[paulbalaji]

Problem

system.MinimumBalanceFor reproduces Rust's u64 -> f64 conversion, f64 multiply and saturating cast with integer code, so SBF needs no float helpers. That costs about 500 to 770 CU per call, mostly in the bit-length loops and the two rounding stages, and every rent-exempt account creation pays it. The pinned runners and current clusters all use a threshold of 2.0 (Fogo, Solana before SIMD-0194) or 1.0 (after it).

Solution

When the threshold bits are exactly 2.0 or 1.0 and n = (128+size)*rate is below 2^53, the conversion is exact and so is the product, so MinimumBalanceFor returns 2n or n. The size and overflow checks still run first. Every other input takes the unchanged emulation, now minimumBalanceExact. MinimumBalance gets it through MinimumBalanceFor. Hyperlane's Go library ships the same shortcut (hyperlane-xyz/hyperlane-monorepo#9852).

Benchmarks

TestRentSBF, whole transaction against main 939b215. The probe's own decoding and output cost ~260 CU of each total.

call	fogo-v0 main	this PR	solana-v3 main	this PR
MinimumBalance(165), runner sysvar (2.0 / 1.0)	889	399	897	396
MinimumBalanceFor(3480, 2.0, 165)	763	273	763	273
MinimumBalanceFor at 2.0, n = 2^53 - 128	1,029	273	1,029	273
MinimumBalanceFor(3480, 2.5, 165), exact path	759	767	759	767

482 to 759 CU less on every fast-path input with a nonzero product. The exact path costs 2 to 8 CU more (the threshold compares and a repeated bounds check). Errors caught before the threshold check are unchanged.

Tests

• TestRentFastPathMatchesExact (native): MinimumBalanceFor equals the emulation alone on both sides of n = 2^53 and of uint64 overflow for every size limit, both thresholds and their neighbouring bit patterns, negative zeros, other thresholds and NaN, plus a million random inputs near the 2^53 boundary, errors included. The existing IEEE oracle test passes unchanged.
• TestRentSBF on fogo-v0 and solana-v3: 532 cases (thresholds 2.0, 1.0, their neighbours, 2.5, 0.5, zeros, -1, infinity, NaN; sizes 0 to 10 MiB + 1 and 2^64 - 1; rates 0, 3,480, 6,960, the 2^53 and overflow limits and one past each) plus MinimumBalance from each runner's Rent sysvar, every result and error code equal to the host's.
• SDK-2 example snapshots of sdk/system/rent.go refreshed, so existing SDK-2 projects need re-scaffolding. Corpus C hashes unchanged.
• make test: same failures as main 939b215 on this machine (stdlib provenance/features tests with Go 1.23.2, TestFullSwapFacts under /tmp).
• verify_rebuild.py + check_rebuild_reference.py pass locally: the full-swap reference ELFs change (clock2 doesn't), so results/sdk/2026-10-19-rent-fast-path/rebuild/metrics.json is the new default reference.

Linked work

Closes #26 (the MinimumBalanceFor row). #27, #28, #29 and #31 cover the other rows; if this merges first, #26 should stay open until the rest land.

Used by hyperlane-xyz/hyperlane-monorepo#9852 through hyperlane-integration-4.