From 28d779ffc83a565f2590a737d95b178606f424b5 Mon Sep 17 00:00:00 2001
From: Adarsh Yoga <adarsh.yoga@intel.com>
Date: Wed, 24 Apr 2024 05:18:56 +0000
Subject: [PATCH 01/10] change private arrays to use new dpex api

---
 dpbench/benchmarks/default/gpairs/gpairs_numba_dpex_k.py | 8 ++++----
 dpbench/benchmarks/default/kmeans/kmeans_numba_dpex_k.py | 4 ++--
 dpbench/benchmarks/default/knn/knn_numba_dpex_k.py       | 2 +-
 3 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/dpbench/benchmarks/default/gpairs/gpairs_numba_dpex_k.py b/dpbench/benchmarks/default/gpairs/gpairs_numba_dpex_k.py
index 52034cef..f5a72b00 100644
--- a/dpbench/benchmarks/default/gpairs/gpairs_numba_dpex_k.py
+++ b/dpbench/benchmarks/default/gpairs/gpairs_numba_dpex_k.py
@@ -38,9 +38,9 @@ def count_weighted_pairs_3d_intel_no_slm_ker(
 
     n_wi = 32
 
-    dsq_mat = dpex.private.array(shape=(32 * 32), dtype=dtype)
-    w0_vec = dpex.private.array(shape=(32), dtype=dtype)
-    w1_vec = dpex.private.array(shape=(32), dtype=dtype)
+    dsq_mat = kapi.PrivateArray(shape=(32 * 32), dtype=dtype)
+    w0_vec = kapi.PrivateArray(shape=(32), dtype=dtype)
+    w1_vec = kapi.PrivateArray(shape=(32), dtype=dtype)
 
     offset0 = gr0 * n_wi * lws0 + lid0
     offset1 = gr1 * n_wi * lws1 + lid1
@@ -80,7 +80,7 @@ def count_weighted_pairs_3d_intel_no_slm_ker(
 
     # update slm_hist. Use work-item private buffer of 16 tfloat elements
     for k in range(0, slm_hist_size, private_hist_size):
-        private_hist = dpex.private.array(shape=(32), dtype=dtype)
+        private_hist = kapi.PrivateArray(shape=(32), dtype=dtype)
         for p in range(private_hist_size):
             private_hist[p] = 0.0
 
diff --git a/dpbench/benchmarks/default/kmeans/kmeans_numba_dpex_k.py b/dpbench/benchmarks/default/kmeans/kmeans_numba_dpex_k.py
index 6bac670c..c89b6b8d 100644
--- a/dpbench/benchmarks/default/kmeans/kmeans_numba_dpex_k.py
+++ b/dpbench/benchmarks/default/kmeans/kmeans_numba_dpex_k.py
@@ -56,7 +56,7 @@ def groupByCluster(
         for i in range(WorkPI):
             point_id = grid * WorkPI * local_size + i * local_size + lid
             if point_id < numpoints:
-                localP = dpex.private.array(dims, dtyp)
+                localP = kapi.PrivateArray(dims, dtyp)
                 for d in range(dims):
                     localP[d] = arrayP[point_id, d]
 
@@ -179,7 +179,7 @@ def updateLabels(
         for i in range(WorkPI):
             point_id = grid * WorkPI * local_size + i * local_size + lid
             if point_id < numpoints:
-                localP = dpex.private.array(dims, dtyp)
+                localP = kapi.PrivateArray(dims, dtyp)
                 for d in range(dims):
                     localP[d] = arrayP[point_id, d]
 
diff --git a/dpbench/benchmarks/default/knn/knn_numba_dpex_k.py b/dpbench/benchmarks/default/knn/knn_numba_dpex_k.py
index 8771efa6..4ec82875 100644
--- a/dpbench/benchmarks/default/knn/knn_numba_dpex_k.py
+++ b/dpbench/benchmarks/default/knn/knn_numba_dpex_k.py
@@ -25,7 +25,7 @@ def _knn_kernel(  # noqa: C901: TODO: can we simplify logic?
     dtype = train.dtype
     i = item.get_id(0)
     # here k has to be 5 in order to match with numpy
-    queue_neighbors = dpex.private.array(shape=(5, 2), dtype=dtype)
+    queue_neighbors = kapi.PrivateArray(shape=(5, 2), dtype=dtype)
 
     for j in range(k):
         x1 = train[j]

From 36e3c36df3973a9515b5ed6686cae55dcdd44a59 Mon Sep 17 00:00:00 2001
From: Adarsh Yoga <adarsh.yoga@intel.com>
Date: Wed, 24 Apr 2024 05:19:40 +0000
Subject: [PATCH 02/10] change dpctl.asarray calls in dpcpp framework

---
 dpbench/infrastructure/frameworks/dpcpp_framework.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/dpbench/infrastructure/frameworks/dpcpp_framework.py b/dpbench/infrastructure/frameworks/dpcpp_framework.py
index 21b4c336..7c47fd12 100644
--- a/dpbench/infrastructure/frameworks/dpcpp_framework.py
+++ b/dpbench/infrastructure/frameworks/dpcpp_framework.py
@@ -59,7 +59,7 @@ def _copy_to_func_impl(ref_array):
             else:
                 order = "K"
             return dpt.asarray(
-                obj=ref_array,
+                ref_array,
                 dtype=ref_array.dtype,
                 device=self.sycl_device,
                 copy=None,

From 45cb63fe9705c1885d166ef9fe1b30fc26a36729 Mon Sep 17 00:00:00 2001
From: Adarsh Yoga <adarsh.yoga@intel.com>
Date: Wed, 24 Apr 2024 05:20:50 +0000
Subject: [PATCH 03/10] fix version pinning in yml file

