[sdcc-devel] The STM8 port, IMO it can be merged soon

[Philipp K. K. <pk...@sp...>]

Dear sdcc developers,

Valentin and me continued work on the stm8 port; for me (on Debian
GNU/Linux x86-64) the port now compiles without warnings, and the list
of regression test failures is down to:

Summary for 'stm8': 48 failures, 6975 tests, 1684 test cases, 2830717
bytes, 18139306 ticks
   Failure: gen/stm8/bug3482146/bug3482146
   Failure: gen/stm8/float_single/float_single_func_ASINF
   Failure: gen/stm8/float_single/float_single_func_ATANF
   Failure: gen/stm8/float_single/float_single_func_FLOORF
   Failure: gen/stm8/float_single/float_single_func_NEG
   Failure: gen/stm8/float_single/float_single_func_SINHF
   Failure:
gen/stm8/gcc-torture-execute-20010224-1/gcc-torture-execute-20010224-1
   Failure:
gen/stm8/gcc-torture-execute-20030714-1/gcc-torture-execute-20030714-1
   Failure:
gen/stm8/gcc-torture-execute-20050826-1/gcc-torture-execute-20050826-1

Please test the port and see if it builds without problems for you, and
if the list of failures looks the same for you. IMO, the port can be me
merged soon; it should already be quite useful in its current state,
since the generated code is mostly correct. I guess some of the failures
might be fixed soon, and if any remain by the time of the merge, we can
just file bug reports for them.

Remaining major shortcomings of the stm8 port:
* Lack of optimization - no peephole optimizer yet, code generation not
very optimized yet.
* No support for large stack, which limits the amount of local variables
we can handle. Regression tests which require lots of local variables
have been disabled for stm8 - these tests are a subset of those disabled
for mcs51-stack-auto.

Philipp