One big problem with such a question is that there is no such thing as "software patents". Software itself is not patentable, it is what the software does or is used to do -- the method -- that is patented. Software may be used to implement the whole invention (many data processing patents) or only one component of the invention (e.g. Diehr.) Many would not even consider Diehr to be a software patent, despite it being considered part of the "patent eligibility trilogy".
There are a number of categories and sub-categories ("art units") that patents are classified as, and you could try to carve out "software patents" based on some of the relevant categories, but that is still not going to be a very rough way of doing so. This is because art units are organized based on the field the invention relates to, rather than how the invention is implemented. Software simply happens to be how inventions could be implemented. As a very inaccurate example, a "security/authentication" art unit could encompass software methods such as the RSA algorithm as well as hardware solutions such as RSA tokens.
As an aside, one of the biggest shortcomings of many early research papers that try to analyze software patents (especially from authors such as Bessen, etc.) is that they are based on very inaccurate classification methods, such as PTO and EPO classifications. As such any conclusions are far from reliable.
So you can guess why the PTO won't even attempt to try to provide such statistics for "software patents": nobody, least of all the Supreme Court, seems to be able to concretely define what a non-abstract software patent is.