llvm-native-core 0.1.6

//! X86 Instruction Bundling — instruction bundle formation for VLIW-style
//! execution on in-order cores, bundle boundary enforcement, bundle locking
//! for precise exception handling, and microarchitecture-specific bundle
//! rules for Intel Atom, Silvermont, Goldmont, and other in-order cores.
//!
//! This module implements instruction bundling: the process of grouping
//! adjacent instructions into fixed-size "bundles" to enable efficient
//! decoding and execution on in-order microarchitectures. While out-of-order
//! cores (Skylake, Zen, etc.) dynamically schedule instructions, in-order
//! cores (Atom, Silvermont, Goldmont, and some embedded x86 cores) benefit
//! from explicit bundle formation that ensures the front-end can decode
//! and issue instructions without pipeline stalls.
//!
//! ## Module Organization
//!
//! ```text
//! ┌─────────────────────────────────────────────────────────────────────────┐
//! │                       X86 Instruction Bundling                           │
//! │  ┌──────────────────┐  ┌──────────────────┐  ┌───────────────────────┐  │
//! │  │  Bundle Formation│  │  Bundle Rules    │  │  Bundle Boundary     │  │
//! │  │  Group instrs    │  │  Atom/Silvermont │  │  Enforcement: no     │  │
//! │  │  into bundles    │  │  Goldmont/Tremont│  │  instrs crossing     │  │
//! │  │                  │  │  per-uarch rules │  │  bundle boundaries   │  │
//! │  └──────────────────┘  └──────────────────┘  └───────────────────────┘  │
//! │  ┌──────────────────┐  ┌──────────────────┐  ┌───────────────────────┐  │
//! │  │  Bundle Locking  │  │  Exception       │  │  Branch @ Bundle     │  │
//! │  │  Precise state   │  │  Handling        │  │  End Requirement     │  │
//! │  │  recovery at     │  │  Bundle-atomic   │  │  Branches must end   │  │
//! │  │  bundle boundary │  │  exception       │  │  at bundle boundary  │  │
//! │  └──────────────────┘  └──────────────────┘  └───────────────────────┘  │
//! └─────────────────────────────────────────────────────────────────────────┘
//! ```
//!
//! ## Background: Why Bundling on X86?
//!
//! X86 is a variable-length CISC ISA. On in-order cores, the front-end
//! decoder must determine instruction boundaries before dispatching.
//! Instruction bundling organizes the instruction stream into fixed-size
//! chunks (bundles) that the decoder can process as a unit, reducing
//! decode complexity and enabling:
//!
//! - **Parallel decode**: Multiple instructions in a bundle can be
//!   decoded simultaneously if they fit within the bundle.
//! - **Precise exceptions**: By committing state only at bundle
//!   boundaries, the CPU can precisely roll back to the start of a
//!   bundle on an exception.
//! - **Reduced front-end stalls**: Fixed bundle boundaries prevent
//!   instructions from spanning decode windows, eliminating
//!   instruction-split stalls.
//!
//! ## Microarchitecture-Specific Rules
//!
//! ### Intel Atom (Bonnell / Saltwell)
//!
//! - **Bundle size**: 8 bytes (aligned to 8-byte boundary).
//! - **Decode width**: 2 instructions per cycle.
//! - **Rules**:
//!   - No instruction may cross an 8-byte boundary.
//!   - Branches must end a bundle.
//!   - Long instructions (≥ 8 bytes) occupy an entire bundle.
//!   - Prefix-heavy instructions may be restricted.
//!
//! ### Intel Silvermont / Airmont
//!
//! - **Bundle size**: 8 bytes (aligned).
//! - **Decode width**: 2-wide decode.
//! - **Improvements over Bonnell**: Slightly more flexible bundling;
//!   some instructions up to 11 bytes allowed if they don't cross
//!   a 16-byte boundary.
//!
//! ### Intel Goldmont / Goldmont Plus
//!
//! - **Bundle size**: 16 bytes (aligned).
//! - **Decode width**: 3-wide decode.
//! - **Rules**:
//!   - Bundles aligned to 16-byte boundaries.
//!   - Up to 3 instructions per bundle.
//!   - Complex instructions that decode to > 3 μops may limit bundle
//!     occupancy.
//!   - Macro-fused pairs occupy 1 slot.
//!
//! ### Intel Tremont / Gracemont / Crestmont / Skymont
//!
//! - **Bundle size**: 16 bytes (aligned) or 32 bytes (dual-cluster).
//! - **Decode width**: 2×3-wide decode clusters.
//! - **Rules**:
//!   - Dual decode clusters each process 16-byte bundles.
//!   - Up to 6 instructions per cycle (3 per cluster).
//!   - Branches end bundles; indirect branches may end a cluster.
//!
//! ## Clean-room reconstruction from:
//!
//! - Intel® 64 and IA-32 Architectures Software Developer's Manual
//!   (volumes 1, 2A, 2B, 2C, 2D, 3A, 3B, 3C, 3D)
//! - Intel® 64 and IA-32 Architectures Optimization Reference Manual
//!   (Order #: 248966)
//! - Intel® Atom™ Microarchitecture Optimization Guide
//! - Intel® Silvermont Microarchitecture Optimization Guide
//! - Intel® Goldmont Microarchitecture Optimization Guide
//! - Intel® Tremont Microarchitecture Optimization Guide
//! - Agner Fog's microarchitecture documentation
//! - uops.info instruction decode measurements
//!
//! Zero LLVM source code consultation. All behavior reconstructed from
//! published specifications and black-box oracle interrogation.

#![allow(non_upper_case_globals, dead_code)]

use std::collections::{HashMap, HashSet};
use std::fmt;

// ============================================================================
// Core Types
// ============================================================================

/// The primary instruction bundling engine for X86/X86-64.
#[derive(Debug, Clone)]
pub struct X86InstrBundling {
    /// Target microarchitecture for bundling rules.
    pub microarch: BundleMicroArch,
    /// Whether targeting 64-bit mode.
    pub is_64bit: bool,
    /// Configuration for bundling behavior.
    pub config: BundleConfig,
    /// Statistics accumulated during bundling.
    pub stats: BundleStats,
    /// Current bundle being formed.
    pub current_bundle: Option<InstrBundle>,
    /// All completed bundles.
    pub bundles: Vec<InstrBundle>,
    /// Current byte offset within the code section.
    pub code_offset: u64,
    /// Label positions (for branch target alignment).
    pub labels: HashMap<String, u64>,
}

/// Microarchitecture-specific bundling target.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub enum BundleMicroArch {
    /// Generic in-order core with safe defaults.
    GenericInOrder,
    /// Intel Atom (Bonnell, 45nm, 2008).
    Bonnell,
    /// Intel Atom (Saltwell, 32nm, 2011).
    Saltwell,
    /// Intel Silvermont (22nm, 2013).
    Silvermont,
    /// Intel Airmont (14nm, 2015).
    Airmont,
    /// Intel Goldmont (14nm, 2016).
    Goldmont,
    /// Intel Goldmont Plus (14nm, 2017).
    GoldmontPlus,
    /// Intel Tremont (10nm, 2020).
    Tremont,
    /// Intel Gracemont (Intel 7, 2021).
    Gracemont,
    /// Intel Crestmont (Intel 4, 2023).
    Crestmont,
    /// Intel Skymont (Intel 3/20A, 2024).
    Skymont,
    /// Intel Darkmont (Intel 18A, 2025).
    Darkmont,
    /// Generic out-of-order core (bundling not strictly needed, but
    /// useful for alignment).
    OutOfOrder,
}

impl BundleMicroArch {
    /// Get the native bundle size for this microarchitecture.
    pub fn bundle_size(&self) -> u32 {
        match self {
            BundleMicroArch::GenericInOrder => 8,
            BundleMicroArch::Bonnell => 8,
            BundleMicroArch::Saltwell => 8,
            BundleMicroArch::Silvermont => 8,
            BundleMicroArch::Airmont => 8,
            BundleMicroArch::Goldmont => 16,
            BundleMicroArch::GoldmontPlus => 16,
            BundleMicroArch::Tremont => 16,
            BundleMicroArch::Gracemont => 16,
            BundleMicroArch::Crestmont => 16,
            BundleMicroArch::Skymont => 16,
            BundleMicroArch::Darkmont => 16,
            BundleMicroArch::OutOfOrder => 16,
        }
    }

    /// Get the decode width (instructions per cycle) for this microarch.
    pub fn decode_width(&self) -> u8 {
        match self {
            BundleMicroArch::GenericInOrder => 2,
            BundleMicroArch::Bonnell => 2,
            BundleMicroArch::Saltwell => 2,
            BundleMicroArch::Silvermont => 2,
            BundleMicroArch::Airmont => 2,
            BundleMicroArch::Goldmont => 3,
            BundleMicroArch::GoldmontPlus => 3,
            BundleMicroArch::Tremont => 6, // 2×3 cluster
            BundleMicroArch::Gracemont => 6,
            BundleMicroArch::Crestmont => 6,
            BundleMicroArch::Skymont => 6,
            BundleMicroArch::Darkmont => 6,
            BundleMicroArch::OutOfOrder => 6,
        }
    }

    /// Whether branches must end at a bundle boundary on this
    /// microarchitecture.
    pub fn requires_branch_at_bundle_end(&self) -> bool {
        match self {
            BundleMicroArch::Bonnell => true,
            BundleMicroArch::Saltwell => true,
            BundleMicroArch::Silvermont => true,
            BundleMicroArch::Airmont => true,
            _ => false,
        }
    }

    /// Whether bundle locking is supported on this microarchitecture.
    pub fn supports_bundle_locking(&self) -> bool {
        match self {
            BundleMicroArch::Bonnell
            | BundleMicroArch::Saltwell
            | BundleMicroArch::Silvermont
            | BundleMicroArch::Airmont => true,
            BundleMicroArch::Goldmont
            | BundleMicroArch::GoldmontPlus
            | BundleMicroArch::Tremont
            | BundleMicroArch::Gracemont
            | BundleMicroArch::Crestmont
            | BundleMicroArch::Skymont
            | BundleMicroArch::Darkmont => true,
            BundleMicroArch::OutOfOrder => false,
            _ => false,
        }
    }

    /// Maximum number of prefix bytes tolerated in a single instruction
    /// within a bundle (exceeding this forces a new bundle).
    pub fn max_prefix_bytes(&self) -> u8 {
        match self {
            BundleMicroArch::Bonnell | BundleMicroArch::Saltwell => 2,
            BundleMicroArch::Silvermont | BundleMicroArch::Airmont => 3,
            BundleMicroArch::Goldmont | BundleMicroArch::GoldmontPlus => 4,
            _ => 4,
        }
    }
}

/// Configuration for instruction bundling.
#[derive(Debug, Clone)]
pub struct BundleConfig {
    /// Bundle size in bytes (overrides microarch default if > 0).
    pub bundle_size_override: u32,
    /// Maximum instructions per bundle (0 = use microarch default).
    pub max_instrs_per_bundle: u8,
    /// Whether to enforce bundle boundary alignment.
    pub enforce_alignment: bool,
    /// Whether to insert NOPs to pad bundles to their full size.
    pub pad_with_nops: bool,
    /// Whether to lock bundles for precise exception handling.
    pub lock_bundles: bool,
    /// Whether branches must end at a bundle boundary.
    pub branch_at_bundle_end: bool,
    /// Whether to use adaptive bundling (vary bundle size based on
    /// instruction mix).
    pub adaptive_bundling: bool,
    /// Whether to allow macro-fused pairs to cross bundle boundaries.
    pub allow_macro_fusion_cross_bundle: bool,
}

impl Default for BundleConfig {
    fn default() -> Self {
        BundleConfig {
            bundle_size_override: 0,
            max_instrs_per_bundle: 0,
            enforce_alignment: true,
            pad_with_nops: true,
            lock_bundles: false,
            branch_at_bundle_end: false,
            adaptive_bundling: false,
            allow_macro_fusion_cross_bundle: false,
        }
    }
}

