Merge pull request #23 from howetuft/work

Work

Author: howetuft

.github/workflows/wheels.yml

@@ -28,6 +28,8 @@ jobs:
         #os: [ubuntu-latest]
         #python-minor: [ 11]
         #os: [windows-latest]
+        #python-minor: [ 11]
+        #os: [macos-13]
         exclude:
           - os: macos-14
             python-minor: 8
@@ -93,6 +95,13 @@ jobs:
             core.exportVariable('CCACHE_DEBUGDIR', String.raw`${{ github.workspace }}/ccache-debug`);
             core.exportVariable('CCACHE_LOGFILE', String.raw`${{ github.workspace }}/ccache.log`);
 
+
+      - name: Update apt (Linux)
+        if: runner.os == 'Linux'
+        shell: bash
+        run: |
+          sudo apt-get update -y
+
       - name: ccache
         uses: hendrikmuhs/[email protected]
         with:
@@ -219,9 +228,10 @@ jobs:
           CIBW_REPAIR_WHEEL_COMMAND_LINUX: |
             TBB=/project/full_deploy/host/onetbb/$TBB_VERSION/Release/x86_64/lib
             OIDN=/project/full_deploy/host/oidn/$OIDN_VERSION/Release/x86_64/lib
+            EMBREE3=/project/full_deploy/host/embree3/$EMBREE3_VERSION/Release/x86_64/lib
             HWLOC=/project/full_deploy/host/hwloc/2.9.3/Release/x86_64/lib
             LEVEL_ZERO=/project/full_deploy/host/level-zero/1.17.39/Release/x86_64/lib
-            LD_LIBRARY_PATH=$TBB:$HWLOC:$LEVEL_ZERO:$OIDN:$LD_LIBRARY_PATH
+            LD_LIBRARY_PATH=$TBB:$HWLOC:$LEVEL_ZERO:$OIDN:$EMBREE3:$LD_LIBRARY_PATH
             echo "LD_LIBRARY_PATH=${LD_LIBRARY_PATH}"
             export LD_LIBRARY_PATH
 

.gitignore

@@ -6,6 +6,12 @@ __pycache__/
 # C extensions
 *.so
 
+# ccache
+.ccache/
+
+# conan-cache content
+conan-cache/*
+
 # Distribution / packaging
 .Python
 build/

conan-profiles/conan-profile-Linux-X64

@@ -8,9 +8,12 @@ compiler.version=12
 compiler.cppstd={{ os.getenv("CXX_VERSION") }}
 compiler.libcxx=libstdc++11
 
-[conf]
-openvdb/*:tools.cmake.cmaketoolchain:extra_variables={ 'USE_CCACHE':'OFF' }
-
 [options]
 openvdb/*:simd=AVX
+embree3/*:avx2=True
 embree3/*:avx512=True
+
+[conf]
+tools.build:cflags+=["-ftree-vectorize", "-march=x86-64-v3", "-O3"]
+tools.build:cxxflags+=["-ftree-vectorize", "-march=x86-64-v3", "-O3"]
+openvdb/*:tools.cmake.cmaketoolchain:extra_variables={ 'USE_CCACHE':'OFF' }

conan-profiles/conan-profile-Windows-X64

@@ -10,8 +10,9 @@ compiler.runtime=dynamic
 
 [conf]
 tools.env.virtualenv:powershell=False
-tools.build:cxxflags=["/openmp", "/MP", "/DEBUG:NONE"]
+tools.build:cxxflags=["/openmp", "/MP", "/DEBUG:NONE", "/O2", "/arch:AVX"]
+embree3/*:tools.build:cxxflags=["/openmp", "/MP", "/DEBUG:NONE", "/O2"]
 
