[X86] [split-stack] Add an option to allocate extra space for stack split functions

[ChuanqiXu]

Summary

As the title describes, this patch add an option to allocate extra space for stack split functions. And if user specify extra spaces, then we can loose the constraint about the varargs of split-stack functions.

This idea comes from a patch from libgcc: patch. In this patch, the author tries to allocate extra space for each stack split functions to avoid frequent allocations. Then in the following reply: reply, the reviewer suggests it could be a good idea to make it as compiler option. So I think it is a good idea to add it in clang.

Test-Plan: check-all

[ChuanqiXu]

fix unit-tests.

[ChuanqiXu]

gentle ping~

llvm/lib/Target/X86/X86FrameLowering.cpp

 #include <cstdlib>
 
 #define DEBUG_TYPE "x86-fl"
 
 STATISTIC(NumFrameLoopProbe, "Number of loop stack probes used in prologue");
 STATISTIC(NumFrameExtraProbe,
           "Number of extra stack probes generated in prologue");
 
+static llvm::cl::opt<unsigned> ExtraSpaceForMoreStackExpansion(
+    "extra-space-for-more-stack-expansion", llvm::cl::Hidden,
+    llvm::cl::desc("allocate extra space for morestack expansion"),
+    llvm::cl::init(0));
+
 using namespace llvm;
 
 X86FrameLowering::X86FrameLowering(const X86Subtarget &STI,
                                    MaybeAlign StackAlignOverride)
     : TargetFrameLowering(StackGrowsDown, StackAlignOverride.valueOrOne(),
                           STI.is64Bit() ? -8 : -4),
       STI(STI), TII(*STI.getInstrInfo()), TRI(STI.getRegisterInfo()) {
   // Cache a bunch of frame-related predicates for this subtarget.
void X86FrameLowering::adjustForSegmentedStacks(
   // To support shrink-wrapping we would need to insert the new blocks
   // at the right place and update the branches to PrologueMBB.
   assert(&(*MF.begin()) == &PrologueMBB && "Shrink-wrapping not supported yet");
 
   unsigned ScratchReg = GetScratchRegister(Is64Bit, IsLP64, MF, true);
   assert(!MF.getRegInfo().isLiveIn(ScratchReg) &&
          "Scratch register is live-in");
 
-  if (MF.getFunction().isVarArg())
-    report_fatal_error("Segmented stacks do not support vararg functions.");
   if (!STI.isTargetLinux() && !STI.isTargetDarwin() && !STI.isTargetWin32() &&
       !STI.isTargetWin64() && !STI.isTargetFreeBSD() &&
       !STI.isTargetDragonFly())
     report_fatal_error("Segmented stacks not supported on this platform.");
 
   // Eventually StackSize will be calculated by a link-time pass; which will
   // also decide whether checking code needs to be injected into this particular
   // prologue.
   StackSize = MFI.getStackSize();
 
+  if (ExtraSpaceForMoreStackExpansion.getNumOccurrences())
+    StackSize += ExtraSpaceForMoreStackExpansion;
+  else if (MF.getFunction().isVarArg())
+    report_fatal_error("Segmented stacks do not support vararg functions if "
+                       "user don't specify extra spaces for split stacks.");
+
   // Do not generate a prologue for leaf functions with a stack of size zero.
   // For non-leaf functions we have to allow for the possibility that the
   // callis to a non-split function, as in PR37807. This function could also
   // take the address of a non-split function. When the linker tries to adjust
   // its non-existent prologue, it would fail with an error. Mark the object
   // file so that such failures are not errors. See this Go language bug-report
   // https://go-review.googlesource.com/c/go/+/148819/
   if (StackSize == 0 && !MFI.hasTailCall()) {

llvm/test/CodeGen/X86/split-stack-extra-space.ll

+; RUN: llc -extra-space-for-more-stack-expansion=100 < %s | FileCheck %s
+source_filename = "/disk2/workspace.xuchuanqi/alibaba-llvm-11.x-for-work/clang/test/CodeGen/split-stacks.c"
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+; CHECK: movabsq  $108, %r10
+; CHECK: movabsq  $0, %r11
+; CHECK: callq  __morestack
+
+; CHECK: movabsq  $124, %r10
+; CHECK: movabsq  $0, %r11
+; CHECK: callq  __morestack
+
+; Function Attrs: noinline nounwind optnone uwtable
+define dso_local i32 @foo() #0 {
+entry:
+  ret i32 0
+}
+
+; Function Attrs: noinline nounwind optnone uwtable
+define dso_local i32 @nosplit() #1 {
+entry:
+  ret i32 0
+}
+
+; Function Attrs: noinline nounwind optnone uwtable
+define dso_local i32 @main() #0 {
+entry:
+  %retval = alloca i32, align 4
+  store i32 0, i32* %retval, align 4
+  %call = call i32 @foo()
+  ret i32 %call
+}
+
+attributes #0 = { noinline nounwind optnone uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="true" "split-stack" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
+attributes #1 = { noinline nounwind optnone uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"clang version 11.1.0 ([e-mail]:clang-llvm/alibaba-llvm-project.git 74aa28fe18b93f7a68bd02d039a60e858d016d7e)"}
 No newline at end of file