Skip to contentSkip to conversation

use gcc builtins for sqrt and fma functions

PATCH In review Vineet Gupta 5 participants 20 messages May 30, 2020 – Jun 2, 2020 Synopsys ARC architecture

lore
Share & tools
lore
#0 lore raw

cover use gcc builtins for sqrt and fma functions

Hi,

This was suggested as part of ARC port review.

Build tested for following targets:

aarch64-linux-gnu arm-linux-gnueabi hppa-linux-gnu x86_64-linux-gnu arm-linux-gnueabihf riscv64-linux-gnu-rv64imac-lp64 riscv64-linux-gnu-rv64imafdc-lp64 powerpc-linux-gnu microblaze-linux-gnu nios2-linux-gnu hppa-linux-gnu

Thx,
-Vineet

Vineet Gupta (5):
ARC/dl-runtime helper macros
iee754: prvoide gcc builtins based generic sqrt functions
iee754: prvoide gcc builtins based generic fma functions aarch/fpu: use generic sqrt, fma functions powerpc/fpu: use generic fma functions

 sysdeps/aarch64/fpu/e_sqrt.c                | 27 --------
 sysdeps/aarch64/fpu/e_sqrtf.c               | 27 --------
 sysdeps/aarch64/fpu/math-use-builtins.h     | 70 +++++++++++++++++++++
 sysdeps/aarch64/fpu/s_fma.c                 | 28 ---------
 sysdeps/aarch64/fpu/s_fmaf.c                | 28 ---------
 sysdeps/arc/dl-runtime.h                    | 42 +++++++++++++
 sysdeps/generic/math-use-builtins.h         |  7 +++
 sysdeps/ieee754/dbl-64/e_sqrt.c             |  6 ++
 sysdeps/ieee754/dbl-64/s_fma.c              |  6 ++
 sysdeps/ieee754/dbl-64/s_fmaf.c             |  6 ++
 sysdeps/ieee754/float128/float128_private.h |  2 +
 sysdeps/ieee754/flt-32/e_sqrtf.c            |  6 ++
 sysdeps/ieee754/ldbl-128/s_fmal.c           |  5 ++
 sysdeps/powerpc/fpu/math-use-builtins.h     | 70 +++++++++++++++++++++
 sysdeps/powerpc/fpu/s_fma.c                 | 27 --------
 sysdeps/powerpc/fpu/s_fmaf.c                | 27 --------
 16 files changed, 220 insertions(+), 164 deletions(-)
 delete mode 100644 sysdeps/aarch64/fpu/e_sqrt.c
 delete mode 100644 sysdeps/aarch64/fpu/e_sqrtf.c
 create mode 100644 sysdeps/aarch64/fpu/math-use-builtins.h
 delete mode 100644 sysdeps/aarch64/fpu/s_fma.c
 delete mode 100644 sysdeps/aarch64/fpu/s_fmaf.c
 create mode 100644 sysdeps/arc/dl-runtime.h
 create mode 100644 sysdeps/powerpc/fpu/math-use-builtins.h
 delete mode 100644 sysdeps/powerpc/fpu/s_fma.c
 delete mode 100644 sysdeps/powerpc/fpu/s_fmaf.c
Signature · 7 lines
2.20.1


_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc
#1 lore raw

PATCH 1/5 ARC/dl-runtime helper macros

This is purely for review purposes to attest the interface defined in prior patch

---
 sysdeps/arc/dl-runtime.h | 42 ++++++++++++++++++++++++++++++++++++++++
 1 file changed, 42 insertions(+)
 create mode 100644 sysdeps/arc/dl-runtime.h
sysdeps/arc/dl-runtime.h new file +42 -0
new file mode 100644
index 000000000000..529d49f5d0a1
@@ -0,0 +1,42 @@
+/* Helpers for On-demand PLT fixup for shared objects. ARC version.
+   Copyright (C) 2017-2020 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
+   <https://www.gnu.org/licenses/>.  */
+
+/* PLT jump into resolver passes PC of PLTn, while _dl_fixup expects the
+   address of corresponding .rela.plt entry.
+
+    - @plt0: runtime pc of first plt entry (DT_PLTGOT)
+    - @pltn: runtime pc of plt entry being resolved
+    - @size: size of .plt.rela entry (unused).  */
+static inline uintptr_t
+reloc_index (uintptr_t plt0, uintptr_t pltn, size_t size)
+{
+  unsigned long int idx = (unsigned long)pltn - (unsigned long)plt0;
+
+  /* PLT trampoline is 16 bytes. */
+  idx /= 16;
+
+  /* Exclude PLT0 and PLT1.  */
+  return idx - 2;
+}
+
+static inline uintptr_t
+reloc_offset (uintptr_t plt0, uintptr_t pltn)
+{
+  size_t sz = sizeof (ElfW(Rela));
+  return reloc_index(plt0, pltn, sz) * sz;
+}
Signature · 7 lines
2.20.1


_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc
#2 lore raw

Re: [PATCH 1/5] ARC/dl-runtime helper macros

Quoted 6 lines · Vineet Gupta

This is purely for review purposes to attest the interface defined in prior patch

---
 sysdeps/arc/dl-runtime.h | 42 ++++++++++++++++++++++++++++++++++++++++
 1 file changed, 42 insertions(+)
 create mode 100644 sysdeps/arc/dl-runtime.h

Oops this was from a different series, please ignore. Don't post patches on friday evening :-(
_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc

#3 lore raw

PATCH 2/5 iee754: prvoide gcc builtins based generic sqrt functions

In reply to Vineet Gupta
---
 sysdeps/generic/math-use-builtins.h | 3 +++
 sysdeps/ieee754/dbl-64/e_sqrt.c     | 6 ++++++
 sysdeps/ieee754/flt-32/e_sqrtf.c    | 6 ++++++
 3 files changed, 15 insertions(+)
sysdeps/generic/math-use-builtins.h +3 -0
index 8a39ef58bc95..fc724c824a17 100644
@@ -60,4 +60,7 @@
 # define USE_COPYSIGNF128_BUILTIN 0
 #endif
 
+#define USE_SQRT_BUILTIN 0
+#define USE_SQRTF_BUILTIN 0
+
 #endif /* math-use-builtins.h */
sysdeps/ieee754/dbl-64/e_sqrt.c +6 -0
index d42a1a4eb6e9..518a8ae5cdaf 100644
@@ -41,6 +41,7 @@
 #include <math_private.h>
 #include <fenv_private.h>
 #include <libm-alias-finite.h>
+#include <math-use-builtins.h>
 
 /*********************************************************************/
 /* An ultimate sqrt routine. Given an IEEE double machine number x   */
@@ -50,6 +51,10 @@
 double
 __ieee754_sqrt (double x)
 {
+#if USE_SQRT_BUILTIN
+  return __builtin_sqrt (x);
+#else
+  /* Use generic implementation.  */
   static const double
     rt0 = 9.99999999859990725855365213134618E-01,
     rt1 = 4.99999999495955425917856814202739E-01,
@@ -138,6 +143,7 @@ __ieee754_sqrt (double x)
 	return (x - x) / (x - x); /* sqrt(-ve)=sNaN */
       return 0x1p-256 * __ieee754_sqrt (x * 0x1p512);
     }
+#endif /* ! USE_SQRT_BUILTIN  */
 }
 #ifndef __ieee754_sqrt
 libm_alias_finite (__ieee754_sqrt, __sqrt)
sysdeps/ieee754/flt-32/e_sqrtf.c +6 -0
index b339444301aa..68fc80e1e1ee 100644
@@ -16,12 +16,17 @@
 #include <math.h>
 #include <math_private.h>
 #include <libm-alias-finite.h>
+#include <math-use-builtins.h>
 
 static	const float	one	= 1.0, tiny=1.0e-30;
 
 float
 __ieee754_sqrtf(float x)
 {
+#if USE_SQRTF_BUILTIN
+	return __builtin_sqrtf (x);
+#else
+	/* Use generic implementation.  */
 	float z;
 	int32_t sign = (int)0x80000000;
 	int32_t ix,s,q,m,t,i;
@@ -83,6 +88,7 @@ __ieee754_sqrtf(float x)
 	ix += (m <<23);
 	SET_FLOAT_WORD(z,ix);
 	return z;
+#endif /* ! USE_SQRTF_BUILTIN  */
 }
 #ifndef __ieee754_sqrtf
 libm_alias_finite (__ieee754_sqrtf, __sqrtf)
Signature · 7 lines
2.20.1


_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Reviewed-by on v1 2/5
#4 lore raw

Re: [PATCH 2/5] iee754: prvoide gcc builtins based generic sqrt functions

On 29/05/2020 23:00, Vineet Gupta wrote:

LGTM, thanks.

