diff mbox series

[04/28] powerpc: ceil/ceilf refactor

Message ID 20190329133529.22523-5-adhemerval.zanella@linaro.org
State Accepted
Commit 6cac323c8dd78668e65aaa29f044cbd33c1a66a5
Headers show
Series powerpc floating-point optimization refactor | expand

Commit Message

Adhemerval Zanella Netto March 29, 2019, 1:35 p.m. UTC
This patches consolidates all the powerpc ceil{f} implementations on
the generic sysdeps/powerpc/fpu/s_ceil{f}.  The generic implementation
uses either the compiler builts for ISA 2.03+ (which generates the frip
instruction) or a generic implementation which uses FP only operations.

It adds a generic implementation (round_to_integer.h) which is shared
with other rounding to integer routines.  The resulting code should be
similar performance-wise to previous assembly one.

The IFUNC organization for powerpc64 is also change to be enabled only
for powerpc64 and not for powerpc64le.

Checked on powerpc-linux-gnu (built without --with-cpu, with
--with-cpu=power4 and with --with-cpu=power5+ and --disable-multi-arch),
powerpc64-linux-gnu (built without --with-cp and with --with-cpu=power5+
and --disable-multi-arch).

	* sysdeps/powerpc/fpu/fenv_libc.h (__fesetround_inline_nocheck): New
	macro.
	* sysdeps/powerpc/fpu/round_to_integer.h: New file.
	* sysdeps/powerpc/fpu/s_ceil.c: Likewise.
	* sysdeps/powerpc/fpu/s_ceilf.c: Likewise.
	* sysdeps/powerpc/powerpc32/fpu/s_ceil.S: Remove file.
	* sysdeps/powerpc/powerpc32/fpu/s_ceilf.S: Likewise.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/Makefile
	(CFLAGS-s_ceil-power5+.c, CFLAGS-s_ceilf-power5+.c): New rule.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-power5+.S:
	Remove file.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-ppc32.S:
	Likewise.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-power5+.S:
	Likewise.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-ppc32.S:
	Likewise.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-power5+.c:
	New file.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-ppc32.c:
	Likewise.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-power5+.c:
	Likewise.
	* sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-ppc32.c:
	Likewise.
	* sysdeps/powerpc/powerpc32/power5+/fpu/s_ceil.S: Remove file.
	* sysdeps/powerpc/powerpc32/power5+/fpu/s_ceilf.S: Likewise.
	* sysdeps/powerpc/powerpc64/be/fpu/multiarch/Makefile: New file.
	* sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil-power5+.c:
	Likewise.
	* sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil-ppc64.c:
	Likewise.
	* sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil.c: Move to ...
	* sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil.c: ... here.
	* sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf-power5+.c: New
	file.
	* sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf-ppc64.c:
	Likewise.
	* sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf.c: Move to ...
	* sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf.c: ...
	* here.
	* sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile
	(libm-sysdep_routines): Remove s_ceil-power5+, s_ceil-ppc64,
	s_ceilf-power5+, and s_ceilf-ppc64.
	* sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-power5+.S: Remove
	file.
	* sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-ppc64.S: Likewise.
	* sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-power5+.S: Likewise.
	* sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-ppc64.S: Likewise.
	* sysdeps/powerpc/powerpc64/fpu/s_ceil.S: Likewise.
	* sysdeps/powerpc/powerpc64/fpu/s_ceilf.S: Likewise.
	* sysdeps/powerpc/powerpc64/power5+/fpu/s_ceil.S: Likewise.
	* sysdeps/powerpc/powerpc64/power5+/fpu/s_ceilf.S: Likewise.
---
 sysdeps/powerpc/fpu/fenv_libc.h               |   8 ++
 sysdeps/powerpc/fpu/round_to_integer.h        | 109 ++++++++++++++++++
 sysdeps/powerpc/fpu/s_ceil.c                  |  35 ++++++
 sysdeps/powerpc/fpu/s_ceilf.c                 |  35 ++++++
 sysdeps/powerpc/powerpc32/fpu/s_ceil.S        |  76 ------------
 sysdeps/powerpc/powerpc32/fpu/s_ceilf.S       |  76 ------------
 .../powerpc32/power4/fpu/multiarch/Makefile   |   2 +
 .../power4/fpu/multiarch/s_ceil-power5+.S     |  33 ------
 .../power4/fpu/multiarch/s_ceil-power5+.c     |   3 +
 .../power4/fpu/multiarch/s_ceil-ppc32.S       |  31 -----
 .../power4/fpu/multiarch/s_ceil-ppc32.c       |   3 +
 .../power4/fpu/multiarch/s_ceilf-power5+.S    |  26 -----
 .../power4/fpu/multiarch/s_ceilf-power5+.c    |   3 +
 .../power4/fpu/multiarch/s_ceilf-ppc32.S      |  27 -----
 .../power4/fpu/multiarch/s_ceilf-ppc32.c      |   3 +
 .../powerpc/powerpc32/power5+/fpu/s_ceil.S    |  29 -----
 .../powerpc/powerpc32/power5+/fpu/s_ceilf.S   |  30 -----
 .../powerpc64/be/fpu/multiarch/Makefile       |   9 ++
 .../be/fpu/multiarch/s_ceil-power5+.c         |   3 +
 .../powerpc64/be/fpu/multiarch/s_ceil-ppc64.c |   3 +
 .../powerpc64/{ => be}/fpu/multiarch/s_ceil.c |   4 +-
 .../be/fpu/multiarch/s_ceilf-power5+.c        |   3 +
 .../be/fpu/multiarch/s_ceilf-ppc64.c          |   3 +
 .../{ => be}/fpu/multiarch/s_ceilf.c          |   4 +-
 .../powerpc/powerpc64/fpu/multiarch/Makefile  |   3 +-
 .../powerpc64/fpu/multiarch/s_ceil-power5+.S  |  30 -----
 .../powerpc64/fpu/multiarch/s_ceil-ppc64.S    |  30 -----
 .../powerpc64/fpu/multiarch/s_ceilf-power5+.S |  24 ----
 .../powerpc64/fpu/multiarch/s_ceilf-ppc64.S   |  24 ----
 sysdeps/powerpc/powerpc64/fpu/s_ceil.S        |  65 -----------
 sysdeps/powerpc/powerpc64/fpu/s_ceilf.S       |  67 -----------
 .../powerpc/powerpc64/power5+/fpu/s_ceil.S    |  30 -----
 .../powerpc/powerpc64/power5+/fpu/s_ceilf.S   |  31 -----
 33 files changed, 225 insertions(+), 637 deletions(-)
 create mode 100644 sysdeps/powerpc/fpu/round_to_integer.h
 create mode 100644 sysdeps/powerpc/fpu/s_ceil.c
 create mode 100644 sysdeps/powerpc/fpu/s_ceilf.c
 delete mode 100644 sysdeps/powerpc/powerpc32/fpu/s_ceil.S
 delete mode 100644 sysdeps/powerpc/powerpc32/fpu/s_ceilf.S
 delete mode 100644 sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-power5+.S
 create mode 100644 sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-power5+.c
 delete mode 100644 sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-ppc32.S
 create mode 100644 sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-ppc32.c
 delete mode 100644 sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-power5+.S
 create mode 100644 sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-power5+.c
 delete mode 100644 sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-ppc32.S
 create mode 100644 sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-ppc32.c
 delete mode 100644 sysdeps/powerpc/powerpc32/power5+/fpu/s_ceil.S
 delete mode 100644 sysdeps/powerpc/powerpc32/power5+/fpu/s_ceilf.S
 create mode 100644 sysdeps/powerpc/powerpc64/be/fpu/multiarch/Makefile
 create mode 100644 sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil-power5+.c
 create mode 100644 sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil-ppc64.c
 rename sysdeps/powerpc/powerpc64/{ => be}/fpu/multiarch/s_ceil.c (95%)
 create mode 100644 sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf-power5+.c
 create mode 100644 sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf-ppc64.c
 rename sysdeps/powerpc/powerpc64/{ => be}/fpu/multiarch/s_ceilf.c (95%)
 delete mode 100644 sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-power5+.S
 delete mode 100644 sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-ppc64.S
 delete mode 100644 sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-power5+.S
 delete mode 100644 sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-ppc64.S
 delete mode 100644 sysdeps/powerpc/powerpc64/fpu/s_ceil.S
 delete mode 100644 sysdeps/powerpc/powerpc64/fpu/s_ceilf.S
 delete mode 100644 sysdeps/powerpc/powerpc64/power5+/fpu/s_ceil.S
 delete mode 100644 sysdeps/powerpc/powerpc64/power5+/fpu/s_ceilf.S

