This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Instruction scheduling for the R5900's 2 integer pipelines


Hi,

I have somehow gotten the problem sorted out. I rolled back the design changes that I made to the divmod stuff, and simply changed the divmod expander to take in the lo register as an operand.
The expander can now tell which lo register was allocated, and everything seems to be working fine for the r5900 and r3000.

Thank you for your time and attention.

Regards,
-W Y


On Saturday, February 6, 2016 12:05 PM, Woon yung Liu <ysai187@yahoo.com> wrote:
Sorry, but I just realized what I had to go with the 3rd option. Other than the difficulties that I described earlier, it's mainly because the divmod split insn will output to a lo register.

When I allocate a hi+lo temporary register, that lo register must be a subset of it.


On Friday, February 5, 2016 11:56 PM, Jeff Law <law@redhat.com> wrote:
On 02/05/2016 05:35 AM, Woon yung Liu wrote:

> The current (GCC 5.3.0) MIPS divmod<mode>4 pattern emits an expand
> that allocates a temporary register (hi+lo) and emits other
> instructions, depending if whether the target is a 32-bit or 64-bit
> MIPS target.
>
>
> However, it uses gen_rtx_REG to allocate the hi+lo register, while I
> need it to freely allocate either one of the two (hi+lo or hi1+lo1)
> that the r5900 has. With the current design, it will only use hi+lo
> in all division ops (as with other MIPS).
>
>
> I think that that I should be using gen_reg_rtx to allocate a pseudo
> register, but that isn't possible in that expand (will result in an
> assertion failure, due to can_create_pseudo_p).
Conditionalize the paths based on can_create_pseudo_p.

Essentially up until the register reloading pass you can freely create 
new pseudos.  Once register reloading has begun, you can't create new 
pseudos.  Though you might also look into why something would be calling 
the divmod expander after reload started -- that seems rather odd.

Jeff


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]