{
  lib,
  stdenv,
  fetchFromGitHub,
  gnumake,
  rocmPackages,
  autoPatchelfHook,
}:

stdenv.mkDerivation {
  pname = "ds4";
  version = "0-unstable-2026-06-17";

  src = fetchFromGitHub {
    owner = "antirez";
    repo = "ds4";
    rev = "80ebbc396aee40eedc1d829222f3362d10fa4c6c";
    hash = "sha256-Ieuc72GHZs20ModQfnvI5Me31n4Pj+WFYtsuqaKJceo=";
  };

  nativeBuildInputs = [
    gnumake
    rocmPackages.hipcc
    autoPatchelfHook
  ];

  buildInputs = [
    rocmPackages.clr
    rocmPackages.hipblas
    rocmPackages.hipblas-common
    rocmPackages.hipblaslt
    rocmPackages.hipcub
    rocmPackages.rocblas
    rocmPackages.rocprim
    rocmPackages.rocwmma
  ];

  # STRIXHALO.md: rocwmma/internal/ headers may be missing in distro packages.
  # nixpkgs builds rocwmma from source so all headers (including internal/) are present.
  # ROCM_ARCH defaults to gfx1151 (Strix Halo / Radeon 8060S) in the upstream Makefile.
  buildPhase = ''
    runHook preBuild

    make strix-halo -j"$NIX_BUILD_CORES" \
      HIPCC="${rocmPackages.hipcc}/bin/hipcc" \
      ROCM_CFLAGS="-O3 -ffast-math -g -fno-finite-math-only -pthread -D__HIP_PLATFORM_AMD__ -Wno-unused-command-line-argument --offload-arch=gfx1151 \
        -I${rocmPackages.clr}/include \
        -I${rocmPackages.hipblas}/include \
        -I${rocmPackages.hipblas-common}/include \
        -I${rocmPackages.hipblaslt}/include \
        -I${rocmPackages.hipcub}/include \
        -I${rocmPackages.rocblas}/include \
        -I${rocmPackages.rocprim}/include \
        -I${rocmPackages.rocwmma}/include" \
      ROCM_LDLIBS="-lm -pthread \
        -L${rocmPackages.hipblas}/lib -lhipblas \
        -L${rocmPackages.hipblaslt}/lib -lhipblaslt \
        -L${rocmPackages.rocblas}/lib -lrocblas \
        -L${rocmPackages.clr}/lib -lamdhip64"

    runHook postBuild
  '';

  installPhase = ''
    runHook preInstall

    mkdir -p $out/bin
    cp ds4 ds4-server ds4-bench ds4-eval ds4-agent $out/bin/

    runHook postInstall
  '';

  passthru = {
    category = "AI Inference";
    updateScript = [
      "nix-update"
      "--flake"
      "--version=branch=main"
      ".#ds4"
    ];
  };

  meta = {
    description = "DeepSeek 4 Flash and PRO local inference engine for ROCm (Strix Halo)";
    longDescription = ''
      DS4 is a self-contained local inference engine specifically built for
      DeepSeek V4 Flash and PRO. It bundles a CLI, HTTP server, and native
      coding agent. This package is built with the ROCm backend for AMD
      Strix Halo GPUs (gfx1151, Radeon 8060S).

      Runtime requirements (from STRIXHALO.md):
      - User must be in 'render' and 'video' groups for /dev/kfd and DRM access
      - GPU-visible memory must be increased via kernel parameters:
        amd_iommu=off amdgpu.gttsize=126976 ttm.pages_limit=32505856 ttm.page_pool_size=32505856
    '';
    homepage = "https://github.com/antirez/ds4";
    changelog = "https://github.com/antirez/ds4/commits/main";
    license = lib.licenses.mit;
    platforms = [ "x86_64-linux" ];
    mainProgram = "ds4";
  };
}