Quoted 11 lines
---
 sysdeps/generic/math-use-builtins.h | 3 +++
 sysdeps/ieee754/dbl-64/e_sqrt.c     | 6 ++++++
 sysdeps/ieee754/flt-32/e_sqrtf.c    | 6 ++++++
 3 files changed, 15 insertions(+)
sysdeps/generic/math-use-builtins.h +0 -0
index 8a39ef58bc95..fc724c824a17 100644

sysdeps/generic/math-use-builtins.h line 66

Quoted 7 lines
sysdeps/generic/math-use-builtins.h +3 -0
@@ -60,4 +60,7 @@
 # define USE_COPYSIGNF128_BUILTIN 0
 #endif
 
+#define USE_SQRT_BUILTIN 0
+#define USE_SQRTF_BUILTIN 0
+
 #endif /* math-use-builtins.h */

Ok.

sysdeps/ieee754/dbl-64/e_sqrt.c line 47

Quoted 12 lines
sysdeps/ieee754/dbl-64/e_sqrt.c +1 -0
index d42a1a4eb6e9..518a8ae5cdaf 100644
@@ -41,6 +41,7 @@
 #include <math_private.h>
 #include <fenv_private.h>
 #include <libm-alias-finite.h>
+#include <math-use-builtins.h>
 
 /*********************************************************************/
 /* An ultimate sqrt routine. Given an IEEE double machine number x   */

Ok.

Quoted 12 lines
sysdeps/ieee754/dbl-64/e_sqrt.c +4 -0
@@ -50,6 +51,10 @@
 double
 __ieee754_sqrt (double x)
 {
+#if USE_SQRT_BUILTIN
+  return __builtin_sqrt (x);
+#else
+  /* Use generic implementation.  */
   static const double
     rt0 = 9.99999999859990725855365213134618E-01,
     rt1 = 4.99999999495955425917856814202739E-01,

sysdeps/ieee754/dbl-64/e_sqrt.c __ieee754_sqrt (double x) · line 149

Quoted 7 lines
sysdeps/ieee754/dbl-64/e_sqrt.c +1 -0
@@ -138,6 +143,7 @@ __ieee754_sqrt (double x)
 	return (x - x) / (x - x); /* sqrt(-ve)=sNaN */
       return 0x1p-256 * __ieee754_sqrt (x * 0x1p512);
     }
+#endif /* ! USE_SQRT_BUILTIN  */
 }
 #ifndef __ieee754_sqrt
 libm_alias_finite (__ieee754_sqrt, __sqrt)

Ok.

Quoted 24 lines
sysdeps/ieee754/flt-32/e_sqrtf.c +5 -0
index b339444301aa..68fc80e1e1ee 100644
@@ -16,12 +16,17 @@
 #include <math.h>
 #include <math_private.h>
 #include <libm-alias-finite.h>
+#include <math-use-builtins.h>
 
 static	const float	one	= 1.0, tiny=1.0e-30;
 
 float
 __ieee754_sqrtf(float x)
 {
+#if USE_SQRTF_BUILTIN
+	return __builtin_sqrtf (x);
+#else
+	/* Use generic implementation.  */
 	float z;
 	int32_t sign = (int)0x80000000;
 	int32_t ix,s,q,m,t,i;

sysdeps/ieee754/flt-32/e_sqrtf.c __ieee754_sqrtf(float x) · line 94

Quoted 7 lines
sysdeps/ieee754/flt-32/e_sqrtf.c +1 -0
@@ -83,6 +88,7 @@ __ieee754_sqrtf(float x)
 	ix += (m <<23);
 	SET_FLOAT_WORD(z,ix);
 	return z;
+#endif /* ! USE_SQRTF_BUILTIN  */
 }
 #ifndef __ieee754_sqrtf
 libm_alias_finite (__ieee754_sqrtf, __sqrtf)

Ok.

_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc

#5 lore raw

Re: [PATCH 2/5] iee754: prvoide gcc builtins based generic sqrt functions

Quoted 76 lines · Adhemerval Zanella

On 29/05/2020 23:00, Vineet Gupta wrote:

LGTM, thanks.

Quoted 11 lines
---
 sysdeps/generic/math-use-builtins.h | 3 +++
 sysdeps/ieee754/dbl-64/e_sqrt.c     | 6 ++++++
 sysdeps/ieee754/flt-32/e_sqrtf.c    | 6 ++++++
 3 files changed, 15 insertions(+)
sysdeps/generic/math-use-builtins.h +0 -0
index 8a39ef58bc95..fc724c824a17 100644

sysdeps/generic/math-use-builtins.h line 66

Quoted 7 lines
sysdeps/generic/math-use-builtins.h +3 -0
@@ -60,4 +60,7 @@
 # define USE_COPYSIGNF128_BUILTIN 0
 #endif
 
+#define USE_SQRT_BUILTIN 0
+#define USE_SQRTF_BUILTIN 0
+
 #endif /* math-use-builtins.h */

Ok.

sysdeps/ieee754/dbl-64/e_sqrt.c line 47

Quoted 12 lines
sysdeps/ieee754/dbl-64/e_sqrt.c +1 -0
index d42a1a4eb6e9..518a8ae5cdaf 100644
@@ -41,6 +41,7 @@
 #include <math_private.h>
 #include <fenv_private.h>
 #include <libm-alias-finite.h>
+#include <math-use-builtins.h>
 
 /*********************************************************************/
 /* An ultimate sqrt routine. Given an IEEE double machine number x   */

Ok.

Quoted 12 lines
sysdeps/ieee754/dbl-64/e_sqrt.c +4 -0
@@ -50,6 +51,10 @@
 double
 __ieee754_sqrt (double x)
 {
+#if USE_SQRT_BUILTIN
+  return __builtin_sqrt (x);
+#else
+  /* Use generic implementation.  */
   static const double
     rt0 = 9.99999999859990725855365213134618E-01,
     rt1 = 4.99999999495955425917856814202739E-01,

sysdeps/ieee754/dbl-64/e_sqrt.c __ieee754_sqrt (double x) · line 149

Quoted 7 lines
sysdeps/ieee754/dbl-64/e_sqrt.c +1 -0
@@ -138,6 +143,7 @@ __ieee754_sqrt (double x)
 	return (x - x) / (x - x); /* sqrt(-ve)=sNaN */
       return 0x1p-256 * __ieee754_sqrt (x * 0x1p512);
     }
+#endif /* ! USE_SQRT_BUILTIN  */
 }
 #ifndef __ieee754_sqrt
 libm_alias_finite (__ieee754_sqrt, __sqrt)

Ok.

