You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When working in Metal (and I presume other languages), there are times where you want to work in say, Float 16 (Half 4) rather than full Float since on Apple Silicon that type is more performant.
While I can change the types for some classes of functions when I use them using say
#define CHROMAAB_TYPE half3
There is a widely held assumption (at least in the sampler folder) that most samplers will return float type:
When working in Metal (and I presume other languages), there are times where you want to work in say, Float 16 (Half 4) rather than full Float since on Apple Silicon that type is more performant.
While I can change the types for some classes of functions when I use them using say
#define CHROMAAB_TYPE half3
There is a widely held assumption (at least in the sampler folder) that most samplers will return float type:
float4 sampleClamp2edge(SAMPLER_TYPE tex, float2 st, float2 texResolution)
for example.
We may want to introduce a define
which we can use like
SAMPLER_PRECISION sampleClamp2edge(SAMPLER_TYPE tex, float2 st, float2 texResolution)
While will allow callees to adjust the sampler type and precision like so:
To opt into fast path on their hardware?
The text was updated successfully, but these errors were encountered: