llvm dev - Sep 2010 - [LLVMdev] [PATCH] increase the max number of physical registers

On Dec 8, 2009, at 3:31 PM, Jakob Stoklund Olesen wrote:
> 
> On Dec 8, 2009, at 11:33 AM, Evan Cheng wrote:
> 
>> This caused a massive slow down to post-ra scheduler (llc -O3 on x86,
-O2 on ARM). I'm going to revert it for now until it has been addressed.
> 
> Probably caused by this member:
> 
>    /// KillIndices - The index of the most recent kill (proceding
bottom-up),
>    /// or ~0u if the register is not live.
>    unsigned KillIndices[TargetRegisterInfo::FirstVirtualRegister];
> 
> And this:
> 
>  std::fill(KillIndices, array_endof(KillIndices), ~0u);
> 
> It should probably be dynamically allocated with TRI->getNumRegs()
members instead.
Yep. David Goodwin is going to fix it.

Evan
> 
> /jakob
>

I have the fix ready, and I'll check it in tomorrow morning.

David

On Dec 8, 2009, at 5:34 PM, Evan Cheng wrote:
> 
> On Dec 8, 2009, at 3:31 PM, Jakob Stoklund Olesen wrote:
> 
>> 
>> On Dec 8, 2009, at 11:33 AM, Evan Cheng wrote:
>> 
>>> This caused a massive slow down to post-ra scheduler (llc -O3 on
x86, -O2 on ARM). I'm going to revert it for now until it has been
addressed.
>> 
>> Probably caused by this member:
>> 
>>   /// KillIndices - The index of the most recent kill (proceding
bottom-up),
>>   /// or ~0u if the register is not live.
>>   unsigned KillIndices[TargetRegisterInfo::FirstVirtualRegister];
>> 
>> And this:
>> 
>> std::fill(KillIndices, array_endof(KillIndices), ~0u);
>> 
>> It should probably be dynamically allocated with TRI->getNumRegs()
members instead.
> 
> Yep. David Goodwin is going to fix it.
> 
> Evan
> 
>> 
>> /jakob
>> 
>

Sending        lib/CodeGen/AggressiveAntiDepBreaker.cpp
Sending        lib/CodeGen/AggressiveAntiDepBreaker.h
Sending        lib/CodeGen/CriticalAntiDepBreaker.cpp
Sending        lib/CodeGen/PostRASchedulerList.cpp
Transmitting file data ....
Committed revision 90970.

On Dec 8, 2009, at 5:34 PM, Evan Cheng wrote:
> 
> On Dec 8, 2009, at 3:31 PM, Jakob Stoklund Olesen wrote:
> 
>> 
>> On Dec 8, 2009, at 11:33 AM, Evan Cheng wrote:
>> 
>>> This caused a massive slow down to post-ra scheduler (llc -O3 on
x86, -O2 on ARM). I'm going to revert it for now until it has been
addressed.
>> 
>> Probably caused by this member:
>> 
>>   /// KillIndices - The index of the most recent kill (proceding
bottom-up),
>>   /// or ~0u if the register is not live.
>>   unsigned KillIndices[TargetRegisterInfo::FirstVirtualRegister];
>> 
>> And this:
>> 
>> std::fill(KillIndices, array_endof(KillIndices), ~0u);
>> 
>> It should probably be dynamically allocated with TRI->getNumRegs()
members instead.
> 
> Yep. David Goodwin is going to fix it.
> 
> Evan
> 
>> 
>> /jakob
>> 
>

Hello,

This patch was reverted and the revert was forgotten to be undone
after the performance regression it introduced was fixed.

Can someone please revert it back (i.e. increase the max physreg size
to 16K or even better to 32K) to enable us to experiment with large
register number machines again? :)

It was this trivial patch:

Index: include/llvm/Target/TargetRegisterInfo.h
==================================================================---
include/llvm/Target/TargetRegisterInfo.h    (revision 89468)
+++ include/llvm/Target/TargetRegisterInfo.h    (working copy)
@@ -299,8 +299,8 @@
      /// FirstVirtualRegister - This is the first register number that is
      /// considered to be a 'virtual' register, which is part of the
SSA
      /// namespace.  This must be the same for all targets, which means that
each
-    /// target is limited to 1024 registers.
-    FirstVirtualRegister = 1024
+    /// target is limited to this fixed number of registers.
+    FirstVirtualRegister = 16384
    };

    /// isPhysicalRegister - Return true if the specified register number is in



On 12/09/2009 03:34 AM, Evan Cheng wrote:
> On Dec 8, 2009, at 3:31 PM, Jakob Stoklund Olesen wrote:
> 
>> On Dec 8, 2009, at 11:33 AM, Evan Cheng wrote:
>>
>>> This caused a massive slow down to post-ra scheduler (llc -O3 on
x86, -O2 on ARM). I'm going to revert it for now until it has been
addressed.
>> Probably caused by this member:
>>
>>    /// KillIndices - The index of the most recent kill (proceding
bottom-up),
>>    /// or ~0u if the register is not live.
>>    unsigned KillIndices[TargetRegisterInfo::FirstVirtualRegister];
>>
>> And this:
>>
>>  std::fill(KillIndices, array_endof(KillIndices), ~0u);
>>
>> It should probably be dynamically allocated with TRI->getNumRegs()
members instead.
> 
> Yep. David Goodwin is going to fix it.
> 
> Evan
> 
>> /jakob
>>
> 
> 
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
> 

-- 
Pekka

Hello,

I haven't yet received any input what this patch breaks (some
LLVM-external project?) and the branch creation date is tomorrow.

