# Rules for optimizing BCC assembler output

# Rules for loading short variables

mov	%0$0[%2],%3
=
mov	%0[%2],%3

mov	%2,%[ax|bx|cx|dx]1
mov	%[ax|bx|cx|dx]1,%2
=
mov	%2,%1

# Breaks:  a = b = 125;
# mov	%[ax|bx|cx|dx]3,%[#|*]0%1
# mov	%2[%4],%[ax|bx|cx|dx]3
# =
# mov	word ptr %2[%4],%0%1

# This gets a few back.
mov	%[ax|bx|cx|dx]3,%[#|*]0%1
mov	%2[%4],%[ax|bx|cx|dx]3
mov	%[ax|bx|cx|dx]3,%5
=
mov	word ptr %2[%4],%0%1
mov	%3,%5

mov	%[ax|bx|cx|dx]3,%[#|*]0%1
mov	%[ax|bx|cx|dx]2,%[ax|bx|cx|dx]3
=
mov	%2,%0%1

# Breaks: a = b = 'a';
# mov	al,%[#|*]0%1
# mov	%2,al
# =
# mov	byte ptr %2,%0%1

xor	ax,ax
mov	bx,%[#|*]0%1
mov	[%2],ax
mov	[%2+2],bx
=
mov	word ptr [%2],#0
mov	word ptr [%2+2],%0%1

mov	ax,%[#|*]0%1
xor	bx,bx
mov	[%2],ax
mov	[%2+2],bx
=
mov	word ptr [%2],%0%1
mov	word ptr [%2+2],#0

mov	ax,%[#|*]4%1
mov	bx,%[#|*]5%3
mov	[%2],ax
mov	[%2+2],bx
=
mov	word ptr [%2],%4%1
mov	word ptr [%2+2],%5%3


# Rules for incrementing short variables

mov	%[ax|bx]2,%1
%[inc|dec]3	%[ax|bx]2
mov	%1,%[ax|bx]2
=
%3	word ptr %1
mov	%2,%1


# Rules for manipulating characters

inc	%[si|di]*
inc	%[si|di]*
mov	al,-1[si]
mov	-1[di],al
=
lodsb
stosb

inc	%[si|di]*
inc	%[si|di]*
mov	al,-1[di]
mov	-1[si],al
=
xchg	si,di
lodsb
stosb
xchg	si,di

inc	si
mov	al,-1[si]
=
lodsb

inc	di
mov	-1[di],al
=
stosb

dec	%[si|di]*
dec	%[si|di]*
mov	al,1[si]
mov	1[di],al
=
std
lodsb
stosb
cld

dec	%[si|di]*
dec	%[si|di]*
mov	al,1[di]
mov	1[si],al
=
std
xchg	si,di
lodsb
stosb
xchg	si,di
cld

dec	si
mov	al,1[si]
=
std
lodsb
cld

dec	di
mov	1[di],al
=
std
stosb
cld


# Rules for manipulating short values

# This is the broken rule - Chad

#mov	%[ax|bx]2,%1
#%[add|and|xor|sub|or]4	%[ax|bx]2,%3
#mov	%1,%[ax|bx]2
#=
#mov	%2,%3
#%4	%1,%2

mov	%[ax|bx]2,%1
%[add|and|xor|or]4	%[ax|bx]2,%3
mov	%3,%[ax|bx]2
=
mov	%2,%1
%4	%3,%2

mov	%[ax|bx]4,%1
%[add|and|xor|sub|or]2	%[si|di]3,%[ax|bx]4
=
%2	%3,%1

mov	al,%1
xor	ah,ah
%[add|and|xor|sub|or]2	ax,%[#|*]0%3
mov	%1,al
=
%2	byte ptr %1,%0%3


