|
|
|
|
@@ -34,246 +34,234 @@ diff -Naur llvm-sycl-nightly-20220208.orig/libdevice/cmake/modules/SYCLLibdevice
|
|
|
|
|
libsycldevice-obj
|
|
|
|
|
libsycldevice-spv)
|
|
|
|
|
|
|
|
|
|
diff --git a/sycl/source/detail/program_manager/program_manager.cpp b/sycl/source/detail/program_manager/program_manager.cpp
|
|
|
|
|
index 17eeaafae194..09e6d2217aaa 100644
|
|
|
|
|
--- a/sycl/source/detail/program_manager/program_manager.cpp
|
|
|
|
|
+++ b/sycl/source/detail/program_manager/program_manager.cpp
|
|
|
|
|
@@ -1647,46 +1647,120 @@ ProgramManager::getSYCLDeviceImagesWithCompatibleState(
|
|
|
|
|
}
|
|
|
|
|
assert(BinImages.size() > 0 && "Expected to find at least one device image");
|
|
|
|
|
diff --git a/sycl/CMakeLists.txt b/sycl/CMakeLists.txt
|
|
|
|
|
index 00ce045f43c3..e044262e628e 100644
|
|
|
|
|
--- a/sycl/CMakeLists.txt
|
|
|
|
|
+++ b/sycl/CMakeLists.txt
|
|
|
|
|
@@ -188,7 +188,6 @@ install(FILES
|
|
|
|
|
COMPONENT sycl-headers)
|
|
|
|
|
|
|
|
|
|
+ // Ignore images with incompatible state. Image is considered compatible
|
|
|
|
|
+ // with a target state if an image is already in the target state or can
|
|
|
|
|
+ // be brought to target state by compiling/linking/building.
|
|
|
|
|
+ //
|
|
|
|
|
+ // Example: an image in "executable" state is not compatible with
|
|
|
|
|
+ // "input" target state - there is no operation to convert the image it
|
|
|
|
|
+ // to "input" state. An image in "input" state is compatible with
|
|
|
|
|
+ // "executable" target state because it can be built to get into
|
|
|
|
|
+ // "executable" state.
|
|
|
|
|
+ for (auto It = BinImages.begin(); It != BinImages.end();) {
|
|
|
|
|
+ if (getBinImageState(*It) > TargetState)
|
|
|
|
|
+ It = BinImages.erase(It);
|
|
|
|
|
+ else
|
|
|
|
|
+ ++It;
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
std::vector<device_image_plain> SYCLDeviceImages;
|
|
|
|
|
- for (RTDeviceBinaryImage *BinImage : BinImages) {
|
|
|
|
|
- const bundle_state ImgState = getBinImageState(BinImage);
|
|
|
|
|
include(AddBoostMp11Headers)
|
|
|
|
|
-include(FetchBoostUnorderedHeaders)
|
|
|
|
|
|
|
|
|
|
# This is workaround to detect changes (add or modify) in subtree which
|
|
|
|
|
# are not detected by copy_directory command.
|
|
|
|
|
diff --git a/sycl/cmake/modules/FetchBoostUnorderedHeaders.cmake b/sycl/cmake/modules/FetchBoostUnorderedHeaders.cmake
|
|
|
|
|
deleted file mode 100644
|
|
|
|
|
index a0f446055026..000000000000
|
|
|
|
|
--- a/sycl/cmake/modules/FetchBoostUnorderedHeaders.cmake
|
|
|
|
|
+++ /dev/null
|
|
|
|
|
@@ -1,129 +0,0 @@
|
|
|
|
|
-# Fetches the unordered boost module and its dependencies
|
|
|
|
|
-function(add_boost_module_headers)
|
|
|
|
|
- cmake_parse_arguments(
|
|
|
|
|
- BOOST_MODULE # prefix
|
|
|
|
|
- "" # options
|
|
|
|
|
- "NAME;SRC_DIR;GIT_TAG;" # one value keywords
|
|
|
|
|
- "" # multi-value keywords
|
|
|
|
|
- ${ARGN}) # arguments
|
|
|
|
|
-
|
|
|
|
|
- // Ignore images with incompatible state. Image is considered compatible
|
|
|
|
|
- // with a target state if an image is already in the target state or can
|
|
|
|
|
- // be brought to target state by compiling/linking/building.
|
|
|
|
|
- //
|
|
|
|
|
- // Example: an image in "executable" state is not compatible with
|
|
|
|
|
- // "input" target state - there is no operation to convert the image it
|
|
|
|
|
- // to "input" state. An image in "input" state is compatible with
|
|
|
|
|
- // "executable" target state because it can be built to get into
|
|
|
|
|
- // "executable" state.
