The ARM processor (Thumb-2), part 7: Bitwise operations
The ARM processor offers the following bitwise operations:
; bitwise and
and Rd, Rn, op2 ; Rd = Rn & op2
; bitwise or
orr Rd, Rn, op2 ; Rd = Rn | op2
; bitwise exclusive or
eor Rd, Rn, op2 ; Rd = Rn ^ op2
; bitwise not
mvn Rd, op2 ; Rd = ~op2
; bitwise and not ("bit clear")
bic Rd, Rn, op2 ; Rd = Rn & ~op2
; bitwise or not
orn Rd, Rn, op2 ; Rd = Rn | ~op2
; all support the S suffix
For bit-testing purposes, there are also discarding versions:
; test for equivalence
teq Rn, op2 ; set flags for Rn ^ op2
; test
tst Rn, op2 ; set flags for Rn & op2
For bitwise operations that set flags, the negative (N) and zero (Z) flags reflect the result, the carry (C) flag reflects any shifting that occurred during the calculation of op2 (noting that calculating constants may also involve shifting, as noted earlier when we discussed constants), and the overflow (V) flag is unchanged.
I don’t see much value in the TEQ instruction. It sets the Z flag the same way as the the CMP instruction. I guess you could use it to see if two registers have the same sign bit, since it sets N based on the exclusive-or of the two inputs. I guess that’s handy when calculating the sign of emulated multiplication or division, but even in those cases, you aren’t going to jump based on the sign; you’re going to save the sign of the result for later application, so you would be better off with the EOR instruction anyway.
Okay, well, you can use the LSL shift on the second register argument in order to compare the high bit of one register with an arbitrary bit of another.
teq Rn, Rm LSL #n ; compare Rn bit 31
; and Rm bit 31 - n
bmi same ; branch if different
Still not particularly compelling. Maybe there’s some specialized workflow where this is useful, like cryptography?
Next time, we’ll look at the bit shifting instructions.
Light
Dark
3 comments
Possible uses of the TEQ instruction:
Not setting the carry flag when you’re not shifting an argument. Never setting the overflow flag.
Directly setting flags. When you use R15 as the LHS of a TEQ, the flags get masked out, so the top bits of the result are always that of the RHS. But when you use R15 as the result of a TEQ (i.e. TEQP), then (rather than the flags of the result) the result gets copied to the flags of R15. As a simple case, TEQP R15, Rn copies the flag bits from Rn to R15.
I see ARM has a bit-reversal instruction these days but without that I have a vague memory of some code that used TEQP to check whether two bits were different and following that up with an EORN to flip both bits at once.
Thumb-2 explicitly disallows pc and sp as destinations for the TEQ instruction. Even in classic ARM, using pc as the destination register is now deprecated. The flag bits haven’t been kept in pc since ARM 3, and the masking/copying behavior you describe no longer exists.
Ah, that explains it – I only got to use ARM 2.