Update software stack to NEO v25.27.34303

.gitignore

@@ -1,2 +1,8 @@
 LocalPreferences.toml
 Manifest.toml
+deps/onemkl_blas.cpp
+deps/onemkl_blas.h
+deps/onemkl_lapack.cpp
+deps/onemkl_lapack.h
+deps/onemkl_sparse.cpp
+deps/onemkl_sparse.h

Project.toml

@@ -37,14 +37,14 @@ GPUCompiler = "1.5"
 GPUToolbox = "0.1, 0.2, 0.3"
 KernelAbstractions = "0.9.1"
 LLVM = "6, 7, 8, 9"
-NEO_jll = "=24.26.30049"
+NEO_jll = "=25.27.34303"
 Preferences = "1"
 SPIRVIntrinsics = "0.2"
 SPIRV_LLVM_Translator_jll = "20"
 SpecialFunctions = "1.3, 2"
 StaticArrays = "1"
 julia = "1.10"
-oneAPI_Level_Zero_Loader_jll = "1.9"
+oneAPI_Level_Zero_Loader_jll = "1.22"
 oneAPI_Support_jll = "0.8"
 
 [extras]

README.md

@@ -302,3 +302,4 @@ The discovered paths will be written to a global file with preferences, typicall
 `$HOME/.julia/environments/vX.Y/LocalPreferences.toml` (where `vX.Y` refers to the Julia
 version you are using). You can modify this file, or remove it when you want to revert to
 default set of binaries.
+

deps/Project.toml

@@ -12,4 +12,4 @@ oneAPI_Level_Zero_Headers_jll = "f4bc562b-d309-54f8-9efb-476e56f0410d"
 oneAPI_Support_Headers_jll = "24f86df5-245d-5634-a4cc-32433d9800b3"
 
 [compat]
-oneAPI_Support_Headers_jll = "=2025.1.0"
+oneAPI_Support_Headers_jll = "=2025.2.0"

deps/build_local.jl

@@ -45,7 +45,7 @@ if !isfile(joinpath(conda_dir, "condarc-julia.yml"))
     touch(joinpath(conda_dir, "conda-meta", "history"))
 end
 Conda.add_channel("https://software.repos.intel.com/python/conda/", conda_dir)
-Conda.add(["dpcpp_linux-64=2025.1.0", "mkl-devel-dpcpp=2025.1.0"], conda_dir)
+Conda.add(["dpcpp_linux-64=2025.2.0", "mkl-devel-dpcpp=2025.2.0"], conda_dir)
 
 Conda.list(conda_dir)
 

deps/generate_helpers.jl

@@ -1,6 +1,6 @@
 non_parametric_routines = ["init_matrix_handle", "release_matrix_handle", "set_matrix_property",
 "init_matmat_descr", "release_matmat_descr", "set_matmat_data", "get_matmat_data", "matmat",
-"omatcopy", "sort_matrix", "optimize_gemv", "optimize_trmv", "optimize_trsv", "optimize_trsm",
+"omatcopy", "sort_matrix", "optimize_gemv", "optimize_gemm", "optimize_trmv", "optimize_trsv", "optimize_trsm",
 "init_omatconvert_descr", "release_omatconvert_descr", "init_omatadd_descr", "release_omatadd_descr",
 "omatconvert_buffer_size", "omatconvert_analyze", "omatconvert_get_nnz", "omatconvert",
 "omatadd_buffer_size", "omatadd_analyze", "omatadd_get_nnz"]

deps/generate_interfaces.jl

@@ -241,6 +241,7 @@ function generate_headers(library::String, filename::Vector{String}, output::Str
         # The routine "optimize_trsm" has two versions.
         suffix = ""
         (name_routine == "optimize_trsm") && occursin("columns", header) && (suffix = "_advanced")
+        (name_routine == "optimize_gemm") && occursin("columns", header) && (suffix = "_advanced")
         name_routine ∈ ("set_csr_data", "set_coo_data") && occursin("int64_t", header) && (suffix = "_64")
         occursin("batch", name_routine) && !occursin("**", header) && (suffix = "_strided")
 

deps/src/onemkl.cpp

@@ -4038,6 +4038,16 @@ extern "C" int onemklXsparse_optimize_trsv(syclQueue_t device_queue, onemklUplo
    return 0;
 }
 
+extern "C" int onemklXsparse_optimize_gemm(syclQueue_t device_queue, onemklTranspose opA, matrix_handle_t A) {
+   auto status = oneapi::mkl::sparse::optimize_gemm(device_queue->val, convert(opA), (oneapi::mkl::sparse::matrix_handle_t) A, {});
+   return 0;
+}
+
+extern "C" int onemklXsparse_optimize_gemm_advanced(syclQueue_t device_queue, onemklLayout layout_val, onemklTranspose opA, onemklTranspose opB, matrix_handle_t A, int64_t columns) {
+   auto status = oneapi::mkl::sparse::optimize_gemm(device_queue->val, convert(layout_val), convert(opA), convert(opB), (oneapi::mkl::sparse::matrix_handle_t) A, columns, {});
+   return 0;
+}
+
 extern "C" int onemklXsparse_optimize_trsm(syclQueue_t device_queue, onemklUplo uplo_val, onemklTranspose opA, onemklDiag diag_val, matrix_handle_t A) {
    auto status = oneapi::mkl::sparse::optimize_trsm(device_queue->val, convert(uplo_val), convert(opA), convert(diag_val), (oneapi::mkl::sparse::matrix_handle_t) A, {});
    return 0;

deps/src/onemkl.h

@@ -2842,6 +2842,13 @@ int onemklXsparse_optimize_trmv(syclQueue_t device_queue, onemklUplo uplo_val, o
 int onemklXsparse_optimize_trsv(syclQueue_t device_queue, onemklUplo uplo_val, onemklTranspose
                                 opA, onemklDiag diag_val, matrix_handle_t A);
 
+int onemklXsparse_optimize_gemm(syclQueue_t device_queue, onemklTranspose opA, matrix_handle_t
+                                A);
+
+int onemklXsparse_optimize_gemm_advanced(syclQueue_t device_queue, onemklLayout layout_val,
+                                         onemklTranspose opA, onemklTranspose opB,
+                                         matrix_handle_t A, int64_t columns);
+
 int onemklXsparse_optimize_trsm(syclQueue_t device_queue, onemklUplo uplo_val, onemklTranspose
                                 opA, onemklDiag diag_val, matrix_handle_t A);
 

lib/level-zero/common.jl

@@ -1,2 +1,15 @@
-#define ZE_MAKE_VERSION( _major, _minor )  (( _major << 16 )|( _minor & 0x0000ffff))
+"""
+    ze_make_version(major::Integer, minor::Integer) -> UInt32
+
+32-bit unsigned integer version number from major and minor components.
+This should be the Julia equivalent of the C macro:
+`#define ZE_MAKE_VERSION( _major, _minor )  (( _major << 16 )|( _minor & 0x0000ffff))`
+"""
+function ZE_MAKE_VERSION(major::Integer, minor::Integer)
+    # Shift the major version 16 bits to the left
+    # and combine it with the minor version using a bitwise OR.
+    # The `& 0xffff` is implicit for standard integer types when combining,
+    # but we can be explicit if needed. The result is cast to UInt32.
+    return (UInt32(major) << 16) | (UInt32(minor) & 0x0000ffff)
+end
 unmake_version(ver) = VersionNumber(Int(ver) >> 16, Int(ver) & 0x0000ffff)