import vanilla x86_64 code as x32

src/math/x32/acosl.s

@@ -0,0 +1,16 @@
+# see ../i386/acos.s
+
+.global acosl
+.type acosl,@function
+acosl:
+	fldt 8(%rsp)
+1:	fld %st(0)
+	fld1
+	fsub %st(0),%st(1)
+	fadd %st(2)
+	fmulp
+	fsqrt
+	fabs
+	fxch %st(1)
+	fpatan
+	ret

src/math/x32/asinl.s

@@ -0,0 +1,12 @@
+.global asinl
+.type asinl,@function
+asinl:
+	fldt 8(%rsp)
+1:	fld %st(0)
+	fld1
+	fsub %st(0),%st(1)
+	fadd %st(2)
+	fmulp
+	fsqrt
+	fpatan
+	ret

src/math/x32/atan2l.s

@@ -0,0 +1,7 @@
+.global atan2l
+.type atan2l,@function
+atan2l:
+	fldt 8(%rsp)
+	fldt 24(%rsp)
+	fpatan
+	ret

src/math/x32/atanl.s

@@ -0,0 +1,7 @@
+.global atanl
+.type atanl,@function
+atanl:
+	fldt 8(%rsp)
+	fld1
+	fpatan
+	ret

src/math/x32/ceill.s

@@ -0,0 +1 @@
+# see floorl.s

src/math/x32/exp2l.s

@@ -0,0 +1,90 @@
+.global expm1l
+.type expm1l,@function
+expm1l:
+	fldt 8(%rsp)
+	fldl2e
+	fmulp
+	movl $0xc2820000,-4(%rsp)
+	flds -4(%rsp)
+	fucomp %st(1)
+	fnstsw %ax
+	sahf
+	fld1
+	jb 1f
+		# x*log2e <= -65, return -1 without underflow
+	fstp %st(1)
+	fchs
+	ret
+1:	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:	fld %st(0)
+	sub $16,%rsp
+	fstpt (%rsp)
+	mov 8(%rsp),%ax
+	and $0x7fff,%ax
+	cmp $0x3fff+13,%ax
+	jb 4f             # |x| < 8192
+	cmp $0x3fff+15,%ax
+	jae 3f            # |x| >= 32768
+	fsts (%rsp)
+	cmpl $0xc67ff800,(%rsp)
+	jb 2f             # x > -16382
+	movl $0x5f000000,(%rsp)
+	flds (%rsp)       # 0x1p63
+	fld %st(1)
+	fsub %st(1)
+	faddp
+	fucomp %st(1)
+	fnstsw
+	sahf
+	je 2f             # x - 0x1p63 + 0x1p63 == x
+	movl $1,(%rsp)
+	flds (%rsp)       # 0x1p-149
+	fdiv %st(1)
+	fstps (%rsp)      # raise underflow
+2:	fld1
+	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)
+	add $16,%rsp
+	ret
+3:	xor %eax,%eax
+4:	cmp $0x3fff-64,%ax
+	fld1
+	jb 1b             # |x| < 0x1p-64
+	fstpt (%rsp)
+	fistl 8(%rsp)
+	fildl 8(%rsp)
+	fsubrp %st(1)
+	addl $0x3fff,8(%rsp)
+	f2xm1
+	fld1
+	faddp             # 2^(x-rint(x))
+	fldt (%rsp)       # 2^rint(x)
+	fmulp
+	add $16,%rsp
+	ret