/// Statistics for bundling operations.
#[derive(Debug, Clone, Default)]
pub struct BundleStats {
    /// Total number of bundles formed.
    pub total_bundles: u64,
    /// Total number of instructions bundled.
    pub total_instructions: u64,
    /// Number of NOP padding bytes inserted.
    pub nop_padding_bytes: u64,
    /// Number of times a bundle boundary was crossed mid-instruction
    /// (forced split).
    pub boundary_splits: u64,
    /// Number of locked bundles.
    pub locked_bundles: u64,
    /// Average instructions per bundle.
    pub avg_instrs_per_bundle: f64,
    /// Average bundle utilization (percentage of bundle bytes used).
    pub avg_utilization: f64,
}

// ============================================================================
// Instruction Representation for Bundling
// ============================================================================

/// A single instruction to be placed within a bundle.
#[derive(Debug, Clone)]
pub struct BundleInstr {
    /// Unique instruction ID.
    pub id: u64,
    /// Mnemonic for debugging.
    pub mnemonic: String,
    /// Instruction byte size (including prefixes, opcode, ModR/M, SIB,
    /// displacement, immediate).
    pub size: u8,
    /// Whether this instruction is a branch (conditional or
    /// unconditional).
    pub is_branch: bool,
    /// Whether this instruction is an indirect branch.
    pub is_indirect_branch: bool,
    /// Whether this instruction is a call.
    pub is_call: bool,
    /// Whether this instruction is a return.
    pub is_return: bool,
    /// Number of prefix bytes.
    pub prefix_bytes: u8,
    /// Number of μops this instruction decodes to (0 = unknown, use
    /// complexity estimate).
    pub uop_count: u8,
    /// Whether this instruction is part of a macro-fused pair.
    pub is_macro_fused: bool,
    /// If macro-fused, the ID of the other instruction in the pair.
    pub macro_fuse_partner: Option<u64>,
    /// Whether this instruction is a NOP (padding).
    pub is_nop: bool,
    /// Original code offset (for debugging).
    pub original_offset: u64,
    /// Whether this instruction requires a lock prefix.
    pub uses_lock_prefix: bool,
    /// Whether this instruction accesses memory.
    pub is_memory: bool,
    /// The instruction's raw bytes.
    pub bytes: Vec<u8>,
}

impl BundleInstr {
    /// Create a new instruction for bundling.
    pub fn new(id: u64, mnemonic: &str, bytes: Vec<u8>) -> Self {
        let size = bytes.len() as u8;
        BundleInstr {
            id,
            mnemonic: mnemonic.to_string(),
            size,
            is_branch: false,
            is_indirect_branch: false,
            is_call: false,
            is_return: false,
            prefix_bytes: 0,
            uop_count: 1,
            is_macro_fused: false,
            macro_fuse_partner: None,
            is_nop: false,
            original_offset: 0,
            uses_lock_prefix: false,
            is_memory: false,
            bytes,
        }
    }

    /// Mark this instruction as a branch.
    pub fn mark_branch(&mut self, is_indirect: bool) {
        self.is_branch = true;
        self.is_indirect_branch = is_indirect;
    }

    /// Mark this instruction as a call.
    pub fn mark_call(&mut self) {
        self.is_call = true;
        self.is_branch = true;
    }

    /// Mark this instruction as a return.
    pub fn mark_return(&mut self) {
        self.is_return = true;
        self.is_branch = true;
    }

    /// Classify the instruction for bundling purposes.
    pub fn classify(&self) -> InstrClass {
        if self.is_nop {
            InstrClass::Nop
        } else if self.is_return {
            InstrClass::Return
        } else if self.is_call {
            InstrClass::Call
        } else if self.is_indirect_branch {
            InstrClass::IndirectBranch
        } else if self.is_branch {
            InstrClass::ConditionalBranch
        } else if self.uses_lock_prefix {
            InstrClass::Locked
        } else if self.is_memory {
            InstrClass::Memory
        } else {
            InstrClass::Alu
        }
    }
}

/// Instruction class for bundle slot assignment.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub enum InstrClass {
    /// Arithmetic/logic instruction.
    Alu,
    /// Memory load/store.
    Memory,
    /// Conditional branch (JCC).
    ConditionalBranch,
    /// Unconditional direct branch (JMP).
    UnconditionalBranch,
    /// Indirect branch (JMP reg/mem).
    IndirectBranch,
    /// Call instruction.
    Call,
    /// Return instruction.
    Return,
    /// Locked instruction (atomic RMW).
    Locked,
    /// NOP/padding.
    Nop,
    /// Complex instruction (CPUID, RDTSC, etc.).
    Complex,
}

impl InstrClass {
    /// Whether this instruction class terminates a bundle.
    pub fn terminates_bundle(&self, microarch: BundleMicroArch) -> bool {
        match self {
            InstrClass::IndirectBranch | InstrClass::Call | InstrClass::Return => true,
            InstrClass::ConditionalBranch | InstrClass::UnconditionalBranch => {
                microarch.requires_branch_at_bundle_end()
            }
            InstrClass::Locked => true,
            _ => false,
        }
    }

    /// Whether this instruction class forces a new bundle.
    pub fn forces_new_bundle(&self, microarch: BundleMicroArch) -> bool {
        match self {
            InstrClass::Locked => true,
            InstrClass::Complex => true,
            InstrClass::IndirectBranch => true,
            _ => false,
        }
    }
}

// ============================================================================
// Instruction Bundle
// ============================================================================

/// An instruction bundle: a fixed-size container holding one or more
/// instructions for in-order execution.
#[derive(Debug, Clone)]
pub struct Bundle {
    /// Bundle ID (sequential).
    pub id: u64,
    /// Byte offset of the bundle in the code section.
    pub offset: u64,
    /// Bundle size in bytes.
    pub size: u32,
    /// Instructions contained in the bundle.
    pub instructions: Vec<BundleInstr>,
    /// Number of bytes used (excluding padding).
    pub used_bytes: u32,
    /// Whether this bundle is locked (atomic exception boundary).
    pub is_locked: bool,
    /// Whether a branch terminates this bundle.
    pub ends_with_branch: bool,
    /// Bundle alignment boundary.
    pub alignment: u32,
    /// Bundle flags.
    pub flags: BundleFlags,
}

/// Bundle flags.
#[derive(Debug, Clone, Default)]
pub struct BundleFlags {
    /// Whether this is the first bundle of a function.
    pub function_entry: bool,
    /// Whether this bundle contains a loop header.
    pub loop_header: bool,
    /// Whether this bundle is a branch target.
    pub branch_target: bool,
    /// Whether NOP padding was added.
    pub has_padding: bool,
    /// Whether this bundle is a "fetch bundle" (aligned to fetch
    /// boundary).
    pub fetch_aligned: bool,
    /// Whether the bundle is full (no more instructions fit).
    pub is_full: bool,
}

/// Legacy alias for `Bundle` used in older code paths.
pub type InstrBundle = Bundle;

// ============================================================================
// Bundle Formation
// ============================================================================

impl X86InstrBundling {
    /// Create a new instruction bundling engine.
    pub fn new(microarch: BundleMicroArch, is_64bit: bool) -> Self {
        X86InstrBundling {
            microarch,
            is_64bit,
            config: BundleConfig::default(),
            stats: BundleStats::default(),
            current_bundle: None,
            bundles: vec![],
            code_offset: 0,
            labels: HashMap::new(),
        }
    }

    /// Create with custom configuration.
    pub fn with_config(microarch: BundleMicroArch, is_64bit: bool, config: BundleConfig) -> Self {
        let mut engine = Self::new(microarch, is_64bit);
        engine.config = config;
        engine
    }

    /// Get the effective bundle size.
    pub fn bundle_size(&self) -> u32 {
        if self.config.bundle_size_override > 0 {
            self.config.bundle_size_override
        } else {
            self.microarch.bundle_size()
        }
    }

    /// Get the maximum instructions per bundle.
    pub fn max_instrs_per_bundle(&self) -> u8 {
        if self.config.max_instrs_per_bundle > 0 {
            self.config.max_instrs_per_bundle
        } else {
            self.microarch.decode_width()
        }
    }

    /// Start a new instruction bundle at the current code offset.
    pub fn begin_bundle(&mut self) {
        let size = self.bundle_size();
        let alignment = if self.config.enforce_alignment {
            size
        } else {
            1
        };

        // Align code offset to bundle boundary if required.
        if self.config.enforce_alignment && self.code_offset % size as u64 != 0 {
            let pad = (size as u64) - (self.code_offset % size as u64);
            self.code_offset += pad;
            self.stats.nop_padding_bytes += pad;
        }

        let bundle = Bundle {
            id: self.bundles.len() as u64,
            offset: self.code_offset,
            size,
            instructions: vec![],
            used_bytes: 0,
            is_locked: self.config.lock_bundles,
            ends_with_branch: false,
            alignment,
            flags: BundleFlags::default(),
        };

        self.current_bundle = Some(bundle);
    }

    /// Add an instruction to the current bundle.
    ///
    /// Returns `true` if the instruction was added successfully.
    /// Returns `false` if the instruction doesn't fit and a new bundle
    /// would be needed.
    pub fn add_instruction(&mut self, instr: &BundleInstr) -> BundleResult {
        let result = self.try_add_instruction(instr);

        match &result {
            BundleResult::Added => {
                self.stats.total_instructions += 1;
            }
            BundleResult::NewBundleNeeded => {
                self.stats.boundary_splits += 1;
            }
            _ => {}
        }

        result
    }

    /// Try to add an instruction without side effects (for speculative
    /// bundling queries).
    pub fn try_add_instruction(&mut self, instr: &BundleInstr) -> BundleResult {
        if self.current_bundle.is_none() {
            self.begin_bundle();
        }

        let bundle = self.current_bundle.as_ref().unwrap();
        let bundle_size = bundle.size;
        let max_instrs = self.max_instrs_per_bundle();

        // Check instruction count limit.
        if bundle.instructions.len() >= max_instrs as usize {
            return BundleResult::NewBundleNeeded;
        }

        // Check: does the instruction fit in the remaining space?
        let remaining = bundle_size - bundle.used_bytes;
        if (instr.size as u32) > remaining {
            // Instruction doesn't fit. Can we split or must we start
            // a new bundle?
            return BundleResult::NewBundleNeeded;
        }

        // Check microarchitecture-specific constraints.
        let class = instr.classify();

        // Some instruction classes force bundle boundary.
        if class.forces_new_bundle(self.microarch) && !bundle.instructions.is_empty() {
            return BundleResult::NewBundleNeeded;
        }

        // Check prefix limit.
        if instr.prefix_bytes > self.microarch.max_prefix_bytes() {
            // Prefix-heavy instruction may need special handling.
            // On older Atom cores, this forces a new bundle.
            if matches!(
                self.microarch,
                BundleMicroArch::Bonnell | BundleMicroArch::Saltwell
            ) {
                if !bundle.instructions.is_empty() {
                    return BundleResult::NewBundleNeeded;
                }
            }
        }

