[Patch PR78561 PowerPC] Revert to old behaviour for counting constant pools
Segher Boessenkool
segher@kernel.crashing.org
Wed Dec 7 11:56:00 GMT 2016
Hi James,
On Wed, Dec 07, 2016 at 10:31:14AM +0000, James Greenhalgh wrote:
> We always keep the entries in the constant pool after we first see them, and
> we figure out what gets used in the constant pool by sweeping through
> looking for referenced entries. That means we can always answer the
> question "has the constant pool ever had entries?" by checking if there
> is anything at all in there.
>
> So that's what this patch does.
>
> Bootstrapped on gcc110 (powerpc64-none-linux-gnu) configured with
> --enable-languages=all,ada,go,obj-c++ with no issues, Segher has also
> tested it with no issues, and I've tested an aarch64-none-elf cross
> compiler to ensure that the original fix still works on AArch64.
>
> OK?
The rs6000 changes are okay for trunk. The rest I cannot approve (but
it does look fine). Thanks,
Segher
> 2016-12-07 James Greenhalgh <james.greenhalgh@arm.com>
>
> PR rtl-optimization/78561
> * config/rs6000/rs6000.c (rs6000_reg_live_or_pic_offset_p) Use
(Missing colon here)
> constant_pool_empty_p in place of get_pool_size_upper_bound.
> (rs6000_stack_info): Likewise.
> (rs6000_emit_prologue): Likewise.
> (rs6000_elf_declare_function_name): Likewise.
> (rs6000_set_up_by_prologue): Likewise.
> (rs6000_can_eliminate): Likewise.
> * output.h (get_pool_size_upper_bound): Delete.
> (constant_pool_empty_p): New.
> * varasm.c (get_pool_size_upper_bound): Delete
> (constant_pool_empty_p): New.
More information about the Gcc-patches
mailing list