griefith/test2
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Windows: Enable OpenPGL on ARM64 machines

Browse Source 
This is required to reach feature completeness with x64 platforms.

Also patches TBB headers in order to fix a few build errors when compiling OpenPGL.

Pull Request: https://projects.blender.org/blender/blender/pulls/127705
This commit is contained in:
Anthony Roberts
2024-09-17 14:57:04 +02:00
parent 943832a841
commit f0de0bc8d3
5 changed files with 271 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
build_files/build_environment/CMakeLists.txt
View File

@@ -100,7 +100,9 @@ if(NOT APPLE)
endif()
endif()
include(cmake/ispc.cmake)
if(NOT BLENDER_PLATFORM_WINDOWS_ARM)
if(BLENDER_PLATFORM_WINDOWS_ARM)
include(cmake/openpgl_windows_arm.cmake)
else()
include(cmake/openpgl.cmake)
endif()
# Embree needs to be included after dpcpp as it uses it for compiling with GPU support

120
build_files/build_environment/cmake/openpgl_windows_arm.cmake Normal file
View File

@@ -0,0 +1,120 @@
# SPDX-FileCopyrightText: 2022-2024 Blender Authors
#
# SPDX-License-Identifier: GPL-2.0-or-later
# Note the utility apps may use png/tiff/gif system libraries, but the
# library itself does not depend on them, so should give no problems.
set(OPENPGL_LLVM_INSTALL_PATH ${LIBDIR}/llvm)
set(OPENPGL_EXTRA_ARGS
-DOPENPGL_BUILD_STATIC=ON
-DOPENPGL_TBB_ROOT=${LIBDIR}/tbb
-DTBB_ROOT=${LIBDIR}/tbb
-DCMAKE_DEBUG_POSTFIX=_d
-DCMAKE_CXX_COMPILER=${OPENPGL_LLVM_INSTALL_PATH}/bin/clang-cl.exe
-DCMAKE_C_COMPILER=${OPENPGL_LLVM_INSTALL_PATH}/bin/clang-cl.exe
-DCMAKE_C_FLAGS_INIT="--target=arm64-pc-windows-msvc"
-DCMAKE_CXX_FLAGS_INIT="--target=arm64-pc-windows-msvc"
-DCMAKE_SHARED_LINKER_FLAGS=-L"${LIBDIR}/../../VS1564R/Release/llvm/lib"
)
if(TBB_STATIC_LIBRARY)
set(OPENPGL_EXTRA_ARGS
${OPENPGL_EXTRA_ARGS}
-DOPENPGL_TBB_COMPONENT=tbb_static
)
endif()
# We want the VS2019 tools for OpenPGL, as they are stable.
# We cannot use VS2022 easily, unless we specify an older (unsupported) toolset such as 17.35,
# as the newer toolsets mandate newer versions of LLVM, which we cannot use currently,
# due to lack of support in OSL and ISPC.
set(OPENPGL_VCTOOLS_REQUIRED_VERSION 14.29)
# Extract the list of installed tools that match the required version from the
# `VCToolsInstallDir` env var
file(TO_CMAKE_PATH $ENV{VCToolsInstallDir} OPENPGL_VCTOOLSINSTALLDIR_PATH)
cmake_path(GET OPENPGL_VCTOOLSINSTALLDIR_PATH PARENT_PATH OPENPGL_VCTOOLSDIR_PATH)
file(GLOB OPENPGL_INSTALLED_VCTOOLS RELATIVE ${OPENPGL_VCTOOLSDIR_PATH} ${OPENPGL_VCTOOLSDIR_PATH}/${OPENPGL_VCTOOLS_REQUIRED_VERSION}*)
# Check that at least one the installed tool versions
# (there may be different subversions) is present.
if(NOT OPENPGL_INSTALLED_VCTOOLS)
message(FATAL_ERROR "When building for Windows ARM64 platforms, OpenPGL requires VC Tools ${OPENPGL_VCTOOLS_REQUIRED_VERSION} to be installed alongside the current version.")
endif()
