Header:

0-3:Hmr<version>
version=0 for normal, 1 for fastspr
4-7:Size of decomped file
8+: data:

<type> <data> <data> <data> ...

Byte info stuff:
<&10 type 1
<&20 type 2
<&40 type 3
<&80 type 4
&FF overrun error (?)
other type 5

Type 1:
<type>
<date>

If <type><15 then repeat <data> <type> + 2 times
Else if <type>=15 then repeat <data> 256 times

Type 2:
<type>
<data>
<data>
<data>
...

Read (<type> & F) + 1 bytes of data straight to output

Type 3:
<type>
<data>

<type> = 000CCCHH
<data> = LLLLLLLL
P = <current output pos> - HHLLLLLLLL
Output CCC+2 bytes from P, subtracting 1 from P each loop

Type 4:
<type> = 00CCCAAF
<data> = BBBBBBBB
If AA>=2 then D=AA+1 else D=AA
D = 1 << D
P = <current output pos> - FBBBBBBBB
S = 0
Output CCC+2 bytes from P+(S >> 2) where S=S+D each loop

Type 5:
<type> = 0CCCCAAA
<data> = BBBBBBBB
P = <current output pos> - AAABBBBBBBB
Output CCCC+2 bytes from P, adding 1 to P each loop

<current output pos> is the pos of the last byte output, not the next free spot

Need output buffer of 4k to work properly

version 1:

a=offset from file start
b=a<<2
a=a-len>>2
if a>=0 then b++,b=b-len
a=a-len>>2
if a>=0 then b++,b=b-len
a=a-len>>2
if a>=0 then b++,b=b-len
return b as address to use

do above translation on formats 3,4,5 just before loading, and just before output on *all* formats. len is the real length rounded up to a whole number of words.