---
 conda-recipe/meta.yaml                    | 2 +-
 dpbench/configs/framework_info/dpcpp.toml | 2 +-
 environments/conda-linux-sycl.yml         | 5 ++---
 environments/conda-win-sycl.yml           | 2 +-
 4 files changed, 5 insertions(+), 6 deletions(-)

diff --git a/conda-recipe/meta.yaml b/conda-recipe/meta.yaml
index cb2ff422..b1e19f30 100644
--- a/conda-recipe/meta.yaml
+++ b/conda-recipe/meta.yaml
@@ -17,7 +17,7 @@ build:
 requirements:
   build:
     - {{ compiler('cxx') }}
-    - {{ compiler('dpcpp') }} ==2024.0.0  # [not osx]
+    - {{ compiler('dpcpp') }}
     - sysroot_linux-64 >=2.28  # [linux]
   host:
     - python
diff --git a/dpbench/configs/framework_info/dpcpp.toml b/dpbench/configs/framework_info/dpcpp.toml
index 77ce097d..c1655e91 100644
--- a/dpbench/configs/framework_info/dpcpp.toml
+++ b/dpbench/configs/framework_info/dpcpp.toml
@@ -10,7 +10,7 @@ postfix = "dpcpp"
 class = "DpcppFramework"
 arch = "cpu"
 sycl_device = "cpu"
-dpcpp_version = "IntelLLVM 2024.0.0"
+dpcpp_version = "IntelLLVM 2024.1.0"
 
 [[framework.postfixes]]
 impl_postfix = "sycl"
diff --git a/environments/conda-linux-sycl.yml b/environments/conda-linux-sycl.yml
index e51648a3..d0b4113a 100644
--- a/environments/conda-linux-sycl.yml
+++ b/environments/conda-linux-sycl.yml
@@ -23,9 +23,8 @@ dependencies:
   - dpctl
   - dpnp
   - numba-dpex
-  # TODO: fix issues on conda-forge build
-  - intel::dpcpp_linux-64==2024.0.0
-  - intel::dpcpp-cpp-rt==2024.0.0
+  - intel::dpcpp_linux-64
+  - intel::dpcpp-cpp-rt
   - cython
   - cmake
   - ninja
diff --git a/environments/conda-win-sycl.yml b/environments/conda-win-sycl.yml
index 89ad4c9b..d3160f22 100644
--- a/environments/conda-win-sycl.yml
+++ b/environments/conda-win-sycl.yml
@@ -25,7 +25,7 @@ dependencies:
   - numba-dpex
   # TODO: switch to conda-forge, but it results in broken OpenCL rt (see below)
   # - conda-forge::dpcpp_win-64
-  - intel::dpcpp_win-64==2024.0.0
+  - intel::dpcpp_win-64
   # fixing cmake version here, because we need to apply patch for IntelLLVM
   - cmake==3.26*
   - cython

From 35ee147deb8a7ce7cb5651f5cffb93e2738d280b Mon Sep 17 00:00:00 2001
From: Adarsh Yoga <adarsh.yoga@intel.com>
Date: Wed, 24 Apr 2024 05:21:19 +0000
Subject: [PATCH 04/10] fix knn failure on windows due to mismatch in types and
 precision

---
 dpbench/benchmarks/default/knn/knn_initialize.py     |  4 +++-
 .../knn/knn_sycl_native_ext/knn_sycl/_knn_kernel.hpp | 12 ++++++------
 .../knn/knn_sycl_native_ext/knn_sycl/_knn_sycl.cpp   |  2 +-
 3 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/dpbench/benchmarks/default/knn/knn_initialize.py b/dpbench/benchmarks/default/knn/knn_initialize.py
index 720b4348..bcab0280 100644
--- a/dpbench/benchmarks/default/knn/knn_initialize.py
+++ b/dpbench/benchmarks/default/knn/knn_initialize.py
@@ -24,7 +24,9 @@ def _gen_data_x(ip_size, data_dim, seed, dtype):
 
     def _gen_data_y(ip_size, classes_num, seed):
         default_rng.seed(seed)
-        data = default_rng.randint(classes_num, size=ip_size)
+        data = default_rng.randint(
+            classes_num, size=ip_size, dtype=types_dict["int"]
+        )
         return data
 
     def _gen_train_data(train_size, data_dim, classes_num, seed_train, dtype):
diff --git a/dpbench/benchmarks/default/knn/knn_sycl_native_ext/knn_sycl/_knn_kernel.hpp b/dpbench/benchmarks/default/knn/knn_sycl_native_ext/knn_sycl/_knn_kernel.hpp
index ac1bcbf3..cd9e2623 100644
--- a/dpbench/benchmarks/default/knn/knn_sycl_native_ext/knn_sycl/_knn_kernel.hpp
+++ b/dpbench/benchmarks/default/knn/knn_sycl_native_ext/knn_sycl/_knn_kernel.hpp
@@ -7,16 +7,16 @@
 
 template <typename FpTy, typename IntTy> class theKernel;
 
-template <typename FpTy> struct neighbors
+template <typename FpTy, typename IntTy> struct neighbors
 {
     FpTy dist;
-    size_t label;
+    IntTy label;
 };
 
 template <typename FpTy, typename IntTy>
 sycl::event knn_impl(sycl::queue q,
                      FpTy *d_train,
-                     size_t *d_train_labels,
+                     IntTy *d_train_labels,
                      FpTy *d_test,
                      size_t k,
                      size_t classes_num,
@@ -33,7 +33,7 @@ sycl::event knn_impl(sycl::queue q,
 
                 // here k has to be 5 in order to match with numpy no. of
                 // neighbors
-                struct neighbors<FpTy> queue_neighbors[5];
+                struct neighbors<FpTy, IntTy> queue_neighbors[5];
 
                 // count distances
                 for (size_t j = 0; j < k; ++j) {
@@ -54,7 +54,7 @@ sycl::event knn_impl(sycl::queue q,
                 for (size_t j = 0; j < k; ++j) {
                     // push queue
                     FpTy new_distance = queue_neighbors[j].dist;
-                    FpTy new_neighbor_label = queue_neighbors[j].label;
+                    IntTy new_neighbor_label = queue_neighbors[j].label;
                     size_t index = j;
                     while (index > 0 &&
                            new_distance < queue_neighbors[index - 1].dist)
@@ -83,7 +83,7 @@ sycl::event knn_impl(sycl::queue q,
 
