misc/py-pytorch: New port: PyTorch: Tensors and dynamic neural networks in Python

misc/Makefile

@@ -450,6 +450,7 @@
     SUBDIR += py-pyprind
     SUBDIR += py-python-geohash
     SUBDIR += py-python-utils
+    SUBDIR += py-pytorch
     SUBDIR += py-qiskit-machine-learning
     SUBDIR += py-scikit-fusion
     SUBDIR += py-serverfiles

misc/py-pytorch/Makefile

@@ -0,0 +1,57 @@
+PORTNAME=	pytorch
+DISTVERSIONPREFIX=	v
+DISTVERSION=	2.0.0
+CATEGORIES=	misc # machine-learning
+MASTER_SITES=	https://github.com/pytorch/pytorch/releases/download/v${DISTVERSION}/
+PKGNAMEPREFIX=	${PYTHON_PKGNAMEPREFIX}
+DIST_SUBDIR=	${PORTNAME}
+
+MAINTAINER=	yuri@FreeBSD.org
+COMMENT=	PyTorch: Tensors and dynamic neural networks in Python
+WWW=		https://pytorch.org/
+
+LICENSE=	BSD3CLAUSE
+LICENSE_FILE=	${WRKSRC}/LICENSE
+
+BUILD_DEPENDS=	cmake:devel/cmake-core \
+		gmake:devel/gmake \
+		pybind11>0:devel/pybind11 \
+		${LOCALBASE}/include/fxdiv.h:devel/fxdiv \
+		${PYTHON_PKGNAMEPREFIX}typing-extensions>0:devel/py-typing-extensions@${PY_FLAVOR} \
+		${PYTHON_PKGNAMEPREFIX}yaml>0:devel/py-yaml@${PY_FLAVOR}
+LIB_DEPENDS=	libopenblas.so:math/openblas \
+		libmpi.so:net/openmpi \
+		libonnx.so:misc/onnx \
+		libpthreadpool.so:devel/pthreadpool \
+		libprotobuf.so:devel/protobuf \
+		libsleef.so:math/sleef
+RUN_DEPENDS=	${PYTHON_PKGNAMEPREFIX}filelock>0:sysutils/py-filelock@${PY_FLAVOR} \
+		${PYTHON_PKGNAMEPREFIX}Jinja2>=0:devel/py-Jinja2@${PY_FLAVOR} \
+		${PYTHON_PKGNAMEPREFIX}networkx>0:math/py-networkx@${PY_FLAVOR} \
+		${PYTHON_PKGNAMEPREFIX}sympy>0:math/py-sympy@${PY_FLAVOR} \
+		${PYTHON_PKGNAMEPREFIX}typing-extensions>0:devel/py-typing-extensions@${PY_FLAVOR}
+RUN_DEPENDS+=	${PYTHON_PKGNAMEPREFIX}dill>0:devel/py-dill@${PY_FLAVOR} # optional dependency
+
+USES=		compiler:c++14-lang localbase:ldflags python
+USE_PYTHON=	distutils autoplist
+
+MAKE_ENV=	USE_NINJA=no # ninja breaks for some reason
+MAKE_ENV+=	BUILD_TEST=0 # ninja breaks for some reason
+LDFLAGS+=	-lexecinfo
+
+BINARY_ALIAS=	make=${GMAKE}
+
+POST_PLIST=	fix-plist
+
+post-install: # strip binaries
+	@${STRIP_CMD} \
+		${STAGEDIR}${PYTHON_SITELIBDIR}/torch/bin/torch_shm_manager \
+		${STAGEDIR}${PYTHON_SITELIBDIR}/torch/_C${PYTHON_EXT_SUFFIX}.so \
+		${STAGEDIR}${PYTHON_SITELIBDIR}/torch/_C_flatbuffer${PYTHON_EXT_SUFFIX}.so \
+		${STAGEDIR}${PYTHON_SITELIBDIR}/functorch/_C${PYTHON_EXT_SUFFIX}.so \
+		${STAGEDIR}${PYTHON_SITELIBDIR}/torch/lib/lib*.so
+
+fix-plist: # remove the stray %%PYTHON_SITELIBDIR%%/caffe2 file
+	@${REINPLACE_CMD} -e "s|.*/caffe2$$||" ${TMPPLIST}
+
+.include <bsd.port.mk>

misc/py-pytorch/distinfo

@@ -0,0 +1,3 @@
+TIMESTAMP = 1683446868
+SHA256 (pytorch/pytorch-v2.0.0.tar.gz) = cecc38b6d4256b810336edfc6119d7a57b701fdf1ba43c50001f31e2724fd8e2
+SIZE (pytorch/pytorch-v2.0.0.tar.gz) = 276643781