        // Check macro-fusion constraints.
        if instr.is_macro_fused {
            // Macro-fused pairs should be in the same bundle.
            let partner_id = instr.macro_fuse_partner;
            let partner_in_bundle = bundle.instructions.iter().any(|i| Some(i.id) == partner_id);

            if partner_in_bundle {
                // Partner already added; this is the second half.
                // Check bundle capacity.
                if (instr.size as u32) > remaining {
                    // If we can't fit the fusion pair, we should ideally
                    // keep them together. This is a policy decision.
                    if self.config.allow_macro_fusion_cross_bundle {
                        // Allow splitting the pair.
                    } else {
                        // Reject: would split a fusion pair.
                        return BundleResult::FusionPairSplit;
                    }
                }
            }
        }

        // If this is a branch and the microarch requires branches at
        // bundle end, verify that the branch is the last instruction
        // in the bundle.
        if class.terminates_bundle(self.microarch) && !bundle.instructions.is_empty() {
            // This instruction would terminate the bundle. That's OK
            // as long as it's the last instruction.
            if (instr.size as u32) <= remaining {
                // It fits. After adding, the bundle will end.
                return BundleResult::Added;
            } else {
                return BundleResult::NewBundleNeeded;
            }
        }

        // NOPs always fit if there's space (they're padding).
        if instr.is_nop && (instr.size as u32) <= remaining {
            return BundleResult::Added;
        }

        BundleResult::Added
    }

    /// Commit the current instruction to the current bundle.
    pub fn commit_instruction(&mut self, instr: BundleInstr) {
        if self.current_bundle.is_none() {
            self.begin_bundle();
        }

        let bundle = self.current_bundle.as_mut().unwrap();
        let class = instr.classify();

        bundle.used_bytes += instr.size as u32;
        bundle.instructions.push(instr);

        if class.terminates_bundle(self.microarch) {
            bundle.ends_with_branch = true;
            self.finish_bundle();
        }
    }

    /// Finish the current bundle: close it and add it to the completed
    /// list, then start a new one.
    pub fn finish_bundle(&mut self) {
        if let Some(mut bundle) = self.current_bundle.take() {
            // Add NOP padding if configured.
            if self.config.pad_with_nops && bundle.used_bytes < bundle.size {
                let pad_size = bundle.size - bundle.used_bytes;
                let nop_instr = BundleInstr {
                    id: u64::MAX,
                    mnemonic: "NOP".to_string(),
                    size: pad_size as u8,
                    is_branch: false,
                    is_indirect_branch: false,
                    is_call: false,
                    is_return: false,
                    prefix_bytes: 0,
                    uop_count: 0,
                    is_macro_fused: false,
                    macro_fuse_partner: None,
                    is_nop: true,
                    original_offset: self.code_offset,
                    uses_lock_prefix: false,
                    is_memory: false,
                    bytes: vec![0x90; pad_size as usize],
                };
                bundle.instructions.push(nop_instr);
                bundle.used_bytes = bundle.size;
                bundle.flags.has_padding = true;
                self.stats.nop_padding_bytes += pad_size as u64;
            }

            bundle.flags.is_full = bundle.used_bytes >= bundle.size;

            // Update bundle flags.
            if bundle.instructions.iter().any(|i| i.is_branch) {
                bundle.ends_with_branch = true;
            }

            self.stats.total_bundles += 1;
            if self.config.lock_bundles {
                self.stats.locked_bundles += 1;
            }

            self.code_offset += bundle.size as u64;
            self.bundles.push(bundle);
        }
    }

    /// Flush the current bundle (finish it) and return all bundles.
    pub fn flush(&mut self) -> Vec<Bundle> {
        self.finish_bundle();
        let bundles = self.bundles.clone();
        self.update_stats();
        bundles
    }

    /// Process a sequence of instructions into bundles.
    pub fn bundle_instructions(&mut self, instrs: &[BundleInstr]) -> Vec<Bundle> {
        self.begin_bundle();

        for instr in instrs {
            match self.add_instruction(instr) {
                BundleResult::Added => {
                    self.commit_instruction(instr.clone());
                }
                BundleResult::NewBundleNeeded => {
                    self.finish_bundle();
                    self.begin_bundle();
                    // Retry adding the instruction.
                    if self.add_instruction(instr) == BundleResult::Added {
                        self.commit_instruction(instr.clone());
                    } else {
                        // Even an empty bundle can't hold this instruction.
                        // Create a bundle just for it.
                        let bundle = self.current_bundle.as_mut().unwrap();
                        bundle.instructions.push(instr.clone());
                        bundle.used_bytes = instr.size as u32;
                        if instr.classify().terminates_bundle(self.microarch) {
                            bundle.ends_with_branch = true;
                        }
                        self.finish_bundle();
                    }
                }
                BundleResult::FusionPairSplit => {
                    // Handle by finishing current bundle and starting
                    // a new one for the fusion pair.
                    self.finish_bundle();
                    self.begin_bundle();
                    self.commit_instruction(instr.clone());
                }
                BundleResult::Illegal => {
                    // Skip illegal instruction in bundling.
                }
            }
        }

        self.flush()
    }

    /// Update derived statistics.
    fn update_stats(&mut self) {
        if self.stats.total_bundles > 0 {
            self.stats.avg_instrs_per_bundle =
                self.stats.total_instructions as f64 / self.stats.total_bundles as f64;

            let total_capacity = self.stats.total_bundles * self.bundle_size() as u64;
            let total_used: u64 = self.bundles.iter().map(|b| b.used_bytes as u64).sum();
            if total_capacity > 0 {
                self.stats.avg_utilization = (total_used as f64 / total_capacity as f64) * 100.0;
            }
        }
    }
}

/// Result of trying to add an instruction to a bundle.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum BundleResult {
    /// Instruction was successfully added to the bundle.
    Added,
    /// A new bundle is needed; the instruction doesn't fit.
    NewBundleNeeded,
    /// Adding this instruction would split a macro-fusion pair.
    FusionPairSplit,
    /// The instruction cannot be bundled (illegal).
    Illegal,
}

// ============================================================================
// Bundle Boundary Enforcement
// ============================================================================

/// Check whether an instruction crosses a bundle boundary and, if so,
/// return the number of NOP bytes needed to align it.
pub fn check_bundle_boundary(instr_offset: u64, instr_size: u8, bundle_size: u32) -> Option<u32> {
    let bundle_end = ((instr_offset / bundle_size as u64) + 1) * bundle_size as u64;
    let instr_end = instr_offset + instr_size as u64;

    if instr_end > bundle_end {
        // Instruction crosses a bundle boundary.
        // Return the number of bytes until the next bundle.
        Some((bundle_end - instr_offset) as u32)
    } else {
        None
    }
}

/// Compute the NOP padding required to ensure the next instruction
/// starts at a bundle boundary.
pub fn compute_bundle_alignment(current_offset: u64, bundle_size: u32) -> u32 {
    let misalignment = current_offset % bundle_size as u64;
    if misalignment == 0 {
        0
    } else {
        (bundle_size as u64 - misalignment) as u32
    }
}

/// Split a long instruction that crosses a bundle boundary into two
/// parts: the bytes fitting in the current bundle, and the remaining
/// bytes that must go into the next bundle.
///
/// Note: On X86, instructions are atomic — you cannot split an
/// instruction across bundles. This function is provided for
/// analysis/diagnostic purposes only.
pub fn diagnose_boundary_crossing(
    instr_offset: u64,
    instr_size: u8,
    bundle_size: u32,
) -> BoundaryCrossingInfo {
    let bundle_end = ((instr_offset / bundle_size as u64) + 1) * bundle_size as u64;
    let instr_end = instr_offset + instr_size as u64;

    BoundaryCrossingInfo {
        crosses_boundary: instr_end > bundle_end,
        bundle_size,
        offset: instr_offset,
        instr_size,
        bytes_in_current_bundle: if instr_end > bundle_end {
            (bundle_end - instr_offset) as u8
        } else {
            instr_size
        },
        bytes_in_next_bundle: if instr_end > bundle_end {
            (instr_end - bundle_end) as u8
        } else {
            0
        },
        next_bundle_offset: bundle_end,
    }
}

/// Information about a bundle boundary crossing.
#[derive(Debug, Clone)]
pub struct BoundaryCrossingInfo {
    /// Whether the instruction crosses a bundle boundary.
    pub crosses_boundary: bool,
    /// The bundle size being checked against.
    pub bundle_size: u32,
    /// Instruction offset in the section.
    pub offset: u64,
    /// Total instruction size.
    pub instr_size: u8,
    /// Bytes of the instruction that fit in the current bundle.
    pub bytes_in_current_bundle: u8,
    /// Bytes of the instruction that would spill into the next bundle.
    pub bytes_in_next_bundle: u8,
    /// The offset of the next bundle boundary.
    pub next_bundle_offset: u64,
}

impl BoundaryCrossingInfo {
    /// Whether this crossing requires a NOP sled to align the instruction.
    pub fn needs_nop_sled(&self) -> bool {
        self.crosses_boundary
    }

    /// The number of NOP bytes to insert before the instruction to
    /// avoid boundary crossing.
    pub fn nop_sled_size(&self) -> u32 {
        if !self.crosses_boundary {
            0
        } else {
            self.bytes_in_current_bundle as u32
        }
    }
}

// ============================================================================
// Bundle Locking for Precise Exception Handling
// ============================================================================

/// Bundle locking ensures that all instructions in a locked bundle
/// either complete together or are rolled back together. This enables
/// precise exception handling on in-order cores where the architectural
/// state is only committed at bundle boundaries.
#[derive(Debug, Clone)]
pub struct BundleLock {
    /// Whether the bundle is locked.
    pub locked: bool,
    /// Lock granularity: bundle-level or instruction-level.
    pub granularity: LockGranularity,
    /// Number of instructions in the locked region.
    pub locked_instr_count: u8,
    /// Exception PC to report on failure.
    pub exception_pc: u64,
}

/// Lock granularity for bundle locking.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum LockGranularity {
    /// Lock the entire bundle: all instructions commit or none do.
    Bundle,
    /// Lock individual instructions within the bundle.
    Instruction,
    /// Lock pairs of instructions (for macro-fused pairs).
    Pair,
    /// No locking; instructions commit independently.
    None,
}

impl BundleLock {
    /// Create a bundle-level lock.
    pub fn bundle_lock() -> Self {
        BundleLock {
            locked: true,
            granularity: LockGranularity::Bundle,
            locked_instr_count: 0,
            exception_pc: 0,
        }
    }

    /// Create an instruction-level lock.
    pub fn instruction_lock() -> Self {
        BundleLock {
            locked: true,
            granularity: LockGranularity::Instruction,
            locked_instr_count: 0,
            exception_pc: 0,
        }
    }

    /// Create an unlocked region.
    pub fn unlocked() -> Self {
        BundleLock {
            locked: false,
            granularity: LockGranularity::None,
            locked_instr_count: 0,
            exception_pc: 0,
        }
    }
}

/// Apply bundle locking to a sequence of bundles.
///
/// On in-order cores, locking is implemented by deferring register
/// writes and memory stores until the bundle boundary. If an exception
/// occurs mid-bundle, all deferred writes are discarded and the
/// architectural state reflects the state at the start of the bundle.
pub fn lock_bundles(bundles: &mut [Bundle], microarch: BundleMicroArch) {
    if !microarch.supports_bundle_locking() {
        return;
    }

    for bundle in bundles.iter_mut() {
        bundle.is_locked = true;
    }
}

/// Check whether an exception at a given instruction offset within a
/// locked bundle is recoverable.
pub fn is_exception_recoverable(bundle: &Bundle, faulting_instr_index: usize) -> bool {
    if !bundle.is_locked {
        // Without locking, exception might leave partial state.
        return false;
    }

