Mark Functions with __attribute__((hot)) and ((cold))

Topic Overview

Split Hot and Cold Code

Placing cold error-handling functions in separate translation units or marking them __attribute__((cold)) keeps icache filled with the fast path. LLVM outlines cold blocks from hot functions during optimization.

Attributes and Layout

hot encourages layout near callers; cold moves unlikely blocks away. -freorder-blocks-and-partition is a GCC knob for similar effect.

cCopy
Loading…

• Pair with unlikely() on branches into cold helpers
• I-cache pressure shows up as frontend stalls in VTune
• Do not mark everything hot; the attribute loses meaning
• Measure icache misses if you suspect bloat

Working Through the Exercise

Keep the relevant documentation open while you implement. When your output disagrees with the reference, trace one failing case by hand before changing random lines.

• Trace first: walk one failing input step by step
• Read the manual: skim the official doc for the mechanism named in this task
• Compare baselines: save measurements from the prior step so speedups are honest

Why for this exercise

You will mark functions hot/cold and inspect section placement or size changes. This exercise requires showing icache-friendly layout for the common case.

Your Task

__attribute__((hot)) and __attribute__((cold)) do more than tune optimisation. GCC puts hot functions in .text.hot and cold ones in .text.unlikely, and the linker groups those sections. The code that actually runs therefore ends up packed together, touching fewer i-cache lines and fewer pages (iTLB entries). Model the linker's placement and measure the footprint of the executed code with and without the attributes.

Input

cCopy
Loading…

Rules

• Placement. Each function starts at the next multiple of 16. Source order places everything in .text in declaration order. With attributes, place .text.hot (hot functions), then .text (the rest), then .text.unlikely (cold functions), each in declaration order.
• Footprint. Count the distinct cache lines and pages overlapped by functions that ran at least once. A function occupies [addr, addr + size).
• Notes after the attribute layout, in declaration order: note: NAME is marked cold but ran N times and note: NAME is marked hot but never ran.
• Errors: func NAME: rejected (duplicate, size < 1, or more than 32), func NAME: unknown attribute X, and run: unknown function NAME.

Output

cCopy
Loading…

Each layout line lists every function by address.

Example

Input:

cCopy
Loading…

Output:

cCopy
Loading…

Requirements

1. Compute addresses from sizes and alignment alone, exactly as a linker concatenates sections.
2. Measure footprint as a set of distinct lines and pages. A line that two executed functions share counts once.

Hidden tests cover small pages, where the attributes save iTLB entries, a cold function that actually runs, a hot function that never runs, functions that straddle page boundaries, 128-byte lines, and invalid declarations.