-- 
2.17.1

Comments

Gabriel F. T. Gomes April 25, 2019, 1:56 a.m. UTC | #1
On Fri, Mar 29 2019, Adhemerval Zanella wrote:
> 

> The IFUNC organization for powerpc64 is also change to be enabled only

> for powerpc64 and not for powerpc64le.


OK.

> 	* sysdeps/powerpc/fpu/fenv_libc.h (__fesetround_inline_nocheck): New

> 	macro.


I really wish we can get rid of the burden of writing changelogs, but
for now shouldn't this be function, instead of macro?

> +/* Same as __fesetround_inline, however without runtime check to use DFP

> +   mtfsfi syntax or if round is valid.  */


What would be a runtime check to use DFP syntax? Are you referring to
the optional third parameter (W) to mtfsfi?  I don't see such mechanism
in __fesetround_inline, so I'm not sure I understand this
comment/comparison.

> +static inline void

> +__fesetround_inline_nocheck (const int round)

> +{

> +  asm volatile ("mtfsfi 7,%0" : : "i" (round));

> +}

> +


The implementation looks OK.

> +enum round_mode

> +{

> +  CEIL,

> +  FLOOR,

> +  ROUND,

> +  TRUNC,

> +  NEARBYINT,

> +};


OK.  To be used in following patches.

> +static inline void

> +set_fenv_mode (enum round_mode mode)

> +{

> +  int rmode;

> +  switch (mode)

> +  {

> +  case CEIL:  rmode = FE_UPWARD; break;

> +  default:    rmode = FE_TONEAREST; break;

> +  }

> +  __fesetround_inline_nocheck (rmode);

> +}


OK.

> +static inline float

> +round_to_integer_float (enum round_mode mode, float x)

> +{

> +  /* Ensure sNaN input is converted to qNaN.  */

> +  if (__glibc_unlikely (isnan (x)))

> +    return x + x;

> +

> +  if (fabs (x) > 0x1p+23)

> +    return x;

> +

> +  float r = x;

> +

> +  /* Save current FPU rounding mode and inexact state.  */

> +  fenv_t fe = fegetenv_register ();

> +  set_fenv_mode (mode);

> +  if (x > 0.0)

> +    {

> +      r += 0x1p+23;

> +      r -= 0x1p+23;

> +      r = fabs (r);

> +    }

> +  else if (x < 0.0)

> +    {

> +      r -= 0x1p+23;

> +      r += 0x1p+23;

> +      r = -fabs (r);

> +    }

> +  __builtin_mtfsf (0xff, fe);

> +

> +  return r;

> +}


Ok.

> +static inline double

> +round_to_integer_double (enum round_mode mode, double x)

> +{

> +  /* Ensure sNaN input is converted to qNaN.  */

> +  if (__glibc_unlikely (isnan (x)))

> +    return x + x;

> +

> +  if (fabs (x) > 0x1p+52)

> +    return x;

> +

> +  double r = x;

> +

> +  /* Save current FPU rounding mode and inexact state.  */

> +  fenv_t fe = fegetenv_register ();

> +  set_fenv_mode (mode);

> +  if (x > 0.0)

> +    {

> +      r += 0x1p+52;

> +      r -= 0x1p+52;

> +      r = fabs (r);

> +    }

> +  else if (x < 0.0)

> +    {

> +      r -= 0x1p+52;

> +      r += 0x1p+52;

> +      r = -fabs (r);

> +    }

> +  __builtin_mtfsf (0xff, fe);

> +

> +  return r;

> +}


OK.

> +double

> +__ceil (double x)

> +{

> +#ifdef _ARCH_PWR5X

> +  return __builtin_ceil (x);

> +#else

> +  return round_to_integer_double (CEIL, x);

> +#endif


The arch check looks correct.
Adhemerval Zanella Netto April 25, 2019, 9:58 p.m. UTC | #2
On 24/04/2019 22:56, Gabriel F. T. Gomes wrote:
> On Fri, Mar 29 2019, Adhemerval Zanella wrote:

>>

>> The IFUNC organization for powerpc64 is also change to be enabled only

>> for powerpc64 and not for powerpc64le.

> 

> OK.

> 

>> 	* sysdeps/powerpc/fpu/fenv_libc.h (__fesetround_inline_nocheck): New

>> 	macro.

> 

> I really wish we can get rid of the burden of writing changelogs, but

> for now shouldn't this be function, instead of macro?

> 

>> +/* Same as __fesetround_inline, however without runtime check to use DFP

>> +   mtfsfi syntax or if round is valid.  */

> 

> What would be a runtime check to use DFP syntax? Are you referring to

> the optional third parameter (W) to mtfsfi?  I don't see such mechanism

> in __fesetround_inline, so I'm not sure I understand this

> comment/comparison.