    // In a locked bundle, any fault is recoverable by rolling back
    // all instructions in the bundle.
    faulting_instr_index < bundle.instructions.len()
}

/// Generate the register checkpoint sequence for a locked bundle.
///
/// Before executing a locked bundle, the CPU (or emulator) must
/// checkpoint the architectural registers. This function returns the
/// list of registers that need checkpointing for precise recovery.
pub fn checkpoint_registers_for_bundle(is_64bit: bool) -> Vec<&'static str> {
    let mut regs = vec!["RAX", "RCX", "RDX", "RBX", "RSP", "RBP", "RSI", "RDI"];
    if is_64bit {
        regs.extend_from_slice(&["R8", "R9", "R10", "R11", "R12", "R13", "R14", "R15"]);
    }
    // XMM registers for floating-point state.
    regs.extend_from_slice(&[
        "XMM0", "XMM1", "XMM2", "XMM3", "XMM4", "XMM5", "XMM6", "XMM7",
    ]);
    if is_64bit {
        regs.extend_from_slice(&[
            "XMM8", "XMM9", "XMM10", "XMM11", "XMM12", "XMM13", "XMM14", "XMM15",
        ]);
    }
    // EFLAGS/RFLAGS
    regs.push("RFLAGS");

    regs
}

// ============================================================================
// Microarchitecture-Specific Bundle Rules
// ============================================================================

/// Atom (Bonnell/Saltwell) bundle rules:
/// - 8-byte bundles aligned to 8-byte boundaries.
/// - No instruction may cross an 8-byte boundary.
/// - Branches must end a bundle.
/// - Long instructions (≥ 8 bytes) occupy a whole bundle.
/// - Max 2 instructions per bundle.
/// - Lock prefix forces a new bundle.
pub mod atom_rules {
    use super::*;

    /// Check if an instruction sequence is valid under Atom bundling rules.
    pub fn validate_atom_bundle(bundle: &Bundle) -> Result<(), String> {
        if bundle.size != 8 {
            return Err(format!("Atom bundle must be 8 bytes, got {}", bundle.size));
        }

        if bundle.instructions.len() > 2 {
            return Err(format!(
                "Atom bundle max 2 instructions, got {}",
                bundle.instructions.len()
            ));
        }

        // Check that no instruction crosses the 8-byte boundary.
        let mut byte_pos = 0u32;
        for (i, instr) in bundle.instructions.iter().enumerate() {
            if byte_pos + instr.size as u32 > 8 {
                return Err(format!(
                    "Instruction {} (id={}) crosses 8-byte boundary at position {}",
                    i, instr.id, byte_pos
                ));
            }

            // Branches must be the last instruction.
            if instr.is_branch && i != bundle.instructions.len() - 1 {
                return Err(format!(
                    "Branch instruction {} (id={}) is not the last in the bundle",
                    i, instr.id
                ));
            }

            byte_pos += instr.size as u32;
        }

        Ok(())
    }

    /// Apply Atom-specific bundling constraints.
    pub fn apply_atom_constraints(instrs: &[BundleInstr], is_64bit: bool) -> Vec<Bundle> {
        let config = BundleConfig {
            bundle_size_override: 8,
            max_instrs_per_bundle: 2,
            enforce_alignment: true,
            pad_with_nops: true,
            lock_bundles: false,
            branch_at_bundle_end: true,
            adaptive_bundling: false,
            allow_macro_fusion_cross_bundle: false,
        };
        let mut engine = X86InstrBundling::with_config(BundleMicroArch::Bonnell, is_64bit, config);
        engine.bundle_instructions(instrs)
    }
}

/// Silvermont/Airmont bundle rules:
/// - 8-byte bundles aligned to 8-byte boundaries.
/// - Instructions up to 11 bytes allowed.
/// - Max 2 instructions per bundle.
/// - Lock prefix allowed within bundle.
/// - Branches end bundles.
pub mod silvermont_rules {
    use super::*;

    /// Check if an instruction sequence is valid under Silvermont
    /// bundling rules.
    pub fn validate_silvermont_bundle(bundle: &Bundle) -> Result<(), String> {
        if bundle.size != 8 {
            return Err(format!(
                "Silvermont bundle must be 8 bytes, got {}",
                bundle.size
            ));
        }

        if bundle.instructions.len() > 2 {
            return Err(format!(
                "Silvermont bundle max 2 instructions, got {}",
                bundle.instructions.len()
            ));
        }

        let mut byte_pos = 0u32;
        for (i, instr) in bundle.instructions.iter().enumerate() {
            // Silvermont allows up to 11-byte instructions (with
            // relaxed alignment).
            if instr.size > 11 {
                return Err(format!(
                    "Instruction {} exceeds Silvermont max size of 11 bytes (got {})",
                    i, instr.size
                ));
            }

            if byte_pos + instr.size as u32 > 8 {
                return Err(format!(
                    "Instruction {} crosses Silvermont 8-byte bundle boundary",
                    i
                ));
            }

            if instr.is_branch && i != bundle.instructions.len() - 1 {
                return Err(format!("Branch must end Silvermont bundle"));
            }

            byte_pos += instr.size as u32;
        }

        Ok(())
    }

    /// Apply Silvermont-specific bundling constraints.
    pub fn apply_silvermont_constraints(instrs: &[BundleInstr], is_64bit: bool) -> Vec<Bundle> {
        let config = BundleConfig {
            bundle_size_override: 8,
            max_instrs_per_bundle: 2,
            enforce_alignment: true,
            pad_with_nops: true,
            lock_bundles: false,
            branch_at_bundle_end: true,
            adaptive_bundling: false,
            allow_macro_fusion_cross_bundle: false,
        };
        let mut engine =
            X86InstrBundling::with_config(BundleMicroArch::Silvermont, is_64bit, config);
        engine.bundle_instructions(instrs)
    }
}

/// Goldmont/Goldmont Plus bundle rules:
/// - 16-byte bundles aligned to 16-byte boundaries.
/// - Max 3 instructions per bundle.
/// - Complex instructions (≥ 3 μops) may limit occupancy.
/// - Macro-fused pairs count as 1 slot.
/// - Branches do not need to end bundles (but check alignment).
pub mod goldmont_rules {
    use super::*;

    /// Validate a Goldmont bundle.
    pub fn validate_goldmont_bundle(bundle: &Bundle) -> Result<(), String> {
        if bundle.size != 16 {
            return Err(format!(
                "Goldmont bundle must be 16 bytes, got {}",
                bundle.size
            ));
        }

        if bundle.instructions.len() > 3 {
            return Err(format!(
                "Goldmont bundle max 3 instructions, got {}",
                bundle.instructions.len()
            ));
        }

        let mut byte_pos = 0u32;
        for (i, instr) in bundle.instructions.iter().enumerate() {
            if instr.is_nop {
                continue; // NOPs don't count.
            }

            if byte_pos + instr.size as u32 > 16 {
                return Err(format!(
                    "Instruction {} crosses Goldmont 16-byte bundle boundary",
                    i
                ));
            }

            byte_pos += instr.size as u32;
        }

        Ok(())
    }

    /// Apply Goldmont-specific bundling constraints.
    pub fn apply_goldmont_constraints(instrs: &[BundleInstr], is_64bit: bool) -> Vec<Bundle> {
        let config = BundleConfig {
            bundle_size_override: 16,
            max_instrs_per_bundle: 3,
            enforce_alignment: true,
            pad_with_nops: true,
            lock_bundles: false,
            branch_at_bundle_end: false,
            adaptive_bundling: false,
            allow_macro_fusion_cross_bundle: true,
        };
        let mut engine = X86InstrBundling::with_config(BundleMicroArch::Goldmont, is_64bit, config);
        engine.bundle_instructions(instrs)
    }
}

/// Tremont/Gracemont/Crestmont/Skymont bundle rules:
/// - 16-byte bundles (or 32-byte dual-cluster).
/// - Dual decode clusters, each processing a 16-byte bundle.
/// - Max 3 instructions per cluster (6 total).
/// - Branches end bundles.
/// - Indirect branches end clusters.
pub mod tremont_rules {
    use super::*;

    /// Validate a Tremont-family bundle.
    pub fn validate_tremont_bundle(bundle: &Bundle) -> Result<(), String> {
        if bundle.size != 16 {
            return Err(format!(
                "Tremont bundle must be 16 bytes, got {}",
                bundle.size
            ));
        }

        if bundle.instructions.len() > 3 {
            return Err(format!(
                "Tremont bundle max 3 instructions, got {}",
                bundle.instructions.len()
            ));
        }

        let mut byte_pos = 0u32;
        for (i, instr) in bundle.instructions.iter().enumerate() {
            if instr.is_nop {
                continue;
            }

            if byte_pos + instr.size as u32 > 16 {
                return Err(format!("Instruction {} crosses Tremont bundle boundary", i));
            }

            // Indirect branches must end a bundle.
            if instr.is_indirect_branch && i != bundle.instructions.len() - 1 {
                return Err(format!("Indirect branch must end Tremont bundle"));
            }

            byte_pos += instr.size as u32;
        }

        Ok(())
    }

    /// Apply Tremont-family-specific bundling constraints.
    pub fn apply_tremont_constraints(instrs: &[BundleInstr], is_64bit: bool) -> Vec<Bundle> {
        let config = BundleConfig {
            bundle_size_override: 16,
            max_instrs_per_bundle: 3,
            enforce_alignment: true,
            pad_with_nops: true,
            lock_bundles: true,
            branch_at_bundle_end: true,
            adaptive_bundling: true,
            allow_macro_fusion_cross_bundle: false,
        };
        let mut engine = X86InstrBundling::with_config(BundleMicroArch::Tremont, is_64bit, config);
        engine.bundle_instructions(instrs)
    }
}

// ============================================================================
// Branch at Bundle End Requirement
// ============================================================================

/// On some microarchitectures, branch instructions must be the last
/// instruction in a bundle. This function checks whether a branch
/// violates this rule.
pub fn check_branch_at_bundle_end(
    bundle: &Bundle,
    microarch: BundleMicroArch,
) -> Result<(), String> {
    if !microarch.requires_branch_at_bundle_end() {
        return Ok(());
    }

    for (i, instr) in bundle.instructions.iter().enumerate() {
        if instr.is_branch && i != bundle.instructions.len() - 1 {
            return Err(format!(
                "Branch at position {} in bundle {} is not the last instruction \
                 (required by {:?})",
                i, bundle.id, microarch
            ));
        }
    }

    Ok(())
}

/// Ensure branches end at bundle boundaries by inserting NOP padding
/// after the branch to fill the bundle.
pub fn enforce_branch_at_bundle_end(bundle: &mut Bundle, _microarch: BundleMicroArch) {
    if bundle.instructions.is_empty() {
        return;
    }

    let last_is_branch = bundle
        .instructions
        .last()
        .map(|i| i.is_branch)
        .unwrap_or(false);

    if !last_is_branch {
        // Check if any non-last instruction is a branch.
        let branch_pos = bundle.instructions.iter().position(|i| i.is_branch);

        if let Some(pos) = branch_pos {
            if pos < bundle.instructions.len() - 1 {
                // Branch is not last. We need to split the bundle
                // or insert NOPs to fill the remainder.
                // Strategy: truncate to keep branch last, push remaining
                // instructions to next bundle.
                let _split_instrs = bundle.instructions.split_off(pos + 1);
                bundle.ends_with_branch = true;

