A Modular Program Analysis Tool Framework for Emerging Accelerators.

AccelProf is a modular program analysis framework for accelerator workloads spanning NVIDIA CUDA, AMD ROCm, and modern deep-learning systems. It abstracts over heterogeneous profiling APIs and deep-learning frameworks, providing a unified interface for capturing and analyzing runtime events at multiple levels. Its extensible architecture enables researchers and practitioners to rapidly prototype custom analysis tools with minimal overhead.

# Download
git clone --recursive https://github.com/AccelProf/AccelProf.git
git submodule update --init --recursive

# Build and install
#   DEBUG=1: enable debug symbols (-g)
#   OPT_LVL=0|1|2|3: compiler optimization level
#   ENABLE_CS=1: enable Compute Sanitizer backend
#   ENABLE_NVBIT=1: enable NVBit backend
#   ENABLE_TORCH=1: enable PyTorch profiling
#   ENABLE_ROCM=1: enable ROCm backend
make \
    DEBUG=0 \
    OPT_LVL=0 \
    ENABLE_CS=0 \
    ENABLE_NVBIT=0 \
    ENABLE_TORCH=0 \
    ENABLE_ROCM=0


# Set env
export ACCEL_PROF_HOME=$(pwd)
export PATH=${ACCEL_PROF_HOME}/bin:${PATH}

Analyze an accelerator application:

accelprof -v -t app_analysis <executable> [args...]

Full user and developer documentation: 👉 https://accelprofdocs.readthedocs.io

• [CGO’26] PASTA: A Modular Program Analysis Tool Framework for Accelerators.
  Mao Lin, Hyeran Jeon, and Keren Zhou.
  Proceedings of the 23rd ACM/IEEE International Symposium on Code Generation and Optimization (CGO 2026).
• [ICCD’26] AutoUVM: Automated Prefetching Framework for LLMs under UVM Oversubscription.
  Mao Lin, Hui Feng, Xianzhong Ding, Guilherme Cox, Qian Wang, Hyeran Jeon.
  Proceedings of the 44th IEEE International Conference on Computer Design (ICCD 2026).

Released under the MIT License. See LICENSE for details.