diff options
| author | Adhemerval Zanella <adhemerval.zanella@linaro.org> | 2020-04-08 17:42:46 -0300 |
|---|---|---|
| committer | Adhemerval Zanella <adhemerval.zanella@linaro.org> | 2020-06-19 10:48:15 -0300 |
| commit | 4b2d8e4442745a09f24804c1fcc97b7c5c4506f1 (patch) | |
| tree | 78b0fe0d725f8ab84c66b978e06216be00d634f2 | |
| parent | 6e98983c0991433fec8cef8702e2028fa6bef12d (diff) | |
| download | glibc-4b2d8e4442745a09f24804c1fcc97b7c5c4506f1.zip glibc-4b2d8e4442745a09f24804c1fcc97b7c5c4506f1.tar.gz glibc-4b2d8e4442745a09f24804c1fcc97b7c5c4506f1.tar.bz2 | |
i386: Use generic exp10f
The generic implementation is twice as fast. Using the exp10f
benchmark:
* master:
"exp10f": {
"workload-spec2017.wrf (adapted)": {
"duration": 1.02967e+09,
"iterations": 4.768e+07,
"reciprocal-throughput": 18.3579,
"latency": 24.8331,
"max-throughput": 5.44725e+07,
"min-throughput": 4.02688e+07
}
}
* patched:
"exp10f": {
"workload-spec2017.wrf (adapted)": {
"duration": 1.01821e+09,
"iterations": 6.1984e+07,
"reciprocal-throughput": 13.1975,
"latency": 19.6563,
"max-throughput": 7.57719e+07,
"min-throughput": 5.08743e+07
}
}
Checked on i686-linux-gnu.
| -rw-r--r-- | sysdeps/i386/fpu/e_exp10f.S | 54 |
1 files changed, 0 insertions, 54 deletions
diff --git a/sysdeps/i386/fpu/e_exp10f.S b/sysdeps/i386/fpu/e_exp10f.S deleted file mode 100644 index 196ce87..0000000 --- a/sysdeps/i386/fpu/e_exp10f.S +++ /dev/null @@ -1,54 +0,0 @@ -/* - * Written by Ulrich Drepper. - */ - -#include <machine/asm.h> -#include <i386-math-asm.h> -#include <libm-alias-finite.h> - -DEFINE_FLT_MIN - -#ifdef PIC -# define MO(op) op##@GOTOFF(%ecx) -#else -# define MO(op) op -#endif - - .text -/* 10^x = 2^(x * log2(10)) */ -ENTRY(__ieee754_exp10f) -#ifdef PIC - LOAD_PIC_REG (cx) -#endif - flds 4(%esp) -/* I added the following ugly construct because exp(+-Inf) resulted - in NaN. The ugliness results from the bright minds at Intel. - For the i686 the code can be written better. - -- drepper@cygnus.com. */ - fxam /* Is NaN or +-Inf? */ - fstsw %ax - movb $0x45, %dh - andb %ah, %dh - cmpb $0x05, %dh - je 1f /* Is +-Inf, jump. */ - fldl2t - fmulp /* x * log2(10) */ - fld %st - frndint /* int(x * log2(10)) */ - fsubr %st,%st(1) /* fract(x * log2(10)) */ - fxch - f2xm1 /* 2^(fract(x * log2(10))) - 1 */ - fld1 - faddp /* 2^(fract(x * log2(10))) */ - fscale /* e^x */ - fstp %st(1) - FLT_NARROW_EVAL_UFLOW_NONNEG_NAN - ret - -1: testl $0x200, %eax /* Test sign. */ - jz 2f /* If positive, jump. */ - fstp %st - fldz /* Set result to 0. */ -2: ret -END (__ieee754_exp10f) -libm_alias_finite (__ieee754_exp10f, __exp10f) |