| Message ID | ork0sr7rqq.fsf@lxoliva.fsfla.org |
|---|---|
| State | New |
| Headers | show |
| Series | calibrate intervals to avoid zero in futures poll test | expand |
On Jan 5, 2021, Alexandre Oliva <oliva@adacore.com> wrote: > We get occasional failures of 30_threads/future/members/poll.cc > on some platforms whose high resolution clock doesn't have such a high > resolution; wait_for_0 ends up as 0, and then some asserts fail as > intervals measured as longer than zero are tested for less than > several times zero. > This patch adds some calibration in the iteration count to set a > measurable base time interval with some additional margin. > Regstrapped on x86_64-linux-gnu, and also tested on > x-arm-wrs-vxworks7r2. Ok to install? Ping? https://gcc.gnu.org/pipermail/gcc-patches/2021-January/562796.html > for libstdc++-v3/ChangeLog > * testsuite/30_threads/future/members/poll.cc: Calibrate > iteration count.
On 05/01/21 04:44 -0300, Alexandre Oliva wrote: > >We get occasional failures of 30_threads/future/members/poll.cc >on some platforms whose high resolution clock doesn't have such a high >resolution; wait_for_0 ends up as 0, and then some asserts fail as >intervals measured as longer than zero are tested for less than >several times zero. > >This patch adds some calibration in the iteration count to set a >measurable base time interval with some additional margin. > >Regstrapped on x86_64-linux-gnu, and also tested on >x-arm-wrs-vxworks7r2. Ok to install? > > >for libstdc++-v3/ChangeLog > > * testsuite/30_threads/future/members/poll.cc: Calibrate > iteration count. >--- > .../testsuite/30_threads/future/members/poll.cc | 33 +++++++++++++++++++- > 1 file changed, 32 insertions(+), 1 deletion(-) > >diff --git a/libstdc++-v3/testsuite/30_threads/future/members/poll.cc b/libstdc++-v3/testsuite/30_threads/future/members/poll.cc >index fff9bea899c90..7b41411a54386 100644 >--- a/libstdc++-v3/testsuite/30_threads/future/members/poll.cc >+++ b/libstdc++-v3/testsuite/30_threads/future/members/poll.cc >@@ -25,7 +25,7 @@ > #include <iostream> > #include <testsuite_hooks.h> > >-const int iterations = 200; >+int iterations = 200; > > using namespace std; > >@@ -45,10 +45,41 @@ int main() > promise<int> p; > future<int> f = p.get_future(); > >+ start_over: > auto start = chrono::high_resolution_clock::now(); > for(int i = 0; i < iterations; i++) > f.wait_for(chrono::seconds(0)); > auto stop = chrono::high_resolution_clock::now(); >+ >+ /* We've run too few iterations for the clock resolution. >+ Attempt to calibrate it. */ >+ if (start == stop) >+ { >+ /* Loop until the clock advances, so that start is right after a >+ time increment. */ >+ do >+ start = chrono::high_resolution_clock::now(); >+ while (start == stop); >+ int i = 0; >+ /* Now until the clock advances again, so that stop is right >+ after another time increment. */ >+ do >+ { >+ f.wait_for(chrono::seconds(0)); >+ stop = chrono::high_resolution_clock::now(); >+ i++; >+ } >+ while (start == stop); >+ /* Got for some 10 cycles, but we're already past that and still I can't parse "Got for some 10 cycles". If that's just a typo that I'm failing to spot ("good for"?) please fix and push the patch. The patch is fine apart from me being unable to understand this comment. >+ get into the calibration loop, double the iteration count and >+ try again. */ >+ if (iterations < i * 10) >+ iterations = i * 10; >+ else >+ iterations *= 2; >+ goto start_over; >+ } >+ > double wait_for_0 = print("wait_for(0s)", stop - start); > > start = chrono::high_resolution_clock::now(); > > >-- >Alexandre Oliva, happy hacker https://FSFLA.org/blogs/lxo/ > Free Software Activist GNU Toolchain Engineer > Vim, Vi, Voltei pro Emacs -- GNUlius Caesar >