# Get the last item in the list (latest, when list is sorted)
list(SORT OPENPGL_INSTALLED_VCTOOLS)
list(GET OPENPGL_INSTALLED_VCTOOLS -1 OPENPGL_VCTOOLS_VERSION)
# Configure our in file and temporarily store it in the build dir
# (with modified extension so nothing else picks it up)
# This feels icky, but boost does something similar, and we haven't called
# `ExternalProject_Add` yet, so the OpenPGL dir does not yet exist.
configure_file(
${PATCH_DIR}/openpgl_Directory.Build.Props.in
${BUILD_DIR}/openpgl_Directory.Build.Props_temp
)
# Set the patch command to copy the configured build props file in,
# and also a newer version of sse2neon
set(OPENPGL_PATCH_COMMAND
COMMAND ${CMAKE_COMMAND} -E copy
${BUILD_DIR}/openpgl_Directory.Build.Props_temp
${BUILD_DIR}/openpgl/src/external_openpgl-build/Directory.Build.Props
COMMAND ${CMAKE_COMMAND} -E copy
${BUILD_DIR}/sse2neon/src/external_sse2neon/sse2neon.h
${BUILD_DIR}/openpgl/src/external_openpgl/third-party/embreeSrc/common/simd/arm/sse2neon.h &&
${PATCH_CMD} -p 1 -d ${BUILD_DIR}/openpgl/src/external_openpgl < ${PATCH_DIR}/openpgl_windows_arm.diff
)
# This all only works if we use the VS generator (with `clangcl` toolset), so ensure we use that
# Note: there is literally no way to get ninja to use a different toolset other than manually
# overwriting every env var, or calling a nested `vcvarsall`, both of which are *messy*.
set(OPENPGL_GENERATOR ${CMAKE_GENERATOR})
set(OPENPGL_GENERATOR_TOOLSET ClangCL)
ExternalProject_Add(external_openpgl
URL file://${PACKAGE_DIR}/${OPENPGL_FILE}
DOWNLOAD_DIR ${DOWNLOAD_DIR}
URL_HASH ${OPENPGL_HASH_TYPE}=${OPENPGL_HASH}
CMAKE_GENERATOR ${OPENPGL_GENERATOR}
CMAKE_GENERATOR_TOOLSET ${OPENPGL_GENERATOR_TOOLSET}
PREFIX ${BUILD_DIR}/openpgl
PATCH_COMMAND ${OPENPGL_PATCH_COMMAND}
CMAKE_ARGS
-DCMAKE_INSTALL_PREFIX=${LIBDIR}/openpgl -DCMAKE_BUILD_TYPE=${BUILD_MODE}
${DEFAULT_CMAKE_FLAGS}
${OPENPGL_EXTRA_ARGS}
INSTALL_DIR ${LIBDIR}/openpgl
)
add_dependencies(
external_openpgl
external_tbb
external_sse2neon
)
if(BUILD_MODE STREQUAL Release)
ExternalProject_Add_Step(external_openpgl after_install
COMMAND ${CMAKE_COMMAND} -E copy_directory
${LIBDIR}/openpgl
${HARVEST_TARGET}/openpgl
DEPENDEES install
)
else()
ExternalProject_Add_Step(external_openpgl after_install
COMMAND ${CMAKE_COMMAND} -E copy
${LIBDIR}/openpgl/lib/openpgl_d.lib
${HARVEST_TARGET}/openpgl/lib/openpgl_d.lib
COMMAND ${CMAKE_COMMAND} -E copy
${LIBDIR}/openpgl/lib/cmake/openpgl-${OPENPGL_SHORT_VERSION}/openpgl_Exports-debug.cmake
${HARVEST_TARGET}/openpgl/lib/cmake/openpgl-${OPENPGL_SHORT_VERSION}/openpgl_Exports-debug.cmake
DEPENDEES install
)
endif()

7
build_files/build_environment/patches/openpgl_Directory.Build.Props.in Normal file
View File

@@ -0,0 +1,7 @@
<Project>
<PropertyGroup>
<LLVMInstallDir>${OPENPGL_LLVM_INSTALL_PATH}</LLVMInstallDir>
<LLVMToolsVersion>${LLVM_VERSION}</LLVMToolsVersion>
<VCToolsVersion>${OPENPGL_VCTOOLS_VERSION}</VCToolsVersion>
</PropertyGroup>
</Project>

