1 |
Shawn Haggett wrote: |
2 |
> |
3 |
> There's two points that come to mind. |
4 |
> |
5 |
> 1) mtune is a request for the compiler to make the code more suited to |
6 |
> the given processor, but without breaking compatibility. march is |
7 |
> telling the compiler, do everything you can to make this code fastest |
8 |
> on this processor. |
9 |
> |
10 |
> From the GCC docs for 4.2.3: |
11 |
> "-mtune=cpu-type: Tune to cpu-type everything applicable about the |
12 |
> generated code, except for the ABI and the set of available |
13 |
> instructions." |
14 |
> "-march=cpu-type: Generate instructions for the machine type cpu-type. |
15 |
> The choices for cpu-type are the same as for -mtune. Moreover, |
16 |
> specifying -march=cpu-type implies -mtune=cpu-type." |
17 |
> |
18 |
> So mtune shouldn't be using any instructions that are in K-6 that |
19 |
> weren't in a 386. |
20 |
> |
21 |
> 2) I believe x86 hardware never goes backwards. That is, if a new |
22 |
> feature is added, all future versions of the chip have that feature, |
23 |
> just with more added. Of course Intel and AMD both have their separate |
24 |
> additions, but since your staying with AMD, moving to a new processor |
25 |
> shouldn't break anything (even if you had used march). |
26 |
> |
27 |
> Disclaimer: I'm not an expert on hardware architectures or compilers, |
28 |
> so I might be wrong. |
29 |
> |
30 |
> Shawn |
31 |
|
32 |
Thanks Shawn, that's probably the best answer I'm going to get, I doubt |
33 |
many of the AMD chip designers hang around here... :) |