This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Fix -mv8plus (was Re: c/838: sparc-solaris2.7 extra testsuite failures when RTL checking)
On Thu, Nov 23, 2000 at 05:53:23AM -0500, Jakub Jelinek wrote:
> BTW: Long term, I think we should allocate additional 16 call-clobbered
> registers which would not be used in normal register allocation and split
> the V8PLUS patterns after reload into some parallels like:
>
> (insn 25 23 29 (parallel[
> (set (reg:SI 3 %g3)
> (const_int 0)
> (set (reg:DI 102 %dg3)
> (ashift:DI (reg:SI 24 %i0)
> (const_int 32 [0x20])))
> ] ) 302 {*ashldi3_v8plus_insn} (nil)
> (nil)) # sllx %i0, 32, %g3
Possibly. Not pretty, but the problem is certainly an evil one.
> * config/sparc/sparc.md (muldi3_v8plus): Remove H constraint.
> Handle CONST_INT as second argument.
> * config/sparc/sparc.c (set_extends): Remove first argument.
> Properly handle AND, CONST_INT and CONST_DOUBLE, handle IOR, XOR and
> MOV.
> (sparc_check_64): Abort if first argument is not REG.
> If it is DImode REG, look at the lower register of the register
> pair.
>
> * config/sparc/sparc.c (load_pic_register, restore_regs,
> output_return, sparc_v8plus_shift, sparc_function_profiler,
> sparc_function_block_profiler, sparc_block_profiler): Fix output
> formatting.
>
> * gcc.dg/ultrasp3.c: New test.
Ok.
r~