Quoted 10 lines
sysdeps/ieee754/flt-32/e_sqrtf.c +1 -0
index b339444301aa..68fc80e1e1ee 100644
@@ -16,12 +16,17 @@
 #include <math.h>
 #include <math_private.h>
 #include <libm-alias-finite.h>
+#include <math-use-builtins.h>
 
Quoted 1 line

sysdeps/ieee754/flt-32/e_sqrtf.c line 21

Quoted 1 line
sysdeps/ieee754/flt-32/e_sqrtf.c +0 -0
 static	const float	one	= 1.0, tiny=1.0e-30;

You will need to move this definitions inside the !USE_SQRTF_BUILTIN to avoid defined by not used warnings. Current practice is to just open code the constants and let compiler optimize the constant pool:

sysdeps/ieee754/flt-32/e_sqrtf.c +4 -6
index 68fc80e..d85a041 100644
@@ -18,8 +18,6 @@
 #include <libm-alias-finite.h>
 #include <math-use-builtins.h>
 
-static const float     one     = 1.0, tiny=1.0e-30;
-
 float
 __ieee754_sqrtf(float x)
 {
@@ -75,10 +73,10 @@ __ieee754_sqrtf(float x)
 
     /* use floating add to find out rounding direction */
        if(ix!=0) {
-           z = one-tiny; /* trigger inexact flag */
-           if (z>=one) {
-               z = one+tiny;
-               if (z>one)
+           z = 0x1p0 - 0x1.4484cp-100; /* trigger inexact flag */
+           if (z >= 0x1p0) {
+               z = 0x1p0 + 0x1.4484cp-100;
+               if (z > 0x1p0)
                    q += 2;
                else
                    q += (q&1);
Quoted 14 lines
Quoted 14 lines
sysdeps/ieee754/flt-32/e_sqrtf.c +4 -0
 float
 __ieee754_sqrtf(float x)
 {
+#if USE_SQRTF_BUILTIN
+	return __builtin_sqrtf (x);
+#else
+	/* Use generic implementation.  */
 	float z;
 	int32_t sign = (int)0x80000000;
 	int32_t ix,s,q,m,t,i;
@@ -83,6 +88,7 @@ __ieee754_sqrtf(float x)
 	ix += (m <<23);
 	SET_FLOAT_WORD(z,ix);
Quoted 9 lines

sysdeps/ieee754/flt-32/e_sqrtf.c __ieee754_sqrtf(float x) · line 94

Quoted 6 lines
sysdeps/ieee754/flt-32/e_sqrtf.c +1 -0
 	return z;
+#endif /* ! USE_SQRTF_BUILTIN  */
 }
 #ifndef __ieee754_sqrtf
 libm_alias_finite (__ieee754_sqrtf, __sqrtf)

Ok.

_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc

#6 lore raw

Re: [PATCH 2/5] iee754: prvoide gcc builtins based generic sqrt functions

Quoted 3 lines · Adhemerval Zanella via Libc-alpha

You will need to move this definitions inside the !USE_SQRTF_BUILTIN to avoid defined by not used warnings. Current practice is to just open code the constants and let compiler optimize the constant pool:

Won't it be better to keep the const variable and trust the compiler to subsume it instead of open coding in multiple places. Makes it more readable ?

-Vineet
_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc

#7 lore raw

Re: [PATCH 2/5] iee754: prvoide gcc builtins based generic sqrt functions

Quoted 7 lines · Vineet Gupta
Quoted 3 lines · Adhemerval Zanella via Libc-alpha

You will need to move this definitions inside the !USE_SQRTF_BUILTIN to avoid defined by not used warnings. Current practice is to just open code the constants and let compiler optimize the constant pool:

Won't it be better to keep the const variable and trust the compiler to subsume it instead of open coding in multiple places. Makes it more readable ?

I don't have a strong preference, it is just the recent math optimizations use the constant directly in the hex format instead of trying to factoring them out.

_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc

#10 lore raw

PATCH 3/5 iee754: prvoide gcc builtins based generic fma functions

In reply to Vineet Gupta
---
 sysdeps/generic/math-use-builtins.h         | 4 ++++
 sysdeps/ieee754/dbl-64/s_fma.c              | 6 ++++++
 sysdeps/ieee754/dbl-64/s_fmaf.c             | 6 ++++++
 sysdeps/ieee754/float128/float128_private.h | 2 ++
 sysdeps/ieee754/ldbl-128/s_fmal.c           | 5 +++++
 5 files changed, 23 insertions(+)
sysdeps/generic/math-use-builtins.h +4 -0
index fc724c824a17..9e96807a3370 100644
@@ -63,4 +63,8 @@
 #define USE_SQRT_BUILTIN 0
 #define USE_SQRTF_BUILTIN 0
 
+#define USE_FMA_BUILTIN 0
+#define USE_FMAF_BUILTIN 0
+#define USE_FMAL_BUILTIN 0
+
 #endif /* math-use-builtins.h */
sysdeps/ieee754/dbl-64/s_fma.c +6 -0
index 876df6e78bdc..1e4b2da1511d 100644
@@ -25,6 +25,7 @@
 #include <fenv_private.h>
 #include <libm-alias-double.h>
 #include <tininess.h>
+#include <math-use-builtins.h>
 
 /* This implementation uses rounding to odd to avoid problems with
    double rounding.  See a paper by Boldo and Melquiond:
@@ -33,6 +34,10 @@
 double
 __fma (double x, double y, double z)
 {
+#if USE_FMA_BUILTIN
+  return __builtin_fma (x);
+#else
+  /* Use generic implementation.  */
   union ieee754_double u, v, w;
   int adjust = 0;
   u.d = x;
@@ -292,6 +297,7 @@ __fma (double x, double y, double z)
       v.ieee.mantissa1 |= j;
       return v.d * 0x1p-108;
     }
+#endif /* ! USE_FMA_BUILTIN  */
 }
 #ifndef __fma
 libm_alias_double (__fma, fma)
sysdeps/ieee754/dbl-64/s_fmaf.c +6 -0
index 57329d0a87fe..f15b18262124 100644
@@ -23,6 +23,7 @@
 #include <math-barriers.h>
 #include <fenv_private.h>
 #include <libm-alias-float.h>
+#include <math-use-builtins.h>
 
 /* This implementation relies on double being more than twice as
    precise as float and uses rounding to odd in order to avoid problems
@@ -33,6 +34,10 @@
 float
 __fmaf (float x, float y, float z)
 {
+#if USE_FMAF_BUILTIN
+  return __builtin_fmaf (x);
+#else
+  /* Use generic implementation.  */
   fenv_t env;
 
   /* Multiplication is always exact.  */
@@ -60,6 +65,7 @@ __fmaf (float x, float y, float z)
 
   /* And finally truncation with round to nearest.  */
   return (float) u.d;
+#endif /* ! USE_FMAF_BUILTIN  */
 }
 #ifndef __fmaf
 libm_alias_float (__fma, fma)
sysdeps/ieee754/float128/float128_private.h +2 -0
index f97463d9dc1b..a697a7c29038 100644
@@ -154,6 +154,8 @@
 #define USE_ROUNDL_BUILTIN USE_ROUNDF128_BUILTIN
 #undef USE_COPYSIGNL_BUILTIN
 #define USE_COPYSIGNL_BUILTIN USE_COPYSIGNF128_BUILTIN
+#undef USE_FMAL_BUILTIN
+#define USE_FMAL_BUILTIN USE_FMA128_BUILTIN
 
