Table data layout question

[BarabasGitHub]

I was looking through the code and I actually wonder about these tables.

zig-ryu/src/ryu64_full_table.zig

Lines 18 to 20 in 724f66d

	// These tables are generated by PrintDoubleLookupTable.
	pub const DOUBLE_POW5_INV_SPLIT = [][]const u64{
	[]const u64{ 1, 288230376151711744 }, []const u64{ 3689348814741910324, 230584300921369395 },

Does the [][]u64 notation mean that it goes through two pointers and that each row can be a different length? If so, wouldn't it then be better to have a [][2]u64 table? Or does it already compile down to that?

[tiehuis]

Yeah I did think that when I did those tables. A quick test shows the code doesn't compile to the same code but I'll have a closer look and come back here before changing it.

[BarabasGitHub]

Yeah I was thinking that if it has to do an extra pointer hop it might explain part of the performance difference between the zig and C implementation. Op vr 10 aug. 2018 14:12 schreef Marc Tiehuis <notifications@github.com>:

…

Yeah I did think that when I did those tables. A quick test shows the code doesn't compile to the same code but I'll have a closer look and come back here before changing it. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AL-sGsTSwIcH5NHj3Aj3dYMPPnMtbsG2ks5uPXjCgaJpZM4Vw4CU> .