std.fmt.formatFloat: implement 32-bit and 64-bit ryu backends #19264

tiehuis · 2024-03-12T02:32:05Z

The ryu floating point formatting algorithm was recently added to std. This implemented only the 128-bit backend, which we use to print all floating point values: #19229

We should also consider implementing the 32-bit and 64-bit backends. At minimum, the 64-bit backend is valuable as it is much faster and reduces the large need for fast 128-bit integer support to make use of the existing backend. The 32-bit backend may be more valuable for smaller code-size specifically for embedded targets.

Upstream reference can be found at: https://github.com/ulfjack/ryu
An old (but outdated) implementation can be found at: https://github.com/tiehuis/zig-ryu

The 64-bit backend supports printing all floats up to 64-bits. The 128-bit continues to be used for larger values. This backend is approximately ~3x faster. Code size is a little smaller in the full table case and much smaller if using the samll tables. The implementation uses the same code-paths, parameterized by a set of tables and their pow5 implementations. We continue to use the same rounding/formatting mechanisms. Initially I explored a separate implementation, as upstream does this and has specific optimizations for these paths but for simplicity we don't. The performance loss is small enough at this point and keeping them combined keeps them in sync. Closes ziglang#19264.

The 64-bit backend supports printing all floats up to 64-bits. The 128-bit continues to be used for larger values. This backend is approximately ~3x faster. Code size is a little smaller in the full table case and much smaller if using the samll tables. The implementation uses the same code-paths, parameterized by a set of tables and their pow5 implementations. We continue to use the same rounding/formatting mechanisms. Initially I explored a separate implementation, as upstream does this and has specific optimizations for these paths but for simplicity we don't. The performance loss is small enough at this point and keeping them combined keeps them in sync. Closes #19264.

The 64-bit backend supports printing all floats up to 64-bits. The 128-bit continues to be used for larger values. This backend is approximately ~3x faster. Code size is a little smaller in the full table case and much smaller if using the samll tables. The implementation uses the same code-paths, parameterized by a set of tables and their pow5 implementations. We continue to use the same rounding/formatting mechanisms. Initially I explored a separate implementation, as upstream does this and has specific optimizations for these paths but for simplicity we don't. The performance loss is small enough at this point and keeping them combined keeps them in sync. Closes ziglang#19264.

tiehuis added contributor friendly This issue is limited in scope and/or knowledge of Zig internals. standard library This issue involves writing Zig code for the standard library. labels Mar 12, 2024

Vexu added this to the 0.13.0 milestone Mar 12, 2024

tiehuis mentioned this issue Mar 29, 2024

implement ryu 64-bit backend #19484

Merged

andrewrk closed this as completed in #19484 Mar 30, 2024

andrewrk modified the milestones: 0.13.0, 0.12.0 Mar 30, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

std.fmt.formatFloat: implement 32-bit and 64-bit ryu backends #19264

std.fmt.formatFloat: implement 32-bit and 64-bit ryu backends #19264

tiehuis commented Mar 12, 2024 •

edited

std.fmt.formatFloat: implement 32-bit and 64-bit ryu backends #19264

std.fmt.formatFloat: implement 32-bit and 64-bit ryu backends #19264

Comments

tiehuis commented Mar 12, 2024 • edited

tiehuis commented Mar 12, 2024 •

edited