On 06/21/2018 11:44 AM, Vladimir Makarov wrote:
>
>
> On 06/13/2018 02:58 PM, Dimitar Dimitrov wrote:
>> From: Dimitar Dimitrov <dddimit...@mm-sol.com>
>>
>> For some targets, Pmode != UNITS_PER_WORD. Take this into account
>> when marking hard registers as being used.
>>
>> I tested C and C++ testsuits for x86_64 with and without this
>> patch. There was no regression, i.e. gcc.sum and g++.sum matched
>> exactly.
>>
>> gcc/ChangeLog:
>>
>> 2018-06-13 Dimitar Dimitrov <dimi...@dinux.eu>
>>
>> * lra-eliminations.c (set_ptr_hard_reg_bits): New function.
>> (update_reg_eliminate): Mark all spanning hw registers.
>>
>> gcc/testsuite/ChangeLog:
>>
>> 2018-06-13 Dimitar Dimitrov <dimi...@dinux.eu>
>>
>> * gcc.target/pru/lra-framepointer-fragmentation-1.c: New test.
>> * gcc.target/pru/lra-framepointer-fragmentation-2.c: New test.
>>
>> Cc: Vladimir Makarov <vmaka...@redhat.com>
>> Cc: Peter Bergner <berg...@vnet.ibm.com>
>> Cc: Kenneth Zadeck <zad...@naturalbridge.com>
>> Cc: Seongbae Park <seongbae.p...@gmail.com>
>> Signed-off-by: Dimitar Dimitrov <dddimit...@mm-sol.com>
>> ---
>> gcc/lra-eliminations.c | 14 ++++-
>> .../pru/lra-framepointer-fragmentation-1.c | 33 ++++++++++++
>> .../pru/lra-framepointer-fragmentation-2.c | 61
>> ++++++++++++++++++++++
>> 3 files changed, 106 insertions(+), 2 deletions(-)
>> create mode 100644
>> gcc/testsuite/gcc.target/pru/lra-framepointer-fragmentation-1.c
>> create mode 100644
>> gcc/testsuite/gcc.target/pru/lra-framepointer-fragmentation-2.c
>>
>> diff --git a/gcc/lra-eliminations.c b/gcc/lra-eliminations.c
>> index 21d8d5f8018..566cc2c8248 100644
>> --- a/gcc/lra-eliminations.c
>> +++ b/gcc/lra-eliminations.c
>> @@ -1180,6 +1180,16 @@ spill_pseudos (HARD_REG_SET set)
>> bitmap_clear (&to_process);
>> }
>> +static void set_ptr_hard_reg_bits (HARD_REG_SET *hard_reg_set, int r)
>> +{
>> + int w;
>> +
>> + for (w = 0; w < GET_MODE_SIZE (Pmode); w += UNITS_PER_WORD, r++)
>> + {
>> + SET_HARD_REG_BIT (*hard_reg_set, r);
>> + }
>> +}
>> +
> The patch itself is ok but for uniformity I'd use
>
> for (int i = hard_regno_nregs (r, Pmode) - 1; i >= 0; i--)
> SET_HARD_REG_BIT (*hard_reg_set, r + i);
I'm a bit surprised we don't already have a utility function to do this.
Hmmm
add_to_hard_reg_set (hard_reg_set, Pmode, r)
So instead LRA ought to be using that function in the places where calls
to set_ptr_hard_reg_bits were introduced.
Dimitar, can you verify that change works?
Jeff