The comment is not clear indeed, to use mtfsfi correctly one should
check hwcap for PPC_FEATURE_HAS_DFP as fesetenv_register do.  I changed
comment to:

/* Same as __fesetround_inline, however without runtime check to use DFP
   mtfsfi syntax (as relax_fenv_state) or if round value is valid.  */

> 

>> +static inline void

>> +__fesetround_inline_nocheck (const int round)

>> +{

>> +  asm volatile ("mtfsfi 7,%0" : : "i" (round));

>> +}

>> +

> 

> The implementation looks OK.

> 

>> +enum round_mode

>> +{

>> +  CEIL,

>> +  FLOOR,

>> +  ROUND,

>> +  TRUNC,

>> +  NEARBYINT,

>> +};

> 

> OK.  To be used in following patches.


In fact I think it would be better to add each round_mode on each
patch it implements. I have changed it locally.

> 

>> +static inline void

>> +set_fenv_mode (enum round_mode mode)

>> +{

>> +  int rmode;

>> +  switch (mode)

>> +  {

>> +  case CEIL:  rmode = FE_UPWARD; break;

>> +  default:    rmode = FE_TONEAREST; break;

>> +  }

>> +  __fesetround_inline_nocheck (rmode);

>> +}

> 

> OK.

> 

>> +static inline float

>> +round_to_integer_float (enum round_mode mode, float x)

>> +{

>> +  /* Ensure sNaN input is converted to qNaN.  */

>> +  if (__glibc_unlikely (isnan (x)))

>> +    return x + x;

>> +

>> +  if (fabs (x) > 0x1p+23)

>> +    return x;

>> +

>> +  float r = x;

>> +

>> +  /* Save current FPU rounding mode and inexact state.  */

>> +  fenv_t fe = fegetenv_register ();

>> +  set_fenv_mode (mode);

>> +  if (x > 0.0)

>> +    {

>> +      r += 0x1p+23;

>> +      r -= 0x1p+23;

>> +      r = fabs (r);

>> +    }

>> +  else if (x < 0.0)

>> +    {

>> +      r -= 0x1p+23;

>> +      r += 0x1p+23;

>> +      r = -fabs (r);

>> +    }

>> +  __builtin_mtfsf (0xff, fe);

>> +

>> +  return r;

>> +}

> 

> Ok.

> 

>> +static inline double

>> +round_to_integer_double (enum round_mode mode, double x)

>> +{

>> +  /* Ensure sNaN input is converted to qNaN.  */

>> +  if (__glibc_unlikely (isnan (x)))

>> +    return x + x;

>> +

>> +  if (fabs (x) > 0x1p+52)

>> +    return x;

>> +

>> +  double r = x;

>> +

>> +  /* Save current FPU rounding mode and inexact state.  */

>> +  fenv_t fe = fegetenv_register ();

>> +  set_fenv_mode (mode);

>> +  if (x > 0.0)

>> +    {

>> +      r += 0x1p+52;

>> +      r -= 0x1p+52;

>> +      r = fabs (r);

>> +    }

>> +  else if (x < 0.0)

>> +    {

>> +      r -= 0x1p+52;

>> +      r += 0x1p+52;

>> +      r = -fabs (r);

>> +    }

>> +  __builtin_mtfsf (0xff, fe);

>> +

>> +  return r;

>> +}

> 

> OK.

> 

>> +double

>> +__ceil (double x)

>> +{

>> +#ifdef _ARCH_PWR5X

>> +  return __builtin_ceil (x);

>> +#else

>> +  return round_to_integer_double (CEIL, x);

>> +#endif

> 

> The arch check looks correct.

>
Gabriel F. T. Gomes May 2, 2019, 6:41 p.m. UTC | #3
On Thu, Apr 25 2019, Adhemerval Zanella wrote:
> On 24/04/2019 22:56, Gabriel F. T. Gomes wrote:

> 

> > What would be a runtime check to use DFP syntax? Are you referring to

> > the optional third parameter (W) to mtfsfi?  I don't see such mechanism

> > in __fesetround_inline, so I'm not sure I understand this

> > comment/comparison.

> 

> The comment is not clear indeed, to use mtfsfi correctly one should

> check hwcap for PPC_FEATURE_HAS_DFP as fesetenv_register do.  I changed

> comment to:

> 

> /* Same as __fesetround_inline, however without runtime check to use DFP

>    mtfsfi syntax (as relax_fenv_state) or if round value is valid.  */


Aha, now I get it, thanks.  With this change, the patch looks good to me.

Reviewed-by: Gabriel F. T. Gomes <gabriel@inconstante.eti.br>


> >> +enum round_mode

> >> +{

> >> +  CEIL,

> >> +  FLOOR,

> >> +  ROUND,

> >> +  TRUNC,

> >> +  NEARBYINT,

> >> +};

> > 

> > OK.  To be used in following patches.

> 

> In fact I think it would be better to add each round_mode on each

> patch it implements. I have changed it locally.


OK.
Gabriel F. T. Gomes May 4, 2019, 9:46 p.m. UTC | #4
Hi, Adhemerval,

During the review of the patch for trunc{,f}, I noticed something that
affects this patch, as well as the patches for floor and round.

Before the changes, the POWER5+ version of the float functions perform a
round to float (via a frsp instruction), because this is explicit in the
assembly code, for instance:

> -ENTRY_TOCLESS (__ceilf, 4)

> -	CALL_MCOUNT 0

> -	frip	fp1, fp1	/* The rounding instructions are double.  */

> -	frsp	fp1, fp1	/* But we need to set ooverflow for float.  */

> -	blr

> -	END (__ceilf)


When this assembly function gets replaced with the builtin, the
generated code does not perform the round to float and will not set
ooverflow (as mentioned in the comment above).

  $ objdump -d math/libm.so.6 | grep -E "<ceilf>:" -A 3
  0000000000060690 <ceilf>:
     60690:       90 0b 20 fc     frip    f1,f1
     60694:       20 00 80 4e     blr
          ...

I still don't know it that is actually a problem.  Do you know?  I'm
sorry that I didn't notice this before.
Adhemerval Zanella Netto May 6, 2019, 12:22 p.m. UTC | #5
On 04/05/2019 18:46, Gabriel F. T. Gomes wrote:
> Hi, Adhemerval,

> 

> During the review of the patch for trunc{,f}, I noticed something that

> affects this patch, as well as the patches for floor and round.