                         // push queue
                         FpTy new_distance = queue_neighbors[k - 1].dist;
-                        FpTy new_neighbor_label = queue_neighbors[k - 1].label;
+                        IntTy new_neighbor_label = queue_neighbors[k - 1].label;
                         size_t index = k - 1;
 
                         while (index > 0 &&
diff --git a/dpbench/benchmarks/default/knn/knn_sycl_native_ext/knn_sycl/_knn_sycl.cpp b/dpbench/benchmarks/default/knn/knn_sycl_native_ext/knn_sycl/_knn_sycl.cpp
index 97938bb0..43f2b1b9 100644
--- a/dpbench/benchmarks/default/knn/knn_sycl_native_ext/knn_sycl/_knn_sycl.cpp
+++ b/dpbench/benchmarks/default/knn/knn_sycl_native_ext/knn_sycl/_knn_sycl.cpp
@@ -43,7 +43,7 @@ void knn_sync(dpctl::tensor::usm_ndarray x_train,
     if (typenum == UAR_FLOAT) {
         sycl::event res_ev = knn_impl<float, unsigned int>(
             x_train.get_queue(), x_train.get_data<float>(),
-            y_train.get_data<size_t>(), x_test.get_data<float>(), k,
+            y_train.get_data<unsigned int>(), x_test.get_data<float>(), k,
             classes_num, train_size, test_size,
             predictions.get_data<unsigned int>(),
             votes_to_classes.get_data<float>(), data_dim);

From 86056f6550e0cbbd331c98f24fa17183c2f5741e Mon Sep 17 00:00:00 2001
From: Yevhenii Havrylko <yevhenii.havrylko@intel.com>
Date: Tue, 30 Apr 2024 17:40:23 -0400
Subject: [PATCH 05/10] Organize package meta

---
 conda-recipe/meta.yaml | 41 +++++++++++++++--------------------------
 pyproject.toml         | 36 +++++++++++++++++++++---------------
 2 files changed, 36 insertions(+), 41 deletions(-)

diff --git a/conda-recipe/meta.yaml b/conda-recipe/meta.yaml
index b1e19f30..dfde4348 100644
--- a/conda-recipe/meta.yaml
+++ b/conda-recipe/meta.yaml
@@ -2,6 +2,10 @@
 #
 # SPDX-License-Identifier: Apache-2.0
 
+{% set pyproject = load_file_data('pyproject.toml') %}
+{% set py_deps = pyproject.get('project', {}).get('dependencies', []) %}
+{% set py_build_deps = pyproject.get('build-system', {}).get('requires', []) %}
+
 package:
   name: dpbench
   version: {{ GIT_DESCRIBE_TAG }}
@@ -18,36 +22,21 @@ requirements:
   build:
     - {{ compiler('cxx') }}
     - {{ compiler('dpcpp') }}
+    # This is required to get compatible headers with the system installed glibc
     - sysroot_linux-64 >=2.28  # [linux]
   host:
     - python
-    - setuptools
-    - cmake==3.26* # [win]
-    - cmake # [not win]
-    - ninja # [not win]
-    - scikit-build
-    - cython
-    - pybind11
-    - versioneer
-    - intel::numpy
-    - numba
-    - dpctl
-    - dpnp
-    - numba-dpex
+    {% for dep in py_build_deps %}
+      {% if dep.startswith('ninja') %}
+    - {{ dep.split(';')[0] }} # [not win]
+      {% else %}
+    - {{ dep|replace('_','-') }}
+      {% endif %}
+    {% endfor %}
   run:
-    - python
-    - tomli
-    - alembic
-    - sqlalchemy
-    - py-cpuinfo
-    - scipy
-    - scikit-learn
-    - pandas
-    - intel::numpy
-    - numba
-    - dpctl
-    - dpnp
-    - numba-dpex
+    {% for dep in py_deps %}
+    - {{ dep|replace('_','-') }}
+    {% endfor %}
 
 test:
   requires:
diff --git a/pyproject.toml b/pyproject.toml
index cfc33785..5c6c1f2c 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -28,26 +28,25 @@ classifiers = [
 license = { text = "Apache-2.0" }
 requires-python = ">=3.9"
 dependencies = [
-    "tomli",
+    "tomli>=2.0.0",
     "alembic>=1.10.0",
     "sqlalchemy>=2.0.0",
-    "py-cpuinfo",
-    "scipy",
-    "scikit_learn",
-    "pandas",
-    "numpy",
-    "dpctl",
-    "dpnp",
-    "numba",
-    "numba_dpex",
+    "py-cpuinfo>=9.0.0",
+    "scipy>=1.13.0",
+    "scikit_learn>=1.4.0",
+    "pandas>=2.2.0",
+    "numpy>=1.24.0",
+    "dpctl>=0.16.1",
+    "dpnp>=0.14.0",
+    "numba>=0.59.0",
+    "numba_dpex>=0.23.0a0",
 ]
-dynamic =["version"]
+dynamic = ["version"]
 
 [project.optional-dependencies]
 numba-mlir = ["numba_mlir"]
 color-config = ["pygments"]
-sycl-build = ["scikit-build>=0.13", "cmake>=3.18", "ninja"]
-npbench = ["dace","dask","legate"]
+npbench = ["dace", "dask", "legate"]
 json-to-toml = ["tomli_w"]
 expected-failure = ["tomlkit"]
 
@@ -62,8 +61,15 @@ dpbench = "dpbench.console.entry:main"
 # TODO: make it optional for no sycl build. Workaround: `--no-deps`.
 # https://github.com/scikit-build/scikit-build/issues/981
 requires = [
-    "setuptools>=42,<64",
-    "versioneer[toml]",
+    "setuptools>=63.0.0",
+    "scikit-build>=0.17.0",                      # sycl build dep
+    "ninja>=1.11.1; platform_system!='Windows'", # sycl build dep
+    "cmake>=3.29.0",                             # sycl build dep
+    # We need dpctl for UsmNdArray integration for dpcpp code
+    "dpctl>=0.16.1",
+    "pybind11>=2.12.0",
+    # WARNING: check with doc how to upgrade
+    "versioneer[toml]==0.29",
 ]
 build-backend = "setuptools.build_meta"
 

