[PATCH 14/15] math: Use tanpif from CORE-MATH
Paul Zimmermann
Paul.Zimmermann@inria.fr
Mon Feb 3 11:05:27 GMT 2025
I confirm all binary32 inputs are correctly rounded on x86-64 for all rounding
modes.
Paul
> From: Adhemerval Zanella <adhemerval.zanella@linaro.org>
> Cc: DJ Delorie <dj@redhat.com>,
> Joseph Myers <josmyers@redhat.com>,
> Paul Zimmermann <Paul.Zimmermann@inria.fr>,
> Alexei Sibidanov <sibid@uvic.ca>
> Date: Fri, 31 Jan 2025 16:17:18 -0300
>
> The CORE-MATH implementation is correctly rounded (for any rounding mode)
> and shows better performance to the generic tanpif.
>
> The code was adapted to glibc style and to use the definition of
> math_config.h (to handle errno, overflow, and underflow).
>
> Benchtest on x64_64 (Ryzen 9 5900X, gcc 14.2.1), aarch64 (Neoverse-N1,
> gcc 13.3.1), and powerpc (POWER10, gcc 13.2.1):
>
> latency master patched improvement
> x86_64 85.1683 47.7990 43.88%
> x86_64v2 76.8219 41.4679 46.02%
> x86_64v3 73.7775 37.7734 48.80%
> aarch64 (Neoverse) 35.4514 18.0742 49.02%
> power8 22.7604 10.1054 55.60%
> power10 22.1358 9.9553 55.03%
>
> reciprocal-throughput master patched improvement
> x86_64 41.0174 19.4718 52.53%
> x86_64v2 34.8565 11.3761 67.36%
> x86_64v3 34.0325 9.6989 71.50%
> aarch64 (Neoverse) 25.4349 9.2017 63.82%
> power8 13.8626 3.8486 72.24%
> power10 11.7933 3.6420 69.12%
> ---
> SHARED-FILES | 4 +
> sysdeps/aarch64/libm-test-ulps | 4 -
> sysdeps/arc/fpu/libm-test-ulps | 4 -
> sysdeps/arc/nofpu/libm-test-ulps | 1 -
> sysdeps/arm/libm-test-ulps | 4 -
> sysdeps/hppa/fpu/libm-test-ulps | 4 -
> sysdeps/i386/fpu/libm-test-ulps | 4 -
> .../i386/i686/fpu/multiarch/libm-test-ulps | 4 -
> sysdeps/ieee754/flt-32/math_config.h | 25 ++++++
> sysdeps/ieee754/flt-32/s_tanpif.c | 88 +++++++++++++++++++
> sysdeps/loongarch/lp64/libm-test-ulps | 4 -
> sysdeps/mips/mips64/libm-test-ulps | 4 -
> sysdeps/or1k/fpu/libm-test-ulps | 4 -
> sysdeps/or1k/nofpu/libm-test-ulps | 1 -
> sysdeps/powerpc/fpu/libm-test-ulps | 4 -
> sysdeps/powerpc/fpu/math_private.h | 1 +
> sysdeps/riscv/nofpu/libm-test-ulps | 1 -
> sysdeps/riscv/rvd/libm-test-ulps | 4 -
> sysdeps/s390/fpu/libm-test-ulps | 4 -
> sysdeps/sparc/fpu/libm-test-ulps | 4 -
> sysdeps/x86_64/fpu/libm-test-ulps | 4 -
> 21 files changed, 118 insertions(+), 59 deletions(-)
> create mode 100644 sysdeps/ieee754/flt-32/s_tanpif.c
>
> diff --git a/SHARED-FILES b/SHARED-FILES
> index c108f3b308..25ece987f1 100644
> --- a/SHARED-FILES
> +++ b/SHARED-FILES
> @@ -358,3 +358,7 @@ sysdeps/ieee754/flt-32/s_sinpif.c:
> (src/binary32/sinpi/sinpif.c in CORE-MATH)
> - the code was adapted to use glibc code style and internal
> functions to handle errno, overflow, and underflow.
