Merge pull request #260299 from Madouura/pr/rocm-meta

4 files changed, 650 insertions, 152 deletions

diff --git a/pkgs/development/rocm-modules/5/default.nix b/pkgs/development/rocm-modules/5/default.nix
index 1e1aa35fa328c..6e263998dbe6c 100644
--- a/pkgs/development/rocm-modules/5/default.nix
+++ b/pkgs/development/rocm-modules/5/default.nix
@@ -1,9 +1,15 @@
 { callPackage
 , recurseIntoAttrs
+, symlinkJoin
+, fetchFromGitHub
 , cudaPackages
 , python3Packages
 , elfutils
 , boost179
+, opencv
+, ffmpeg_4
+, libjpeg_turbo
+, rapidjson-unstable
 }:
 
 let
@@ -187,7 +193,7 @@ in rec {
   };
 
   rocblas = callPackage ./rocblas {
-    inherit rocmUpdateScript rocm-cmake clr tensile;
+    inherit rocblas rocmUpdateScript rocm-cmake clr tensile;
     inherit (llvm) openmp;
     stdenv = llvm.rocmClangStdenv;
   };
@@ -269,4 +275,254 @@ in rec {
     stdenv = llvm.rocmClangStdenv;
     rocmlir = rocmlir-rock;
   };
+
+  ## GPUOpen-ProfessionalCompute-Libraries ##
+  rpp = callPackage ./rpp {
+    inherit rocmUpdateScript rocm-cmake rocm-docs-core clr half;
+    inherit (llvm) openmp;
+    stdenv = llvm.rocmClangStdenv;
+  };
+
+  rpp-hip = rpp.override {
+    useOpenCL = false;
+    useCPU = false;
+  };
+
+  rpp-opencl = rpp.override {
+    useOpenCL = true;
+    useCPU = false;
+  };
+
+  rpp-cpu = rpp.override {
+    useOpenCL = false;
+    useCPU = true;
+  };
+
+  mivisionx = callPackage ./mivisionx {
+    inherit rocmUpdateScript rocm-cmake rocm-device-libs clr rpp rocblas miopengemm miopen migraphx half rocm-docs-core;
+    inherit (llvm) clang openmp;
+    opencv = opencv.override { enablePython = true; };
+    ffmpeg = ffmpeg_4;
+    rapidjson = rapidjson-unstable;
+    stdenv = llvm.rocmClangStdenv;
+
+    # Unfortunately, rocAL needs a custom libjpeg-turbo until further notice
+    # See: https://github.com/GPUOpen-ProfessionalCompute-Libraries/MIVisionX/issues/1051
+    libjpeg_turbo = libjpeg_turbo.overrideAttrs {
+      version = "2.0.6.1";
+
+      src = fetchFromGitHub {
+        owner = "rrawther";
+        repo = "libjpeg-turbo";
+        rev = "640d7ee1917fcd3b6a5271aa6cf4576bccc7c5fb";
+        sha256 = "sha256-T52whJ7nZi8jerJaZtYInC2YDN0QM+9tUDqiNr6IsNY=";
+      };
+    };
+  };
+
+  mivisionx-hip = mivisionx.override {
+    rpp = rpp-hip;
+    useOpenCL = false;
+    useCPU = false;
+  };
+
+  mivisionx-opencl = mivisionx.override {
+    rpp = rpp-opencl;
+    miopen = miopen-opencl;
+    useOpenCL = true;
+    useCPU = false;
+  };
+
+  mivisionx-cpu = mivisionx.override {
+    rpp = rpp-cpu;
+    useOpenCL = false;
+    useCPU = true;
+  };
+
+  ## Meta ##
+  # Emulate common ROCm meta layout
+  # These are mainly for users. I strongly suggest NOT using these in nixpkgs derivations
+  # Don't put these into `propagatedBuildInputs` unless you want PATH/PYTHONPATH issues!