124
build_files/build_environment/patches/openpgl_windows_arm.diff Normal file
View File

@@ -0,0 +1,124 @@
diff --git a/CMakeLists.txt b/CMakeLists.txt
index ec486ee..cc77876 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -35,7 +35,7 @@ SET(OPENPGL_ARM OFF)
IF (APPLE AND CMAKE_SYSTEM_NAME STREQUAL "Darwin" AND (CMAKE_SYSTEM_PROCESSOR STREQUAL "arm64" OR CMAKE_OSX_ARCHITECTURES MATCHES "arm64"))
MESSAGE(STATUS "Building for Apple silicon")
SET(OPENPGL_ARM ON)
-ELSEIF(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64")
+ELSEIF(CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" OR CMAKE_SYSTEM_PROCESSOR STREQUAL "ARM64")
MESSAGE(STATUS "Building for AArch64")
SET(OPENPGL_ARM ON)
ENDIF()
diff --git a/openpgl/CMakeLists.txt b/openpgl/CMakeLists.txt
index 4bcccbe..1199f22 100644
--- a/openpgl/CMakeLists.txt
+++ b/openpgl/CMakeLists.txt
@@ -60,7 +60,8 @@ message(STATUS "Compiler: ${CMAKE_CXX_COMPILER_ID}")
message(STATUS "Arch: ${CMAKE_SYSTEM_PROCESSOR}")
if(WIN32)
- if(CMAKE_CXX_COMPILER_ID STREQUAL "MSVC")
+ # Here we chack for MSVC, or Clang pretending to be MSVC via Clang-CL
+ if(CMAKE_CXX_COMPILER_ID STREQUAL "MSVC" OR (CMAKE_CXX_COMPILER_ID STREQUAL "Clang" AND CMAKE_CXX_SIMULATE_ID STREQUAL "MSVC"))
set(OPENPGL_RELEASE_OPTIONS /Ox /Oi)
set(OPENPGL_COMMON_OPTIONS /fp:precise)
#set(OPENPGL_RELEASE_OPTIONS ${OPENPGL_RELEASE_OPTIONS} -ftree-vectorize -mfpmath=sse -funsafe-math-optimizations -fno-rounding-math -fno-signaling-nans -fno-math-errno -fomit-frame-pointer )
@@ -77,6 +78,14 @@ if(WIN32)
if(OPENPGL_ISA_AVX512)
set_source_files_properties(api/deviceCPU16.cpp PROPERTIES COMPILE_FLAGS "/D__SSE__ /D__SSE2__ /D__SSE3__ /D__SSE4_1__ /D__SSE4_2__ /arch:AVX /arch:AVX2 /arch:AVX512")
endif()
+ if(CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
+ if(OPENPGL_ISA_NEON)
+ set_source_files_properties(api/deviceCPU4.cpp PROPERTIES COMPILE_FLAGS "/D__SSE4_2__ /D__SSE4_1__")
+ endif()
+ if(OPENPGL_ISA_NEON2X)
+ set_source_files_properties(api/deviceCPU8.cpp PROPERTIES COMPILE_FLAGS "/D__AVX2__ /D__AVX__ /D__SSE4_2__ /D__SSE4_1__ /D__BMI__ /D__BMI2__ /D__LZCNT__")
+ endif()
+ endif()
elseif(CMAKE_CXX_COMPILER_ID STREQUAL "IntelLLVM" OR CMAKE_CXX_COMPILER_ID STREQUAL "dpcpp")
set(OPENPGL_RELEASE_OPTIONS -O3)
set(OPENPGL_COMMON_OPTIONS -Wall)
diff --git a/third-party/embreeSrc/common/simd/arm/sse2neon.h b/third-party/embreeSrc/common/simd/arm/sse2neon.h
index 99831e3..a2cefbc 100644
--- a/third-party/embreeSrc/common/simd/arm/sse2neon.h
+++ b/third-party/embreeSrc/common/simd/arm/sse2neon.h
@@ -336,6 +336,14 @@ FORCE_INLINE void _sse2neon_smp_mb(void)
* argument "a" of mm_shuffle_ps that will be places in fp1 of result.
* fp0 is the same for fp0 of result.
*/
+#if defined(__aarch64__)
+#define _MN_SHUFFLE(fp3,fp2,fp1,fp0) ( (uint8x16_t){ (((fp3)*4)+0), (((fp3)*4)+1), (((fp3)*4)+2), (((fp3)*4)+3), (((fp2)*4)+0), (((fp2)*4)+1), (((fp2)*4)+\
+2), (((fp2)*4)+3), (((fp1)*4)+0), (((fp1)*4)+1), (((fp1)*4)+2), (((fp1)*4)+3), (((fp0)*4)+0), (((fp0)*4)+1), (((fp0)*4)+2), (((fp0)*4)+3) } )
+#define _MF_SHUFFLE(fp3,fp2,fp1,fp0) ( (uint8x16_t){ (((fp3)*4)+0), (((fp3)*4)+1), (((fp3)*4)+2), (((fp3)*4)+3), (((fp2)*4)+0), (((fp2)*4)+1), (((fp2)*4)+\
