Y Khan writes: > Please let me know if this is not the right place to > ask this question. OK, a bit off-topic, but ask away. > I had read somewhere that a 32 bit processor can > read/write only and only 32 bits in any one cycle. That's not strictly true, as most 32-bit processors these days are multiple-issue, but let's carry on. > So, I thought that if I needed a 'char' then 32 bits > would be read from memory into registers and then > would be manipulated there to get a 'char'. As such I > thought that it was logical that operations on int are > faster than on chars. Not really, no, because reading a byte and extending it to 32 bits can be done in a single operation. > I was reading an ARM processor manual some time ago > and I saw that processor has instructions to read a > single byte too. > > So, I am confused now. I earlier thought that the > compiler generates such code that reads in 32 bits and > then manipulates those bits to get an 8-bit char > value. That's not true. > Now, ARM processor manual says that the processor can > be instructed to read only one byte too. But the > processor registers are 32 bits. so, this is again > confusing. > > This doesn't look very straightforward to me because > if the data bus is 32 bits wide then probably we are > still reading 32 bits. > > So, the question is: > > 1. Can only 8 bits be read from memory in a 32 bit > register? Yes. > 2. When a processor gives an instruction to read only > 8 bits, then is it manipulating the 32 bits by itself? No. > 3. If the address of my 'char' is 0x5, then do we only > read the memory location 0x5, or from 0x4 to 0x7, and > then manipulate it to get the value at 0x5 in the > lowest 8 bits of the 32 bit register. OK, this is how it usually works: The processor reads a whole word. The word is loaded into a barrel shifter (see http://en.wikipedia.org/wiki/Barrel_shifter) and the byte we want is shifted to the correct position in the word. The high-order bits in the word are masked to zero (or the byte is sign extended). Writing a single byte is harder: we either have to do a read/modify/write cycle or use byte select lines to the memory devices. This is where caches come into play: it is very likely that the memory word we want to write is already in the cache, so even if we do want to do a read/modify/write, that doesn't involve a read access to external memory. Andrew.