 /* IEEE function renames.  */
 #define __ieee754_acoshl __ieee754_acoshf128
sysdeps/ieee754/ldbl-128/s_fmal.c +5 -0
index 7475015bcec6..1403734a5aeb 100644
@@ -25,6 +25,7 @@
 #include <math_private.h>
 #include <libm-alias-ldouble.h>
 #include <tininess.h>
+#include <math-use-builtins.h>
 
 /* This implementation uses rounding to odd to avoid problems with
    double rounding.  See a paper by Boldo and Melquiond:
@@ -33,6 +34,9 @@
 _Float128
 __fmal (_Float128 x, _Float128 y, _Float128 z)
 {
+#if USE_FMAL_BUILTIN
+  return __builtin_fmal (x);
+#else
   union ieee854_long_double u, v, w;
   int adjust = 0;
   u.d = x;
@@ -296,5 +300,6 @@ __fmal (_Float128 x, _Float128 y, _Float128 z)
       v.ieee.mantissa3 |= j;
       return v.d * L(0x1p-228);
     }
+#endif /* ! USE_FMAL_BUILTIN  */
 }
 libm_alias_ldouble (__fma, fma)
Signature · 7 lines
2.20.1


_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Reviewed-by on v1 3/5
#11 lore raw

Re: [PATCH 3/5] iee754: prvoide gcc builtins based generic fma functions

On 29/05/2020 23:00, Vineet Gupta wrote:

Ok with the fixes below.

Quoted 13 lines
---
 sysdeps/generic/math-use-builtins.h         | 4 ++++
 sysdeps/ieee754/dbl-64/s_fma.c              | 6 ++++++
 sysdeps/ieee754/dbl-64/s_fmaf.c             | 6 ++++++
 sysdeps/ieee754/float128/float128_private.h | 2 ++
 sysdeps/ieee754/ldbl-128/s_fmal.c           | 5 +++++
 5 files changed, 23 insertions(+)
sysdeps/generic/math-use-builtins.h +0 -0
index fc724c824a17..9e96807a3370 100644

sysdeps/generic/math-use-builtins.h line 70

Quoted 8 lines
sysdeps/generic/math-use-builtins.h +4 -0
@@ -63,4 +63,8 @@
 #define USE_SQRT_BUILTIN 0
 #define USE_SQRTF_BUILTIN 0
 
+#define USE_FMA_BUILTIN 0
+#define USE_FMAF_BUILTIN 0
+#define USE_FMAL_BUILTIN 0
+
 #endif /* math-use-builtins.h */

For float128 support it should also contain a:

  #define USE_FMAF128_BUILTIN
Quoted 13 lines
sysdeps/ieee754/dbl-64/s_fma.c +1 -0
index 876df6e78bdc..1e4b2da1511d 100644
@@ -25,6 +25,7 @@
 #include <fenv_private.h>
 #include <libm-alias-double.h>
 #include <tininess.h>
+#include <math-use-builtins.h>
 
 /* This implementation uses rounding to odd to avoid problems with
    double rounding.  See a paper by Boldo and Melquiond:

sysdeps/ieee754/dbl-64/s_fma.c line 38

Quoted 5 lines
sysdeps/ieee754/dbl-64/s_fma.c +2 -0
@@ -33,6 +34,10 @@
 double
 __fma (double x, double y, double z)
 {
+#if USE_FMA_BUILTIN
+  return __builtin_fma (x);

It should be:

  return __builtin_fma (x, y, z);

Same for float and long double variant.

Quoted 47 lines
sysdeps/ieee754/dbl-64/s_fma.c +3 -0
+#else
+  /* Use generic implementation.  */
   union ieee754_double u, v, w;
   int adjust = 0;
   u.d = x;
@@ -292,6 +297,7 @@ __fma (double x, double y, double z)
       v.ieee.mantissa1 |= j;
       return v.d * 0x1p-108;
     }
+#endif /* ! USE_FMA_BUILTIN  */
 }
 #ifndef __fma
 libm_alias_double (__fma, fma)> diff --git a/sysdeps/ieee754/dbl-64/s_fmaf.c b/sysdeps/ieee754/dbl-64/s_fmaf.c

index 57329d0a87fe..f15b18262124 100644

sysdeps/ieee754/dbl-64/s_fmaf.c +6 -0
@@ -23,6 +23,7 @@
 #include <math-barriers.h>
 #include <fenv_private.h>
 #include <libm-alias-float.h>
+#include <math-use-builtins.h>
 
 /* This implementation relies on double being more than twice as
    precise as float and uses rounding to odd in order to avoid problems
@@ -33,6 +34,10 @@
 float
 __fmaf (float x, float y, float z)
 {
+#if USE_FMAF_BUILTIN
+  return __builtin_fmaf (x);
+#else
+  /* Use generic implementation.  */
   fenv_t env;
 
   /* Multiplication is always exact.  */
@@ -60,6 +65,7 @@ __fmaf (float x, float y, float z)
 
   /* And finally truncation with round to nearest.  */
   return (float) u.d;
+#endif /* ! USE_FMAF_BUILTIN  */
 }
 #ifndef __fmaf
 libm_alias_float (__fma, fma)> diff --git a/sysdeps/ieee754/float128/float128_private.h b/sysdeps/ieee754/float128/float128_private.h

index f97463d9dc1b..a697a7c29038 100644

sysdeps/ieee754/float128/float128_private.h line 161

Quoted 8 lines
sysdeps/ieee754/float128/float128_private.h +2 -0
@@ -154,6 +154,8 @@
 #define USE_ROUNDL_BUILTIN USE_ROUNDF128_BUILTIN
 #undef USE_COPYSIGNL_BUILTIN
 #define USE_COPYSIGNL_BUILTIN USE_COPYSIGNF128_BUILTIN
+#undef USE_FMAL_BUILTIN
+#define USE_FMAL_BUILTIN USE_FMA128_BUILTIN
 
 /* IEEE function renames.  */
 #define __ieee754_acoshl __ieee754_acoshf128

Ok.

Quoted 24 lines
sysdeps/ieee754/ldbl-128/s_fmal.c +4 -0
index 7475015bcec6..1403734a5aeb 100644
@@ -25,6 +25,7 @@
 #include <math_private.h>
 #include <libm-alias-ldouble.h>
 #include <tininess.h>
+#include <math-use-builtins.h>
 
 /* This implementation uses rounding to odd to avoid problems with
    double rounding.  See a paper by Boldo and Melquiond:
@@ -33,6 +34,9 @@
 _Float128
 __fmal (_Float128 x, _Float128 y, _Float128 z)
 {
+#if USE_FMAL_BUILTIN
+  return __builtin_fmal (x);
+#else
   union ieee854_long_double u, v, w;
   int adjust = 0;
   u.d = x;

sysdeps/ieee754/ldbl-128/s_fmal.c __fmal (_Float128 x, _Float128 y, _Float128 z) · line 305

Quoted 6 lines
sysdeps/ieee754/ldbl-128/s_fmal.c +1 -0
@@ -296,5 +300,6 @@ __fmal (_Float128 x, _Float128 y, _Float128 z)
       v.ieee.mantissa3 |= j;
       return v.d * L(0x1p-228);
     }
+#endif /* ! USE_FMAL_BUILTIN  */
 }
 libm_alias_ldouble (__fma, fma)

