Accepting request 1281194 from science

OBS-URL: https://build.opensuse.org/request/show/1281194
OBS-URL: https://build.opensuse.org/package/show/openSUSE:Factory/pocl?expand=0&rev=36

link_against_libclang-cpp_so.patch

@@ -1,13 +0,0 @@
-diff --git a/cmake/LLVM.cmake b/cmake/LLVM.cmake
-index 2827d64..e186bc1 100644
---- a/cmake/LLVM.cmake
-+++ b/cmake/LLVM.cmake
-@@ -239,7 +239,7 @@ string(STRIP "${LLVM_SYSLIBS}" LLVM_SYSLIBS)
- ####################################################################
- 
- # llvm-config does not include clang libs
--if((9 LESS LLVM_MAJOR) AND (NOT STATIC_LLVM))
-+if((8 LESS LLVM_MAJOR) AND (NOT STATIC_LLVM))
-   # For Clang 10+, link against a single shared library instead of multiple component shared
-   # libraries.
-   if("${LLVM_LIBNAMES}" MATCHES "LLVMTCE")

maint.README

@@ -0,0 +1,2 @@
+By default POCL uses the TBB backend. If libpocl-devices-tbb is not installed, the OpenCL runtime will not be able to find any devices.
+To force POCl to use the CPU backend, set the environment variable POCL_DEVICES=cpu.

pocl-3.1.tar.gz

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:82314362552e050aff417318dd623b18cf0f1d0f84f92d10a7e3750dd12d3a9a
-size 1928607

pocl-7.0.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f55caba8c3ce12bec7b683ce55104c7555e19457fc2ac72c6f035201e362be08
+size 3766036

