UPSTREAM PR #18015: Add message for pre-RDNA AMD GPU support via opencl

[loci-dev]

Mirrored from ggml-org/llama.cpp#18015

Make sure to read the contributing guidelines before submitting a PR

[loci-review]

Explore the complete analysis inside the Version Insights

Performance Analysis Summary - PR #561

Analysis Overview

This PR introduces a documentation-only change to docs/backend/OPENCL.md, adding clarification about pre-RDNA AMD GPU support via the OpenCL backend. No source code, build configuration, or executable binaries were modified.

Performance Impact

No performance changes detected. The analysis confirms zero modifications to performance-critical code paths. All core binaries remain unchanged with identical power consumption profiles between versions.

Key Metrics:

• Response Time: No change across all functions
• Throughput Time: No change across all functions
• Power Consumption: No change across all binaries
• Tokens per Second: No impact (inference functions unmodified)

Binary Analysis:

• libllama.so: 195280 nJ (baseline) → 195280 nJ (target), delta: 0 ns
• libggml-cpu.so: 116389 nJ (baseline) → 116389 nJ (target), delta: 0 ns
• All other binaries show identical power consumption

This documentation update provides hardware compatibility information without affecting runtime behavior or performance characteristics.