w2yehia wrote:
> It doesn't look like we create a clone for pwr6 and pwr4. Those are no longer
> supported so it makes sense but is it a good idea to add a follow on PR to
> warn that using those won't do anything?
In my build of this PR, I see the clones:
```
clang --target=powerpc-ibm-aix-xcoff tt.c -emit-llvm -S -o -|grep "define.*cpu"
define internal i32 @compute_sum.cpu_pwr9(ptr noundef %arr, i32 noundef %size)
#0 {
define internal i32 @compute_sum.cpu_pwr7(ptr noundef %arr, i32 noundef %size)
#1 {
define internal i32 @compute_sum.cpu_pwr6(ptr noundef %arr, i32 noundef %size)
#2 {
define internal i32 @compute_sum.cpu_pwr4(ptr noundef %arr, i32 noundef %size)
#3 {
define internal i32 @compute_sum.cpu_pwr8(ptr noundef %arr, i32 noundef %size)
#4 {
```
> But using arch instead of cpu causes a crash:
We don't support the `-march` flag on AIX, neither does GCC:
```
> clang tt.c -march=power7
clang: error: unsupported option '-march=' for target
'powerpc64le-unknown-linux-gnu'
> clang --target=powerpc-ibm-aix-xcoff tt.c -march=power7
clang: error: unsupported option '-march=' for target 'powerpc-ibm-aix-xcoff'
> gcc -c -march=power7 tt.c
gcc: error: unrecognized command-line option "-march=power7" did you mean
"-mcpu=power7"?
```
but we shouldn't crash, I'll make the FE error on "arch=<cpu>" in the
target_clones.
GCC errors too:
```
$ gcc -c tt.c
tt.c:5:5: error: __attribute__((__target__("arch=power9"))) is invalid
5 | int compute_sum(int *arr, int size) {
| ^~~~~~~~~~~
```
https://github.com/llvm/llvm-project/pull/177428
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