 [options]
 openvdb/*:simd=AVX
-embree3/*:avx=True
+# NB: MSVC does not support avx2 nor avx512

conan-profiles/conan-profile-common

@@ -16,3 +16,45 @@ yasm/*
 
 [options]
 oidn/*:shared=True
+embree3/*:shared=True
+embree3/*:with_tbb=True
+embree3/*:sse2=True
+embree3/*:sse42=True
+embree3/*:avx=True
+
+
+# boost/*:without_atomic=True
+boost/*:without_charconv=True
+# boost/*:without_chrono=True
+boost/*:without_cobalt=True
+# boost/*:without_container=True
+boost/*:without_context=True
+boost/*:without_contract=True
+boost/*:without_coroutine=True
+# boost/*:without_date_time=True
+# boost/*:without_exception=True
+boost/*:without_fiber=True
+# boost/*:without_filesystem=True
+# boost/*:without_graph=True
+boost/*:without_graph_parallel=True
+# boost/*:without_iostreams=True
+boost/*:without_json=True
+boost/*:without_locale=True
+boost/*:without_log=True
+# boost/*:without_math=True
+boost/*:without_mpi=True
+boost/*:without_nowide=True
+boost/*:without_process=True
+# boost/*:without_program_options=True
+boost/*:without_python=True
+# boost/*:without_random=True
+# boost/*:without_regex=True
+# boost/*:without_serialization=True
+boost/*:without_stacktrace=True
+# boost/*:without_system=True
+boost/*:without_test=True
+# boost/*:without_thread=True
+boost/*:without_timer=True
+boost/*:without_type_erasure=True
+boost/*:without_url=True
+boost/*:without_wave=True

conan-profiles/conan-profile-macOS-ARM64

@@ -30,10 +30,15 @@ compiler.libcxx=libc++
 [options]
 minizip-ng/*:with_zlib=True
 minizip-ng/*:with_libcomp=False
+embree3/*:sse2=False
+embree3/*:sse42=False
+embree3/*:avx=False
+embree3/*:avx2=False
 embree3/*:neon2x=True
 
 {% endif %}
 
 [buildenv]
 CXX=clang++
 CC=clang
+&:tools.build:cxxflags=["-O3"]

conan-profiles/conan-profile-macOS-X64

@@ -13,9 +13,11 @@ compiler.libcxx=libc++
 minizip-ng/*:with_zlib=True
 minizip-ng/*:with_libcomp=False
 openvdb/*:simd=AVX
-embree3/*:avx512=True
-embree3/*:shared=True
 
 [buildenv]
 CXX=clang++
 CC=clang
+
+[conf]
+&:tools.build:cxxflags+=["-ftree-vectorize", "-march=x86-64-v3", "-O3", "-mno-avx2"]
+# NB: Github MacOS X64 runner does not support avx2 nor avx512

doc/notes.md

@@ -91,7 +91,12 @@ author).
 tried to have it minimal and it's 10.13. For the delocation to succeed, we also
 have to set `MACOSX_DEPLOYMENT_TARGET` environment variable to 10.13.
 
+## Vectorization
+For Intel, we stick to x86-64-v3 (Haswell, 2013), no more. For all platform
+except MacOS Intel, we rely on "-O" option.
+For MacOS Intel: Github runner is not standard, we have to deactivate avx2
+extensions. Otherwise `pyluxcore.so` cannot be tested ("Illegal instruction").
+See https://github.com/ggerganov/whisper.cpp/issues/358 and `sysctl -a
+machdep.cpu`
 
 
-we 
-

pyproject.toml

@@ -8,7 +8,7 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "pyluxcore"
-version = "2.9a1.post12"
+version = "2.9a1.post13"
 dependencies = [
   "numpy >= 2; python_version > '3.8'",
   "numpy < 2; python_version <= '3.8'",

scripts/repair_wheel_windows.sh

@@ -27,6 +27,7 @@ delvewheel repair -v \
   --add-path="${redist_paths}" \
   --add-path="C:/Users/runneradmin/full_deploy/host/onetbb/${TBB_VERSION}/Release/x86_64/bin" \
   --add-path="C:/Users/runneradmin/full_deploy/host/oidn/${OIDN_VERSION}/Release/x86_64/bin" \
+  --add-path="C:/Users/runneradmin/full_deploy/host/embree3/${EMBREE3_VERSION}/Release/x86_64/bin" \
   -w "${dest_dir}" \
   "${wheel}"