Modified UnrealMathNeon.h and put this in. For some reason vcvt_f16_f32 is no longer available. So I’m thinking that assigning f16x4[0…3] to Vec[0…3] may function correctly
Since it’s a vector that stores 4 floats, if I’m not mistaken. Could be really wrong, but it seems to compile for mobile with this fine.:
Code:
template <bool bAligned>
FORCEINLINE void VectorStoreHalf4(VectorRegister Vec, void* RESTRICT Ptr)
{
float16x4_t f16x4;
for (int x=0;x<4;x++)
{
f16x4[x] = Vec[x];
}
if (bAligned)
{
vst1_u8( (uint8_t *)Ptr, f16x4 );
}
else
{
uint32_t buf[2];
vst1_u8( (uint8_t *)buf, f16x4 );
*(float32_t *)Ptr = buf[0];
}
}