|
|
|
|
|
- if (ImgState > TargetState)
|
|
|
|
|
- continue;
|
|
|
|
|
|
|
|
|
|
- for (const sycl::device &Dev : Devs) {
|
|
|
|
|
+ // If a non-input state is requested, we can filter out some compatible
|
|
|
|
|
+ // images and return only those with the highest compatible state for each
|
|
|
|
|
+ // device-kernel pair. This map tracks how many kernel-device pairs need each
|
|
|
|
|
+ // image, so that any unneeded ones are skipped.
|
|
|
|
|
+ // TODO this has no effect if the requested state is input, consider having
|
|
|
|
|
+ // a separate branch for that case to avoid unnecessary tracking work.
|
|
|
|
|
+ struct DeviceBinaryImageInfo {
|
|
|
|
|
+ std::shared_ptr<std::vector<sycl::kernel_id>> KernelIDs;
|
|
|
|
|
+ bundle_state State = bundle_state::input;
|
|
|
|
|
+ int RequirementCounter = 0;
|
|
|
|
|
+ };
|
|
|
|
|
+ std::unordered_map<RTDeviceBinaryImage *, DeviceBinaryImageInfo> ImageInfoMap;
|
|
|
|
|
+
|
|
|
|
|
+ for (const sycl::device &Dev : Devs) {
|
|
|
|
|
+ // Track the highest image state for each requested kernel.
|
|
|
|
|
+ using StateImagesPairT =
|
|
|
|
|
+ std::pair<bundle_state, std::vector<RTDeviceBinaryImage *>>;
|
|
|
|
|
+ using KernelImageMapT =
|
|
|
|
|
+ std::map<kernel_id, StateImagesPairT, LessByNameComp>;
|
|
|
|
|
+ KernelImageMapT KernelImageMap;
|
|
|
|
|
+ if (!KernelIDs.empty())
|
|
|
|
|
+ for (const kernel_id &KernelID : KernelIDs)
|
|
|
|
|
+ KernelImageMap.insert({KernelID, {}});
|
|
|
|
|
+
|
|
|
|
|
+ for (RTDeviceBinaryImage *BinImage : BinImages) {
|
|
|
|
|
if (!compatibleWithDevice(BinImage, Dev) ||
|
|
|
|
|
!doesDevSupportImgAspects(Dev, *BinImage))
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
- std::shared_ptr<std::vector<sycl::kernel_id>> KernelIDs;
|
|
|
|
|
- // Collect kernel names for the image
|
|
|
|
|
- {
|
|
|
|
|
- std::lock_guard<std::mutex> KernelIDsGuard(m_KernelIDsMutex);
|
|
|
|
|
- KernelIDs = m_BinImg2KernelIDs[BinImage];
|
|
|
|
|
- // If the image does not contain any non-service kernels we can skip it.
|
|
|
|
|
- if (!KernelIDs || KernelIDs->empty())
|
|
|
|
|
- continue;
|
|
|
|
|
+ auto InsertRes = ImageInfoMap.insert({BinImage, {}});
|
|
|
|
|
+ DeviceBinaryImageInfo &ImgInfo = InsertRes.first->second;
|
|
|
|
|
+ if (InsertRes.second) {
|
|
|
|
|
+ ImgInfo.State = getBinImageState(BinImage);
|
|
|
|
|
+ // Collect kernel names for the image
|
|
|
|
|
+ {
|
|
|
|
|
+ std::lock_guard<std::mutex> KernelIDsGuard(m_KernelIDsMutex);
|
|
|
|
|
+ ImgInfo.KernelIDs = m_BinImg2KernelIDs[BinImage];
|
|
|
|
|
+ }
|
|
|
|
|
}
|
|
|
|
|
+ const bundle_state ImgState = ImgInfo.State;
|
|
|
|
|
+ const std::shared_ptr<std::vector<sycl::kernel_id>> &ImageKernelIDs =
|
|
|
|
|
+ ImgInfo.KernelIDs;
|
|
|
|
|
+ int &ImgRequirementCounter = ImgInfo.RequirementCounter;
|
|
|
|
|
|
|
|
|
|
- DeviceImageImplPtr Impl = std::make_shared<detail::device_image_impl>(
|
|
|
|
|
- BinImage, Ctx, Devs, ImgState, KernelIDs, /*PIProgram=*/nullptr);
|
|
|
|
|
+ // If the image does not contain any non-service kernels we can skip it.
|
|
|
|
|
+ if (!ImageKernelIDs || ImageKernelIDs->empty())
|
|
|
|
|
+ continue;
|
|
|
|
|
|
|
|
|
|
- SYCLDeviceImages.push_back(
|
|
|
|
|
- createSyclObjFromImpl<device_image_plain>(Impl));
|
|
|
|
|
- break;
|
|
|
|
|
+ // Update tracked information.