                // Pad remaining space with NOPs.
                let used = bundle
                    .instructions
                    .iter()
                    .map(|i| i.size as u32)
                    .sum::<u32>();
                if used < bundle.size {
                    let pad = bundle.size - used;
                    let nop = BundleInstr {
                        id: u64::MAX,
                        mnemonic: "NOP".to_string(),
                        size: pad as u8,
                        is_branch: false,
                        is_indirect_branch: false,
                        is_call: false,
                        is_return: false,
                        prefix_bytes: 0,
                        uop_count: 0,
                        is_macro_fused: false,
                        macro_fuse_partner: None,
                        is_nop: true,
                        original_offset: 0,
                        uses_lock_prefix: false,
                        is_memory: false,
                        bytes: vec![0x90; pad as usize],
                    };
                    bundle.instructions.push(nop);
                    bundle.used_bytes = bundle.size;
                    bundle.flags.has_padding = true;
                }
            }
        }
    }
}

// ============================================================================
// Bundle Optimization
// ============================================================================

/// Optimize bundle placement to maximize utilization and minimize NOP
/// padding.
///
/// This pass reorders instructions within bundles (while respecting
/// data dependencies) to pack more instructions per bundle.
pub fn optimize_bundle_packing(bundles: &mut [Bundle], microarch: BundleMicroArch) {
    for bundle in bundles.iter_mut() {
        if bundle.flags.is_full || bundle.instructions.len() <= 1 {
            continue;
        }

        // Simple greedy reorder: sort by size (smallest first) to
        // maximize packing density. This is a heuristic; a full
        // implementation would respect data dependencies.
        let mut instrs = bundle.instructions.clone();

        // Keep branches as the last instruction.
        let branch_idx = instrs.iter().position(|i| i.is_branch);
        let has_branch = branch_idx.is_some();

        // Sort non-branch, non-nop instructions by size ascending.
        let mut non_branch: Vec<_> = instrs
            .iter()
            .enumerate()
            .filter(|(_, i)| !i.is_branch && !i.is_nop)
            .collect();
        non_branch.sort_by_key(|(_, i)| i.size);

        // Rebuild.
        let mut new_instrs = Vec::new();
        let mut used = 0u32;

        for (_orig_idx, instr) in &non_branch {
            if used + instr.size as u32 <= bundle.size {
                used += instr.size as u32;
                new_instrs.push((*instr).clone());
            }
        }

        // Re-add branch at end if present.
        if let Some(idx) = branch_idx {
            if used + instrs[idx].size as u32 <= bundle.size {
                new_instrs.push(instrs[idx].clone());
                used += instrs[idx].size as u32;
            }
        }

        // Add NOP padding.
        if used < bundle.size {
            let pad = bundle.size - used;
            let nop = BundleInstr {
                id: u64::MAX,
                mnemonic: "NOP".to_string(),
                size: pad as u8,
                is_branch: false,
                is_indirect_branch: false,
                is_call: false,
                is_return: false,
                prefix_bytes: 0,
                uop_count: 0,
                is_macro_fused: false,
                macro_fuse_partner: None,
                is_nop: true,
                original_offset: 0,
                uses_lock_prefix: false,
                is_memory: false,
                bytes: vec![0x90; pad as usize],
            };
            new_instrs.push(nop);
        }

        bundle.instructions = new_instrs;
        bundle.used_bytes = bundle.size;
    }
}

/// Compute the bundle utilization efficiency as a percentage.
pub fn bundle_utilization(bundle: &Bundle) -> f64 {
    if bundle.size == 0 {
        return 100.0;
    }
    (bundle.used_bytes as f64 / bundle.size as f64) * 100.0
}

/// Find the bundle with the lowest utilization (most wasted space).
pub fn find_least_utilized_bundle(bundles: &[Bundle]) -> Option<&Bundle> {
    bundles.iter().filter(|b| b.size > 0).min_by(|a, b| {
        a.used_bytes
            .partial_cmp(&b.used_bytes)
            .unwrap_or(std::cmp::Ordering::Equal)
    })
}

// ============================================================================
// Bundle Display / Debugging
// ============================================================================

impl fmt::Display for Bundle {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        write!(
            f,
            "Bundle #{} @ 0x{:04X} ({}B, {} used, {} instrs",
            self.id,
            self.offset,
            self.size,
            self.used_bytes,
            self.instructions.len()
        )?;

        if self.is_locked {
            write!(f, ", LOCKED")?;
        }
        if self.ends_with_branch {
            write!(f, ", ENDS_BR")?;
        }
        if self.flags.has_padding {
            write!(f, ", PADDED")?;
        }
        if self.flags.loop_header {
            write!(f, ", LOOP_HDR")?;
        }
        if self.flags.function_entry {
            write!(f, ", FUNC_ENTRY")?;
        }

        write!(f, "):")?;

        for instr in &self.instructions {
            write!(
                f,
                "\n    {:04X}: {} ({:?}, {}B)",
                self.offset,
                instr.mnemonic,
                instr.classify(),
                instr.size
            )?;
        }

        Ok(())
    }
}

impl X86InstrBundling {
    /// Dump all bundles as a human-readable listing.
    pub fn dump_bundles(&self) -> String {
        let mut output = String::new();
        output.push_str(&format!(
            "=== X86 Instruction Bundles (microarch: {:?}) ===\n",
            self.microarch
        ));
        output.push_str(&format!(
            "Bundle size: {} bytes, max {} instrs/bundle\n",
            self.bundle_size(),
            self.max_instrs_per_bundle()
        ));
        output.push_str(&format!("Total bundles: {}\n", self.bundles.len()));
        output.push_str(&format!(
            "Total instructions: {}\n\n",
            self.stats.total_instructions
        ));

        for bundle in &self.bundles {
            output.push_str(&format!("{}\n", bundle));
        }

        output.push_str(&format!(
            "\nAverage instructions/bundle: {:.2}\n",
            self.stats.avg_instrs_per_bundle
        ));
        output.push_str(&format!(
            "Average utilization: {:.1}%\n",
            self.stats.avg_utilization
        ));
        output.push_str(&format!(
            "NOP padding bytes: {}\n",
            self.stats.nop_padding_bytes
        ));

        output
    }
}

// ============================================================================
// Factory Functions for Testing
// ============================================================================

/// Create an X86InstrBundling engine for Atom (Bonnell).
pub fn make_atom_bundling() -> X86InstrBundling {
    X86InstrBundling::new(BundleMicroArch::Bonnell, true)
}

/// Create an X86InstrBundling engine for Silvermont.
pub fn make_silvermont_bundling() -> X86InstrBundling {
    X86InstrBundling::new(BundleMicroArch::Silvermont, true)
}

/// Create an X86InstrBundling engine for Goldmont.
pub fn make_goldmont_bundling() -> X86InstrBundling {
    X86InstrBundling::new(BundleMicroArch::Goldmont, true)
}

/// Create an X86InstrBundling engine for Tremont.
pub fn make_tremont_bundling() -> X86InstrBundling {
    X86InstrBundling::new(BundleMicroArch::Tremont, true)
}

/// Create a test instruction with given size.
pub fn make_test_instr(id: u64, mnemonic: &str, size: u8) -> BundleInstr {
    BundleInstr {
        id,
        mnemonic: mnemonic.to_string(),
        size,
        is_branch: false,
        is_indirect_branch: false,
        is_call: false,
        is_return: false,
        prefix_bytes: 0,
        uop_count: 1,
        is_macro_fused: false,
        macro_fuse_partner: None,
        is_nop: false,
        original_offset: 0,
        uses_lock_prefix: false,
        is_memory: false,
        bytes: vec![0x90; size as usize],
    }
}

/// Create a test branch instruction.
pub fn make_test_branch(id: u64, mnemonic: &str, size: u8, is_indirect: bool) -> BundleInstr {
    let mut instr = make_test_instr(id, mnemonic, size);
    instr.mark_branch(is_indirect);
    instr
}

/// Create a test call instruction.
pub fn make_test_call(id: u64, size: u8) -> BundleInstr {
    let mut instr = make_test_instr(id, "CALL", size);
    instr.mark_call();
    instr
}

/// Create a test return instruction.
pub fn make_test_return(id: u64, size: u8) -> BundleInstr {
    let mut instr = make_test_instr(id, "RET", size);
    instr.mark_return();
    instr
}

// ============================================================================
// Tests
// ============================================================================

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_bundle_creation() {
        let mut engine = make_atom_bundling();
        engine.begin_bundle();
        assert!(engine.current_bundle.is_some());
        let bundle = engine.current_bundle.as_ref().unwrap();
        assert_eq!(bundle.size, 8);
    }

    #[test]
    fn test_simple_bundling() {
        let mut engine = make_atom_bundling();
        let instrs = vec![
            make_test_instr(1, "ADD", 2),
            make_test_instr(2, "MOV", 3),
            make_test_instr(3, "SUB", 2),
            make_test_instr(4, "XOR", 2),
        ];
        let bundles = engine.bundle_instructions(&instrs);
        assert!(bundles.len() >= 2);
    }

    #[test]
    fn test_branch_ends_bundle_atom() {
        let mut engine = make_atom_bundling();
        let instrs = vec![
            make_test_instr(1, "CMP", 2),
            make_test_branch(2, "JE", 2, false),
            make_test_instr(3, "MOV", 3),
        ];
        let bundles = engine.bundle_instructions(&instrs);
        // Branch should end its bundle.
        let branch_bundle = &bundles[0];
        assert!(branch_bundle.ends_with_branch);
        assert!(branch_bundle.instructions.len() <= 2);
    }

    #[test]
    fn test_atom_bundle_validation() {
        let mut bundle = Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![make_test_instr(1, "ADD", 2), make_test_instr(2, "MOV", 2)],
            used_bytes: 4,
            is_locked: false,
            ends_with_branch: false,
            alignment: 8,
            flags: BundleFlags::default(),
        };
        assert!(atom_rules::validate_atom_bundle(&bundle).is_ok());

        // Add a third instruction — invalid.
        bundle.instructions.push(make_test_instr(3, "SUB", 2));
        assert!(atom_rules::validate_atom_bundle(&bundle).is_err());
    }

    #[test]
    fn test_silvermont_bundle_validation() {
        let bundle = Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![make_test_instr(1, "ADD", 2), make_test_instr(2, "MOV", 2)],
            used_bytes: 4,
            is_locked: false,
            ends_with_branch: false,
            alignment: 8,
            flags: BundleFlags::default(),
        };
        assert!(silvermont_rules::validate_silvermont_bundle(&bundle).is_ok());
    }

    #[test]
    fn test_goldmont_bundle_validation() {
        let bundle = Bundle {
            id: 0,
            offset: 0,
            size: 16,
            instructions: vec![
                make_test_instr(1, "ADD", 2),
                make_test_instr(2, "MOV", 3),
                make_test_instr(3, "SUB", 2),
            ],
            used_bytes: 7,
            is_locked: false,
            ends_with_branch: false,
            alignment: 16,
            flags: BundleFlags::default(),
        };
        assert!(goldmont_rules::validate_goldmont_bundle(&bundle).is_ok());

