Branch Conditional (bc) Instruction

Overview

The Branch Conditional (bc) instruction is a conditional branch operation found in instruction-set architectures. Its execution path can be governed by flag values held in the Condition Register (CR), meaning that depending on operand encodings (and, in special cases, the CTR register contents), the instruction either always reads and is affected by the CR, or only sometimes is.

Operands and Influencing State

The bc instruction carries two key operand fields that determine its CR-affected behavior:

• bo.data — the branch-options operand field.
• bi.data — the bit-index operand field that selects which CR bit is tested.

In addition, the value of the CTR (Count Register) can be a determining factor in certain special encodings of bo.data.

Classification-Derived Rules for CR Dependence

A sample-driven approach using a decision-tree classifier on execution traces produced the following architectural rules describing when a bc instruction is affected by the CR (claim "CR.Affected = True"):

These rules illustrate that a subset of bo.data values (e.g., 0x0–0x1, 0x8–0x9, 0x4–0x7, 0xC–0xF) cause the branch decision to always consult the CR, while remaining encodings produce mixed behavior dependent on other microarchitectural state.

Method of Discovery

The CR-affected rules above were discovered automatically rather than hand-authored. The procedure:

1. Generate a bc instruction sample, varying its operands and surrounding register state.
2. Collect input-feature information (opcode, operand fields such as bo.data / bi.data, register values such as CTR, MSR flags, etc.).
3. Execute the instruction on a reference model / simulator.
4. Determine whether the instruction's operation was affected by CR by toggling the relevant CR flag, re-executing, and comparing simulated machine states — flag-driven divergence indicates the instruction reads and is affected by the flag.
5. Feed collected (features, CR-affected?) samples to a decision-tree classifier that deduces the input-feature combinations predicting CR-affected behavior.
6. Translate the classifier's rules into a generator-usable form so that subsequent instruction-generation requests satisfying those rules will reproduce the CR-affected event with the requested probability.

This automated architectural-knowledge extraction enables test generators to construct interesting sequences such as an instruction that writes to CR followed by a bc branch that is provably affected by CR.

Representative Sample Rows

The following rows illustrate the feature set and the corresponding CR-affected outcome observed in collected samples:

The sample table shows that, for example, bo.data == 0xC correlates with CR.Affected == True, consistent with the always-affected rule for bo.data ∈ {0x4,0x5,0x6,0x7,0xC,0xD,0xE,0xF}.

Significance

Automatically deriving when a bc branch depends on the CR is valuable for generation-based verification because it removes the need for a verification engineer to hand-encode the architecture's branch-conditional semantics, and it lets the generator deliberately produce CR-influencing branch sequences to stress test pipelines and branch-prediction logic.