| Message ID | 20200710224109.18663-1-eshandhawan51@gmail.com |
|---|---|
| State | New |
| Headers | show |
| Series |
|
| Related | show |
On 11/07/2020 00:41, Eshan dhawan wrote: > Signed-off-by: Eshan dhawan<eshandhawan51@gmail.com> > --- > newlib/libm/machine/arm/fenv-vfp.c | 7 ++++- > newlib/libm/machine/arm/fenv.c | 47 +++++++++++++++++++----------- > 2 files changed, 36 insertions(+), 18 deletions(-) What are the design goals of this implementation? Should it work with some sort of run time VFP detection? Should it select the support using builtin defines? I would use builtin defines only and then just use a single file.
On Sat, Jul 11, 2020 at 8:46 PM Sebastian Huber < sebastian.huber@embedded-brains.de> wrote: > On 11/07/2020 00:41, Eshan dhawan wrote: > > > Signed-off-by: Eshan dhawan<eshandhawan51@gmail.com> > > --- > > newlib/libm/machine/arm/fenv-vfp.c | 7 ++++- > > newlib/libm/machine/arm/fenv.c | 47 +++++++++++++++++++----------- > > 2 files changed, 36 insertions(+), 18 deletions(-) > > What are the design goals of this implementation? Should it work with > some sort of run time VFP detection? Should it select the support using > builtin defines? > It used the GCC flags to detect VPF if VPF is not selected all the symbols of vfp won't appear :) > > I would use builtin defines only and then just use a single file. > I tried converting into a single file but the file becomes too complex to handle and check for errors. :)
On 11/07/2020 19:18, Eshan Dhawan wrote: > > > On Sat, Jul 11, 2020 at 8:46 PM Sebastian Huber > <sebastian.huber@embedded-brains.de > <mailto:sebastian.huber@embedded-brains.de>> wrote: > > On 11/07/2020 00:41, Eshan dhawan wrote: > > > Signed-off-by: Eshan dhawan<eshandhawan51@gmail.com > <mailto:eshandhawan51@gmail.com>> > > --- > > newlib/libm/machine/arm/fenv-vfp.c | 7 ++++- > > newlib/libm/machine/arm/fenv.c | 47 > +++++++++++++++++++----------- > > 2 files changed, 36 insertions(+), 18 deletions(-) > > What are the design goals of this implementation? Should it work with > some sort of run time VFP detection? Should it select the support using > builtin defines? > > It used the GCC flags to detect VPF > if VPF is not selected all the symbols of vfp won't appear :) > > > I would use builtin defines only and then just use a single file. > > I tried converting into a single file but the file becomes too complex > to handle and check for errors. :) Complexity is always a bit subjective. How did you test the ARM implementation? I doubt the soft-float support works at all. FreeBSD ships its own soft-float library. Newlib uses the one from libgcc by default. They are not compatible. I guess the MIPS soft-float fenv support has this issue as well. Having an GCC incompatible soft-float fenv support in Newlib makes no sense to me. How can we fix this? Disable it completely or return some run time errors? Maybe we should remove the ARM fenv support from Newlib until these issues are fixed and a working implementation is available. -- Sebastian Huber, embedded brains GmbH Address : Dornierstr. 4, D-82178 Puchheim, Germany Phone : +49 89 189 47 41-16 Fax : +49 89 189 47 41-09 E-Mail : sebastian.huber@embedded-brains.de PGP : Public key available on request. Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG.