> 

> Before the changes, the POWER5+ version of the float functions perform a

> round to float (via a frsp instruction), because this is explicit in the

> assembly code, for instance:

> 

>> -ENTRY_TOCLESS (__ceilf, 4)

>> -	CALL_MCOUNT 0

>> -	frip	fp1, fp1	/* The rounding instructions are double.  */

>> -	frsp	fp1, fp1	/* But we need to set ooverflow for float.  */

>> -	blr

>> -	END (__ceilf)

> 

> When this assembly function gets replaced with the builtin, the

> generated code does not perform the round to float and will not set

> ooverflow (as mentioned in the comment above).

> 

>   $ objdump -d math/libm.so.6 | grep -E "<ceilf>:" -A 3

>   0000000000060690 <ceilf>:

>      60690:       90 0b 20 fc     frip    f1,f1

>      60694:       20 00 80 4e     blr

>           ...

> 

> I still don't know it that is actually a problem.  Do you know?  I'm

> sorry that I didn't notice this before.

> 


My understanding is the frsp was added to align the implementation to 
POSIX 2001 [1] description which contains a possible range error.
However, as described by man-pages [2], this error for IEEE 754
formats can not happen in practice (the maximum value of exponent for
float, 2^127, is larger than maximum value by significand, 2^23).

This was in fact corrected by both C99 and POSIX 2008 [3] by removing 
the possible return error. So, as gcc builtin implements, the frsp
is superfluous. 

[1] http://pubs.opengroup.org/onlinepubs/009695399/functions/ceil.html
[2] http://man7.org/linux/man-pages/man3/ceil.3.html
[3] https://pubs.opengroup.org/onlinepubs/9699919799/functions/ceil.html
Gabriel F. T. Gomes May 9, 2019, 7:56 p.m. UTC | #6
On Mon, May 06 2019, Adhemerval Zanella wrote:
> 

> My understanding is the frsp was added to align the implementation to 

> POSIX 2001 [1] description which contains a possible range error.

> However, as described by man-pages [2], this error for IEEE 754

> formats can not happen in practice (the maximum value of exponent for

> float, 2^127, is larger than maximum value by significand, 2^23).


I see, and the check (if (fabs (x) > 0x1p+23) return x), at the
beginning of the function, guarantees that an overflow is never caused
by the following addition and subtraction (or vice versa) of 2^23.
Thanks for the clarification.

> This was in fact corrected by both C99 and POSIX 2008 [3] by removing 

> the possible return error. So, as gcc builtin implements, the frsp

> is superfluous. 

> 

> [1] http://pubs.opengroup.org/onlinepubs/009695399/functions/ceil.html

> [2] http://man7.org/linux/man-pages/man3/ceil.3.html

> [3] https://pubs.opengroup.org/onlinepubs/9699919799/functions/ceil.html


Thanks for the references, too.
diff mbox series

Patch

diff --git a/sysdeps/powerpc/fpu/fenv_libc.h b/sysdeps/powerpc/fpu/fenv_libc.h
index 67a9c9a51a..01dccb15a8 100644
--- a/sysdeps/powerpc/fpu/fenv_libc.h
+++ b/sysdeps/powerpc/fpu/fenv_libc.h
@@ -98,6 +98,14 @@  __fesetround_inline (int round)
   return 0;
 }
 