misc/py-pytorch/files/patch-CMakeLists.txt

@@ -0,0 +1,39 @@
+--- CMakeLists.txt.orig	2023-04-03 19:45:59 UTC
++++ CMakeLists.txt
+@@ -138,7 +138,7 @@ endif()
+ set(CPU_AARCH64 OFF)
+ set(CPU_INTEL OFF)
+ 
+-if(CMAKE_SYSTEM_PROCESSOR MATCHES "(AMD64|x86_64)")
++if(CMAKE_SYSTEM_PROCESSOR MATCHES "(amd64|x86_64)")
+   set(CPU_INTEL ON)
+ elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|arm64)")
+   set(CPU_AARCH64 ON)
+@@ -163,7 +163,7 @@ include(CMakeDependentOption)
+ option(ATEN_NO_TEST "Do not build ATen test binaries" OFF)
+ option(BUILD_BINARY "Build C++ binaries" OFF)
+ option(BUILD_DOCS "Build Caffe2 documentation" OFF)
+-option(BUILD_CUSTOM_PROTOBUF "Build and use Caffe2's own protobuf under third_party" ON)
++option(BUILD_CUSTOM_PROTOBUF "Build and use Caffe2's own protobuf under third_party" OFF)
+ option(BUILD_PYTHON "Build Python binaries" ON)
+ option(BUILD_CAFFE2 "Master flag to build Caffe2" OFF)
+ option(BUILD_LITE_INTERPRETER "Master flag to build Lite Interpreter" OFF)
+@@ -398,15 +398,15 @@ endif()
+ # USE_SYSTEM_LIBS being "OFF".
+ option(USE_SYSTEM_LIBS "Use all available system-provided libraries." OFF)
+ option(USE_SYSTEM_CPUINFO "Use system-provided cpuinfo." OFF)
+-option(USE_SYSTEM_SLEEF "Use system-provided sleef." OFF)
++option(USE_SYSTEM_SLEEF "Use system-provided sleef." ON)
+ option(USE_SYSTEM_GLOO "Use system-provided gloo." OFF)
+ option(USE_SYSTEM_FP16 "Use system-provided fp16." OFF)
+-option(USE_SYSTEM_PYBIND11 "Use system-provided PyBind11." OFF)
++option(USE_SYSTEM_PYBIND11 "Use system-provided PyBind11." ON)
+ option(USE_SYSTEM_PTHREADPOOL "Use system-provided pthreadpool." OFF)
+ option(USE_SYSTEM_PSIMD "Use system-provided psimd." OFF)
+ option(USE_SYSTEM_FXDIV "Use system-provided fxdiv." OFF)
+ option(USE_SYSTEM_BENCHMARK "Use system-provided google benchmark." OFF)
+-option(USE_SYSTEM_ONNX "Use system-provided onnx." OFF)
++option(USE_SYSTEM_ONNX "Use system-provided onnx." ON)
+ option(USE_SYSTEM_XNNPACK "Use system-provided xnnpack." OFF)
+ option(USE_GOLD_LINKER "Use ld.gold to link" OFF)
+ if(USE_SYSTEM_LIBS)

misc/py-pytorch/files/patch-aten_src_ATen_cpu_vec_vec256_vec256__bfloat16.h

@@ -0,0 +1,11 @@
+--- aten/src/ATen/cpu/vec/vec256/vec256_bfloat16.h.orig	2023-05-07 16:59:15 UTC
++++ aten/src/ATen/cpu/vec/vec256/vec256_bfloat16.h
+@@ -206,7 +206,7 @@ template <> class Vectorized<BFloat16> { (public)
+     }
+     return b;
+   }
+-  Vectorized<BFloat16> map(const __m256 (*const vop)(__m256)) const {
++  Vectorized<BFloat16> map(__m256 (*const vop)(__m256)) const {
+     __m256 lo, hi;
+     cvtbf16_fp32(values, lo, hi);
+     const auto o1 = vop(lo);

misc/py-pytorch/files/patch-aten_src_ATen_cpu_vec_vec512_vec512__bfloat16.h

@@ -0,0 +1,11 @@
+--- aten/src/ATen/cpu/vec/vec512/vec512_bfloat16.h.orig	2023-05-07 17:07:36 UTC
++++ aten/src/ATen/cpu/vec/vec512/vec512_bfloat16.h
+@@ -283,7 +283,7 @@ template <> class Vectorized<BFloat16> { (public)
+   }
+   #pragma clang diagnostic push
+   #pragma clang diagnostic ignored "-Wignored-qualifiers"
+-  Vectorized<BFloat16> map(const __m512 (*const vop)(__m512)) const {
++  Vectorized<BFloat16> map(__m512 (*const vop)(__m512)) const {
+     __m512 lo, hi;
+     cvtbf16_fp32(values, lo, hi);
+     const auto o1 = vop(lo);

misc/py-pytorch/files/patch-aten_src_ATen_native_sparse_ValidateCompressedIndicesCommon.h

