Status Update
Samy Bahra
- Data from the fast path latency benchmarks has been added, please see
http://concurrencykit.org/doc/appendixZ.html
This provides "best case" approximations (in NS) of the latency
associated with various CK primitives. It will be regularly updated
to include different architectures and/or micro-architectures.
Build
- People have requested that the configure script behave more like an
autoconf-generated script. Certain flags have been dropped and behavior
has slightly changed. Please see ./configure --help for more information.
Some bugs in options parsing has been fixed.
- Tip: If you are on an unsupported platform which GCC provides atomic
primitives for you may be able to build CK using:
CC=<compiler of choice> ./configure --profile=<PROFILE>. See example
build profiles in the build directory.
- Tip: If you require 32-bit IA32 support, until the 32-bit port is cleaned,
you may be able to use the GCC atomics interface by using:
CC=gcc ./configure --profile=x86
Regressions
- Common utility functions (CPU affinity and highly-granular timers) were
factored out of the regressions into a regressions/common.h
- The timer for SPARCv9 (%ticks) has been re-enabled.
- Fast path latency micro-benchmarks have been added for the spinlocks,
bytelocks, FIFO and stack. Output is in CPU ticks.
ck_pr
- The CK_F_PR_CAS_32_VALUE feature flag has been added for sparcv9.
- include/ck_pr.h now provides convenience functions for implementing
fixed-width CAS-based atomics based on what lower-level MD-specific
atomics provide. For example, the SPARCv9 port only provides
CAS/LOAD/STORE/SWAP. If the generic include/ck_pr.h detects the
presence of these it will provide an implementation of ck_pr_add,
etc... This will greatly ease porting efforts as a bare-bones port
to a platform will only require fixed-width CAS/LOAD/STORE
(some examples include ck_pr_cas_32_value, ck_pr_cas_64_value,
ck_pr_load_16).
ck_hp_stack:
- Added CK_HP_*_SLOTS_* constants.
Regards.
--
Samy Al Bahra [http://repnop.org/]