+/* Same as __fesetround_inline, however without runtime check to use DFP
+   mtfsfi syntax or if round is valid.  */
+static inline void
+__fesetround_inline_nocheck (const int round)
+{
+  asm volatile ("mtfsfi 7,%0" : : "i" (round));
+}
+
 /* Definitions of all the FPSCR bit numbers */
 enum {
   FPSCR_FX = 0,    /* exception summary */
diff --git a/sysdeps/powerpc/fpu/round_to_integer.h b/sysdeps/powerpc/fpu/round_to_integer.h
new file mode 100644
index 0000000000..417aae6242
--- /dev/null
+++ b/sysdeps/powerpc/fpu/round_to_integer.h
@@ -0,0 +1,109 @@ 
+/* Round to integer generic implementation.
+   Copyright (C) 2019 Free Software Foundation, Inc.
+   This file is part of the GNU C Library
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Library General Public License as
+   published by the Free Software Foundation; either version 2 of the
+   License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Library General Public License for more details.
+
+   You should have received a copy of the GNU Library General Public
+   License along with the GNU C Library; see the file COPYING.LIB.  If
+   not, see <http://www.gnu.org/licenses/>.  */
+
+#ifndef _ROUND_TO_INTEGER_H
+#define _ROUND_TO_INTEGER_H
+
+#include <fenv_private.h>
+
+enum round_mode
+{
+  CEIL,
+  FLOOR,
+  ROUND,
+  TRUNC,
+  NEARBYINT,
+};
+
+static inline void
+set_fenv_mode (enum round_mode mode)
+{
+  int rmode;
+  switch (mode)
+  {
+  case CEIL:  rmode = FE_UPWARD; break;
+  default:    rmode = FE_TONEAREST; break;
+  }
+  __fesetround_inline_nocheck (rmode);
+}
+
+static inline float
+round_to_integer_float (enum round_mode mode, float x)
+{
+  /* Ensure sNaN input is converted to qNaN.  */
+  if (__glibc_unlikely (isnan (x)))
+    return x + x;
+
+  if (fabs (x) > 0x1p+23)
+    return x;
+
+  float r = x;
+
+  /* Save current FPU rounding mode and inexact state.  */
+  fenv_t fe = fegetenv_register ();
+  set_fenv_mode (mode);
+  if (x > 0.0)
+    {
+      r += 0x1p+23;
+      r -= 0x1p+23;
+      r = fabs (r);
+    }
+  else if (x < 0.0)
+    {
+      r -= 0x1p+23;
+      r += 0x1p+23;
+      r = -fabs (r);
+    }
+  __builtin_mtfsf (0xff, fe);
+
+  return r;
+}
+
+static inline double
+round_to_integer_double (enum round_mode mode, double x)
+{
+  /* Ensure sNaN input is converted to qNaN.  */
+  if (__glibc_unlikely (isnan (x)))
+    return x + x;
+
+  if (fabs (x) > 0x1p+52)
+    return x;
+
+  double r = x;
+
+  /* Save current FPU rounding mode and inexact state.  */
+  fenv_t fe = fegetenv_register ();
+  set_fenv_mode (mode);
+  if (x > 0.0)
+    {
+      r += 0x1p+52;
+      r -= 0x1p+52;
+      r = fabs (r);
+    }
+  else if (x < 0.0)
+    {
+      r -= 0x1p+52;
+      r += 0x1p+52;
+      r = -fabs (r);
+    }
+  __builtin_mtfsf (0xff, fe);
+
+  return r;
+}
+
+#endif
diff --git a/sysdeps/powerpc/fpu/s_ceil.c b/sysdeps/powerpc/fpu/s_ceil.c
new file mode 100644
index 0000000000..49008c7af8
--- /dev/null
+++ b/sysdeps/powerpc/fpu/s_ceil.c
@@ -0,0 +1,35 @@ 
+/* Smallest integral value not less than argument.  PowerPC version.
+   Copyright (C) 2019 Free Software Foundation, Inc.
+   This file is part of the GNU C Library
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Library General Public License as
+   published by the Free Software Foundation; either version 2 of the
+   License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Library General Public License for more details.
+
+   You should have received a copy of the GNU Library General Public
+   License along with the GNU C Library; see the file COPYING.LIB.  If
+   not, see <http://www.gnu.org/licenses/>.  */
+
+#define NO_MATH_REDIRECT
+#include <math.h>
+#include <libm-alias-double.h>
+#include <round_to_integer.h>
+
+double
+__ceil (double x)
+{
+#ifdef _ARCH_PWR5X
+  return __builtin_ceil (x);
+#else
+  return round_to_integer_double (CEIL, x);
+#endif
+}
+#ifndef __ceil
+libm_alias_double (__ceil, ceil)
+#endif
diff --git a/sysdeps/powerpc/fpu/s_ceilf.c b/sysdeps/powerpc/fpu/s_ceilf.c
new file mode 100644
index 0000000000..8c86bf30a3
--- /dev/null
+++ b/sysdeps/powerpc/fpu/s_ceilf.c
@@ -0,0 +1,35 @@ 
+/* Smallest integral value not less than argument.  PowerPC version.
+   Copyright (C) 2019 Free Software Foundation, Inc.
+   This file is part of the GNU C Library
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Library General Public License as
+   published by the Free Software Foundation; either version 2 of the
+   License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Library General Public License for more details.
+
+   You should have received a copy of the GNU Library General Public
+   License along with the GNU C Library; see the file COPYING.LIB.  If
+   not, see <http://www.gnu.org/licenses/>.  */
+
+#define NO_MATH_REDIRECT
+#include <math.h>
+#include <libm-alias-float.h>
+#include <round_to_integer.h>
+
+float
+__ceilf (float x)
+{
+#ifdef _ARCH_PWR5X
+  return __builtin_ceilf (x);
+#else
+  return round_to_integer_float (CEIL, x);
+#endif
+}
+#ifndef __ceilf
+libm_alias_float (__ceil, ceil)
+#endif
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_ceil.S b/sysdeps/powerpc/powerpc32/fpu/s_ceil.S
deleted file mode 100644
index 4d0fdad2c1..0000000000
--- a/sysdeps/powerpc/powerpc32/fpu/s_ceil.S
+++ /dev/null
@@ -1,76 +0,0 @@ 
-/* ceil function.  PowerPC32 version.
-   Copyright (C) 2004-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <math_ldbl_opt.h>
-#include <libm-alias-double.h>
-
-	.section	.rodata.cst4,"aM",@progbits,4
-	.align	2
-.LC0:	/* 2**52 */
-	.long 0x59800000
-
-	.section	".text"
-ENTRY (__ceil)
-#ifdef SHARED
-	mflr	r11
-	cfi_register(lr,r11)
-	SETUP_GOT_ACCESS(r9,got_label)
-	addis	r9,r9,.LC0-got_label@ha
-	lfs	fp13,.LC0-got_label@l(r9)
-	mtlr	r11
-	cfi_same_value (lr)
-#else
-	lis	r9,.LC0@ha
-	lfs	fp13,.LC0@l(r9)
-#endif
-	fabs	fp0,fp1
-	fsub	fp12,fp13,fp13	/* generate 0.0  */
-	fcmpu	cr7,fp0,fp13	/* if (fabs(x) > TWO52)  */
-	mffs	fp11		/* Save current FPU rounding mode and
-				   "inexact" state.  */
-	fcmpu	cr6,fp1,fp12	/* if (x > 0.0)  */
-	bnl-	cr7,.L10
-	mtfsfi	7,2		/* Set rounding mode toward +inf.  */
-	ble-	cr6,.L4
-	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
-	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-	fabs	fp1,fp1		/* if (x == 0.0)  */
-				/* x = 0.0; */
-	mtfsf	0xff,fp11	/* Restore previous rounding mode and
-				   "inexact" state.  */
-	blr
-.L4:
-	bge-	cr6,.L9		/* if (x < 0.0)  */
-	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
-	fnabs	fp1,fp1		/* if (x == 0.0)  */
-				/* x = -0.0; */
-.L9:
-	mtfsf	0xff,fp11	/* Restore previous rounding mode and
-				   "inexact" state.  */
-	blr
-.L10:
-	/* Ensure sNaN input is converted to qNaN.  */
-	fcmpu	cr7,fp1,fp1
-	beqlr	cr7
-	fadd	fp1,fp1,fp1
-	blr
-	END (__ceil)
-
-libm_alias_double (__ceil, ceil)
diff --git a/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S b/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S
deleted file mode 100644
index c32044afa6..0000000000
--- a/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S
+++ /dev/null
@@ -1,76 +0,0 @@ 
-/* float ceil function.  PowerPC32 version.
-   Copyright (C) 2004-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <libm-alias-float.h>
-
-	.section	.rodata.cst4,"aM",@progbits,4
-	.align	2
-.LC0:	/* 2**23 */
-	.long 0x4b000000
-
-	.section	".text"
-ENTRY (__ceilf)
-#ifdef SHARED
-	mflr	r11
-	cfi_register(lr,r11)
-	SETUP_GOT_ACCESS(r9,got_label)
-	addis	r9,r9,.LC0-got_label@ha
-	lfs	fp13,.LC0-got_label@l(r9)
-	mtlr	r11
-	cfi_same_value (lr)
-#else
-	lis	r9,.LC0@ha
-	lfs	fp13,.LC0@l(r9)
-#endif
-	fabs	fp0,fp1
-	fsubs	fp12,fp13,fp13	/* generate 0.0  */
-	fcmpu	cr7,fp0,fp13	/* if (fabs(x) > TWO23)  */
-	mffs	fp11		/* Save current FPU rounding mode and
-				   "inexact" state.  */
-	fcmpu	cr6,fp1,fp12	/* if (x > 0.0)  */
-	bnl-	cr7,.L10
-	mtfsfi	7,2		/* Set rounding mode toward +inf.  */
-	ble-	cr6,.L4
-	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
-	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-	fabs	fp1,fp1		/* if (x == 0.0)  */
-				/* x = 0.0; */
-	mtfsf	0xff,fp11	/* Restore previous rounding mode and
-				   "inexact" state.  */
-	blr
-.L4:
-	bge-	cr6,.L9		/* if (x < 0.0)  */
-	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
-	fnabs	fp1,fp1		/* if (x == 0.0)  */
-				/* x = -0.0; */
-.L9:
-	mtfsf	0xff,fp11	/* Restore previous rounding mode and
-				   "inexact" state.  */
-	blr
-.L10:
-	/* Ensure sNaN input is converted to qNaN.  */
-	fcmpu	cr7,fp1,fp1
-	beqlr	cr7
-	fadds	fp1,fp1,fp1
-	blr
-	END (__ceilf)
-
-libm_alias_float (__ceil, ceil)
-
diff --git a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/Makefile b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/Makefile
index 4e85021d50..cf38e347f2 100644
--- a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/Makefile
+++ b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/Makefile
@@ -26,6 +26,8 @@  libm-sysdep_routines += s_llrintf-power6 s_llrintf-ppc32 s_llrint-power6 \
 			s_logbf-power7 s_logbf-ppc32 e_hypot-power7 \
 			e_hypot-ppc32 e_hypotf-power7 e_hypotf-ppc32
 