@@ -0,0 +1,78 @@
+--- aten/src/ATen/native/sparse/ValidateCompressedIndicesCommon.h.orig	2023-05-07 08:51:40 UTC
++++ aten/src/ATen/native/sparse/ValidateCompressedIndicesCommon.h
+@@ -39,7 +39,7 @@ namespace {
+ // use `cidx/idx` to refer to `compressed_indices/plain_indices` respectively.
+ 
+ INVARIANT_CHECK_FUNC_API
+-_assert(const bool cond, const char* const message) {
++__assert(const bool cond, const char* const message) {
+ #ifdef GPUCC
+   CUDA_KERNEL_ASSERT(cond && message);
+ #else
+@@ -57,9 +57,9 @@ INVARIANT_CHECK_FUNC_API _check_first_cidx_is_zero(
+     const index_t& zero) {
+   const bool invariant = cidx == zero;
+   if (cdim_name == CDimName::CRow) {
+-    _assert(invariant, "`crow_indices[..., 0] == 0` is not satisfied.");
++    __assert(invariant, "`crow_indices[..., 0] == 0` is not satisfied.");
+   } else {
+-    _assert(invariant, "`ccol_indices[..., 0] == 0` is not satisfied.");
++    __assert(invariant, "`ccol_indices[..., 0] == 0` is not satisfied.");
+   }
+ }
+ 
+@@ -71,9 +71,9 @@ INVARIANT_CHECK_FUNC_API _check_last_cidx_is_nnz(
+     const index_t& nnz) {
+   const bool invariant = cidx == nnz;
+   if (cdim_name == CDimName::CRow) {
+-    _assert(invariant, "`crow_indices[..., -1] == nnz` is not satisfied.");
++    __assert(invariant, "`crow_indices[..., -1] == nnz` is not satisfied.");
+   } else {
+-    _assert(invariant, "`ccol_indices[..., -1] == nnz` is not satisfied.");
++    __assert(invariant, "`ccol_indices[..., -1] == nnz` is not satisfied.");
+   }
+ }
+ 
+@@ -88,11 +88,11 @@ INVARIANT_CHECK_FUNC_API _check_cidx_nondecreasing_loc
+   const auto s_cidx = cidx_next - cidx;
+   const bool invariant = zero <= s_cidx && s_cidx <= dim;
+   if (cdim_name == CDimName::CRow) {
+-    _assert(
++    __assert(
+         invariant,
+         "`0 <= crow_indices[..., 1:] - crow_indices[..., :-1] <= ncols` is not satisfied.");
+   } else {
+-    _assert(
++    __assert(
+         invariant,
+         "`0 <= ccol_indices[..., 1:] - ccol_indices[..., :-1] <= nrows` is not satisfied.");
+   }
+@@ -107,9 +107,9 @@ INVARIANT_CHECK_FUNC_API _check_idx_bounds(
+     const index_t& dim) {
+   const bool invariant = zero <= idx && idx < dim;
+   if (cdim_name == CDimName::CRow) {
+-    _assert(invariant, "`0 <= col_indices < ncols` is not satisfied.");
++    __assert(invariant, "`0 <= col_indices < ncols` is not satisfied.");
+   } else {
+-    _assert(invariant, "`0 <= row_indices < nrows` is not satisfied.");
++    __assert(invariant, "`0 <= row_indices < nrows` is not satisfied.");
+   }
+ }
+ 
+@@ -128,14 +128,14 @@ INVARIANT_CHECK_FUNC_API _check_idx_sorted_distinct_va
+   for (auto* RESTRICT curr = slice_begin + 1; curr < slice_end; ++curr) {
+     const auto invariant = *(curr - 1) < *curr;
+     if (cdim_name == CDimName::CRow) {
+-      _assert(
++      __assert(
+           invariant,
+           "`col_indices[..., crow_indices[..., i - 1]:crow_indices[..., i]] "
+           "for all i = 1, ..., nrows "
+           "are sorted and distinct along the last dimension values` "
+           "is not satisfied.");
+     } else {
+-      _assert(
++      __assert(
+           invariant,
+           "`row_indices[..., ccol_indices[..., i - 1]:ccol_indices[..., i]] "
+           "for all i = 1, ..., ncols "

