diff --git a/var/spack/repos/builtin/packages/quda/package.py b/var/spack/repos/builtin/packages/quda/package.py
new file mode 100644
index 0000000000000000000000000000000000000000..16f3b2716217e705d14d0bd8025bf12c24023546
--- /dev/null
+++ b/var/spack/repos/builtin/packages/quda/package.py
@@ -0,0 +1,198 @@
+# Copyright Spack Project Developers. See COPYRIGHT file for details.
+#
+# SPDX-License-Identifier: (Apache-2.0 OR MIT)
+
+from spack.package import *
+
+
+class Quda(CMakePackage, CudaPackage, ROCmPackage):
+ """QUDA is a library for performing calculations in lattice QCD on GPUs."""
+
+ homepage = "https://lattice.github.io/quda/"
+ url = "https://github.com/lattice/quda/archive/refs/tags/v1.1.0.tar.gz"
+ git = "https://github.com/lattice/quda.git"
+
+ tags = ["hep", "lattice"]
+
+ maintainers("chaoos")
+
+ license("MIT OR BSD-3-Clause", checked_by="chaoos")
+
+ version("develop", branch="develop")
+
+ # git describe --tags --match 'v*' 18bf43ed40c75ae276e55bb8ddf2f64aa5510c37
+ version(
+ "1.1.0-4597-g18bf43ed4", preferred=True, commit="18bf43ed40c75ae276e55bb8ddf2f64aa5510c37"
+ )
+
+ version("1.1.0", sha256="b4f635c993275010780ea09d8e593e0713a6ca1af1db6cc86c64518714fcc745")
+ version(
+ "1.0.0",
+ deprecated=True,
+ sha256="32b883bd4af45b76a832d8a070ab020306c94ff6590410cbe7c3eab3b630b938",
+ )
+ version(
+ "0.9.0",
+ deprecated=True,
+ sha256="0a9f2e028fb40e4a09f78af51702d2de4099a9bf10fb8ce350eacb2d8327e481",
+ )
+ version(
+ "0.8.0",
+ deprecated=True,
+ sha256="58d9a94b7fd38ec1f79bea7a0e9b470f0fa517cbf5fce42d5776e1c65607aeda",
+ )
+
+ # build dependencies
+ generator("ninja")
+ depends_on("cmake@3.18:", type="build")
+ depends_on("ninja", type="build")
+ depends_on("c", type="build")
+ depends_on("cxx", type="build")
+ depends_on("fortran", type="build", when="+tifr")
+ depends_on("fortran", type="build", when="+bqcd")
+
+ variant("mpi", default=False, description="Enable MPI support")
+ variant("qmp", default=False, description="Enable QMP")
+ variant("qio", default=False, description="Enable QIO")
+ variant("openqcd", default=False, description="Enable openQCD interface")
+ variant("milc", default=False, description="Enable MILC interface")
+ variant("qdp", default=False, description="Enable QDP interface")
+ variant("bqcd", default=False, description="Enable BQCD interface")
+ variant("cps", default=False, description="Enable CPS interface")
+ variant("qdpjit", default=False, description="Enable QDPJIT interface")
+ variant("tifr", default=False, description="Enable TIFR interface")
+ variant("multigrid", default=False, description="Enable multigrid")
+ variant("nvshmem", default=False, description="Enable NVSHMEM", when="+cuda")
+
+ variant("clover", default=False, description="Build clover Dirac operators")
+ variant(
+ "clover_hasenbusch", default=False, description="Build clover Hasenbusch twist operators"
+ )
+ variant("domain_wall", default=False, description="Build domain wall Dirac operators")
+ variant("laplace", default=False, description="Build laplace operator")
+ variant(
+ "ndeg_twisted_clover",
+ default=False,
+ description="Build non-degenerate twisted clover Dirac operators",
+ )
+ variant(
+ "ndeg_twisted_mass",
+ default=False,
+ description="Build non-degenerate twisted mass Dirac operators",
+ )
+ variant("staggered", default=False, description="Build staggered Dirac operators")
+ variant("twisted_clover", default=False, description="Build twisted clover Dirac operators")
+ variant("twisted_mass", default=False, description="Build twisted mass Dirac operators")
+ variant("wilson", default=True, description="Build Wilson Dirac operators")
+
+ with when("+multigrid"):
+ variant(
+ "mg_mrhs_list",
+ default=16,
+ multi=True,
+ description="The list of multi-rhs sizes that get compiled",
+ )
+ variant(
+ "mg_nvec_list",
+ default="6,24,32",
+ multi=True,
+ description="The list of null space vector sizes that get compiled",
+ )
+
+ # dependencies
+ depends_on("mpi", when="+mpi")
