From 54142c44e963f410262fa868c159b6df858f3c53 Mon Sep 17 00:00:00 2001 From: Joseph Myers Date: Wed, 23 Sep 2015 18:14:57 +0000 Subject: Use math_narrow_eval more consistently. Where glibc code needs to avoid excess range and precision in floating-point arithmetic, code variously uses either asms or volatile to force the results of that arithmetic to memory; mostly this is conditional on FLT_EVAL_METHOD, but in the case of lrint / llrint functions some use of volatile is unconditional (and is present unnecessarily in versions for long double). This patch make such code use the recently-added math_narrow_eval macro consistently, removing the unnecessary uses of volatile in long double lrint / llrint implementations completely. Tested for x86_64, x86, mips64 and powerpc. * math/s_nexttowardf.c (__nexttowardf): Use math_narrow_eval. * stdlib/strtod_l.c: Include . (overflow_value): Use math_narrow_eval. (underflow_value): Likewise. * sysdeps/i386/fpu/s_nexttoward.c (__nexttoward): Likewise. * sysdeps/i386/fpu/s_nexttowardf.c (__nexttowardf): Likewise. * sysdeps/ieee754/dbl-64/e_gamma_r.c (gamma_positive): Likewise. (__ieee754_gamma_r): Likewise. * sysdeps/ieee754/dbl-64/gamma_productf.c (__gamma_productf): Likewise. * sysdeps/ieee754/dbl-64/k_rem_pio2.c (__kernel_rem_pio2): Likewise. * sysdeps/ieee754/dbl-64/lgamma_neg.c (__lgamma_neg): Likewise. * sysdeps/ieee754/dbl-64/s_erf.c (__erfc): Likewise. * sysdeps/ieee754/dbl-64/s_llrint.c (__llrint): Likewise. * sysdeps/ieee754/dbl-64/s_lrint.c (__lrint): Likewise. * sysdeps/ieee754/flt-32/e_gammaf_r.c (gammaf_positive): Likewise. (__ieee754_gammaf_r): Likewise. * sysdeps/ieee754/flt-32/k_rem_pio2f.c (__kernel_rem_pio2f): Likewise. * sysdeps/ieee754/flt-32/lgamma_negf.c (__lgamma_negf): Likewise. * sysdeps/ieee754/flt-32/s_erff.c (__erfcf): Likewise. * sysdeps/ieee754/flt-32/s_llrintf.c (__llrintf): Likewise. * sysdeps/ieee754/flt-32/s_lrintf.c (__lrintf): Likewise. * sysdeps/ieee754/ldbl-128/s_llrintl.c (__llrintl): Do not use volatile. * sysdeps/ieee754/ldbl-128/s_lrintl.c (__lrintl): Likewise. * sysdeps/ieee754/ldbl-128/s_nexttoward.c (__nexttoward): Use math_narrow_eval. * sysdeps/ieee754/ldbl-128ibm/s_nexttoward.c (__nexttoward): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_nexttowardf.c (__nexttowardf): Likewise. * sysdeps/ieee754/ldbl-96/gamma_product.c (__gamma_product): Likewise. * sysdeps/ieee754/ldbl-96/s_llrintl.c (__llrintl): Do not use volatile. * sysdeps/ieee754/ldbl-96/s_lrintl.c (__lrintl): Likewise. * sysdeps/ieee754/ldbl-96/s_nexttoward.c (__nexttoward): Use math_narrow_eval. * sysdeps/ieee754/ldbl-96/s_nexttowardf.c (__nexttowardf): Likewise. * sysdeps/ieee754/ldbl-opt/s_nexttowardfd.c (__nldbl_nexttowardf): Likewise. --- sysdeps/ieee754/flt-32/e_gammaf_r.c | 20 +++++++------------- sysdeps/ieee754/flt-32/k_rem_pio2f.c | 19 +++++-------------- sysdeps/ieee754/flt-32/lgamma_negf.c | 12 ++---------- sysdeps/ieee754/flt-32/s_erff.c | 5 +---- sysdeps/ieee754/flt-32/s_llrintf.c | 4 ++-- sysdeps/ieee754/flt-32/s_lrintf.c | 4 ++-- 6 files changed, 19 insertions(+), 45 deletions(-) (limited to 'sysdeps/ieee754/flt-32') diff --git a/sysdeps/ieee754/flt-32/e_gammaf_r.c b/sysdeps/ieee754/flt-32/e_gammaf_r.c index 250e412aa7..183db1e033 100644 --- a/sysdeps/ieee754/flt-32/e_gammaf_r.c +++ b/sysdeps/ieee754/flt-32/e_gammaf_r.c @@ -69,11 +69,7 @@ gammaf_positive (float x, int *exp2_adj) /* Adjust into the range for applying Stirling's approximation. */ float n = __ceilf (4.0f - x); -#if FLT_EVAL_METHOD != 0 - volatile -#endif - float x_tmp = x + n; - x_adj = x_tmp; + x_adj = math_narrow_eval (x + n); x_eps = (x - (x_adj - n)); prod = __gamma_productf (x_adj - n, x_eps, n, &eps); } @@ -111,9 +107,6 @@ float __ieee754_gammaf_r (float x, int *signgamp) { int32_t hx; -#if FLT_EVAL_METHOD != 0 - volatile -#endif float ret; GET_FLOAT_WORD (hx, x); @@ -149,7 +142,7 @@ __ieee754_gammaf_r (float x, int *signgamp) { /* Overflow. */ *signgamp = 0; - ret = FLT_MAX * FLT_MAX; + ret = math_narrow_eval (FLT_MAX * FLT_MAX); return ret; } else @@ -193,27 +186,28 @@ __ieee754_gammaf_r (float x, int *signgamp) } } } + ret = math_narrow_eval (ret); } if (isinf (ret) && x != 0) { if (*signgamp < 0) { - ret = -__copysignf (FLT_MAX, ret) * FLT_MAX; + ret = math_narrow_eval (-__copysignf (FLT_MAX, ret) * FLT_MAX); ret = -ret; } else - ret = __copysignf (FLT_MAX, ret) * FLT_MAX; + ret = math_narrow_eval (__copysignf (FLT_MAX, ret) * FLT_MAX); return ret; } else if (ret == 0) { if (*signgamp < 0) { - ret = -__copysignf (FLT_MIN, ret) * FLT_MIN; + ret = math_narrow_eval (-__copysignf (FLT_MIN, ret) * FLT_MIN); ret = -ret; } else - ret = __copysignf (FLT_MIN, ret) * FLT_MIN; + ret = math_narrow_eval (__copysignf (FLT_MIN, ret) * FLT_MIN); return ret; } else diff --git a/sysdeps/ieee754/flt-32/k_rem_pio2f.c b/sysdeps/ieee754/flt-32/k_rem_pio2f.c index 6f14d5bac7..0c7685c3ab 100644 --- a/sysdeps/ieee754/flt-32/k_rem_pio2f.c +++ b/sysdeps/ieee754/flt-32/k_rem_pio2f.c @@ -167,30 +167,21 @@ recompute: break; case 1: case 2:; -#if __FLT_EVAL_METHOD__ != 0 - volatile -#endif float fv = 0.0; - for (i=jz;i>=0;i--) fv += fq[i]; + for (i=jz;i>=0;i--) fv = math_narrow_eval (fv + fq[i]); y[0] = (ih==0)? fv: -fv; - fv = fq[0]-fv; - for (i=1;i<=jz;i++) fv += fq[i]; + fv = math_narrow_eval (fq[0]-fv); + for (i=1;i<=jz;i++) fv = math_narrow_eval (fv + fq[i]); y[1] = (ih==0)? fv: -fv; break; case 