lat_mem_rd: add count override

[nicolas-bouton]

Add a -c option to lat_mem_rd so the user can limit dynamically the number of instructions executed per iteration. Additionaly, this patch keeps the existing default behavior for normal runs.

This changes allow to perform the same amount of instruction until the default count, computed with the size of the array currently tested and the stride, reach the maximum count given by the user.

• src/lat_mem_rd.c,
  src/lib_mem.h: Here.

[Copilot]

Pull request overview

Adds a new -c <max-count> option to lat_mem_rd to cap the per-iteration work performed in the pointer-chase loop while preserving the default behavior when the option is not provided.

Changes:

• Extend struct mem_state with a count field to carry the computed (and optionally capped) loop count into the benchmark callback.
• Add -c CLI parsing and thread the max-count value through loads() to cap the benchmark’s inner loop.
• Adjust the nanoseconds/load calculation to use the new state.count-based accounting.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
src/lib_mem.h	Adds count to mem_state so the benchmark loop can use a precomputed/capped iteration count.
src/lat_mem_rd.c	Implements -c option, computes capped count, uses it in the benchmark loop, and updates result scaling.

Suppressed comments (1)

src/lat_mem_rd.c:71

• The new -c flag caps state.count, but each count unit executes HUNDRED (100 pointer-chase loads). The help text currently suggests it is a raw instruction/load count; please clarify the unit to avoid misinterpretation.

	char   *usage = "[-P <parallelism>] [-W <warmup>] [-N <repetitions>] [-s <start>] [-e <end>] [-c <max-count>] [-t [-p <chase-page-size>]] len [stride...]\n";

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.