[Openmp-commits] [PATCH] D86082: [libomptarget][NFC] Sort list of plugins in chronological order

George Rokos via Phabricator via Openmp-commits openmp-commits at lists.llvm.org
Mon Aug 17 08:29:04 PDT 2020

grokos created this revision.
grokos added reviewers: JonChesterfield, jdoerfert, RaviNarayanaswamy, gregrodgers, ABataev.
grokos added a project: OpenMP.
Herald added subscribers: kristof.beyls, tpr.
grokos requested review of this revision.
Herald added a subscriber: sstefan1.

At the Aug 12 2020 telecon we agreed to reorder the list of plugins chronologically.

Libomptarget was checked into the trunk when the generic-elf-64 and CUDA plugins had already been developed, so there is no evidence which plugin was developed first (see https://reviews.llvm.org/rG2467df6e4f04e3d0e8e78d662473ba1b87c0a885). Looking into IBM's clang-ykt branch doesn't help either - the oldest commit already contains both plugins. Because I was working for IBM when we started libomptarget development I can more or less remember what we did first and what we did last but this isn't documented anywhere.

In the very early days we developed the generic-elf-64 plugin as a means of having some prototype to test libomptarget with. Because development was done on PPC machines I assume that the ppc build of the generic-elf-64 plugin should come first. The x86 variant follows. The CUDA plugin was developed slightly later.

The rest of the plugins were enabled after libomptarget was upstreamed and the order can be found by browsing the commits. First we had support for aarch64 via the generic-elf-64 plugin (https://reviews.llvm.org/rG1e254c5295502526362efa1757359fdfaf92495d), then came support for SX-Aurora (https://reviews.llvm.org/rG6b9e43c67e0b109881db2524ec5207c4a888e7f7) and finally amdgpu (https://reviews.llvm.org/rGd0b312955f12beba5b03c8a524a8456cb4174bd7).

  rG LLVM Github Monorepo



Index: openmp/libomptarget/src/rtl.cpp
--- openmp/libomptarget/src/rtl.cpp
+++ openmp/libomptarget/src/rtl.cpp
@@ -23,12 +23,12 @@
 // List of all plugins that can support offloading.
 static const char *RTLNames[] = {
+    /* PowerPC target       */ "libomptarget.rtl.ppc64.so",
+    /* x86_64 target        */ "libomptarget.rtl.x86_64.so",
+    /* CUDA target          */ "libomptarget.rtl.cuda.so",
+    /* AArch64 target       */ "libomptarget.rtl.aarch64.so",
     /* SX-Aurora VE target  */ "libomptarget.rtl.ve.so",
-    /* PowerPC target */ "libomptarget.rtl.ppc64.so",
-    /* x86_64 target  */ "libomptarget.rtl.x86_64.so",
-    /* CUDA target    */ "libomptarget.rtl.cuda.so",
-    /* AArch64 target */ "libomptarget.rtl.aarch64.so",
-    /* AMDGPU target  */ "libomptarget.rtl.amdgpu.so",
+    /* AMDGPU target        */ "libomptarget.rtl.amdgpu.so",
 RTLsTy *RTLs;

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D86082.286022.patch
Type: text/x-patch
Size: 954 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/openmp-commits/attachments/20200817/189c507b/attachment-0001.bin>

More information about the Openmp-commits mailing list