3: /* painful */ for (i=jz;i>0;i--) { -#if __FLT_EVAL_METHOD__ != 0 - volatile -#endif - float fv = fq[i-1]+fq[i]; + float fv = math_narrow_eval (fq[i-1]+fq[i]); fq[i] += fq[i-1]-fv; fq[i-1] = fv; } for (i=jz;i>1;i--) { -#if __FLT_EVAL_METHOD__ != 0 - volatile -#endif - float fv = fq[i-1]+fq[i]; + float fv = math_narrow_eval (fq[i-1]+fq[i]); fq[i] += fq[i-1]-fv; fq[i-1] = fv; } diff --git a/sysdeps/ieee754/flt-32/lgamma_negf.c b/sysdeps/ieee754/flt-32/lgamma_negf.c index ed9659801e..68bee8572a 100644 --- a/sysdeps/ieee754/flt-32/lgamma_negf.c +++ b/sysdeps/ieee754/flt-32/lgamma_negf.c @@ -245,17 +245,9 @@ __lgamma_negf (float x, int *signgamp) } float log_gamma_ratio; -#if FLT_EVAL_METHOD != 0 - volatile -#endif - float y0_tmp = 1 - x0_hi; - float y0 = y0_tmp; + float y0 = math_narrow_eval (1 - x0_hi); float y0_eps = -x0_hi + (1 - y0) - x0_lo; -#if FLT_EVAL_METHOD != 0 - volatile -#endif - float y_tmp = 1 - x; - float y = y_tmp; + float y = math_narrow_eval (1 - x); float y_eps = -x + (1 - y); /* We now wish to compute LOG_GAMMA_RATIO = log (gamma (Y0 + Y0_EPS) / gamma (Y + Y_EPS)). XDIFF diff --git a/sysdeps/ieee754/flt-32/s_erff.c b/sysdeps/ieee754/flt-32/s_erff.c index 2be44cc40c..1aabce5cb8 100644 --- a/sysdeps/ieee754/flt-32/s_erff.c +++ b/sysdeps/ieee754/flt-32/s_erff.c @@ -213,10 +213,7 @@ float __erfcf(float x) r = __ieee754_expf(-z*z-(float)0.5625)* __ieee754_expf((z-x)*(z+x)+R/S); if(hx>0) { -#if FLT_EVAL_METHOD != 0 - volatile -#endif - float ret = r/x; + float ret = math_narrow_eval (r/x); if (ret == 0) __set_errno (ERANGE); return ret; diff --git a/sysdeps/ieee754/flt-32/s_llrintf.c b/sysdeps/ieee754/flt-32/s_llrintf.c index ac8c1a2216..d7588d04cf 100644 --- a/sysdeps/ieee754/flt-32/s_llrintf.c +++ b/sysdeps/ieee754/flt-32/s_llrintf.c @@ -34,7 +34,7 @@ __llrintf (float x) { int32_t j0; u_int32_t i0; - volatile float w; + float w; float t; long long int result; int sx; @@ -52,7 +52,7 @@ __llrintf (float x) result = (long long int) i0 << (j0 - 23); else { - w = two23[sx] + x; + w = math_narrow_eval (two23[sx] + x); t = w - two23[sx]; GET_FLOAT_WORD (i0, t); j0 = ((i0 >> 23) & 0xff) - 0x7f; diff --git a/sysdeps/ieee754/flt-32/s_lrintf.c b/sysdeps/ieee754/flt-32/s_lrintf.c index 7581a8d286..bc60fd2311 100644 --- a/sysdeps/ieee754/flt-32/s_lrintf.c +++ b/sysdeps/ieee754/flt-32/s_lrintf.c @@ -34,7 +34,7 @@ __lrintf (float x) { int32_t j0; u_int32_t i0; - volatile float w; + float w; float t; long int result; int sx; @@ -52,7 +52,7 @@ __lrintf (float x) result = (long int) i0 << (j0 - 23); else { - w = two23[sx] + x; + w = math_narrow_eval (two23[sx] + x); t = w - two23[sx]; GET_FLOAT_WORD (i0, t); j0 = ((i0 >> 23) & 0xff) - 0x7f; -- cgit 1.4.1