execution: reduce interpreter, syscall and runtime overhead - #281

Open

7layermagik wants to merge 22 commits into

7layer/review-performance-corefrom

7layer/execution-performance

Commits on Sep 24, 2026

fix: preserve boolean flag defaults and explicit config overrides
7layermagik
committed
915f619
fix: own retained vote deque storage before pooled reuse
7layermagik
committed
cb360c4
sbpf: reject overflow in contiguous virtual memory ranges
7layermagik
committed
be7a914
sealevel: remove SHA-256 syscall descriptor and digest allocations
7layermagik
committed
584791f
test: measure captured SHA loop and document Zen 5 acceleration
7layermagik
committed
f944a0e
sbpf: cut per-instruction overhead in the interpreter (~2x on SPL Token transfer)

claude
authored and
7layermagik
committed
75db2e0
sbpf: add interpreter benchmarks and a differential test corpus

claude
authored and
7layermagik
committed
5b8f33f
sealevel: add native program micro-benchmarks (System transfer, Vote TowerSync)

claude
authored and
7layermagik
committed
3f0ec6c
test: update VASA stack registers for fixed-size interpreter state
7layermagik
committed
e8dcadc
test: benchmark verified token arithmetic and CPI workloads with warm programs
7layermagik
committed
203e105
docs: record individual-program and matched Alpenglow replay measurements
7layermagik
committed
1f4a69b
sealevel: copy, compare and fill VM memory without temporaries or byte loops

claude
authored and
7layermagik
committed
2361bc9
lthash: vectorize MixIn/MixOut with AVX2 on amd64

claude
authored and
7layermagik
committed
32e149e
test: preserve zero-length memory validation and repair VM fixtures
7layermagik
committed
bfddc29
test: cover unaligned and aliased AVX2 hash lanes
7layermagik
committed
4ad9805
sbpf: account for resolved call targets in program cache cost
7layermagik
committed
1ca0fb3
test: retain syscall differential coverage and execution reproduction guide
7layermagik
committed
a6f3586
test: trim execution benchmark experiments and documentation
7layermagik
committed
14c5404
sbpf: restore v2 memory opcodes and differential coverage
7layermagik
committed
1992ab4
sbpf: guard pooled write tracking beyond bitmap capacity
7layermagik
committed
a520264
sealevel: match Agave zero-length memory copy behavior
7layermagik
committed
3830a18
sealevel: track sibling header writes in pooled VM memory
7layermagik
committed
a317eb6

Navigation Menu

execution: reduce interpreter, syscall and runtime overhead - #281

execution: reduce interpreter, syscall and runtime overhead#281
7layermagik wants to merge 22 commits into
7layer/review-performance-corefrom
7layer/execution-performance

Commits on Sep 24, 2026

fix: preserve boolean flag defaults and explicit config overrides

fix: own retained vote deque storage before pooled reuse

sbpf: reject overflow in contiguous virtual memory ranges

sealevel: remove SHA-256 syscall descriptor and digest allocations

test: measure captured SHA loop and document Zen 5 acceleration

sbpf: cut per-instruction overhead in the interpreter (~2x on SPL Token transfer)

sbpf: add interpreter benchmarks and a differential test corpus

sealevel: add native program micro-benchmarks (System transfer, Vote TowerSync)

test: update VASA stack registers for fixed-size interpreter state

test: benchmark verified token arithmetic and CPI workloads with warm programs

docs: record individual-program and matched Alpenglow replay measurements

sealevel: copy, compare and fill VM memory without temporaries or byte loops

lthash: vectorize MixIn/MixOut with AVX2 on amd64

test: preserve zero-length memory validation and repair VM fixtures

test: cover unaligned and aliased AVX2 hash lanes

sbpf: account for resolved call targets in program cache cost

test: retain syscall differential coverage and execution reproduction guide

test: trim execution benchmark experiments and documentation

sbpf: restore v2 memory opcodes and differential coverage

sbpf: guard pooled write tracking beyond bitmap capacity

sealevel: match Agave zero-length memory copy behavior

sealevel: track sibling header writes in pooled VM memory