On 13/07/2020 07:36, Sebastian Huber wrote: > On 11/07/2020 19:18, Eshan Dhawan wrote: >> >> >> On Sat, Jul 11, 2020 at 8:46 PM Sebastian Huber >> <sebastian.huber@embedded-brains.de >> <mailto:sebastian.huber@embedded-brains.de>> wrote: >> >> On 11/07/2020 00:41, Eshan dhawan wrote: >> >> > Signed-off-by: Eshan dhawan<eshandhawan51@gmail.com >> <mailto:eshandhawan51@gmail.com>> >> > --- >> > newlib/libm/machine/arm/fenv-vfp.c | 7 ++++- >> > newlib/libm/machine/arm/fenv.c | 47 >> +++++++++++++++++++----------- >> > 2 files changed, 36 insertions(+), 18 deletions(-) >> >> What are the design goals of this implementation? Should it work with >> some sort of run time VFP detection? Should it select the support >> using >> builtin defines? >> >> It used the GCC flags to detect VPF >> if VPF is not selected all the symbols of vfp won't appear :) >> >> >> I would use builtin defines only and then just use a single file. >> >> I tried converting into a single file but the file becomes too complex >> to handle and check for errors. :) > > Complexity is always a bit subjective. > > How did you test the ARM implementation? I doubt the soft-float support > works at all. FreeBSD ships its own soft-float library. Newlib uses the > one from libgcc by default. They are not compatible. > > I guess the MIPS soft-float fenv support has this issue as well. > > Having an GCC incompatible soft-float fenv support in Newlib makes no > sense to me. How can we fix this? Disable it completely or return some > run time errors? > > Maybe we should remove the ARM fenv support from Newlib until these > issues are fixed and a working implementation is available. I noticed that the PowerPC fenv support unconditionally assumes that a floating-point unit is available. Not all PowerPC processors support a floating-point unit and some applications may intentionally use soft-float. Using this code may result in an unimplemented exception if MSR[FP]=1, or an FP Unavailable exception if MSR[FP]=0. If this is a desirable behaviour, then we can use the same on ARM. Just provide the VFP implementation. If you use the stuff in other setups you get an exception. -- Sebastian Huber, embedded brains GmbH Address : Dornierstr. 4, D-82178 Puchheim, Germany Phone : +49 89 189 47 41-16 Fax : +49 89 189 47 41-09 E-Mail : sebastian.huber@embedded-brains.de PGP : Public key available on request. Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG.
On Mon, Jul 13, 2020 at 12:48 AM Sebastian Huber < sebastian.huber@embedded-brains.de> wrote: > On 13/07/2020 07:36, Sebastian Huber wrote: > > On 11/07/2020 19:18, Eshan Dhawan wrote: > >> > >> > >> On Sat, Jul 11, 2020 at 8:46 PM Sebastian Huber > >> <sebastian.huber@embedded-brains.de > >> <mailto:sebastian.huber@embedded-brains.de>> wrote: > >> > >> On 11/07/2020 00:41, Eshan dhawan wrote: > >> > >> > Signed-off-by: Eshan dhawan<eshandhawan51@gmail.com > >> <mailto:eshandhawan51@gmail.com>> > >> > --- > >> > newlib/libm/machine/arm/fenv-vfp.c | 7 ++++- > >> > newlib/libm/machine/arm/fenv.c | 47 > >> +++++++++++++++++++----------- > >> > 2 files changed, 36 insertions(+), 18 deletions(-) > >> > >> What are the design goals of this implementation? Should it work > with > >> some sort of run time VFP detection? Should it select the support > >> using > >> builtin defines? > >> > >> It used the GCC flags to detect VPF > >> if VPF is not selected all the symbols of vfp won't appear :) > >> > >> > >> I would use builtin defines only and then just use a single file. > >> > >> I tried converting into a single file but the file becomes too complex > >> to handle and check for errors. :) > > > > Complexity is always a bit subjective. > > > > How did you test the ARM implementation? I doubt the soft-float support > > works at all. FreeBSD ships its own soft-float library. Newlib uses the > > one from libgcc by default. They are not compatible. > > > > I