On 19/06/14 21:19, Yuri Gribov wrote: >> Thirdly, we also need to fix movhi_bytes (for pre-v4) thumb2_movhi_insn >> (for thumb2) and, quite possibly, thumb1_movhi_insn (for thumb1). There >> may well be additional changes for movqi variants as well. > > A general question: how should one test ARM backend patches? Is it > enough to regtest ARM and Thumb2 on some modern Cortex? If not - what > other variants should be tested? > > -Y >
The answer to this question is that you have to be realistic. It's clearly impossible to test every permutation, but you should, ideally, test a reasonable selection of permutations that might be affected by a change. Ie, if you make a change that is likely to affect thumb1 code generation, it doesn't make sense to only test the arm or thumb2 code generation (and vice versa). The tricky thing here is that although thumb1 and thumb2 sound quite similar in name, in practice thumb2 code generation is closer to ARM than thumb1. R.
