[PATCH] D13363: [DeadStoreElimination] Add support for non-local DSE

[Ivan Baev via llvm-commits]

ivanbaev added a comment.

Hi Bruno,

Here are LNT compile-time:  ibaev-linux__clang_DEV__aarch64 test results
Previous: aarch64  compiler with this patch, non-local DSE
Current: aarch64 compiler without this patch, that is top-of-trunk DSE

Tests Summary
Status Group 	#
Performance Regressions 	141
Performance Improvements 	134
Added Tests 	1
Existing Failures 	489
Unchanged Tests 	1191
Total Tests 	1956

- Top regressions with Current (top-of-trunk DSE) ***

Performance Regressions - Compile Time 	Δ 	Previous 	Current 	σ
SingleSource/UnitTests/2002-08-02-CastTest 	400.00% 	0.0080 	0.0400 	-
SingleSource/UnitTests/C++11/stdthreadbug 	300.00% 	0.0040 	0.0160 	-
SingleSource/UnitTests/2003-05-31-LongShifts 	250.00% 	0.0080 	0.0280 	-
SingleSource/UnitTests/2008-04-20-LoopBug2 	200.00% 	0.0120 	0.0360 	-
SingleSource/Regression/C/2004-02-03-AggregateCopy 	150.00% 	0.0160 	0.0400 	-
MultiSource/Benchmarks/Prolangs-C++/vcirc/vcirc 	100.00% 	0.0240 	0.0480 	-
SingleSource/Benchmarks/Misc/mandel-2 	100.00% 	0.0360 	0.0720 	-
SingleSource/UnitTests/2003-04-22-Switch 	100.00% 	0.0200 	0.0400 	-
SingleSource/UnitTests/Vector/multiplies 	100.00% 	0.0200 	0.0400 	-
SingleSource/UnitTests/block-byref-test 	100.00% 	0.0120 	0.0240 	-
....

- Top improvents with Current (top-of-trunk DSE) ***

Performance Improvements - Compile Time 	Δ 	Previous 	Current 	σ
SingleSource/UnitTests/2002-10-13-BadLoad 	-75.00% 	0.0160 	0.0040 	-
SingleSource/UnitTests/2006-12-01-float_varg 	-71.43% 	0.0280 	0.0080 	-
SingleSource/Regression/C++/2003-06-08-VirtualFunctions 	-66.67% 	0.0240 	0.0080 	-
SingleSource/Regression/C++/2003-09-29-NonPODsByValue 	-66.67% 	0.0360 	0.0120 	-
SingleSource/Regression/C++/EH/function_try_block 	-66.67% 	0.0480 	0.0160 	-
SingleSource/UnitTests/byval-alignment 	-63.64% 	0.0440 	0.0160 	-
SingleSource/Benchmarks/Misc/pi 	-58.33% 	0.0480 	0.0200 	-
SingleSource/Regression/C++/short_circuit_dtor 	-58.33% 	0.0480 	0.0200 	-
SingleSource/UnitTests/2005-05-11-Popcount-ffs-fls 	-55.56% 	0.0720 	0.0320 	-
SingleSource/UnitTests/Vector/simple 	-55.56% 	0.0360 	0.0160 	-
SingleSource/Regression/C++/EH/simple_rethrow 	-54.55% 	0.0440 	0.0200 	-
...

Since this is my first LNT testing, could someone comment on these results. Any outliers?

Thanks,
Ivan


http://reviews.llvm.org/D13363