Data Type Optimization Based on CPU Architecture

Data Type Optimization Based on CPU Architecture (C++)

---

Prerequisites

---

1. Why CPU Architecture Matters

Modern CPUs are optimized for specific data widths and alignment.

Choosing the right data type improves both performance and memory efficiency.

Using mismatched data types can cause:

• Slower execution
• Extra instructions
• Cache inefficiency

Use data types that match the native word size of the CPU

Architecture	Native Size
32-bit CPU	32 bits (4 bytes)
64-bit CPU	64 bits (8 bytes)

Most modern systems = 64-bit

2. Optimal Data Types

✔ On 64-bit CPU

int64_t a;   // optimal
size_t b;    // optimal

Matches CPU register size

❌ Suboptimal

int8_t a;

May require:

• Extra masking
• Additional instructions

CPUs operate on registers:

• 64-bit CPU → 64-bit registers
• Smaller types often promoted internally

Example

uint8_t a, b;
uint8_t c = a + b;

Internally:

• Promoted to int
• Then truncated back

2-1. Alignment Consideration

✔ Proper alignment

int64_t a;

Aligned → fast access

❌ Misaligned

#pragma pack(1)
struct A 
{
    char c;
    int64_t x;
};

Misaligned access → slower

2-2. SIMD & Vectorization

SIMD prefers aligned, consistent data types

float arr[8];

Works well with AVX (256-bit)

❌ Mixed / irregular layout

struct A 
{
    char c;
    float x;
};

Hard to vectorize

When Smaller Types Are Better

✔ Large arrays (memory-bound)
✔ Cache-sensitive workloads

• uint8_t for image processing
• uint16_t for compressed data

✔ DO

• Use size_t for indexing
• Use native types (int64_t on 64-bit systems)
• Align data properly
• Profile performance

❌ DON’T

• Use small types blindly
• Ignore alignment
• Mix data types unnecessarily

Factor	Small Type	Native Type
Memory	Efficient	Larger
Cache	Better	Worse
CPU ops	Slower	Faster
SIMD	Harder	Easier