1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
|
<h1><a name="Tricky_Reassociation_of_Addressi"> </a> Tricky Reassociation of Addressing Modes </h1>
<p>
Given an expression like
</p><pre>r1 + 2*(r2 + 1234)
</pre>
we can reorganize it to fit in a single addressing mode, like this
<pre>r1 + 2*r2 + 2*1234
</pre>
Currently, I use four grammar rules to catch all the feasible cases:
<pre>addr : PLUS_DI(base64 | MULT_DI(PLUS_DI(r64 | imm8) | mcon))
addr : PLUS_DI(base64 | MULT_DI(PLUS_DI(r64 | imm32) | mcon))
addr : PLUS_DI(base64 | ASHIFT_DI(PLUS_DI(r64 | imm8), scon))
addr : PLUS_DI(base64 | ASHIFT_DI(PLUS_DI(r64 | imm32), scon))
</pre>
where mcon = 1, 2, 4, or 8 and scon = 0, 1, 2, or 3.
<p>
This is a little sloppy, since a imm32*mcon may overflow it's 4-byte encoding.
To avoid this problem, we need to introduce 3 new non-terminals: imm29, imm30, and imm31, with associated rules
</p><pre>addr : PLUS_DI(base64 | MULT_DI(PLUS_DI(r64 | imm29) | CONST_P8))
addr : PLUS_DI(base64 | MULT_DI(PLUS_DI(r64 | imm30) | CONST_P4))
addr : PLUS_DI(base64 | MULT_DI(PLUS_DI(r64 | imm31) | CONST_P2))
addr : PLUS_DI(base64 | ASHIFT_DI(PLUS_DI(r64 | imm29), CONST_P3))
addr : PLUS_DI(base64 | ASHIFT_DI(PLUS_DI(r64 | imm30), CONST_P2))
addr : PLUS_DI(base64 | ASHIFT_DI(PLUS_DI(r64 | imm31), CONST_P1))
</pre>
We'll need to do something analogous for the imm8. While it can't
really overflow, the cost (size in bytes) can become incorrect if we're
not careful. Therefore, need to introduce 3 more non-terminals: imm5,
imm6, and imm7, with associated rules
<pre>addr : PLUS_DI(base64 | MULT_DI(PLUS_DI(r64 | imm5) | CONST_P8))
addr : PLUS_DI(base64 | MULT_DI(PLUS_DI(r64 | imm6) | CONST_P4))
addr : PLUS_DI(base64 | MULT_DI(PLUS_DI(r64 | imm7) | CONST_P2))
addr : PLUS_DI(base64 | ASHIFT_DI(PLUS_DI(r64 | imm5), CONST_P3))
addr : PLUS_DI(base64 | ASHIFT_DI(PLUS_DI(r64 | imm6), CONST_P2))
addr : PLUS_DI(base64 | ASHIFT_DI(PLUS_DI(r64 | imm7), CONST_P1))
|
