From: Adhemerval Zanella Date: Mon, 21 Apr 2025 16:45:31 +0000 (-0300) Subject: math: Optimize float ilogb/llogb X-Git-Tag: glibc-2.42~179 X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=39775f00b194e5f10b3b398de57b04a18846814e;p=thirdparty%2Fglibc.git math: Optimize float ilogb/llogb It removes the wrapper by moving the error/EDOM handling to an out-of-line implementation (__math_invalidf_i/__math_invalidf_li). Also, __glibc_unlikely is used on errors case since it helps code generation on recent gcc. The code now builds to with gcc-14 on aarch64: 0000000000000000 <__ilogbf>: 0: 1e260000 fmov w0, s0 4: d3577801 ubfx x1, x0, #23, #8 8: 340000e1 cbz w1, 24 <__ilogbf+0x24> c: 5101fc20 sub w0, w1, #0x7f 10: 7103fc3f cmp w1, #0xff 14: 54000040 b.eq 1c <__ilogbf+0x1c> // b.none 18: d65f03c0 ret 1c: 12b00000 mov w0, #0x7fffffff // #2147483647 20: 14000000 b 0 <__math_invalidf_i> 24: 53175800 lsl w0, w0, #9 28: 340000a0 cbz w0, 3c <__ilogbf+0x3c> 2c: 5ac01000 clz w0, w0 30: 12800fc1 mov w1, #0xffffff81 // #-127 34: 4b000020 sub w0, w1, w0 38: d65f03c0 ret 3c: 320107e0 mov w0, #0x80000001 // #-2147483647 40: 14000000 b 0 <__math_invalidf_i> Some ABI requires additional adjustments: * i386 and m68k requires to use the template version, since both provide __ieee754_ilogb implementatations. * loongarch uses a custom implementation as well. * powerpc64le also has a custom implementation for POWER9, which is also used for float and float128 version. The generic e_ilogb.c implementation is moved on powerpc to keep the current code as-is. Checked on aarch64-linux-gnu and x86_64-linux-gnu. Reviewed-by: Wilco Dijkstra --- diff --git a/sysdeps/i386/fpu/w_ilogbf.c b/sysdeps/i386/fpu/w_ilogbf.c new file mode 100644 index 0000000000..047ad4bf11 --- /dev/null +++ b/sysdeps/i386/fpu/w_ilogbf.c @@ -0,0 +1,2 @@ +#include +#include diff --git a/sysdeps/i386/fpu/w_llogbf.c b/sysdeps/i386/fpu/w_llogbf.c new file mode 100644 index 0000000000..edb7e9a9e6 --- /dev/null +++ b/sysdeps/i386/fpu/w_llogbf.c @@ -0,0 +1,2 @@ +#include +#include diff --git a/sysdeps/ieee754/flt-32/e_ilogbf.c b/sysdeps/ieee754/flt-32/e_ilogbf.c index 710fb598a8..a27fb94732 100644 --- a/sysdeps/ieee754/flt-32/e_ilogbf.c +++ b/sysdeps/ieee754/flt-32/e_ilogbf.c @@ -1,41 +1 @@ -/* Get integer exponent of a floating-point value. - Copyright (C) 1999-2025 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 - . */ - -#include -#include -#include -#include "math_config.h" - -int -__ieee754_ilogbf (float x) -{ - uint32_t ux = asuint (x); - int ex = (ux & ~SIGN_MASK) >> MANTISSA_WIDTH; - if (ex == 0) /* zero or subnormal */ - { - /* Clear sign and exponent. */ - ux <<= 1 + EXPONENT_WIDTH; - if (ux == 0) - return FP_ILOGB0; - /* subnormal */ - return -127 - stdc_leading_zeros (ux); - } - if (ex == EXPONENT_MASK >> MANTISSA_WIDTH) /* NaN or Inf */ - return ux << (1 + EXPONENT_WIDTH) ? FP_ILOGBNAN : INT_MAX; - return ex - 127; -} +/* ilogbf is implemented at w_ilogbf.c */ diff --git a/sysdeps/ieee754/flt-32/math_config.h b/sysdeps/ieee754/flt-32/math_config.h index 8d9c8ee3ad..fd9481a016 100644 --- a/sysdeps/ieee754/flt-32/math_config.h +++ b/sysdeps/ieee754/flt-32/math_config.h @@ -208,6 +208,8 @@ attribute_hidden float __math_uflowf (uint32_t); attribute_hidden float __math_may_uflowf (uint32_t); attribute_hidden float __math_divzerof (uint32_t); attribute_hidden float __math_invalidf (float); +attribute_hidden int __math_invalidf_i (int); +attribute_hidden long int __math_invalidf_li (long int); attribute_hidden float __math_edomf (float x); /* Shared between expf, exp2f, exp10f, and powf. */ diff --git a/sysdeps/ieee754/flt-32/math_errf.c b/sysdeps/ieee754/flt-32/math_errf.c index edcc4c094a..244e38a4ce 100644 --- a/sysdeps/ieee754/flt-32/math_errf.c +++ b/sysdeps/ieee754/flt-32/math_errf.c @@ -16,6 +16,7 @@ License along with the GNU C Library; if not, see . */ +#include #include "math_config.h" #if WANT_ERRNO @@ -27,8 +28,24 @@ with_errnof (float y, int e) errno = e; return y; } + +NOINLINE static int +with_errnof_i (int y, int e) +{ + errno = e; + return y; +} + +NOINLINE static long int +with_errnof_li (long int y, int e) +{ + errno = e; + return y; +} #else # define with_errnof(x, e) (x) +# define with_errnof_i(x, x) (x) +# define with_errnof_li(x, x) (x) #endif attribute_hidden float @@ -80,3 +97,19 @@ __math_invalidf (float x) float y = (x - x) / (x - x); return isnan (x) ? y : with_errnof (y, EDOM); } + +attribute_hidden int +__math_invalidf_i (int x) +{ + float y = 0.0f / 0.0f; + math_force_eval (y); + return with_errnof_i (x, EDOM); +} + +attribute_hidden long int +__math_invalidf_li (long int x) +{ + float y = 0.0f / 0.0f; + math_force_eval (y); + return with_errnof_li (x, EDOM); +} diff --git a/sysdeps/ieee754/flt-32/w_ilogbf-impl.h b/sysdeps/ieee754/flt-32/w_ilogbf-impl.h new file mode 100644 index 0000000000..5aa8bf0934 --- /dev/null +++ b/sysdeps/ieee754/flt-32/w_ilogbf-impl.h @@ -0,0 +1,38 @@ +/* Get integer exponent of a floating-point value. + Copyright (C) 1999-2025 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 + . */ + +static inline RET_TYPE +IMPL_NAME (float x) +{ + uint32_t ux = asuint (x); + int ex = (ux & ~SIGN_MASK) >> MANTISSA_WIDTH; + if (__glibc_unlikely (ex == 0)) + { + /* Zero or subnormal. + Clear sign and exponent. */ + ux <<= 1 + EXPONENT_WIDTH; + if (ux == 0) + return RET_INVALID (RET_LOGB0); + /* subnormal */ + return (RET_TYPE)-127 - stdc_leading_zeros (ux); + } + if (__glibc_unlikely (ex == EXPONENT_MASK >> MANTISSA_WIDTH)) + /* NaN or Inf */ + return RET_INVALID (ux << (1 + EXPONENT_WIDTH) ? RET_LOGBNAN : RET_LOGMAX); + return ex - 127; +} diff --git a/sysdeps/ieee754/flt-32/w_ilogbf.c b/sysdeps/ieee754/flt-32/w_ilogbf.c new file mode 100644 index 0000000000..4e2a707b64 --- /dev/null +++ b/sysdeps/ieee754/flt-32/w_ilogbf.c @@ -0,0 +1,53 @@ +/* Get integer exponent of a floating-point value. + Copyright (C) 1999-2025 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 + . */ + +#include +#include +#include +#include +#include +#include "math_config.h" + +#ifdef DEF_AS_LLOGBF +# define DECL_NAME __llogb +# define FUNC_NAME llogb +# define RET_TYPE long int +# define RET_LOGB0 FP_LLOGB0 +# define RET_LOGBNAN FP_LLOGBNAN +# define RET_LOGMAX LONG_MAX +# define RET_INVALID __math_invalidf_li +#else +# define DECL_NAME __ilogb +# define FUNC_NAME ilogb +# define RET_TYPE int +# define RET_LOGB0 FP_ILOGB0 +# define RET_LOGBNAN FP_ILOGBNAN +# define RET_LOGMAX INT_MAX +# define RET_INVALID __math_invalidf_i +#endif +#define __IMPL_NAME(x,y) x ## _ ## y +#define _IMPL_NAME(x,y) __IMPL_NAME(x,y) +#define IMPL_NAME _IMPL_NAME(FUNC_NAME, impl) +#include + +RET_TYPE +M_DECL_FUNC (DECL_NAME) (float x) +{ + return IMPL_NAME (x); +} +libm_alias_float (DECL_NAME, FUNC_NAME); diff --git a/sysdeps/ieee754/flt-32/w_llogbf.c b/sysdeps/ieee754/flt-32/w_llogbf.c new file mode 100644 index 0000000000..8676434f6e --- /dev/null +++ b/sysdeps/ieee754/flt-32/w_llogbf.c @@ -0,0 +1,2 @@ +#define DEF_AS_LLOGBF +#include "w_ilogbf.c" diff --git a/sysdeps/loongarch/fpu/e_ilogbf.c