aboutsummaryrefslogtreecommitdiff
path: root/gcc/opts.c
diff options
context:
space:
mode:
authorWilco Dijkstra <wdijkstr@arm.com>2017-01-11 16:38:42 +0000
committerWilco Dijkstra <wilco@gcc.gnu.org>2017-01-11 16:38:42 +0000
commitd2471b3168260e0c467672b308a2bb5eda10ff95 (patch)
tree41d764d66c719c39a698fd982608534d57c2593f /gcc/opts.c
parentdbd8471076ba7d89e3c62a6f2fc82cc3abc82a9d (diff)
downloadgcc-d2471b3168260e0c467672b308a2bb5eda10ff95.zip
gcc-d2471b3168260e0c467672b308a2bb5eda10ff95.tar.gz
gcc-d2471b3168260e0c467672b308a2bb5eda10ff95.tar.bz2
My previous change to the Cortex-A53 scheduler resulted in a 13% regression on a proprietary benchmark.
My previous change to the Cortex-A53 scheduler resulted in a 13% regression on a proprietary benchmark. This turned out to be due to non-optimal scheduling of int to float conversions. This patch separates int to FP transfers from int to float conversions based on experiments to determine the best schedule. As a result of these tweaks the performance of the benchmark improves by 20%. gcc/ * config/arm/cortex-a53.md: Add bypasses for cortex_a53_r2f_cvt. (cortex_a53_r2f): Only use for transfers. (cortex_a53_f2r): Likewise. (cortex_a53_r2f_cvt): Add reservation for conversions. (cortex_a53_f2r_cvt): Likewise. From-SVN: r244322
Diffstat (limited to 'gcc/opts.c')
0 files changed, 0 insertions, 0 deletions