This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
[ping][PATCH v2] Add systemtap markers to math function slow paths
- From: Siddhesh Poyarekar <siddhesh at redhat dot com>
- To: libc-alpha at sourceware dot org
- Cc: law at redhat dot com, carlos at redhat dot com
- Date: Fri, 4 Oct 2013 11:21:26 +0530
- Subject: [ping][PATCH v2] Add systemtap markers to math function slow paths
- Authentication-results: sourceware.org; auth=none
- References: <20130911053106 dot GH4306 at spoyarek dot pnq dot redhat dot com> <20130911053427 dot GI4306 at spoyarek dot pnq dot redhat dot com> <20130917061817 dot GF13304 at spoyarek dot pnq dot redhat dot com>
Ping!
On Tue, Sep 17, 2013 at 11:48:18AM +0530, Siddhesh Poyarekar wrote:
> Hi.
>
> Here is take two of the probes patch with documentation added to
> probes.texi that Alexandre Oliva proposed[1] to add to the glibc
> manual. These probes are hit only in the slow paths of the respective
> functions and are a good aid for developers to see if their
> application uses any inputs that may be resulting in these slow paths
> being invoked.
>
> Regards,
> Siddhesh
>
> 2013-09-11 Siddhesh Poyarekar <siddhesh@redhat.com>
> Jeff Law <law@redhat.com>
>
> * sysdeps/ieee754/dbl-64/e_atan2.c: Include stap-probe.h.
> (atan2Mp): Add systemtap probe marker.
> * sysdeps/ieee754/dbl-64/e_log.c: include stap-probe.h.
> (__ieee754_log): Add systemtap probe marker.
> * sysdeps/ieee754/dbl-64/s_atan.c: Include stap-probe.h.
> (atanMp): Add systemtap probe marker.
> * sysdeps/ieee754/dbl-64/s_tan.c: Include stap-probe.h.
> (tanMp): Add systemtap probe marker.
> * sysdeps/ieee754/dbl-64/slowexp.c: Include stap-probe.h.
> (__slowexp): Add systemtap probe marker.
> * sysdeps/ieee754/dbl-64/slowpow.c: Include stap-probe.h.
> (__slowpow): Add systemtap probe marker.
> * manual/probes.texi: Document probes.
>
> diff --git a/manual/probes.texi b/manual/probes.texi
> index 650bea7..7b5f192 100644
> --- a/manual/probes.texi
> +++ b/manual/probes.texi
> @@ -16,6 +16,7 @@ arguments.
>
> @menu
> * Memory Allocation Probes:: Probes in the memory allocation subsystem
> +* Mathematical Function Probes:: Probes in mathematical functions
> @end menu
>
> @node Memory Allocation Probes
> @@ -249,3 +250,100 @@ This probe is hit when function @code{free} decides to adjust the
> dynamic brk/mmap thresholds. Argument @var{$arg1} and @var{$arg2} are
> the adjusted mmap and trim thresholds, respectively.
> @end deftp
> +
> +@node Mathematical Function Probes
> +@section Mathematical Function Probes
> +
> +Some mathematical functions fall back to multiple precision arithmetic for
> +some inputs to get last bit precision for their return values. This multiple
> +precision fallback is much slower than the default algorithms and may have a
> +significant impact on application performance. The systemtap probe markers
> +described in this section may help you determine if your application calls
> +mathematical functions with inputs that may result in multiple-precision
> +arithmetic.
> +
> +Unless explicitly mentioned otherwise, a precision of 1 implies 24 bits of
> +precision in the mantissa of the multiple precision number. Hence, a precision
> +level of 32 implies 768 bits of precision in the mantissa.
> +
> +@deftp Probe slowexp_p6 (double @var{$arg1}, double @var{$arg2})
> +This probe is hit when the @code{exp} function is called with an input that
> +results in multiple precision computation with precision 6. Argument
> +@var{$arg1} is the input value and @var{$arg2} is the computed output.
> +@end deftp
> +
> +@deftp Probe slowexp_p32 (double @var{$arg1}, double @var{$arg2})
> +This probe is hit when the @code{exp} function is called with an input that
> +results in multiple precision computation with precision 32. Argument
> +@var{$arg1} is the input value and @var{$arg2} is the computed output.
> +@end deftp
> +
> +@deftp Probe slowpow_p10 (double @var{$arg1}, double @var{$arg2}, double @var{$arg3}, double @var{$arg4})
> +This probe is hit when the @code{pow} function is called with inputs that
> +result in multiple precision computation with precision 10. Arguments
> +@var{$arg1} and @var{$arg2} are the input values, @code{$arg3} is the value
> +computed in the fast phase of the algorithm and @code{$arg4} is the final
> +accurate value.
> +@end deftp
> +
> +@deftp Probe slowpow_p32 (double @var{$arg1}, double @var{$arg2}, double @var{$arg3}, double @var{$arg4})
> +This probe is hit when the @code{pow} function is called with an input that
> +results in multiple precision computation with precision 32. Arguments
> +@var{$arg1} and @var{$arg2} are the input values, @code{$arg3} is the value
> +computed in the fast phase of the algorithm and @code{$arg4} is the final
> +accurate value.