guess the MIPS soft-float fenv support has this issue as well. > > > > Having an GCC incompatible soft-float fenv support in Newlib makes no > > sense to me. How can we fix this? Disable it completely or return some > > run time errors? > > > > Maybe we should remove the ARM fenv support from Newlib until these > > issues are fixed and a working implementation is available. > > I noticed that the PowerPC fenv support unconditionally assumes that a > floating-point unit is available. Not all PowerPC processors support a > floating-point unit and some applications may intentionally use > soft-float. Using this code may result in an unimplemented exception if > MSR[FP]=1, or an FP Unavailable exception if MSR[FP]=0. > > If this is a desirable behaviour, then we can use the same on ARM. Just > provide the VFP implementation. If you use the stuff in other setups you > get an exception. > The RISC-V would be a good model. Jim Wilson provided some default behavior when there is no hard FPU. --joel > > -- > Sebastian Huber, embedded brains GmbH > > Address : Dornierstr. 4, D-82178 Puchheim, Germany > Phone : +49 89 189 47 41-16 > Fax : +49 89 189 47 41-09 > E-Mail : sebastian.huber@embedded-brains.de > PGP : Public key available on request. > > Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG. >
On Tue, Jul 14, 2020 at 6:11 AM Joel Sherrill <joel@rtems.org> wrote: > The RISC-V would be a good model. Jim Wilson provided some default > behavior when there is no hard FPU. It wasn't me personally. This predates my involvement. I'd guess either Andrew Waterman, Palmer Dabbelt, or Kito Cheng. Anyways, we use the hard FPU if available, both for hard-float and soft-float ABIs. If no hard FPU, then the various fenv routines do very little, just returning some default values. Jim
On 14/07/2020 18:59, Jim Wilson wrote: > On Tue, Jul 14, 2020 at 6:11 AM Joel Sherrill<joel@rtems.org> wrote: >> The RISC-V would be a good model. Jim Wilson provided some default >> behavior when there is no hard FPU. > It wasn't me personally. This predates my involvement. I'd guess > either Andrew Waterman, Palmer Dabbelt, or Kito Cheng. Anyways, we > use the hard FPU if available, both for hard-float and soft-float > ABIs. If no hard FPU, then the various fenv routines do very little, > just returning some default values. Ok, good. Eshan, could you please use the RISC-V as a model for the ARM support. Please reduce the file count. I think we just need one fenv.c and not additional source and header files. Alternatively, you can split it up into one file per function similar to RISC-V.
diff --git a/newlib/libm/machine/arm/fenv-vfp.c b/newlib/libm/machine/arm/fenv-vfp.c index 297e81296..11a05ea54 100644 --- a/newlib/libm/machine/arm/fenv-vfp.c +++ b/newlib/libm/machine/arm/fenv-vfp.c @@ -26,7 +26,12 @@ * $FreeBSD$ */ +#if defined(__VFP_FP__) && !defined(__ARM_PCS_VFP) +#define SOFTFP_ABI +#endif + +#ifndef SOFTFP_ABI #define FENV_MANGLE(x) __vfp_ ##x #include <machine/fenv-mangle.h> #include "fenv.c" - +#endif diff --git a/newlib/libm/machine/arm/fenv.c b/newlib/libm/machine/arm/fenv.c index 5b61ab81f..98125cacf 100644 --- a/newlib/libm/machine/arm/fenv.c +++ b/newlib/libm/machine/arm/fenv.c @@ -43,7 +43,6 @@ #define SOFTFP_ABI #endif - #ifndef FENV_MANGLE /* * Hopefully the system ID byte is immutable, so it's valid to use @@ -102,8 +101,6 @@ extern inline int fegetexcept(void); #else /* !FENV_MANGLE && SOFTFP_ABI */ /* Set by libc when the VFP unit is enabled */ -int _libc_arm_fpu_present; - int __softfp_feclearexcept(int excepts); int __softfp_fegetexceptflag(fexcept_t *flagp, int excepts); int __softfp_fesetexceptflag(const fexcept_t *flagp, int excepts); @@ -119,6 +116,7 @@ int __softfp_feenableexcept(int __mask); int __softfp_fedisableexcept(int __mask); int __softfp_fegetexcept(void); +#ifndef SOFTFP_ABI int __vfp_feclearexcept(int excepts); int __vfp_fegetexceptflag(fexcept_t *flagp, int excepts); int __vfp_fesetexceptflag(const fexcept_t *flagp, int excepts); @@ -133,6 +131,7 @@ int __vfp_feupdateenv(const fenv_t *envp); int __vfp_feenableexcept(int __mask); int __vfp_fedisableexcept(int __mask); int __vfp_fegetexcept(void); +#endif static int __softfp_round_to_vfp(int round) @@ -171,8 +170,9 @@ __softfp_round_from_vfp(int round) int feclearexcept(int excepts) { - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI __vfp_feclearexcept(excepts); +#endif __softfp_feclearexcept(excepts); return (0); @@ -183,8 +183,9 @@ int fegetexceptflag(fexcept_t *flagp, int excepts) fexcept_t __vfp_flagp; __vfp_flagp = 0; - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI __vfp_fegetexceptflag(&__vfp_flagp, excepts); +#endif __softfp_fegetexceptflag(flagp, excepts); *flagp |= __vfp_flagp; @@ -195,8 +196,9 @@ int fegetexceptflag(fexcept_t *flagp, int excepts) int fesetexceptflag(const fexcept_t *flagp, int excepts) { - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI __vfp_fesetexceptflag(flagp, excepts); +#endif __softfp_fesetexceptflag(flagp, excepts); return (0); @@ -205,8 +207,9 @@ int fesetexceptflag(const fexcept_t *flagp, int excepts) int feraiseexcept(int excepts) { - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI __vfp_feraiseexcept(excepts); +#endif __softfp_feraiseexcept(excepts); return (0); @@ -217,8 +220,9 @@ int fetestexcept(int excepts) int __got_excepts; __got_excepts = 0; - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI __got_excepts = __vfp_fetestexcept(excepts); +#endif __got_excepts |= __softfp_fetestexcept(excepts); return (__got_excepts); @@ -227,16 +231,18 @@ int fetestexcept(int excepts) int fegetround(void) { - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI return __softfp_round_from_vfp(__vfp_fegetround()); +#endif return __softfp_fegetround(); } int fesetround(int round) { - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI __vfp_fesetround(__softfp_round_to_vfp(round)); +#endif __softfp_fesetround(round); return (0); @@ -247,8 +253,9 @@ int fegetenv(fenv_t *envp) fenv_t __vfp_envp; __vfp_envp = 0; - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI __vfp_fegetenv(&__vfp_envp); +#endif __softfp_fegetenv(envp); *envp |= __vfp_envp; @@ -260,8 +267,9 @@ int feholdexcept(fenv_t *envp) fenv_t __vfp_envp; __vfp_envp = 0; - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI __vfp_feholdexcept(&__vfp_envp); +#endif __softfp_feholdexcept(envp); *envp |= __vfp_envp; @@ -271,8 +279,9 @@ int feholdexcept(fenv_t *envp) int fesetenv(const fenv_t *envp) { - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI __vfp_fesetenv(envp); +#endif __softfp_fesetenv(envp); return (0); @@ -281,8 +290,9 @@ int fesetenv(const fenv_t *envp) int feupdateenv(const fenv_t *envp) { - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI __vfp_feupdateenv(envp); +#endif __softfp_feupdateenv(envp); return (0); @@ -293,8 +303,9 @@ int feenableexcept(int __mask) int __unmasked; __unmasked = 0; - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI __unmasked = __vfp_feenableexcept(__mask); +#endif __unmasked |= __softfp_feenableexcept(__mask); return (__unmasked); @@ -305,8 +316,9 @@ int fedisableexcept(int __mask) int __unmasked; __unmasked = 0; - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI __unmasked = __vfp_fedisableexcept(__mask); +#endif __unmasked |= __softfp_fedisableexcept(__mask); return (__unmasked); @@ -317,8 +329,9 @@ int fegetexcept(void) int __unmasked; __unmasked = 0; - if (_libc_arm_fpu_present) +#ifndef SOFTFP_ABI __unmasked = __vfp_fegetexcept(); +#endif __unmasked |= __softfp_fegetexcept(); return (__unmasked);
Signed-off-by: Eshan dhawan <eshandhawan51@gmail.com> --- newlib/libm/machine/arm/fenv-vfp.c | 7 ++++- newlib/libm/machine/arm/fenv.c | 47 +++++++++++++++++++----------- 2 files changed, 36 insertions(+), 18 deletions(-) -- 2.17.1