pocl-llvm16.patch

@@ -1,203 +0,0 @@
-diff --git a/CMakeLists.txt b/CMakeLists.txt
-index 8c8741b6..85de3968 100644
---- a/CMakeLists.txt
-+++ b/CMakeLists.txt
-@@ -520,7 +520,7 @@ if(ENABLE_LLVM)
-   # of the code, since this macro adds another layer of complexity.
-   if(LLVM_VERSION VERSION_LESS_EQUAL 14.0)
-     set(ENABLE_LLVM_OPAQUE_POINTERS OFF CACHE INTERNAL "llvm opaque pointers" FORCE)
--  elseif(LLVM_VERSION VERSION_EQUAL 15.0)
-+  elseif(LLVM_VERSION VERSION_EQUAL 15.0 OR LLVM_VERSION VERSION_EQUAL 16.0)
-     option(ENABLE_LLVM_OPAQUE_POINTERS "Handle the change to llvm opaque pointers." ON)
-   else()
-     set(ENABLE_LLVM_OPAQUE_POINTERS ON CACHE INTERNAL "llvm opaque pointers" FORCE)
-diff --git a/cmake/LLVM.cmake b/cmake/LLVM.cmake
-index 18138f36..00ad2186 100644
---- a/cmake/LLVM.cmake
-+++ b/cmake/LLVM.cmake
-@@ -122,8 +122,10 @@ string(REPLACE "${LLVM_PREFIX}" "${LLVM_PREFIX_CMAKE}" LLVM_LIBDIR "${LLVM_LIBDI
- run_llvm_config(LLVM_INCLUDEDIR --includedir)
- string(REPLACE "${LLVM_PREFIX}" "${LLVM_PREFIX_CMAKE}" LLVM_INCLUDEDIR "${LLVM_INCLUDEDIR}")
- 
--run_llvm_config(LLVM_SRC_ROOT --src-root)
--run_llvm_config(LLVM_OBJ_ROOT --obj-root)
-+if(LLVM_VERSION_MAJOR LESS 16)
-+  run_llvm_config(LLVM_SRC_ROOT --src-root)
-+  run_llvm_config(LLVM_OBJ_ROOT --obj-root)
-+endif()
- string(REPLACE "${LLVM_PREFIX}" "${LLVM_PREFIX_CMAKE}" LLVM_OBJ_ROOT "${LLVM_OBJ_ROOT}")
- run_llvm_config(LLVM_ALL_TARGETS --targets-built)
- run_llvm_config(LLVM_HOST_TARGET --host-target)
-@@ -772,12 +774,35 @@ endif()
- 
- ####################################################################
- 
--if(ENABLE_HOST_CPU_DEVICES AND NOT DEFINED ${CL_DISABLE_HALF})
--  set(CL_DISABLE_HALF 0)
--  message(STATUS "Checking fp16 support")
--  custom_try_compile_clang_silent("__fp16 callfp16(__fp16 a) { return a * (__fp16)1.8; };" "__fp16 x=callfp16((__fp16)argc);" RESV ${CLANG_TARGET_OPTION}${LLC_TRIPLE} ${CLANG_MARCH_FLAG}${LLC_HOST_CPU})
--  if(RESV)
-+# Clang documentation on Language Extensions:
-+# __fp16 is supported on every target, as it is purely a storage format
-+# _Float16 is currently only supported on the following targets... SPIR, x86
-+# DIsabled for non-x86-64 because of limitations:
-+#     The _Float16 type requires SSE2 feature and above due to the instruction
-+#        limitations. When using it on i386 targets, you need to specify -msse2
-+#        explicitly.
-+#     For targets without F16C feature or above, please make sure:
-+#     Use GCC 12.0 and above if you are using libgcc.
-+#     If you are using compiler-rt, use the same version with the compiler.
-+#        Early versions provided FP16 builtins in a different ABI. A workaround is
-+#        to use a small code snippet to check the ABI if you cannot make sure of it.
-+
-+if(ENABLE_HOST_CPU_DEVICES AND NOT DEFINED CL_DISABLE_HALF)
-+  # LLVM <15 doesn't support FP16 emulation
-+  # LLVM 15 crashes on some code, with the FP16 emulation
-+  # CONFORMANCE disables FP16 b/c it's incomplete
-+  # enabled on x86-64 only for now
-+  if((LLVM_VERSION_MAJOR LESS 16) OR ENABLE_CONFORMANCE OR (NOT X86_64))
-+    message(STATUS "FP16 support disabled")
-     set(CL_DISABLE_HALF 1)
-+  else()
-+    set(CL_DISABLE_HALF 0)
-+    message(STATUS "Checking fp16 support")
-+    custom_try_compile_clang_silent("_Float16 callfp16(_Float16 a) { return a * 1.8f16; };" "_Float16 x=callfp16((_Float16)argc);"
-+      RESV ${CLANG_TARGET_OPTION}${LLC_TRIPLE} ${CLANG_MARCH_FLAG}${LLC_HOST_CPU})
-+    if(RESV)
-+      set(CL_DISABLE_HALF 1)
-+    endif()
-   endif()
- endif()
- 
-diff --git a/cmake/bitcode_rules.cmake b/cmake/bitcode_rules.cmake
-index 3d1a373f..10a309c2 100644
---- a/cmake/bitcode_rules.cmake
-+++ b/cmake/bitcode_rules.cmake
-@@ -167,7 +167,7 @@ function(compile_ll_to_bc FILENAME SUBDIR BCLIST)
-     list(APPEND ${BCLIST} "${BC_FILE}")
-     set(${BCLIST} ${${BCLIST}} PARENT_SCOPE)
- 
--    if(LLVM_VERSION VERSION_EQUAL 15.0)
-+    if(LLVM_VERSION VERSION_EQUAL 15.0 OR LLVM_VERSION VERSION_EQUAL 16.0)
-       # both of these are necesssary. some of the files (like barrier.ll)
-       # don't contain any pointers and thus cannot be guessed; if llvm-as
-       # produces the wrong opaque-type file, later llvm-link will fail
-diff --git a/include/_kernel_c.h b/include/_kernel_c.h
