search

Protofall / Crayon-Game-Framework

My own library for making games on the Dreamcast
BSD 3-Clause "New" or "Revised" License
5 stars 0 forks source link

Crayon: SH4 function optimisations #216

Open Protofall opened 4 years ago

Protofall commented 4 years ago

I've been notified of some quirks with the SH4 that effect performance. Its recommended that your function's arguments sum up to 32 bits/4 bytes. This can be done by passing structs by reference. Also inside the function they recommend using 10 vars of up to 32 bits / 4 bytes each (Not 320 bits total). If you fit these requirements you're only doing register accesses (5 nanoseconds), but any more and you need to read the stack which is a memory access at 1 millisecond. This is only really important for the rendering side since for 60FPS each frame is 16 milliseconds.

https://www.st.com/content/ccc/resource/technical/document/reference_manual/37/ad/a9/94/a1/8d/43/a1/CD17839242.pdf/files/CD17839242.pdf/jcr:content/translations/en.CD17839242.pdf

https://cdn.discordapp.com/attachments/488332893324836866/657738059214749710/SH4_ABI.PNG

Protofall commented 4 years ago

A bunch more notes https://dreamcast.wiki/GCC-SH4_tips#Use_struct_copies_.28instead_of_copying_individual_elements_of_a_struct.29

Protofall commented 4 years ago

Also not an SH4 optimisation, but using fs_mmap() should help with memory fragmentation apparently?