        // 4 instructions on Goldmont — invalid.
        let invalid_bundle = Bundle {
            id: 1,
            offset: 16,
            size: 16,
            instructions: vec![
                make_test_instr(1, "ADD", 2),
                make_test_instr(2, "MOV", 2),
                make_test_instr(3, "SUB", 2),
                make_test_instr(4, "XOR", 2),
            ],
            used_bytes: 8,
            is_locked: false,
            ends_with_branch: false,
            alignment: 16,
            flags: BundleFlags::default(),
        };
        assert!(goldmont_rules::validate_goldmont_bundle(&invalid_bundle).is_err());
    }

    #[test]
    fn test_tremont_bundle_validation() {
        let bundle = Bundle {
            id: 0,
            offset: 0,
            size: 16,
            instructions: vec![
                make_test_instr(1, "ADD", 3),
                make_test_instr(2, "MOV", 4),
                make_test_instr(3, "SUB", 2),
            ],
            used_bytes: 9,
            is_locked: false,
            ends_with_branch: false,
            alignment: 16,
            flags: BundleFlags::default(),
        };
        assert!(tremont_rules::validate_tremont_bundle(&bundle).is_ok());
    }

    #[test]
    fn test_boundary_crossing_check() {
        // Instruction at offset 6, size 4, bundle size 8.
        let crossing = diagnose_boundary_crossing(6, 4, 8);
        assert!(crossing.crosses_boundary);
        assert_eq!(crossing.bytes_in_current_bundle, 2);
        assert_eq!(crossing.bytes_in_next_bundle, 2);
        assert_eq!(crossing.next_bundle_offset, 8);
    }

    #[test]
    fn test_no_boundary_crossing() {
        // Instruction at offset 4, size 2, bundle size 8.
        let crossing = diagnose_boundary_crossing(4, 2, 8);
        assert!(!crossing.crosses_boundary);
        assert_eq!(crossing.bytes_in_current_bundle, 2);
        assert_eq!(crossing.bytes_in_next_bundle, 0);
    }

    #[test]
    fn test_boundary_crossing_nop_sled() {
        let crossing = diagnose_boundary_crossing(7, 3, 8);
        assert!(crossing.needs_nop_sled());
        assert_eq!(crossing.nop_sled_size(), 1); // 1 byte until boundary
    }

    #[test]
    fn test_bundle_alignment_computation() {
        assert_eq!(compute_bundle_alignment(0, 8), 0);
        assert_eq!(compute_bundle_alignment(8, 8), 0);
        assert_eq!(compute_bundle_alignment(3, 8), 5);
        assert_eq!(compute_bundle_alignment(15, 16), 1);
    }

    #[test]
    fn test_bundle_locking() {
        let mut bundles = vec![Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![make_test_instr(1, "ADD", 2)],
            used_bytes: 2,
            is_locked: false,
            ends_with_branch: false,
            alignment: 8,
            flags: BundleFlags::default(),
        }];
        lock_bundles(&mut bundles, BundleMicroArch::Silvermont);
        assert!(bundles[0].is_locked);
    }

    #[test]
    fn test_exception_recoverability() {
        let bundle = Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![make_test_instr(1, "ADD", 2), make_test_instr(2, "MOV", 2)],
            used_bytes: 4,
            is_locked: true,
            ends_with_branch: false,
            alignment: 8,
            flags: BundleFlags::default(),
        };
        assert!(is_exception_recoverable(&bundle, 0));
        assert!(is_exception_recoverable(&bundle, 1));
        assert!(!is_exception_recoverable(&bundle, 2)); // out of bounds

        let unlocked = Bundle {
            is_locked: false,
            ..bundle.clone()
        };
        assert!(!is_exception_recoverable(&unlocked, 0));
    }

    #[test]
    fn test_checkpoint_registers() {
        let regs_64 = checkpoint_registers_for_bundle(true);
        assert!(regs_64.contains(&"RAX"));
        assert!(regs_64.contains(&"R15"));
        assert!(regs_64.contains(&"XMM15"));
        assert!(regs_64.contains(&"RFLAGS"));

        let regs_32 = checkpoint_registers_for_bundle(false);
        assert!(regs_32.contains(&"RAX"));
        assert!(!regs_32.contains(&"R8"));
    }

    #[test]
    fn test_bundle_flush() {
        let mut engine = make_atom_bundling();
        engine.begin_bundle();
        engine.commit_instruction(make_test_instr(1, "ADD", 2));
        engine.commit_instruction(make_test_instr(2, "MOV", 2));
        let bundles = engine.flush();
        assert_eq!(bundles.len(), 1);
        assert_eq!(bundles[0].instructions.len(), 2);
    }

    #[test]
    fn test_bundle_padding() {
        let mut engine = make_atom_bundling();
        engine.begin_bundle();
        engine.commit_instruction(make_test_instr(1, "ADD", 2));
        // Only 2 bytes used; bundle should be padded to 8.
        let bundles = engine.flush();
        assert_eq!(bundles[0].used_bytes, 8); // padded
        assert!(bundles[0].flags.has_padding);
    }

    #[test]
    fn test_bundle_optimization() {
        let mut bundles = vec![Bundle {
            id: 0,
            offset: 0,
            size: 16,
            instructions: vec![make_test_instr(1, "LONG", 8), make_test_instr(2, "SHRT", 2)],
            used_bytes: 10,
            is_locked: false,
            ends_with_branch: false,
            alignment: 16,
            flags: BundleFlags::default(),
        }];
        optimize_bundle_packing(&mut bundles, BundleMicroArch::Goldmont);
        // Short instruction should be placed first for better packing.
        assert_eq!(bundles[0].instructions[0].mnemonic, "SHRT");
    }

    #[test]
    fn test_bundle_utilization() {
        let bundle = Bundle {
            id: 0,
            offset: 0,
            size: 16,
            instructions: vec![],
            used_bytes: 8,
            is_locked: false,
            ends_with_branch: false,
            alignment: 16,
            flags: BundleFlags::default(),
        };
        assert_eq!(bundle_utilization(&bundle), 50.0);
    }

    #[test]
    fn test_find_least_utilized() {
        let bundles = vec![
            Bundle {
                id: 0,
                offset: 0,
                size: 8,
                instructions: vec![],
                used_bytes: 2,
                is_locked: false,
                ends_with_branch: false,
                alignment: 8,
                flags: BundleFlags::default(),
            },
            Bundle {
                id: 1,
                offset: 8,
                size: 8,
                instructions: vec![],
                used_bytes: 7,
                is_locked: false,
                ends_with_branch: false,
                alignment: 8,
                flags: BundleFlags::default(),
            },
        ];
        let least = find_least_utilized_bundle(&bundles);
        assert!(least.is_some());
        assert_eq!(least.unwrap().id, 0);
    }

    #[test]
    fn test_instr_classification() {
        let mut instr = make_test_instr(1, "ADD", 2);
        assert_eq!(instr.classify(), InstrClass::Alu);

        instr.mark_branch(false);
        assert_eq!(instr.classify(), InstrClass::ConditionalBranch);

        instr.mark_branch(true);
        assert_eq!(instr.classify(), InstrClass::IndirectBranch);

        instr.mark_call();
        assert_eq!(instr.classify(), InstrClass::Call);

        instr.mark_return();
        assert_eq!(instr.classify(), InstrClass::Return);
    }

    #[test]
    fn test_instr_class_terminates_bundle() {
        assert!(InstrClass::IndirectBranch.terminates_bundle(BundleMicroArch::Bonnell));
        assert!(InstrClass::Call.terminates_bundle(BundleMicroArch::Bonnell));
        assert!(InstrClass::Return.terminates_bundle(BundleMicroArch::Bonnell));
        assert!(!InstrClass::Alu.terminates_bundle(BundleMicroArch::Bonnell));
    }

    #[test]
    fn test_instr_class_forces_new_bundle() {
        assert!(InstrClass::Locked.forces_new_bundle(BundleMicroArch::Bonnell));
        assert!(InstrClass::Complex.forces_new_bundle(BundleMicroArch::Bonnell));
        assert!(!InstrClass::Alu.forces_new_bundle(BundleMicroArch::Bonnell));
    }

    #[test]
    fn test_branch_at_bundle_end_enforcement() {
        let mut bundle = Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![
                make_test_instr(1, "ADD", 2),
                make_test_branch(2, "JE", 2, false),
                make_test_instr(3, "MOV", 2),
            ],
            used_bytes: 6,
            is_locked: false,
            ends_with_branch: false,
            alignment: 8,
            flags: BundleFlags::default(),
        };

        enforce_branch_at_bundle_end(&mut bundle, BundleMicroArch::Bonnell);

        // Branch should now be last.
        let last = bundle.instructions.last().unwrap();
        assert!(last.is_branch);
        assert!(bundle.ends_with_branch);
    }

    #[test]
    fn test_check_branch_at_bundle_end() {
        let valid_bundle = Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![
                make_test_instr(1, "ADD", 2),
                make_test_branch(2, "JE", 2, false),
            ],
            used_bytes: 4,
            is_locked: false,
            ends_with_branch: true,
            alignment: 8,
            flags: BundleFlags::default(),
        };
        assert!(check_branch_at_bundle_end(&valid_bundle, BundleMicroArch::Bonnell).is_ok());

        let invalid_bundle = Bundle {
            id: 1,
            offset: 8,
            size: 8,
            instructions: vec![
                make_test_branch(1, "JE", 2, false),
                make_test_instr(2, "MOV", 2),
            ],
            used_bytes: 4,
            is_locked: false,
            ends_with_branch: false,
            alignment: 8,
            flags: BundleFlags::default(),
        };
        assert!(check_branch_at_bundle_end(&invalid_bundle, BundleMicroArch::Bonnell).is_err());
    }

    #[test]
    fn test_bundle_display() {
        let bundle = Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![make_test_instr(1, "ADD", 2)],
            used_bytes: 2,
            is_locked: true,
            ends_with_branch: false,
            alignment: 8,
            flags: BundleFlags {
                function_entry: true,
                ..BundleFlags::default()
            },
        };
        let display = format!("{}", bundle);
        assert!(display.contains("Bundle #0"));
        assert!(display.contains("ADD"));
        assert!(display.contains("LOCKED"));
        assert!(display.contains("FUNC_ENTRY"));
    }

    #[test]
    fn test_dump_bundles() {
        let mut engine = make_goldmont_bundling();
        engine.bundle_instructions(&[make_test_instr(1, "ADD", 3), make_test_instr(2, "MOV", 4)]);
        let dump = engine.dump_bundles();
        assert!(dump.contains("Goldmont"));
        assert!(dump.contains("ADD"));
        assert!(dump.contains("MOV"));
    }

    #[test]
    fn test_microarch_bundle_sizes() {
        assert_eq!(BundleMicroArch::Bonnell.bundle_size(), 8);
        assert_eq!(BundleMicroArch::Silvermont.bundle_size(), 8);
        assert_eq!(BundleMicroArch::Goldmont.bundle_size(), 16);
        assert_eq!(BundleMicroArch::Tremont.bundle_size(), 16);
        assert_eq!(BundleMicroArch::OutOfOrder.bundle_size(), 16);
    }

    #[test]
    fn test_microarch_decode_widths() {
        assert_eq!(BundleMicroArch::Bonnell.decode_width(), 2);
        assert_eq!(BundleMicroArch::Goldmont.decode_width(), 3);
        assert_eq!(BundleMicroArch::Tremont.decode_width(), 6);
    }