+  # See: https://rocm.docs.amd.com/en/latest/_images/image.004.png
+  # See: https://rocm.docs.amd.com/en/latest/deploy/linux/os-native/package_manager_integration.html
+  meta = rec {
+    rocm-developer-tools = symlinkJoin {
+      name = "rocm-developer-tools-meta";
+
+      paths = [
+        hsa-amd-aqlprofile-bin
+        rocm-core
+        rocr-debug-agent
+        roctracer
+        rocdbgapi
+        rocprofiler
+        rocgdb
+        rocm-language-runtime
+      ];
+    };
+
+    rocm-ml-sdk = symlinkJoin {
+      name = "rocm-ml-sdk-meta";
+
+      paths = [
+        rocm-core
+        miopen-hip
+        rocm-hip-sdk
+        rocm-ml-libraries
+      ];
+    };
+
+    rocm-ml-libraries = symlinkJoin {
+      name = "rocm-ml-libraries-meta";
+
+      paths = [
+        llvm.clang
+        llvm.mlir
+        llvm.openmp
+        rocm-core
+        miopen-hip
+        rocm-hip-libraries
+      ];
+    };
+
+    rocm-hip-sdk = symlinkJoin {
+      name = "rocm-hip-sdk-meta";
+
+      paths = [
+        rocprim
+        rocalution
+        hipfft
+        rocm-core
+        hipcub
+        hipblas
+        rocrand
+        rocfft
+        rocsparse
+        rccl
+        rocthrust
+        rocblas
+        hipsparse
+        hipfort
+        rocwmma
+        hipsolver
+        rocsolver
+        rocm-hip-libraries
+        rocm-hip-runtime-devel
+      ];
+    };
+
+    rocm-hip-libraries = symlinkJoin {
+      name = "rocm-hip-libraries-meta";
+
+      paths = [
+        rocblas
+        hipfort
+        rocm-core
+        rocsolver
+        rocalution
+        rocrand
+        hipblas
+        rocfft
+        hipfft
+        rccl
+        rocsparse
+        hipsparse
+        hipsolver
+        rocm-hip-runtime
+      ];
+    };
+
+    rocm-openmp-sdk = symlinkJoin {
+      name = "rocm-openmp-sdk-meta";
+
+      paths = [
+        rocm-core
+        llvm.clang
+        llvm.mlir
+        llvm.openmp # openmp-extras-devel (https://github.com/ROCm-Developer-Tools/aomp)
+        rocm-language-runtime
+      ];
+    };
+
+    rocm-opencl-sdk = symlinkJoin {
+      name = "rocm-opencl-sdk-meta";
+
+      paths = [
+        rocm-core
+        rocm-runtime
+        clr
+        clr.icd
+        rocm-thunk
+        rocm-opencl-runtime
+      ];
+    };
+
+    rocm-opencl-runtime = symlinkJoin {
+      name = "rocm-opencl-runtime-meta";
+
+      paths = [
+        rocm-core
+        clr
+        clr.icd
+        rocm-language-runtime
+      ];
+    };
+
+    rocm-hip-runtime-devel = symlinkJoin {
+      name = "rocm-hip-runtime-devel-meta";
+
+      paths = [
+        clr
+        rocm-core
+        hipify
+        rocm-cmake
+        llvm.clang
+        llvm.mlir
+        llvm.openmp
+        rocm-thunk
+        rocm-runtime
+        rocm-hip-runtime
+      ];
+    };
+
+    rocm-hip-runtime = symlinkJoin {
+      name = "rocm-hip-runtime-meta";
+
+      paths = [
+        rocm-core
+        rocminfo
+        clr
+        rocm-language-runtime
+      ];
+    };
+
+    rocm-language-runtime = symlinkJoin {
+      name = "rocm-language-runtime-meta";
+
+      paths = [
+        rocm-runtime
+        rocm-core
+        rocm-comgr
