| author |  Andrew Kelley <andrew@ziglang.org> 2022-06-27 19:05:51-07:00 |
| committer | Andrew Kelley <andrew@ziglang.org> 2022-06-27 19:11:55-07:00 |
| log | a71d00a4d504edfdb09cd169d29ca1bbc0b909c4 |
| tree | 1b4f5a419ee46ccc26d605f00109f026d9a40c26 |
| parent | 0b8bd9b2b4f609fb4ae7d31da7e7a0fd4f5ad987 |
std.crypto.25519.field: avoid excessive inlining
This valid zig code produces reasonable LLVM IR, however, on the
wasm32-wasi target, when using the wasmtime runtime, the number of
locals of the `isSquare` function exceeds 50000, causing wasmtime
to refuse to execute the binary.
The `inline` keyword in Zig is intended to be used only where it is
semantically necessary; not as an optimization hint. Otherwise, this may
produce unwanted binary bloat for the -OReleaseSmall use case.
In the future, it is possible that we may end up with both `inline`
keyword, which operates as it does in status quo, and additionally
`callconv(.inline_hint)` which has no semantic impact, but may be
observed by optimization passes.
In this commit, I also cleaned up `isSquare` by eliminating an
unnecessary mutable variable, replacing it with several local constants.
Closes #11947.1 files changed, 7 insertions(+), 8 deletions(-)
lib- std
- crypto
- 25519
lib/std/crypto/25519/field.zig+7-8
| ... | ... | @@ -341,7 +341,7 @@ pub const Fe = struct { |
| 341 | 341 | } |
| 342 | 342 | |
| 343 | 343 | /// Square a field element `n` times |
| 344 | inline fn sqn(a: Fe, comptime n: comptime_int) Fe { | |
| 344 | fn sqn(a: Fe, n: usize) Fe { | |
| 345 | 345 | var i: usize = 0; |
| 346 | 346 | var fe = a; |
| 347 | 347 | while (i < n) : (i += 1) { |
| ... | ... | @@ -390,13 +390,12 @@ pub const Fe = struct { |
| 390 | 390 | const _11 = a.mul(a.sq()); |
| 391 | 391 | const _1111 = _11.mul(_11.sq().sq()); |
| 392 | 392 | const _11111111 = _1111.mul(_1111.sq().sq().sq().sq()); |
| 393 | var t = _11111111.sqn(2).mul(_11); | |
| 394 | const u = t; | |
| 395 | t = t.sqn(10).mul(u).sqn(10).mul(u); | |
| 396 | t = t.sqn(30).mul(t); | |
| 397 | t = t.sqn(60).mul(t); | |
| 398 | t = t.sqn(120).mul(t).sqn(10).mul(u).sqn(3).mul(_11).sq(); | |
| 399 | return @bitCast(bool, @truncate(u1, ~(t.toBytes()[1] & 1))); | |
| 393 | const u = _11111111.sqn(2).mul(_11); | |
| 394 | const t = u.sqn(10).mul(u).sqn(10).mul(u); | |
| 395 | const t2 = t.sqn(30).mul(t); | |
| 396 | const t3 = t2.sqn(60).mul(t2); | |
| 397 | const t4 = t3.sqn(120).mul(t3).sqn(10).mul(u).sqn(3).mul(_11).sq(); | |
| 398 | return @bitCast(bool, @truncate(u1, ~(t4.toBytes()[1] & 1))); | |
| 400 | 399 | } |
| 401 | 400 | |
| 402 | 401 | fn uncheckedSqrt(x2: Fe) Fe { |