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

Bruno Cardoso Lopes via llvm-commits llvm-commits at lists.llvm.org
Mon Nov 16 10:54:00 PST 2015


bruno added a comment.

Hi Ivan,

Sorry for the delay. Below the x86 results, of a 5x run in a stable machine.
Highlights include: CINT2006/458_sjeng performing 1.1% slower while CINT2000/175_vpr is 1.5% faster.

Performance Regressions - Execution Time	Δ	Previous	Current	σ
MultiSource/Benchmarks/TSVC/IndirectAddressing-flt/IndirectAddressing-flt	3.80%	2.5611	2.6583	0.0179
MultiSource/Benchmarks/sim/sim	2.93%	3.2302	3.3250	0.0019
SingleSource/Benchmarks/Polybench/medley/floyd-warshall/floyd-warshall	2.70%	1.6456	1.6900	0.0172
MultiSource/Benchmarks/NPB-serial/is/is	1.17%	5.6993	5.7660	0.0172
External/SPEC/CINT2006/458_sjeng/458_sjeng	1.14%	X	Y	Z
SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trmm/trmm	1.13%	1.4695	1.4861	0.0055
MultiSource/Benchmarks/TSVC/LoopRerolling-flt/LoopRerolling-flt	1.13%	2.2458	2.2711	0.0082

Performance Improvements - Compile Time	Δ	Previous	Current	σ
MultiSource/Benchmarks/Trimaran/enc-pc1/enc-pc1	-16.00%	0.0844	0.0709	0.0002

Performance Improvements - Execution Time	Δ	Previous	Current	σ
MultiSource/Benchmarks/Trimaran/enc-pc1/enc-pc1	-12.79%	0.4956	0.4322	0.0000
MultiSource/Benchmarks/mafft/pairlocalalign	-5.84%	18.7424	17.6486	0.0154
External/SPEC/CINT2000/175_vpr/175_vpr	-1.50%	X	Y	Z
MultiSource/Applications/sqlite3/sqlite3	 -1.27%	2.5050	2.4732	0.0107


http://reviews.llvm.org/D13363





More information about the llvm-commits mailing list