+        llvm.openmp # openmp-extras-runtime (https://github.com/ROCm-Developer-Tools/aomp)
+      ];
+    };
+
+    rocm-all = symlinkJoin {
+      name = "rocm-all-meta";
+
+      paths = [
+        rocm-developer-tools
+        rocm-ml-sdk
+        rocm-ml-libraries
+        rocm-hip-sdk
+        rocm-hip-libraries
+        rocm-openmp-sdk
+        rocm-opencl-sdk
+        rocm-opencl-runtime
+        rocm-hip-runtime-devel
+        rocm-hip-runtime
+        rocm-language-runtime
+      ];
+    };
+  };
 }
diff --git a/pkgs/development/rocm-modules/5/mivisionx/default.nix b/pkgs/development/rocm-modules/5/mivisionx/default.nix
new file mode 100644
index 0000000000000..d6e198eb4c778
--- /dev/null
+++ b/pkgs/development/rocm-modules/5/mivisionx/default.nix
@@ -0,0 +1,145 @@
+{ lib
+, stdenv
+, fetchFromGitHub
+, rocmUpdateScript
+, cmake
+, rocm-cmake
+, rocm-device-libs
+, clr
+, pkg-config
+, rpp
+, rocblas
+, miopengemm
+, miopen
+, migraphx
+, clang
+, openmp
+, protobuf
+, qtcreator
+, opencv
+, ffmpeg
+, boost
+, libjpeg_turbo
+, half
+, lmdb
+, rapidjson
+, rocm-docs-core
+, python3Packages
+, useOpenCL ? false
+, useCPU ? false
+, buildDocs ? false # Needs internet
+, gpuTargets ? [ ]
+}:
+
+stdenv.mkDerivation (finalAttrs: {
+  pname = "mivisionx-" + (
+    if (!useOpenCL && !useCPU) then "hip"
+    else if (!useOpenCL && !useCPU) then "opencl"
+    else "cpu"
+  );
+
+  version = "5.7.0";
+
+  src = fetchFromGitHub {
+    owner = "GPUOpen-ProfessionalCompute-Libraries";
+    repo = "MIVisionX";
+    rev = "rocm-${finalAttrs.version}";
+    hash = "sha256-Z7UIqJWuSD+/FoZ1VIbITp4R/bwaqFCQqsL8CRW73Ek=";
+  };
+
+  nativeBuildInputs = [
+    cmake
+    rocm-cmake
+    clr
+    pkg-config
+  ] ++ lib.optionals buildDocs [
+    rocm-docs-core
+    python3Packages.python
+  ];
+
+  buildInputs = [
+    miopengemm
+    miopen
+    migraphx
+    rpp
+    rocblas
+    openmp
+    half
+    protobuf
+    qtcreator
+    opencv
+    ffmpeg
+    boost
+    libjpeg_turbo
+    lmdb
+    rapidjson
+    python3Packages.pybind11
+    python3Packages.numpy
+    python3Packages.torchWithRocm
+  ];
+
+  cmakeFlags = [
+    "-DROCM_PATH=${clr}"
+    "-DAMDRPP_PATH=${rpp}"
+    # Manually define CMAKE_INSTALL_<DIR>
+    # See: https://github.com/NixOS/nixpkgs/pull/197838
+    "-DCMAKE_INSTALL_BINDIR=bin"
+    "-DCMAKE_INSTALL_LIBDIR=lib"
+    "-DCMAKE_INSTALL_INCLUDEDIR=include"
+    "-DCMAKE_INSTALL_PREFIX_PYTHON=lib"
+    # "-DAMD_FP16_SUPPORT=ON" `error: typedef redefinition with different types ('__half' vs 'half_float::half')`
+  ] ++ lib.optionals (gpuTargets != [ ]) [
+    "-DAMDGPU_TARGETS=${lib.concatStringsSep ";" gpuTargets}"
+  ] ++ lib.optionals (!useOpenCL && !useCPU) [
+    "-DBACKEND=HIP"
+  ] ++ lib.optionals (useOpenCL && !useCPU) [
+    "-DBACKEND=OCL"
+  ] ++ lib.optionals useCPU [
+    "-DBACKEND=CPU"
+  ];
+
+  postPatch = ''
+    # We need to not use hipcc and define the CXXFLAGS manually due to `undefined hidden symbol: tensorflow:: ...`
+    export CXXFLAGS+="--rocm-path=${clr} --rocm-device-lib-path=${rocm-device-libs}/amdgcn/bitcode"
+    patchShebangs rocAL/rocAL_pybind/examples
+
+    # Properly find miopengemm and miopen
+    substituteInPlace amd_openvx_extensions/CMakeLists.txt \
