PowerPC and the clrlslwi -or- rlwinm instruction

Question

I've read the other posts here, but still have a question about PPC and rlwinm.

The example I'm looking at is:

li  r0, 0x100
clrlslwi  r9, r0, 27,5 (which afaik is the same as rlwinm r9, r0,5,22,26)

First I load r0 = 0x100 Then Left Shift r0 by 5 so r0 = 0x2000 (No overflow, so no rotate needed)

Make a Mask with bits 22 to 26 set = 0x3e0

AND r0 with the Mask = 0x2000 AND 0x3E0 = 0

I must be doing something wrong...this would always net 0 What am I missing?

Allisonallissa

You should format your question (wrap inline code segments in backticks, prefix code blocks with four spaces) and include the actual code you're having trouble with. – Cristobal

Yah, the value is always 0. So is 2 - 2. What's the problem? – Parodic

Is it just me, or are PPC mnemonics way harder to decipher than most other architectures? It was pretty obvious from looking at compiler output what most of the insns do in something like ARM asm, even before I learned much of anything about ARM. But PPC just looks like gibberish. – Tamayo

@PeterCordes: Well, ARM used the excess bits in their 32-bit instructions for predicated execution, PPC used them to make instructions do more complex operations. Guess which ISA needed to be completely redone for 64-bit. – Batfowl

Your analysis looks correct. But what's the question? – Drowsy

My question is, since this is just going to result in 0, why so many instructions? Like RR said above, why not just li r0,0 ? I guess this just showing how inefficient the compiler used to make this code is. – Allisonallissa

Answer 1

yes that's true (Clear left and shift left word immediate)

clrlslwi rA,rS,b,n (n ≤ b ≤ 31) = rlwinm rA,rS,n,b – n,31 – n

so clrlslwi r9, r0, 27,5 = rlwinm r9, r0, 5, 22(27-5), 26(31-5)

Windbag