From 525ad96e0ef6e46b6ac28438ef2ffa1d5317fb64 Mon Sep 17 00:00:00 2001 From: Szabolcs Nagy Date: Sun, 16 Dec 2012 17:28:18 +0100 Subject: [PATCH] math: move x86_64 exp2l implementation to exp2l.s from expl.s --- src/math/x86_64/exp2l.s | 76 +++++++++++++++++++++++++++++++++++- src/math/x86_64/expl.s | 83 ---------------------------------------- src/math/x86_64/expm1l.s | 2 +- 3 files changed, 76 insertions(+), 85 deletions(-) diff --git a/src/math/x86_64/exp2l.s b/src/math/x86_64/exp2l.s index 510c256d..1f8ed7bb 100644 --- a/src/math/x86_64/exp2l.s +++ b/src/math/x86_64/exp2l.s @@ -1 +1,75 @@ -# see expl.s +.global expm1l +.type expm1l,@function +expm1l: + fldt 8(%rsp) + fldl2e + fmulp + fld1 + fld %st(1) + fabs + fucom %st(1) + fnstsw %ax + fstp %st(0) + fstp %st(0) + sahf + ja 1f + f2xm1 + ret +1: push %rax + call 1f + pop %rax + fld1 + fsubrp + ret + +.global exp2l +.type exp2l,@function +exp2l: + fldt 8(%rsp) +1: mov $0x467ff000,%eax + mov %eax,-16(%rsp) + mov $0x80000000,%eax + mov %eax,-20(%rsp) + xor %eax,%eax + mov %eax,-24(%rsp) + flds -16(%rsp) # 16380 + fld %st(1) + fabs + fucom %st(1) + fnstsw + fstp %st(0) + fstp %st(0) + sahf + ja 3f # |x| > 16380 + jp 2f # x is nan (avoid invalid except in fistp) + fld %st(0) + fistpl -16(%rsp) + fildl -16(%rsp) + fxch %st(1) + fsub %st(1) + mov $0x3fff,%eax + add %eax,-16(%rsp) + f2xm1 + fld1 + faddp # 2^(x-rint(x)) + fldt -24(%rsp) # 2^rint(x) + fmulp +2: fstp %st(1) + ret + +3: fld %st(0) + fstpt -24(%rsp) + fld1 + mov -15(%rsp),%ax + and $0x7fff,%ax + cmp $0x7fff,%ax + je 1f # x = +-inf + fld %st(1) + frndint + fxch %st(2) + fsub %st(2) # st(0)=x-rint(x), st(1)=1, st(2)=rint(x) + f2xm1 + faddp # 2^(x-rint(x)) +1: fscale + fstp %st(1) + ret diff --git a/src/math/x86_64/expl.s b/src/math/x86_64/expl.s index 740bc779..e69de29b 100644 --- a/src/math/x86_64/expl.s +++ b/src/math/x86_64/expl.s @@ -1,83 +0,0 @@ -.global expm1l -.type expm1l,@function -expm1l: - fldt 8(%rsp) -1: fldl2e - fmulp - fld1 - fld %st(1) - fabs - fucom %st(1) - fnstsw %ax - fstp %st(0) - fstp %st(0) - sahf - ja 1f - f2xm1 - ret -1: push %rax - call 1f - pop %rax - fld1 - fsubrp - ret - -.global expl -.type expl,@function -expl: - fldt 8(%rsp) - fldl2e - fmulp - jmp 1f - -.global exp2l -.type exp2l,@function -exp2l: - fldt 8(%rsp) -1: mov $0x467ff000,%eax - mov %eax,-16(%rsp) - mov $0x80000000,%eax - mov %eax,-20(%rsp) - xor %eax,%eax - mov %eax,-24(%rsp) - flds -16(%rsp) # 16380 - fld %st(1) - fabs - fucom %st(1) - fnstsw - fstp %st(0) - fstp %st(0) - sahf - ja 3f # |x| > 16380 - jp 2f # x is nan (avoid invalid except in fistp) - fld %st(0) - fistpl -16(%rsp) - fildl -16(%rsp) - fxch %st(1) - fsub %st(1) - mov $0x3fff,%eax - add %eax,-16(%rsp) - f2xm1 - fld1 - faddp # 2^(x-rint(x)) - fldt -24(%rsp) # 2^rint(x) - fmulp -2: fstp %st(1) - ret - -3: fld %st(0) - fstpt -24(%rsp) - fld1 - mov -15(%rsp),%ax - and $0x7fff,%ax - cmp $0x7fff,%ax - je 1f # x = +-inf - fld %st(1) - frndint - fxch %st(2) - fsub %st(2) # st(0)=x-rint(x), st(1)=1, st(2)=rint(x) - f2xm1 - faddp # 2^(x-rint(x)) -1: fscale - fstp %st(1) - ret diff --git a/src/math/x86_64/expm1l.s b/src/math/x86_64/expm1l.s index 89c2d290..e773f080 100644 --- a/src/math/x86_64/expm1l.s +++ b/src/math/x86_64/expm1l.s @@ -1 +1 @@ -# see expm1.s +# see exp2l.s -- 2.25.1