Branch Divergence & Predication: Comparison Matrix

Comparison matrix

Round-robin, greedy, and priority issue policies trade throughput, fairness, and tail behavior.

Use the matrix as a decision aid, not as a scoring shortcut. GPU design choices are strongly workload-dependent: the same policy can be correct for dense regular compute, wrong for sparse or divergent kernels, and dangerous for mixed graphics+compute scenes.

+------------------+----------------+----------------+----------------+
| Approach         | Strength       | Weakness       | Best when      |
+------------------+----------------+----------------+----------------+
| Conservative     | stable closure | lower peak     | new product    |
| Balanced         | good efficiency | needs profiling | broad mix      |
| Aggressive       | max throughput | sensitive tails | premium SKU    |
| Refactor         | scales cleaner | longer cycle   | chronic stalls |
+------------------+----------------+----------------+----------------+

When to choose each approach

• Choose architecture and tuning policy from measured bottleneck mix and product phase

Interview traps

• Copying tactics across unrelated workloads

• Ignoring inter-stage coupling in graphics+compute paths

Evidence comparison

GPU EVIDENCE MATRIX - Branch Divergence & Predication

+---------------------------+--------------------------------+--------------------------------+---------------------------+
| Evidence                  | Tells you                      | Does not prove                 | Next action               |
+---------------------------+--------------------------------+--------------------------------+---------------------------+
| occupancy + issue report  | warp residency and issue shape | memory transaction quality     | inspect coalescing        |
| cache + bandwidth counters| hierarchy pressure             | scheduler fairness causes      | profile warp arbitration  |
| lane-mask / branch trace  | divergence and reconvergence   | thermal or voltage stability   | pair with power telemetry |
| NoC/controller snapshots  | congestion and queue hotspots  | source-level mapping quality   | correlate with kernel map |
| release benchmark matrix  | end-to-end workload behavior   | root cause depth               | run focused reproducer    |
+---------------------------+--------------------------------+--------------------------------+---------------------------+

GPU deep dive

Warp scheduling quality determines whether latency hiding survives real control-flow and memory variance.

Concept diagram

WARP SCHEDULING LOOP

ready warp? -> issue -> dependency wait -> reconverge -> issue

Metric graph

STALL REASON SHARE

long scoreboard    ███████
divergence replay  █████
barrier wait       ███

Reports and artifacts

• eligible warp ratio

• stall reason histogram

• barrier wait cycles

• scheduler fairness report

Mini case study

A barrier-heavy kernel looked occupancy-safe, but warp arrival imbalance turned sync points into dominant stalls.

Debug branches

• Compare scheduler policy traces under bursty workloads

• Measure reconvergence delay and predication side effects

• Quantify barrier idle time before tuning launch size

Senior review question

Ask: which metric and benchmark pairing proves this topic is truly closed in production context?

Key takeaways

• Always pair micro-kernel metrics with end-to-end workload impact.

• Lock toolchain, driver, and launch metadata before comparing performance results.

Common pitfalls

• Optimizing occupancy without checking memory-system saturation.

• Comparing profiler captures from different driver or compiler builds.

• Declaring wins without reproducible accuracy and performance gates.

Principal GPU review addendum

Branch Divergence & Predication is not just a definition to memorize. In a real GPU program it becomes an interaction between software shape, compiler mapping, warp execution, memory movement, interconnect policy, and physical limits. The first senior move is to name which layer is being exercised before interpreting a counter.

Divergent control flow serializes paths under lane masks; predication can reduce branch overhead but may execute extra instructions. This mechanism matters because GPUs are throughput machines: a small inefficiency repeated across lanes, warps, SMs, frames, or dispatches can dominate product performance even when a unit-level diagram looks balanced.

Use divergence rate, reconvergence delay, and branch efficiency as an entry point, not as the conclusion. A metric shift only becomes actionable after it is tied to a workload slice, a profiler capture, an architectural path, and a reproducible artifact such as branch mask timeline, reconvergence stack trace, and predication tradeoff report.

Warp scheduling is a latency-hiding discipline, not a pure fairness problem. The review posture is therefore evidence-first: explain what the kernel or graphics workload asked for, how the GPU mapped it onto hardware, where useful work stopped, and which owner can change the smallest boundary safely.

In review, insist on a concrete chain from workload to hardware behavior: workload shape -> compiler/runtime mapping -> warp or pipeline behavior -> memory/fabric pressure -> measured product impact. That chain prevents generic GPU tuning advice from replacing engineering evidence.