On Jan 14, 2021, Jonathan Wakely <jwakely@redhat.com> wrote: >> + /* Got for some 10 cycles, but we're already past that and still > I can't parse "Got for some 10 cycles". If that's just a typo Yeah, I meant "Go for ... but if ..." and managed to double-mangle it. Thanks for spotting it. Here's the patch I'm installing, with the typos fixed. Thanks! calibrate intervals to avoid zero in futures poll test From: Alexandre Oliva <oliva@adacore.com> We get occasional failures of 30_threads/future/members/poll.cc on some platforms whose high resolution clock doesn't have such a high resolution; wait_for_0 ends up as 0, and then some asserts fail as intervals measured as longer than zero are tested for less than several times zero. This patch adds some calibration in the iteration count to set a measurable base time interval with some additional margin. for libstdc++-v3/ChangeLog * testsuite/30_threads/future/members/poll.cc: Calibrate iteration count. --- .../testsuite/30_threads/future/members/poll.cc | 33 +++++++++++++++++++- 1 file changed, 32 insertions(+), 1 deletion(-) diff --git a/libstdc++-v3/testsuite/30_threads/future/members/poll.cc b/libstdc++-v3/testsuite/30_threads/future/members/poll.cc index 91f685b172d73..133dae15ac471 100644 --- a/libstdc++-v3/testsuite/30_threads/future/members/poll.cc +++ b/libstdc++-v3/testsuite/30_threads/future/members/poll.cc @@ -25,7 +25,7 @@ #include <iostream> #include <testsuite_hooks.h> -const int iterations = 200; +int iterations = 200; using namespace std; @@ -45,10 +45,41 @@ int main() promise<int> p; future<int> f = p.get_future(); + start_over: auto start = chrono::high_resolution_clock::now(); for(int i = 0; i < iterations; i++) f.wait_for(chrono::seconds(0)); auto stop = chrono::high_resolution_clock::now(); + + /* We've run too few iterations for the clock resolution. + Attempt to calibrate it. */ + if (start == stop) + { + /* Loop until the clock advances, so that start is right after a + time increment. */ + do + start = chrono::high_resolution_clock::now(); + while (start == stop); + int i = 0; + /* Now until the clock advances again, so that stop is right + after another time increment. */ + do + { + f.wait_for(chrono::seconds(0)); + stop = chrono::high_resolution_clock::now(); + i++; + } + while (start == stop); + /* Go for some 10 cycles, but if we're already past that and + still get into the calibration loop, double the iteration + count and try again. */ + if (iterations < i * 10) + iterations = i * 10; + else + iterations *= 2; + goto start_over; + } + double wait_for_0 = print("wait_for(0s)", stop - start); start = chrono::high_resolution_clock::now();
On Thu, 14 Jan 2021 at 19:57, Alexandre Oliva <oliva@adacore.com> wrote: > > On Jan 14, 2021, Jonathan Wakely <jwakely@redhat.com> wrote: > > >> + /* Got for some 10 cycles, but we're already past that and still > > > I can't parse "Got for some 10 cycles". If that's just a typo > > Yeah, I meant "Go for ... but if ..." and managed to double-mangle it. > Thanks for spotting it. Here's the patch I'm installing, with the typos > fixed. Thanks! > > > calibrate intervals to avoid zero in futures poll test > > From: Alexandre Oliva <oliva@adacore.com> > > We get occasional failures of 30_threads/future/members/poll.cc > on some platforms whose high resolution clock doesn't have such a high > resolution; wait_for_0 ends up as 0, and then some asserts fail as > intervals measured as longer than zero are tested for less than > several times zero. > > This patch adds some calibration in the iteration count to set a > measurable base time interval with some additional margin. > Seeing such random errors when testing arm target under qemu on shared servers. I noticed several such errors on gcc-testresults, too. So