[PATCH] D84363: [XCOFF][AIX] Handle llvm.used and llvm.compiler.used global array
Jason Liu via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Wed Jul 22 14:24:32 PDT 2020
jasonliu created this revision.
jasonliu added reviewers: daltenty, DiggerLin, Xiangling_L, hubert.reinterpretcast.
Herald added subscribers: llvm-commits, kbarton, hiraditya, nemanjai.
Herald added a project: LLVM.
For now, just return and do nothing when we see llvm.used and llvm.compiler.used global array.
Hopefully, we could come up with a good solution later to prevent linker from eliminating symbols in llvm.used array.
https://reviews.llvm.org/D84363
Files:
llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
llvm/test/CodeGen/PowerPC/aix-xcoff-used.ll
Index: llvm/test/CodeGen/PowerPC/aix-xcoff-used.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/PowerPC/aix-xcoff-used.ll
@@ -0,0 +1,20 @@
+;; This test verifies llc on AIX would not crash when llvm.used and
+;; llvm.compiler.used is presented in the IR.
+
+; RUN: llc -verify-machineinstrs -mcpu=pwr4 -mtriple powerpc-ibm-aix-xcoff < %s | \
+; RUN: FileCheck %s
+
+; RUN: llc -verify-machineinstrs -mcpu=pwr4 -mtriple powerpc64-ibm-aix-xcoff < %s | \
+; RUN: FileCheck %s
+
+ at keep_this = internal global i32 2, align 4
+ at keep_this2 = internal global i32 3, align 4
+ at llvm.used = appending global [1 x i8*] [i8* bitcast (i32* @keep_this to i8*)], section "llvm.metadata"
+ at llvm.compiler.used = appending global [1 x i8*] [i8* bitcast (i32* @keep_this2 to i8*)], section "llvm.metadata"
+
+; CHECK-NOT: llvm.metadata
+; CHECK: .lglobl keep_this
+; CHECK: keep_this:
+; CHECK: .lglobl keep_this2
+; CHECK: keep_this2:
+; CHECK-NOT: llvm.metadata
Index: llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
===================================================================
--- llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
+++ llvm/lib/Target/PowerPC/PPCAsmPrinter.cpp
@@ -1678,22 +1678,28 @@
report_fatal_error("COMDAT not yet supported by AIX.");
}
-static bool isSpecialLLVMGlobalArrayForStaticInit(const GlobalVariable *GV) {
+static bool isSpecialLLVMGlobalArrayToSkip(const GlobalVariable *GV) {
return StringSwitch<bool>(GV->getName())
+ // TODO: Update the handling of global arrays for static init when
+ // we support the ".ref" directive.
+ // Otherwise, we can skip these arrays, because the AIX linker collects
+ // static init functions simply based on their name.
.Cases("llvm.global_ctors", "llvm.global_dtors", true)
+ // TODO: Linker could still eliminate the GV if we just skip handling
+ // llvm.used array. Skipping them for now until we or the AIX OS team
+ // come up with a good solution.
+ .Case("llvm.used", true)
+ // It's correct to just skip llvm.compiler.used array here.
+ .Case("llvm.compiler.used", true)
.Default(false);
}
void PPCAIXAsmPrinter::emitGlobalVariable(const GlobalVariable *GV) {
- ValidateGV(GV);
-
- // TODO: Update the handling of global arrays for static init when we support
- // the ".ref" directive.
- // Otherwise, we can skip these arrays, because the AIX linker collects
- // static init functions simply based on their name.
- if (isSpecialLLVMGlobalArrayForStaticInit(GV))
+ if (isSpecialLLVMGlobalArrayToSkip(GV))
return;
+ ValidateGV(GV);
+
// Create the symbol, set its storage class.
MCSymbolXCOFF *GVSym = cast<MCSymbolXCOFF>(getSymbol(GV));
GVSym->setStorageClass(
@@ -1836,8 +1842,11 @@
// We need to know, up front, the alignment of csects for the assembly path,
// because once a .csect directive gets emitted, we could not change the
// alignment value on it.
- for (const auto &G : M.globals())
+ for (const auto &G : M.globals()) {
+ if (isSpecialLLVMGlobalArrayToSkip(&G))
+ continue;
setCsectAlignment(&G);
+ }
for (const auto &F : M)
setCsectAlignment(&F);
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D84363.279943.patch
Type: text/x-patch
Size: 3225 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200722/b25d5ca3/attachment.bin>
More information about the llvm-commits
mailing list