+CFLAGS-s_ceil-power5+.c = -mcpu=power5+
+CFLAGS-s_ceilf-power5+.c = -mcpu=power5+
 CFLAGS-s_modf-power5+.c = -mcpu=power5+
 CFLAGS-s_modff-power5+.c = -mcpu=power5+
 CFLAGS-s_logbl-power7.c = -mcpu=power7
diff --git a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-power5+.S b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-power5+.S
deleted file mode 100644
index b13a56667e..0000000000
--- a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-power5+.S
+++ /dev/null
@@ -1,33 +0,0 @@ 
-/* ceil function.  PowerPC32/power5+ version.
-   Copyright (C) 2013-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <math_ldbl_opt.h>
-
-#undef hidden_def
-#define hidden_def(name)
-#undef weak_alias
-#define weak_alias(name, alias)
-#undef strong_alias
-#define strong_alias(name, alias)
-#undef compat_symbol
-#define compat_symbol(lib, name, alias, ver)
-
-#define __ceil __ceil_power5plus
-
-#include <sysdeps/powerpc/powerpc32/power5+/fpu/s_ceil.S>
diff --git a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-power5+.c b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-power5+.c
new file mode 100644
index 0000000000..87bc66cdb0
--- /dev/null
+++ b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-power5+.c
@@ -0,0 +1,3 @@ 
+#include <math.h>
+#define __ceil __ceil_power5plus
+#include <sysdeps/powerpc/fpu/s_ceil.c>
diff --git a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-ppc32.S b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-ppc32.S
deleted file mode 100644
index 899dc140c6..0000000000
--- a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-ppc32.S
+++ /dev/null
@@ -1,31 +0,0 @@ 
-/* ceil function.  PowerPC32 default version.
-   Copyright (C) 2013-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <math_ldbl_opt.h>
-
-#undef weak_alias
-#define weak_alias(a,b)
-#undef strong_alias
-#define strong_alias(a,b)
-#undef compat_symbol
-#define compat_symbol(a,b,c,d)
-
-#define __ceil __ceil_ppc32
-
-#include <sysdeps/powerpc/powerpc32/fpu/s_ceil.S>
diff --git a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-ppc32.c b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-ppc32.c
new file mode 100644
index 0000000000..93c098476b
--- /dev/null
+++ b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceil-ppc32.c
@@ -0,0 +1,3 @@ 
+#include <math.h>
+#define __ceil __ceil_ppc32
+#include <sysdeps/powerpc/fpu/s_ceil.c>
diff --git a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-power5+.S b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-power5+.S
deleted file mode 100644
index 6279638031..0000000000
--- a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-power5+.S
+++ /dev/null
@@ -1,26 +0,0 @@ 
-/* ceilf function.  PowerPC32/power5+ version.
-   Copyright (C) 2013-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-
-#undef weak_alias
-#define weak_alias(name, alias)
-
-#define __ceilf __ceilf_power5plus
-
-#include <sysdeps/powerpc/powerpc32/power5+/fpu/s_ceilf.S>
diff --git a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-power5+.c b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-power5+.c
new file mode 100644
index 0000000000..a5bfa98535
--- /dev/null
+++ b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-power5+.c
@@ -0,0 +1,3 @@ 
+#include <math.h>
+#define __ceilf __ceilf_power5plus
+#include <sysdeps/powerpc/fpu/s_ceilf.c>
diff --git a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-ppc32.S b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-ppc32.S
deleted file mode 100644
index a84edf7a5b..0000000000
--- a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-ppc32.S
+++ /dev/null
@@ -1,27 +0,0 @@ 
-/* ceilf function.  PowerPC32 default version.
-   Copyright (C) 2013-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <math_ldbl_opt.h>
-
-#undef weak_alias
-#define weak_alias(a,b)
-
-#define __ceilf __ceilf_ppc32
-
-#include <sysdeps/powerpc/powerpc32/fpu/s_ceilf.S>
diff --git a/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-ppc32.c b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-ppc32.c
new file mode 100644
index 0000000000..a4dcdcb406
--- /dev/null
+++ b/sysdeps/powerpc/powerpc32/power4/fpu/multiarch/s_ceilf-ppc32.c
@@ -0,0 +1,3 @@ 
+#include <math.h>
+#define __ceilf __ceilf_ppc32
+#include <sysdeps/powerpc/fpu/s_ceilf.c>
diff --git a/sysdeps/powerpc/powerpc32/power5+/fpu/s_ceil.S b/sysdeps/powerpc/powerpc32/power5+/fpu/s_ceil.S
deleted file mode 100644
index 7dab4023f7..0000000000
--- a/sysdeps/powerpc/powerpc32/power5+/fpu/s_ceil.S
+++ /dev/null
@@ -1,29 +0,0 @@ 
-/* ceil function.  PowerPC32/power5+ version.
-   Copyright (C) 2006-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <math_ldbl_opt.h>
-#include <libm-alias-double.h>
-
-	.machine	"power5"
-EALIGN (__ceil, 4, 0)
-	frip	fp1, fp1
-	blr
-	END (__ceil)
-
-libm_alias_double (__ceil, ceil)
diff --git a/sysdeps/powerpc/powerpc32/power5+/fpu/s_ceilf.S b/sysdeps/powerpc/powerpc32/power5+/fpu/s_ceilf.S
deleted file mode 100644
index 15a6f27d28..0000000000
--- a/sysdeps/powerpc/powerpc32/power5+/fpu/s_ceilf.S
+++ /dev/null
@@ -1,30 +0,0 @@ 
-/* ceilf function.  PowerPC32/power5+ version.
-   Copyright (C) 2006-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <libm-alias-float.h>
-
-	.machine	"power5"
-EALIGN (__ceilf, 4, 0)
-	frip	fp1, fp1	/* The rounding instructions are double.  */
-	frsp	fp1, fp1	/* But we need to set ooverflow for float.  */
-	blr
-	END (__ceilf)
-
-libm_alias_float (__ceil, ceil)
-
diff --git a/sysdeps/powerpc/powerpc64/be/fpu/multiarch/Makefile b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/Makefile
new file mode 100644
index 0000000000..932c3c7e6c
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/Makefile
@@ -0,0 +1,9 @@ 
+ifeq ($(subdir),math)
+libm-sysdep_routines += s_ceil-power5+ \
+			s_ceil-ppc64 \
+			s_ceilf-power5+ \
+			s_ceilf-ppc64
+
+CFLAGS-s_ceil-power5+.c = -mcpu=power5+
+CFLAGS-s_ceilf-power5+.c = -mcpu=power5+
+endif
diff --git a/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil-power5+.c b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil-power5+.c
new file mode 100644
index 0000000000..87bc66cdb0
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil-power5+.c
@@ -0,0 +1,3 @@ 
+#include <math.h>
+#define __ceil __ceil_power5plus
+#include <sysdeps/powerpc/fpu/s_ceil.c>
diff --git a/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil-ppc64.c b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil-ppc64.c
new file mode 100644
index 0000000000..8711ff3229
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil-ppc64.c
@@ -0,0 +1,3 @@ 
+#include <math.h>
+#define __ceil __ceil_ppc64
+#include <sysdeps/powerpc/fpu/s_ceil.c>
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil.c b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil.c
similarity index 95%
rename from sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil.c
rename to sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil.c
index b72f4b1cc0..36976a0404 100644
--- a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil.c
+++ b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceil.c
@@ -18,10 +18,8 @@ 
 
