added __llshift_alt

Author: ZERICO2005

src/crt/dtof.src

@@ -75,8 +75,8 @@ __dtof:
 	add	hl, hl
 	add	hl, hl
 	; offset = -$380 - -$47F = $FF = -1 ; therefore decrement
-	; H = exponent + 1
-	ld	l, 29		; f64_mant_bits - f32_mant_bits = 52 - 23 = 29
+	ld	l, h		; L = exponent + 1
+	ld	h, 29		; f64_mant_bits - f32_mant_bits = 52 - 23 = 29
 	ex	(sp), hl	; (SP) = exponent/shift, HL = lo24
 
 	; clear exponent
@@ -101,14 +101,15 @@ __dtof:
 	and	a, $1F
 .L.round_up:
 .L.round_down:
-	call	__llshru
+	call	__llshru_alt
 	; B, C, and UDE are zero here
+	pop	bc	; C = exponent + 1
 	or	a, a
 	jr	z, .L.no_round
 	inc	hl	; does not overflow
 .L.no_round:
-	pop	af	; a = exponent + 1, flags = 29 = ---5H3V-C
-	sbc	a, b	; decrement exponent and clear carry
+	ld	a, c
+	dec	a
 	rra
 	jr	nc, .L.even_exponent
 	ld	bc, $800000
@@ -216,5 +217,5 @@ __dtof:
 
 	.extern	__lland
 	.extern	__llcmpzero
-	.extern	__llshru
+	.extern	__llshru_alt
 	.extern	__lshru

src/crt/dtoll.src

@@ -47,19 +47,19 @@ __dtoull:
 	inc	a	; A is one less than it should be here to allow for the CPL trick in shift_right
 	; C is [16, 31]
 	cp	a, c	; only call __llshl if the shift amount is [0, 63]
-	ld	l, a
+	ld	h, a
 	ex	(sp), hl
 	; shift is non-zero and [1, 11] in the non-UB case
-	call	c, __llshl
+	call	c, __llshl_alt
 	jr	.finish
 .shift_right:
 	; A is [-1, -52]
 	; expon is [0, 51]
 	cpl
-	ld	l, a
+	ld	h, a
 	ex	(sp), hl
 	; shift is [0, 51]
-	call	__llshru
+	call	__llshru_alt
 .finish:
 	pop	af	; reset SP
 	pop	af
@@ -85,5 +85,5 @@ __dtoull:
 	jr	.finish_zero_or_one
 
 	.extern	__llneg
-	.extern	__llshl
-	.extern	__llshru
+	.extern	__llshl_alt
+	.extern	__llshru_alt

src/crt/llshl_alt.src

@@ -0,0 +1,33 @@
+	.assume	adl=1
+
+	.section	.text
+	.global	__llshl_alt
+	.type	__llshl_alt, @function
+
+__llshl_alt:
+; Suboptimal for large shift amounts
+	push	af
+	ld	a, b
+
+	push	hl
+	ld	hl, 10
+	add	hl, sp
+	ld	b, (hl)
+	pop	hl
+
+	inc	b
+	jr	.L.begin
+
+.L.loop:
+	add	hl, hl
+	ex	de, hl
+	adc	hl, hl
+	ex	de, hl
+	rl	c
+	rla
+.L.begin:
+	djnz	.L.loop
+
+	ld	b, a
+	pop	af
+	ret

src/crt/llshr_alt.src

@@ -0,0 +1,62 @@
+	.assume	adl=1
+
+	.section	.text
+
+	.global	__llshru_alt
+	.type	__llshru_alt, @function
+	.global	__llshrs_alt
+	.type	__llshrs_alt, @function
+
+__llshru_alt:
+; Suboptimal for large shift amounts
+	push	af
+	push	iy
+	ld	iy, 0
+	add	iy, sp
+	ld	a, (iy + 10)
+	or	a, a
+	jr	z, .L.finish
+	push	de
+	push	hl
+	srl	b
+	jr	.L.hijack_llshru_alt
+
+__llshrs_alt:
+; Suboptimal for large shift amounts
+	push	af
+	push	iy
+	ld	iy, 0
+	add	iy, sp
+	ld	a, (iy + 10)
+	or	a, a
+	jr	z, .L.finish
+	push	de
+	push	hl
+
+	.local	__llshr_alt_common
+__llshr_alt_common:
+.L.loop:
+	sra	b
+.L.hijack_llshru_alt:
+	rr	c
+	rr	(iy - 1)
+	rr	d
+	rr	e
+	rr	(iy - 4)
+	rr	h
+	rr	l
+	dec	a
+	jr	nz, .L.loop
+
+	ld	(iy - 3), e
+	ld	(iy - 2), d
+	ex	de, hl
+	pop	hl
+	ld	l, e
+	ld	h, d
+	pop	de
+	.local	.L.finish
+.L.finish:
+	pop	iy
+	pop	af
+	ret

