Difference between ARM ADD with 2 or 3 operands?

Question

Looking into the documentation of armkeil http://www.keil.com/support/man/docs/armasm/armasm_dom1361289861747.htm you always need 3 variables when you do an ADD function, for example:

"MOV r2,#4 \n\t"
"ADD r2,r2,#3 \n\t"

this has an outcome of 7 in register 2. But when I try the following:

"MOV r2,#4 \n\t"
"ADD r2,#3 \n\t"

The outcome is the same, what is the fundamental difference between those 2? Is it allowed to use the second one?

Answer 1

ARM syntax allows omitting the middle operand (first source) when it's the same as the destination. (At least in GAS "Unified" syntax; originally ARM Thumb syntax only allowed it when the machine encoding actually had a separate destination, or something. GAS has different syntaxes; GCC normally puts it in .syntax unified mode IIRC, so that's the context for inline asm)

ADD r2,r2,#3 is exactly identical to ADD r2,#3 for assemblers that accept both. Keil's assembler may not, IDK.

Keil vs. GAS / GCC use different directives, but mostly the same syntax for instructions I think.

---

Some thumb instruction encodings only have room to encode 2 operands, so they are like dst += immediate not dst = src + immediate at the machine-code level. At the asm source level, Keil documents it as ADDS Rd, Rd, #imm (in the page you linked).

The other 16-bit encoding of adds only has a 3-bit immediate (0..7), but uses those bits to allow encoding a separate destination, adds Rd, Rn, #imm.

(Note that add instead of adds generally requires a 32-bit instruction, so don't do that unless you need to preserve flags.)

I mention this because your previous questions were about Cortex-M. GAS probably still accepts add r2, #3 when assembling in ARM mode. The first source is just implicitly the destination.