implemented dtol and dtoll in assembly

Author: ZERICO2005

src/crt/dtol.src

@@ -5,26 +5,17 @@
 	public	__dtol
 	public	__dtoul
 
-; __dtol_c correctly handles all non-UB cases for both
+; correctly handles all non-UB cases for both
 ; (long)long double and (unsigned long)long double
 __dtol:
 __dtoul:
-	; f64_ret_i32
-	push	af, iy
-	ld	a, b
-	push	bc, de, hl
-	ld	hl, 7
-	add	hl, sp
-	res	7, (hl)	; fabsl(x)
-	inc	hl
-	rlca
-	ld	(hl), a	; store the sign of x in the padding byte
-	call	__dtol_c
-	pop	af
-	ld	a, e
+	push	bc
+	push	de
+	call	__dtoll	; same as __dtoull
+	ld	c, e
 	pop	de
-	ld	e, a
-	pop	bc, iy, af
+	ld	e, c
+	pop	bc
 	ret
 
-	extern	__dtol_c
+	extern	__dtoll

src/crt/dtoll.src

@@ -5,21 +5,77 @@
 	public	__dtoll
 	public	__dtoull
 
-; __dtoll_c correctly handles all non-UB cases for both
+; correctly handles all non-UB cases for both
 ; (long long)long double and (unsigned long long)long double
 __dtoll:
 __dtoull:
-	; f64_ret_i64
-	push	af, iy
-	ld	a, b
-	res	7, b	; fabsl(x)
-	push	bc, de, hl
-	ld	hl, 8
-	add	hl, sp
-	rlca
-	ld	(hl), a	; store the sign of x in the padding byte
-	call	__dtoll_c
-	pop	af, af, af, iy, af
+	bit	6, b	; set if |x| >= 2.0L
+	jr	z, .zero_or_one
+	bit	7, b
+	push	af
+	res	7, b
+	push	hl
+	; -((Float64_mant_bits + Float64_bias) << 4)
+	ld	hl, $FFC010 ; -16368 ; -$3FF0
+
+	; clears the exponent field without touching the mantissa
+	; sets the LSB of the exponent since x is normalized
+	ld	a, c
+	or	a, l	; or a, $10
+	and	a, $1F
+
+	add	hl, bc
+	; HL <<= 4
+	add	hl, hl
+	add	hl, hl
+	add	hl, hl
+	add	hl, hl
+
+	ld	c, a
+	ld	b, 0
+	ld	a, h
+	sub	a, 52 + 1	; float64 mantissa bits
+	jr	c, .shift_right
+	; shift_left
+	; expon >= 52 or [52, 63]
+	; A is one less than it should be here to allow for the CPL trick in shift_right
+	; A is [-1, 10]
+	cp	a, 11	; only call __llshl if the shift amount is [0, 63]
+	inc	a	; positioning inc a after cp a allows __llshl to be skipped when the shift amount is zero
+	ld	l, a
+	ex	(sp), hl
+	call	c, __llshl
+	jr	.finish
+.shift_right:
+	; expon is [0, 51]
+	cpl
+	ld	l, a
+	ex	(sp), hl
+	call	__llshru
+.finish:
+	pop	af	; reset SP
+	pop	af
+.finish_zero_or_one:
+	jp	nz, __llneg
 	ret
 
-	extern	__dtoll_c
+.zero_or_one:
+	ld	hl, 16
+	ld	d, h
+	ld	e, h
+	add	hl, bc	; adds one to the exponent
+	bit	6, h	; if |x| was [1, 2)
+	jr	z, .zero
+	inc	de
+	bit	7, b	; sets NZ if the result should be -1
+.zero:
+	ld	c, d
+	ld	b, d
+	ld	h, d
+	ld	l, d
+	ex.s	de, hl
+	jr	.finish_zero_or_one
+
+	extern	__llneg
+	extern	__llshl
+	extern	__llshru

src/crt/float64_to_int.c

@@ -96,6 +96,8 @@ static uint64_t f64_to_unsigned(F64_pun val) {
     return val.bin;
 }
 
+
+
 /**
  * @brief the exact same routine is used for (long long)long double and
  * (unsigned long long)long double. If the input long double is out of range,