|
|
|
|
|
+ for (kernel_id &KernelID : *ImageKernelIDs) {
|
|
|
|
|
+ StateImagesPairT *StateImagesPair;
|
|
|
|
|
+ // If only specific kernels are requested, ignore the rest.
|
|
|
|
|
+ if (!KernelIDs.empty()) {
|
|
|
|
|
+ auto It = KernelImageMap.find(KernelID);
|
|
|
|
|
+ if (It == KernelImageMap.end())
|
|
|
|
|
+ continue;
|
|
|
|
|
+ StateImagesPair = &It->second;
|
|
|
|
|
+ } else
|
|
|
|
|
+ StateImagesPair = &KernelImageMap[KernelID];
|
|
|
|
|
+
|
|
|
|
|
+ auto &[KernelImagesState, KernelImages] = *StateImagesPair;
|
|
|
|
|
+
|
|
|
|
|
+ if (KernelImages.empty()) {
|
|
|
|
|
+ KernelImagesState = ImgState;
|
|
|
|
|
+ KernelImages.push_back(BinImage);
|
|
|
|
|
+ ++ImgRequirementCounter;
|
|
|
|
|
+ } else if (KernelImagesState < ImgState) {
|
|
|
|
|
+ for (RTDeviceBinaryImage *Img : KernelImages) {
|
|
|
|
|
+ auto It = ImageInfoMap.find(Img);
|
|
|
|
|
+ assert(It != ImageInfoMap.end());
|
|
|
|
|
+ assert(It->second.RequirementCounter > 0);
|
|
|
|
|
+ --(It->second.RequirementCounter);
|
|
|
|
|
+ }
|
|
|
|
|
+ KernelImages.clear();
|
|
|
|
|
+ KernelImages.push_back(BinImage);
|
|
|
|
|
+ KernelImagesState = ImgState;
|
|
|
|
|
+ ++ImgRequirementCounter;
|
|
|
|
|
+ } else if (KernelImagesState == ImgState) {
|
|
|
|
|
+ KernelImages.push_back(BinImage);
|
|
|
|
|
+ ++ImgRequirementCounter;
|
|
|
|
|
+ }
|
|
|
|
|
+ }
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
+ for (const auto &ImgInfoPair : ImageInfoMap) {
|
|
|
|
|
+ if (ImgInfoPair.second.RequirementCounter == 0)
|
|
|
|
|
+ continue;
|
|
|
|
|
+
|
|
|
|
|
+ DeviceImageImplPtr Impl = std::make_shared<detail::device_image_impl>(
|
|
|
|
|
+ ImgInfoPair.first, Ctx, Devs, ImgInfoPair.second.State,
|
|
|
|
|
+ ImgInfoPair.second.KernelIDs, /*PIProgram=*/nullptr);
|
|
|
|
|
+
|
|
|
|
|
+ SYCLDeviceImages.push_back(createSyclObjFromImpl<device_image_plain>(Impl));
|
|
|
|
|
+ }
|
|
|
|
|
+
|
|
|
|
|
return SYCLDeviceImages;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
diff --git a/sycl/source/detail/pi.cpp b/sycl/source/detail/pi.cpp
|
|
|
|
|
index fb228cd85979..94e8438ee964 100644
|
|
|
|
|
--- a/sycl/source/detail/pi.cpp
|
|
|
|
|
+++ b/sycl/source/detail/pi.cpp
|
|
|
|
|
@@ -635,45 +635,47 @@ static uint16_t getELFHeaderType(const unsigned char *ImgData, size_t ImgSize) {
|
|
|
|
|
RT::PiDeviceBinaryType getBinaryImageFormat(const unsigned char *ImgData,
|
|
|
|
|
size_t ImgSize) {
|
|
|
|
|
// Top-level magic numbers for the recognized binary image formats.