+      --replace "miopengemm PATHS \''${ROCM_PATH} QUIET" "miopengemm PATHS ${miopengemm} QUIET" \
+      --replace "miopen     PATHS \''${ROCM_PATH} QUIET" "miopen PATHS ${miopen} QUIET" \
+      --replace "\''${ROCM_PATH}/include/miopen/config.h" "${miopen}/include/miopen/config.h"
+
+    # Properly find turbojpeg
+    substituteInPlace amd_openvx/cmake/FindTurboJpeg.cmake \
+      --replace "\''${TURBO_JPEG_PATH}/include" "${libjpeg_turbo.dev}/include" \
+      --replace "\''${TURBO_JPEG_PATH}/lib" "${libjpeg_turbo.out}/lib"
+
+    # Fix bad paths
+    substituteInPlace rocAL/rocAL/rocAL_hip/CMakeLists.txt amd_openvx_extensions/amd_nn/nn_hip/CMakeLists.txt amd_openvx/openvx/hipvx/CMakeLists.txt \
+      --replace "COMPILER_FOR_HIP \''${ROCM_PATH}/llvm/bin/clang++" "COMPILER_FOR_HIP ${clang}/bin/clang++"
+  '';
+
+  postBuild = lib.optionalString buildDocs ''
+    python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en ../docs _build/html
+  '';
+
+  postInstall = lib.optionalString (!useOpenCL && !useCPU) ''
+    patchelf $out/lib/rocal_pybind*.so --shrink-rpath --allowed-rpath-prefixes "$NIX_STORE"
+    chmod +x $out/lib/rocal_pybind*.so
+  '';
+
+  passthru.updateScript = rocmUpdateScript {
+    name = finalAttrs.pname;
+    owner = finalAttrs.src.owner;
+    repo = finalAttrs.src.repo;
+  };
+
+  meta = with lib; {
+    description = "Set of comprehensive computer vision and machine intelligence libraries, utilities, and applications";
+    homepage = "https://github.com/GPUOpen-ProfessionalCompute-Libraries/MIVisionX";
+    license = with licenses; [ mit ];
+    maintainers = teams.rocm.members;
+    platforms = platforms.linux;
+    broken = versions.minor finalAttrs.version != versions.minor stdenv.cc.version;
+  };
+})
diff --git a/pkgs/development/rocm-modules/5/rocblas/default.nix b/pkgs/development/rocm-modules/5/rocblas/default.nix
index 76dc38850d57f..17732fd516814 100644
--- a/pkgs/development/rocm-modules/5/rocblas/default.nix
+++ b/pkgs/development/rocm-modules/5/rocblas/default.nix
@@ -1,4 +1,5 @@
-{ lib
+{ rocblas
+, lib
 , stdenv
 , fetchFromGitHub
 , rocmUpdateScript
@@ -25,147 +26,54 @@
 , tensileLibFormat ? "msgpack"
 , gpuTargets ? [ "all" ]
 }:
+
 let
-  rocblas = stdenv.mkDerivation (finalAttrs: {
-    pname = "rocblas";
-    version = "5.7.0";
-
-    outputs = [
-      "out"
-    ] ++ lib.optionals buildTests [
-      "test"
-    ] ++ lib.optionals buildBenchmarks [
-      "benchmark"
+  # NOTE: Update the default GPU targets on every update
+  gfx80 = (rocblas.override {
+    gpuTargets = [
+      "gfx803"
     ];
-
-    src = fetchFromGitHub {
-      owner = "ROCmSoftwarePlatform";
-      repo = "rocBLAS";
-      rev = "rocm-${finalAttrs.version}";
-      hash = "sha256-3wKnwvAra8u9xqlC05wUD+gSoBILTVJFU2cIV6xv3Lk=";
-    };
-
-    nativeBuildInputs = [
-      cmake
-      rocm-cmake
-      clr
+  }).overrideAttrs { pname = "rocblas-tensile-gfx80"; };
+
+  gfx90 = (rocblas.override {
+    gpuTargets = [
+      "gfx900"
+      "gfx906:xnack-"
+      "gfx908:xnack-"
+      "gfx90a:xnack+"
+      "gfx90a:xnack-"
     ];
+  }).overrideAttrs { pname = "rocblas-tensile-gfx90"; };
 
-    buildInputs = [
-      python3