From 26f0ceb04784232148cb547308129a4cb0003f77 Mon Sep 17 00:00:00 2001
From: Yevhenii Havrylko <yevhenii.havrylko@intel.com>
Date: Tue, 30 Apr 2024 18:01:21 -0400
Subject: [PATCH 06/10] Migrate package build to python-build

---
 conda-recipe/bld.bat   | 45 +++++++++++++++++++++++++++++++-----------
 conda-recipe/build.sh  | 34 ++++++++++++++++++++-----------
 conda-recipe/meta.yaml | 19 ++++++++++++------
 pyproject.toml         |  3 +++
 4 files changed, 73 insertions(+), 28 deletions(-)

diff --git a/conda-recipe/bld.bat b/conda-recipe/bld.bat
index f08875ae..8b4ca425 100644
--- a/conda-recipe/bld.bat
+++ b/conda-recipe/bld.bat
@@ -17,8 +17,8 @@ set "DPBENCH_SYCL=1"
 set "CMAKE_GENERATOR=Ninja"
 set "CC=icx"
 set "CXX=icx"
-
-"%PYTHON%" setup.py clean --all
+:: Make CMake verbose
+set "VERBOSE=1"
 
 FOR %%V IN (14.0.0 14 15.0.0 15 16.0.0 16 17.0.0 17) DO @(
   REM set DIR_HINT if directory exists
@@ -41,18 +41,41 @@ if EXIST "%PLATFORM_DIR%" (
   if errorlevel 1 exit 1
 )
 
-@REM TODO: switch to pip build. Currently results in broken binary
-@REM %PYTHON% -m pip install --no-index --no-deps --no-build-isolation . -v
+:: -wnx flags mean: --wheel --no-isolation --skip-dependency-check
+%PYTHON% -m build -w -n -x
+if %ERRORLEVEL% neq 0 exit 1
+
+:: `pip install dist\dpbench*.whl` does not work on windows,
+:: so use a loop; there's only one wheel in dist/ anyway
+for /f %%f in ('dir /b /S .\dist') do (
+    %PYTHON% -m wheel tags --remove --build %GIT_DESCRIBE_NUMBER% %%f
+    if %ERRORLEVEL% neq 0 exit 1
+)
+
+:: wheel file was renamed
+for /f %%f in ('dir /b /S .\dist') do (
+    %PYTHON% -m pip install %%f ^
+      --no-build-isolation ^
+      --no-deps ^
+      --only-binary :all: ^
+      --no-index ^
+      --prefix %PREFIX% ^
+      -vv
+    if %ERRORLEVEL% neq 0 exit 1
+)
+
+:: Must be consistent with pyproject.toml project.scritps. Currently pip does
+:: not allow to ignore scripts installation, so we have to remove them manually.
+:: https://github.com/pypa/pip/issues/3980
+:: We have to let conda-build manage it for use in order to set proper python
+:: path.
+:: https://docs.conda.io/projects/conda-build/en/stable/resources/define-metadata.html#python-entry-points
+rm %PREFIX%\Scripts\dpbench.exe
+
+:: Copy wheel package
 if NOT "%WHEELS_OUTPUT_FOLDER%"=="" (
-    rem Install and assemble wheel package from the build bits
-    "%PYTHON%" setup.py install --single-version-externally-managed --record=record.txt bdist_wheel --build-number %GIT_DESCRIBE_NUMBER%
-    if errorlevel 1 exit 1
     copy dist\dpbench*.whl %WHEELS_OUTPUT_FOLDER%
     if errorlevel 1 exit 1
-) ELSE (
-    rem Only install
-    "%PYTHON%" setup.py install --single-version-externally-managed --record=record.txt
-    if errorlevel 1 exit 1
 )
 
 rem copy back
diff --git a/conda-recipe/build.sh b/conda-recipe/build.sh
index 544ce140..c06023cb 100644
--- a/conda-recipe/build.sh
+++ b/conda-recipe/build.sh
@@ -16,18 +16,30 @@ export CMAKE_GENERATOR="Ninja"
 export CC=icx
 export CXX=icpx
 
-if [ -e "_skbuild" ]; then
-    ${PYTHON} setup.py clean --all
-fi
-
 # TODO: switch to pip build. Currently results in broken binary on Windows
 # $PYTHON -m pip install --no-index --no-deps --no-build-isolation . -v
 
-# Build wheel package
-if [ -n "${WHEELS_OUTPUT_FOLDER}" ]; then
-    $PYTHON setup.py install --single-version-externally-managed --record=record.txt bdist_wheel -p manylinux2014_x86_64 --build-number $GIT_DESCRIBE_NUMBER
-    mkdir -p ${WHEELS_OUTPUT_FOLDER}
-    cp dist/dpbench*.whl ${WHEELS_OUTPUT_FOLDER}
-else
-    $PYTHON setup.py install --single-version-externally-managed --record=record.txt
+# -wnx flags mean: --wheel --no-isolation --skip-dependency-check
+${PYTHON} -m build -w -n -x
+${PYTHON} -m wheel tags --remove --build "$GIT_DESCRIBE_NUMBER" \
+    --platform-tag manylinux2014_x86_64 dist/dpbench*.whl
+${PYTHON} -m pip install dist/dpbench*.whl \
+    --no-build-isolation \
+    --no-deps \
+    --only-binary :all: \
+    --no-index \
+    --prefix ${PREFIX} \
+    -vv
+
+# Must be consistent with pyproject.toml project.scritps. Currently pip does
+# not allow to ignore scripts installation, so we have to remove them manually.
+# https://github.com/pypa/pip/issues/3980
+# We have to let conda-build manage it for use in order to set proper python
+# path.
+# https://docs.conda.io/projects/conda-build/en/stable/resources/define-metadata.html#python-entry-points
+rm ${PREFIX}/bin/dpbench
+
+# Copy wheel package
+if [[ -v WHEELS_OUTPUT_FOLDER ]]; then
+    cp dist/dpbench*.whl "${WHEELS_OUTPUT_FOLDER[@]}"
 fi
diff --git a/conda-recipe/meta.yaml b/conda-recipe/meta.yaml
index dfde4348..1c8561f3 100644
--- a/conda-recipe/meta.yaml
+++ b/conda-recipe/meta.yaml
@@ -5,6 +5,7 @@
 {% set pyproject = load_file_data('pyproject.toml') %}
 {% set py_deps = pyproject.get('project', {}).get('dependencies', []) %}
 {% set py_build_deps = pyproject.get('build-system', {}).get('requires', []) %}
+{% set project_scripts = pyproject.get('project', {}).get('scripts', {}) %}
 
