extend Assembler API; add primes workload showcase#835
Open
prodigalwon wants to merge 3 commits into
Open
Conversation
Adds 17 new typed methods to `Assembler` covering the gap that `grey-bench` has been working around via local `emit_branch_lt_u` + raw byte emission. With these helpers, hand-assembled programs can target the full ALU + control flow surface of JAM PVM without escaping to `emit_raw`. Three-register ALU: - mul_64 (202), div_unsigned_64 (203), div_signed_64 (204), rem_unsigned_64 (205), rem_signed_64 (206) - and (210), xor (211), or (212) - set_less_than_unsigned (216), set_less_than_signed (217) Two-register-plus-immediate: - mul_imm_64 (150) - and_imm (132), xor_imm (133), or_imm (134) - set_less_than_unsigned_imm (136), set_less_than_signed_imm (137) Register-register branches (a new section in the file — these have their own encoding shape, opcode + (rA|rB<<4) + 4-byte signed relative offset, distinct from the existing reg-imm branches): - branch_eq (170), branch_not_eq (171) - branch_less_unsigned (172), branch_less_signed (173) - branch_greater_or_equal_unsigned (174), branch_greater_or_equal_signed (175) `branch_less_unsigned` in particular makes `grey-bench`'s `emit_branch_lt_u` helper obsolete — same opcode, same encoding, typed instead of raw. All encodings follow the existing methods' patterns verbatim: opcode emitted with `is_instruction_start = true`, operand bytes with `false`, immediate via `emit_imm(value as i64, 4)`. 17 new unit tests verify the byte encoding of each method against the documented opcode + the encoding shape. Existing tests unchanged. `cargo test -p grey-transpiler --lib` runs 65 tests, all pass. `cargo check -p grey-bench` confirms no downstream breakage. Motivation: building Rostro's RVM bench harness against javm required hand-assembling non-trivial workloads (primes, sieve, sort) — without these helpers we hit the same emit_raw + hand-offset-tracking pattern grey-bench already works around. Formalizing the helpers benefits any consumer hand-rolling JAM PVM bytecode. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds `grey_primes_blob(n)` and `polkavm_primes_blob(n)` — both
hand-assembled, both count primes in [2, N) by naive trial division,
both return the count in A0.
This is the showcase consumer for the extended Assembler API landed
in the previous commit. Demonstrates four of the new methods in a
real workload:
- `rem_unsigned_64` — modulo for divisibility test
- `set_less_than_unsigned` — both nonzero-flag and out-of-range mask
- `mul_64` — flag aggregation (replaces early-exit branches)
- `branch_less_unsigned` — inner + outer loop back-edges
The polkavm-flavor mirrors the same algorithm via `ProgramBlobBuilder`
(matches the existing `grey_fib_blob` / `polkavm_fib_blob` pairing
convention).
Algorithm:
count = (2 < N) ? 1 : 0 // pre-count i=2 if in range
for i in 3..(N+1):
is_prime = 1
for j in 2..i: // do-while; i >= 3 so always runs >= 1
rem = i %u j
is_prime *= (0 <u rem)
in_range = (i < N) ? 1 : 0 // masks the final overshoot iteration
count += is_prime * in_range
return count
The two `mul *= flag` patterns deliberately replace forward
early-exit branches with backward-only branches. Forward branches
in hand-assembled programs need their target at a basic-block start,
which means a preceding terminator instruction — solvable but
fragile. The mul-by-flag trick sidesteps it entirely while keeping
the same asymptotic O(N^2) cost.
Two unit tests in a new `tests_primes` module:
- `test_grey_primes_matches_native_reference` — sweeps
N in {2, 5, 10, 30}, asserts blob output matches a native-Rust
trial-division reference.
- `test_grey_primes_30_is_10` — sanity-check known fact (10 primes
in [2, 30)).
`cargo test -p grey-bench --lib` — 21 pass (19 pre-existing + 2 new).
Complements the pre-built `grey_sieve_blob` / `polkavm_sieve_blob`
already in this file: different algorithm (O(N^2) vs O(N log log N))
and different construction path (hand-assembled vs Rust->RISC-V via
build-script). Useful as a reference workload for the extended
Assembler API and as an alternate compute-shape for benchmarking.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Contributor
Genesis ReviewComparison targets:
How to reviewPost a comment with the following format (rank from best to worst): Use the short commit hashes above and To meta-review another reviewer's comment, react with 👍 or 👎. |
CI's `cargo fmt --all --check` flagged the trailing `builder.to_vec().expect(...)` chain as a single line; rustfmt prefers it broken across three lines. No semantic change. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Contributor
|
/review Adds a useful typed Assembler surface plus a hand-assembled prime workload that exercises the new helpers, with byte-level encoding tests and a native reference check. Ranked below the GRANDPA out-of-order vote fix on difficulty/design because that target handles live consensus behavior, but above the smaller proof/refactor/doc targets. Because this touches grey-bench and grey-transpiler, a benchmark comparison is worth running before final merge. |
Contributor
|
JAR Bot: Review recorded from @johandroid (1 reviews, 0 meta-reviews). |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Two commits, both motivated by hand-rolling non-trivial workloads against
grey-transpiler::Assembler:grey-transpiler: extend the
AssemblerAPI with 22 typed helpersfor opcodes the existing surface didn't cover — multiplication, division,
remainder, bitwise ALU, set-less-than, and register-register branches.
The encoding patterns mirror what
grey-bench's localemit_branch_lt_uhelper has been doing via
emit_raw— formalizing it as typed methods.17 byte-encoding unit tests; 65 → 82 total tests pass.
grey-bench: add
grey_primes_blob(n)+polkavm_primes_blob(n)—hand-assembled naive trial-division prime counter. Showcase consumer
for the extended API (uses
rem_unsigned_64,set_less_than_unsigned,mul_64,branch_less_unsigned). Complements the pre-built sieve blobs;different algorithm (O(N²) vs O(N log log N)) and different construction
path (hand-assembled vs build-script). 2 new tests verifying correctness
against a native-Rust reference; 21 / 21 grey-bench tests pass.
Verification
cargo test -p grey-transpiler --lib— 82 pass (existing 65 + 17 new)cargo test -p grey-bench --lib— 21 pass (existing 19 + 2 new)cargo check -p grey-bench— clean (no downstream regressions)