-    ] ++ lib.optionals buildTensile [
-      msgpack
-      libxml2
-      python3Packages.msgpack
-      python3Packages.joblib
-    ] ++ lib.optionals buildTests [
-      gtest
-    ] ++ lib.optionals (buildTests || buildBenchmarks) [
-      gfortran
-      openmp
-      amd-blis
-    ] ++ lib.optionals (buildTensile || buildTests || buildBenchmarks) [
-      python3Packages.pyyaml
+  gfx94 = (rocblas.override {
+    gpuTargets = [
+      "gfx940"
+      "gfx941"
+      "gfx942"
     ];
+  }).overrideAttrs { pname = "rocblas-tensile-gfx94"; };
 
-    cmakeFlags = [
-      "-DCMAKE_C_COMPILER=hipcc"
-      "-DCMAKE_CXX_COMPILER=hipcc"
-      "-Dpython=python3"
-      "-DAMDGPU_TARGETS=${lib.concatStringsSep ";" gpuTargets}"
-      "-DBUILD_WITH_TENSILE=${if buildTensile then "ON" else "OFF"}"
-      # Manually define CMAKE_INSTALL_<DIR>
-      # See: https://github.com/NixOS/nixpkgs/pull/197838
-      "-DCMAKE_INSTALL_BINDIR=bin"
-      "-DCMAKE_INSTALL_LIBDIR=lib"
-      "-DCMAKE_INSTALL_INCLUDEDIR=include"
-    ] ++ lib.optionals buildTensile [
-      "-DVIRTUALENV_HOME_DIR=/build/source/tensile"
-      "-DTensile_TEST_LOCAL_PATH=/build/source/tensile"
-      "-DTensile_ROOT=/build/source/tensile/lib/python${python3.pythonVersion}/site-packages/Tensile"
-      "-DTensile_LOGIC=${tensileLogic}"
-      "-DTensile_CODE_OBJECT_VERSION=${tensileCOVersion}"
-      "-DTensile_SEPARATE_ARCHITECTURES=${if tensileSepArch then "ON" else "OFF"}"
-      "-DTensile_LAZY_LIBRARY_LOADING=${if tensileLazyLib then "ON" else "OFF"}"
-      "-DTensile_LIBRARY_FORMAT=${tensileLibFormat}"
-    ] ++ lib.optionals buildTests [
-      "-DBUILD_CLIENTS_TESTS=ON"
-    ] ++ lib.optionals buildBenchmarks [
-      "-DBUILD_CLIENTS_BENCHMARKS=ON"
-    ] ++ lib.optionals (buildTests || buildBenchmarks) [
-      "-DCMAKE_CXX_FLAGS=-I${amd-blis}/include/blis"
+  gfx10 = (rocblas.override {
+    gpuTargets = [
+      "gfx1010"
+      "gfx1012"
+      "gfx1030"
     ];
+  }).overrideAttrs { pname = "rocblas-tensile-gfx10"; };
 
-    # Tensile REALLY wants to write to the nix directory if we include it normally
-    postPatch = lib.optionalString buildTensile ''
-      cp -a ${tensile} tensile
-      chmod +w -R tensile
-
-      # Rewrap Tensile
-      substituteInPlace tensile/bin/{.t*,.T*,*} \
-        --replace "${tensile}" "/build/source/tensile"
-
-      substituteInPlace CMakeLists.txt \
-        --replace "include(virtualenv)" "" \
-        --replace "virtualenv_install(\''${Tensile_TEST_LOCAL_PATH})" ""
-    '';
-
-    postInstall = lib.optionalString buildTests ''
-      mkdir -p $test/bin
-      cp -a $out/bin/* $test/bin
-      rm $test/bin/*-bench || true
-    '' + lib.optionalString buildBenchmarks ''
-      mkdir -p $benchmark/bin
-      cp -a $out/bin/* $benchmark/bin
-      rm $benchmark/bin/*-test || true
-    '' + lib.optionalString (buildTests || buildBenchmarks ) ''
-      rm -rf $out/bin
-    '';
-
-    passthru.updateScript = rocmUpdateScript {
-      name = finalAttrs.pname;
-      owner = finalAttrs.src.owner;
-      repo = finalAttrs.src.repo;
-    };
-
-    requiredSystemFeatures = [ "big-parallel" ];
-
-    meta = with lib; {
-      description = "BLAS implementation for ROCm platform";