src/math/x32/expl.s

@@ -0,0 +1,101 @@
+# exp(x) = 2^hi + 2^hi (2^lo - 1)
+# where hi+lo = log2e*x with 128bit precision
+# exact log2e*x calculation depends on nearest rounding mode
+# using the exact multiplication method of Dekker and Veltkamp
+
+.global expl
+.type expl,@function
+expl:
+	fldt 8(%rsp)
+
+		# interesting case: 0x1p-32 <= |x| < 16384
+		# check if (exponent|0x8000) is in [0xbfff-32, 0xbfff+13]
+	mov 16(%rsp), %ax
+	or $0x8000, %ax
+	sub $0xbfdf, %ax
+	cmp $45, %ax
+	jbe 2f
+	test %ax, %ax
+	fld1
+	js 1f
+		# if |x|>=0x1p14 or nan return 2^trunc(x)
+	fscale
+	fstp %st(1)
+	ret
+		# if |x|<0x1p-32 return 1+x
+1:	faddp
+	ret
+
+		# should be 0x1.71547652b82fe178p0L == 0x3fff b8aa3b29 5c17f0bc
+		# it will be wrong on non-nearest rounding mode
+2:	fldl2e
+	subq $48, %rsp
+		# hi = log2e_hi*x
+		# 2^hi = exp2l(hi)
+	fmul %st(1),%st
+	fld %st(0)
+	fstpt (%rsp)
+	fstpt 16(%rsp)
+	fstpt 32(%rsp)
+	call exp2l
+		# if 2^hi == inf return 2^hi
+	fld %st(0)
+	fstpt (%rsp)
+	cmpw $0x7fff, 8(%rsp)
+	je 1f
+	fldt 32(%rsp)
+	fldt 16(%rsp)
+		# fpu stack: 2^hi x hi
+		# exact mult: x*log2e
+	fld %st(1)
+		# c = 0x1p32+1
+	movq $0x41f0000000100000,%rax
+	pushq %rax
+	fldl (%rsp)
+		# xh = x - c*x + c*x
+		# xl = x - xh
+	fmulp
+	fld %st(2)
+	fsub %st(1), %st
+	faddp
+	fld %st(2)
+	fsub %st(1), %st
+		# yh = log2e_hi - c*log2e_hi + c*log2e_hi
+	movq $0x3ff7154765200000,%rax
+	pushq %rax
+	fldl (%rsp)
+		# fpu stack: 2^hi x hi xh xl yh
+		# lo = hi - xh*yh + xl*yh
+	fld %st(2)
+	fmul %st(1), %st
+	fsubp %st, %st(4)
+	fmul %st(1), %st
+	faddp %st, %st(3)
+		# yl = log2e_hi - yh
+	movq $0x3de705fc2f000000,%rax
+	pushq %rax
+	fldl (%rsp)
+		# fpu stack: 2^hi x lo xh xl yl
+		# lo += xh*yl + xl*yl
+	fmul %st, %st(2)
+	fmulp %st, %st(1)
+	fxch %st(2)
+	faddp
+	faddp
+		# log2e_lo
+	movq $0xbfbe,%rax
+	pushq %rax
+	movq $0x82f0025f2dc582ee,%rax
+	pushq %rax
+	fldt (%rsp)
+	addq $40,%rsp
+		# fpu stack: 2^hi x lo log2e_lo
+		# lo += log2e_lo*x
+		# return 2^hi + 2^hi (2^lo - 1)
+	fmulp %st, %st(2)
+	faddp
+	f2xm1
+	fmul %st(1), %st
+	faddp
+1:	addq $48, %rsp
+	ret

src/math/x32/expm1l.s

@@ -0,0 +1 @@
+# see exp2l.s

src/math/x32/fabs.s

@@ -0,0 +1,9 @@
+.global fabs
+.type fabs,@function
+fabs:
+	xor %eax,%eax
+	dec %rax
+	shr %rax
+	movq %rax,%xmm1
+	andpd %xmm1,%xmm0
+	ret

src/math/x32/fabsf.s

@@ -0,0 +1,7 @@
+.global fabsf
+.type fabsf,@function
+fabsf:
+	mov $0x7fffffff,%eax
+	movq %rax,%xmm1
+	andps %xmm1,%xmm0
+	ret

src/math/x32/fabsl.s

@@ -0,0 +1,6 @@
+.global fabsl
+.type fabsl,@function
+fabsl:
+	fldt 8(%rsp)
+	fabs
+	ret

src/math/x32/floorl.s

@@ -0,0 +1,27 @@
+.global floorl
+.type floorl,@function
+floorl:
+	fldt 8(%rsp)
+1:	mov $0x7,%al
+1:	fstcw 8(%rsp)
+	mov 9(%rsp),%ah
+	mov %al,9(%rsp)
+	fldcw 8(%rsp)
+	frndint
+	mov %ah,9(%rsp)
+	fldcw 8(%rsp)
+	ret
+
+.global ceill
+.type ceill,@function
+ceill:
+	fldt 8(%rsp)
+	mov $0xb,%al
+	jmp 1b
+
+.global truncl
+.type truncl,@function
+truncl:
+	fldt 8(%rsp)
+	mov $0xf,%al
+	jmp 1b

