d/numam-dpdk
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

examples/performance-thread: fix compilation on Suse 11 SP2

Browse Source 
Fixes following compilation error, using uint64_t type,
instead of int128_t unnecessarily:

In file included from ./common/lthread.c:82:0:
./common/lthread_timer.h: In function ‘_ns_to_clks’:
./common/lthread_timer.h:49:20: error: expected ‘=’, ‘,’, ‘;’,
 ‘asm’ or ‘__attribute__’ before ‘clkns’
compilation terminated due to -Wfatal-errors.

Fixes: 116819b9ed ("examples/performance-thread: add lthread subsystem")
Cc: stable@dpdk.org

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
This commit is contained in:
Bruce Richardson 2017-04-24 14:04:22 +01:00 committed by Thomas Monjalon
parent 9aba2a3f3d
commit 1e6d5a9608
1 changed files with 14 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
examples/performance-thread/common/lthread_timer.h
View File

@ -46,11 +46,22 @@ extern "C" {
static inline uint64_t
_ns_to_clks(uint64_t ns)
{
unsigned __int128 clkns = rte_get_tsc_hz();
/*
* clkns needs to be divided by 1E9 to get ns clocks. However,
* dividing by this first would lose a lot of accuracy.
* Dividing after a multiply by ns, could cause overflow of
* uint64_t if ns is about 5 seconds [if we assume a max tsc
* rate of 4GHz]. Therefore we first divide by 1E4, then
* multiply and finally divide by 1E5. This allows ns to be
* values many hours long, without overflow, while still keeping
* reasonable accuracy.
*/
uint64_t clkns = rte_get_tsc_hz() / 1e4;
clkns *= ns;
clkns /= 1000000000;
return (uint64_t) clkns;
clkns /= 1e5;
return clkns;
}