Two recent regressions, both related to recent tcg changes.

Our CI does not test TCI with --enable-debug-tcg, which given timeout
constraints is probably correct, but in this case resulted in an
infinite loop on aarch64 multiarch/memory.c with FEAT_LSE2 enabled.

r~

Richard Henderson (2):
  tcg/tci: Adjust passing of MemOpIdx
  tcg/tci: Adjust call-clobbered regs for int128_t

 tcg/tci.c                | 30 +++++++++++++-----------------
 tcg/tci/tcg-target.c.inc | 30 +++++++++---------------------
 2 files changed, 22 insertions(+), 38 deletions(-)

-- 
2.34.1

On 6/6/23 22:46, Richard Henderson wrote:
> Two recent regressions, both related to recent tcg changes.
> 
> Our CI does not test TCI with --enable-debug-tcg, which given timeout
> constraints is probably correct, but in this case resulted in an
> infinite loop on aarch64 multiarch/memory.c with FEAT_LSE2 enabled.

To expand on that: with --enable-debug-tcg, assertions fire and catch the bug; without the 
assertions, the generated bytecode is incorrect, which leads to the loop.


r~