|
|
This patch replaces i386 assembly versions of e_logf with generic
e_logf.c. For workload-spec2017.wrf, on Nehalem, it improves
performance by:
Before After Improvement
reciprocal-throughput 73.3865 40.0454 83%
latency 90.0985 54.4479 65%
On Skylake, it improves performance by:
Before After Improvement
reciprocal-throughput 75.1384 22.1452 239%
latency 91.9441 50.7925 81%
On IvyBridge with --disable-multi-arch, it improves performance by:
Before After Improvement
reciprocal-throughput 84.5575 28.7879 193%
latency 103.971 57.5231 80%
* sysdeps/i386/fpu/e_logf.S: Removed.
* sysdeps/i386/fpu/e_logf_data.c: Likewise.
* sysdeps/i386/fpu/w_logf.c: Likewise.
* sysdeps/i386/i686/fpu/e_logf.S: Likewise.
* sysdeps/i386/fpu/libm-test-ulps: Updated for generic e_logf.c.
* sysdeps/i386/i686/fpu/multiarch/libm-test-ulps: Likewise.
* sysdeps/i386/i686/fpu/multiarch/Makefile (libm-sysdep_routines):
Add e_logf-sse2.
(CFLAGS-e_logf-sse2.c): New.
* sysdeps/i386/i686/fpu/multiarch/e_logf-sse2.c: New file.
* sysdeps/i386/i686/fpu/multiarch/e_logf.c: Likewise.
|