+2), (((fp2)*4)+3), (((fp1)*4)+16+0), (((fp1)*4)+16+1), (((fp1)*4)+16+2), (((fp1)*4)+16+3), (((fp0)*4)+16+0), (((fp0)*4)+16+1), (((fp0)*4)+16+2), (((fp0)*\
+4)+16+3) } )
+#endif
+
#define _MM_SHUFFLE(fp3, fp2, fp1, fp0) \
(((fp3) << 6) | ((fp2) << 4) | ((fp1) << 2) | ((fp0)))
@@ -2822,7 +2830,7 @@ FORCE_INLINE void _mm_stream_pi(__m64 *p, __m64 a)
FORCE_INLINE void _mm_stream_ps(float *p, __m128 a)
{
#if __has_builtin(__builtin_nontemporal_store)
- __builtin_nontemporal_store(a, (float32x4_t *) p);
+ __builtin_nontemporal_store(reinterpret_cast<float32x4_t>(a), (float32x4_t *) p);
#else
vst1q_f32(p, vreinterpretq_f32_m128(a));
#endif
@@ -5660,7 +5668,7 @@ FORCE_INLINE void _mm_storeu_si32(void *p, __m128i a)
FORCE_INLINE void _mm_stream_pd(double *p, __m128d a)
{
#if __has_builtin(__builtin_nontemporal_store)
- __builtin_nontemporal_store(a, (__m128d *) p);
+ __builtin_nontemporal_store(reinterpret_cast<float32x4_t>(a), (float32x4_t *) p);
#elif defined(__aarch64__) || defined(_M_ARM64)
vst1q_f64(p, vreinterpretq_f64_m128d(a));
#else
@@ -6809,14 +6817,14 @@ FORCE_INLINE __m64 _mm_sign_pi8(__m64 _a, __m64 _b)
_sse2neon_define2( \
__m128i, a, b, \
const uint16_t _mask[8] = \
- _sse2neon_init(((imm) & (1 << 0)) ? (uint16_t) -1 : 0x0, \
- ((imm) & (1 << 1)) ? (uint16_t) -1 : 0x0, \
- ((imm) & (1 << 2)) ? (uint16_t) -1 : 0x0, \
- ((imm) & (1 << 3)) ? (uint16_t) -1 : 0x0, \
- ((imm) & (1 << 4)) ? (uint16_t) -1 : 0x0, \
- ((imm) & (1 << 5)) ? (uint16_t) -1 : 0x0, \
- ((imm) & (1 << 6)) ? (uint16_t) -1 : 0x0, \
- ((imm) & (1 << 7)) ? (uint16_t) -1 : 0x0); \
+ _sse2neon_init(((imm) & (1 << 0)) ? (uint16_t)0xffff : (uint16_t)0x0000, \
+ ((imm) & (1 << 1)) ? (uint16_t)0xffff : (uint16_t)0x0000, \
+ ((imm) & (1 << 2)) ? (uint16_t)0xffff : (uint16_t)0x0000, \
+ ((imm) & (1 << 3)) ? (uint16_t)0xffff : (uint16_t)0x0000, \
+ ((imm) & (1 << 4)) ? (uint16_t)0xffff : (uint16_t)0x0000, \
+ ((imm) & (1 << 5)) ? (uint16_t)0xffff : (uint16_t)0x0000, \
+ ((imm) & (1 << 6)) ? (uint16_t)0xffff : (uint16_t)0x0000, \
+ ((imm) & (1 << 7)) ? (uint16_t)0xffff : (uint16_t)0x0000); \
uint16x8_t _mask_vec = vld1q_u16(_mask); \
uint16x8_t __a = vreinterpretq_u16_m128i(_a); \
uint16x8_t __b = vreinterpretq_u16_m128i(_b); _sse2neon_return( \
diff --git a/third-party/embreeSrc/common/sys/intrinsics.h b/third-party/embreeSrc/common/sys/intrinsics.h
index f5074bb..b0511b5 100644
--- a/third-party/embreeSrc/common/sys/intrinsics.h
+++ b/third-party/embreeSrc/common/sys/intrinsics.h
@@ -91,7 +91,7 @@ namespace embree
#if defined(__X86_64__) || defined (__aarch64__)
__forceinline size_t bsf(size_t v) {
-#if defined(__AVX2__)
+#if defined(__AVX2__) && !defined(__aarch64__)
return _tzcnt_u64(v);
#else
unsigned long r = 0; _BitScanForward64(&r,v); return r;
@@ -140,7 +140,7 @@ namespace embree
#if defined(__X86_64__) || defined (__aarch64__)
__forceinline size_t bsr(size_t v) {
-#if defined(__AVX2__)
+#if defined(__AVX2__) && !defined(__aarch64__)
return 63 -_lzcnt_u64(v);
#else
unsigned long r = 0; _BitScanReverse64(&r, v); return r;