> +sysdeps/ieee754/flt-32/s_tanpif.c:
> + (src/binary32/tanpi/tanpif.c in CORE-MATH)
> + - the code was adapted to use glibc code style and internal
> + functions to handle errno, overflow, and underflow.
> diff --git a/sysdeps/aarch64/libm-test-ulps b/sysdeps/aarch64/libm-test-ulps
> index c6c93aa0e4..3bcd0e5ae4 100644
> --- a/sysdeps/aarch64/libm-test-ulps
> +++ b/sysdeps/aarch64/libm-test-ulps
> @@ -1681,7 +1681,6 @@ ldouble: 3
>
> Function: "tanpi":
> double: 3
> -float: 3
> ldouble: 3
>
> Function: "tanpi_advsimd":
> @@ -1690,7 +1689,6 @@ float: 2
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
> ldouble: 4
>
> Function: "tanpi_sve":
> @@ -1699,12 +1697,10 @@ float: 2
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
> ldouble: 4
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
> ldouble: 4
>
> Function: "tgamma":
> diff --git a/sysdeps/arc/fpu/libm-test-ulps b/sysdeps/arc/fpu/libm-test-ulps
> index 65ebf6f9a0..24e6036954 100644
> --- a/sysdeps/arc/fpu/libm-test-ulps
> +++ b/sysdeps/arc/fpu/libm-test-ulps
> @@ -1137,19 +1137,15 @@ double: 3
>
> Function: "tanpi":
> double: 3
> -float: 3
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
>
> Function: "tgamma":
> double: 9
> diff --git a/sysdeps/arc/nofpu/libm-test-ulps b/sysdeps/arc/nofpu/libm-test-ulps
> index 3ba4f01cbf..6359d6e038 100644
> --- a/sysdeps/arc/nofpu/libm-test-ulps
> +++ b/sysdeps/arc/nofpu/libm-test-ulps
> @@ -271,7 +271,6 @@ double: 2
>
> Function: "tanpi":
> double: 3
> -float: 3
>
> Function: "tgamma":
> double: 9
> diff --git a/sysdeps/arm/libm-test-ulps b/sysdeps/arm/libm-test-ulps
> index f887712d8e..273c54dd4d 100644
> --- a/sysdeps/arm/libm-test-ulps
> +++ b/sysdeps/arm/libm-test-ulps
> @@ -1130,19 +1130,15 @@ double: 3
>
> Function: "tanpi":
> double: 3
> -float: 3
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
>
> Function: "tgamma":
> double: 9
> diff --git a/sysdeps/hppa/fpu/libm-test-ulps b/sysdeps/hppa/fpu/libm-test-ulps
> index 10f7f2ebde..723cb79d12 100644
> --- a/sysdeps/hppa/fpu/libm-test-ulps
> +++ b/sysdeps/hppa/fpu/libm-test-ulps
> @@ -1160,19 +1160,15 @@ double: 3
>
> Function: "tanpi":
> double: 3
> -float: 3
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
>
> Function: "tgamma":
> double: 9
> diff --git a/sysdeps/i386/fpu/libm-test-ulps b/sysdeps/i386/fpu/libm-test-ulps
> index 77aa7155db..8107d2fa2c 100644
> --- a/sysdeps/i386/fpu/libm-test-ulps
> +++ b/sysdeps/i386/fpu/libm-test-ulps
> @@ -1750,25 +1750,21 @@ ldouble: 4
>
> Function: "tanpi":
> double: 3
> -float: 3
> float128: 3
> ldouble: 3
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
> float128: 4
> ldouble: 4
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
> float128: 4
> ldouble: 4
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
> float128: 4
> ldouble: 4
>
> diff --git a/sysdeps/i386/i686/fpu/multiarch/libm-test-ulps b/sysdeps/i386/i686/fpu/multiarch/libm-test-ulps
> index 7168d577d8..b99c50214c 100644
> --- a/sysdeps/i386/i686/fpu/multiarch/libm-test-ulps
> +++ b/sysdeps/i386/i686/fpu/multiarch/libm-test-ulps
> @@ -1755,25 +1755,21 @@ ldouble: 4
>
> Function: "tanpi":
> double: 3
> -float: 3
> float128: 3
