[Reading Notes] Wedege-Parallel Triangle Counting for GPUs
A novel GPU algorithm--WETRIC, counting triangles in static undirected sparse graphs.
Akejyo — personal journal
Boundless Gift
“The vision of the Cyclops is now shared with you. Deep in the icefields, the full picture of the prophecy is yet to be revealed.”
Horizon Invitation
An invitation to the end of the icefield, distributed or sold through secret channels to those deserving. However, obtaining this invitation is merely the first step, and the test is yet to come—
Dimensional Fluid
“The circular gate lies within the cold winds, waiting to be opened once more.”
Old Fan
When it rains for several days on end, the old Yanese Tianshi complains to his friends about the pain in his knees, before casually waving his fan and driving away the rainclouds.
Golden Chalice
Saints, knights, and pietists. All of them are madmen. Dear God, even the chalice itself is made of pure gold!
Hot Water Kettle
Some people often use the water boiler in the Rhodes Island Office to cook instant noodles in the middle of the night. That's not a very healthy lifestyle...
01 / The latest entries
View all records Recent records
READ
[Reading Notes] Triangle Counting over Large-scale Directed Graphs
Exact and scalable triangle counting on large-scale directed graphs, where triangles have seven different types determined by edge directions and the graph may be too large for a s
[Reading Notes] BeaconKV: Key-Value Cache Compression Guided by Beacon Queries for Efficient Large Reasoning Model Inference
CoT (Chain-of-Thought) brings memory bottleneck (KV Cache), limits the practical deployment of LRMs (Large Reasoning Models) under constrained GPU resources.
02 / Work in progress
All expeditions Current expedition
FIELD STUDY / AN IMAGINED SHOREEXPEDITION 01 ● ACTIVE
Triangle counting on multiple GPUs
A research project on triangle counting on multiple GPUs, with a focus on optimizing performance and scalability.
- FIELD
- Research / HPC
- LAST RECORD
- 10 September 2026
03 / A few places to begin
Selected writing
What is Sepeculative Decoding
We have a slow, big target model M_p, inference from which we're trying to accelerate (distribution p(x)). We also have a quick, efficient approximation model M_q (distribution q(x
Read the record →Parallel Computing with MPI — The Sieve of Eratosthenes and Performance Optimization
Implement optimize4 to improve performance as much as possible without enabling O2 or changing the algorithm.
Read the record →TX·IEG菁英班3期作业wp
使用Windows API CreateRemoteThread 远程注入模块到 crackme 进程中并且不崩溃，然后使用 ark 工具(例如 processhacker 等)能够查看注入模块的信息；
Read the record →The person behind the pages
I’m Akejyo. This is a place to think in public, keep useful questions, and make things slowly.
A little about me ↗Ride with things, let the mind roam.