23:05karolherbst: is there a list somewhere to check what vec2 fp16 instructions AMD GPUs support?
23:12pendingchaos: aco_nir_op_supports_packed_math_16bit(), the VOPP instructions in aco_opcodes.py or the ISA docs should work
23:26karolherbst: I see thanks. I was looking into vec2 ldexp, but seems like LLVM haandles it, and aco simply gives me wrong results
23:40karolherbst: mhh looks like ldexp is the only instruction which has different dneorm handling than everything else? weird