_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc

#13 lore raw

PATCH 4/5 aarch/fpu: use generic sqrt, fma functions

In reply to Vineet Gupta
---
 sysdeps/aarch64/fpu/e_sqrt.c            | 27 ----------
 sysdeps/aarch64/fpu/e_sqrtf.c           | 27 ----------
 sysdeps/aarch64/fpu/math-use-builtins.h | 70 +++++++++++++++++++++++++
 sysdeps/aarch64/fpu/s_fma.c             | 28 ----------
 sysdeps/aarch64/fpu/s_fmaf.c            | 28 ----------
 5 files changed, 70 insertions(+), 110 deletions(-)
 delete mode 100644 sysdeps/aarch64/fpu/e_sqrt.c
 delete mode 100644 sysdeps/aarch64/fpu/e_sqrtf.c
 create mode 100644 sysdeps/aarch64/fpu/math-use-builtins.h
 delete mode 100644 sysdeps/aarch64/fpu/s_fma.c
 delete mode 100644 sysdeps/aarch64/fpu/s_fmaf.c
sysdeps/aarch64/fpu/e_sqrt.c deleted +0 -27
deleted file mode 100644
index abb67ef7b061..000000000000
@@ -1,27 +0,0 @@
-/* Square root of floating point number.
-   Copyright (C) 2015-2020 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
-   <https://www.gnu.org/licenses/>.  */
-
-#include <math_private.h>
-#include <libm-alias-finite.h>
-
-double
-__ieee754_sqrt (double d)
-{
-  return __builtin_sqrt (d);
-}
-libm_alias_finite (__ieee754_sqrt, __sqrt)
sysdeps/aarch64/fpu/e_sqrtf.c deleted +0 -27
deleted file mode 100644
index 13008a4f45d6..000000000000
@@ -1,27 +0,0 @@
-/* Single-precision floating point square root.
-   Copyright (C) 2015-2020 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
-   <https://www.gnu.org/licenses/>.  */
-
-#include <math_private.h>
-#include <libm-alias-finite.h>
-
-float
-__ieee754_sqrtf (float s)
-{
-  return __builtin_sqrtf (s);
-}
-libm_alias_finite (__ieee754_sqrtf, __sqrtf)
sysdeps/aarch64/fpu/math-use-builtins.h new file +70 -0

70 lines not on this page · raw message

sysdeps/aarch64/fpu/s_fma.c deleted +0 -28
deleted file mode 100644
index 48bb40482dc9..000000000000
@@ -1,28 +0,0 @@
-/* Copyright (C) 1996-2020 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
-   <https://www.gnu.org/licenses/>.  */
-
-#include <math.h>
-#include <libm-alias-double.h>
-
-double
-__fma (double x, double y, double z)
-{
-  return __builtin_fma (x, y, z);
-}
-
-libm_alias_double (__fma, fma)
sysdeps/aarch64/fpu/s_fmaf.c deleted +0 -28
deleted file mode 100644
index 544f32e27aec..000000000000
@@ -1,28 +0,0 @@
-/* Copyright (C) 2011-2020 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
-   <https://www.gnu.org/licenses/>.  */
-
-#include <math.h>
-#include <libm-alias-float.h>
-
-float
-__fmaf (float x, float y, float z)
-{
-  return __builtin_fmaf (x, y, z);
-}
-
-libm_alias_float (__fma, fma)
Signature · 7 lines
2.20.1


_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc
#14 lore raw

Re: [PATCH 4/5] aarch/fpu: use generic sqrt, fma functions

Quoted 1 line · Vineet Gupta

LGTM, some comments below.

Quoted 36 lines
---
 sysdeps/aarch64/fpu/e_sqrt.c            | 27 ----------
 sysdeps/aarch64/fpu/e_sqrtf.c           | 27 ----------
 sysdeps/aarch64/fpu/math-use-builtins.h | 70 +++++++++++++++++++++++++
 sysdeps/aarch64/fpu/s_fma.c             | 28 ----------
 sysdeps/aarch64/fpu/s_fmaf.c            | 28 ----------
 5 files changed, 70 insertions(+), 110 deletions(-)
 delete mode 100644 sysdeps/aarch64/fpu/e_sqrt.c
 delete mode 100644 sysdeps/aarch64/fpu/e_sqrtf.c
 create mode 100644 sysdeps/aarch64/fpu/math-use-builtins.h
 delete mode 100644 sysdeps/aarch64/fpu/s_fma.c
 delete mode 100644 sysdeps/aarch64/fpu/s_fmaf.c
sysdeps/aarch64/fpu/e_sqrt.c deleted +0 -17
deleted file mode 100644
index abb67ef7b061..000000000000
@@ -1,27 +0,0 @@
-/* Square root of floating point number.
-   Copyright (C) 2015-2020 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
-   <https://www.gnu.org/licenses/>.  */

sysdeps/aarch64/fpu/e_sqrt.c line 27

Quoted 10 lines
sysdeps/aarch64/fpu/e_sqrt.c deleted +0 -10
-
-#include <math_private.h>
-#include <libm-alias-finite.h>
-
-double
-__ieee754_sqrt (double d)
-{
-  return __builtin_sqrt (d);
-}
-libm_alias_finite (__ieee754_sqrt, __sqrt)

Ok.

Quoted 23 lines
sysdeps/aarch64/fpu/e_sqrtf.c deleted +0 -17
deleted file mode 100644
index 13008a4f45d6..000000000000
@@ -1,27 +0,0 @@
-/* Single-precision floating point square root.
-   Copyright (C) 2015-2020 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
-   <https://www.gnu.org/licenses/>.  */

sysdeps/aarch64/fpu/e_sqrtf.c line 27

Quoted 10 lines
sysdeps/aarch64/fpu/e_sqrtf.c deleted +0 -10
-
-#include <math_private.h>
-#include <libm-alias-finite.h>
-
-float
-__ieee754_sqrtf (float s)
-{
-  return __builtin_sqrtf (s);
-}
-libm_alias_finite (__ieee754_sqrtf, __sqrtf)

Ok.

sysdeps/aarch64/fpu/math-use-builtins.h line 2

Quoted 8 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +2 -0
new file mode 100644
index 000000000000..52f0a0dad6dd
@@ -0,0 +1,70 @@
+/* Using math gcc builtins instead of generic implementation.  aarch64 version.
+   Copyright (C) 2019-2020 Free Software Foundation, Inc.

I think it should be just 2020 here.

Quoted 17 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +17 -0
+   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
+   <https://www.gnu.org/licenses/>.  */
+
+#ifndef MATH_USE_BUILTINS_H

sysdeps/aarch64/fpu/math-use-builtins.h line 29

Quoted 10 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +10 -0
+#define MATH_USE_BUILTINS_H	1
+
+#include <features.h> /* For __GNUC_PREREQ.  */
+
+/* Define these macros to 1 to use __builtin_xyz instead of the
+   generic implementation.  */
+#define USE_NEARBYINT_BUILTIN 0
+#define USE_NEARBYINTF_BUILTIN 0
+#define USE_NEARBYINTL_BUILTIN 0
+#define USE_NEARBYINTF128_BUILTIN 0

Since we are adding this new file for aarch64, we could also enable it fo nearbyint{f} and remove sysdeps/aarch64/fpu/s_nearbyint{f}.c as well.

sysdeps/aarch64/fpu/math-use-builtins.h line 34

Quoted 5 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +5 -0
+
+#define USE_RINT_BUILTIN 0
+#define USE_RINTF_BUILTIN 0
+#define USE_RINTL_BUILTIN 0
+#define USE_RINTF128_BUILTIN 0

Ditto.

sysdeps/aarch64/fpu/math-use-builtins.h line 39

Quoted 5 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +5 -0
+
+#define USE_FLOOR_BUILTIN 0
+#define USE_FLOORF_BUILTIN 0
+#define USE_FLOORL_BUILTIN 0
+#define USE_FLOORF128_BUILTIN 0

Ditto.

sysdeps/aarch64/fpu/math-use-builtins.h line 44

Quoted 5 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +5 -0
+
+#define USE_CEIL_BUILTIN 0
+#define USE_CEILF_BUILTIN 0
+#define USE_CEILL_BUILTIN 0
+#define USE_CEILF128_BUILTIN 0

