perf(new-execution): optimize tracing memory functions (#1941)

- make `PAGE_SIZE` a const generic in `PagedVec`
- use fixed size `[T; PAGE_SIZE]` array instead of slice
- update `PagedVec::get` to return a copy of data and not allocate pages
- reduce bounds checking - maybe we can just remove all bounds checking
in `PagedVec` if we can guarantee the pointers are bounds-checked
somewhere else? 🤔
- store a `u8` offset in `AccessMetadata` instead of a `u32` start
pointer
- pack `timestamp` and `block_size` values into a single `u32`
- make `prev_access_time` function a bit more readable
- use a stack allocated array (smallvec) instead of heap allocated
vector for intermediate variables

Author: shuklaayush

.github/workflows/lints.yml

@@ -46,7 +46,7 @@ jobs:
           # list of all unique features across workspace generated using:
           # cargo metadata --format-version=1 --no-deps | jq -r '.packages[].features | to_entries[] | .key' | sort -u | tr '\n' ' ' && echo ""
           # (exclude mimalloc since it conflicts with jemalloc)
-          cargo clippy --all-targets --all --tests --features "aggregation bls12_381 bn254 build-binaries default entrypoint evm-prove evm-verify export-intrinsics export-libm function-span getrandom-unsupported halo2-compiler halo2curves heap-embedded-alloc jemalloc jemalloc-prof metrics nightly-features panic-handler parallel perf-metrics rust-runtime static-verifier std test-utils" -- -D warnings
+          cargo clippy --all-targets --all --tests --features "aggregation bls12_381 bn254 build-elfs default entrypoint evm-prove evm-verify export-intrinsics export-libm function-span getrandom-unsupported halo2-compiler halo2curves heap-embedded-alloc jemalloc jemalloc-prof metrics nightly-features panic-handler parallel perf-metrics rust-runtime static-verifier std test-utils" -- -D warnings
           cargo clippy --all-targets --all --tests --no-default-features --features "mimalloc" -- -D warnings
 
       - name: Run fmt, clippy for guest

benchmarks/execute/benches/execute.rs

@@ -298,7 +298,7 @@ fn setup_leaf_verifier() -> (
     (vm, leaf_exe, input_stream)
 }
 
-#[divan::bench(sample_count = 3)]
+#[divan::bench(sample_count = 5)]
 fn benchmark_leaf_verifier_execute(bencher: Bencher) {
     bencher
         .with_inputs(|| {
@@ -321,7 +321,7 @@ fn benchmark_leaf_verifier_execute(bencher: Bencher) {
         });
 }
 
-#[divan::bench(sample_count = 3)]
+#[divan::bench(sample_count = 5)]
 fn benchmark_leaf_verifier_execute_metered(bencher: Bencher) {
     bencher
         .with_inputs(|| {
@@ -349,7 +349,7 @@ fn benchmark_leaf_verifier_execute_metered(bencher: Bencher) {
         });
 }
 
-#[divan::bench(sample_count = 3)]
+#[divan::bench(sample_count = 5)]
 fn benchmark_leaf_verifier_execute_preflight(bencher: Bencher) {
     bencher
         .with_inputs(|| {

benchmarks/utils/Cargo.toml

@@ -29,7 +29,7 @@ openvm-stark-sdk = { workspace = true, optional = true }
 
 [features]
 default = []
-build-binaries = []
+build-elfs = []
 generate-fixtures = [
     "dep:bitcode",
     "dep:openvm-circuit",

crates/vm/src/arch/execution_mode/metered/memory_ctx.rs

@@ -252,6 +252,7 @@ impl<const PAGE_BITS: usize> MemoryCtx<PAGE_BITS> {
         }
     }
 
+    // TODO(ayush): check if batching is actually even faster
     /// Resolve all lazy updates of each memory access for memory adapters/poseidon2/merkle chip.
     #[inline(always)]
     pub(crate) fn lazy_update_boundary_heights(&mut self, trace_heights: &mut [u32]) {