From 2517438cae9831332178378078fc273cb8ffb466 Mon Sep 17 00:00:00 2001 From: Marek Szuba Date: Fri, 31 Aug 2018 14:27:16 +0100 Subject: [PATCH] dev-libs/beignet: disable optimisations broken on some platforms Certain optimisation introduced in 1.3.2 is now known not to work correctly on Skylake and Haswell systems. Upstream has been notified but has yet to respond. Upstream-Bug: https://bugs.freedesktop.org/show_bug.cgi?id=102137 Closes: https://bugs.gentoo.org/662760 Package-Manager: Portage-2.3.40, Repoman-2.3.9 --- dev-libs/beignet/beignet-1.3.2-r2.ebuild | 107 ++++++++++++++++++ ...t-1.3.2_disable-doNegAddOptimization.patch | 66 +++++++++++ 2 files changed, 173 insertions(+) create mode 100644 dev-libs/beignet/beignet-1.3.2-r2.ebuild create mode 100644 dev-libs/beignet/files/beignet-1.3.2_disable-doNegAddOptimization.patch diff --git a/dev-libs/beignet/beignet-1.3.2-r2.ebuild b/dev-libs/beignet/beignet-1.3.2-r2.ebuild new file mode 100644 index 000000000000..bc69b887c3d5 --- /dev/null +++ b/dev-libs/beignet/beignet-1.3.2-r2.ebuild @@ -0,0 +1,107 @@ +# Copyright 1999-2018 Gentoo Foundation +# Distributed under the terms of the GNU General Public License v2 + +EAPI=6 + +PYTHON_COMPAT=( python{2_7,3_4,3_5,3_6} ) +CMAKE_BUILD_TYPE="Release" + +inherit python-any-r1 cmake-multilib flag-o-matic llvm toolchain-funcs + +DESCRIPTION="OpenCL implementation for Intel GPUs" +HOMEPAGE="https://01.org/beignet" + +LICENSE="LGPL-2.1+" +SLOT="0" +IUSE="ocl-icd ocl20" + +if [[ "${PV}" == "9999" ]]; then + inherit git-r3 + EGIT_REPO_URI="https://anongit.freedesktop.org/git/beignet.git" + KEYWORDS="" +else + KEYWORDS="~amd64" + SRC_URI="https://01.org/sites/default/files/${P}-source.tar.gz" + S=${WORKDIR}/Beignet-${PV}-Source +fi + +COMMON="media-libs/mesa[${MULTILIB_USEDEP}] + =x11-libs/libdrm-2.4.70[video_cards_intel,${MULTILIB_USEDEP}] + x11-libs/libXext[${MULTILIB_USEDEP}] + x11-libs/libXfixes[${MULTILIB_USEDEP}]" +RDEPEND="${COMMON} + app-eselect/eselect-opencl" +DEPEND="${COMMON} + ${PYTHON_DEPS} + ocl-icd? ( dev-libs/ocl-icd ) + virtual/pkgconfig" + +LLVM_MAX_SLOT=6 + +PATCHES=( + "${FILESDIR}"/no-debian-multiarch.patch + "${FILESDIR}"/${PN}-1.3.2_disable-doNegAddOptimization.patch + "${FILESDIR}"/${PN}-1.3.2_cmake-llvm-config-multilib.patch + "${FILESDIR}"/${PN}-1.3.2_llvm6.patch + "${FILESDIR}"/${PN}-1.3.1-oclicd_no_upstream_icdfile.patch + "${FILESDIR}"/${PN}-1.2.0_no-hardcoded-cflags.patch + "${FILESDIR}"/llvm-terminfo.patch +) + +DOCS=( + docs/. +) + +pkg_pretend() { + if [[ ${MERGE_TYPE} != "binary" ]]; then + if tc-is-gcc; then + if [[ $(gcc-major-version) -eq 4 ]] && [[ $(gcc-minor-version) -lt 6 ]]; then + eerror "Compilation with gcc older than 4.6 is not supported" + die "Too old gcc found." + fi + fi + fi +} + +pkg_setup() { + llvm_pkg_setup + python_setup +} + +src_prepare() { + # See Bug #593968 + append-flags -fPIC + + cmake-utils_src_prepare + # We cannot run tests because they require permissions to access + # the hardware, and building them is very time-consuming. + cmake_comment_add_subdirectory utests +} + +multilib_src_configure() { + VENDOR_DIR="/usr/$(get_libdir)/OpenCL/vendors/${PN}" + + local mycmakeargs=( + -DCMAKE_INSTALL_PREFIX="${EPREFIX}${VENDOR_DIR}" + -DOCLICD_COMPAT=$(usex ocl-icd) + $(usex