> ldouble: 3
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
> float128: 4
> ldouble: 4
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
> float128: 4
> ldouble: 4
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
> float128: 4
> ldouble: 4
>
> diff --git a/sysdeps/ieee754/flt-32/math_config.h b/sysdeps/ieee754/flt-32/math_config.h
> index 035461199c..8d9c8ee3ad 100644
> --- a/sysdeps/ieee754/flt-32/math_config.h
> +++ b/sysdeps/ieee754/flt-32/math_config.h
> @@ -84,6 +84,31 @@ roundeven_finite (double x)
> #endif
> }
>
> +#ifndef ROUNDEVENF_INTRINSICS
> +/* When set, roundevenf_finite will route to the internal roundevenf function. */
> +# define ROUNDEVENF_INTRINSICS 1
> +#endif
> +
> +static inline float
> +roundevenf_finite (float x)
> +{
> + if (!isfinite (x))
> + __builtin_unreachable ();
> +#if ROUNDEVENF_INTRINSICS
> + return roundevenf (x);
> +#else
> + float y = roundf (x);
> + if (fabs (x - y) == 0.5)
> + {
> + union { float f; uint32_t i; } u = {y};
> + union { float f; uint32_t i; } v = {y - copysignf (1.0, x)};
> + if (__builtin_ctzl (v.i) > __builtin_ctzl (u.i))
> + y = v.f;
> + }
> + return y;
> +#endif
> +}
> +
> static inline uint32_t
> asuint (float f)
> {
> diff --git a/sysdeps/ieee754/flt-32/s_tanpif.c b/sysdeps/ieee754/flt-32/s_tanpif.c
> new file mode 100644
> index 0000000000..efbc47b507
> --- /dev/null
> +++ b/sysdeps/ieee754/flt-32/s_tanpif.c
> @@ -0,0 +1,88 @@
> +/* Correctly-rounded tangent of binary32 value for angles in half-revolutions
> +
> +Copyright (c) 2022-2025 Alexei Sibidanov.
> +
> +The original version of this file was copied from the CORE-MATH
> +project (src/binary32/tanpi/tanpif.c, revision 3bbf907).
> +
> +Permission is hereby granted, free of charge, to any person obtaining a copy
> +of this software and associated documentation files (the "Software"), to deal
> +in the Software without restriction, including without limitation the rights
> +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
> +copies of the Software, and to permit persons to whom the Software is
> +furnished to do so, subject to the following conditions:
> +
> +The above copyright notice and this permission notice shall be included in all
> +copies or substantial portions of the Software.
> +
> +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
> +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
> +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
> +SOFTWARE.
> +
> +*/
> +
> +#include <stdint.h>
> +#include <errno.h>
> +#include <libm-alias-float.h>
> +#include "math_config.h"
> +
> +float
> +__tanpif (float x)
> +{
> + uint32_t ix = asuint (x);
> + uint32_t e = ix & (0xff << 23);
> + if (__glibc_unlikely (e > (150 << 23)))
> + {
> + if (e == (0xff << 23))
> + {
> + if (!(ix << 9))
> + return __math_invalidf (x);
> + return x + x; /* nan */
> + }
> + return copysign (0.0f, x);
> + }
> + float x4 = 4.0f * x;
> + float nx4 = roundevenf_finite (x4);
> + float dx4 = x4 - nx4;
> + float ni = roundevenf_finite (x);
> + float zf = x - ni;
> + if (__glibc_unlikely (dx4 == 0.0f))
> + {
> + int k = x4;
> + if (k & 1)
> + return copysignf (1.0f, zf);
> + k &= 7;
> + if (k == 0)