-index e1038ea5..73ad9d18 100644
---- a/include/_kernel_c.h
-+++ b/include/_kernel_c.h
-@@ -112,7 +112,7 @@ typedef uint uint16 __attribute__((__ext_vector_type__(16)));
- /* NOTE: the Clang's __fp16 does not work robustly in C mode,
-    it might produce invalid code at least with half vectors.
-    Using the native 'half' type in OpenCL C mode works better. */
--typedef __fp16 half;
-+typedef _Float16 half;
- 
- typedef half half2  __attribute__((__ext_vector_type__(2)));
- typedef half half3  __attribute__((__ext_vector_type__(3)));
-diff --git a/lib/CL/pocl_llvm_build.cc b/lib/CL/pocl_llvm_build.cc
-index 3ef6e0c5..84413a48 100644
---- a/lib/CL/pocl_llvm_build.cc
-+++ b/lib/CL/pocl_llvm_build.cc
-@@ -263,7 +263,7 @@ int pocl_llvm_build_program(cl_program program,
- 
-   size_t fastmath_flag = user_options.find("-cl-fast-relaxed-math");
- 
--#if (CLANG_MAJOR == 15)
-+#if (CLANG_MAJOR == 15) || (CLANG_MAJOR == 16)
- #ifdef LLVM_OPAQUE_POINTERS
-   ss << "-opaque-pointers ";
- #else
-diff --git a/lib/CL/pocl_llvm_utils.cc b/lib/CL/pocl_llvm_utils.cc
-index 3cc7fcc5..0690af17 100644
---- a/lib/CL/pocl_llvm_utils.cc
-+++ b/lib/CL/pocl_llvm_utils.cc
-@@ -273,7 +273,9 @@ void InitializeLLVM() {
-     initializeAnalysis(Registry);
-     initializeTransformUtils(Registry);
-     initializeInstCombine(Registry);
-+#ifdef LLVM_OLDER_THAN_16_0
-     initializeInstrumentation(Registry);
-+#endif
-     initializeTarget(Registry);
-   }
- 
-@@ -390,7 +392,7 @@ void pocl_llvm_create_context(cl_context ctx) {
- 
-   data->Context = new llvm::LLVMContext();
-   assert(data->Context);
--#if (CLANG_MAJOR == 15)
-+#if (CLANG_MAJOR == 15) || (CLANG_MAJOR == 16)
- #ifdef LLVM_OPAQUE_POINTERS
-   data->Context->setOpaquePointers(true);
- #else
-diff --git a/lib/llvmopencl/BarrierTailReplication.cc b/lib/llvmopencl/BarrierTailReplication.cc
-index 98bc0635..43802f73 100644
---- a/lib/llvmopencl/BarrierTailReplication.cc
-+++ b/lib/llvmopencl/BarrierTailReplication.cc
-@@ -352,7 +352,11 @@ BarrierTailReplication::ReplicateBasicBlocks(BasicBlockVector &new_graph,
-          i2 != e2; ++i2) {
-       Instruction *i = i2->clone();
-       reference_map.insert(std::make_pair(&*i2, i));
-+#ifdef LLVM_OLDER_THAN_16_0
-       new_b->getInstList().push_back(i);
-+#else
-+      i->insertInto(new_b, new_b->end());
-+#endif
-     }
- 
-     // Add predicates to PHINodes of basic blocks the replicated
-diff --git a/lib/llvmopencl/ParallelRegion.cc b/lib/llvmopencl/ParallelRegion.cc
-index dfe56e03..639bd686 100644
---- a/lib/llvmopencl/ParallelRegion.cc
-+++ b/lib/llvmopencl/ParallelRegion.cc
-@@ -201,12 +201,17 @@ ParallelRegion::chainAfter(ParallelRegion *region)
- #endif
- 
-   BasicBlock *successor = t->getSuccessor(0);
--  Function::BasicBlockListType &bb_list = 
--    successor->getParent()->getBasicBlockList();
--  
--  for (iterator i = begin(), e = end(); i != e; ++i)
-+  Function *F = successor->getParent();
- 
-+#ifdef LLVM_OLDER_THAN_16_0
-+  Function::BasicBlockListType &bb_list =
-+    F->getBasicBlockList();
-+  for (iterator i = begin(), e = end(); i != e; ++i)
-     bb_list.insertAfter(tail->getIterator(), *i);
-+#else
-+  for (iterator i = begin(), e = end(); i != e; ++i)
-+    F->insert(tail->getIterator(), *i);
-+#endif
-   t->setSuccessor(0, entryBB());
- 
-   t = exitBB()->getTerminator();
-diff --git a/lib/llvmopencl/WorkItemAliasAnalysis.cc b/lib/llvmopencl/WorkItemAliasAnalysis.cc
-index 02dcd2ec..8a85b25a 100644
---- a/lib/llvmopencl/WorkItemAliasAnalysis.cc
-+++ b/lib/llvmopencl/WorkItemAliasAnalysis.cc
-@@ -51,6 +51,12 @@ using namespace llvm;
- typedef llvm::MemoryLocation Location;
- typedef llvm::AliasResult AliasResult;
- 
-+#ifdef LLVM_OLDER_THAN_16_0
-+#define AAResultB AAResultBase<WorkItemAAResult>
-+#else
-+#define AAResultB AAResultBase
-+#endif
-+
- /// WorkItemAliasAnalysis - This is a simple alias analysis
- /// implementation that uses pocl metadata to make sure memory accesses from
- /// different work items are not aliasing.
-@@ -58,8 +64,8 @@ typedef llvm::AliasResult AliasResult;
- 
- // LLVM 3.8+
- 
--class WorkItemAAResult : public AAResultBase<WorkItemAAResult> {
--    friend AAResultBase<WorkItemAAResult>;
-+class WorkItemAAResult : public AAResultB {
-+    friend AAResultB;
- 
- public:
-     static char ID;