src/crt/ltod.src

@@ -63,13 +63,13 @@ __lltod:
 .else
 .L.no_round_inexact:
 .endif
-	ld	h, b
+	ld	l, b
 	ld	a, c
-	ld	l, c
+	ld	h, c
 	pop	bc
 
 	ex	(sp), hl	; (SP) = shift
-	call	__llshru
+	call	__llshru_alt
 	add	a, 51 - 1	; compensate for the implicit mantissa bit
 
 	; BC/exponent = [$434*, $43E*]
@@ -79,11 +79,10 @@ __lltod:
 	add	a, a
 	add	a, c
 	ld	c, a
-	pop	af		; A = rounding
-	or	a, a		; NZ = round-up, Z = no-round
+	pop	af		; C = round-up, NC = no-round
 	ld	b, $43
 .if 0
-	jr	z, .L.int_to_f64_shl.no_round
+	jr	nc, .L.int_to_f64_shl.no_round
 	; inlined __lladd_1
 	inc	hl
 	add	hl, de
@@ -97,7 +96,7 @@ __lltod:
 	inc	bc
 	jr	.L.int_to_f64_shl.finish
 .else
-	call	nz, __lladd_1	; round up to even
+	call	c, __lladd_1	; round up to even
 	jr	.L.int_to_f64_shl.finish
 .endif
 
@@ -138,13 +137,13 @@ __ltod:
 	; Maximum exponent: $434 (2^52)
 	; It is assumed that A is [0, 51] here, or [-52, -1] before adding 52
 	push	hl
-	ld	l, a
+	ld	h, a
 	ex	(sp), hl	; (SP) = shift
-	call	__llshl
-	ex	(sp), hl	; (SP) = shifted HL, L = shift
+	call	__llshl_alt
+	ex	(sp), hl	; (SP) = shifted HL, H = shift
 
 	ld	a, 51
-	sub	a, l
+	sub	a, h
 
 	; exponent = ($400 + (base2_logarithm - 1)) << 4
 	; BC = $4PPM
@@ -173,8 +172,8 @@ __ltod:
 	.extern	__lneg
 	.extern	__lctlz
 	.extern	__llctlz
-	.extern	__llshl
-	.extern	__llshru
+	.extern	__llshl_alt
+	.extern	__llshru_alt
 	.extern	__llneg
 	.extern	__lladd_1
 	.extern	___fe_cur_env

src/softfloat/s_normSubnormalF64Sig.src

@@ -26,9 +26,9 @@ _softfloat_normSubnormalF64Sig:
 	inc	hl
 	ld	(iy + 0), hl
 	; z.sig = sig<<shiftDist;
-	ld	l, a
+	ld	h, a
 	ex	(sp), hl
-	call	__llshl
+	call	__llshl_alt
 	pop	af	; reset SP
 	ld	(iy + 2), hl
 	ld	(iy + 5), de
@@ -37,6 +37,6 @@ _softfloat_normSubnormalF64Sig:
 	lea	hl, iy + 0	; ABI struct ptr
 	ret
 
-	.extern	__llshl
+	.extern	__llshl_alt
 	.extern	__llctlz
 	.extern	__ineg

src/softfloat/s_shiftRightJam64.src

@@ -33,13 +33,14 @@ _softfloat_shiftRightJam64:
 	ld	de, (iy + 7)
 	ld	bc, (iy + 10)
 	jr	nc, .L.overflow_shift
+	ld	h, l
 	push	hl	; >> shift
 	sub	a, l	; A = 0 - L
 	and	a, $3F
-	ld	l, a
+	ld	h, a
 	push	hl	; << shift
 	ld	hl, (iy + 4)
-	call	__llshl
+	call	__llshl_alt
 	xor	a, a
 	call	__llcmpzero
 	jr	z, .L.no_round
@@ -49,7 +50,7 @@ _softfloat_shiftRightJam64:
 	ld	hl, (iy + 4)
 	ld	de, (iy + 7)
 	ld	bc, (iy + 10)
-	call	__llshru
+	call	__llshru_alt
 	or	a, l
 	ld	l, a
 	pop	af	; reset SP
@@ -70,5 +71,5 @@ _softfloat_shiftRightJam64:
 	ret
 
 	.extern	__llcmpzero
-	.extern	__llshl
-	.extern	__llshru
+	.extern	__llshl_alt
+	.extern	__llshru_alt