Re: Building llama.cpp for AMD GPU using only Debian packages?

[Cordell Bloor <cgmb@slerp.xyz>]

A closer examination confirms that this is caused by the -mf16c flag in 
rocblas.

On 2024-07-01 07:56, Cordell Bloor wrote:
> vex amd64->IR: unhandled instruction bytes: 0xC4 0xE2 0x79 0x13 0xC0 
> 0xC5 0xF0 0x57 0xC9 0xC5

The disassembly of those bytes [1]:

0:  c4 e2 79 13 c0          vcvtph2ps xmm0,xmm0
5:  c5 f0 57 c9             vxorps xmm1,xmm1,xmm1
9:  c5                      .byte 0xc5

Sincerely,
Cory Bloor

[1]: https://defuse.ca/online-x86-assembler.htm