2026
- 09-20 A Function Name Cannot Hold Its Numerical Contract
- 09-16 Same Output Shape, Different Answers: Pooling, Resizing, and Index Reductions
- 09-09 It Only Moves Data. How Can It Still Be Wrong?
- 09-04 An Instruction’s Four Identities: Why Code Generation Needs More Than Strings
- 08-29 You Found the Operator’s Name. Does That Make It Supported?
- 08-25 The Most Useful Words in an Operator Specification May Be “Not Yet Confirmed”
- 08-18 The Assembly Looks Right. Does That Mean the Machine Code Is Right?
- 08-13 Why Does the Same Release Package Get a Different Hash Every Time?
- 08-07 Compilation Failed. Why Is There Still a “Successful Model” in the Directory?
- 08-03 Why Is Packing Compiler Outputs into One File Hard?
- 07-27 The Compiler Moved. Its Lookup-Table Module Didn’t Get the Address
- 07-22 The Library Is in the Package. Why Can’t the Program Find It?
- 07-16 One Version of the Headers, Another Version of the Generator
- 07-12 “I Only Rebuilt It.” Why Did the Installation Directory Disappear?
- 07-05 Debug Works, but Release “Optimizes Away” a Write
- 06-30 Fewer Operations—So Why Disable the Optimization?
- 06-25 “I Parsed a Number” Does Not Mean the Input Is Valid
- 06-20 One More Loop, and the Binary Snapshot Looks Like a Different Program
- 06-14 You Caught the Exception. What Happens to the Half-Written Machine Code?
- 06-08 I Waited for the Event. Why Isn't the Data Ready?
- 06-03 Why Write the Same Final State Twice?
- 05-29 The Address Fits. Why Does the Compiler Say Memory Is Full?
- 05-23 The View Moved No Data. Why Can’t We Return the Memory?
- 05-17 A Few More Branches in the Graph—and Synchronization Registers Run Out First?
- 05-12 Why Does Someone Keep Borrowing My “Temporary” Register?
- 05-07 When Weights Move to the Left: A Transpose Identity That Moves the Whole Compiler Chain
- 05-01 A Sigmoid After MatMul: Why the Compiler Still Refuses to Fuse It
- 04-25 Putting Activation Inside Convolution: Preserve the Quantization Boundary Before Removing the Node
- 04-20 Who Moved My Scratch Register? Teaching an Emitter to Borrow and Return
- 04-15 Abs and Sign: One Line of Arithmetic Does Not Make Their Parameter Contracts Interchangeable
- 04-09 Subtracting Zero Can Disappear; Subtracting from Zero Cannot
- 04-03 The First Row Works, the Second Disappears: Alignment, Strides, and Padded Constants
- 03-29 Broadcasting: Sometimes the Right Address Is the One That Does Not Move
- 03-25 When the Debugger Starts Lying: One Permute, Three Ways to Fix Dump
- 03-18 Why Does Reshape Need Code? The Elements Stay the Same; Their Route Changes
- 03-13 Transposed Convolution Weights: The Shape Is Honest, but Memory Speaks Another Dialect
- 03-07 Grouped Convolution: Once You Slice the Cake, Slice the Labels Too
- 03-03 The Importer Got Clever Too Soon: Algebraic Optimization before Quantization Metadata
- 02-24 Floating Point Passes, Quantization Fails: A Rewrite Gives the Output the Input’s Identity
- 02-19 Epsilon Was Nonzero—Until It Nearly Vanished in the Integer Domain
- 02-13 Build a Dedicated Operator, Then Remove It: A Design Change Worth Recording
- 02-09 Just a Few Channel Parameters—So Why Is Local Memory Full?
- 02-02 Once BatchNorm Becomes Multiply-Add, the Numerical Work Begins
- 01-28 The Multiplication Is Right; the Scale Is Wrong: An Integer Kernel’s Two Ledgers
- 01-22 This Reduction Does Nothing—Why Can’t We Delete It?
- 01-18 Max, Min, and Sum: Similar Operators, Different Mathematical Homework
- 01-11 The Vanishing Dimension: How keep_dims=false Can Break an Inverse Permutation
- 01-06 Just Take the Mean—Why Is the Compiler Writing Loops?