misc/py-pytorch/files/patch-cmake_Dependencies.cmake

@@ -0,0 +1,11 @@
+--- cmake/Dependencies.cmake.orig	2022-12-16 00:23:46 UTC
++++ cmake/Dependencies.cmake
+@@ -339,7 +339,7 @@ if(USE_NNPACK OR USE_QNNPACK OR USE_PYTORCH_QNNPACK OR
+       set(DISABLE_NNPACK_AND_FAMILY ON)
+     endif()
+   else()
+-    if(NOT IOS AND NOT (CMAKE_SYSTEM_NAME MATCHES "^(Android|Linux|Darwin|Windows)$"))
++    if(NOT IOS AND NOT (CMAKE_SYSTEM_NAME MATCHES "^(Android|Linux|FreeBSD|Darwin|Windows)$"))
+       message(WARNING
+         "Target platform \"${CMAKE_SYSTEM_NAME}\" is not supported in {Q/X}NNPACK. "
+         "Supported platforms are Android, iOS, Linux, and macOS. "

misc/py-pytorch/files/patch-cmake_public_mkldnn.cmake

@@ -0,0 +1,11 @@
+--- cmake/public/mkldnn.cmake.orig	2022-12-16 00:23:46 UTC
++++ cmake/public/mkldnn.cmake
+@@ -4,7 +4,7 @@ if(CPU_AARCH64)
+   include(${CMAKE_CURRENT_LIST_DIR}/ComputeLibrary.cmake)
+ endif()
+ 
+-find_package(MKLDNN QUIET)
++find_package(MKLDNN REQUIRED)
+ 
+ if(NOT TARGET caffe2::mkldnn)
+   add_library(caffe2::mkldnn INTERFACE IMPORTED)

