fix 128bits ctlz intrinsincs UB

[dvermd]

Fixes #604

This is the GIMPLE code generated

      if (param0 == 0) goto then; else goto else;
      then:
      zeros = 128;
      goto after;
      else:
      _2 = param0 >> 64;
      _3 = (size_t) _2;
      if (_3 != 0) goto ctlz_then; else goto ctlz_else;
      after:
      stack_var_1.15_4 = &stack_var_1;
      MEM[(unsigned int *)stack_var_1.15_4] = zeros;
      stack_var_1.16_5 = &stack_var_1;
      loadedValue2 = MEM[(unsigned int *)stack_var_1.16_5];
      D.4444 = loadedValue2;
      return D.4444;
      ctlz_then:
      _6 = param0 >> 64;
      _7 = (size_t) _6;
      _8 = __builtin_clzll (_7);
      zeros = (unsigned int) _8;
      goto ctlz_after;
      ctlz_else:
      _9 = (size_t) param0;
      _10 = __builtin_clzll (_9);
      _11 = (unsigned int) _10;
      zeros = _11 + 64;
      goto ctlz_after;
      ctlz_after:
      zeros = zeros;
      goto after;

[dvermd]

It looks like the some CI jobs are failing on

thread 'rustc' panicked at src/int.rs:176:13:
assertion failed: a_type.dyncast_array().is_some()

which is some code I didn't touch.

Can my changes have some side effects ?
I see that #631 also have these jobs failing. Perhaps it's not my changes which bring these failures.

Any hints ?

[antoyo]

Not sure it's related. I opened a new dummy PR to check if the tests pass, though.

[antoyo]

The CI passed, so the problem is in here.

I suspect this is because you changed the return type: it used to return an array type and your change makes it return a u32.

[dvermd]

The return type of the if width == 128 was already a u32 because the block ended with

                let res = self.context.new_array_access(self.location, result, index);

                return self.gcc_int_cast(res.to_rvalue(), result_type);

which will get an element in the array and cast it as result_type which is u32.

Edit: formatting

[dvermd]

Ok, finally found the line pointing in my change let leading_zeroes = self.add(low_leading_zeroes, sixty_four); where low_leading_zeroesand sixty_four are of different types.

The problem I see is that the if branch should have been taken.

[antoyo]

Sorry for the delay: I've been very busy.

[antoyo]

It seems your new algorithm significantly diverges from the one linked here.
If so, could you please replace this link with another pointing to the algorithm you used? If you can't, could you please add comments below to make this easier to understand?