 #define NO_MATH_REDIRECT
 #include <math.h>
-#include <math_ldbl_opt.h>
-#include <shlib-compat.h>
-#include "init-arch.h"
 #include <libm-alias-double.h>
+#include "init-arch.h"
 
 extern __typeof (__ceil) __ceil_ppc64 attribute_hidden;
 extern __typeof (__ceil) __ceil_power5plus attribute_hidden;
diff --git a/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf-power5+.c b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf-power5+.c
new file mode 100644
index 0000000000..a5bfa98535
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf-power5+.c
@@ -0,0 +1,3 @@ 
+#include <math.h>
+#define __ceilf __ceilf_power5plus
+#include <sysdeps/powerpc/fpu/s_ceilf.c>
diff --git a/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf-ppc64.c b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf-ppc64.c
new file mode 100644
index 0000000000..086251dc09
--- /dev/null
+++ b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf-ppc64.c
@@ -0,0 +1,3 @@ 
+#include <math.h>
+#define __ceilf __ceilf_ppc64
+#include <sysdeps/powerpc/fpu/s_ceilf.c>
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf.c b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf.c
similarity index 95%
rename from sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf.c
rename to sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf.c
index 707f07c0bb..34811330d5 100644
--- a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf.c
+++ b/sysdeps/powerpc/powerpc64/be/fpu/multiarch/s_ceilf.c
@@ -18,10 +18,8 @@ 
 
 #define NO_MATH_REDIRECT
 #include <math.h>
-#include <math_ldbl_opt.h>
-#include <shlib-compat.h>
-#include "init-arch.h"
 #include <libm-alias-float.h>
+#include "init-arch.h"
 
 extern __typeof (__ceilf) __ceilf_ppc64 attribute_hidden;
 extern __typeof (__ceilf) __ceilf_power5plus attribute_hidden;
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile b/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile
index 39b557604c..ac6d162f91 100644
--- a/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile
+++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/Makefile
@@ -14,8 +14,7 @@  sysdep_calls := s_copysign-power6 s_copysign-ppc64 \
 
