A command-line tool for analyzing SIMD instruction usage in compiled binaries. Inspects x86-64 binaries (ELF, Mach-O, PE) and provides detailed statistics on SIMD operations by ISA extension.

• SIMD Detection: Automatically identifies and counts SIMD instructions in binary files
• ISA Classification: Categorizes instructions by extension (SSE, SSE2, SSE4, AVX, AVX2, etc.)
• Multiple Output Formats: Support for JSON and YAML output
• Detailed Breakdown: Optional per-ISA instruction analysis with occurrence counts
• Cross-Platform: Works with ELF, Mach-O, and PE binary formats

• Linux system with objdump installed
• Python 3.6+
• Optional: PyYAML for YAML output format

# Clone or download simdscan.py
chmod +x simdscan.py

./simdscan.py <binary_file>

• -f, --format {json,yaml}: Output format (default: json)
• --show-insts: Include detailed per-ISA instruction breakdown
• -h, --help: Show help message

Analyze a library file:

./simdscan.py libfaiss.a

Output with instruction details:

./simdscan.py --show-insts libmath.so

YAML format output:

./simdscan.py -f yaml binary_file

{
  "binary": "libfaiss.a",
  "has_simd": true,
  "isa_summary": {
    "SSE": 96629,
    "SSE2": 26378
  },
  "total_simd_insts": 123007
}

{
  "binary": "libfaiss_avx2.a",
  "has_simd": true,
  "isa_summary": {
    "AVX": 58825,
    "SSE4": 4078
  },
  "total_simd_insts": 62903
}

{
  "binary": "example.so",
  "has_simd": true,
  "isa_summary": {
    "SSE": 1024,
    "AVX": 512
  },
  "total_simd_insts": 1536,
  "isa_details": {
    "SSE": {
      "unique_mnemonics": 15,
      "occurrences": {
        "movaps": 256,
        "addps": 128,
        "mulps": 64
      }
    },
    "AVX": {
      "unique_mnemonics": 12,
      "occurrences": {
        "vmovaps": 128,
        "vaddps": 96,
        "vmulps": 48
      }
    }
  }
}

1. Disassembly: Uses objdump to disassemble the target binary
2. Pattern Matching: Identifies SIMD instructions using mnemonic patterns
3. Classification: Categorizes instructions by their corresponding ISA extensions
4. Aggregation: Counts occurrences and provides summary statistics

• SSE (Streaming SIMD Extensions)
• SSE2, SSE3, SSSE3, SSE4.1, SSE4.2
• AVX (Advanced Vector Extensions)
• AVX2, AVX-512
• And other x86-64 SIMD instruction sets

• Performance Analysis: Identify SIMD utilization in optimized code
• Compiler Verification: Verify that auto-vectorization is working
• Library Comparison: Compare SIMD usage between different builds
• Architecture Targeting: Ensure binaries use appropriate instruction sets

• Requires objdump to be available in PATH
• Only supports x86-64 architecture
• Static analysis only (doesn't consider runtime execution frequency)

Apache License 2.0

As you wish!