# Rules for comparing short values

mov	%[ax|bx]3,%1[%4]
cmp	%[ax|bx]3,%[#|*]0%2
=
cmp	word ptr %1[%4],%0%2

mov	al,%1[%4]
cmp	al,%[#|*]0%2
=
cmp	byte ptr %1[%4],%0%2

mov	%[ax|bx]3,%[ax|bx|cx|dx|si|di]2
cmp	%[ax|bx]3,%1
=
cmp	%2,%1


# General comparison rules

xor	%1,%1
mov	%2,%3
cmp	%2,%1
=
mov	%2,%3
cmp	%2,#0

mov	%1,%[#|*]0%2
mov	%3,%4
cmp	%3,%1
=
mov	%3,%4
cmp	%3,%0%2

mov	%1,%2
cmp	%1,%[#|*]00
%[jne |je  ]4	%3
=
mov	%1,%2
test	%1,%1
%4	%3

%[and|xor|or]1	%2,%3
test	%2,%2
=
%1	%2,%3


# General incrementation and decrementation rules

inc	%3 ptr %1
mov	%2,%1
dec	%2
=
mov	%2,%1
inc	%3 ptr %1

dec	%3 ptr %1
mov	%2,%1
inc	%2
=
mov	%2,%1
dec	%3 ptr %1

mov	%1,%2
inc	%3 ptr %2
push	%1
=
push	%3 ptr %2
inc	%3 ptr %2

mov	%1,%2
dec	%3 ptr %2
push	%1
=
push	%3 ptr %2
dec	%3 ptr %2


# Misc. rules

push	word %[#|*]0%1
inc	%[si|di]2
inc	%[si|di]2
mov	%[ax|bx|cx|dx]3,%*[bp]
mov	%4[%[si|di]2],%[ax|bx|cx|dx]3
inc	sp
inc	sp
=
mov	word ptr %(%4+2)[%2],%0%1
inc	%2
inc	%2

push	%1
mov	%[ax|bx|cx|dx]3,%2
%[add|sub|and|xor|or]4	%[ax|bx|cx|dx]3,%*[bp]
inc	sp
inc	sp
=
mov	%3,%2
%4	%3,%1

push	%1
mov	%[ax|bx|cx|dx]3,%2
%[add|sub|and|xor|or]5	%[ax|bx|cx|dx]3,%6
%[add|sub|and|xor|or]4	%[ax|dx|cx|dx]3,%*[bp]
inc	sp
inc	sp
=
mov	%3,%2
%5	%3,%6
%4	%3,%1

push	%1
mov	%[ax|bx|cx|dx]3,%*[bp]
%[add|sub|and|xor|or]4	%2,%[ax|bx|cx|dx]3
inc	sp
inc	sp
=
mov	%3,%1
%4	%2,%3

mov	%1,%[ax|bx|cx|dx]2
push	%1
=
mov	%1,%2
push	%2

# Rules to remove superflous mov instructions

mov	%1,%2
mov	%1,%2
=
mov	%1,%2

mov	%1,%2
.%3:
mov	%1,%2
=
.%3:
mov	%1,%2

mov	%1,%2
mov	%2,%1
=
mov	%1,%2

mov	%1,%[ax|bx|cx|dx|eax|ebx|ecx|edx]2
mov	%[ax|bx|cx|dx|eax|ebx|ecx|edx]3,%1
=
mov	%1,%2
mov	%3,%2

mov	%1,%1
=
!mov	%1,%1

# Sometimes the compiler puts a jump right after a ret or another jump
# (in if-else-if constructions). Eliminate the second unnecessary jump.
# Also it puts a jump to the next instruction in extensive if-else
# constructions. That jump can also be removed.

jmp	.%1
jmp	.%2
=
jmp	%1

ret
jmp	%2
=
ret

jmp .%1
.%1:
=
.%1:

jmp .%1
.%2:
.%1:
=
.%2:
.%1:

jmp .%1
.%3:
.%2:
.%1:
=
.%3:
.%2:
.%1:


# When using register variables the compiler initializes them when
# they are found in the source code, and sets up the stack frame for
# other local variables lateron. Move initialization of register variables
# behind the add-sp instruction to make further optimization easier.

proc_start
mov	%[si|di]3,%1
add	sp,*%2
=
proc_start
add	sp,*%2
mov	%3,%1

proc_start
mov	%[si|di]3,%1
dec	sp
dec	sp
=
proc_start
dec	sp
dec	sp
mov	%3,%1

add	sp,*%1
add	sp,*%2
=
add	sp,*%1+%2

dec	sp
dec	sp
add	sp,*%1
=
add	sp,*%1-2

add	sp,*%1
dec	sp
dec	sp
=
add	sp,*%1-2

inc	sp
inc	sp
%1	%2
inc	sp
inc	sp
=
%1	%2
add	sp,*4

inc	sp
inc	sp
%1	%2
add	sp,*%3
=
%1	%2
add	sp,*%3+2

add	sp,*%3
%1	%2
inc	sp
inc	sp
=
%1	%2
add	sp,*%3+2

add	sp,*%3
%1	%2
add	sp,*%4
=
%1	%2
add	sp,*%3+%4

# And a great bug floating push/pop
push	dx
push	cx
push	bx
push	ax
pop	ax
pop	bx
pop	cx
pop	dx
=
! push/pop