Ditto.

sysdeps/aarch64/fpu/math-use-builtins.h line 49

Quoted 5 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +5 -0
+
+#define USE_TRUNC_BUILTIN 0
+#define USE_TRUNCF_BUILTIN 0
+#define USE_TRUNCL_BUILTIN 0
+#define USE_TRUNCF128_BUILTIN 0

Ditto.

sysdeps/aarch64/fpu/math-use-builtins.h line 54

Quoted 5 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +5 -0
+
+#define USE_ROUND_BUILTIN 0
+#define USE_ROUNDF_BUILTIN 0
+#define USE_ROUNDL_BUILTIN 0
+#define USE_ROUNDF128_BUILTIN 0

Ditto.

sysdeps/aarch64/fpu/math-use-builtins.h line 62

Quoted 8 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +8 -0
+
+#define USE_COPYSIGNL_BUILTIN 1
+#if __GNUC_PREREQ (7, 0)
+# define USE_COPYSIGNF128_BUILTIN 1
+#else
+# define USE_COPYSIGNF128_BUILTIN 0
+#endif
+

It should be described in commit message as well (although generated instruction are essentially the same).

sysdeps/aarch64/fpu/math-use-builtins.h line 70

Quoted 8 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +8 -0
+#define USE_SQRT_BUILTIN 1
+#define USE_SQRTF_BUILTIN 1
+
+#define USE_FMA_BUILTIN 1
+#define USE_FMAF_BUILTIN 1
+#define USE_FMAL_BUILTIN 0
+
+#endif /* math-use-builtins.h */

Ok.

Quoted 24 lines
sysdeps/aarch64/fpu/s_fma.c deleted +0 -18
deleted file mode 100644
index 48bb40482dc9..000000000000
@@ -1,28 +0,0 @@
-/* Copyright (C) 1996-2020 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
-   <https://www.gnu.org/licenses/>.  */
-

sysdeps/aarch64/fpu/s_fma.c line 28

Quoted 10 lines
sysdeps/aarch64/fpu/s_fma.c deleted +0 -10
-#include <math.h>
-#include <libm-alias-double.h>
-
-double
-__fma (double x, double y, double z)
-{
-  return __builtin_fma (x, y, z);
-}
-
-libm_alias_double (__fma, fma)

Ok.

Quoted 25 lines
sysdeps/aarch64/fpu/s_fmaf.c deleted +0 -18
deleted file mode 100644
index 544f32e27aec..000000000000
@@ -1,28 +0,0 @@
-/* Copyright (C) 2011-2020 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
-   <https://www.gnu.org/licenses/>.  */
-

sysdeps/aarch64/fpu/s_fmaf.c line 28

Quoted 10 lines
sysdeps/aarch64/fpu/s_fmaf.c deleted +0 -10
-#include <math.h>
-#include <libm-alias-float.h>
-
-float
-__fmaf (float x, float y, float z)
-{
-  return __builtin_fmaf (x, y, z);
-}
-
-libm_alias_float (__fma, fma)

Ok.

_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc

#15 lore raw

Re: [PATCH 4/5] aarch/fpu: use generic sqrt, fma functions

Quoted 7 lines · Adhemerval Zanella via Libc-alpha
Quoted 1 line · Vineet Gupta

LGTM, some comments below.

Quoted 11 lines

sysdeps/aarch64/fpu/e_sqrt.c line 27

Quoted 9 lines
sysdeps/aarch64/fpu/e_sqrt.c deleted +0 -9
-#include <math_private.h>
-#include <libm-alias-finite.h>
-
-double
-__ieee754_sqrt (double d)
-{
-  return __builtin_sqrt (d);
-}
-libm_alias_finite (__ieee754_sqrt, __sqrt)

Ok.

How is one to test aarch64 port with hard-float. build-many-glibc doesn't have a hf variant and hacking one didn't do the right thing either.

Quoted 10 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 2

Quoted 8 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +2 -0
new file mode 100644
index 000000000000..52f0a0dad6dd
@@ -0,0 +1,70 @@
+/* Using math gcc builtins instead of generic implementation.  aarch64 version.
+   Copyright (C) 2019-2020 Free Software Foundation, Inc.

I think it should be just 2020 here.

Fixed.

Quoted 10 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 29

Quoted 7 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +7 -0
+
+/* Define these macros to 1 to use __builtin_xyz instead of the
+   generic implementation.  */
+#define USE_NEARBYINT_BUILTIN 0
+#define USE_NEARBYINTF_BUILTIN 0
+#define USE_NEARBYINTL_BUILTIN 0
+#define USE_NEARBYINTF128_BUILTIN 0

Since we are adding this new file for aarch64, we could also enable it fo nearbyint{f} and remove sysdeps/aarch64/fpu/s_nearbyint{f}.c as well.

OK, but only

+#define USE_NEARBYINT_BUILTIN 1
+#define USE_NEARBYINTF_BUILTIN 1

The other 2 are not defined currently. If they are, then I prefer they be enabled as a separate commit for bisectability.

Quoted 6 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 34

Quoted 4 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +4 -0
+#define USE_RINT_BUILTIN 0
+#define USE_RINTF_BUILTIN 0
+#define USE_RINTL_BUILTIN 0
+#define USE_RINTF128_BUILTIN 0

Ditto.

OK and again only RINT and RINTF.
Also I don't understand one thing. Both the generic and aarch64 code have this

float
__rintf (float x)
...
libm_alias_float (__rint, rint)

The alias arg 1 __rint seems to lack suffix 'f' ?

Quoted 3 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 37

Quoted 3 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +3 -0
+
+#define USE_FLOOR_BUILTIN 0
+#define USE_FLOORF_BUILTIN 0

Again FLOOR, FLOORF only

Quoted 4 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 39

Quoted 2 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +2 -0
+#define USE_FLOORL_BUILTIN 0
+#define USE_FLOORF128_BUILTIN 0

Ditto.

Ditto

Quoted 7 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 44

Quoted 5 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +5 -0
+
+#define USE_CEIL_BUILTIN 0
+#define USE_CEILF_BUILTIN 0
+#define USE_CEILL_BUILTIN 0
+#define USE_CEILF128_BUILTIN 0

Ditto.

Ditto

Quoted 8 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 49

Quoted 6 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +5 -0
+
+#define USE_TRUNC_BUILTIN 0
+#define USE_TRUNCF_BUILTIN 0
+#define USE_TRUNCL_BUILTIN 0
+#define USE_TRUNCF128_BUILTIN 0

Ditto.

Ditto

Quoted 7 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 54

Quoted 5 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +5 -0
+
+#define USE_ROUND_BUILTIN 0
+#define USE_ROUNDF_BUILTIN 0
+#define USE_ROUNDL_BUILTIN 0
+#define USE_ROUNDF128_BUILTIN 0

Ditto.

Ditto

Quoted 11 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 62

Quoted 8 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +8 -0
+
+#define USE_COPYSIGNL_BUILTIN 1
+#if __GNUC_PREREQ (7, 0)
+# define USE_COPYSIGNF128_BUILTIN 1
+#else
+# define USE_COPYSIGNF128_BUILTIN 0
+#endif
+

It should be described in commit message as well (although generated instruction are essentially the same).

Well this one is even more special as it was already using the same code, except that the math-use-builtins.h was generic vs. aarch64 specific one.

Thx
-Vineet
_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc

#16 lore raw

Re: [PATCH 4/5] aarch/fpu: use generic sqrt, fma functions

Quoted 21 lines · Vineet Gupta via Libc-alpha
Quoted 7 lines · Adhemerval Zanella via Libc-alpha
Quoted 1 line · Vineet Gupta

LGTM, some comments below.

