Turn on list-ilp scheduling by default on x86 and x86-64, fix up
authorEric Christopher <echristo@apple.com>
Tue, 8 Mar 2011 02:42:25 +0000 (02:42 +0000)
committerEric Christopher <echristo@apple.com>
Tue, 8 Mar 2011 02:42:25 +0000 (02:42 +0000)
commit7c2cdb1c05cd94b494729f5e034aec0562165230
treeae978bf809c0deef18d6faf6459c121c43e18f98
parent0bf56c821c454a960afc34cb5746006bc1f9e1a1
Turn on list-ilp scheduling by default on x86 and x86-64, fix up
testcases accordingly. Some are currently xfailed and will be filed
as bugs to be fixed or understood.

Performance results:

roughly neutral on SPEC
some micro benchmarks in the llvm suite are up between 100 and 150%, only
a pair of regressions that are due to be investigated

john-the-ripper saw:
10% improvement in traditional DES
8% improvement in BSDI DES
59% improvement in FreeBSD MD5
67% improvement in OpenBSD Blowfish
14% improvement in LM DES

Small compile time impact.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@127208 91177308-0d34-0410-b5e6-96231b3b80d8
19 files changed:
lib/Target/X86/X86ISelLowering.cpp
test/CodeGen/X86/2006-05-01-SchedCausingSpills.ll
test/CodeGen/X86/2008-07-11-SpillerBug.ll
test/CodeGen/X86/2008-07-19-movups-spills.ll
test/CodeGen/X86/2008-08-05-SpillerBug.ll
test/CodeGen/X86/2008-08-31-EH_RETURN32.ll
test/CodeGen/X86/2009-04-20-LinearScanOpt.ll
test/CodeGen/X86/2010-05-03-CoalescerSubRegClobber.ll
test/CodeGen/X86/2010-09-17-SideEffectsInChain.ll
test/CodeGen/X86/break-anti-dependencies.ll
test/CodeGen/X86/coalesce-esp.ll
test/CodeGen/X86/commute-two-addr.ll
test/CodeGen/X86/fold-pcmpeqd-0.ll
test/CodeGen/X86/lsr-loop-exit-cond.ll
test/CodeGen/X86/lsr-reuse.ll
test/CodeGen/X86/pr1505b.ll
test/CodeGen/X86/remat-scalar-zero.ll
test/CodeGen/X86/v-binop-widen.ll
test/CodeGen/X86/zext-sext.ll