> +@end deftp
> +
> +@deftp Probe slowlog (int @var{$arg1}, double @var{$arg2}, double @var{$arg3})
> +This probe is hit when the @code{log} function is called with an input that
> +results in multiple precision computation. Argument @var{$arg1} is the
> +precision with which the computation succeeded. Argument @var{$arg2} is the
> +input and @var{$arg3} is the computed output.
> +@end deftp
> +
> +@deftp Probe slowlog_inexact (int @var{$arg1}, double @var{$arg2}, double @var{$arg3})
> +This probe is hit when the @code{log} function is called with an input that
> +results in multiple precision computation and none of the multiple precision
> +computations result in an accurate result. Argument @var{$arg1} is the maximum
> +precision with which computations were performed. Argument @var{$arg2} is the
> +input and @var{$arg3} is the computed output.
> +@end deftp
> +
> +@deftp Probe slowatan2 (int @var{$arg1}, double @var{$arg2}, double @var{$arg3}, double @var{$arg4})
> +This probe is hit when the @code{atan2} function is called with an input that
> +results in multiple precision computation. Argument @var{$arg1} is the
> +precision with which computation succeeded. Arguments @var{$arg2} and
> +@var{$arg3} are inputs to the @code{atan2} function and @var{$arg4} is the
> +computed result.
> +@end deftp
> +
> +@deftp Probe slowatan2_inexact (int @var{$arg1}, double @var{$arg2}, double @var{$arg3}, double @var{$arg4})
> +This probe is hit when the @code{atan} function is called with an input that
> +results in multiple precision computation and none of the multiple precision
> +computations result in an accurate result. Argument @var{$arg1} is the maximum
> +precision with which computations were performed. Arguments @var{$arg2} and
> +@var{$arg3} are inputs to the @code{atan2} function and @var{$arg4} is the
> +computed result.
> +@end deftp
> +
> +@deftp Probe slowatan (int @var{$arg1}, double @var{$arg2}, double @var{$arg3})
> +This probe is hit when the @code{atan} function is called with an input that
> +results in multiple precision computation. Argument @var{$arg1} is the
> +precision with which computation succeeded. Argument @var{$arg2} is the
> +input to the @code{atan} function and @var{$arg3} is the computed result.
> +@end deftp
> +
> +@deftp Probe slowatan_inexact (int @var{$arg1}, double @var{$arg2}, double @var{$arg3})
> +This probe is hit when the @code{atan} function is called with an input that
> +results in multiple precision computation and none of the multiple precision
> +computations result in an accurate result. Argument @var{$arg1} is the maximum
> +precision with which computations were performed. Argument @var{$arg2} is the
> +input to the @code{atan} function and @var{$arg3} is the computed result.
> +@end deftp
> +
> +@deftp Probe slowtan (double @var{$arg1}, double @var{$arg2})
> +This probe is hit when the @code{tan} function is called with an input that
> +results in multiple precision computation with precision 32. Argument
> +@var{$arg1} is the input to the function and @var{$arg2} is the computed
> +result.
> +@end deftp
> diff --git a/sysdeps/ieee754/dbl-64/e_atan2.c b/sysdeps/ieee754/dbl-64/e_atan2.c
> index 4ebe9c0..4075622 100644
> --- a/sysdeps/ieee754/dbl-64/e_atan2.c
> +++ b/sysdeps/ieee754/dbl-64/e_atan2.c
> @@ -42,6 +42,7 @@
> #include "uatan.tbl"
> #include "atnat2.h"
> #include <math_private.h>
> +#include <stap-probe.h>
>
> #ifndef SECTION
> # define SECTION
> @@ -597,7 +598,11 @@ atan2Mp (double x, double y, const int pr[])
> __mp_dbl (&mpz1, &z1, p);
> __mp_dbl (&mpz2, &z2, p);
> if (z1 == z2)
> - return z1;
> + {
> + LIBC_PROBE (slowatan2, 4, &p, &x, &y, &z1);
> + return z1;
> + }
> }
> + LIBC_PROBE (slowatan2_inexact, 4, &p, &x, &y, &z1);
> return z1; /*if impossible to do exact computing */
> }
> diff --git a/sysdeps/ieee754/dbl-64/e_log.c b/sysdeps/ieee754/dbl-64/e_log.c
