labath added a comment. In https://reviews.llvm.org/D32813#746012, @tberghammer wrote:
> I am a bit confused by the correlation between your change and commit > message. In the commit message you say that 32 byte structs I mean 32-byte vectors. I.e. variables declared as `float foo __attribute__((__vector_size__(32)));` > are passed as x8 pointers but the implementation of > LoadValueFromConsecutiveGPRRegisters seems to read it out from the v0-v8 > registers for vectors of up to 8 elements independently of there size. LoadValueFromConsecutiveGPRRegisters does this for "homogeneous structs", which is a different concept than vector: """Note that for short-vector types the fundamental types are 64-bit vector and 128-bit vector; the type of the elements in the short vector does not form part of the test for homogeneity. """ So an 8-byte and 16-byte vector (and probably structures containing them) are passed in v0..v7 registers. However, a 32-byte vector is not a short-vector type, nor a homogeneous aggregate, so it is passed as a generic struct, via the v8 pointer. Also based on that code I have the suspicion that the first branch (where byte_size <= 16) is not actually used or necessary and also I don't see anything in the ABI documentation indicating otherwise (it would be a pretty crazy ABI if they say that if you have 4 double then passed in a single 32 byte register while if you have 8 double then passed in 8 different 32 byte registers). Can you make sure that branch is necessary (e.g. removing it breaks at least 1 test)? Removing the branch makes the test for 8 and 16-byte vectors fail. https://reviews.llvm.org/D32813 _______________________________________________ lldb-commits mailing list lldb-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits
