https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55212
--- Comment #343 from Kazumoto Kojima <kkojima at gcc dot gnu.org> ---
(In reply to Oleg Endo from comment #342)
> (In reply to John Paul Adrian Glaubitz from comment #339)
> > /home/glaubitz/webkit2gtk-sh4-new-new/webkit2gtk-2.46.0/build-soup3/
> > JavaScriptCore/DerivedSources/unified-sources/UnifiedSource-f2e18ffc-36.cpp
> > (...)
> > {standard input}: Assembler messages:
> > {standard input}:6139: Error: syntax error in @(disp,[Rn, gbr, pc])
> > {standard input}:6139: Error: invalid operands for opcode
> > {standard input}:54249: Error: syntax error in @(disp,[Rn, gbr, pc])
> > {standard input}:54249: Error: invalid operands for opcode
> > {standard input}:59575: Error: syntax error in @(disp,[Rn, gbr, pc])
> > {standard input}:59575: Error: invalid operands for opcode
> > {standard input}:61031: Error: syntax error in @(disp,[Rn, gbr, pc])
> > {standard input}:61031: Error: invalid operands for opcode
> >
>
> I was able to reproduce it sh-elf copmiler by compiling the attached .ii
> file with the following options:
>
> sh-elf-gcc -c -m4 -ml -matomic-model=soft-gusa -Wextra -Wall -pipe
> -fmax-errors=20 -Wno-odr -Wno-stringop-overread -Wno-stringop-overflow
> -Wno-nonnull -Wno-array-bounds -Wno-expansion-to-defined -Wno-noexcept-type
> -Wno-psabi -Wno-misleading-indentation -Wno-maybe-uninitialized -Wundef
> -Wpointer-arith -Wmissing-format-attribute -Wformat-security -Wcast-align
> -Wno-tautological-compare -fasynchronous-unwind-tables -fdebug-types-section
> -g1 -O2 fstack-protector-strong -fPIE -Wformat -Werror=format-security
> -fno-strict-aliasing -fno-exceptions -fno-rtti -fcoroutines
> -ffunction-sections -fdata-sections -fPIC -fvisibility=hidden
> -ffp-contract=off -std=c++23 -save-temps -dp test.ii
>
>
> The problematic insn is:
> flds fr1,fpul ! 1297 [c=4 l=2] movsi_ie/21
> mov.b @(5,fpul),r0 ! 517 [c=2 l=2] *movqi/8 <<<<<<<<<<
>
>
> now that looks like quite some adventurous register usage.
>
> This should actually be rejected by the 'sh_legitimate_address_p' function.
> I guess that a check is missing somewhere in the code that does the
> transformation?
Yes. The wrong instruction
mov.b @(5,fpul),r0 ! 517 [c=2 l=2] *movqi/8
is generated with *movqi insn which is defined by
(define_insn "*mov<mode>"
[(set (match_operand:QIHI 0 "general_movdst_operand"
"=r,r,r,Sid,^zr,Ssd,r, Sdd,z, r,l")
(match_operand:QIHI 1 "general_movsrc_operand"
"Q,r,i,^zr,Sid,r, Ssd,z, Sdd,l,r"))]
"TARGET_SH1
&& (arith_reg_operand (operands[0], <MODE>mode)
|| arith_reg_operand (operands[1], <MODE>mode))"
and the 8-th alternative is an Sdd memory load. general_mov{dst,src}_operand
predicates check the memory displacement expression. They only check the
address register with REG_P, though.