b/sysdeps/loongarch/fpu/e_ilogbf.c index adced63566..a27fb94732 100644 --- a/sysdeps/loongarch/fpu/e_ilogbf.c +++ b/sysdeps/loongarch/fpu/e_ilogbf.c @@ -1,39 +1 @@ -/* __ieee754_ilogbf(). LoongArch version. - Copyright (C) 2022-2025 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 - . */ - -#define NO_MATH_REDIRECT -#include -#include - -int -__ieee754_ilogbf (float x) -{ - int x_cond; - asm volatile ("fclass.s \t%0, %1" : "=f" (x_cond) : "f" (x)); - - if (__glibc_unlikely (x_cond & _FCLASS_ZERO)) - return FP_ILOGB0; - else if (__glibc_unlikely (x_cond & ( _FCLASS_NAN | _FCLASS_INF))) - return FP_ILOGBNAN; - else - { - asm volatile ("fabs.s \t%0, %1" : "=f" (x) : "f" (x)); - asm volatile ("flogb.s \t%0, %1" : "=f" (x) : "f" (x)); - return x; - } -} +/* ilogbf is implemented at w_ilogbf.c */ diff --git a/sysdeps/loongarch/fpu/w_ilogbf-impl.h b/sysdeps/loongarch/fpu/w_ilogbf-impl.h new file mode 100644 index 0000000000..9cb4172c26 --- /dev/null +++ b/sysdeps/loongarch/fpu/w_ilogbf-impl.h @@ -0,0 +1,37 @@ +/* Get integer exponent of a floating-point value. LoongArch version. + Copyright (C) 2022-2025 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 + . */ + +#include + +static inline RET_TYPE +IMPL_NAME (float x) +{ + int x_cond; + asm volatile ("fclass.s \t%0, %1" : "=f" (x_cond) : "f" (x)); + + if (__glibc_unlikely (x_cond & _FCLASS_ZERO)) + return RET_INVALID (RET_LOGB0); + else if (__glibc_unlikely (x_cond & ( _FCLASS_NAN | _FCLASS_INF))) + return RET_INVALID (RET_LOGBNAN); + else + { + asm volatile ("fabs.s \t%0, %1" : "=f" (x) : "f" (x)); + asm volatile ("flogb.s \t%0, %1" : "=f" (x) : "f" (x)); + return x; + } +} diff --git a/sysdeps/m68k/m680x0/w_ilogbf.c b/sysdeps/m68k/m680x0/w_ilogbf.c new file mode 100644 index 0000000000..047ad4bf11 --- /dev/null +++ b/sysdeps/m68k/m680x0/w_ilogbf.c @@ -0,0 +1,2 @@ +#include +#include diff --git a/sysdeps/m68k/m680x0/w_llogbf.c b/sysdeps/m68k/m680x0/w_llogbf.c new file mode 100644 index 0000000000..edb7e9a9e6 --- /dev/null +++ b/sysdeps/m68k/m680x0/w_llogbf.c @@ -0,0 +1,2 @@ +#include +#include diff --git a/sysdeps/powerpc/powerpc64/le/fpu/e_ilogbf.c b/sysdeps/powerpc/powerpc64/le/fpu/e_ilogbf.c new file mode 100644 index 0000000000..1c2a8a5ade --- /dev/null +++ b/sysdeps/powerpc/powerpc64/le/fpu/e_ilogbf.c @@ -0,0 +1,41 @@ +/* Get integer exponent of a floating-point value. + Copyright (C) 1999-2025 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 + . */ + +#include +#include +#include +#include "sysdeps/ieee754/flt-32/math_config.h" + +int +__ieee754_ilogbf (float x) +{ + uint32_t ux = asuint (x); + int ex = (ux & ~SIGN_MASK) >> MANTISSA_WIDTH; + if (ex == 0) /* zero or subnormal */ + { + /* Clear sign and exponent. */ + ux <<= 1 + EXPONENT_WIDTH; + if (ux == 0) + return FP_ILOGB0; + /* sbunormal */ + return -127 - stdc_leading_zeros (ux); + } + if (ex == EXPONENT_MASK >> MANTISSA_WIDTH) /* NaN or Inf */ + return ux << (1 + EXPONENT_WIDTH) ? FP_ILOGBNAN : INT_MAX; + return ex - 127; +} diff --git a/sysdeps/powerpc/powerpc64/le/fpu/w_ilogbf.c b/sysdeps/powerpc/powerpc64/le/fpu/w_ilogbf.c new file mode 100644 index 0000000000..047ad4bf11 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/le/fpu/w_ilogbf.c @@ -0,0 +1,2 @@ +#include +#include diff --git a/sysdeps/powerpc/powerpc64/le/fpu/w_llogbf.c b/sysdeps/powerpc/powerpc64/le/fpu/w_llogbf.c new file mode 100644 index 0000000000..edb7e9a9e6 --- /dev/null +++ b/sysdeps/powerpc/powerpc64/le/fpu/w_llogbf.c @@ -0,0 +1,2 @@ +#include +#include