 package:
   name: dpbench
@@ -17,6 +18,10 @@ build:
     number: {{ GIT_DESCRIBE_NUMBER }}
     script_env:
         - WHEELS_OUTPUT_FOLDER
+    entry_points:
+        {% for script, module in project_scripts | dictsort %}
+        - {{ script ~ " = " ~ module }}
+        {% endfor %}
 
 requirements:
   build:
@@ -26,9 +31,15 @@ requirements:
     - sysroot_linux-64 >=2.28  # [linux]
   host:
     - python
+    - pip
     {% for dep in py_build_deps %}
       {% if dep.startswith('ninja') %}
     - {{ dep.split(';')[0] }} # [not win]
+      {% elif dep.startswith('cmake') %}
+    - cmake=3.26 # [win]
+    - {{ dep }} # [not win]
+      {% elif dep.startswith('build>=') %}
+    - {{ 'python-' ~ dep }}
       {% else %}
     - {{ dep|replace('_','-') }}
       {% endif %}
@@ -39,12 +50,8 @@ requirements:
     {% endfor %}
 
 test:
-  requires:
-    - dpctl
-    - dpnp
-    - numba-dpex
-    - numba
-    - numpy
+  commands:
+    - dpbench --help
 
 about:
   home: https://github.com/IntelPython/dpbench
diff --git a/pyproject.toml b/pyproject.toml
index 5c6c1f2c..dded7fec 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -55,12 +55,15 @@ expected-failure = ["tomlkit"]
 "Homepage" = "https://https://github.com/IntelPython/dpbench"
 
 [project.scripts]
+# Keep consistent with conda build scripts
 dpbench = "dpbench.console.entry:main"
 
 [build-system]
 # TODO: make it optional for no sycl build. Workaround: `--no-deps`.
 # https://github.com/scikit-build/scikit-build/issues/981
 requires = [
+    "wheel>=0.43",
+    "build>=1.1",
     "setuptools>=63.0.0",
     "scikit-build>=0.17.0",                      # sycl build dep
     "ninja>=1.11.1; platform_system!='Windows'", # sycl build dep

From 6542b52919e4a6611975430316f75fd15f933965 Mon Sep 17 00:00:00 2001
From: Yevhenii Havrylko <yevhenii.havrylko@intel.com>
Date: Wed, 1 May 2024 16:10:20 -0400
Subject: [PATCH 07/10] Use conda build over mambabuild

---
 .github/workflows/build_and_run.yml |  1 -
 .github/workflows/conda-package.yml | 29 ++++++++++++++---------------
 2 files changed, 14 insertions(+), 16 deletions(-)

diff --git a/.github/workflows/build_and_run.yml b/.github/workflows/build_and_run.yml
index 959092aa..b2d90013 100644
--- a/.github/workflows/build_and_run.yml
+++ b/.github/workflows/build_and_run.yml
@@ -77,7 +77,6 @@ jobs:
         uses: conda-incubator/setup-miniconda@v2
         with:
           python-version: ${{ matrix.python }}
-          miniforge-variant: Mambaforge
           miniforge-version: latest
           activate-environment: "build"
           channel-priority: "disabled"
diff --git a/.github/workflows/conda-package.yml b/.github/workflows/conda-package.yml
index e11d8430..15e0960d 100644
--- a/.github/workflows/conda-package.yml
+++ b/.github/workflows/conda-package.yml
@@ -55,15 +55,13 @@ jobs:
           fetch-depth: 0
 
       - name: Setup miniconda
-        uses: conda-incubator/setup-miniconda@v2
+        uses: conda-incubator/setup-miniconda@v3
         with:
-          python-version: ${{ matrix.python }}
-          miniforge-variant: Mambaforge
-          miniforge-version: latest
-          activate-environment: "build"
+          auto-activate-base: true
+          activate-environment: ""
           channels: ${{ env.CHANNELS }}
           channel-priority: "disabled"
-          run-post: false
+          miniforge-version: latest
 
       - name: Disable defautls
         run: conda config --remove channels defaults
@@ -74,9 +72,8 @@ jobs:
           echo "CONDA_BLD=$CONDA_PREFIX/conda-bld/${{ runner.os == 'Linux' && 'linux' || 'win' }}-64/" | tr "\\" '/' >> $GITHUB_ENV
           echo "WHEELS_OUTPUT_FOLDER=$GITHUB_WORKSPACE${{ runner.os == 'Linux' && '/' || '\\' }}" >> $GITHUB_ENV
 
-      # boa is an extention to conda so we can use mamba resolver in conda build
       - name: Install conda-build
-        run: mamba install boa
+        run: conda install conda-build
 
       - name: Configure MSBuild
         if: runner.os == 'Windows'
@@ -84,8 +81,12 @@ jobs:
         with:
           vs-version: '14.35'
 
+      - name: Show conda-rc
+        shell: bash -el {0}
+        run: cat ~/.condarc
+
       - name: Build conda package
-        run: conda mambabuild --no-test --python ${{ matrix.python }} conda-recipe
+        run: conda build --python ${{ matrix.python }} conda-recipe
 
       - name: Upload artifact
         uses: actions/upload-artifact@v3.1.2
@@ -122,15 +123,13 @@ jobs:
 
     steps:
       - name: Setup miniconda
-        uses: conda-incubator/setup-miniconda@v2
+        uses: conda-incubator/setup-miniconda@v3
         with:
           python-version: ${{ matrix.python }}
-          miniforge-variant: Mambaforge
-          miniforge-version: latest
           activate-environment: "build"
           channels: ${{ env.CHANNELS }}
           channel-priority: "disabled"
-          run-post: false
+          miniforge-version: latest
 
