llvm dev - Jul 2016 - HEAD compilation causes gcc internal error

After worked around the problem in SimplifyCFG.cpp (calling isCast()
instead of comparing opcode), I hit another gcc crash for
FunctionImport.cpp line 480, which I have no idea what's wrong with the
code. "Luckily", I found gcc4.8.2 and gave it a try, both crashes are
gone.

*New problem though:*
/llvm-clang-trunk/src/tools/clang/lib/CodeGen/CGBuiltin.cpp: In member
function ?llvm::Value*
clang::CodeGen::CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned int, const
clang::CallExpr*)?:
/llvm-clang-trunk/src/tools/clang/lib/CodeGen/CGBuiltin.cpp:7283:39:
error: *?r600_rsq?
is not a member of ?llvm::Intrinsic?*
     return emitUnaryBuiltin(*this, E, Intrinsic::r600_rsq);
                                       ^
/llvm-clang-trunk/src/tools/clang/lib/CodeGen/CGBuiltin.cpp:7289:39:
error: *?AMDGPU_ldexp?
is not a member of ?llvm::Intrinsic?*
     return emitFPIntBuiltin(*this, E, Intrinsic::AMDGPU_ldexp);
                                       ^

Actually, if I didn't configure to target AMDGPU, why bothering compiling
these code?

On Fri, Jul 22, 2016 at 11:52 AM, Reid Kleckner <rnk at google.com> wrote:
> We can definitely accept a non-invasive patch to work around the issue.
> GCC 4.8 is definitely inside our toolchain support window.
>
> On Fri, Jul 22, 2016 at 2:51 PM, Welson Sun via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
>> Sure this is more likely a gcc bug. However, same toolchain compiled
>> without any problems a week ago. Also, in some organizations, upgrading
gcc
>> is very hard if not impossible.
>>
>> On Fri, Jul 22, 2016 at 11:11 AM, Paulo Matos via llvm-dev <
>> llvm-dev at lists.llvm.org> wrote:
>>
>>>
>>>
>>> On 22/07/16 20:08, Welson Sun via llvm-dev wrote:
>>> > This is gcc4.8.0 compiling HEAD synced on July 22:
>>> >
>>>
>>> My guess is that you're better off posting this to the GCC
mailing list.
>>> It's not LLVM's fault GCC crashed on this. On the other
hand, you might
>>> want to try reproducing this with a more recent compiler before
>>> announcing the issue on the GCC side.
>>>
>>> --
>>> Paulo Matos
>>> _______________________________________________
>>> LLVM Developers mailing list
>>> llvm-dev at lists.llvm.org
>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>>
>>
>>
>>
>> --
>> - Welson
>>
>>
>> _______________________________________________
>> LLVM Developers mailing list
>> llvm-dev at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>
>>
>

-- 
- Welson
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160722/419d9165/attachment-0001.html>

> On Jul 22, 2016, at 14:18, Welson Sun via llvm-dev <llvm-dev at
lists.llvm.org> wrote:
> 
> After worked around the problem in SimplifyCFG.cpp (calling isCast()
instead of comparing opcode), I hit another gcc crash for FunctionImport.cpp
line 480, which I have no idea what's wrong with the code.
"Luckily", I found gcc4.8.2 and gave it a try, both crashes are gone.
> 
> New problem though:
> /llvm-clang-trunk/src/tools/clang/lib/CodeGen/CGBuiltin.cpp: In member
function ?llvm::Value*
clang::CodeGen::CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned int, const
clang::CallExpr*)?:
> /llvm-clang-trunk/src/tools/clang/lib/CodeGen/CGBuiltin.cpp:7283:39: error:
?r600_rsq? is not a member of ?llvm::Intrinsic?
>      return emitUnaryBuiltin(*this, E, Intrinsic::r600_rsq);
>                                        ^
> /llvm-clang-trunk/src/tools/clang/lib/CodeGen/CGBuiltin.cpp:7289:39: error:
?AMDGPU_ldexp? is not a member of ?llvm::Intrinsic?
>      return emitFPIntBuiltin(*this, E, Intrinsic::AMDGPU_ldexp);
>                                        ^
> 
> Actually, if I didn't configure to target AMDGPU, why bothering
compiling these code?
Your clang is out of date. Intrinsics for all targets for now are included in
every build
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160722/bf5eccf1/attachment.html>

Ah, thanks Matt! Forgot that.


On Fri, Jul 22, 2016 at 2:22 PM, Matt Arsenault <arsenm2 at gmail.com>
wrote:
>
> On Jul 22, 2016, at 14:18, Welson Sun via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
> After worked around the problem in SimplifyCFG.cpp (calling isCast()
> instead of comparing opcode), I hit another gcc crash for
> FunctionImport.cpp line 480, which I have no idea what's wrong with the
> code. "Luckily", I found gcc4.8.2 and gave it a try, both crashes
are gone.
>
> *New problem though:*
> /llvm-clang-trunk/src/tools/clang/lib/CodeGen/CGBuiltin.cpp: In member
> function ?llvm::Value*
> clang::CodeGen::CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned int, const
> clang::CallExpr*)?:
> /llvm-clang-trunk/src/tools/clang/lib/CodeGen/CGBuiltin.cpp:7283:39: error:
>  *?r600_rsq? is not a member of ?llvm::Intrinsic?*
>      return emitUnaryBuiltin(*this, E, Intrinsic::r600_rsq);
>                                        ^
> /llvm-clang-trunk/src/tools/clang/lib/CodeGen/CGBuiltin.cpp:7289:39: error:
>  *?AMDGPU_ldexp? is not a member of ?llvm::Intrinsic?*
>      return emitFPIntBuiltin(*this, E, Intrinsic::AMDGPU_ldexp);
>                                        ^
>
> Actually, if I didn't configure to target AMDGPU, why bothering
compiling
> these code?
>
>
> Your clang is out of date. Intrinsics for all targets for now are included
> in every build
>


-- 
- Welson
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20160722/f57cddfa/attachment.html>