-      homepage = "https://github.com/ROCmSoftwarePlatform/rocBLAS";
-      license = with licenses; [ mit ];
-      maintainers = teams.rocm.members;
-      platforms = platforms.linux;
-      broken = versions.minor finalAttrs.version != versions.minor stdenv.cc.version;
-    };
-  });
-
-  gfx80 = runCommand "rocblas-gfx80" { preferLocalBuild = true; } ''
-    mkdir -p $out/lib/rocblas/library
-    cp -a ${rocblas}/lib/rocblas/library/*gfx80* $out/lib/rocblas/library
-  '';
-
-  gfx90 = runCommand "rocblas-gfx90" { preferLocalBuild = true; } ''
-    mkdir -p $out/lib/rocblas/library
-    cp -a ${rocblas}/lib/rocblas/library/*gfx90* $out/lib/rocblas/library
-  '';
-
-  gfx94 = runCommand "rocblas-gfx94" { preferLocalBuild = true; } ''
-    mkdir -p $out/lib/rocblas/library
-    cp -a ${rocblas}/lib/rocblas/library/*gfx94* $out/lib/rocblas/library
-  '';
-
-  gfx10 = runCommand "rocblas-gfx10" { preferLocalBuild = true; } ''
-    mkdir -p $out/lib/rocblas/library
-    cp -a ${rocblas}/lib/rocblas/library/*gfx10* $out/lib/rocblas/library
-  '';
+  gfx11 = (rocblas.override {
+    gpuTargets = [
+      "gfx1100"
+      "gfx1101"
+      "gfx1102"
+    ];
+  }).overrideAttrs { pname = "rocblas-tensile-gfx11"; };
 
-  gfx11 = runCommand "rocblas-gfx11" { preferLocalBuild = true; } ''
-    mkdir -p $out/lib/rocblas/library
-    cp -a ${rocblas}/lib/rocblas/library/*gfx11* $out/lib/rocblas/library
-  '';
+  # Unfortunately, we have to do two full builds, otherwise we get overlapping _fallback.dat files
+  fallbacks = rocblas.overrideAttrs { pname = "rocblas-tensile-fallbacks"; };
 in stdenv.mkDerivation (finalAttrs: {
-  inherit (rocblas) pname version src passthru meta;
+  pname = "rocblas";
+  version = "5.7.0";
 
   outputs = [
     "out"
@@ -175,26 +83,127 @@ in stdenv.mkDerivation (finalAttrs: {
     "benchmark"
   ];
 
-  dontUnpack = true;
-  dontPatch = true;
-  dontConfigure = true;
-  dontBuild = true;
-
-  installPhase = ''
-    runHook preInstall
-
-    mkdir -p $out
-    cp -a --no-preserve=mode ${rocblas}/* $out
-    ln -sf ${gfx80}/lib/rocblas/library/* $out/lib/rocblas/library
-    ln -sf ${gfx90}/lib/rocblas/library/* $out/lib/rocblas/library
-    ln -sf ${gfx94}/lib/rocblas/library/* $out/lib/rocblas/library
-    ln -sf ${gfx10}/lib/rocblas/library/* $out/lib/rocblas/library
-    ln -sf ${gfx11}/lib/rocblas/library/* $out/lib/rocblas/library
+  src = fetchFromGitHub {
+    owner = "ROCmSoftwarePlatform";
+    repo = "rocBLAS";
+    rev = "rocm-${finalAttrs.version}";
+    hash = "sha256-3wKnwvAra8u9xqlC05wUD+gSoBILTVJFU2cIV6xv3Lk=";
+  };
+
+  nativeBuildInputs = [
+    cmake
+    rocm-cmake
+    clr
+  ];
+
+  buildInputs = [
+    python3
+  ] ++ lib.optionals buildTensile [
+    msgpack
+    libxml2
+    python3Packages.msgpack
+    python3Packages.joblib
+  ] ++ lib.optionals buildTests [
+    gtest
+  ] ++ lib.optionals (buildTests || buildBenchmarks) [
+    gfortran
+    openmp
+    amd-blis
+  ] ++ lib.optionals (buildTensile || buildTests || buildBenchmarks) [
+    python3Packages.pyyaml
+  ];
+
+  cmakeFlags = [
+    "-DCMAKE_C_COMPILER=hipcc"
+    "-DCMAKE_CXX_COMPILER=hipcc"
+    "-Dpython=python3"
+    "-DAMDGPU_TARGETS=${lib.concatStringsSep ";" gpuTargets}"
