feat(nightly): execution becomes faster (#2013)

Uses the nightly `become` keyword to tell LLVM to `musttail`. This did
not work at first when the `Handler` type returned `Result` but after I
changed it to return nothing it does.

Ideas taken from
https://github.com/xacrimon/tcvm/blob/main/src/interp.rs

To remove code duplication, I created local declarative macros
`dispatch!` for any "enum dispatch" logic of function pointers we were
doing. The benefit is that this is now shared across four functions:
{e1,e2} x {pre_compute,handler}. I did not make a single general purpose
macro because Rust token parsing rules actually make this hard because
it doesn't let you reference ident's you don't declare, and the ident's
used actually vary based on the function.

Removed `pc_base` and just leave some empty space to avoid `pc -
pc_base` calculation. This is because `self.pc_base` is a runtime
variable, so we don't want to have to use a register (or worse
load/store) to access it.

To complete this:
- [x] For each crate, add "tco" feature, and then add
`#[create_tco_handler]` attribute above `execute_e1_impl` functions.
Then for each `Executor` implementation, copy the `pre_compute` function
implementation verbatim but switch to `handler` function signature and
return the tco handler fn pointer instead.
- [x] Do the same for metered execution.
- [x] ~~Switch to x86 global asm instead of relying on LLVM if we want
to be extra safe.~~ this seemed complicated and hard to do fully
properly so I prefer `become`.

Closes INT-4309

Author: jonathanpwang

.github/workflows/benchmark-call.yml

@@ -107,7 +107,7 @@ on:
 env:
   S3_METRICS_PATH: s3://openvm-public-data-sandbox-us-east-1/benchmark/github/metrics
   S3_FLAMEGRAPHS_PATH: s3://openvm-public-data-sandbox-us-east-1/benchmark/github/flamegraphs
-  FEATURE_FLAGS: "metrics,parallel,nightly-features"
+  FEATURE_FLAGS: "metrics,parallel,nightly-features,tco"
   INPUT_ARGS: ""
   CARGO_NET_GIT_FETCH_WITH_CLI: "true"
 

.github/workflows/benchmarks-execute.yml

@@ -2,8 +2,7 @@ name: "Execution benchmarks"
 
 on:
   push:
-    # TODO(ayush): remove after feat/new-execution is merged
-    branches: ["main", "feat/new-execution"]
+    branches: ["main"]
   pull_request:
     types: [opened, synchronize, reopened, labeled]
     branches: ["**"]
@@ -28,6 +27,7 @@ env:
   CARGO_TERM_COLOR: always
   S3_FIXTURES_PATH: s3://openvm-public-data-sandbox-us-east-1/benchmark/fixtures
   JEMALLOC_SYS_WITH_MALLOC_CONF: "retain:true,background_thread:true,metadata_thp:always,thp:always,dirty_decay_ms:10000,muzzy_decay_ms:10000,abort_conf:true"
+  TOOLCHAIN: "+nightly-2025-08-19"
 
 jobs:
   codspeed-walltime-benchmarks:
@@ -66,12 +66,12 @@ jobs:
 
       - name: Build benchmarks
         working-directory: benchmarks/execute
-        run: cargo codspeed build --profile maxperf
+        run: cargo $TOOLCHAIN codspeed build --profile maxperf --features tco
       - name: Run benchmarks
         uses: CodSpeedHQ/action@v3
         with:
           working-directory: benchmarks/execute
-          run: cargo codspeed run
+          run: cargo $TOOLCHAIN codspeed run
           token: ${{ secrets.CODSPEED_TOKEN }}
 
   codspeed-instrumentation-benchmarks:
@@ -111,10 +111,10 @@ jobs:
 
       - name: Build benchmarks
         working-directory: benchmarks/execute
-        run: cargo codspeed build
+        run: cargo $TOOLCHAIN codspeed build --features tco
       - name: Run benchmarks
         uses: CodSpeedHQ/action@v3
         with:
           working-directory: benchmarks/execute
-          run: cargo codspeed run
+          run: cargo $TOOLCHAIN codspeed run
           token: ${{ secrets.CODSPEED_TOKEN }}

benchmarks/execute/Cargo.toml

@@ -46,6 +46,7 @@ divan = { package = "codspeed-divan-compat", version = "3.0.2" }
 
