hold acquisition and no agreement on unified data processing. This is even more challenging in small-animal MRI, with lower SNR and much higher cardiac and respiratory rates. We propose an improved ASL method for measuring cardiac perfusion without a contrast agent, which is based on the Radial Tiny Golden Angle Look Locker FAIR ASL acquisition. We improve upon the current methods, which were mainly based on simple Cartesian sampling and ECG triggering, by incorporating sparse radial k-space sampling, retrospectively self-gated acquisition and spatially regularized linear subspace reconstruction. This methodology improves the achievable SNR of the perfusion maps and allows a shorter acquisition time. The method is mainly focused on preclinical small-animal applications, where high cardiac and respiratory rates are present and pose a major challenge. With the proposed method we can estimate single-slice myocardial-perfusion maps, without external gating, in all phases of the cardiac cycle. As a byproduct of this method, myocardial T1 maps are also estimated, which can be used as an additional biomarker. We tested our method in vivo using a preclinical 9.4 T 94/30 Bruker MRI scanner on a rat test subject. We were able to obtain perfusion maps of the whole cardiac cycle with 315 um spatial resolution and 2 mm slice thickness from a 12-minutes acquisition. The acquisition time could possibly be reduced further, by additional optimization of the pulse sequence.