|
|
|
|
|
- struct {
|
|
|
|
|
- RT::PiDeviceBinaryType Fmt;
|
|
|
|
|
- const uint32_t Magic;
|
|
|
|
|
- } Fmts[] = {{PI_DEVICE_BINARY_TYPE_SPIRV, 0x07230203},
|
|
|
|
|
- {PI_DEVICE_BINARY_TYPE_LLVMIR_BITCODE, 0xDEC04342},
|
|
|
|
|
- // 'I', 'N', 'T', 'C' ; Intel native
|
|
|
|
|
- {PI_DEVICE_BINARY_TYPE_NATIVE, 0x43544E49}};
|
|
|
|
|
- if (NOT DEFINED BOOST_MODULE_SRC_DIR)
|
|
|
|
|
- set(BOOST_MODULE_GIT_REPO "https://github.com/boostorg/${BOOST_MODULE_NAME}.git")
|
|
|
|
|
- message(STATUS "Source dir not set for boost module ${BOOST_MODULE_NAME}, downloading headers from ${BOOST_MODULE_GIT_REPO}")
|
|
|
|
|
-
|
|
|
|
|
- if (ImgSize >= sizeof(Fmts[0].Magic)) {
|
|
|
|
|
- detail::remove_const_t<decltype(Fmts[0].Magic)> Hdr = 0;
|
|
|
|
|
- std::copy(ImgData, ImgData + sizeof(Hdr), reinterpret_cast<char *>(&Hdr));
|
|
|
|
|
- set(BOOST_MODULE_FULL_NAME "boost_${BOOST_MODULE_NAME}")
|
|
|
|
|
- FetchContent_Declare(${BOOST_MODULE_FULL_NAME}
|
|
|
|
|
- GIT_REPOSITORY ${BOOST_MODULE_GIT_REPO}
|
|
|
|
|
- GIT_TAG ${BOOST_MODULE_GIT_TAG}
|
|
|
|
|
- )
|
|
|
|
|
- FetchContent_GetProperties(${BOOST_MODULE_FULL_NAME})
|
|
|
|
|
- FetchContent_MakeAvailable(${BOOST_MODULE_FULL_NAME})
|
|
|
|
|
-
|
|
|
|
|
- // Check headers for direct formats.
|
|
|
|
|
- for (const auto &Fmt : Fmts) {
|
|
|
|
|
- if (Hdr == Fmt.Magic)
|
|
|
|
|
- return Fmt.Fmt;
|
|
|
|
|
- }
|
|
|
|
|
- set(BOOST_MODULE_SRC_DIR ${${BOOST_MODULE_FULL_NAME}_SOURCE_DIR})
|
|
|
|
|
- else (NOT DEFINED BOOST_MODULE_SRC_DIR)
|
|
|
|
|
- message(STATUS "Using boost/${BOOST_MODULE_NAME} headers from ${BOOST_MODULE_SRC_DIR}")
|
|
|
|
|
- endif(NOT DEFINED BOOST_MODULE_SRC_DIR)
|
|
|
|
|
-
|
|
|
|
|
- // ELF e_type for recognized binary image formats.
|
|
|
|
|
- struct {
|
|
|
|
|
- RT::PiDeviceBinaryType Fmt;
|
|
|
|
|
- const uint16_t Magic;
|
|
|
|
|
- } ELFFmts[] = {{PI_DEVICE_BINARY_TYPE_NATIVE, 0xFF04}, // OpenCL executable
|
|
|
|
|
- {PI_DEVICE_BINARY_TYPE_NATIVE, 0xFF12}}; // ZEBIN executable
|
|
|
|
|
- set(BOOST_UNORDERED_INCLUDE_DIRS ${BOOST_UNORDERED_INCLUDE_DIRS} "${BOOST_MODULE_SRC_DIR}/include" PARENT_SCOPE)
|
|
|
|
|
-endfunction(add_boost_module_headers)
|
|
|
|
|
-
|
|
|
|
|
- // ELF files need to be parsed separately. The header type ends after 18
|
|
|
|
|
- // bytes.
|
|
|
|
|
- if (Hdr == 0x464c457F && ImgSize >= 18) {
|
|
|
|
|
- uint16_t HdrType = getELFHeaderType(ImgData, ImgSize);
|
|
|
|
|
- for (const auto &ELFFmt : ELFFmts) {
|
|
|
|
|
- if (HdrType == ELFFmt.Magic)
|
|
|
|
|
- return ELFFmt.Fmt;
|
|
|
|
|
- }
|
|
|
|
|
- // Newer ZEBIN format does not have a special header type, but can instead
|
|
|
|
|
- // be identified by having a required .ze_info section.
|
|
|
|
|
- if (checkELFSectionPresent(".ze_info", ImgData, ImgSize))
|
|
|
|
|
- return PI_DEVICE_BINARY_TYPE_NATIVE;
|
|
|
|
|
- }
|
|
|
|
|
+ auto MatchMagicNumber = [&](auto Number) {
|
|
|
|
|
+ return ImgSize >= sizeof(Number) &&
|
|
|
|
|
+ std::memcmp(ImgData, &Number, sizeof(Number)) == 0;
|
|
|
|
|
+ };
|
|
|
|
|
+
|
|
|
|
|
+ if (MatchMagicNumber(uint32_t{0x07230203}))
|
|
|
|
|
+ return PI_DEVICE_BINARY_TYPE_SPIRV;
|
|
|
|
|
+
|
|
|
|
|
+ if (MatchMagicNumber(uint32_t{0xDEC04342}))
|
|
|
|
|
+ return PI_DEVICE_BINARY_TYPE_LLVMIR_BITCODE;
|
|
|
|
|
+
|
|
|
|
|
+ if (MatchMagicNumber(uint32_t{0x43544E49}))
|
|
|
|
|
+ // 'I', 'N', 'T', 'C' ; Intel native
|
|
|
|
|
+ return PI_DEVICE_BINARY_TYPE_LLVMIR_BITCODE;
|
|
|
|
|
+
|
|
|
|
|
+ // Check for ELF format, size requirements include data we'll read in case of
|
|
|
|
|
+ // succesful match.