I guess this is a ping for this patch, to clear this noise in the results? Thanks, Christophe > > for libstdc++-v3/ChangeLog > > * testsuite/30_threads/future/members/poll.cc: Calibrate > iteration count. > --- > .../testsuite/30_threads/future/members/poll.cc | 33 +++++++++++++++++++- > 1 file changed, 32 insertions(+), 1 deletion(-) > > diff --git a/libstdc++-v3/testsuite/30_threads/future/members/poll.cc b/libstdc++-v3/testsuite/30_threads/future/members/poll.cc > index 91f685b172d73..133dae15ac471 100644 > --- a/libstdc++-v3/testsuite/30_threads/future/members/poll.cc > +++ b/libstdc++-v3/testsuite/30_threads/future/members/poll.cc > @@ -25,7 +25,7 @@ > #include <iostream> > #include <testsuite_hooks.h> > > -const int iterations = 200; > +int iterations = 200; > > using namespace std; > > @@ -45,10 +45,41 @@ int main() > promise<int> p; > future<int> f = p.get_future(); > > + start_over: > auto start = chrono::high_resolution_clock::now(); > for(int i = 0; i < iterations; i++) > f.wait_for(chrono::seconds(0)); > auto stop = chrono::high_resolution_clock::now(); > + > + /* We've run too few iterations for the clock resolution. > + Attempt to calibrate it. */ > + if (start == stop) > + { > + /* Loop until the clock advances, so that start is right after a > + time increment. */ > + do > + start = chrono::high_resolution_clock::now(); > + while (start == stop); > + int i = 0; > + /* Now until the clock advances again, so that stop is right > + after another time increment. */ > + do > + { > + f.wait_for(chrono::seconds(0)); > + stop = chrono::high_resolution_clock::now(); > + i++; > + } > + while (start == stop); > + /* Go for some 10 cycles, but if we're already past that and > + still get into the calibration loop, double the iteration > + count and try again. */ > + if (iterations < i * 10) > + iterations = i * 10; > + else > + iterations *= 2; > + goto start_over; > + } > + > double wait_for_0 = print("wait_for(0s)", stop - start); > > start = chrono::high_resolution_clock::now(); > > > -- > Alexandre Oliva, happy hacker https://FSFLA.org/blogs/lxo/ > Free Software Activist GNU Toolchain Engineer > Vim, Vi, Voltei pro Emacs -- GNUlius Caesar
diff --git a/libstdc++-v3/testsuite/30_threads/future/members/poll.cc b/libstdc++-v3/testsuite/30_threads/future/members/poll.cc index fff9bea899c90..7b41411a54386 100644 --- a/libstdc++-v3/testsuite/30_threads/future/members/poll.cc +++ b/libstdc++-v3/testsuite/30_threads/future/members/poll.cc @@ -25,7 +25,7 @@ #include <iostream> #include <testsuite_hooks.h> -const int iterations = 200; +int iterations = 200; using namespace std; @@ -45,10 +45,41 @@ int main() promise<int> p; future<int> f = p.get_future(); + start_over: auto start = chrono::high_resolution_clock::now(); for(int i = 0; i < iterations; i++) f.wait_for(chrono::seconds(0)); auto stop = chrono::high_resolution_clock::now(); + + /* We've run too few iterations for the clock resolution. + Attempt to calibrate it. */ + if (start == stop) + { + /* Loop until the clock advances, so that start is right after a + time increment. */ + do + start = chrono::high_resolution_clock::now(); + while (start == stop); + int i = 0; + /* Now until the clock advances again, so that stop is right + after another time increment. */ + do + { + f.wait_for(chrono::seconds(0)); + stop = chrono::high_resolution_clock::now(); + i++; + } + while (start == stop); + /* Got for some 10 cycles, but we're already past that and still + get into the calibration loop, double the iteration count and + try again. */ + if (iterations < i * 10) + iterations = i * 10; + else + iterations *= 2; + goto start_over; + } + double wait_for_0 = print("wait_for(0s)", stop - start); start = chrono::high_resolution_clock::now();