 sysdep_routines += $(sysdep_calls)
 libm-sysdep_routines += s_llround-power6x \
-			s_llround-power5+ s_llround-ppc64 s_ceil-power5+ \
-			s_ceil-ppc64 s_ceilf-power5+ s_ceilf-ppc64 \
+			s_llround-power5+ s_llround-ppc64 \
 			s_floor-power5+ s_floor-ppc64 s_floorf-power5+ \
 			s_floorf-ppc64 s_round-power5+ s_round-ppc64 \
 			s_roundf-power5+ s_roundf-ppc64 s_trunc-power5+ \
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-power5+.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-power5+.S
deleted file mode 100644
index 8782e6378e..0000000000
--- a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-power5+.S
+++ /dev/null
@@ -1,30 +0,0 @@ 
-/* ceil function.  PowerPC64/power5+ version.
-   Copyright (C) 2013-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <shlib-compat.h>
-
-#undef weak_alias
-#define weak_alias(a,b)
-#undef strong_alias
-#define strong_alias(a,b)
-#undef compat_symbol
-#define compat_symbol(a,b,c,d)
-
-#define __ceil __ceil_power5plus
-
-#include <sysdeps/powerpc/powerpc64/power5+/fpu/s_ceil.S>
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-ppc64.S
deleted file mode 100644
index 5d9b30d50d..0000000000
--- a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceil-ppc64.S
+++ /dev/null
@@ -1,30 +0,0 @@ 
-/* ceil function.  PowerPC64 default version.
-   Copyright (C) 2013-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <shlib-compat.h>
-
-#undef weak_alias
-#define weak_alias(a,b)
-#undef strong_alias
-#define strong_alias(a,b)
-#undef compat_symbol
-#define compat_symbol(a,b,c,d)
-
-#define __ceil __ceil_ppc64
-
-#include <sysdeps/powerpc/powerpc64/fpu/s_ceil.S>
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-power5+.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-power5+.S
deleted file mode 100644
index 6c9d18880e..0000000000
--- a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-power5+.S
+++ /dev/null
@@ -1,24 +0,0 @@ 
-/* ceilf function.  PowerPC64/power5+ version.
-   Copyright (C) 2013-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#undef weak_alias
-#define weak_alias(a,b)
-
-#define __ceilf __ceilf_power5plus
-
-#include <sysdeps/powerpc/powerpc64/power5+/fpu/s_ceilf.S>
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-ppc64.S b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-ppc64.S
deleted file mode 100644
index 2aaa8c4e4a..0000000000
--- a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_ceilf-ppc64.S
+++ /dev/null
@@ -1,24 +0,0 @@ 
-/* ceilf function.  PowerPC64 default version.
-   Copyright (C) 2013-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#undef weak_alias
-#define weak_alias(a,b)
-
-#define __ceilf __ceilf_ppc64
-
-#include <sysdeps/powerpc/powerpc64/fpu/s_ceilf.S>
diff --git a/sysdeps/powerpc/powerpc64/fpu/s_ceil.S b/sysdeps/powerpc/powerpc64/fpu/s_ceil.S
deleted file mode 100644
index 3e0a968abe..0000000000
--- a/sysdeps/powerpc/powerpc64/fpu/s_ceil.S
+++ /dev/null
@@ -1,65 +0,0 @@ 
-/* ceil function.  PowerPC64 version.
-   Copyright (C) 2004-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <math_ldbl_opt.h>
-#include <libm-alias-double.h>
-
-	.section	".toc","aw"
-.LC0:	/* 2**52 */
-	.tc FD_43300000_0[TC],0x4330000000000000
-	.section	".text"
-
-ENTRY (__ceil, 4)
-	CALL_MCOUNT 0
-	lfd	fp13,.LC0@toc(2)
-	fabs	fp0,fp1
-	fsub	fp12,fp13,fp13	/* generate 0.0  */
-	fcmpu	cr7,fp0,fp13	/* if (fabs(x) > TWO52)  */
-	mffs	fp11		/* Save current FPU rounding mode and
-				   "inexact" state.  */
-	fcmpu	cr6,fp1,fp12	/* if (x > 0.0)  */
-	bnl-	cr7,.L10
-	mtfsfi	7,2		/* Set rounding mode toward +inf.  */
-	ble-	cr6,.L4
-	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
-	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-	fabs	fp1,fp1		/* if (x == 0.0)  */
-				/* x = 0.0; */
-	mtfsf	0xff,fp11	/* Restore previous rounding mode and
-				   "inexact" state.  */
-	blr
-.L4:
-	bge-	cr6,.L9		/* if (x < 0.0)  */
-	fsub	fp1,fp1,fp13	/* x-= TWO52;  */
-	fadd	fp1,fp1,fp13	/* x+= TWO52;  */
-	fnabs	fp1,fp1		/* if (x == 0.0)  */
-				/* x = -0.0; */
-.L9:
-	mtfsf	0xff,fp11	/* Restore previous rounding mode and
-				   "inexact" state.  */
-	blr
-.L10:
-	/* Ensure sNaN input is converted to qNaN.  */
-	fcmpu	cr7,fp1,fp1
-	beqlr	cr7
-	fadd	fp1,fp1,fp1
-	blr
-	END (__ceil)
-
-libm_alias_double (__ceil, ceil)
diff --git a/sysdeps/powerpc/powerpc64/fpu/s_ceilf.S b/sysdeps/powerpc/powerpc64/fpu/s_ceilf.S
deleted file mode 100644
index a189f9fd58..0000000000
--- a/sysdeps/powerpc/powerpc64/fpu/s_ceilf.S
+++ /dev/null
@@ -1,67 +0,0 @@ 
-/* float ceil function.  PowerPC64 version.
-   Copyright (C) 2004-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <libm-alias-float.h>
-
-	.section	".toc","aw"
-	.p2align 3
-.LC0:	/* 2**23 */
-	.long 0x4b000000
-	.long 0x0
-	.section	".text"
-
-ENTRY (__ceilf, 4)
-	CALL_MCOUNT 0
-	lfs	fp13,.LC0@toc(2)
-	fabs	fp0,fp1
-	fsubs	fp12,fp13,fp13	/* generate 0.0  */
-	fcmpu	cr7,fp0,fp13	/* if (fabs(x) > TWO23)  */
-	mffs	fp11		/* Save current FPU rounding mode and
-				   "inexact" state.  */
-	fcmpu	cr6,fp1,fp12	/* if (x > 0.0)  */
-	bnl-	cr7,.L10
-	mtfsfi	7,2		/* Set rounding mode toward +inf.  */
-	ble-	cr6,.L4
-	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
-	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-	fabs	fp1,fp1		/* if (x == 0.0)  */
-				/* x = 0.0; */
-	mtfsf	0xff,fp11	/* Restore previous rounding mode and
-				   "inexact" state.  */
-	blr
-.L4:
-	bge-	cr6,.L9		/* if (x < 0.0)  */
-	fsubs	fp1,fp1,fp13	/* x-= TWO23;  */
-	fadds	fp1,fp1,fp13	/* x+= TWO23;  */
-	fnabs	fp1,fp1		/* if (x == 0.0)  */
-				/* x = -0.0; */
-.L9:
-	mtfsf	0xff,fp11	/* Restore previous rounding mode and
-				   "inexact" state.  */
-	blr
-.L10:
-	/* Ensure sNaN input is converted to qNaN.  */
-	fcmpu	cr7,fp1,fp1
-	beqlr	cr7
-	fadds	fp1,fp1,fp1
-	blr
-	END (__ceilf)
-
-libm_alias_float (__ceil, ceil)
-
diff --git a/sysdeps/powerpc/powerpc64/power5+/fpu/s_ceil.S b/sysdeps/powerpc/powerpc64/power5+/fpu/s_ceil.S
deleted file mode 100644
index 8feb687697..0000000000
--- a/sysdeps/powerpc/powerpc64/power5+/fpu/s_ceil.S
+++ /dev/null
@@ -1,30 +0,0 @@ 
-/* ceil function.  PowerPC64/power5+ version.
-   Copyright (C) 2006-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <math_ldbl_opt.h>
-#include <libm-alias-double.h>
-
-	.machine	"power5"
-ENTRY_TOCLESS (__ceil, 4)
-	CALL_MCOUNT 0
-	frip	fp1, fp1
-	blr
-	END (__ceil)
-
-libm_alias_double (__ceil, ceil)
diff --git a/sysdeps/powerpc/powerpc64/power5+/fpu/s_ceilf.S b/sysdeps/powerpc/powerpc64/power5+/fpu/s_ceilf.S
deleted file mode 100644
index 11d13abd11..0000000000
--- a/sysdeps/powerpc/powerpc64/power5+/fpu/s_ceilf.S
+++ /dev/null
@@ -1,31 +0,0 @@ 
-/* ceilf function.  PowerPC64/power5+ version.
-   Copyright (C) 2006-2019 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <libm-alias-float.h>
-
-	.machine	"power5"
-ENTRY_TOCLESS (__ceilf, 4)
-	CALL_MCOUNT 0
-	frip	fp1, fp1	/* The rounding instructions are double.  */
-	frsp	fp1, fp1	/* But we need to set ooverflow for float.  */
-	blr
-	END (__ceilf)
-
-libm_alias_float (__ceil, ceil)
-