|
|
|
|
|
+ if (ImgSize >= 18 && MatchMagicNumber(uint32_t{0x464c457F})) {
|
|
|
|
|
+ uint16_t ELFHdrType = getELFHeaderType(ImgData, ImgSize);
|
|
|
|
|
+ if (ELFHdrType == 0xFF04)
|
|
|
|
|
+ // OpenCL executable.
|
|
|
|
|
+ return PI_DEVICE_BINARY_TYPE_NATIVE;
|
|
|
|
|
+
|
|
|
|
|
+ if (ELFHdrType == 0xFF12)
|
|
|
|
|
+ // ZEBIN executable.
|
|
|
|
|
+ return PI_DEVICE_BINARY_TYPE_NATIVE;
|
|
|
|
|
+
|
|
|
|
|
+ // Newer ZEBIN format does not have a special header type, but can instead
|
|
|
|
|
+ // be identified by having a required .ze_info section.
|
|
|
|
|
+ if (checkELFSectionPresent(".ze_info", ImgData, ImgSize))
|
|
|
|
|
+ return PI_DEVICE_BINARY_TYPE_NATIVE;
|
|
|
|
|
}
|
|
|
|
|
+
|
|
|
|
|
+ // "ar" format is used to pack binaries for multiple devices, e.g. via
|
|
|
|
|
+ //
|
|
|
|
|
+ // -Xsycl-target-backend=spir64_gen "-device acm-g10,acm-g11"
|
|
|
|
|
+ //
|
|
|
|
|
+ // option.
|
|
|
|
|
+ if (MatchMagicNumber(std::array{'!', '<', 'a', 'r', 'c', 'h', '>', '\n'}))
|
|
|
|
|
+ return PI_DEVICE_BINARY_TYPE_NATIVE;
|
|
|
|
|
+
|
|
|
|
|
return PI_DEVICE_BINARY_TYPE_NONE;
|
|
|
|
|
}
|
|
|
|
|
-set(BOOST_UNORDERED_GIT_TAG bd24dfd284dbc70e7521915af0d8d049f74a1e85)
|
|
|
|
|
-# Author: joaquintides <joaquin@tid.es>
|
|
|
|
|
-# Date: Tue Jul 18 18:19:13 2023 +0200
|
|
|
|
|
-#
|
|
|
|
|
-# updated concurrent map benchmark plots
|
|
|
|
|
-add_boost_module_headers(NAME "unordered" SRC_DIR ${BOOST_UNORDERED_SOURCE_DIR} GIT_TAG ${BOOST_UNORDERED_GIT_TAG})
|
|
|
|
|
-
|
|
|
|
|
-set(BOOST_ASSERT_GIT_TAG 02256c84fd0cd58a139d9dc1b25b5019ca976ada)
|
|
|
|
|
-# Author: Peter Dimov <pdimov@gmail.com>
|
|
|
|
|
-# Date: Thu Jun 22 18:11:58 2023 +0300
|
|
|
|
|
-#
|
|
|
|
|
-# Do not use std::source_location::current under nvcc. Fixes #32.