src/math/x32/fmodl.s

@@ -0,0 +1,11 @@
+.global fmodl
+.type fmodl,@function
+fmodl:
+	fldt 24(%rsp)
+	fldt 8(%rsp)
+1:	fprem
+	fstsw %ax
+	sahf
+	jp 1b
+	fstp %st(1)
+	ret

src/math/x32/llrint.s

@@ -0,0 +1,5 @@
+.global llrint
+.type llrint,@function
+llrint:
+	cvtsd2si %xmm0,%rax
+	ret

src/math/x32/llrintf.s

@@ -0,0 +1,5 @@
+.global llrintf
+.type llrintf,@function
+llrintf:
+	cvtss2si %xmm0,%rax
+	ret

src/math/x32/llrintl.s

@@ -0,0 +1,7 @@
+.global llrintl
+.type llrintl,@function
+llrintl:
+	fldt 8(%rsp)
+	fistpll 8(%rsp)
+	mov 8(%rsp),%rax
+	ret

src/math/x32/log10l.s

@@ -0,0 +1,7 @@
+.global log10l
+.type log10l,@function
+log10l:
+	fldlg2
+	fldt 8(%rsp)
+	fyl2x
+	ret

src/math/x32/log1pl.s

@@ -0,0 +1,15 @@
+.global log1pl
+.type log1pl,@function
+log1pl:
+	mov 14(%rsp),%eax
+	fldln2
+	and $0x7fffffff,%eax
+	fldt 8(%rsp)
+	cmp $0x3ffd9400,%eax
+	ja 1f
+	fyl2xp1
+	ret
+1:	fld1
+	faddp
+	fyl2x
+	ret

src/math/x32/log2l.s

@@ -0,0 +1,7 @@
+.global log2l
+.type log2l,@function
+log2l:
+	fld1
+	fldt 8(%rsp)
+	fyl2x
+	ret

src/math/x32/logl.s

@@ -0,0 +1,7 @@
+.global logl
+.type logl,@function
+logl:
+	fldln2
+	fldt 8(%rsp)
+	fyl2x
+	ret

src/math/x32/lrint.s

@@ -0,0 +1,5 @@
+.global lrint
+.type lrint,@function
+lrint:
+	cvtsd2si %xmm0,%rax
+	ret

src/math/x32/lrintf.s

@@ -0,0 +1,5 @@
+.global lrintf
+.type lrintf,@function
+lrintf:
+	cvtss2si %xmm0,%rax
+	ret

src/math/x32/lrintl.s

@@ -0,0 +1,7 @@
+.global lrintl
+.type lrintl,@function
+lrintl:
+	fldt 8(%rsp)
+	fistpll 8(%rsp)
+	mov 8(%rsp),%rax
+	ret

src/math/x32/remainderl.s

@@ -0,0 +1,11 @@
+.global remainderl
+.type remainderl,@function
+remainderl:
+	fldt 24(%rsp)
+	fldt 8(%rsp)
+1:	fprem1
+	fstsw %ax
+	sahf
+	jp 1b
+	fstp %st(1)
+	ret

src/math/x32/rintl.s

@@ -0,0 +1,6 @@
+.global rintl
+.type rintl,@function
+rintl:
+	fldt 8(%rsp)
+	frndint
+	ret

src/math/x32/sqrt.s

@@ -0,0 +1,4 @@
+.global sqrt
+.type sqrt,@function
+sqrt:	sqrtsd %xmm0, %xmm0
+	ret

src/math/x32/sqrtf.s

@@ -0,0 +1,4 @@
+.global sqrtf
+.type sqrtf,@function
+sqrtf:  sqrtss %xmm0, %xmm0
+	ret

src/math/x32/sqrtl.s

@@ -0,0 +1,5 @@
+.global sqrtl
+.type sqrtl,@function
+sqrtl:	fldt 8(%rsp)
+	fsqrt
+	ret

src/math/x32/truncl.s

@@ -0,0 +1 @@
+# see floorl.s