<Project Name> — Migration Readiness Report

Analysis Date: <date>

Target Architecture: <arch, e.g., RISC-V rv64gc>

Reference Architecture: <x86_64 (Intel/AMD)>

Analyst Agent: amphimixis-orchestrator


1. Repository & Project Status

Metric

Value

Repository

<url>

Latest Commit

<date> (<N> days ago)

Total Commits

<N>

Latest Tag

<tag>

Activity

<actively maintained / sporadically maintained / archived>

Build Systems

<CMake / Makefile / Meson / …>

Test Count

<N>

External Dependencies

<count> — <list>

Distro Packages

<available in Debian / Arch / Yocto / …>


2. Platform-Specific Code Analysis

Architecture Macros Found

Macro

File

Line

Scope of Guarded Code

<_MSC_VER>

<file>

<line>

<just an #include / critical algorithm>

Platform Preprocessor Guards

#ifdef

Platform

What It Guards

<__APPLE__>

<macOS>

<file I/O variant>

Portability Verdict

  • No exceptions:

  • Alignment safe:

  • Embedded usability:

  • Overall: <N concerns found — list them>


3. Build & Test Results

Platform

Build

Tests Passed

Tests Failed

Notes

<reference x86> (-O3 -march=native -g)

✅

<N>

<M>

<flags>

<target> (\<flags>)

✅/❌

<N>

<M>

<cross-compilation / remote / emulator>

Build Failures / Test Failures Detail

<if any failures, describe root cause>


4. Performance Comparison

IMPORTANT: Don’t think, say the gist, if you don’t know, don’t write

Experimental Conditions

  • CPU: <x86 model> vs <target model>

  • Cores pinned: <list>

  • Warmup runs: <N>

  • Measurement runs: <N>

Key Metrics

Metric

x86

<target>

Ratio

Elapsed time (avg)

<μs>

<μs>

<N>

Instructions per cycle

<N>

<N>

<N>

L1-dcache miss rate

<%>

<%>

<N>

LLC miss rate

<%>

<%>

<N>

Branch misprediction rate

<%>

<%>

<N>

Frontend Bound

<%>

<%>

<N>

Backend Bound

<%>

<%>

<N>

Retiring

<%>

<%>

<N>

Hotspot Analysis (x86)

% Time

Function

Module

Analysis

Hotspot Analysis (<target>)

% Time

Function

Module

Analysis

Bottleneck Summary

<Identify the top 1-3 bottlenecks with causal explanations>

  • <#1: function X accounts for N% of samples because …>

  • <#2: high LLC miss rate suggests …>

Vectorization Intrinsics

  • Hand-written: <count> — <list of intrinsics by ISA>

  • Compiler auto-generated: <found in binary / none>

  • Portability concern: <yes/no — explanation>


5. Optimization Results

Vector Instructions in Binary

Architecture

Count

ISAs Found

<reference x86>

<N>

<SSE/AVX/AVX-512/NEON*>

<target>

<N>

<RVV/NEON/none>

* Note: NEON should not appear in an x86 binary — if it does, the analysis tool has a false positive.

Optimization Attempts

#

Optimization

Before (ms)

After (ms)

Δ

Causal Analysis

1

\<flag>

<N>

<N>

±X%

<explanation>

2

\<allocator>

<N>

<N>

±X%

<explanation>

3

LTO

<N>

<N>

±X%

<explanation>


6. Notes About Exploration process

<…>


7. Migration Readiness Summary

Aspect

Status

Notes

Builds on <reference x86>

✅/❌

Tests pass on <reference x86>

✅/❌

Builds on <target>

✅/❌

Tests pass on <target>

✅/❌

Zero external dependencies

✅/❌

No hand-written intrinsics

✅/❌

Alignment safe

✅/❌

Exceptions handled

✅/⚠️

Auto-vectorization

✅/⚠️

<compiler generates vector code / no vectorization on target>

Migration Verdict

<READY / MINOR CONCERNS / NOT READY> — <one-sentence summary>

Required Actions Before Migration

  1. <action item>

  2. <action item>