 [features]
 default = ["jemalloc"]
+tco = ["openvm-sdk/tco"]
 mimalloc = ["openvm-circuit/mimalloc"]
 jemalloc = ["openvm-circuit/jemalloc"]
 jemalloc-prof = ["openvm-circuit/jemalloc-prof"]

benchmarks/prove/Cargo.toml

@@ -33,8 +33,9 @@ metrics.workspace = true
 [dev-dependencies]
 
 [features]
-default = ["parallel", "jemalloc", "metrics", "evm"]
+default = ["parallel", "jemalloc", "metrics"]
 metrics = ["openvm-sdk/metrics"]
+tco = ["openvm-sdk/tco"]
 perf-metrics = ["openvm-sdk/perf-metrics", "metrics"]
 stark-debug = ["openvm-sdk/stark-debug"]
 # runs leaf aggregation benchmarks:

ci/scripts/bench.py

@@ -15,9 +15,12 @@ def run_cargo_command(
     kzg_params_dir,
     profile="release"
 ):
+    toolchain = "+1.86"
+    if "tco" in feature_flags:
+        toolchain = "+nightly-2025-08-19"
     # Command to run (for best performance but slower builds, use --profile maxperf)
     command = [
-        "cargo", "run", "--no-default-features", "-p", "openvm-benchmarks-prove", "--bin", bin_name, "--profile", profile, "--features", ",".join(feature_flags), "--"
+        "cargo", toolchain, "run", "--no-default-features", "-p", "openvm-benchmarks-prove", "--bin", bin_name, "--profile", profile, "--features", ",".join(feature_flags), "--"
     ]
 
     if app_log_blowup is not None:

crates/cli/Cargo.toml

@@ -45,6 +45,7 @@ default = ["parallel", "jemalloc", "evm-verify", "metrics"]
 evm-prove = ["openvm-sdk/evm-prove"]
 evm-verify = ["evm-prove", "openvm-sdk/evm-verify"]
 metrics = ["openvm-sdk/metrics"]
+tco = ["openvm-sdk/tco"]
 # for guest profiling:
 perf-metrics = ["openvm-sdk/perf-metrics", "metrics"]
 # performance features:

crates/cli/src/lib.rs

@@ -1,3 +1,6 @@
+#![cfg_attr(feature = "tco", allow(incomplete_features))]
+#![cfg_attr(feature = "tco", feature(explicit_tail_calls))]
+
 pub mod commands;
 pub mod default;
 pub mod input;

crates/sdk/Cargo.toml

@@ -79,6 +79,17 @@ metrics = [
     "openvm-native-recursion/metrics",
     "openvm-native-compiler/metrics",
 ]
+tco = [
+    "openvm-circuit/tco",
+    "openvm-rv32im-circuit/tco",
+    "openvm-native-circuit/tco",
+    "openvm-sha256-circuit/tco",
+    "openvm-keccak256-circuit/tco",
+    "openvm-bigint-circuit/tco",
+    "openvm-algebra-circuit/tco",
+    "openvm-ecc-circuit/tco",
+    "openvm-pairing-circuit/tco"
+]
 # for guest profiling:
 perf-metrics = ["openvm-circuit/perf-metrics", "openvm-transpiler/function-span"]
 # turns on stark-backend debugger in all proofs

crates/sdk/src/lib.rs

@@ -1,3 +1,5 @@
+#![cfg_attr(feature = "tco", allow(incomplete_features))]
+#![cfg_attr(feature = "tco", feature(explicit_tail_calls))]
 use std::{
     borrow::Borrow,
     fs::read,

crates/vm/Cargo.toml

@@ -68,6 +68,9 @@ basic-memory = []
 # turns on stark-backend debugger in all proofs
 stark-debug = []
 test-utils = ["openvm-stark-sdk"]
+# Tail call optimizations. This requires nightly for the `become` keyword (https://github.com/rust-lang/rust/pull/144232).
+# However tail call elimination is still an incomplete feature in Rust, so the `tco` feature remains experimental until then.
+tco = ["openvm-circuit-derive/tco"]
 # performance features:
 mimalloc = ["openvm-stark-backend/mimalloc"]
 jemalloc = ["openvm-stark-backend/jemalloc"]