JIT: Exclude BB_UNITY_WEIGHT scaling from BasicBlock::isBBWeightCold

[amanasifkhalid]

Motivating example. BB_COLD_WEIGHT is currently 0.01, suggesting blocks that execute less than 1% of the time on average should be considered cold by block layout. However, BasicBlock::isBBWeightCold compares the block's normalized weight (with BB_UNITY_WEIGHT scaling) against BB_COLD_WEIGHT, so the actual cutoff for cold blocks is currently 0.01%. This change removes the BB_UNITY_WEIGHT scaling to get the intended behavior.

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull Request Overview

This PR adjusts BasicBlock::isBBWeightCold to compare the raw block weight against the product of call count and the cold-weight threshold, removing the previous unity-weight normalization.

• Swaps getBBWeight(comp) for bbWeight < (getCalledCount(comp) * BB_COLD_WEIGHT)
• Aims to restore the intended 1% cutoff for cold blocks rather than 0.01%

Comments suppressed due to low confidence (2)

src/coreclr/jit/block.h:1309

• There are no tests covering the updated isBBWeightCold behavior; consider adding unit tests for blocks exactly at, just below, and just above the computed cold threshold.

return bbWeight < (getCalledCount(comp) * BB_COLD_WEIGHT);

src/coreclr/jit/block.h:1309

• The new logic may misclassify blocks when getCalledCount(comp) returns zero, causing all blocks to appear hot. Consider guarding against a zero call count or falling back to a default behavior for unknown call counts.

return bbWeight < (getCalledCount(comp) * BB_COLD_WEIGHT);

[amanasifkhalid]

cc @dotnet/jit-contrib, @AndyAyersMS PTAL. Diffs have quite a bit of churn in both directions, though I'm taking solace in the relatively small PerfScore diffs, since this primarily affects the placement of cold code. I expect this to be a notable regression only in cases where we mistakenly give hot blocks near-zero weight, and we were likely already doing a subpar job with such cases (example). Thanks!

[amanasifkhalid]

ping @AndyAyersMS

[amanasifkhalid]

/ba-g unrelated wasm build failure