diff options
| author | Szabolcs Nagy <nsz@port70.net> | 2014-11-05 21:40:29 +0100 |
|---|---|---|
| committer | Szabolcs Nagy <nsz@port70.net> | 2014-11-05 21:40:29 +0100 |
| commit | a732e80d33b4fd6f510f7cec4f5573ef5d89bc4e (patch) | |
| tree | 66494d38f5374808c3816c83395d5f5850abb607 | |
| parent | de2b9c21d94e0b76b629fec0060d043f535eef01 (diff) | |
| download | musl-a732e80d33b4fd6f510f7cec4f5573ef5d89bc4e.tar.gz musl-a732e80d33b4fd6f510f7cec4f5573ef5d89bc4e.tar.bz2 musl-a732e80d33b4fd6f510f7cec4f5573ef5d89bc4e.tar.xz musl-a732e80d33b4fd6f510f7cec4f5573ef5d89bc4e.zip | |
math: fix x86_64 and x32 asm not to use sahf instruction
Some early x86_64 cpus (released before 2006) did not support sahf/lahf
instructions so they should be avoided (intel manual says they are only
supported if CPUID.80000001H:ECX.LAHF-SAHF[bit 0] = 1).
The workaround simplifies exp2l and expm1l because fucomip can be
used instead of the fucomp;fnstsw;sahf sequence copied from i386.
In fmodl and remainderl sahf is replaced by a simple bit test.
| -rw-r--r-- | src/math/x32/exp2l.s | 13 | ||||
| -rw-r--r-- | src/math/x32/fmodl.s | 4 | ||||
| -rw-r--r-- | src/math/x32/remainderl.s | 4 | ||||
| -rw-r--r-- | src/math/x86_64/exp2l.s | 13 | ||||
| -rw-r--r-- | src/math/x86_64/fmodl.s | 4 | ||||
| -rw-r--r-- | src/math/x86_64/remainderl.s | 4 |
6 files changed, 14 insertions, 28 deletions
diff --git a/src/math/x32/exp2l.s b/src/math/x32/exp2l.s index d9f4d6ed..dfb2bc7c 100644 --- a/src/math/x32/exp2l.s +++ b/src/math/x32/exp2l.s @@ -6,9 +6,7 @@ expm1l: fmulp movl $0xc2820000,-4(%esp) flds -4(%esp) - fucomp %st(1) - fnstsw %ax - sahf + fucomip %st(1) fld1 jb 1f # x*log2e <= -65, return -1 without underflow @@ -17,11 +15,8 @@ expm1l: ret 1: fld %st(1) fabs - fucom %st(1) - fnstsw %ax + fucomip %st(1) fstp %st(0) - fstp %st(0) - sahf ja 1f f2xm1 ret @@ -53,9 +48,7 @@ exp2l: fld %st(1) fsub %st(1) faddp - fucomp %st(1) - fnstsw - sahf + fucomip %st(1) je 2f # x - 0x1p63 + 0x1p63 == x movl $1,(%esp) flds (%esp) # 0x1p-149 diff --git a/src/math/x32/fmodl.s b/src/math/x32/fmodl.s index 9e4378ab..b9513204 100644 --- a/src/math/x32/fmodl.s +++ b/src/math/x32/fmodl.s @@ -5,7 +5,7 @@ fmodl: fldt 8(%esp) 1: fprem fstsw %ax - sahf - jp 1b + testb $4,%ah + jnz 1b fstp %st(1) ret diff --git a/src/math/x32/remainderl.s b/src/math/x32/remainderl.s index c97f68ad..79bf4feb 100644 --- a/src/math/x32/remainderl.s +++ b/src/math/x32/remainderl.s @@ -5,7 +5,7 @@ remainderl: fldt 8(%esp) 1: fprem1 fstsw %ax - sahf - jp 1b + testb $4,%ah + jnz 1b fstp %st(1) ret diff --git a/src/math/x86_64/exp2l.s b/src/math/x86_64/exp2l.s index 0d6cd563..0e9bdf9f 100644 --- a/src/math/x86_64/exp2l.s +++ b/src/math/x86_64/exp2l.s @@ -6,9 +6,7 @@ expm1l: fmulp movl $0xc2820000,-4(%rsp) flds -4(%rsp) - fucomp %st(1) - fnstsw %ax - sahf + fucomip %st(1) fld1 jb 1f # x*log2e <= -65, return -1 without underflow @@ -17,11 +15,8 @@ expm1l: ret 1: fld %st(1) fabs - fucom %st(1) - fnstsw %ax + fucomip %st(1) fstp %st(0) - fstp %st(0) - sahf ja 1f f2xm1 ret @@ -53,9 +48,7 @@ exp2l: fld %st(1) fsub %st(1) faddp - fucomp %st(1) - fnstsw - sahf + fucomip %st(1) je 2f # x - 0x1p63 + 0x1p63 == x movl $1,(%rsp) flds (%rsp) # 0x1p-149 diff --git a/src/math/x86_64/fmodl.s b/src/math/x86_64/fmodl.s index ca81e60c..cd8d2b7c 100644 --- a/src/math/x86_64/fmodl.s +++ b/src/math/x86_64/fmodl.s @@ -5,7 +5,7 @@ fmodl: fldt 8(%rsp) 1: fprem fstsw %ax - sahf - jp 1b + testb $4,%ah + jnz 1b fstp %st(1) ret diff --git a/src/math/x86_64/remainderl.s b/src/math/x86_64/remainderl.s index 75c12374..2c337cf5 100644 --- a/src/math/x86_64/remainderl.s +++ b/src/math/x86_64/remainderl.s @@ -5,7 +5,7 @@ remainderl: fldt 8(%rsp) 1: fprem1 fstsw %ax - sahf - jp 1b + testb $4,%ah + jnz 1b fstp %st(1) ret |