    #[test]
    fn test_microarch_branch_at_end_requirement() {
        assert!(BundleMicroArch::Bonnell.requires_branch_at_bundle_end());
        assert!(BundleMicroArch::Silvermont.requires_branch_at_bundle_end());
        assert!(!BundleMicroArch::Goldmont.requires_branch_at_bundle_end());
        assert!(!BundleMicroArch::OutOfOrder.requires_branch_at_bundle_end());
    }

    #[test]
    fn test_microarch_bundle_locking_support() {
        assert!(BundleMicroArch::Silvermont.supports_bundle_locking());
        assert!(BundleMicroArch::Goldmont.supports_bundle_locking());
        assert!(!BundleMicroArch::OutOfOrder.supports_bundle_locking());
    }

    #[test]
    fn test_config_override_bundle_size() {
        let config = BundleConfig {
            bundle_size_override: 32,
            ..BundleConfig::default()
        };
        let engine = X86InstrBundling::with_config(BundleMicroArch::Bonnell, true, config);
        assert_eq!(engine.bundle_size(), 32);
    }

    #[test]
    fn test_config_max_instrs_override() {
        let config = BundleConfig {
            max_instrs_per_bundle: 4,
            ..BundleConfig::default()
        };
        let engine = X86InstrBundling::with_config(BundleMicroArch::Bonnell, true, config);
        assert_eq!(engine.max_instrs_per_bundle(), 4);
    }

    #[test]
    fn test_large_instruction_bundling() {
        // An 8-byte instruction on Atom should occupy a whole bundle.
        let mut engine = make_atom_bundling();
        let results = engine.bundle_instructions(&[
            make_test_instr(1, "VERYLONG", 8),
            make_test_instr(2, "ADD", 2),
        ]);
        // The 8-byte instruction should be alone in its bundle.
        assert!(results.len() >= 2);
        assert_eq!(results[0].instructions[0].size, 8);
    }

    #[test]
    fn test_atom_rules_apply() {
        let instrs = vec![
            make_test_instr(1, "ADD", 2),
            make_test_instr(2, "MOV", 2),
            make_test_instr(3, "SUB", 2),
        ];
        let bundles = atom_rules::apply_atom_constraints(&instrs, true);
        assert!(bundles.len() >= 2);
        // Max 2 per bundle.
        for b in &bundles {
            let non_nop_count = b.instructions.iter().filter(|i| !i.is_nop).count();
            assert!(non_nop_count <= 2);
        }
    }

    #[test]
    fn test_silvermont_rules_apply() {
        let instrs = vec![make_test_instr(1, "ADD", 2), make_test_instr(2, "MOV", 2)];
        let bundles = silvermont_rules::apply_silvermont_constraints(&instrs, true);
        assert_eq!(bundles.len(), 1);
        assert_eq!(bundles[0].size, 8);
    }

    #[test]
    fn test_goldmont_rules_apply() {
        let instrs = vec![
            make_test_instr(1, "ADD", 3),
            make_test_instr(2, "MOV", 4),
            make_test_instr(3, "SUB", 2),
        ];
        let bundles = goldmont_rules::apply_goldmont_constraints(&instrs, true);
        assert!(!bundles.is_empty());
        assert_eq!(bundles[0].size, 16);
    }

    #[test]
    fn test_tremont_rules_apply() {
        let instrs = vec![
            make_test_instr(1, "ADD", 3),
            make_test_instr(2, "MOV", 4),
            make_test_instr(3, "SUB", 2),
        ];
        let bundles = tremont_rules::apply_tremont_constraints(&instrs, true);
        assert!(!bundles.is_empty());
        assert_eq!(bundles[0].size, 16);
    }

    #[test]
    fn test_bundle_result_equality() {
        assert_eq!(BundleResult::Added, BundleResult::Added);
        assert_ne!(BundleResult::Added, BundleResult::NewBundleNeeded);
    }

    #[test]
    fn test_boundary_crossing_info_display() {
        let info = diagnose_boundary_crossing(7, 3, 8);
        assert!(info.crosses_boundary);
        assert_eq!(info.bytes_in_current_bundle, 1);
        assert_eq!(info.bytes_in_next_bundle, 2);
    }

    #[test]
    fn test_empty_bundle() {
        let bundle = Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![],
            used_bytes: 0,
            is_locked: false,
            ends_with_branch: false,
            alignment: 8,
            flags: BundleFlags::default(),
        };
        // Empty bundle should be valid on all microarchs.
        assert!(atom_rules::validate_atom_bundle(&bundle).is_ok());
    }

    #[test]
    fn test_lock_granularity() {
        let lock = BundleLock::bundle_lock();
        assert!(lock.locked);
        assert_eq!(lock.granularity, LockGranularity::Bundle);

        let ilock = BundleLock::instruction_lock();
        assert_eq!(ilock.granularity, LockGranularity::Instruction);

        let unlocked = BundleLock::unlocked();
        assert!(!unlocked.locked);
    }

    #[test]
    fn test_macro_fusion_cross_bundle() {
        let config = BundleConfig {
            allow_macro_fusion_cross_bundle: false,
            ..BundleConfig::default()
        };
        let mut engine = X86InstrBundling::with_config(BundleMicroArch::Goldmont, true, config);

        let mut fused1 = make_test_instr(1, "CMP", 3);
        fused1.is_macro_fused = true;
        fused1.macro_fuse_partner = Some(2);

        let mut fused2 = make_test_instr(2, "JE", 2);
        fused2.is_macro_fused = true;
        fused2.macro_fuse_partner = Some(1);

        engine.bundle_instructions(&[fused1, fused2]);
        // Should keep fusion pair together (if possible).
        assert!(engine.bundles.len() >= 1);
    }

    #[test]
    fn test_stats_after_bundling() {
        let mut engine = make_atom_bundling();
        engine.bundle_instructions(&[make_test_instr(1, "ADD", 2), make_test_instr(2, "MOV", 2)]);
        assert!(engine.stats.total_bundles >= 1);
        assert_eq!(engine.stats.total_instructions, 2);
    }

    #[test]
    fn test_adaptive_bundling() {
        let config = BundleConfig {
            adaptive_bundling: true,
            bundle_size_override: 0,
            max_instrs_per_bundle: 0,
            ..BundleConfig::default()
        };
        let mut engine = X86InstrBundling::with_config(BundleMicroArch::Tremont, true, config);
        engine.bundle_instructions(&[
            make_test_instr(1, "ADD", 3),
            make_test_instr(2, "MOV", 4),
            make_test_instr(3, "SUB", 2),
        ]);
        // Adaptive bundling should still produce valid bundles.
        for b in &engine.bundles {
            assert!(b.instructions.len() <= 3);
        }
    }

    #[test]
    fn test_bundle_stats_accumulation() {
        let mut engine = make_atom_bundling();
        engine.bundle_instructions(&[
            make_test_instr(1, "ADD", 2),
            make_test_instr(2, "MOV", 2),
            make_test_instr(3, "SUB", 2),
            make_test_instr(4, "XOR", 2),
        ]);
        assert_eq!(engine.stats.total_instructions, 4);
        assert!(engine.stats.total_bundles >= 2);
    }

    #[test]
    fn test_bundle_ends_with_call() {
        let mut engine = X86InstrBundling::new(BundleMicroArch::Goldmont, true);
        let call = make_test_call(1, 5);
        engine.bundle_instructions(&[call]);
        let bundle = &engine.bundles[0];
        assert!(bundle.ends_with_branch);
    }

    #[test]
    fn test_bundle_ends_with_return() {
        let mut engine = X86InstrBundling::new(BundleMicroArch::Goldmont, true);
        let ret = make_test_return(1, 1);
        engine.bundle_instructions(&[ret]);
        let bundle = &engine.bundles[0];
        assert!(bundle.ends_with_branch);
    }

    #[test]
    fn test_bundle_rejects_oversized_instr() {
        let mut engine = make_atom_bundling();
        let instrs = vec![make_test_instr(1, "BIG", 10)];
        let bundles = engine.bundle_instructions(&instrs);
        // A 10-byte instruction doesn't fit in an 8-byte bundle,
        // so it should be placed in its own (possibly oversized) bundle.
        assert!(!bundles.is_empty());
    }

    #[test]
    fn test_bundle_fusion_pair_handling() {
        let config = BundleConfig {
            allow_macro_fusion_cross_bundle: false,
            ..BundleConfig::default()
        };
        let mut engine = X86InstrBundling::with_config(BundleMicroArch::Goldmont, true, config);

        let mut c = make_test_instr(1, "CMP", 3);
        c.is_macro_fused = true;
        c.macro_fuse_partner = Some(2);

        let mut j = make_test_branch(2, "JE", 2, false);
        j.is_macro_fused = true;
        j.macro_fuse_partner = Some(1);

        engine.bundle_instructions(&[c, j]);
        assert!(!engine.bundles.is_empty());
    }

    #[test]
    fn test_boundary_crossing_exact_fit() {
        // Instruction at offset 0, size 8, bundle size 8 — exact fit.
        let crossing = diagnose_boundary_crossing(0, 8, 8);
        assert!(!crossing.crosses_boundary);
        assert_eq!(crossing.bytes_in_current_bundle, 8);
        assert_eq!(crossing.bytes_in_next_bundle, 0);
    }

    #[test]
    fn test_boundary_crossing_just_over() {
        // Instruction at offset 5, size 4, bundle size 8 — overflows by 1.
        let crossing = diagnose_boundary_crossing(5, 4, 8);
        assert!(crossing.crosses_boundary);
        assert_eq!(crossing.bytes_in_current_bundle, 3);
        assert_eq!(crossing.bytes_in_next_bundle, 1);
    }

    #[test]
    fn test_bundle_alignment_large_offset() {
        // Offset 100 with 16-byte bundles.
        assert_eq!(compute_bundle_alignment(100, 16), 12);
        assert_eq!(compute_bundle_alignment(112, 16), 0);
    }

    #[test]
    fn test_lock_bundles_silvermont() {
        let mut bundles = vec![
            Bundle {
                id: 0,
                offset: 0,
                size: 8,
                instructions: vec![make_test_instr(1, "ADD", 2)],
                used_bytes: 2,
                is_locked: false,
                ends_with_branch: false,
                alignment: 8,
                flags: BundleFlags::default(),
            },
            Bundle {
                id: 1,
                offset: 8,
                size: 8,
                instructions: vec![make_test_instr(2, "MOV", 2)],
                used_bytes: 2,
                is_locked: false,
                ends_with_branch: false,
                alignment: 8,
                flags: BundleFlags::default(),
            },
        ];
        lock_bundles(&mut bundles, BundleMicroArch::Silvermont);
        assert!(bundles.iter().all(|b| b.is_locked));
    }

    #[test]
    fn test_checkpoint_registers_count() {
        let regs_64 = checkpoint_registers_for_bundle(true);
        // 8 GPRs + 8 R8-R15 + 16 XMMs + RFLAGS = 33
        assert_eq!(regs_64.len(), 33);

        let regs_32 = checkpoint_registers_for_bundle(false);
        // 8 GPRs + 8 XMMs + RFLAGS = 17
        assert_eq!(regs_32.len(), 17);
    }

    #[test]
    fn test_instr_class_discriminants() {
        // Verify each variant exists and is distinct.
        let classes = [
            InstrClass::Alu,
            InstrClass::Memory,
            InstrClass::ConditionalBranch,
            InstrClass::UnconditionalBranch,
            InstrClass::IndirectBranch,
            InstrClass::Call,
            InstrClass::Return,
            InstrClass::Locked,
            InstrClass::Nop,
            InstrClass::Complex,
        ];
        for (i, a) in classes.iter().enumerate() {
            for (j, b) in classes.iter().enumerate() {
                if i == j {
                    assert_eq!(a, b);
                } else {
                    assert_ne!(a, b);
                }
            }
        }
    }