       - name: Disable defautls
         run: conda config --remove channels defaults
@@ -159,7 +158,7 @@ jobs:
 
       # Needed to be able to run conda index
       - name: Install conda-build
-        run: mamba install conda-build conda-index
+        run: conda install conda-build conda-index
 
       - name: Create conda channel
         run: python -m conda_index ${{ env.CHANNEL_PATH }}
@@ -170,7 +169,7 @@ jobs:
           cat ${{ env.VER_JSON_PATH }}
 
       - name: Install dpbench
-        run: mamba install ${{ env.PACKAGE_NAME }}=${{ env.PACKAGE_VERSION }} pytest intel::intel-opencl-rt python=${{ matrix.python }} -c ${{ env.CHANNEL_PATH }}
+        run: conda install ${{ env.PACKAGE_NAME }}=${{ env.PACKAGE_VERSION }} pytest intel::intel-opencl-rt python=${{ matrix.python }} -c ${{ env.CHANNEL_PATH }}
 
       - name: Setup OpenCL CPU device
         if: runner.os == 'Windows'

From e1bebacad3169fc4dcf4abe0f50b3b32878cc4e1 Mon Sep 17 00:00:00 2001
From: Yevhenii Havrylko <yevhenii.havrylko@intel.com>
Date: Thu, 2 May 2024 18:19:46 -0400
Subject: [PATCH 08/10] Fix dpbench conda dep on python

---
 conda-recipe/meta.yaml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/conda-recipe/meta.yaml b/conda-recipe/meta.yaml
index 1c8561f3..055a9c25 100644
--- a/conda-recipe/meta.yaml
+++ b/conda-recipe/meta.yaml
@@ -45,6 +45,7 @@ requirements:
       {% endif %}
     {% endfor %}
   run:
+    - python
     {% for dep in py_deps %}
     - {{ dep|replace('_','-') }}
     {% endfor %}

From ee1357c7ee9b8ae4fe49ea9fe4b8caf2049816a7 Mon Sep 17 00:00:00 2001
From: Yevhenii Havrylko <yevhenii.havrylko@intel.com>
Date: Fri, 3 May 2024 11:07:37 -0400
Subject: [PATCH 09/10] Use experimental flag with nopybench & npbench

---
 .github/workflows/conda-package.yml |  4 +--
 README.md                           | 38 +++++++++++++++--------------
 dpbench/console/run.py              |  9 ++++---
 3 files changed, 28 insertions(+), 23 deletions(-)

diff --git a/.github/workflows/conda-package.yml b/.github/workflows/conda-package.yml
index 15e0960d..abadd19e 100644
--- a/.github/workflows/conda-package.yml
+++ b/.github/workflows/conda-package.yml
@@ -196,11 +196,11 @@ jobs:
       # we want to make sure that configuration files are geting populated
       - name: Run npbench benchmark
         run: |
-          dpbench -i numpy -b azimint_hist run --npbench --precision=${{matrix.precision}}
+          dpbench -i numpy -b azimint_hist run --experimental-npbench --precision=${{matrix.precision}}
 
       - name: Run rodinia benchmark
         run: |
-          dpbench run --rodinia --no-dpbench --no-validate -r 1 --precision=${{matrix.precision}}
+          dpbench run --experimental-rodinia --no-dpbench --no-validate -r 1 --precision=${{matrix.precision}}
 
   upload_anaconda:
     name: Upload dppy/label/dev ['${{ matrix.os }}', python='${{ matrix.python }}']
diff --git a/README.md b/README.md
index 7adde796..2bc8489e 100644
--- a/README.md
+++ b/README.md
@@ -117,9 +117,9 @@ SPDX-License-Identifier: Apache-2.0
 4. All available options are available using `dpbench --help` and `dpbench <command> --help`:
 
     ```
-    usage: dpbench [-h] [-b [BENCHMARKS]] [-i [IMPLEMENTATIONS]] [-a | --all-implementations | --no-all-implementations] [--version] [-r [RUN_ID]] [--last-run | --no-last-run]
-                [-d [RESULTS_DB]] [--log-level [{critical,fatal,error,warning,info,debug}]]
-                {run,report,config} ...
+    usage: dpbench [-h] [-b [BENCHMARKS]] [-i [IMPLEMENTATIONS]] [-a | --all-implementations | --no-all-implementations] [--version] [-r [RUN_ID]] [--last-run | --no-last-run] [-d [RESULTS_DB]]
+               [--log-level [{critical,fatal,error,warning,info,debug}]]
+               {run,report,config} ...
 
     positional arguments:
     {run,report,config}
@@ -131,12 +131,12 @@ SPDX-License-Identifier: Apache-2.0
     -i [IMPLEMENTATIONS], --implementations [IMPLEMENTATIONS]
                             Comma separated list of implementations. Use --all-implementations to load all available implementations.
     -a, --all-implementations, --no-all-implementations
-                            If set, all available implementations will be loaded. (default: False)
+                            If set, all available implementations will be loaded.
     --version             show program's version number and exit
     -r [RUN_ID], --run-id [RUN_ID]
                             run_id to perform actions on. Use --last-run to use latest available run, or leave empty to create new one.
     --last-run, --no-last-run
-                            Sets run_id to the latest run_id from the database. (default: False)
+                            Sets run_id to the latest run_id from the database.
     -d [RESULTS_DB], --results-db [RESULTS_DB]
                             Path to a database to store results.
     --log-level [{critical,fatal,error,warning,info,debug}]
@@ -144,24 +144,26 @@ SPDX-License-Identifier: Apache-2.0
     ```
 