> + return copysignf (0.0f, x);
> + if (k == 4)
> + return -copysignf (0.0f, x);
> + __set_errno (ERANGE);
> + if (k == 2)
> + return 1.0f / 0.0f;
> + if (k == 6)
> + return -1.0f / 0.0f;
> + }
> + ix = asuint (zf);
> + uint32_t a = ix & (~0u >> 1);
> + if (__glibc_unlikely (a == 0x3e933802u))
> + return copysignf (0x1.44cfbap+0f, zf) + copysignf (0x1p-25f, zf);
> + if (__glibc_unlikely (a == 0x38f26685u))
> + return copysignf (0x1.7cc304p-12, zf) + copysignf (0x1p-37f, zf);
> +
> + double z = zf, z2 = z * z;
> +
> + static const double cn[] = { 0x1.921fb54442d19p-1, -0x1.1f458b3e1f8d6p-2,
> + 0x1.68a34bd0b8f6ap-6, -0x1.e4866f7a25f99p-13 };
> + static const double cd[] = { 0x1p+0, -0x1.4b4b98d2df3a7p-1,
> + 0x1.8e9926d2bb901p-4, -0x1.a6f77fd847eep-9 };
> + double z4
> + = z2 * z2,
> + r = (z - z * z2) * ((cn[0] + z2 * cn[1]) + z4 * (cn[2] + z2 * cn[3]))
> + / (((cd[0] + z2 * cd[1]) + z4 * (cd[2] + z2 * cd[3])) * (0.25 - z2));
> + return r;
> +}
> +libm_alias_float (__tanpi, tanpi)
> diff --git a/sysdeps/loongarch/lp64/libm-test-ulps b/sysdeps/loongarch/lp64/libm-test-ulps
> index 4fadba43c2..b4a6a3ba35 100644
> --- a/sysdeps/loongarch/lp64/libm-test-ulps
> +++ b/sysdeps/loongarch/lp64/libm-test-ulps
> @@ -1437,22 +1437,18 @@ ldouble: 3
>
> Function: "tanpi":
> double: 3
> -float: 3
> ldouble: 3
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
> ldouble: 4
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
> ldouble: 4
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
> ldouble: 4
>
> Function: "tgamma":
> diff --git a/sysdeps/mips/mips64/libm-test-ulps b/sysdeps/mips/mips64/libm-test-ulps
> index 5177b54557..3b1c725fae 100644
> --- a/sysdeps/mips/mips64/libm-test-ulps
> +++ b/sysdeps/mips/mips64/libm-test-ulps
> @@ -1449,22 +1449,18 @@ ldouble: 3
>
> Function: "tanpi":
> double: 3
> -float: 3
> ldouble: 3
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
> ldouble: 4
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
> ldouble: 4
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
> ldouble: 4
>
> Function: "tgamma":
> diff --git a/sysdeps/or1k/fpu/libm-test-ulps b/sysdeps/or1k/fpu/libm-test-ulps
> index 1fb4ec57c0..accf30904d 100644
> --- a/sysdeps/or1k/fpu/libm-test-ulps
> +++ b/sysdeps/or1k/fpu/libm-test-ulps
> @@ -1115,19 +1115,15 @@ double: 3
>
> Function: "tanpi":
> double: 3
> -float: 3
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
>
> Function: "tgamma":
> double: 9
> diff --git a/sysdeps/or1k/nofpu/libm-test-ulps b/sysdeps/or1k/nofpu/libm-test-ulps
> index aff11b5148..f3d5604e6a 100644
> --- a/sysdeps/or1k/nofpu/libm-test-ulps
> +++ b/sysdeps/or1k/nofpu/libm-test-ulps
> @@ -1015,7 +1015,6 @@ double: 3
>
> Function: "tanpi":
> double: 3
> -float: 3
>
> Function: "tgamma":
> double: 9
> diff --git a/sysdeps/powerpc/fpu/libm-test-ulps b/sysdeps/powerpc/fpu/libm-test-ulps
> index e59c3e47ef..404d3afc27 100644
> --- a/sysdeps/powerpc/fpu/libm-test-ulps
> +++ b/sysdeps/powerpc/fpu/libm-test-ulps
> @@ -1857,25 +1857,21 @@ ldouble: 6
>
> Function: "tanpi":
> double: 3
> -float: 3
> float128: 2
> ldouble: 2
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
> float128: 4
> ldouble: 8