+ depends_on("cuda", when="+cuda")
+ depends_on("nvshmem", when="+nvshmem")
+ depends_on("gdrcopy", when="+nvshmem")
+ with when("+rocm"):
+ depends_on("hip")
+ depends_on("hipblas")
+ depends_on("hipfft")
+ depends_on("hiprand")
+ depends_on("hipcub")
+
+ conflicts("+qmp +mpi", msg="Specifying both QMP and MPI might result in undefined behavior")
+ conflicts("+cuda +rocm", msg="CUDA and ROCm support are mutually exclusive")
+ conflicts("~cuda ~rocm", msg="Either CUDA or ROCm support is required")
+ conflicts("cuda_arch=none", when="+cuda", msg="Please indicate a cuda_arch value")
+ conflicts(
+ "+nvshmem", when="~mpi ~qmp", msg="NVSHMEM requires either +mpi or +qmp to be enabled"
+ )
+
+ # CMAKE_BUILD_TYPE
+ variant(
+ "build_type",
+ default="STRICT",
+ description="The build type to build",
+ values=("STRICT", "RELEASE", "DEVEL", "DEBUG", "HOSTDEBUG", "SANITIZE"),
+ )
+
+ def cmake_args(self):
+ if self.spec.satisfies("+cuda"):
+ target = "CUDA"
+ cuda_archs = self.spec.variants["cuda_arch"].value
+ arch = " ".join(f"sm_{i}" for i in cuda_archs)
+ elif self.spec.satisfies("+rocm"):
+ target = "HIP"
+ arch = self.spec.variants["amdgpu_target"].value
+
+ args = [
+ self.define("QUDA_BUILD_ALL_TESTS", False),
+ self.define("QUDA_TARGET_TYPE", target),
+ self.define("QUDA_GPU_ARCH", arch),
+ self.define("QUDA_PRECISION", 14),
+ self.define("QUDA_RECONSTRUCT", 7),
+ self.define("QUDA_DOWNLOAD_USQCD", False),
+ self.define("QUDA_DIRAC_DEFAULT_OFF", True),
+ self.define_from_variant("QUDA_DIRAC_CLOVER", "clover"),
+ self.define_from_variant("QUDA_DIRAC_CLOVER_HASENBUSCH", "clover_hasenbusch"),
+ self.define_from_variant("QUDA_DIRAC_DOMAIN_WALL", "domain_wall"),
+ self.define_from_variant("QUDA_DIRAC_LAPLACE", "laplace"),
+ self.define_from_variant("QUDA_DIRAC_NDEG_TWISTED_CLOVER", "ndeg_twisted_clover"),
+ self.define_from_variant("QUDA_DIRAC_NDEG_TWISTED_MASS", "ndeg_twisted_mass"),
+ self.define_from_variant("QUDA_DIRAC_STAGGERED", "staggered"),
+ self.define_from_variant("QUDA_DIRAC_TWISTED_CLOVER", "twisted_clover"),
+ self.define_from_variant("QUDA_DIRAC_TWISTED_MASS", "twisted_mass"),
+ self.define_from_variant("QUDA_DIRAC_WILSON", "wilson"),
+ self.define_from_variant("QUDA_MPI", "mpi"),
+ self.define_from_variant("QUDA_QMP", "qmp"),
+ self.define_from_variant("QUDA_QIO", "qio"),
+ self.define_from_variant("QUDA_INTERFACE_OPENQCD", "openqcd"),
+ self.define_from_variant("QUDA_INTERFACE_MILC", "milc"),
+ self.define_from_variant("QUDA_INTERFACE_QDP", "qdp"),
+ self.define_from_variant("QUDA_INTERFACE_BQCD", "bqcd"),
+ self.define_from_variant("QUDA_INTERFACE_CPS", "cps"),
+ self.define_from_variant("QUDA_INTERFACE_QDPJIT", "qdpjit"),
+ self.define_from_variant("QUDA_INTERFACE_TIFR", "tifr"),
+ self.define_from_variant("QUDA_MULTIGRID", "multigrid"),
+ self.define_from_variant("QUDA_NVSHMEM", "nvshmem"),
+ ]
+ if self.spec.satisfies("+multigrid"):
+ args.append(
+ self.define(
+ "QUDA_MULTIGRID_NVEC_LIST", ",".join(self.spec.variants["mg_nvec_list"].value)
+ )
+ )
+ args.append(
+ self.define(
+ "QUDA_MULTIGRID_MRHS_LIST", ",".join(self.spec.variants["mg_mrhs_list"].value)
+ )
+ )
+
+ if self.spec.satisfies("+nvshmem"):
+ args.append(self.define("QUDA_NVSHMEM_HOME", self.spec["nvshmem"].prefix))
+ args.append(self.define("QUDA_GDRCOPY_HOME", self.spec["gdrcopy"].prefix))
+
+ if self.spec.satisfies("+cuda"):
+ args.append(self.define("QUDA_GPU_ARCH_SUFFIX", "real")) # real or virtual
+ elif self.spec.satisfies("+rocm"):
+ args.append(self.define("CMAKE_C_COMPILER", self.spec["hip"].hipcc))
+ args.append(self.define("CMAKE_CXX_COMPILER", self.spec["hip"].hipcc))
+ # args.append(self.define("ROCM_PATH", self.spec["hip"].prefix))
+
+ # required when building on a machine with no AMD GPU present
+ args.append(self.define("AMDGPU_TARGETS", arch))
+
+ # suppress _GLIBCXX17_DEPRECATED warnings when compiling c++17
+ args.append(self.define("CMAKE_CXX_FLAGS", "-Wno-deprecated-declarations"))
+ return args