Hello, The global variable first_pass_instance exists to hack jump threading, see tree-pass.h:
/* Set to true if the pass is called the first time during compilation of the
current function. Note that using this information in the optimization
passes is considered not to be clean, and it should be avoided if possible.
This flag is currently used to prevent loops from being peeled repeatedly
in jump threading; it will be removed once we preserve loop structures
throughout the compilation -- we will be able to mark the affected loops
directly in jump threading, and avoid peeling them next time. */
extern bool first_pass_instance;
AFAIU, GCC now actually does preserve loops, so it should be possible
to remove this hack.
But the recent changes in tree-ssa-reassoc.c introduced a new user of
first_pass_instance (apparently a fat comment discouraging its use is
not enough...).
The attached patch fixes this by splitting pass_reassoc in an early
and a late pass. The pass names are reassoc1 and reassoc2 to avoid
changing a large number of test cases. Admittedly also not ideal, but
better than first_pass_instance.
Richi, could you please have a look at jump threading, and see if
first_pass_instance can be remove?
Bootstrapped&tested on x86_64-unknown-linux-gnu. OK for trunk?
Ciao!
Steven
* tree-ssa-reassoc.c (linearize_expr_tree, undistribute_ops_list,
reassociate_bb, do_reassoc, execute_reassoc): Take extra EARLY_PASS
bool argument to distinguish between the early and late reassociation
pass.
(acceptable_pow_call): Do not look at first_pass_instance.
(execute_early_reassoc, execute_late_reassoc): New functions.
(pass_early_reassoc, pass_late_reassoc): New gimple_opt_pass structs.
(pass_reassoc): Removed.
* tree-pass.h (pass_reassoc): Removed.
(pass_early_reassoc, pass_late_reassoc): Declare.
* passes.c (init_optimization_passes): Schedule pass_early_reassoc and
pass_late_reassoc instead of pass_reassoc.
early_late_reassoc.diff
Description: Binary data