Quoted 11 lines

sysdeps/aarch64/fpu/e_sqrt.c line 27

Quoted 9 lines
sysdeps/aarch64/fpu/e_sqrt.c deleted +0 -9
-#include <math_private.h>
-#include <libm-alias-finite.h>
-
-double
-__ieee754_sqrt (double d)
-{
-  return __builtin_sqrt (d);
-}
-libm_alias_finite (__ieee754_sqrt, __sqrt)

Ok.

Quoted 2 lines

How is one to test aarch64 port with hard-float. build-many-glibc doesn't have a hf variant and hacking one didn't do the right thing either.

AArch64 is hard-float by default : there is no soft-float variant.

regards
Ramana

Quoted 109 lines
Quoted 11 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 2

Quoted 9 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +2 -0
new file mode 100644
index 000000000000..52f0a0dad6dd
@@ -0,0 +1,70 @@
+/* Using math gcc builtins instead of generic implementation.  aarch64 version.
+   Copyright (C) 2019-2020 Free Software Foundation, Inc.

I think it should be just 2020 here.

Fixed.

Quoted 10 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 29

Quoted 7 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +7 -0
+
+/* Define these macros to 1 to use __builtin_xyz instead of the
+   generic implementation.  */
+#define USE_NEARBYINT_BUILTIN 0
+#define USE_NEARBYINTF_BUILTIN 0
+#define USE_NEARBYINTL_BUILTIN 0
+#define USE_NEARBYINTF128_BUILTIN 0

Since we are adding this new file for aarch64, we could also enable it fo nearbyint{f} and remove sysdeps/aarch64/fpu/s_nearbyint{f}.c as well.

OK, but only

(quoted diff) +2 -0
+#define USE_NEARBYINT_BUILTIN 1
+#define USE_NEARBYINTF_BUILTIN 1

The other 2 are not defined currently. If they are, then I prefer they be enabled as a separate commit for bisectability.

Quoted 6 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 34

Quoted 4 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +4 -0
+#define USE_RINT_BUILTIN 0
+#define USE_RINTF_BUILTIN 0
+#define USE_RINTL_BUILTIN 0
+#define USE_RINTF128_BUILTIN 0

Ditto.

OK and again only RINT and RINTF.
Also I don't understand one thing. Both the generic and aarch64 code have this

float
__rintf (float x)
...
libm_alias_float (__rint, rint)

The alias arg 1 __rint seems to lack suffix 'f' ?

Quoted 3 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 37

Quoted 3 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +3 -0
+
+#define USE_FLOOR_BUILTIN 0
+#define USE_FLOORF_BUILTIN 0

Again FLOOR, FLOORF only

Quoted 4 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 39

Quoted 2 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +2 -0
+#define USE_FLOORL_BUILTIN 0
+#define USE_FLOORF128_BUILTIN 0

Ditto.

Ditto

Quoted 7 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 44

Quoted 5 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +5 -0
+
+#define USE_CEIL_BUILTIN 0
+#define USE_CEILF_BUILTIN 0
+#define USE_CEILL_BUILTIN 0
+#define USE_CEILF128_BUILTIN 0

Ditto.

Ditto

Quoted 8 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 49

Quoted 6 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +5 -0
+
+#define USE_TRUNC_BUILTIN 0
+#define USE_TRUNCF_BUILTIN 0
+#define USE_TRUNCL_BUILTIN 0
+#define USE_TRUNCF128_BUILTIN 0

Ditto.

Ditto

Quoted 7 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 54

Quoted 5 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +5 -0
+
+#define USE_ROUND_BUILTIN 0
+#define USE_ROUNDF_BUILTIN 0
+#define USE_ROUNDL_BUILTIN 0
+#define USE_ROUNDF128_BUILTIN 0

Ditto.

Ditto

Quoted 11 lines

sysdeps/aarch64/fpu/math-use-builtins.h line 62

Quoted 8 lines
sysdeps/aarch64/fpu/math-use-builtins.h new file +8 -0
+
+#define USE_COPYSIGNL_BUILTIN 1
+#if __GNUC_PREREQ (7, 0)
+# define USE_COPYSIGNF128_BUILTIN 1
+#else
+# define USE_COPYSIGNF128_BUILTIN 0
+#endif
+

It should be described in commit message as well (although generated instruction are essentially the same).

Quoted 5 lines

Well this one is even more special as it was already using the same code, except that the math-use-builtins.h was generic vs. aarch64 specific one.

Thx
-Vineet

_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc

#17 lore raw

Re: [PATCH 4/5] aarch/fpu: use generic sqrt, fma functions

In reply to Vineet Gupta
Quoted 8 lines · Vineet Gupta via Libc-alpha

Also I don't understand one thing. Both the generic and aarch64 code have this

float
__rintf (float x)
...
libm_alias_float (__rint, rint)

The alias arg 1 __rint seems to lack suffix 'f' ?

Please see the comments in sysdeps/generic/libm-alias-float.h. The arguments are *unsuffixed* names.

Signature · 7 lines
Joseph S. Myers
[email protected]

_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc
#18 lore raw

PATCH 5/5 powerpc/fpu: use generic fma functions

In reply to Vineet Gupta
---
 sysdeps/powerpc/fpu/math-use-builtins.h | 70 +++++++++++++++++++++++++
 sysdeps/powerpc/fpu/s_fma.c             | 27 ----------
 sysdeps/powerpc/fpu/s_fmaf.c            | 27 ----------
 3 files changed, 70 insertions(+), 54 deletions(-)
 create mode 100644 sysdeps/powerpc/fpu/math-use-builtins.h
 delete mode 100644 sysdeps/powerpc/fpu/s_fma.c
 delete mode 100644 sysdeps/powerpc/fpu/s_fmaf.c
sysdeps/powerpc/fpu/math-use-builtins.h new file +70 -0
new file mode 100644
index 000000000000..9bdde66cf1de
@@ -0,0 +1,70 @@
+/* Using math gcc builtins instead of generic implementation.  PowerPC version.
+   Copyright (C) 2019-2020 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
+   <https://www.gnu.org/licenses/>.  */
+
+#ifndef MATH_USE_BUILTINS_H
+#define MATH_USE_BUILTINS_H	1
+
+#include <features.h> /* For __GNUC_PREREQ.  */
+
+/* Define these macros to 1 to use __builtin_xyz instead of the
+   generic implementation.  */
+#define USE_NEARBYINT_BUILTIN 0
+#define USE_NEARBYINTF_BUILTIN 0
+#define USE_NEARBYINTL_BUILTIN 0
+#define USE_NEARBYINTF128_BUILTIN 0
+
+#define USE_RINT_BUILTIN 0
+#define USE_RINTF_BUILTIN 0
+#define USE_RINTL_BUILTIN 0
+#define USE_RINTF128_BUILTIN 0
+
+#define USE_FLOOR_BUILTIN 0
+#define USE_FLOORF_BUILTIN 0
+#define USE_FLOORL_BUILTIN 0
+#define USE_FLOORF128_BUILTIN 0
+
+#define USE_CEIL_BUILTIN 0
+#define USE_CEILF_BUILTIN 0
+#define USE_CEILL_BUILTIN 0
+#define USE_CEILF128_BUILTIN 0
+
+#define USE_TRUNC_BUILTIN 0
+#define USE_TRUNCF_BUILTIN 0
+#define USE_TRUNCL_BUILTIN 0
+#define USE_TRUNCF128_BUILTIN 0
+
+#define USE_ROUND_BUILTIN 0
+#define USE_ROUNDF_BUILTIN 0
+#define USE_ROUNDL_BUILTIN 0
+#define USE_ROUNDF128_BUILTIN 0
+
+#define USE_COPYSIGNL_BUILTIN 1
+#if __GNUC_PREREQ (7, 0)
+# define USE_COPYSIGNF128_BUILTIN 1
+#else
+# define USE_COPYSIGNF128_BUILTIN 0
+#endif
+
+#define USE_SQRT_BUILTIN 0
+#define USE_SQRTF_BUILTIN 0
+
+#define USE_FMA_BUILTIN 1
+#define USE_FMAF_BUILTIN 1
+#define USE_FMAL_BUILTIN 0
+
+#endif /* math-use-builtins.h */
sysdeps/powerpc/fpu/s_fma.c deleted +0 -27
deleted file mode 100644
index 9ddd13253485..000000000000
@@ -1,27 +0,0 @@
-/* Compute x * y + z as ternary operation.  PowerPC version.
-   Copyright (C) 2019-2020 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
-   <https://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <libm-alias-double.h>
-
-double
-__fma (double x, double y, double z)
-{
-  return __builtin_fma (x, y, z);
-}
-libm_alias_double (__fma, fma)
sysdeps/powerpc/fpu/s_fmaf.c deleted +0 -27
deleted file mode 100644
index d132c20125bb..000000000000
@@ -1,27 +0,0 @@
-/* Compute x * y + z as ternary operation.  PowerPC version.
-   Copyright (C) 2019-2020 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
-   <https://www.gnu.org/licenses/>.  */
-
-#include <sysdep.h>
-#include <libm-alias-float.h>
-
-float
-__fmaf (float x, float y, float z)
-{
-  return __builtin_fmaf (x, y, z);
-}
-libm_alias_float (__fma, fma)
Signature · 7 lines
2.20.1