17
build_files/build_environment/patches/tbb.diff
View File

@@ -171,6 +171,23 @@ index 00000000..13d56678
+ _InterlockedAnd((long volatile *)operand, (long)addend);
+}
+
diff --git a/include/tbb/tbb_config.h b/include/tbb/tbb_config.h
index 7a8d06a0..80b4e3a6 100644
--- a/include/tbb/tbb_config.h
+++ b/include/tbb/tbb_config.h
@@ -209,10 +209,10 @@
#elif __clang__
/** TODO: these options need to be rechecked **/
#define __TBB_CPP11_VARIADIC_TEMPLATES_PRESENT __has_feature(__cxx_variadic_templates__)
- #define __TBB_CPP11_RVALUE_REF_PRESENT (__has_feature(__cxx_rvalue_references__) && (_LIBCPP_VERSION || __TBB_GLIBCXX_VERSION >= 40500))
+ #define __TBB_CPP11_RVALUE_REF_PRESENT (__has_feature(__cxx_rvalue_references__) && (_LIBCPP_VERSION || __TBB_GLIBCXX_VERSION >= 40500 || _MSC_VER >= 1700))
#define __TBB_IMPLICIT_MOVE_PRESENT __has_feature(cxx_implicit_moves)
/** TODO: extend exception_ptr related conditions to cover libstdc++ **/
- #define __TBB_EXCEPTION_PTR_PRESENT (__cplusplus >= 201103L && (_LIBCPP_VERSION || __TBB_GLIBCXX_VERSION >= 40600))
+ #define __TBB_EXCEPTION_PTR_PRESENT (__cplusplus >= 201103L && (_LIBCPP_VERSION || __TBB_GLIBCXX_VERSION >= 40600 || _MSC_VER >= 1600))
#define __TBB_STATIC_ASSERT_PRESENT __has_feature(__cxx_static_assert__)
#if (__cplusplus >= 201103L && __has_include(<tuple>))
#define __TBB_CPP11_TUPLE_PRESENT 1
diff --git a/include/tbb/tbb_machine.h b/include/tbb/tbb_machine.h
index 9752be58..ebb98ec2 100644
--- a/include/tbb/tbb_machine.h