- 01-02 Compiler Engineering Notes: 48 Bilingual Articles and Reading Paths
2025
- 04-20 ROS2 Fleet Coordinator V1: From Task Submission to Route Dispatch
- 03-28 Why Planning Uses Services While State and Assignments Use Topics
- 03-25 Why the Project Has Not Chosen Nav2-First Integration
- 03-24 DDS Sandbox Restrictions and the Limits of ROS 2 Runtime Verification
- 03-21 Why Conflict Detection and Reservations Are the Next Major Complexity Increase
- 03-17 robot_agent: Execution States and Waypoint Progression
- 03-15 path_planner: Graph Model, BFS, and Route Stitching
- 03-10 fleet_manager: Task Queue, Planner Dependency, and Assignment State Retention
- 01-08 Unit Impulse and Unit Step in Continuous and Discrete Time
2024
- 10-10 Beyond Technical Depth: What Else Makes a Capable Engineer?
- 10-01 Task Contexts, Stack Initialization, and Context Switching
- 09-30 Memory Management and Task Switching in a Multiprogram Environment
- 09-20 Building Multiple Applications and Managing the Compilation Pipeline
- 09-15 Implementing RISC-V Trap Handling and System Calls
- 09-15 Compiling, Loading, and Executing a User Program
- 09-15 From Compilation to Execution: Loading Navy Applications in Nanos-lite
- 09-01 Device Interfaces, Tracing, and Memory-Mapped I/O in NEMU
- 08-30 Startup Assembly and Experiments With printf
- 08-25 ELF Symbol Handling and Abstract Machine Infrastructure
- 08-20 Implementing and Using Memory and Function Tracing
- 08-18 Tracing Infrastructure and an Instruction Ring Buffer
- 08-08 Timing, Logging, Argument Passing, and Automated Makefile Builds
- 07-13 A Possible Implementation of sprintf and stdarg
- 07-13 Project Makefiles and NEMU Batch Mode
- 07-13 Reading the Source: Traps, Abstract Machine, and the Runtime Environment
- 05-31 Engineering Debugging Notes
- 05-10 Variadic Macros, Token Pasting, and Callback-Style Macro Expansion
- 05-09 Commenting Multiline Macros: A Mistake That Cost an Afternoon and an Evening
- 05-08 How the YEMU Emulator Works
- 05-08 Notes on Operand Decoding and Implementing RISC-V Instructions
- 05-08 C Preprocessor Token-Pasting Operator
- 04-08 Tracing the Nested Calls Behind cmd_c
- 04-08 Instruction Pattern Matching: The Macros and Functions Behind INSTPAT
- 03-23 Using a Macro to Check Whether Another Macro Is Defined
- 02-26 Cache Capacity, Block Size, Miss Penalties, and Set Associativity
- 02-14 Precise Traps, Multiple Issue, Register Renaming, and Memory Aliasing
- 02-03 Stage-Specific Forwarding, Load-Use Stalls, Early Branches, and Pipeline Flushes
- 01-26 Pipeline Hazards, Forwarding, Register Timing, and the x0 Trap
- 01-14 Rounding, Intermediate Results, Parallel Floating-Point Arithmetic, and Processor Design
- 01-06 Multiplication Hardware, Subnormal Floating-Point Values, and the RISC-V F/D Extensions
- 01-03 Bounds Checks, Calling Conventions, Large Immediates, and PC-Relative Addressing
- 01-03 Long Branches, Synchronization, Compiler Optimizations, and x86 Complexity
- 01-01 Fundamental RISC-V Design Principles
2023
- 12-12 Where I Left Off in My x86 Assembly Notes
- 10-11 x86 Programming Notes, Part 3: Execution, Descriptors, and Mode Switching
- 09-23 x86 Programming Notes, Part 2: I/O, Calls, Interrupts, and Protected Mode
- 09-14 Assembly Environments and Conventions Used in This Blog
- 09-11 x86 Programming Notes, Part 1: Techniques and Instruction Details
- 08-10 Preparing for the Long Haul
- 08-01 8086 Interrupts: Sources, the Vector Table, and Handler Execution
- 07-30 A Simple 8086 Assembly Lab: Displaying a Number on Screen
- 07-25 AI: What We Should Understand and How We Might Develop
- 07-15 A Beginner 8086 Assembly Lab: Copying Program Code
- 07-14 DOS Basics and the Beginning of Assembly Programming
VISITORS
FEATURED TAGS
ABOUT ME
RECENT POSTS
- A Function Name Cannot Hold Its Numerical Contract
- Same Output Shape, Different Answers: Pooling, Resizing, and Index Reductions
- It Only Moves Data. How Can It Still Be Wrong?
- An Instruction’s Four Identities: Why Code Generation Needs More Than Strings
- You Found the Operator’s Name. Does That Make It Supported?