|
|
|
|
|
-add_boost_module_headers(NAME "assert" SRC_DIR ${BOOST_ASSERT_SOURCE_DIR} GIT_TAG ${BOOST_ASSERT_GIT_TAG})
|
|
|
|
|
-
|
|
|
|
|
-set(BOOST_CONFIG_GIT_TAG a1cf5d531405e62927b0257b5cbecc66a545b508)
|
|
|
|
|
-# Merge: f5726a26 a1edcd56
|
|
|
|
|
-# Author: jzmaddock <john@johnmaddock.co.uk>
|
|
|
|
|
-# Date: Sat Apr 15 13:20:12 2023 +0100
|
|
|
|
|
-#
|
|
|
|
|
-# Merge pull request #475 from boostorg/ci_2023_04
|
|
|
|
|
-add_boost_module_headers(NAME "config" SRC_DIR ${BOOST_CONFIG_SOURCE_DIR} GIT_TAG ${BOOST_CONFIG_GIT_TAG})
|
|
|
|
|
-
|
|
|
|
|
-set(BOOST_CONTAINER_HASH_GIT_TAG 226eb066e949adbf37b220e993d64ecefeeaae99)
|
|
|
|
|
-# Author: Peter Dimov <pdimov@gmail.com>
|
|
|
|
|
-# Date: Thu Jun 29 14:38:53 2023 +0300
|
|
|
|
|
-#
|
|
|
|
|
-# Update .drone.jsonnet
|
|
|
|
|
-add_boost_module_headers(NAME "container_hash" SRC_DIR ${BOOST_CONTAINER_HASH_SOURCE_DIR} GIT_TAG ${BOOST_CONTAINER_HASH_GIT_TAG})
|
|
|
|
|
-
|
|
|
|
|
-set(BOOST_CORE_GIT_TAG 216999e552e7f73e63c7bcc88b8ce9c179bbdbe2)
|
|
|
|
|
-# Author: Peter Dimov <pdimov@gmail.com>
|
|
|
|
|
-# Date: Sun Jun 25 13:46:53 2023 +0300
|
|
|
|
|
-#
|
|
|
|
|
-# Avoid -Wsign-conversion warning in checked_delete.hpp
|
|
|
|
|
-add_boost_module_headers(NAME "core" SRC_DIR ${BOOST_CORE_SOURCE_DIR} GIT_TAG ${BOOST_CORE_GIT_TAG})
|
|
|
|
|
-
|
|
|
|
|
-# Describe is a dependency of container_hash
|
|
|
|
|
-set(BOOST_DESCRIBE_GIT_TAG a0eafb08100eb15a57b6dae6d270c0012a56aa21)
|
|
|
|
|
-# Merge: 1692c3e b54fda5
|
|
|
|
|
-# Author: Peter Dimov <pdimov@gmail.com>
|
|
|
|
|
-# Date: Sun May 21 04:51:35 2023 +0300
|
|
|
|
|
-#
|
|
|
|
|
-# Merge branch 'fix-deprecated-inline-static-variables' of https://github.com/Romain-Geissler-1A/describe into feature/pr-40
|
|
|
|
|
-add_boost_module_headers(NAME "describe" SRC_DIR ${BOOST_DESCRIBE_SOURCE_DIR} GIT_TAG ${BOOST_DESCRIBE_GIT_TAG})
|
|
|
|
|
-
|
|
|
|
|
-set(BOOST_MOVE_GIT_TAG f1fbb45134065deebe95249c616a967d4b66c809)
|
|
|
|
|
-# Author: Ion Gaztañaga <igaztanaga@gmail.com>
|
|
|
|
|
-# Date: Mon Mar 13 13:32:29 2023 +0100
|
|
|
|
|
-#
|
|
|
|
|
-# Use [[msvc::intrinsic] attribute if available in move/forward in order to improve debug experience
|
|
|
|
|
-add_boost_module_headers(NAME "move" SRC_DIR ${BOOST_MOVE_SOURCE_DIR} GIT_TAG ${BOOST_MOVE_GIT_TAG})
|
|
|
|
|
-
|
|
|
|
|
-# Reuse mp11 fetched earlier for DPC++ headers
|
|
|
|
|
-set(BOOST_UNORDERED_INCLUDE_DIRS ${BOOST_UNORDERED_INCLUDE_DIRS} "${BOOST_MP11_SOURCE_DIR}/include/")
|
|
|
|
|
-
|
|
|
|
|
-set(BOOST_PREDEF_GIT_TAG 392e4e767469e3469c9390f0d9cca16724dc3fc8)
|
|
|
|
|
-# Merge: a12c7fd 499d28e
|
|
|
|
|
-# Author: Rene Rivera <grafikrobot@gmail.com>
|
|
|
|
|
-# Date: Sun Feb 27 14:44:35 2022 -0600
|
|
|
|
|
-#
|
|
|
|
|
-# Release 1.14.