pocl.changes

@@ -1,3 +1,68 @@
+-------------------------------------------------------------------
+Thu May 22 15:18:21 UTC 2025 - Eyad Issa <eyadlorenzo@gmail.com>
+
+- Drop the _service file
+- Use Source Urls instead of obs_scm
+- Add option to build with the vulkan experimental device
+- Build with level0 experimental device
+- Update to version 7.0:
+  * PoCL received the official OpenCL 3.0 conformance stamp with the
+    CPU (x86-64) and Level Zero drivers.
+  * Support for LLVM versions 19 and 20
+  * Support for dynamic device addition and network discovery.
+    This new feature allows discovery of remote servers located in
+    LAN or WAN environments and enables runtime addition of
+    discovered devices to the remote client's platform. Network
+    discovery is performed thorugh mDNS, unicast-DNS-SD, and
+    DHT-based mechanisms, using Avahi and OpenDHT libraries.
+  * New experimental support for Defined Built-in Kernels (DBK) has
+    been added to the CPU drivers
+  * The old "work-item replication" work-group function generation
+    method was removed to clean up the kernel compiler.
+  * Full changes at
+    /usr/share/doc/packages/pocl/notes_7_0.rst
+
+-------------------------------------------------------------------
+Mon Jan  6 13:30:37 UTC 2025 - Eyad Issa <eyadlorenzo@gmail.com>
+
+- Disable SPIR-V Support on Leap 15.x (missing llvm-spirv)
+- Enable CPU TBB driver
+- Add note for users/packagers in maint.README.
+
+-------------------------------------------------------------------
+Sat Dec 28 01:52:48 UTC 2024 - Eyad Issa <eyadlorenzo@gmail.com>
+
+- Use _service file to create tarball from git
+- Enable SPIR-V support
+
+- Remove pocl-llvm16.patch
+- Remove link_against_libclang-cpp_so.patch
+
+- Update to version 6.0:
+  * New CPU driver which uses Threading Building Blocks for
+    task scheduling
+  * Experimental cl_ext_buffer_device_address prototype
+  * Multi-device command buffer infrastructure
+  * Command queue priority/throttle hints
+  * Support for using OpenMP for task scheduling in the
+    "cpu" driver
+  * Improvements to the Remote driver
+  * Improvements to the Level Zero driver
+  * Improvements to the CUDA driver
+
+- Update to version 5.0:
+  * New backend: Remote
+  * CUDA driver: More OpenCL 3.0 features and extensions
+    implemented
+  * RISC-V CPU support improved
+  * Other CPU improvements
+  * Experimental built-in-kernel library based FPGA backend
+
+- Update to version 4.0:
+  * Support for Clang/LLVM 16.0
+  * CPU driver improvements
+  * New Level Zero driver
+
 -------------------------------------------------------------------
 Thu Feb 29 14:57:59 UTC 2024 - Dominique Leuenberger <dimstar@opensuse.org>
 
