On Wed, Jan 10, 2024 at 07:05:39PM +0000, Andre Vieira (lists) wrote: > This patch is still work in progress, but posting to show failure with > bitint-7 test where handle_stmt called from lower_mergeable_stmt ICE's > because the idx (3) is out of range for the __BitInt(135) with a limb_prec > of 64.
I can reproduce it, will debug it momentarily. > I hacked gcc locally to work around this issue and still have one > outstanding failure, so will look to resolve that failure before posting a > new version. I see 2 issues in your patch: 1) as written earlier, the bitint lowering pass doesn't yet support big endian, so I think at least temporarily we shouldn't pretend it works until it is fixed; https://gcc.gnu.org/pipermail/gcc-patches/2023-August/626847.html mentions the big-endian issues 2) for n in [65, 128] I believe the aarch64 ABI says that _BitInt should be treated the same as __int128; so for those cases we want limb_mode = abi_limb_mode = TImode; those cases are just lowered to TImode arithmetics, there is no attempt to treat those as arrays of smaller limbs. It is just the [129, 65535] cases where we want limb_mode DImode (for efficiency and implementability) and abi_limb_mode TImode (for the ABI reasons, alignment, sizing, ...), i.e. the cases where the ABI says it is treated as an array of the 128-bit limbs --- gcc/config/aarch64/aarch64.cc.jj 2024-01-11 09:37:28.000000000 +0100 +++ gcc/config/aarch64/aarch64.cc 2024-01-11 09:41:03.440202903 +0100 @@ -28279,12 +28279,20 @@ aarch64_excess_precision (enum excess_pr bool aarch64_bitint_type_info (int n, struct bitint_info *info) { + /* For now, the middle-end bitint lowering doesn't handle big endian. */ + if (TARGET_BIG_END) + return false; + if (n <= 8) info->limb_mode = QImode; else if (n <= 16) info->limb_mode = HImode; else if (n <= 32) info->limb_mode = SImode; + else if (n <= 64) + info->limb_mode = DImode; + else if (n <= 128) + info->limb_mode = TImode; else info->limb_mode = DImode; Jakub