misc/py-pytorch/files/patch-third__party_cpuinfo_CMakeLists.txt

@@ -0,0 +1,56 @@
+--- third_party/cpuinfo/CMakeLists.txt.orig	2023-04-03 19:46:00 UTC
++++ third_party/cpuinfo/CMakeLists.txt
+@@ -65,7 +65,7 @@ IF(NOT CMAKE_SYSTEM_PROCESSOR)
+       "cpuinfo will compile, but cpuinfo_initialize() will always fail.")
+     SET(CPUINFO_SUPPORTED_PLATFORM FALSE)
+   ENDIF()
+-ELSEIF(NOT CPUINFO_TARGET_PROCESSOR MATCHES "^(i[3-6]86|AMD64|x86(_64)?|armv[5-8].*|aarch64|arm64|ARM64)$")
++ELSEIF(NOT CPUINFO_TARGET_PROCESSOR MATCHES "^(i[3-6]86|amd64|x86(_64)?|armv[5-8].*|aarch64|arm64|ARM64)$")
+   MESSAGE(WARNING
+     "Target processor architecture \"${CPUINFO_TARGET_PROCESSOR}\" is not supported in cpuinfo. "
+     "cpuinfo will compile, but cpuinfo_initialize() will always fail.")
+@@ -77,7 +77,7 @@ IF(NOT CMAKE_SYSTEM_NAME)
+       "Target operating system is not specified. "
+       "cpuinfo will compile, but cpuinfo_initialize() will always fail.")
+   SET(CPUINFO_SUPPORTED_PLATFORM FALSE)
+-ELSEIF(NOT CMAKE_SYSTEM_NAME MATCHES "^(Windows|CYGWIN|MSYS|Darwin|Linux|Android)$")
++ELSEIF(NOT CMAKE_SYSTEM_NAME MATCHES "^(Windows|CYGWIN|MSYS|Darwin|Linux|FreeBSD|Android)$")
+   IF(${CMAKE_VERSION} VERSION_GREATER_EQUAL "3.14" AND NOT CMAKE_SYSTEM_NAME STREQUAL "iOS")
+     MESSAGE(WARNING
+       "Target operating system \"${CMAKE_SYSTEM_NAME}\" is not supported in cpuinfo. "
+@@ -123,7 +123,7 @@ SET(CPUINFO_SRCS
+   src/cache.c)
+ 
+ IF(CPUINFO_SUPPORTED_PLATFORM)
+-  IF(NOT CMAKE_SYSTEM_NAME STREQUAL "Emscripten" AND (CPUINFO_TARGET_PROCESSOR MATCHES "^(i[3-6]86|AMD64|x86(_64)?)$" OR IOS_ARCH MATCHES "^(i386|x86_64)$"))
++  IF(NOT CMAKE_SYSTEM_NAME STREQUAL "Emscripten" AND (CPUINFO_TARGET_PROCESSOR MATCHES "^(i[3-6]86|amd64|x86(_64)?)$" OR IOS_ARCH MATCHES "^(i386|x86_64)$"))
+     LIST(APPEND CPUINFO_SRCS
+       src/x86/init.c
+       src/x86/info.c
+@@ -324,7 +324,7 @@ ENDIF()
+ # ---[ cpuinfo mock library and mock tests
+ IF(CPUINFO_SUPPORTED_PLATFORM AND CPUINFO_BUILD_MOCK_TESTS)
+   SET(CPUINFO_MOCK_SRCS "${CPUINFO_SRCS}")
+-  IF(CPUINFO_TARGET_PROCESSOR MATCHES "^(i[3-6]86|AMD64|x86(_64)?)$")
++  IF(CPUINFO_TARGET_PROCESSOR MATCHES "^(i[3-6]86|amd64|x86(_64)?)$")
+     LIST(APPEND CPUINFO_MOCK_SRCS src/x86/mockcpuid.c)
+   ENDIF()
+   IF(CMAKE_SYSTEM_NAME STREQUAL "Linux" OR CMAKE_SYSTEM_NAME STREQUAL "Android")
+@@ -768,7 +768,7 @@ IF(CPUINFO_SUPPORTED_PLATFORM AND CPUINFO_BUILD_UNIT_T
+     ADD_TEST(NAME get-current-test COMMAND get-current-test)
+   ENDIF()
+ 
+-  IF(CPUINFO_TARGET_PROCESSOR MATCHES "^(i[3-6]86|AMD64|x86(_64)?)$")
++  IF(CPUINFO_TARGET_PROCESSOR MATCHES "^(i[3-6]86|amd64|x86(_64)?)$")
+     ADD_EXECUTABLE(brand-string-test test/name/brand-string.cc)
+     CPUINFO_TARGET_ENABLE_CXX11(brand-string-test)
+     CPUINFO_TARGET_RUNTIME_LIBRARY(brand-string-test)
+@@ -835,7 +835,7 @@ IF(CPUINFO_SUPPORTED_PLATFORM AND CPUINFO_BUILD_TOOLS)
+     CPUINFO_TARGET_RUNTIME_LIBRARY(cpuinfo-dump)
+   ENDIF()
+ 
+-  IF(CPUINFO_TARGET_PROCESSOR MATCHES "^(i[3-6]86|AMD64|x86(_64)?)$")
++  IF(CPUINFO_TARGET_PROCESSOR MATCHES "^(i[3-6]86|amd64|x86(_64)?)$")
+     ADD_EXECUTABLE(cpuid-dump tools/cpuid-dump.c)
+     CPUINFO_TARGET_ENABLE_C99(cpuid-dump)
+     CPUINFO_TARGET_RUNTIME_LIBRARY(cpuid-dump)

misc/py-pytorch/files/patch-third__party_fbgemm_third__party_asmjit_src_asmjit_core_virtmem.cpp

@@ -0,0 +1,10 @@
+--- third_party/fbgemm/third_party/asmjit/src/asmjit/core/virtmem.cpp.orig	2022-12-16 00:23:48 UTC
++++ third_party/fbgemm/third_party/asmjit/src/asmjit/core/virtmem.cpp
+@@ -45,6 +45,7 @@
+ #endif
+ 
+ #include <atomic>
++#include <sys/stat.h>
+ 
+ #if defined(__APPLE__) || defined(__BIONIC__)
+   #define ASMJIT_VM_SHM_DETECT 0

misc/py-pytorch/files/patch-third__party_kineto_libkineto_src_ThreadUtil.cpp

@@ -0,0 +1,11 @@
+--- third_party/kineto/libkineto/src/ThreadUtil.cpp.orig	2023-04-03 19:46:02 UTC
++++ third_party/kineto/libkineto/src/ThreadUtil.cpp
+@@ -57,7 +57,7 @@ int32_t systemThreadId() {
+ #elif defined _MSC_VER
+     _sysTid = (int32_t)GetCurrentThreadId();
+ #else
+-    _sysTid = (int32_t)syscall(SYS_gettid);
++    _sysTid = (int32_t)syscall(SYS_getpid);
+ #endif
+   }
+   return _sysTid;

misc/py-pytorch/pkg-descr

@@ -0,0 +1,3 @@
+PyTorch is a Python package that provides two high-level features:
+* Tensor computation (like NumPy) with strong GPU acceleration
+* Deep neural networks built on a tape-based autograd system