[PATCH] D30903: [AArch64] Use alias analysis in the load/store optimization pass.
Chad Rosier via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Mon Mar 13 12:55:23 PDT 2017
mcrosier added a comment.
Here are the relative stats for SPEC2000/SPEC2006 combine using llvm statistics:
Message Diff %age
-------------------------------------------------------------------------------------- ------- -------
aarch64-ldst-opt - Number of load/store pair instructions generated +969 1.63%
aarch64-ldst-opt - Number of loads from stores promoted +232 362.50%
aarch64-ldst-opt - Number of narrow zero stores promoted +20 2.41%
aarch64-ldst-opt - Number of post-index updates folded -3 -0.12%
asm-printer - Number of machine instrs printed -974 -0.03%
assembler - Number of emitted object file bytes -3616 -0.01%
assembler - Number of evaluated fixups +8 0.00%
assembler - Number of fragment layouts +0 0.00%
basicaa - Number of times a GEP is decomposed +125690 0.28%
basicaa - Number of times the limit to decompose GEPs is reached +43 0.04%
bdce - Number of instructions removed (unused) +0 0.00%
bdce - Number of instructions trivialized (dead bits) +0 0.00%
bitcode-reader - Number of MDStrings loaded +0 0.00%
branch-relaxation - Number of conditional branches relaxed +0 0.00%
branchfolding - Number of block tails merged -6 -0.01%
mccodeemitter - Number of MC fixups created. +8 0.00%
mccodeemitter - Number of MC instructions emitted. -974 -0.03%
mcexpr - Number of MCExpr evaluations +16 0.00%
memory-builtins - Number of arguments with unsolved size and offset +32 0.04%
memory-builtins - Number of load instructions with unsolved size and offset +84 0.16%
https://reviews.llvm.org/D30903
More information about the llvm-commits
mailing list