llvm dev - Sep 2013 - [LLVMdev] [PATCH] Detect Haswell subarchitecture (i.e. using -march=native)

Hello,

This is my first patch on this list, however I've already submitted several
trough bug tracking system. Since it probably needs some review and it's not
a bug I am submitting it here.

The main intent of this patch is to detect "core-avx2" platform on
Haswell i7 CPUs when running -march=native. Currently it detects it as generic
x86_64.

lib/Support/Host.cpp:
* Haswell is detected for CPUID Family 6 Model 60
* Similar to Ivy and Sandy Bridge we check for AVX2 since some Haswell Pentiums
are SSE4.x only
* I have marked HasAVX2 as "volatile", since otherwise it gets
magically zeroed (by optimizer?) when compiling clang with latest clang build
from trunk

lib/Target/X86/X86Subtarget.cpp:
* Also enabling X86::FeatureFastUAMem for Haswell

Regards,
-- 
Adam Strzelecki | nanoant.com | twitter.com/nanoant
-------------- next part --------------
A non-text attachment was scrubbed...
Name: llvm-detect-haswell.patch
Type: application/octet-stream
Size: 1765 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130912/b6842ec0/attachment.obj>

Hi Adam,
> * I have marked HasAVX2 as "volatile", since otherwise it gets
> magically zeroed (by optimizer?) when compiling clang with latest
> clang build from trunk
That's far more worrying to me than not being able to detect Haswell.
I can't reproduce the problem here at the moment: both debug and
release builds give identical assembly for Host.cpp.

I don't suppose you could post the command-line clang is using to
build Host.cpp (and perhaps the differing object files if you have
them handy)?

Cheers.

Tim.

> That's far more worrying to me than not being able to detect Haswell.
> I can't reproduce the problem here at the moment: both debug and
> release builds give identical assembly for Host.cpp.
OK. I know the reason you cannot reproduce it, before posting the patch I've
decided to check for AVX before checking AVX2, just not to cpuid AVX2 when we
don't have AVX1 anyway.

So the problem exists with following predicate: (1)
  bool HasAVX2 = !GetX86CpuIDAndInfo(0x7, &EAX, &EBX, &ECX,
&EDX) &&
                 (EBX & 0x20);

However it is working absolutely fine if I add "volatile": (2)
  volatile bool HasAVX2 = !GetX86CpuIDAndInfo(0x7, &EAX, &EBX, &ECX,
&EDX) &&
                          (EBX & 0x20);

Or add extra check for HasAVX: (3)
  bool HasAVX2 = HasAVX && !GetX86CpuIDAndInfo(0x7, &EAX, &EBX,
&ECX, &EDX) &&
                 (EBX & 0x20);

Attaching object files related to all of these cases below.

Also attaching patch that removes volatile, but leaves HasAVX check that makes
the code run fine here.

My platform is Ubuntu 12.04 LTS 64-bit & i7-4470k.

llvm[1]: Compiling Host.cpp for Release build
if  clang++ -I/home/ono/Documents/llvm/include
-I/home/ono/Documents/llvm/lib/Support -I/tmp/llvm.q35LGwjHmR/include
-I/tmp/llvm.q35LGwjHmR/lib/Support  -DNDEBUG -D_GNU_SOURCE
-D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -O3
-fomit-frame-pointer -fvisibility-inlines-hidden -fno-exceptions -fno-rtti -fPIC
-Woverloaded-virtual -Wcast-qual    -Wall -W -Wno-unused-parameter
-Wwrite-strings -Wcovered-switch-default -Wno-uninitialized
-Wno-missing-field-initializers -c -MMD -MP -MF
"/tmp/llvm.q35LGwjHmR/lib/Support/Release/Host.d.tmp" -MT
"/tmp/llvm.q35LGwjHmR/lib/Support/Release/Host.o" -MT
"/tmp/llvm.q35LGwjHmR/lib/Support/Release/Host.d"
/home/ono/Documents/llvm/lib/Support/Host.cpp -o
/tmp/llvm.q35LGwjHmR/lib/Support/Release/Host.o ; \
	        then /bin/mv -f
"/tmp/llvm.q35LGwjHmR/lib/Support/Release/Host.d.tmp"
"/tmp/llvm.q35LGwjHmR/lib/Support/Release/Host.d"; else /bin/rm
"/tmp/llvm.q35LGwjHmR/lib/Support/Release/Host.d.tmp"; exit 1; fi

Regards,
-- 
Adam Strzelecki | nanoant.com | twitter.com/nanoant

-------------- next part --------------
A non-text attachment was scrubbed...
Name: Host-objects.tbz2
Type: application/octet-stream
Size: 6366 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130912/6c851eb1/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: llvm-detect-haswell-r2.patch
Type: application/octet-stream
Size: 1720 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20130912/6c851eb1/attachment-0001.obj>