|
|
|
|
|
-add_boost_module_headers(NAME "predef" SRC_DIR ${BOOST_PREDEF_SOURCE_DIR} GIT_TAG ${BOOST_PREDEF_GIT_TAG})
|
|
|
|
|
-
|
|
|
|
|
-set(BOOST_PREPROCESSOR_GIT_TAG 667e87b3392db338a919cbe0213979713aca52e3)
|
|
|
|
|
-# Author: Peter Dimov <pdimov@gmail.com>
|
|
|
|
|
-# Date: Tue Aug 16 20:59:52 2022 +0300
|
|
|
|
|
-#
|
|
|
|
|
-# Change C test names to not conflict with the C++ ones
|
|
|
|
|
-add_boost_module_headers(NAME "preprocessor" SRC_DIR ${BOOST_PREPROCESSOR_SOURCE_DIR} GIT_TAG ${BOOST_PREPROCESSOR_GIT_TAG})
|
|
|
|
|
-
|
|
|
|
|
-set(BOOST_STATIC_ASSERT_GIT_TAG 45eec41c293bc5cd36ec3ed83671f70bc1aadc9f)
|
|
|
|
|
-# Merge: ba72d33 a1abfec
|
|
|
|
|
-# Author: jzmaddock <john@johnmaddock.co.uk>
|
|
|
|
|
-# Date: Tue Mar 8 09:35:50 2022 +0000
|
|
|
|
|
-#
|
|
|
|
|
-# Merge pull request #15 from sdarwin/githubactions
|
|
|
|
|
-add_boost_module_headers(NAME "static_assert" SRC_DIR ${BOOST_STATIC_ASSERT_SOURCE_DIR} GIT_TAG ${BOOST_STATIC_ASSERT_GIT_TAG})
|
|
|
|
|
-
|
|
|
|
|
-set(BOOST_THROW_EXCEPTION_GIT_TAG 23dd41e920ecd91237500ac6428f7d392a7a875c)
|
|
|
|
|
-# Author: Peter Dimov <pdimov@gmail.com>
|
|
|
|
|
-# Date: Sun Jun 25 16:12:57 2023 +0300
|
|
|
|
|
-#
|
|
|
|
|
-# Update ci.yml
|
|
|
|
|
-add_boost_module_headers(NAME "throw_exception" SRC_DIR ${BOOST_THROW_EXCEPTION_SOURCE_DIR} GIT_TAG ${BOOST_THROW_EXCEPTION_GIT_TAG})
|
|
|
|
|
-
|
|
|
|
|
-set(BOOST_TUPLE_GIT_TAG 500e4fa0a2845b96c0dd919e7485e0f216438a01)
|
|
|
|
|
-# Merge: aa16ae3 ded3c1d
|
|
|
|
|
-# Author: Joel de Guzman <djowel@gmail.com>
|
|
|
|
|
-# Date: Thu Dec 30 23:20:18 2021 +0800
|
|
|
|
|
-#
|
|
|
|
|
-# Merge pull request #21 from igaztanaga/patch-1
|
|
|
|
|
-add_boost_module_headers(NAME "tuple" SRC_DIR ${BOOST_TUPLE_SOURCE_DIR} GIT_TAG ${BOOST_TUPLE_GIT_TAG})
|
|
|
|
|
-
|
|
|
|
|
-set(BOOST_TYPE_TRAITS_GIT_TAG 89f5011b4a79d91e42735670e39f72cb25c86c72)
|
|
|
|
|
-# Merge: 55feb75 1ebd31e
|
|
|
|
|
-# Author: John Maddock <john@johnmaddock.co.uk>
|
|
|
|
|
-# Date: Fri Feb 24 18:02:30 2023 +0000
|
|
|
|
|
-#
|
|
|
|
|
-# Merge branch 'develop'
|
|
|
|
|
-add_boost_module_headers(NAME "type_traits" SRC_DIR ${BOOST_TYPE_TRAITS_SOURCE_DIR} GIT_TAG ${BOOST_TYPE_TRAITS_GIT_TAG})
|
|
|
|
|
diff --git a/sycl/source/CMakeLists.txt b/sycl/source/CMakeLists.txt
|
|
|
|
|
index ead8f2c83ab7..6fb8305a1a88 100644
|
|
|
|
|
--- a/sycl/source/CMakeLists.txt
|
|
|
|
|
+++ b/sycl/source/CMakeLists.txt
|
|
|
|
|
@@ -69,8 +69,6 @@ function(add_sycl_rt_library LIB_NAME LIB_OBJ_NAME)
|
|
|
|
|
target_link_libraries(${LIB_NAME} PRIVATE ${ARG_XPTI_LIB})
|
|
|
|
|
endif()
|
|
|
|
|
|
|
|
|
|
- target_include_directories(${LIB_OBJ_NAME} PRIVATE ${BOOST_UNORDERED_INCLUDE_DIRS})
|
|
|
|
|
-
|
|
|
|
|
# pi_win_proxy_loader
|
|
|
|
|
if (WIN32)
|
|
|
|
|
include_directories(${LLVM_EXTERNAL_SYCL_SOURCE_DIR}/pi_win_proxy_loader)
|
|
|
|
|
diff --git a/sycl/source/detail/kernel_program_cache.hpp b/sycl/source/detail/kernel_program_cache.hpp
|
|
|
|
|
index 87a41d9fe105..c0a572b4d144 100644
|
|
|
|
|
--- a/sycl/source/detail/kernel_program_cache.hpp
|
|
|
|
|
+++ b/sycl/source/detail/kernel_program_cache.hpp