ocl20 "" "-DENABLE_OPENCL_20=OFF") + ) + + cmake-utils_src_configure +} + +multilib_src_install() { + VENDOR_DIR="/usr/$(get_libdir)/OpenCL/vendors/${PN}" + + cmake-utils_src_install + + insinto /etc/OpenCL/vendors/ + echo "${EPREFIX}${VENDOR_DIR}/lib/${PN}/libcl.so" > "${PN}-${ABI}.icd" || die "Failed to generate ICD file" + doins "${PN}-${ABI}.icd" + + dosym "lib/${PN}/libcl.so" "${VENDOR_DIR}"/libOpenCL.so.1 + dosym "lib/${PN}/libcl.so" "${VENDOR_DIR}"/libOpenCL.so + dosym "lib/${PN}/libcl.so" "${VENDOR_DIR}"/libcl.so.1 + dosym "lib/${PN}/libcl.so" "${VENDOR_DIR}"/libcl.so +} diff --git a/dev-libs/beignet/files/beignet-1.3.2_disable-doNegAddOptimization.patch b/dev-libs/beignet/files/beignet-1.3.2_disable-doNegAddOptimization.patch new file mode 100644 index 000000000000..ae399b17f053 --- /dev/null +++ b/dev-libs/beignet/files/beignet-1.3.2_disable-doNegAddOptimization.patch @@ -0,0 +1,66 @@ +--- b/backend/src/backend/gen_insn_selection_optimize.cpp ++++ a/backend/src/backend/gen_insn_selection_optimize.cpp +@@ -74,7 +74,8 @@ namespace gbe + const GenRegister& replacement) : + insn(insn), intermedia(intermedia), replacement(replacement) + { +- assert(insn.opcode == SEL_OP_MOV || insn.opcode == SEL_OP_ADD); ++ assert(insn.opcode == SEL_OP_MOV); ++ assert(&(insn.src(0)) == &replacement); + assert(&(insn.dst(0)) == &intermedia); + this->elements = CalculateElements(intermedia, insn.state.execWidth); + replacementOverwritten = false; +@@ -101,7 +102,6 @@ namespace gbe + void doReplacement(ReplaceInfo* info); + bool CanBeReplaced(const ReplaceInfo* info, const SelectionInstruction& insn, const GenRegister& var); + void cleanReplaceInfoMap(); +- void doNegAddOptimization(SelectionInstruction &insn); + + SelectionBlock &bb; + const ir::Liveness::LiveOut& liveout; +@@ -159,13 +159,8 @@ namespace gbe + + void SelBasicBlockOptimizer::addToReplaceInfoMap(SelectionInstruction& insn) + { +- assert(insn.opcode == SEL_OP_MOV || insn.opcode == SEL_OP_ADD); +- GenRegister &src = insn.src(0); +- if (insn.opcode == SEL_OP_ADD) { +- if (src.file == GEN_IMMEDIATE_VALUE) +- src = insn.src(1); +- } +- ++ assert(insn.opcode == SEL_OP_MOV); ++ const GenRegister& src = insn.src(0); + const GenRegister& dst = insn.dst(0); + if (src.type != dst.type || src.file != dst.file) + return; +@@ -254,29 +249,10 @@ namespace gbe + + if (insn.opcode == SEL_OP_MOV) + addToReplaceInfoMap(insn); +- +- doNegAddOptimization(insn); + } + cleanReplaceInfoMap(); + } + +- /* LLVM transform Mad(a, -b, c) to +- Add b, -b, 0 +- Mad val, a, b, c +- for Gen support negtive modifier, mad(a, -b, c) is native suppoted. +- Also it can be used for the same like instruction sequence. +- Do it just like a: mov b, -b, so it is a Mov operation like LocalCopyPropagation +- */ +- void SelBasicBlockOptimizer::doNegAddOptimization(SelectionInstruction &insn) { +- if (insn.opcode == SEL_OP_ADD) { +- GenRegister src0 = insn.src(0); +- GenRegister src1 = insn.src(1); +- if ((src0.negation && src1.file == GEN_IMMEDIATE_VALUE && src1.value.f == 0.0f) || +- (src1.negation && src0.file == GEN_IMMEDIATE_VALUE && src0.value.f == 0.0f)) +- addToReplaceInfoMap(insn); +- } +- } +- + void SelBasicBlockOptimizer::run() + { + for (size_t i = 0; i < MaxTries; ++i) { -- 2.26.2