    #[test]
    fn test_microarch_max_prefix_bytes() {
        assert_eq!(BundleMicroArch::Bonnell.max_prefix_bytes(), 2);
        assert_eq!(BundleMicroArch::Silvermont.max_prefix_bytes(), 3);
        assert_eq!(BundleMicroArch::Goldmont.max_prefix_bytes(), 4);
    }

    #[test]
    fn test_optimize_bundle_packing_idempotent() {
        let mut bundles = vec![Bundle {
            id: 0,
            offset: 0,
            size: 16,
            instructions: vec![make_test_instr(1, "B", 2), make_test_instr(2, "A", 10)],
            used_bytes: 12,
            is_locked: false,
            ends_with_branch: false,
            alignment: 16,
            flags: BundleFlags::default(),
        }];
        optimize_bundle_packing(&mut bundles, BundleMicroArch::Goldmont);
        // Should still be valid.
        assert!(goldmont_rules::validate_goldmont_bundle(&bundles[0]).is_ok());
    }

    #[test]
    fn test_find_least_utilized_empty_list() {
        let bundles: Vec<Bundle> = vec![];
        assert!(find_least_utilized_bundle(&bundles).is_none());
    }

    #[test]
    fn test_bundle_utilization_full() {
        let bundle = Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![],
            used_bytes: 8,
            is_locked: false,
            ends_with_branch: false,
            alignment: 8,
            flags: BundleFlags::default(),
        };
        assert_eq!(bundle_utilization(&bundle), 100.0);
    }

    #[test]
    fn test_bundle_utilization_empty() {
        let bundle = Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![],
            used_bytes: 0,
            is_locked: false,
            ends_with_branch: false,
            alignment: 8,
            flags: BundleFlags::default(),
        };
        assert_eq!(bundle_utilization(&bundle), 0.0);
    }

    #[test]
    fn test_bundle_utilization_zero_size() {
        let bundle = Bundle {
            id: 0,
            offset: 0,
            size: 0,
            instructions: vec![],
            used_bytes: 0,
            is_locked: false,
            ends_with_branch: false,
            alignment: 1,
            flags: BundleFlags::default(),
        };
        assert_eq!(bundle_utilization(&bundle), 100.0);
    }

    #[test]
    fn test_enforce_branch_at_bundle_end_empty() {
        let mut bundle = Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![],
            used_bytes: 0,
            is_locked: false,
            ends_with_branch: false,
            alignment: 8,
            flags: BundleFlags::default(),
        };
        enforce_branch_at_bundle_end(&mut bundle, BundleMicroArch::Bonnell);
        // Should not panic on empty bundle.
        assert!(bundle.instructions.is_empty());
    }

    #[test]
    fn test_enforce_branch_at_bundle_end_branch_already_last() {
        let mut bundle = Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![
                make_test_instr(1, "ADD", 2),
                make_test_branch(2, "JE", 2, false),
            ],
            used_bytes: 4,
            is_locked: false,
            ends_with_branch: true,
            alignment: 8,
            flags: BundleFlags::default(),
        };
        enforce_branch_at_bundle_end(&mut bundle, BundleMicroArch::Bonnell);
        // Branch was already last; should remain last.
        assert!(bundle.instructions.last().unwrap().is_branch);
    }

    #[test]
    fn test_dump_bundles_output_format() {
        let mut engine = make_atom_bundling();
        engine.bundle_instructions(&[make_test_instr(1, "PUSH", 1), make_test_instr(2, "MOV", 3)]);
        let dump = engine.dump_bundles();
        assert!(dump.contains("Bonnell"));
        assert!(dump.contains("Bundle size:"));
        assert!(dump.contains("Total bundles:"));
        assert!(dump.contains("Average"));
    }

    #[test]
    fn test_make_test_instr_defaults() {
        let instr = make_test_instr(42, "TEST", 4);
        assert_eq!(instr.id, 42);
        assert_eq!(instr.mnemonic, "TEST");
        assert_eq!(instr.size, 4);
        assert!(!instr.is_branch);
        assert!(!instr.is_call);
        assert!(!instr.is_return);
        assert!(!instr.is_nop);
        assert_eq!(instr.classify(), InstrClass::Alu);
    }

    #[test]
    fn test_config_default_values() {
        let config = BundleConfig::default();
        assert_eq!(config.bundle_size_override, 0);
        assert_eq!(config.max_instrs_per_bundle, 0);
        assert!(config.enforce_alignment);
        assert!(config.pad_with_nops);
        assert!(!config.lock_bundles);
        assert!(!config.branch_at_bundle_end);
        assert!(!config.adaptive_bundling);
        assert!(!config.allow_macro_fusion_cross_bundle);
    }

    #[test]
    fn test_engine_stats_updated_after_flush() {
        let mut engine = make_goldmont_bundling();
        engine.bundle_instructions(&[make_test_instr(1, "ADD", 3), make_test_instr(2, "MOV", 4)]);
        engine.flush();
        // After flush, stats should be updated.
        assert!(engine.stats.total_bundles > 0);
        assert!(engine.stats.avg_instrs_per_bundle > 0.0);
        assert!(engine.stats.avg_utilization > 0.0);
    }

    #[test]
    fn test_locked_bundle_count() {
        let config = BundleConfig {
            lock_bundles: true,
            ..BundleConfig::default()
        };
        let mut engine = X86InstrBundling::with_config(BundleMicroArch::Tremont, true, config);
        engine.bundle_instructions(&[make_test_instr(1, "ADD", 3), make_test_instr(2, "MOV", 4)]);
        assert!(engine.stats.locked_bundles > 0);
    }

    #[test]
    fn test_bundle_instruction_ordering() {
        // Instructions should appear in order within a bundle.
        let mut engine = make_atom_bundling();
        engine.bundle_instructions(&[
            make_test_instr(1, "FIRST", 2),
            make_test_instr(2, "SECOND", 2),
        ]);
        let bundle = &engine.bundles[0];
        assert_eq!(bundle.instructions[0].mnemonic, "FIRST");
        assert_eq!(bundle.instructions[1].mnemonic, "SECOND");
    }

    #[test]
    fn test_all_microarch_bundle_sizes_consistent() {
        let microarchs = [
            BundleMicroArch::Bonnell,
            BundleMicroArch::Saltwell,
            BundleMicroArch::Silvermont,
            BundleMicroArch::Airmont,
            BundleMicroArch::Goldmont,
            BundleMicroArch::GoldmontPlus,
            BundleMicroArch::Tremont,
            BundleMicroArch::Gracemont,
            BundleMicroArch::Crestmont,
            BundleMicroArch::Skymont,
            BundleMicroArch::Darkmont,
            BundleMicroArch::OutOfOrder,
        ];
        for ma in &microarchs {
            let size = ma.bundle_size();
            assert!(
                size == 8 || size == 16,
                "{:?} has unexpected bundle size {}",
                ma,
                size
            );
        }
    }

    #[test]
    fn test_lock_granularity_variants() {
        assert_eq!(LockGranularity::Bundle, LockGranularity::Bundle);
        assert_ne!(LockGranularity::Bundle, LockGranularity::None);
        assert_ne!(LockGranularity::Instruction, LockGranularity::None);
        assert_ne!(LockGranularity::Pair, LockGranularity::None);
    }

    #[test]
    fn test_bundle_flags_default() {
        let flags = BundleFlags::default();
        assert!(!flags.function_entry);
        assert!(!flags.loop_header);
        assert!(!flags.branch_target);
        assert!(!flags.has_padding);
        assert!(!flags.fetch_aligned);
        assert!(!flags.is_full);
    }

    #[test]
    fn test_bundle_display_includes_flags() {
        let mut bundle = Bundle {
            id: 5,
            offset: 0x40,
            size: 8,
            instructions: vec![make_test_branch(1, "JMP", 2, true)],
            used_bytes: 2,
            is_locked: true,
            ends_with_branch: true,
            alignment: 8,
            flags: BundleFlags {
                loop_header: true,
                branch_target: true,
                has_padding: true,
                ..BundleFlags::default()
            },
        };
        let display = format!("{}", bundle);
        assert!(display.contains("Bundle #5"));
        assert!(display.contains("0x0040"));
        assert!(display.contains("LOCKED"));
        assert!(display.contains("ENDS_BR"));
        assert!(display.contains("PADDED"));
        assert!(display.contains("LOOP_HDR"));
    }

    #[test]
    fn test_instr_class_terminates_bundle_all_microarchs() {
        // Returns and indirect branches should always terminate.
        for ma in &[
            BundleMicroArch::Bonnell,
            BundleMicroArch::Silvermont,
            BundleMicroArch::Goldmont,
            BundleMicroArch::Tremont,
            BundleMicroArch::OutOfOrder,
        ] {
            assert!(InstrClass::Return.terminates_bundle(*ma));
            assert!(InstrClass::IndirectBranch.terminates_bundle(*ma));
            assert!(InstrClass::Call.terminates_bundle(*ma));
        }
    }

    #[test]
    fn test_try_add_to_full_bundle() {
        let mut engine = make_atom_bundling();
        engine.begin_bundle();
        // Fill the bundle with a 6-byte instruction.
        engine.commit_instruction(make_test_instr(1, "BIG", 6));
        // Try to add a 3-byte instruction — should need new bundle.
        let result = engine.try_add_instruction(&make_test_instr(2, "ADD", 3));
        assert_eq!(result, BundleResult::NewBundleNeeded);
    }

    #[test]
    fn test_begin_bundle_alignment() {
        let mut engine = make_atom_bundling();
        engine.code_offset = 5;
        engine.begin_bundle();
        // Should have aligned to 8.
        assert_eq!(engine.code_offset, 8);
    }

    #[test]
    fn test_silvermont_allows_larger_instructions() {
        let bundle = Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![make_test_instr(1, "LONG", 8)],
            used_bytes: 8,
            is_locked: false,
            ends_with_branch: false,
            alignment: 8,
            flags: BundleFlags::default(),
        };
        assert!(silvermont_rules::validate_silvermont_bundle(&bundle).is_ok());
    }

    #[test]
    fn test_silvermont_rejects_oversize() {
        let bundle = Bundle {
            id: 0,
            offset: 0,
            size: 8,
            instructions: vec![make_test_instr(1, "TOOLONG", 12)],
            used_bytes: 12,
            is_locked: false,
            ends_with_branch: false,
            alignment: 8,
            flags: BundleFlags::default(),
        };
        assert!(silvermont_rules::validate_silvermont_bundle(&bundle).is_err());
    }

    #[test]
    fn test_boundary_crossing_info_nop_sled_zero_when_no_crossing() {
        let info = diagnose_boundary_crossing(0, 4, 8);
        assert!(!info.needs_nop_sled());
        assert_eq!(info.nop_sled_size(), 0);
    }

    #[test]
    fn test_bundle_microarch_comparison() {
        // Verify each microarch has different characteristics.
        assert_ne!(
            BundleMicroArch::Bonnell.bundle_size(),
            BundleMicroArch::Goldmont.bundle_size()
        );
        assert_ne!(
            BundleMicroArch::Silvermont.decode_width(),
            BundleMicroArch::Tremont.decode_width()
        );
    }
}