_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc
Reviewed-by on v1 5/5
#19 lore raw

Re: [PATCH 5/5] powerpc/fpu: use generic fma functions

On 29/05/2020 23:00, Vineet Gupta wrote:

LGTM, thanks.

This new addition, along with the aarch64 one, makes me wonder if it would be better to decompose the USE_* defined in multiple files so the architecture adds only the required ones (instead of copy/paste the whole file with the generic ones).

Also for powerpc, maybe a future cleanup would to move the logic to select the builtin to math-use-builtins.h. Something like:

  #ifdef _ARCH_PWR5X
  #  define USE_FLOOR_BUILTIN 1
  #  define USE_FLOORF_BUILTIN 1
  #else
  #  define USE_FLOOR_BUILTIN 0
  #  define USE_FLOORF_BUILTIN 0
  #endif
  #define USE_FLOORL_BUILTIN 0
  #define USE_FLOORF128_BUILTIN 0
Quoted 75 lines
---
 sysdeps/powerpc/fpu/math-use-builtins.h | 70 +++++++++++++++++++++++++
 sysdeps/powerpc/fpu/s_fma.c             | 27 ----------
 sysdeps/powerpc/fpu/s_fmaf.c            | 27 ----------
 3 files changed, 70 insertions(+), 54 deletions(-)
 create mode 100644 sysdeps/powerpc/fpu/math-use-builtins.h
 delete mode 100644 sysdeps/powerpc/fpu/s_fma.c
 delete mode 100644 sysdeps/powerpc/fpu/s_fmaf.c
sysdeps/powerpc/fpu/math-use-builtins.h new file +60 -0
new file mode 100644
index 000000000000..9bdde66cf1de
@@ -0,0 +1,70 @@
+/* Using math gcc builtins instead of generic implementation.  PowerPC version.
+   Copyright (C) 2019-2020 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
+   <https://www.gnu.org/licenses/>.  */
+
+#ifndef MATH_USE_BUILTINS_H
+#define MATH_USE_BUILTINS_H	1
+
+#include <features.h> /* For __GNUC_PREREQ.  */
+
+/* Define these macros to 1 to use __builtin_xyz instead of the
+   generic implementation.  */
+#define USE_NEARBYINT_BUILTIN 0
+#define USE_NEARBYINTF_BUILTIN 0
+#define USE_NEARBYINTL_BUILTIN 0
+#define USE_NEARBYINTF128_BUILTIN 0
+
+#define USE_RINT_BUILTIN 0
+#define USE_RINTF_BUILTIN 0
+#define USE_RINTL_BUILTIN 0
+#define USE_RINTF128_BUILTIN 0
+
+#define USE_FLOOR_BUILTIN 0
+#define USE_FLOORF_BUILTIN 0
+#define USE_FLOORL_BUILTIN 0
+#define USE_FLOORF128_BUILTIN 0
+
+#define USE_CEIL_BUILTIN 0
+#define USE_CEILF_BUILTIN 0
+#define USE_CEILL_BUILTIN 0
+#define USE_CEILF128_BUILTIN 0
+
+#define USE_TRUNC_BUILTIN 0
+#define USE_TRUNCF_BUILTIN 0
+#define USE_TRUNCL_BUILTIN 0
+#define USE_TRUNCF128_BUILTIN 0
+
+#define USE_ROUND_BUILTIN 0
+#define USE_ROUNDF_BUILTIN 0
+#define USE_ROUNDL_BUILTIN 0
+#define USE_ROUNDF128_BUILTIN 0
+
+#define USE_COPYSIGNL_BUILTIN 1
+#if __GNUC_PREREQ (7, 0)
+# define USE_COPYSIGNF128_BUILTIN 1
+#else
+# define USE_COPYSIGNF128_BUILTIN 0

sysdeps/powerpc/fpu/math-use-builtins.h line 70

Quoted 10 lines
sysdeps/powerpc/fpu/math-use-builtins.h new file +10 -0
+#endif
+
+#define USE_SQRT_BUILTIN 0
+#define USE_SQRTF_BUILTIN 0
+
+#define USE_FMA_BUILTIN 1
+#define USE_FMAF_BUILTIN 1
+#define USE_FMAL_BUILTIN 0
+
+#endif /* math-use-builtins.h */

Ok.

Quoted 23 lines
sysdeps/powerpc/fpu/s_fma.c deleted +0 -17
deleted file mode 100644
index 9ddd13253485..000000000000
@@ -1,27 +0,0 @@
-/* Compute x * y + z as ternary operation.  PowerPC version.
-   Copyright (C) 2019-2020 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
-   <https://www.gnu.org/licenses/>.  */

sysdeps/powerpc/fpu/s_fma.c line 27

Quoted 10 lines
sysdeps/powerpc/fpu/s_fma.c deleted +0 -10
-
-#include <sysdep.h>
-#include <libm-alias-double.h>
-
-double
-__fma (double x, double y, double z)
-{
-  return __builtin_fma (x, y, z);
-}
-libm_alias_double (__fma, fma)

Ok.

Quoted 24 lines
sysdeps/powerpc/fpu/s_fmaf.c deleted +0 -17
deleted file mode 100644
index d132c20125bb..000000000000
@@ -1,27 +0,0 @@
-/* Compute x * y + z as ternary operation.  PowerPC version.
-   Copyright (C) 2019-2020 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
-   <https://www.gnu.org/licenses/>.  */

sysdeps/powerpc/fpu/s_fmaf.c line 27

Quoted 10 lines
sysdeps/powerpc/fpu/s_fmaf.c deleted +0 -10
-
-#include <sysdep.h>
-#include <libm-alias-float.h>
-
-float
-__fmaf (float x, float y, float z)
-{
-  return __builtin_fmaf (x, y, z);
-}
-libm_alias_float (__fma, fma)

Ok.

_______________________________________________
linux-snps-arc mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-snps-arc