math: move x87-family remainder functions to C with inline asm

src/math/i386/remainder.c

+#include <math.h>
+
+double remainder(double x, double y)
+{
+	unsigned short fpsr;
+	// fprem1 does not introduce excess precision into x
+	do __asm__ ("fprem1; fnstsw %%ax" : "+t"(x), "=a"(fpsr) : "u"(y));
+	while (fpsr & 0x400);
+	return x;
+}
+
+weak_alias(remainder, drem);

src/math/i386/remainder.s

-.global remainder
-.type remainder,@function
-remainder:
-.weak drem
-.type drem,@function
-drem:
-	fldl 12(%esp)
-	fldl 4(%esp)
-1:	fprem1
-	fnstsw %ax
-	sahf
-	jp 1b
-	fstp %st(1)
-	ret

src/math/i386/remainderf.c

+#include <math.h>
+
+float remainderf(float x, float y)
+{
+	unsigned short fpsr;
+	// fprem1 does not introduce excess precision into x
+	do __asm__ ("fprem1; fnstsw %%ax" : "+t"(x), "=a"(fpsr) : "u"(y));
+	while (fpsr & 0x400);
+	return x;
+}
+
+weak_alias(remainderf, dremf);

src/math/i386/remainderf.s

-.global remainderf
-.type remainderf,@function
-remainderf:
-.weak dremf
-.type dremf,@function
-dremf:
-	flds 8(%esp)
-	flds 4(%esp)
-1:	fprem1
-	fnstsw %ax
-	sahf
-	jp 1b
-	fstp %st(1)
-	ret

src/math/i386/remainderl.c

+#include <math.h>
+
+long double remainderl(long double x, long double y)
+{
+	unsigned short fpsr;
+	do __asm__ ("fprem1; fnstsw %%ax" : "+t"(x), "=a"(fpsr) : "u"(y));
+	while (fpsr & 0x400);
+	return x;
+}

src/math/i386/remainderl.s

-.global remainderl
-.type remainderl,@function
-remainderl:
-	fldt 16(%esp)
-	fldt 4(%esp)
-1:	fprem1
-	fnstsw %ax
-	sahf
-	jp 1b
-	fstp %st(1)
-	ret

src/math/x86_64/remainderl.c

+#include <math.h>
+
+long double remainderl(long double x, long double y)
+{
+	unsigned short fpsr;
+	do __asm__ ("fprem1; fnstsw %%ax" : "+t"(x), "=a"(fpsr) : "u"(y));
+	while (fpsr & 0x400);
+	return x;
+}

src/math/x86_64/remainderl.s

-.global remainderl
-.type remainderl,@function
-remainderl:
-	fldt 24(%rsp)
-	fldt 8(%rsp)
-1:	fprem1
-	fnstsw %ax
-	testb $4,%ah
-	jnz 1b
-	fstp %st(1)
-	ret