GCN Assembler

A specialized assembler library for AMD Graphics Core Next (GCN) architecture, enabling low-level GPU kernel generation.

🏛️ Architecture

graph TB
    subgraph "GCN Compilation Flow"
        A[GCN Instructions] --> B[GCN Program Builder]
        B --> C[Instruction Formatter]
        C --> D[ELF Wrapper]
        D --> E[GPU Binary (.bin / .elf)]
        
        subgraph "Instruction Types"
            F[SOP - Scalar Ops]
            G[VOP - Vector Ops]
            H[SMEM - Scalar Memory]
            I[VMEM - Vector Memory]
        end
        
        B --> F
        B --> G
        B --> H
        B --> I
    end

🚀 Features

Core Capabilities

• Architecture Support: Comprehensive support for GCN ISA (Instruction Set Architecture) covering scalar and vector operations.
• Kernel Orchestration: Simplifies the construction of GPU kernels with automatic text section and global symbol generation.
• Memory Mapping: Handles complex GCN memory operand formats, including buffer descriptors and displacement offsets.

Advanced Features

• ELF Encapsulation: Automatically wraps generated GCN machine code into an ELF container compatible with AMD's GPU loaders.
• Instruction Formatting: Provides a human-readable formatter for debugging and verifying instruction sequences before binary emission.
• Error Tracking: Integrated with Gaia's diagnostic system to catch invalid operand combinations or illegal instruction pairings.

💻 Usage

Generating a GCN Kernel

The following example shows how to create a simple GCN program and write it to a binary file.

use gcn_assembler::{GcnProgram, GcnKernel, GcnInstruction, GcnWriter};
use std::fs;

fn main() {
    // 1. Define instructions (e.g., s_endpgm)
    let instructions = vec![
        GcnInstruction::new("s_endpgm"),
    ];

    // 2. Build kernel and program
    let kernel = GcnKernel {
        name: "my_kernel".to_string(),
        instructions,
    };
    let program = GcnProgram { kernels: vec![kernel] };

    // 3. Write to binary
    let writer = GcnWriter::new();
    let binary = writer.write(&program).expect("Failed to generate GCN binary");
    
    fs::write("kernel.bin", binary).unwrap();
}

🛠️ Support Status

Legend: ✅ Supported, 🚧 In Progress, ❌ Not Supported

🔗 Relations

• gaia-types: Uses basic type definitions and LEB128 utilities for metadata encoding.
• gaia-assembler: Acts as a GPU-specific backend for offloading Gaia IR compute tasks to AMD hardware.