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
> float128: 4
> ldouble: 8
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
> float128: 4
> ldouble: 8
>
> diff --git a/sysdeps/powerpc/fpu/math_private.h b/sysdeps/powerpc/fpu/math_private.h
> index aace1a8708..7065d276c0 100644
> --- a/sysdeps/powerpc/fpu/math_private.h
> +++ b/sysdeps/powerpc/fpu/math_private.h
> @@ -62,6 +62,7 @@ __ieee754_sqrtf128 (_Float128 __x)
> #ifdef _ARCH_PWR6
> /* ISA 2.03 provides frin/round() and cntlzw/ctznll(). */
> # define ROUNDEVEN_INTRINSICS 0
> +# define ROUNDEVENF_INTRINSICS 0
> #endif
>
> #endif /* _PPC_MATH_PRIVATE_H_ */
> diff --git a/sysdeps/riscv/nofpu/libm-test-ulps b/sysdeps/riscv/nofpu/libm-test-ulps
> index 2545d0e166..720250e208 100644
> --- a/sysdeps/riscv/nofpu/libm-test-ulps
> +++ b/sysdeps/riscv/nofpu/libm-test-ulps
> @@ -1306,7 +1306,6 @@ ldouble: 3
>
> Function: "tanpi":
> double: 3
> -float: 3
> ldouble: 3
>
> Function: "tgamma":
> diff --git a/sysdeps/riscv/rvd/libm-test-ulps b/sysdeps/riscv/rvd/libm-test-ulps
> index 94534a4f80..ee5df22f81 100644
> --- a/sysdeps/riscv/rvd/libm-test-ulps
> +++ b/sysdeps/riscv/rvd/libm-test-ulps
> @@ -1452,22 +1452,18 @@ ldouble: 3
>
> Function: "tanpi":
> double: 3
> -float: 3
> ldouble: 3
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
> ldouble: 4
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
> ldouble: 4
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
> ldouble: 4
>
> Function: "tgamma":
> diff --git a/sysdeps/s390/fpu/libm-test-ulps b/sysdeps/s390/fpu/libm-test-ulps
> index 2c89048b56..1491089e84 100644
> --- a/sysdeps/s390/fpu/libm-test-ulps
> +++ b/sysdeps/s390/fpu/libm-test-ulps
> @@ -1434,22 +1434,18 @@ ldouble: 3
>
> Function: "tanpi":
> double: 3
> -float: 3
> ldouble: 3
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
> ldouble: 4
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
> ldouble: 4
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
> ldouble: 4
>
> Function: "tgamma":
> diff --git a/sysdeps/sparc/fpu/libm-test-ulps b/sysdeps/sparc/fpu/libm-test-ulps
> index 3af2355545..d894901bbd 100644
> --- a/sysdeps/sparc/fpu/libm-test-ulps
> +++ b/sysdeps/sparc/fpu/libm-test-ulps
> @@ -1449,22 +1449,18 @@ ldouble: 3
>
> Function: "tanpi":
> double: 3
> -float: 3
> ldouble: 3
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
> ldouble: 4
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
> ldouble: 4
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
> ldouble: 4
>
> Function: "tgamma":
> diff --git a/sysdeps/x86_64/fpu/libm-test-ulps b/sysdeps/x86_64/fpu/libm-test-ulps
> index f6da5ba186..a4bd2edcbc 100644
> --- a/sysdeps/x86_64/fpu/libm-test-ulps
> +++ b/sysdeps/x86_64/fpu/libm-test-ulps
> @@ -2292,25 +2292,21 @@ double: 1
>
> Function: "tanpi":
> double: 3
> -float: 3
> float128: 3
> ldouble: 3
>
> Function: "tanpi_downward":
> double: 2
> -float: 3
> float128: 4
> ldouble: 4
>
> Function: "tanpi_towardzero":
> double: 2
> -float: 3
> float128: 4
> ldouble: 4
>
> Function: "tanpi_upward":
> double: 2
> -float: 4
> float128: 4
> ldouble: 4
>
> --
> 2.43.0
>
>
More information about the Libc-alpha
mailing list