@@ -445,4 +510,3 @@ Tue Oct  7 19:16:42 UTC 2014 - mardnh@gmx.de
 
 - initial stable package, version 0.10
   based on home:mnhauke:opencl:testing/pocl
-

pocl.spec

@@ -1,7 +1,7 @@
 #
 # spec file for package pocl
 #
-# Copyright (c) 2023 SUSE LLC
+# Copyright (c) 2025 SUSE LLC
 # Copyright (c) 2014 Guillaume GARDET <guillaume@opensuse.org>
 #
 # All modifications and additions to the file contributed by third parties
@@ -17,38 +17,65 @@
 #
 
 
-%define sover  2
+%global sover  2
+
+%if 0%{?suse_version} >= 1600
+  %bcond_without  spirv
+%else
+  %bcond_with     spirv
+%endif
+
+%if 0%{?suse_version} >= 1600
+  %bcond_without  ttb
+%else
+  %bcond_with     ttb
+%endif
+%if 0%{?suse_version} >= 1600
+  %bcond_without  levelzero
+%else
+  %bcond_with     levelzero
+%endif
+
+%bcond_with vulkan
+
 Name:           pocl
-Version:        3.1
+Version:        7.0
 Release:        0
 Summary:        Portable Computing Language - an OpenCL implementation
 # The whole code is under MIT
 # except include/utlist.h which is under BSD (and unbundled)
 License:        MIT
 Group:          Development/Tools/Other
-URL:            http://portablecl.org/
+URL:            https://portablecl.org/
 Source0:        https://github.com/pocl/pocl/archive/v%{version}.tar.gz#/%{name}-%{version}.tar.gz
+Source98:       maint.README
 Source99:       pocl-rpmlintrc
-Patch0:         link_against_libclang-cpp_so.patch
+# Version 7.0: Supports LLVM versions 19 and 20
-# PATCH-FIX-UPSTREAM - pocl-llvm16.patch - enables support for LLVM 16
+BuildRequires:  ((clang-devel >= 19 with clang-devel < 21) or clang20-devel)
-# https://github.com/pocl/pocl/commit/20d1bfa9bfd301964f7b2fc6d7f4589dd04e1b5c
+BuildRequires:  cmake >= 3.15
-# https://github.com/pocl/pocl/commit/bf50f0052e4248cd1acfaaa8da95c5e4ca52f815
-Patch1:         pocl-llvm16.patch
-BuildRequires:  cmake
 BuildRequires:  gcc-c++
 BuildRequires:  ninja
-BuildRequires:  ocl-icd-devel
 BuildRequires:  opencl-headers
 BuildRequires:  pkgconfig
-%if 0%{?suse_version} <= 1500 && 0%{?sle_version} > 150500
+BuildRequires:  pkgconfig(OpenCL)
-BuildRequires:  ((clang-devel >= 6.0.0 with clang-devel < 17) or clang15-devel)
-%else
-BuildRequires:  ((clang-devel >= 6.0.0 with clang-devel < 17) or clang16-devel)
-%endif
 BuildRequires:  pkgconfig(hwloc)
+BuildRequires:  pkgconfig(ocl-icd)
 # PPC has limited support/testing from upstream
 # s390(x) is also not supported, so use ExclusiveArch
 ExclusiveArch:  %{ix86} x86_64 %{arm} aarch64 riscv64
+%if %{with ttb}
+BuildRequires:  pkgconfig(tbb)
+%endif
+%if %{with spirv}
+BuildRequires:  libLLVMSPIRVLib-devel
+BuildRequires:  spirv-tools
+%endif
+%if %{with levelzero}
+BuildRequires:  pkgconfig(level-zero)
+%endif
+%if %{with vulkan}
+BuildRequires:  pkgconfig(vulkan)
+%endif
 
