Minetest logo

IRC log for #minetest-dev, 2021-09-18

| Channels | #minetest-dev index | Today | | Google Search | Plaintext

All times shown according to UTC.

Enable nick filtering
Time Nick Message
00:22 MTDiscord <josiah_wi> Ah, good point! These nuances probably aren't as important as understanding the algorithm, so I might file them away investigate once I actually understand enough to have any business doing so.
22 more elements. Show/hide.
15:38 MTDiscord <exe_virus> The underlying variables of a float are stored in a double in the assembly code. I.e. a float uses the same amount of memory and address space on 64bit machines as a double. Compiler specific, but that's how both GCC and VC++ do it since at least VC13+ and GCC 6+ in my testing.   So unless we really want the lower precision, it's just that: lower precision.
15:41 MTDiscord <Warr1024> Haha, that's actually pretty funny.  Actually makes some sense too, i.e. in simplifying the FPU interface.
15:45 sfan5 I don't know what you tested but that's trivially proven wrong https://godbolt.org/z/96rGrsfjb
15:48 hendursaga joined #minetest-dev
15:50 sfan5 if you meant that they use the same registers (on x86 and x86-64) that's true
15:50 Extex joined #minetest-dev
16:06 pgimeno not true for SSE2+ though
16:14 sfan5 it's xmm[0-9] either way, just that if you're working with multiple values only two doubles fit but four floats
16:14 sfan5 right?
2 more elements. Show/hide.
16:55 pgimeno right
17:00 pgimeno https://godbolt.org/z/ncofrncqP
10 more elements. Show/hide.

| Channels | #minetest-dev index | Today | | Google Search | Plaintext