https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82192
Jeffrey A. Law <law at redhat dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |law at redhat dot com
--- Comment #10 from Jeffrey A. Law <law at redhat dot com> ---
Does the oddity that shifts truncate on x86, but bit operations do not come
into play here?
ie,
(lshiftrt:SI (subreg:SI (reg:DI 94 [ a ]) 0)
(subreg:QI (reg:SI 97) 0))
Has a well defined meaning for the hardware (setting aside the RTL issues
momentarily). THe shift count will be truncated in the expected way.
When we substitute into:
(insn 14 10 15 2 (parallel [
(set (reg:HI 102)
(and:HI (subreg:HI (reg:SI 98) 0)
(const_int 8191 [0x1fff])))
(clobber (reg:CC 17 flags))
]) "pr82192-2.c":7 391 {*andhi_1}
(expr_list:REG_DEAD (reg:SI 98)
(expr_list:REG_UNUSED (reg:CC 17 flags)
(nil))))
Generating:
(parallel [
(set (subreg:DI (reg:HI 102) 0)
(zero_extract:DI (reg:DI 94 [ a ])
(const_int 13 [0xd])
(zero_extend:SI (subreg:QI (reg:SI 97) 0))))
(clobber (reg:CC 17 flags))
])
We no longer have the same behavior because we're not emitting a shift, but
instead a bit instruction which have different semantics at the hardware level.
So maybe what's needed is just guarding the transformation on
SHIFT_COUNT_TRUNCATED?
I don't really have time to dig into this today, but wanted to get my initial
thoughts recorded.