|
|
|
|
|
@@ -18,12 +18,10 @@
|
|
|
|
|
|
|
|
|
|
#include <atomic>
|
|
|
|
|
#include <condition_variable>
|
|
|
|
|
+#include <map>
|
|
|
|
|
#include <mutex>
|
|
|
|
|
#include <type_traits>
|
|
|
|
|
|
|
|
|
|
-#include <boost/unordered/unordered_flat_map.hpp>
|
|
|
|
|
-#include <boost/unordered_map.hpp>
|
|
|
|
|
-
|
|
|
|
|
// For testing purposes
|
|
|
|
|
class MockKernelProgramCache;
|
|
|
|
|
|
|
|
|
|
@@ -113,8 +111,8 @@ public:
|
|
|
|
|
std::pair<std::uintptr_t, sycl::detail::pi::PiDevice>;
|
|
|
|
|
|
|
|
|
|
struct ProgramCache {
|
|
|
|
|
- ::boost::unordered_map<ProgramCacheKeyT, ProgramBuildResultPtr> Cache;
|
|
|
|
|
- ::boost::unordered_multimap<CommonProgramKeyT, ProgramCacheKeyT> KeyMap;
|
|
|
|
|
+ std::map<ProgramCacheKeyT, ProgramBuildResultPtr> Cache;
|
|
|
|
|
+ std::multimap<CommonProgramKeyT, ProgramCacheKeyT> KeyMap;
|
|
|
|
|
|
|
|
|
|
size_t size() const noexcept { return Cache.size(); }
|
|
|
|
|
};
|
|
|
|
|
@@ -138,10 +136,8 @@ public:
|
|
|
|
|
};
|
|
|
|
|
using KernelBuildResultPtr = std::shared_ptr<KernelBuildResult>;
|
|
|
|
|
|
|
|
|
|
- using KernelByNameT =
|
|
|
|
|
- ::boost::unordered_map<std::string, KernelBuildResultPtr>;
|
|
|
|
|
- using KernelCacheT =
|
|
|
|
|
- ::boost::unordered_map<sycl::detail::pi::PiProgram, KernelByNameT>;
|
|
|
|
|
+ using KernelByNameT = std::map<std::string, KernelBuildResultPtr>;
|
|
|
|
|
+ using KernelCacheT = std::map<sycl::detail::pi::PiProgram, KernelByNameT>;
|
|
|
|
|
|
|
|
|
|
using KernelFastCacheKeyT =
|
|
|
|
|
std::tuple<SerializedObj, sycl::detail::pi::PiDevice, std::string,
|
|
|
|
|
@@ -149,13 +145,7 @@ public:
|
|
|
|
|
using KernelFastCacheValT =
|
|
|
|
|
std::tuple<sycl::detail::pi::PiKernel, std::mutex *,
|
|
|
|
|
const KernelArgMask *, sycl::detail::pi::PiProgram>;
|
|
|
|
|
- // This container is used as a fast path for retrieving cached kernels.
|
|
|
|
|
- // unordered_flat_map is used here to reduce lookup overhead.
|
|
|
|
|
- // The slow path is used only once for each newly created kernel, so the
|
|
|
|
|
- // higher overhead of insertion that comes with unordered_flat_map is more
|
|
|
|
|
- // of an issue there. For that reason, those use regular unordered maps.
|
|
|
|
|
- using KernelFastCacheT =
|
|
|
|
|
- ::boost::unordered_flat_map<KernelFastCacheKeyT, KernelFastCacheValT>;
|
|
|
|
|
+ using KernelFastCacheT = std::map<KernelFastCacheKeyT, KernelFastCacheValT>;
|
|
|
|
|
|
|
|
|
|
~KernelProgramCache() = default;
|
|
|
|
|
|
|
|
|
|
diff --git a/sycl/unittests/CMakeLists.txt b/sycl/unittests/CMakeLists.txt
|
|
|
|
|
index 71d2413c2974..0c55c870c4b8 100644
|
|
|
|
|
--- a/sycl/unittests/CMakeLists.txt
|
|
|
|
|
+++ b/sycl/unittests/CMakeLists.txt
|
|
|
|
|
@@ -1,6 +1,5 @@
|
|
|
|
|
add_custom_target(SYCLUnitTests)
|
|
|
|
|
set_target_properties(SYCLUnitTests PROPERTIES FOLDER "SYCL tests")
|
|
|
|
|
-include_directories(${BOOST_UNORDERED_INCLUDE_DIRS})
|
|
|
|
|
|
|
|
|
|
foreach(flag_var
|
|
|
|
|
CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELEASE
|
|
|
|
|
--
|
|
|
|
|
2.30.1.windows.1
|
|
|
|
|
|
|
|
|
|
|
Xavier Hallade