error casting between dtypes.double and dtypes.half on CLANG

[chenyuxyz]

probably target dependent, __fp16 is for ARM only? but CI passes too

example, with CLANG=1

a = Tensor([1, 2, 3], dtype=dtypes.double)
print(a.numpy())
print(a.cast(dtypes.half).numpy())

source

#include <stdbool.h>
#include <tgmath.h>
#define max(x,y) ((x>y)?x:y)
#define half __fp16
void E_3(half* restrict data0, const double* restrict data1) {
  for (int ridx0 = 0; ridx0 < 3; ridx0++) {
    double val0 = data1[ridx0];
    data0[ridx0] = (half)(val0);
  }
}

on mac m1 max

[1. 2. 3.]
[1. 2. 3.]

on tinybox, ubuntu 22.04

[1. 2. 3.]
[0.00e+00 0.00e+00 3.05e-05]

[mooihi]

The new runners do not have gcc-13 installed, this is inline with Ubuntu 22.04 default gcc package versions (up to gcc-12). The old runner had gcc-13 installed and clang seemed to be able to use those library files and compile properly.

EDIT: Having libgcc-s1=13.1.0-8ubuntu1~22.04 makes the dynamic linking works properly but libgcc-s1=12.3.0-1ubuntu1~22.04 does not.

[chenyuxyz]

fixed in #8265