| Message ID | 20250102143444.3574078-1-adhemerval.zanella@linaro.org |
|---|---|
| State | New |
| Headers | show |
| Series | math: Fix acosf when building with gcc <= 11 | expand |
On Thu, 2 Jan 2025, Adhemerval Zanella wrote: > GCC <= 11 wrongly assumes the rounding is to nearest and performs a > constant folding where it should evaluate since the result is not > exact [1]. > > [1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57245 Note that the problematic folding still happens a few lines below the fixed instance where '+ 0x1p-25' without the 'f' suffix causes the sum to be evaluated in double precision, and then converted to float. With 0x1p-25f, gcc doesn't fold the summation. Alexander
On 02/01/25 11:43, Alexander Monakov wrote: > > On Thu, 2 Jan 2025, Adhemerval Zanella wrote: > >> GCC <= 11 wrongly assumes the rounding is to nearest and performs a >> constant folding where it should evaluate since the result is not >> exact [1]. >> >> [1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57245 > > Note that the problematic folding still happens a few lines below the > fixed instance where '+ 0x1p-25' without the 'f' suffix causes the > sum to be evaluated in double precision, and then converted to float. > With 0x1p-25f, gcc doesn't fold the summation. > > Alexander Yes, I will let Paul fix it and sync with core-math changes. At least this is not triggering any regressions so far, although our current test does not check for an extensive range as core-math tests do.
diff --git a/sysdeps/aarch64/libm-test-ulps b/sysdeps/aarch64/libm-test-ulps index f1ab172f27..190685f893 100644 --- a/sysdeps/aarch64/libm-test-ulps +++ b/sysdeps/aarch64/libm-test-ulps @@ -11,7 +11,6 @@ float: 1 Function: "acos_downward": double: 1 -float: 1 ldouble: 1 Function: "acos_sve": @@ -20,7 +19,6 @@ float: 1 Function: "acos_towardzero": double: 1 -float: 1 ldouble: 1 Function: "acos_upward": diff --git a/sysdeps/i386/fpu/libm-test-ulps b/sysdeps/i386/fpu/libm-test-ulps index 27f0f71b0a..423061e12e 100644 --- a/sysdeps/i386/fpu/libm-test-ulps +++ b/sysdeps/i386/fpu/libm-test-ulps @@ -7,12 +7,10 @@ float128: 1 ldouble: 2 Function: "acos_downward": -float: 1 float128: 1 ldouble: 2 Function: "acos_towardzero": -float: 1 float128: 1 ldouble: 2 diff --git a/sysdeps/i386/i686/fpu/multiarch/libm-test-ulps b/sysdeps/i386/i686/fpu/multiarch/libm-test-ulps index 66733c2d1d..466e68dae0 100644 --- a/sysdeps/i386/i686/fpu/multiarch/libm-test-ulps +++ b/sysdeps/i386/i686/fpu/multiarch/libm-test-ulps @@ -7,12 +7,10 @@ float128: 1 ldouble: 2 Function: "acos_downward": -float: 1 float128: 1 ldouble: 2 Function: "acos_towardzero": -float: 1 float128: 1 ldouble: 2 diff --git a/sysdeps/ieee754/flt-32/e_acosf.c b/sysdeps/ieee754/flt-32/e_acosf.c index cba01221dc..a5a4de4fc2 100644 --- a/sysdeps/ieee754/flt-32/e_acosf.c +++ b/sysdeps/ieee754/flt-32/e_acosf.c @@ -3,7 +3,7 @@ Copyright (c) 2023-2024 Alexei Sibidanov. The original version of this file was copied from the CORE-MATH -project (file src/binary32/acos/acosf.c, revision 61d7bef). +project (file src/binary32/acos/acosf.c, revision 56dd347). Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal @@ -28,6 +28,7 @@ SOFTWARE. #include <math.h> #include <math_private.h> #include <libm-alias-finite.h> +#include <math-barriers.h> #include "math_config.h" static __attribute__ ((noinline)) float @@ -66,7 +67,7 @@ poly12 (double z, const double *c) float __ieee754_acosf (float x) { - const double pi2 = 0x1.921fb54442d18p+0; + double pi2 = 0x1.921fb54442d18p+0; static const double o[] = { 0, 0x1.921fb54442d18p+1 }; double xs = x; double r; @@ -87,7 +88,10 @@ __ieee754_acosf (float x) }; /* Avoid spurious underflow exception. */ if (__glibc_unlikely (ax <= 0x40000000u)) /* |x| < 2^-63 */ - return (float) pi2; + /* GCC <= 11 wrongly assumes the rounding is to nearest and + performs a constant folding here: + https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57245 */ + return math_opt_barrier (pi2); double z = xs; double z2 = z * z; double z4 = z2 * z2; diff --git a/sysdeps/powerpc/fpu/libm-test-ulps b/sysdeps/powerpc/fpu/libm-test-ulps index 89dcef3603..07b91f8534 100644 --- a/sysdeps/powerpc/fpu/libm-test-ulps +++ b/sysdeps/powerpc/fpu/libm-test-ulps @@ -8,13 +8,11 @@ ldouble: 1 Function: "acos_downward": double: 1 -float: 1 float128: 1 ldouble: 3 Function: "acos_towardzero": double: 1 -float: 1 float128: 1 ldouble: 3 diff --git a/sysdeps/x86_64/fpu/libm-test-ulps b/sysdeps/x86_64/fpu/libm-test-ulps index 624186f3f6..ad936ad32f 100644 --- a/sysdeps/x86_64/fpu/libm-test-ulps +++ b/sysdeps/x86_64/fpu/libm-test-ulps @@ -8,13 +8,11 @@ ldouble: 2 Function: "acos_downward": double: 1 -float: 1 float128: 1 ldouble: 2 Function: "acos_towardzero": double: 1 -float: 1 float128: 1 ldouble: 2