+    "-DBUILD_WITH_TENSILE=${if buildTensile then "ON" else "OFF"}"
+    # Manually define CMAKE_INSTALL_<DIR>
+    # See: https://github.com/NixOS/nixpkgs/pull/197838
+    "-DCMAKE_INSTALL_BINDIR=bin"
+    "-DCMAKE_INSTALL_LIBDIR=lib"
+    "-DCMAKE_INSTALL_INCLUDEDIR=include"
+  ] ++ lib.optionals buildTensile [
+    "-DVIRTUALENV_HOME_DIR=/build/source/tensile"
+    "-DTensile_TEST_LOCAL_PATH=/build/source/tensile"
+    "-DTensile_ROOT=/build/source/tensile/${python3.sitePackages}/Tensile"
+    "-DTensile_LOGIC=${tensileLogic}"
+    "-DTensile_CODE_OBJECT_VERSION=${tensileCOVersion}"
+    "-DTensile_SEPARATE_ARCHITECTURES=${if tensileSepArch then "ON" else "OFF"}"
+    "-DTensile_LAZY_LIBRARY_LOADING=${if tensileLazyLib then "ON" else "OFF"}"
+    "-DTensile_LIBRARY_FORMAT=${tensileLibFormat}"
+  ] ++ lib.optionals buildTests [
+    "-DBUILD_CLIENTS_TESTS=ON"
+  ] ++ lib.optionals buildBenchmarks [
+    "-DBUILD_CLIENTS_BENCHMARKS=ON"
+  ] ++ lib.optionals (buildTests || buildBenchmarks) [
+    "-DCMAKE_CXX_FLAGS=-I${amd-blis}/include/blis"
+  ];
+
+  postPatch = lib.optionalString (finalAttrs.pname != "rocblas") ''
+    # Return early and install tensile files manually
+    substituteInPlace library/src/CMakeLists.txt \
+      --replace "set_target_properties( TensileHost PROPERTIES OUTPUT_NAME" "return()''\nset_target_properties( TensileHost PROPERTIES OUTPUT_NAME"
+  '' + lib.optionalString (buildTensile && finalAttrs.pname == "rocblas") ''
+    # Link the prebuilt Tensile files
+    mkdir -p build/Tensile/library
+
+    for path in ${gfx80} ${gfx90} ${gfx94} ${gfx10} ${gfx11} ${fallbacks}; do
+      ln -s $path/lib/rocblas/library/* build/Tensile/library
+    done
+
+    unlink build/Tensile/library/TensileManifest.txt
+  '' + lib.optionalString buildTensile ''
+    # Tensile REALLY wants to write to the nix directory if we include it normally
+    cp -a ${tensile} tensile
+    chmod +w -R tensile
+
+    # Rewrap Tensile
+    substituteInPlace tensile/bin/{.t*,.T*,*} \
+      --replace "${tensile}" "/build/source/tensile"
+
+    substituteInPlace CMakeLists.txt \
+      --replace "include(virtualenv)" "" \
+      --replace "virtualenv_install(\''${Tensile_TEST_LOCAL_PATH})" ""
+  '';
+
+  postInstall = lib.optionalString (finalAttrs.pname == "rocblas") ''
+    ln -sf ${fallbacks}/lib/rocblas/library/TensileManifest.txt $out/lib/rocblas/library
+  '' + lib.optionalString (finalAttrs.pname != "rocblas") ''
+    mkdir -p $out/lib/rocblas/library
+    rm -rf $out/share
+  '' + lib.optionalString (finalAttrs.pname != "rocblas" && finalAttrs.pname != "rocblas-tensile-fallbacks") ''
+    rm Tensile/library/{TensileManifest.txt,*_fallback.dat}
+    mv Tensile/library/* $out/lib/rocblas/library
+  '' + lib.optionalString (finalAttrs.pname == "rocblas-tensile-fallbacks") ''
+    mv Tensile/library/{TensileManifest.txt,*_fallback.dat} $out/lib/rocblas/library
   '' + lib.optionalString buildTests ''
-    cp -a ${rocblas.test} $test
+    mkdir -p $test/bin
+    cp -a $out/bin/* $test/bin
+    rm $test/bin/*-bench || true
   '' + lib.optionalString buildBenchmarks ''
-    cp -a ${rocblas.benchmark} $benchmark
-  '' + ''
-    runHook postInstall
