Added some more incredibly untested helper functions.

mach/mips/libem/bls4.s

@@ -0,0 +1,30 @@
+#
+.sect .text; .sect .rom; .sect .data; .sect .bss
+
+/* Does a block move of words between non-overlapping buffers.
+ * Stack: ( src dst len -- )
+ */
+
+.sect .text
+.define .bls4
+.bls4:
+	lw r4, 0(sp)	! r4=len
+	lw r5, 4(sp)	! r5=dst
+	lw r6, 8(sp)	! r6=src
+	addiu sp, sp, 12
+
+	srl r4, r4, 2   ! convert len to words
+1:
+	beq r4, zero, 2f
+	nop
+
+	lw at, 0(r6)
+	sw at, 0(r5)
+	addiu r6, r6, 4
+	addiu r5, r5, 4
+	addiu r4, r4, -1
+	b 1b
+
+2:
+	jr ra
+	nop

mach/mips/libem/c_ud_i.s

@@ -1,8 +1,5 @@
 #
-.sect .text
-.sect .rom
-.sect .data
-.sect .bss
+.sect .text; .sect .rom; .sect .data; .sect .bss
 
 .sect .text
 .define .c_ud_i
@@ -30,3 +27,10 @@ toobig:
 	addiu r2, r2, 0x8000
 	jr ra
 	nop
+
+/* 2147483648 as a double. */
+.sect .rom
+.define .fd_80000000
+.fd_80000000:
+	.data4 0, 0x41e00000
+

mach/mips/libem/c_uf_i.s

@@ -31,3 +31,8 @@ toobig:
 	jr ra
 	nop
 
+/* 2147483648 as a float. */
+.sect .rom
+.ff_80000000:
+	.data4 0x4f000000
+

mach/mips/libem/c_ui_d.s

@@ -0,0 +1,26 @@
+#
+.sect .text; .sect .rom; .sect .data; .sect .bss
+
+.sect .text
+.define .c_ui_d
+.c_ui_d:
+	/* Input: r2
+	 * Output: f0
+	 * Only at and f30/f31 may be used.
+	 */
+	mtc1 r2, f0
+	cvt.d.w f0, f0
+	bgez r2, nonnegative
+	nop
+
+	ori at, zero, hi16[.fd_100000000]
+	ldc1 f30, lo16[.fd_100000000] (at)
+	add.d f0, f0, f30
+nonnegative:
+	jr ra
+	nop
+
+/* 4294967296 as a double. */
+.sect .rom
+.fd_100000000:
+	.data4 0, 0x41f00000

mach/mips/libem/c_ui_f.s

@@ -0,0 +1,26 @@
+#
+.sect .text; .sect .rom; .sect .data; .sect .bss
+
+.sect .text
+.define .c_ui_f
+.c_ui_f:
+	/* Input: r2
+	 * Output: f0
+	 * Only at and f30/f31 may be used.
+	 */
+	mtc1 r2, f0
+	cvt.s.w f0, f0
+	bgez r2, nonnegative
+	nop
+
+	ori at, zero, hi16[.fs_100000000]
+	ldc1 f30, lo16[.fs_100000000] (at)
+	add.d f0, f0, f30
+nonnegative:
+	jr ra
+	nop
+
+/* 4294967296 as a float. */
+.sect .rom
+.fs_100000000:
+	.data4 0x4f800000

mach/mips/libem/fd_80000000.s

@@ -1,10 +0,0 @@
-#
-.sect .text; .sect .rom; .sect .data; .sect .bss
-
-/* 2147483648 as a double; used as a pivot for double->unsigned and unsigned->double. */
-
-.sect .rom
-.define .fd_80000000
-.fd_80000000:
-	.data4 0x41e00000, 0
-

mach/mips/libem/fef8.s

@@ -0,0 +1,56 @@
+#
+.sect .text; .sect .rom; .sect .data; .sect .bss
+
+/* Split a double-precision float into fraction and exponent, like
+ * frexp(3) in C, http://en.cppreference.com/w/c/numeric/math/frexp
+ *
+ * Stack: ( double -- fraction exponent )
+ */
+
+#define DBL_EXPBITS	11
+#define DBL_FRACHBITS	20
+#define DBL_FRACLBITS	32
+#define DBL_FRACBITS	52
+#define DBL_EXP_INFNAN	2047
+
+#define DBL_EXP_BIAS 1023
+
+.sect .text
+.define .fef8
+.fef8:
+	lw r4, 0(sp)			! r4 = low word (bits 0..31)
+	lw r5, 4(sp)			! r5 = high word (bits 32..63)
+
+	! IEEE double = sign * 1.fraction * 2**(exponent - 1023)
+	!   sign  exponent  fraction
+	!   31     30..19     18..0, 31..0
+	!
+	! IEEE exponent = 1022 in [0.5, 1) or (-1, -0.5].
+
+	ext r7, r5, DBL_FRACHBITS, DBL_EXPBITS ! r7 = IEEE exponent
+	beq r7, zero, zeroexp   ! this number is zero or denormalised, treat specially
+	nop
+
+	li at, DBL_EXP_INFNAN
+	beq r7, at, return      ! just return if infinity or NaN
+	nop
+
+	addiu r7, r7, -[DBL_EXP_BIAS-1] ! undo exponent bias
+	li at, DBL_EXP_BIAS-1
+	ins r5, at, DBL_FRACHBITS, DBL_EXPBITS ! replace exponent
+return:
+	addiu sp, sp, -4        ! returning one more quad than we got
+	sw r5, 8(sp)
+	sw r6, 4(sp)
+	sw r7, 0(sp)
+	jr ra
+	nop
+
+	/* We received a denormalised number or zero. */
+zeroexp:
+	/* TODO: we just assume that the number is zero here. */
+	mov r5, zero
+	mov r6, zero
+	mov r7, zero
+	b return
+	nop

mach/mips/libem/ff_80000000.s

@@ -1,10 +0,0 @@
-#
-.sect .text; .sect .rom; .sect .data; .sect .bss
-
-/* 2147483648 as a float; used as a pivot for double->float and unsigned->float. */
-
-.sect .rom
-.define .ff_80000000
-.ff_80000000:
-	.data4 0x4f000000
-