 %description
 Portable Computing Language (pocl) is an implementation of the OpenCL standard
@@ -67,6 +94,7 @@ and multithread.
 %package -n libpocl%{sover}
 Summary:        Shared Library part of pocl
 Group:          System/Libraries
+Recommends:     libpocl-devices-tbb = %{version}
 
 %description -n libpocl%{sover}
 Portable Computing Language (pocl) is an implementation of the OpenCL standard
@@ -75,6 +103,26 @@ heterogenous GPUs/accelerators.
 
 This subpackage contains the shared library part of pocl.
 
+%package -n libpocl-devices-tbb
+Summary:        TBB device for pocl
+
+%description -n libpocl-devices-tbb
+Portable Computing Language (pocl) is an implementation of the OpenCL standard
+which can be adapted for new targets and devices, both for homogeneous CPU and
+heterogenous GPUs/accelerators.
+
+This subpackage contains the Thread Building Blocks (TBB) device for pocl.
+
+%package -n libpocl-devices-levelzero
+Summary:        Level Zero device for pocl
+
+%description -n libpocl-devices-levelzero
+Portable Computing Language (pocl) is an implementation of the OpenCL standard
+which can be adapted for new targets and devices, both for homogeneous CPU and
+heterogenous GPUs/accelerators.
+
+This subpackage contains the Level Zero device for pocl.
+
 %package devel
 Summary:        Development files for the Portable Computing Language
 Group:          Development/Languages/Other
@@ -93,11 +141,31 @@ This subpackage provides the development files needed for pocl.
 %autosetup -p1
 
 %build
-%define __builder ninja
+%global __builder ninja
 %cmake \
-  -DENABLE_CUDA=0 \
-  -DENABLE_ICD=ON \
   -DPOCL_INSTALL_ICD_VENDORDIR=%{_datadir}/OpenCL/vendors \
+  -DENABLE_LLVM=ON \
+  -DWITH_LLVM_CONFIG=%{_bindir}/llvm-config \
+  -DENABLE_ICD=ON \
+  -DSTATIC_LLVM=OFF \
+  -DENABLE_REMOTE_DISCOVERY_AVAHI=ON \
+  -DENABLE_REMOTE_ADVERTISEMENT_AVAHI=ON \
+  -DENABLE_REMOTE_DISCOVERY_DHT=ON \
+  -DENABLE_REMOTE_ADVERTISEMENT_DHT=ON \
+  -DENABLE_CUDA=OFF \
+  -DINSTALL_OPENCL_HEADERS=OFF \
+%if %{with spirv}
+  -DENABLE_SPIRV=ON \
+%endif
+%if %{with ttb}
+  -DENABLE_TBB_DEVICE=ON \
+%endif
+%if %{with levelzero}
+  -DENABLE_LEVEL0=ON \
+%endif
+%if %{with vulkan}
+  -DENABLE_VULKAN=ON \
+%endif
 %ifarch %{ix86} x86_64
   -DKERNELLIB_HOST_CPU_VARIANTS=distro \
 %endif
@@ -113,7 +181,7 @@ This subpackage provides the development files needed for pocl.
 %if 0%{?suse_version} <= 1500 && 0%{?sle_version} <= 150300
   -DCMAKE_INSTALL_LIBDIR:PATH=%{_lib} \
 %endif
-  -DWITH_LLVM_CONFIG=%{_bindir}/llvm-config
+  %{nil}
 
 %cmake_build
 
@@ -142,4 +210,14 @@ This subpackage provides the development files needed for pocl.
 %{_libdir}/libpocl.so
 %{_libdir}/pkgconfig/pocl.pc
 
+%if %{with ttb}
+%files -n libpocl-devices-tbb
+%{_libdir}/pocl/libpocl-devices-tbb.so
+%endif
+
+%if %{with levelzero}
+%files -n libpocl-devices-levelzero
+%{_libdir}/pocl/libpocl-devices-level0.so
+%endif
+
 %changelog