diff options
author | Szabolcs Nagy <szabolcs.nagy@arm.com> | 2021-12-31 09:50:50 +0000 |
---|---|---|
committer | Szabolcs Nagy <szabolcs.nagy@arm.com> | 2022-01-10 14:27:17 +0000 |
commit | 347a5b592c91f0230d52a5117fd83e1f75028207 (patch) | |
tree | cef1eb7585b59b17118ad72f82856b39f94cfd14 /sysdeps/ieee754 | |
parent | e72ef23ee88187284b4b1ca9b2e314e618429d35 (diff) | |
download | glibc-347a5b592c91f0230d52a5117fd83e1f75028207.tar.gz glibc-347a5b592c91f0230d52a5117fd83e1f75028207.tar.xz glibc-347a5b592c91f0230d52a5117fd83e1f75028207.zip |
math: Fix float conversion regressions with gcc-12 [BZ #28713]
Converting double precision constants to float is now affected by the runtime dynamic rounding mode instead of being evaluated at compile time with default rounding mode (except static object initializers). This can change the computed result and cause performance regression. The known correctness issues (increased ulp errors) are already fixed, this patch fixes remaining cases of unnecessary runtime conversions. Add float M_* macros to math.h as new GNU extension API. To avoid conversions the new M_* macros are used and instead of casting double literals to float, use float literals (only required if the conversion is inexact). The patch was tested on aarch64 where the following symbols had new spurious conversion instructions that got fixed: __clog10f __gammaf_r_finite@GLIBC_2.17 __j0f_finite@GLIBC_2.17 __j1f_finite@GLIBC_2.17 __jnf_finite@GLIBC_2.17 __kernel_casinhf __lgamma_negf __log1pf __y0f_finite@GLIBC_2.17 __y1f_finite@GLIBC_2.17 cacosf cacoshf casinhf catanf catanhf clogf gammaf_positive Fixes bug 28713. Reviewed-by: Paul Zimmermann <Paul.Zimmermann@inria.fr>
Diffstat (limited to 'sysdeps/ieee754')
-rw-r--r-- | sysdeps/ieee754/flt-32/e_gammaf_r.c | 12 | ||||
-rw-r--r-- | sysdeps/ieee754/flt-32/e_j0f.c | 4 | ||||
-rw-r--r-- | sysdeps/ieee754/flt-32/e_j1f.c | 4 | ||||
-rw-r--r-- | sysdeps/ieee754/flt-32/e_jnf.c | 2 | ||||
-rw-r--r-- | sysdeps/ieee754/flt-32/lgamma_negf.c | 8 | ||||
-rw-r--r-- | sysdeps/ieee754/flt-32/s_log1pf.c | 2 |
6 files changed, 16 insertions, 16 deletions
diff --git a/sysdeps/ieee754/flt-32/e_gammaf_r.c b/sysdeps/ieee754/flt-32/e_gammaf_r.c index 782beaa144..3db8463407 100644 --- a/sysdeps/ieee754/flt-32/e_gammaf_r.c +++ b/sysdeps/ieee754/flt-32/e_gammaf_r.c @@ -85,7 +85,7 @@ gammaf_positive (float x, int *exp2_adj) float x_adj_frac = x_adj - x_adj_int; int x_adj_log2; float x_adj_mant = __frexpf (x_adj, &x_adj_log2); - if (x_adj_mant < (float) M_SQRT1_2) + if (x_adj_mant < M_SQRT1_2f) { x_adj_log2--; x_adj_mant *= 2.0f; @@ -94,7 +94,7 @@ gammaf_positive (float x, int *exp2_adj) float ret = (__ieee754_powf (x_adj_mant, x_adj) * __ieee754_exp2f (x_adj_log2 * x_adj_frac) * __ieee754_expf (-x_adj) - * sqrtf (2 * (float) M_PI / x_adj) + * sqrtf (2 * M_PIf / x_adj) / prod); exp_adj += x_eps * __ieee754_logf (x_adj); float bsum = gamma_coeff[NCOEFF - 1]; @@ -176,11 +176,11 @@ __ieee754_gammaf_r (float x, int *signgamp) if (frac > 0.5f) frac = 1.0f - frac; float sinpix = (frac <= 0.25f - ? __sinf ((float) M_PI * frac) - : __cosf ((float) M_PI * (0.5f - frac))); + ? __sinf (M_PIf * frac) + : __cosf (M_PIf * (0.5f - frac))); int exp2_adj; - float tret = (float) M_PI / (-x * sinpix - * gammaf_positive (-x, &exp2_adj)); + float tret = M_PIf / (-x * sinpix + * gammaf_positive (-x, &exp2_adj)); ret = __scalbnf (tret, -exp2_adj); math_check_force_underflow_nonneg (ret); } diff --git a/sysdeps/ieee754/flt-32/e_j0f.c b/sysdeps/ieee754/flt-32/e_j0f.c index 0453a30109..9ae91a9dcc 100644 --- a/sysdeps/ieee754/flt-32/e_j0f.c +++ b/sysdeps/ieee754/flt-32/e_j0f.c @@ -233,7 +233,7 @@ j0f_near_root (float x, float z) float index_f; int index; - index_f = roundf ((x - FIRST_ZERO_J0) / (float) M_PI); + index_f = roundf ((x - FIRST_ZERO_J0) / M_PIf); /* j0f_asympt fails to give an error <= 9 ulps for x=0x1.324e92p+7 (index 48) thus we can't reduce SMALL_SIZE below 49. */ if (index_f >= SMALL_SIZE) @@ -514,7 +514,7 @@ y0f_near_root (float x, float z) float index_f; int index; - index_f = roundf ((x - FIRST_ZERO_Y0) / (float) M_PI); + index_f = roundf ((x - FIRST_ZERO_Y0) / M_PIf); if (index_f >= SMALL_SIZE) return y0f_asympt (x); index = (int) index_f; diff --git a/sysdeps/ieee754/flt-32/e_j1f.c b/sysdeps/ieee754/flt-32/e_j1f.c index 052c661db9..ade10402fd 100644 --- a/sysdeps/ieee754/flt-32/e_j1f.c +++ b/sysdeps/ieee754/flt-32/e_j1f.c @@ -243,7 +243,7 @@ j1f_near_root (float x, float z) x = -x; sign = -1.0f; } - index_f = roundf ((x - FIRST_ZERO_J1) / (float) M_PI); + index_f = roundf ((x - FIRST_ZERO_J1) / M_PIf); if (index_f >= SMALL_SIZE) return sign * j1f_asympt (x); index = (int) index_f; @@ -525,7 +525,7 @@ y1f_near_root (float x, float z) float index_f; int index; - index_f = roundf ((x - FIRST_ZERO_Y1) / (float) M_PI); + index_f = roundf ((x - FIRST_ZERO_Y1) / M_PIf); if (index_f >= SMALL_SIZE) return y1f_asympt (x); index = (int) index_f; diff --git a/sysdeps/ieee754/flt-32/e_jnf.c b/sysdeps/ieee754/flt-32/e_jnf.c index 80b684dd97..ff0a9d7c63 100644 --- a/sysdeps/ieee754/flt-32/e_jnf.c +++ b/sysdeps/ieee754/flt-32/e_jnf.c @@ -134,7 +134,7 @@ __ieee754_jnf(int n, float x) tmp = n; v = two/x; tmp = tmp*__ieee754_logf(fabsf(v*tmp)); - if(tmp<(float)8.8721679688e+01) { + if(tmp<8.8721679688e+01f) { for(i=n-1,di=(float)(i+i);i>0;i--){ temp = b; b *= di; diff --git a/sysdeps/ieee754/flt-32/lgamma_negf.c b/sysdeps/ieee754/flt-32/lgamma_negf.c index 756742edd8..541b5a3ab1 100644 --- a/sysdeps/ieee754/flt-32/lgamma_negf.c +++ b/sysdeps/ieee754/flt-32/lgamma_negf.c @@ -165,9 +165,9 @@ static float lg_sinpi (float x) { if (x <= 0.25f) - return __sinf ((float) M_PI * x); + return __sinf (M_PIf * x); else - return __cosf ((float) M_PI * (0.5f - x)); + return __cosf (M_PIf * (0.5f - x)); } /* Compute cos (pi * X) for -0.25 <= X <= 0.5. */ @@ -176,9 +176,9 @@ static float lg_cospi (float x) { if (x <= 0.25f) - return __cosf ((float) M_PI * x); + return __cosf (M_PIf * x); else - return __sinf ((float) M_PI * (0.5f - x)); + return __sinf (M_PIf * (0.5f - x)); } /* Compute cot (pi * X) for -0.25 <= X <= 0.5. */ diff --git a/sysdeps/ieee754/flt-32/s_log1pf.c b/sysdeps/ieee754/flt-32/s_log1pf.c index 6630f5a78e..555f0f82c8 100644 --- a/sysdeps/ieee754/flt-32/s_log1pf.c +++ b/sysdeps/ieee754/flt-32/s_log1pf.c @@ -92,7 +92,7 @@ __log1pf(float x) if(k==0) return zero; else {c += k*ln2_lo; return k*ln2_hi+c;} } - R = hfsq*((float)1.0-(float)0.66666666666666666*f); + R = hfsq*(1.0f-0.66666666666666666f*f); if(k==0) return f-R; else return k*ln2_hi-((R-(k*ln2_lo+c))-f); } |