     ```
-    usage: dpbench run [-h] [-p [{S,M,L}]] [-s | --validate | --no-validate] [--dpbench | --no-dpbench] [--npbench | --no-npbench] [--polybench | --no-polybench] [-r [REPEAT]] [-t [TIMEOUT]]
-                    [--precision [{single,double}]] [--print-results | --no-print-results] [--save | --no-save] [--sycl-device [SYCL_DEVICE]]
-                    [--skip-expected-failures | --no-skip-expected-failures]
+    usage: dpbench run [-h] [-p [{S,M16Gb,M,L}]] [-s | --validate | --no-validate] [--dpbench | --no-dpbench] [--experimental-npbench | --no-experimental-npbench] [--experimental-polybench | --no-experimental-polybench]
+                   [--experimental-rodinia | --no-experimental-rodinia] [-r [REPEAT]] [-t [TIMEOUT]] [--precision [{single,double}]] [--print-results | --no-print-results] [--save | --no-save] [--sycl-device [SYCL_DEVICE]]
+                   [--skip-expected-failures | --no-skip-expected-failures]
 
     Subcommand to run benchmark executions.
 
     options:
     -h, --help            show this help message and exit
-    -p [{S,M,L}], --preset [{S,M,L}]
+    -p [{S,M16Gb,M,L}], --preset [{S,M16Gb,M,L}]
                             Preset to use for benchmark execution.
     -s, --validate, --no-validate
-                            Set if the validation will be run for each benchmark. (default: True)
+                            Set if the validation will be run for each benchmark.
     --dpbench, --no-dpbench
-                            Set if run dpbench benchmarks. (default: True)
-    --npbench, --no-npbench
-                            Set if run npbench benchmarks. (default: False)
-    --polybench, --no-polybench
-                            Set if run polybench benchmarks. (default: False)
+                            Set if run dpbench benchmarks.
+    --experimental-npbench, --no-experimental-npbench
+                            Set if run npbench benchmarks.
+    --experimental-polybench, --no-experimental-polybench
+                            Set if run polybench benchmarks.
+    --experimental-rodinia, --no-experimental-rodinia
+                            Set if run rodinia benchmarks.
     -r [REPEAT], --repeat [REPEAT]
                             Number of repeats for each benchmark.
     -t [TIMEOUT], --timeout [TIMEOUT]
@@ -169,12 +171,12 @@ SPDX-License-Identifier: Apache-2.0
     --precision [{single,double}]
                             Data precision to use for array initialization.
     --print-results, --no-print-results
-                            Show the result summary or not (default: True)
-    --save, --no-save     Either to save execution into database. (default: True)
+                            Show the result summary or not
+    --save, --no-save     Either to save execution into database.
     --sycl-device [SYCL_DEVICE]
                             Sycl device to overwrite for framework configurations.
     --skip-expected-failures, --no-skip-expected-failures
-                            Either to save execution into database. (default: True)
+                            Either to save execution into database.
     ```
 
     ```
diff --git a/dpbench/console/run.py b/dpbench/console/run.py
index 9e44c8b8..da4bee9c 100644
--- a/dpbench/console/run.py
+++ b/dpbench/console/run.py
@@ -47,19 +47,22 @@ def add_run_arguments(parser: argparse.ArgumentParser):
         help="Set if run dpbench benchmarks.",
     )
     parser.add_argument(
-        "--npbench",
+        "--experimental-npbench",
+        dest="npbench",
         action=argparse.BooleanOptionalAction,
         default=False,
         help="Set if run npbench benchmarks.",
     )
     parser.add_argument(
-        "--polybench",
+        "--experimental-polybench",
+        dest="polybench",
         action=argparse.BooleanOptionalAction,
         default=False,
         help="Set if run polybench benchmarks.",
     )
     parser.add_argument(
-        "--rodinia",
+        "--experimental-rodinia",
+        dest="rodinia",
         action=argparse.BooleanOptionalAction,
         default=False,
         help="Set if run rodinia benchmarks.",

From 583ee26d690f96603bd87d07df18b132ca8818f0 Mon Sep 17 00:00:00 2001
From: Yevhenii Havrylko <yevhenii.havrylko@intel.com>
Date: Fri, 3 May 2024 12:08:40 -0400
Subject: [PATCH 10/10] Add required packages to frameworks

---
 dpbench/config/reader.py                          | 13 ++++++++++++-
 .../infrastructure/frameworks/cupy_framework.py   |  4 ++++
 .../infrastructure/frameworks/dpnp_framework.py   |  4 ++++
 dpbench/infrastructure/frameworks/fabric.py       |  7 ++++++-
 dpbench/infrastructure/frameworks/framework.py    | 15 +++++++++++++++
 .../frameworks/numba_cuda_framework.py            |  4 ++++
 .../frameworks/numba_dpex_framework.py            |  4 ++++
 .../infrastructure/frameworks/numba_framework.py  |  4 ++++
 .../frameworks/numba_mlir_framework.py            |  4 ++++
 9 files changed, 57 insertions(+), 2 deletions(-)

diff --git a/dpbench/config/reader.py b/dpbench/config/reader.py
index 4a37aedc..d89cf91b 100644
--- a/dpbench/config/reader.py
+++ b/dpbench/config/reader.py
@@ -13,6 +13,8 @@
 
 import tomli
 
+from dpbench.infrastructure.frameworks.fabric import get_framework_class
+
 from .benchmark import Benchmark, BenchmarkImplementation, Presets
 from .config import Config
 from .framework import Framework
@@ -117,7 +119,7 @@ def read_configs(  # noqa: C901: TODO: move modules into config
     for framework in config.frameworks:
         config.implementations += framework.postfixes
 
-    if implementations is None:
+    if implementations is None or len(implementations) == 0:
         implementations = {impl.postfix for impl in config.implementations}
 
     if load_implementations:
@@ -228,6 +230,15 @@ def read_frameworks(
         if len(framework.postfixes) == 0:
             continue
 
+        cls = get_framework_class(framework)
+        unavailable_pkgs = cls.get_missing_required_packages()
+        if len(unavailable_pkgs) > 0:
+            logging.warning(
+                f"Framework {framework.simple_name} unavailable "
+                + f"due to missing packages {unavailable_pkgs}"
+            )
+            continue
+
         config.frameworks.append(framework)
 
 
diff --git a/dpbench/infrastructure/frameworks/cupy_framework.py b/dpbench/infrastructure/frameworks/cupy_framework.py
index e1c781a9..273995d6 100644
--- a/dpbench/infrastructure/frameworks/cupy_framework.py
+++ b/dpbench/infrastructure/frameworks/cupy_framework.py
@@ -19,6 +19,10 @@ def __init__(self, fname: str = None, config: cfg.Framework = None):
 
         super().__init__(fname, config)
 