> index f9300f9..17ab75a 100644
> --- a/sysdeps/ieee754/dbl-64/e_log.c
> +++ b/sysdeps/ieee754/dbl-64/e_log.c
> @@ -39,6 +39,7 @@
> #include "mpa.h"
> #include "MathLib.h"
> #include <math_private.h>
> +#include <stap-probe.h>
>
> #ifndef SECTION
> # define SECTION
> @@ -242,8 +243,12 @@ stage_n:
> __mp_dbl (&mpy1, &y1, p);
> __mp_dbl (&mpy2, &y2, p);
> if (y1 == y2)
> - return y1;
> + {
> + LIBC_PROBE (slowlog, 3, &p, &x, &y1);
> + return y1;
> + }
> }
> + LIBC_PROBE (slowlog_inexact, 3, &p, &x, &y1);
> return y1;
> }
>
> diff --git a/sysdeps/ieee754/dbl-64/s_atan.c b/sysdeps/ieee754/dbl-64/s_atan.c
> index 7b6c83f..35ab5c1 100644
> --- a/sysdeps/ieee754/dbl-64/s_atan.c
> +++ b/sysdeps/ieee754/dbl-64/s_atan.c
> @@ -42,6 +42,7 @@
> #include "uatan.tbl"
> #include "atnat.h"
> #include <math.h>
> +#include <stap-probe.h>
>
> void __mpatan (mp_no *, mp_no *, int); /* see definition in mpatan.c */
> static double atanMp (double, const int[]);
> @@ -306,8 +307,12 @@ atanMp (double x, const int pr[])
> __mp_dbl (&mpy1, &y1, p);
> __mp_dbl (&mpy2, &y2, p);
> if (y1 == y2)
> - return y1;
> + {
> + LIBC_PROBE (slowatan, 3, &p, &x, &y1);
> + return y1;
> + }
> }
> + LIBC_PROBE (slowatan_inexact, 3, &p, &x, &y1);
> return y1; /*if impossible to do exact computing */
> }
>
> diff --git a/sysdeps/ieee754/dbl-64/s_tan.c b/sysdeps/ieee754/dbl-64/s_tan.c
> index 54f863e..cc0dbbb 100644
> --- a/sysdeps/ieee754/dbl-64/s_tan.c
> +++ b/sysdeps/ieee754/dbl-64/s_tan.c
> @@ -41,6 +41,7 @@
> #include <math.h>
> #include <math_private.h>
> #include <fenv.h>
> +#include <stap-probe.h>
>
> #ifndef SECTION
> # define SECTION
> @@ -838,6 +839,7 @@ tanMp (double x)
> p = 32;
> __mptan (x, &mpy, p);
> __mp_dbl (&mpy, &y, p);
> + LIBC_PROBE (slowtan, 2, &x, &y);
> return y;
> }
>
> diff --git a/sysdeps/ieee754/dbl-64/slowexp.c b/sysdeps/ieee754/dbl-64/slowexp.c
> index 8f353f6..525224f 100644
> --- a/sysdeps/ieee754/dbl-64/slowexp.c
> +++ b/sysdeps/ieee754/dbl-64/slowexp.c
> @@ -29,6 +29,8 @@
> /**************************************************************************/
> #include <math_private.h>
>
> +#include <stap-probe.h>
> +
> #ifndef USE_LONG_DOUBLE_FOR_MP
> # include "mpa.h"
> void __mpexp (mp_no *x, mp_no *y, int p);
> @@ -60,13 +62,22 @@ __slowexp (double x)
> __mp_dbl (&mpw, &w, p);
> __mp_dbl (&mpz, &z, p);
> if (w == z)
> - return w;
> + {
> + /* Track how often we get to the slow exp code plus
> + its input/output values. */
> + LIBC_PROBE (slowexp_p6, 2, &x, &w);
> + return w;
> + }
> else
> {
> p = 32;
> __dbl_mp (x, &mpx, p);
> __mpexp (&mpx, &mpy, p);
> __mp_dbl (&mpy, &res, p);
> +
> + /* Track how often we get to the uber-slow exp code plus
> + its input/output values. */
> + LIBC_PROBE (slowexp_p32, 2, &x, &res);
> return res;
> }
> #else
> diff --git a/sysdeps/ieee754/dbl-64/slowpow.c b/sysdeps/ieee754/dbl-64/slowpow.c
> index a379728..d200c39 100644
> --- a/sysdeps/ieee754/dbl-64/slowpow.c
> +++ b/sysdeps/ieee754/dbl-64/slowpow.c
> @@ -34,6 +34,8 @@
> #include "mpa.h"
> #include <math_private.h>
>
> +#include <stap-probe.h>
> +
> #ifndef SECTION
> # define SECTION
> #endif
> @@ -97,7 +99,12 @@ __slowpow (double x, double y, double z)
> __sub (&mpp, &eps, &mpr1, p);
> __mp_dbl (&mpr1, &res1, p);
> if (res == res1)
> - return res;
> + {
> + /* Track how often we get to the slow pow code plus
> + its input/output values. */
> + LIBC_PROBE (slowpow_p10, 4, &x, &y, &z, &res);
> + return res;
> + }
>
> /* If we don't, then we repeat using a higher precision. 768 bits of
> precision ought to be enough for anybody. */
> @@ -109,5 +116,10 @@ __slowpow (double x, double y, double z)
> __mul (&mpy, &mpz, &mpw, p);
> __mpexp (&mpw, &mpp, p);
> __mp_dbl (&mpp, &res, p);
> +
> + /* Track how often we get to the uber-slow pow code plus
> + its input/output values. */
> + LIBC_PROBE (slowpow_p32, 4, &x, &y, &z, &res);
> +
> return res;
> }