[SystemZ][z/OS] Fix the entry point marker for leaf functions

[Everybody0523]

Summary

The function emitFunctionEntryLabel does not look at whether or not a function is a leaf when setting the entry flags, and instead blindly marks all functions as non-leaf routines.

[uweigand]

LGTM, thanks!

[uweigand]

Still LGTM, thanks.

llvm/lib/Target/SystemZ/SystemZAsmPrinter.cpp

if (Subtarget.getTargetTriple().isOSzOS()) {
     CurrentFnEPMarkerSym =
         OutContext.createTempSymbol(Twine("EPM_").concat(N).str(), true);
     CurrentFnPPA1Sym =
         OutContext.createTempSymbol(Twine("PPA1_").concat(N).str(), true);
 
     // EntryPoint Marker
     const MachineFrameInfo &MFFrame = MF->getFrameInfo();
     bool IsUsingAlloca = MFFrame.hasVarSizedObjects();
+    uint32_t DSASize = MFFrame.getStackSize();
+    bool IsLeaf = DSASize == 0 && MFFrame.getCalleeSavedInfo().empty();
 
     // Set Flags
     uint8_t Flags = 0;
+    if (IsLeaf)
+      Flags |= 0x08;
     if (IsUsingAlloca)
       Flags |= 0x04;
 
-    uint32_t DSASize = MFFrame.getStackSize();
-
     // Combine into top 27 bits of DSASize and bottom 5 bits of Flags.
     uint32_t DSAAndFlags = DSASize & 0xFFFFFFE0; // (x/32) << 5
     DSAAndFlags |= Flags;
 
     // Emit entry point marker section.
     OutStreamer->AddComment("XPLINK Routine Layout Entry");
     OutStreamer->emitLabel(CurrentFnEPMarkerSym);
     OutStreamer->AddComment("Eyecatcher 0x00C300C500C500");
     OutStreamer->emitIntValueInHex(0x00C300C500C500, 7); // Eyecatcher.
     OutStreamer->AddComment("Mark Type C'1'");
     OutStreamer->emitInt8(0xF1); // Mark Type.
     OutStreamer->AddComment("Offset to PPA1");
     OutStreamer->emitAbsoluteSymbolDiff(CurrentFnPPA1Sym, CurrentFnEPMarkerSym,
                                         4);
     if (OutStreamer->isVerboseAsm()) {
       OutStreamer->AddComment("DSA Size 0x" + Twine::utohexstr(DSASize));
       OutStreamer->AddComment("Entry Flags");
+      if (Flags & 0x08)
+        OutStreamer->AddComment("  Bit 1: 1 = Leaf function");
+      else
+        OutStreamer->AddComment("  Bit 1: 0 = Non-leaf function");
       if (Flags & 0x04)
         OutStreamer->AddComment("  Bit 2: 1 = Uses alloca");
       else
         OutStreamer->AddComment("  Bit 2: 0 = Does not use alloca");
     }
     OutStreamer->emitInt32(DSAAndFlags);
   }
 
   AsmPrinter::emitFunctionEntryLabel();
 }
 
 // Force static initialization.
 extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeSystemZAsmPrinter() {
   RegisterAsmPrinter<SystemZAsmPrinter> X(getTheSystemZTarget());
 }

llvm/test/CodeGen/SystemZ/zos-prologue-epilog.ll

define void @large_stack2(i64 %n1, i64 %n2, i64 %n3) {
   %arr0 = alloca [131072 x i64], align 8
   %arr1 = alloca i64, i64 %n1, align 8
   call i64 (ptr, ptr, i64, i64, i64) @fun4(ptr %arr0,
             ptr %arr1, i64 %n1, i64 %n2, i64 %n3)
   ret void
 }
 
 ; CHECK-LABEL: leaf_func
+; CHECK: .long  8 * DSA Size 0x0
+; CHECK-NEXT:     * Entry Flags
+; CHECK-NEXT:     *   Bit 1: 1 = Leaf function
+; CHECK-NEXT:     *   Bit 2: 0 = Does not use alloca
 ; CHECK-NOT: aghi  4,
 ; CHECK-NOT: stmg
 ; CHECK: agr  1, 2
 ; CHECK: msgr 1, 3
 ; CHECK: aghik  3, 1, -4
 ; CHECK-NOT: aghi  4,
 ; CHECK-NOT: lmg
 define i64 @leaf_func0(i64 %a, i64 %b, i64 %c) {

llvm/test/MC/GOFF/ppa1.ll

 ; RUN: llc -mtriple s390x-ibm-zos < %s | FileCheck %s
 ; REQUIRES: systemz-registered-target
 
 ; CHECK: @@EPM_void_test_0: * @void_test
 ; CHECK: * XPLINK Routine Layout Entry
 ; CHECK: .long   12779717 * Eyecatcher 0x00C300C500C500
 ; CHECK: .short  197
 ; CHECK: .byte   0
 ; CHECK: .byte   241 * Mark Type C'1'
-; CHECK: .long   0 * DSA Size 0x0
+; CHECK: .long   8 * DSA Size 0x0
 ; CHECK: * Entry Flags
+; CHECK: *   Bit 1: 1 = Leaf function
 ; CHECK: *   Bit 2: 0 = Does not use alloca
 ; CHECK: @@func_end0:
 ; CHECK: .section        ".ppa1"
 ; CHECK: @@PPA1_void_test_0:                     * PPA1
 ; CHECK:        .byte   2                               * Version
 ; CHECK:        .byte   206                             * LE Signature X'CE'
 ; CHECK:        .short  0                               * Saved GPR Mask
 ; CHECK:        .byte   128                             * PPA1 Flags 1