What is the faster way to add byte before each byte in char *

If I have a char* for example :

Code:

---

{0x01,0x02,0x03,0x04,0x05}

---

and I want to add 0x00 before each byte .

the result will be :

Code:

---

{0x00,0x01,0x00,0x02,0x00,0x03,0x00,0x04,0x00,0x05}

---

What is the faster way to do that ?

Probably picking up more than one character at a time, which would be processor dependent. In assembly on X86 with xmm registers, you could load 16 bytes at a time, copy to another xmm register, then use pshufb on both xmm registers to insert the zeroes, then store 16 bytes at a time (twice).

> What is the faster way to do that ?
Faster than what?
You haven't shown us what you're doing at the moment.

Unless you have some woefully underpowered micro-controller, or needing to do 100's of MB/sec, this looks like a premature optimisation.

Quote:

---

Originally Posted by rcgldr

Probably picking up more than one character at a time, which would be processor dependent. In assembly on X86 with xmm registers, you could load 16 bytes at a time, copy to another xmm register, then use pshufb on both xmm registers to insert the zeroes, then store 16 bytes at a time (twice).

---

For those of us not familiar with assembly, couldn't the same be achieved by memcpy'ing into a uint16_t pointer? Would probably need to account for endianness though.





EDIT: offtopic, BTW, do u guys know if the forums were under ddos attack? It was hard to access the site the past 24 hours. I was getting this error:Attachment 15622

Quote:

---

Originally Posted by Dren

For those of us not familiar with assembly, couldn't the same be achieved by memcpy'ing into a uint16_t pointer? Would probably need to account for endianness though.

---

Not as fast. The instruction pshufb will shuffle the bytes and insert 0's in 1 cycle, and the operations on two separate xmm registers occur in parallel.