+    mkdir -p $benchmark/bin
+    cp -a $out/bin/* $benchmark/bin
+    rm $benchmark/bin/*-test || true
+  '' + lib.optionalString (buildTests || buildBenchmarks ) ''
+    rm -rf $out/bin
   '';
+
+  passthru.updateScript = rocmUpdateScript {
+    name = finalAttrs.pname;
+    owner = finalAttrs.src.owner;
+    repo = finalAttrs.src.repo;
+  };
+
+  requiredSystemFeatures = [ "big-parallel" ];
+
+  meta = with lib; {
+    description = "BLAS implementation for ROCm platform";
+    homepage = "https://github.com/ROCmSoftwarePlatform/rocBLAS";
+    license = with licenses; [ mit ];
+    maintainers = teams.rocm.members;
+    platforms = platforms.linux;
+    broken = versions.minor finalAttrs.version != versions.minor stdenv.cc.version;
+  };
 })
diff --git a/pkgs/development/rocm-modules/5/rpp/default.nix b/pkgs/development/rocm-modules/5/rpp/default.nix
new file mode 100644
index 0000000000000..eee10f2ab7c82
--- /dev/null
+++ b/pkgs/development/rocm-modules/5/rpp/default.nix
@@ -0,0 +1,88 @@
+{ lib
+, stdenv
+, fetchFromGitHub
+, rocmUpdateScript
+, cmake
+, rocm-cmake
+, rocm-docs-core
+, half
+, clr
+, openmp
+, boost
+, python3Packages
+, buildDocs ? false # Needs internet
+, useOpenCL ? false
+, useCPU ? false
+, gpuTargets ? [ ]
+}:
+
+stdenv.mkDerivation (finalAttrs: {
+  pname = "rpp-" + (
+    if (!useOpenCL && !useCPU) then "hip"
+    else if (!useOpenCL && !useCPU) then "opencl"
+    else "cpu"
+  );
+
+  version = "5.7.0";
+
+  src = fetchFromGitHub {
+    owner = "GPUOpen-ProfessionalCompute-Libraries";
+    repo = "rpp";
+    rev = "rocm-${finalAttrs.version}";
+    hash = "sha256-s6ODmxPBLpR5f8VALaW6F0p0rZSxSd2LH2+60SEfLCk=";
+  };
+
+  nativeBuildInputs = [
+    cmake
+    rocm-cmake
+    clr
+  ] ++ lib.optionals buildDocs [
+    rocm-docs-core
+    python3Packages.python
+  ];
+
+  buildInputs = [
+    half
+    openmp
+    boost
+  ];
+
+  cmakeFlags = [
+    "-DROCM_PATH=${clr}"
+  ] ++ lib.optionals (gpuTargets != [ ]) [
+    "-DAMDGPU_TARGETS=${lib.concatStringsSep ";" gpuTargets}"
+  ] ++ lib.optionals (!useOpenCL && !useCPU) [
+    "-DCMAKE_C_COMPILER=hipcc"
+    "-DCMAKE_CXX_COMPILER=hipcc"
+    "-DBACKEND=HIP"
+  ] ++ lib.optionals (useOpenCL && !useCPU) [
+    "-DBACKEND=OCL"
+  ] ++ lib.optionals useCPU [
+    "-DBACKEND=CPU"
+  ];
+
+  postPatch = lib.optionalString (!useOpenCL && !useCPU) ''
+    # Bad path
+    substituteInPlace CMakeLists.txt \
+      --replace "COMPILER_FOR_HIP \''${ROCM_PATH}/llvm/bin/clang++" "COMPILER_FOR_HIP ${clr}/bin/hipcc"
+  '';
+
+  postBuild = lib.optionalString buildDocs ''
+    python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en ../docs _build/html
+  '';
+
+  passthru.updateScript = rocmUpdateScript {
+    name = finalAttrs.pname;
+    owner = finalAttrs.src.owner;
+    repo = finalAttrs.src.repo;
+  };
+
+  meta = with lib; {
+    description = "Comprehensive high-performance computer vision library for AMD processors";
+    homepage = "https://github.com/GPUOpen-ProfessionalCompute-Libraries/rpp";
+    license = with licenses; [ mit ];
+    maintainers = teams.rocm.members;
+    platforms = platforms.linux;
+    broken = versions.minor finalAttrs.version != versions.minor stdenv.cc.version;
+  };
+})