Can someone please tell what we can do to make this patch merged
in for 2.8?

Thanks!

On 07/06/2010 01:26 PM, Pekka Jääskeläinen wrote:
> Hello,
> 
> This patch was reverted and the revert was forgotten to be undone
> after the performance regression it introduced was fixed.
> 
> Can someone please revert it back (i.e. increase the max physreg size
> to 16K or even better to 32K) to enable us to experiment with large
> register number machines again? :)
> 
> It was this trivial patch:
> 
> Index: include/llvm/Target/TargetRegisterInfo.h
> ==================================================================> ---
include/llvm/Target/TargetRegisterInfo.h    (revision 89468)
> +++ include/llvm/Target/TargetRegisterInfo.h    (working copy)
> @@ -299,8 +299,8 @@
>       /// FirstVirtualRegister - This is the first register number that is
>       /// considered to be a 'virtual' register, which is part of
the SSA
>       /// namespace.  This must be the same for all targets, which means
that each
> -    /// target is limited to 1024 registers.
> -    FirstVirtualRegister = 1024
> +    /// target is limited to this fixed number of registers.
> +    FirstVirtualRegister = 16384
>     };
> 
>     /// isPhysicalRegister - Return true if the specified register number
is in
> 
> 
> 
> On 12/09/2009 03:34 AM, Evan Cheng wrote:
>> On Dec 8, 2009, at 3:31 PM, Jakob Stoklund Olesen wrote:
>>
>>> On Dec 8, 2009, at 11:33 AM, Evan Cheng wrote:
>>>
>>>> This caused a massive slow down to post-ra scheduler (llc -O3
on x86, -O2 on ARM). I'm going to revert it for now until it has been
addressed.
>>> Probably caused by this member:
>>>
>>>    /// KillIndices - The index of the most recent kill (proceding
bottom-up),
>>>    /// or ~0u if the register is not live.
>>>    unsigned KillIndices[TargetRegisterInfo::FirstVirtualRegister];
>>>
>>> And this:
>>>
>>>  std::fill(KillIndices, array_endof(KillIndices), ~0u);
>>>
>>> It should probably be dynamically allocated with
TRI->getNumRegs() members instead.
>> Yep. David Goodwin is going to fix it.
>>
>> Evan
>>
>>> /jakob
>>>
>>
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>>
> 
> 

-- 
Pekka

hi Jääskeläinen,

i am very interesting in the way you write the RegisterInfo.td for the
machine with such a lot registers, or it is generated by some others
program?

best regards
ether

2010/7/6 Pekka Jääskeläinen <pekka.jaaskelainen at
tut.fi>:
> Hello,
>
> This patch was reverted and the revert was forgotten to be undone
> after the performance regression it introduced was fixed.
>
> Can someone please revert it back (i.e. increase the max physreg size
> to 16K or even better to 32K) to enable us to experiment with large
> register number machines again? :)
>
> It was this trivial patch:
>
> Index: include/llvm/Target/TargetRegisterInfo.h
> ==================================================================> ---
include/llvm/Target/TargetRegisterInfo.h    (revision 89468)
> +++ include/llvm/Target/TargetRegisterInfo.h    (working copy)
> @@ -299,8 +299,8 @@
>      /// FirstVirtualRegister - This is the first register number that is
>      /// considered to be a 'virtual' register, which is part of
the SSA
>      /// namespace.  This must be the same for all targets, which means
that each
> -    /// target is limited to 1024 registers.
> -    FirstVirtualRegister = 1024
> +    /// target is limited to this fixed number of registers.
> +    FirstVirtualRegister = 16384
>    };
>
>    /// isPhysicalRegister - Return true if the specified register number is
in
>
>
>
> On 12/09/2009 03:34 AM, Evan Cheng wrote:
>> On Dec 8, 2009, at 3:31 PM, Jakob Stoklund Olesen wrote:
>>
>>> On Dec 8, 2009, at 11:33 AM, Evan Cheng wrote:
>>>
>>>> This caused a massive slow down to post-ra scheduler (llc -O3
on x86, -O2 on ARM). I'm going to revert it for now until it has been
addressed.
>>> Probably caused by this member:
>>>
>>>    /// KillIndices - The index of the most recent kill (proceding
bottom-up),
>>>    /// or ~0u if the register is not live.
>>>    unsigned KillIndices[TargetRegisterInfo::FirstVirtualRegister];
>>>
>>> And this:
>>>
>>>  std::fill(KillIndices, array_endof(KillIndices), ~0u);
>>>
>>> It should probably be dynamically allocated with
TRI->getNumRegs() members instead.
>>
>> Yep. David Goodwin is going to fix it.
>>
>> Evan
>>
>>> /jakob
>>>
>>
>>
>> _______________________________________________
>> LLVM Developers mailing list
>> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>>
>
>
> --
> Pekka
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>