+    @staticmethod
+    def required_packages() -> list[str]:
+        return ["cupy"]
+
     def copy_to_func(self) -> Callable:
         """Returns the copy-method that should be used
         for copying the benchmark arguments."""
diff --git a/dpbench/infrastructure/frameworks/dpnp_framework.py b/dpbench/infrastructure/frameworks/dpnp_framework.py
index 459e3ff0..5071cb63 100644
--- a/dpbench/infrastructure/frameworks/dpnp_framework.py
+++ b/dpbench/infrastructure/frameworks/dpnp_framework.py
@@ -35,6 +35,10 @@ def __init__(self, fname: str = None, config: cfg.Framework = None):
             )
             raise sdce
 
+    @staticmethod
+    def required_packages() -> list[str]:
+        return ["dpnp"]
+
     def device_filter_string(self) -> str:
         """Returns the sycl device's filter string if the framework has an
         associated sycl device."""
diff --git a/dpbench/infrastructure/frameworks/fabric.py b/dpbench/infrastructure/frameworks/fabric.py
index 0cdc8202..527f3fdb 100644
--- a/dpbench/infrastructure/frameworks/fabric.py
+++ b/dpbench/infrastructure/frameworks/fabric.py
@@ -39,7 +39,7 @@ def build_framework_map() -> dict[str, Framework]:
     return result
 
 
-def build_framework(framework_config: cfg.Framework) -> Framework:
+def get_framework_class(framework_config: cfg.Framework) -> Framework:
     available_classes = [
         Framework,
         DpcppFramework,
@@ -61,4 +61,9 @@ def build_framework(framework_config: cfg.Framework) -> Framework:
         )
         constructor = Framework
 
+    return constructor
+
+
+def build_framework(framework_config: cfg.Framework) -> Framework:
+    constructor = get_framework_class(framework_config)
     return constructor(config=framework_config)
diff --git a/dpbench/infrastructure/frameworks/framework.py b/dpbench/infrastructure/frameworks/framework.py
index d0bd5887..757e94da 100644
--- a/dpbench/infrastructure/frameworks/framework.py
+++ b/dpbench/infrastructure/frameworks/framework.py
@@ -5,6 +5,7 @@
 # SPDX-License-Identifier: BSD-3-Clause
 
 import logging
+from importlib.util import find_spec
 from typing import Any, Callable, Dict, final
 
 import pkg_resources
@@ -45,6 +46,20 @@ def __init__(
 
         self.device_info = cpuinfo.get_cpu_info().get("brand_raw")
 
+    @staticmethod
+    def required_packages() -> list[str]:
+        return []
+
+    @classmethod
+    def get_missing_required_packages(cls) -> None:
+        unavailable_packages = []
+        for pkg in cls.required_packages():
+            spec = find_spec(pkg)
+            if spec is None:
+                unavailable_packages.append(pkg)
+
+        return unavailable_packages
+
     def device_filter_string(self) -> str:
         """Returns the sycl device's filter string if the framework has an
         associated sycl device."""
diff --git a/dpbench/infrastructure/frameworks/numba_cuda_framework.py b/dpbench/infrastructure/frameworks/numba_cuda_framework.py
index 5c5940d0..32fb5dfc 100644
--- a/dpbench/infrastructure/frameworks/numba_cuda_framework.py
+++ b/dpbench/infrastructure/frameworks/numba_cuda_framework.py
@@ -19,6 +19,10 @@ def __init__(self, fname: str = None, config: cfg.Framework = None):
 
         super().__init__(fname, config)
 
+    @staticmethod
+    def required_packages() -> list[str]:
+        return ["cupy"]
+
     def copy_to_func(self) -> Callable:
         """Returns the copy-method that should be used
         for copying the benchmark arguments."""
diff --git a/dpbench/infrastructure/frameworks/numba_dpex_framework.py b/dpbench/infrastructure/frameworks/numba_dpex_framework.py
index e1cdb5b5..d6868a59 100644
--- a/dpbench/infrastructure/frameworks/numba_dpex_framework.py
+++ b/dpbench/infrastructure/frameworks/numba_dpex_framework.py
@@ -33,6 +33,10 @@ def __init__(self, fname: str = None, config: cfg.Framework = None):
             )
             raise sdce
 
+    @staticmethod
+    def required_packages() -> list[str]:
+        return ["numba_dpex"]
+
     def device_filter_string(self) -> str:
         """Returns the sycl device's filter string if the framework has an
         associated sycl device."""
diff --git a/dpbench/infrastructure/frameworks/numba_framework.py b/dpbench/infrastructure/frameworks/numba_framework.py
index 12180d1d..7eeadb0d 100644
--- a/dpbench/infrastructure/frameworks/numba_framework.py
+++ b/dpbench/infrastructure/frameworks/numba_framework.py
@@ -27,3 +27,7 @@ def __init__(self, fname: str = None, config: cfg.Framework = None):
         """
 
         super().__init__(fname, config)
+
+    @staticmethod
+    def required_packages() -> list[str]:
+        return ["numba"]
diff --git a/dpbench/infrastructure/frameworks/numba_mlir_framework.py b/dpbench/infrastructure/frameworks/numba_mlir_framework.py
index c2a87170..8dfd9305 100644
--- a/dpbench/infrastructure/frameworks/numba_mlir_framework.py
+++ b/dpbench/infrastructure/frameworks/numba_mlir_framework.py
@@ -28,6 +28,10 @@ def __init__(self, fname: str = None, config: cfg.Framework = None):
 
             self.device_info = dpctl.SyclDevice(self.sycl_device).name
 
+    @staticmethod
+    def required_packages() -> list[str]:
+        return ["numba_mlir"]
+
     def copy_to_func(self) -> Callable:
         """Returns the copy-method that should be used
         for copying the benchmark arguments to device."""