Benjamin Kramer 59957500f9 CodeGenPrepare: Add a transform to turn selects into branches in some cases. ... This came up when a change in block placement formed a cmov and slowed down a hot loop by 50%: ucomisd (%rdi), %xmm0 cmovbel %edx, %esi cmov is a really bad choice in this context because it doesn't get branch prediction. If we emit it as a branch, an out-of-order CPU can do a better job (if the branch is predicted right) and avoid waiting for the slow load+compare instruction to finish. Of course it won't help if the branch is unpredictable, but those are really rare in practice. This patch uses a dumb conservative heuristic, it turns all cmovs that have one use and a direct memory operand into branches. cmovs usually save some code size, so we disable the transform in -Os mode. In-Order architectures are unlikely to benefit as well, those are included in the "predictableSelectIsExpensive" flag. It would be better to reuse branch probability info here, but BPI doesn't support select instructions currently. It would make sense to use the same heuristics as the if-converter pass, which does the opposite direction of this transform. Test suite shows a small improvement here and there on corei7-level machines, but the actual results depend a lot on the used microarchitecture. The transformation is currently disabled by default and available by passing the -enable-cgp-select2branch flag to the code generator. Thanks to Chandler for the initial test case to him and Evan Cheng for providing me with comments and test-suite numbers that were more stable than mine :) git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@156234 91177308-0d34-0410-b5e6-96231b3b80d8		2012-05-05 12:49:22 +00:00
..
ADCE.cpp	The landingpad instruction isn't dead simply because it's value isn't used.	2011-08-19 21:52:06 +00:00
BasicBlockPlacement.cpp	Get rid of static constructors for pass registration. Instead, every pass exposes an initializeMyPassFunction(), which	2010-10-19 17:21:58 +00:00
CMakeLists.txt	build/CMake: Finish removal of add_llvm_library_dependencies.	2011-11-29 19:25:30 +00:00
CodeGenPrepare.cpp	CodeGenPrepare: Add a transform to turn selects into branches in some cases.	2012-05-05 12:49:22 +00:00
ConstantProp.cpp	Last bit of TargetLibraryInfo propagation. Also fixed a case for TargetData	2011-12-01 21:29:16 +00:00
CorrelatedValuePropagation.cpp	llvm::SwitchInst	2012-03-11 06:09:17 +00:00
DCE.cpp	Remove unused STL header includes.	2011-04-23 19:53:52 +00:00
DeadStoreElimination.cpp	Reverted r152620 - DSE: Shorten memset when a later store overwrites the start of it. There were all sorts of buildbot issues	2012-02-28 05:06:24 +00:00
EarlyCSE.cpp	bz11794 : EarlyCSE stack overflow on long functions.	2012-01-31 23:14:41 +00:00
GlobalMerge.cpp	Fix various issues (or do cleanups) found by enabling certain MSVC warnings.	2012-02-13 06:30:56 +00:00
GVN.cpp	Change recurse depth limit to uint32 to fix warning.	2012-04-27 19:30:32 +00:00
IndVarSimplify.cpp	More IndVarSimplify cleanup.	2012-03-24 00:51:17 +00:00
JumpThreading.cpp	enhance jump threading to preserve TBAA information when PRE'ing loads,	2012-03-13 18:07:41 +00:00
LICM.cpp	An instruction in a loop is not guaranteed to be executed just because the loop	2012-05-01 04:03:01 +00:00
LLVMBuild.txt	LLVMBuild: Remove trailing newline, which irked me.	2011-12-12 19:48:00 +00:00
LoopDeletion.cpp	Don't use PN->replaceUsesOfWith() to change a PHINode's incoming blocks,	2011-06-21 10:02:43 +00:00
LoopIdiomRecognize.cpp	Correct grammar.	2011-09-13 00:44:16 +00:00
LoopInstSimplify.cpp	Propagate TargetLibraryInfo throughout ConstantFolding.cpp and	2011-12-01 03:08:23 +00:00
LoopRotation.cpp	loop-rotate shouldn't hoist alloca instructions out of a loop. Patch by Patrik Hägglund, with slightly modified test. Issue reported by Patrik Hägglund on llvmdev.	2012-02-16 00:41:10 +00:00
LoopStrengthReduce.cpp	Make sure HoistInsertPosition finds a position that is dominated by all	2012-04-30 03:53:06 +00:00
LoopUnrollPass.cpp	LoopUnrollPass: Use variable "Threshold" instead of "CurrentThreshold" when	2012-04-04 11:44:08 +00:00
LoopUnswitch.cpp	Second attempt at PR12573:	2012-04-30 10:44:54 +00:00
LowerAtomic.cpp	Remove the old atomic instrinsics. autoupgrade functionality is included with this patch.	2011-10-06 23:20:49 +00:00
Makefile
MemCpyOptimizer.cpp	Fix various issues (or do cleanups) found by enabling certain MSVC warnings.	2012-02-13 06:30:56 +00:00
ObjCARC.cpp	Miscellaneous accumulated cleanups.	2012-04-27 18:56:31 +00:00
Reassociate.cpp	Add 'landingpad' instructions to the list of instructions to ignore.	2012-05-04 04:22:32 +00:00
Reg2Mem.cpp	Fix a ton of comment typos found by codespell. Patch by	2011-04-15 05:18:47 +00:00
Scalar.cpp	Add a new ObjC ARC optimization pass to eliminate unneeded	2012-01-17 20:52:24 +00:00
ScalarReplAggregates.cpp	Typo.	2012-04-11 19:21:58 +00:00
SCCP.cpp	Fix 80-column violation.	2012-03-28 00:35:33 +00:00
SimplifyCFGPass.cpp	Convert CallInst and InvokeInst APIs to use ArrayRef.	2011-07-15 08:37:34 +00:00
SimplifyLibCalls.cpp	fix pr12559: mark unavailable win32 math libcalls	2012-04-17 23:05:54 +00:00
Sink.cpp	Move Instruction::isSafeToSpeculativelyExecute out of VMCore and	2011-12-14 23:49:11 +00:00
TailRecursionElimination.cpp	A pile of long over-due refactorings here. There are some very, *very*	2012-05-04 00:58:03 +00:00