Segmented stacks: omit __morestack call when there's no frame.

Patch by Florian Zeitz

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@209436 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 77 insertions, 18 deletions

diff --git a/lib/Target/ARM/ARMFrameLowering.cpp b/lib/Target/ARM/ARMFrameLowering.cpp
index c0f8a8d902..0caf4bfd77 100644
--- a/lib/Target/ARM/ARMFrameLowering.cpp
+++ b/lib/Target/ARM/ARMFrameLowering.cpp
@@ -1746,6 +1746,12 @@ void ARMFrameLowering::adjustForSegmentedStacks(MachineFunction &MF) const {
   ARMFunctionInfo *ARMFI = MF.getInfo<ARMFunctionInfo>();
   DebugLoc DL;
 
+  uint64_t StackSize = MFI->getStackSize();
+
+  // Do not generate a prologue for functions with a stack of size zero
+  if (StackSize == 0)
+    return;
+
   // Use R4 and R5 as scratch registers.
   // We save R4 and R5 before use and restore them before leaving the function.
   unsigned ScratchReg0 = ARM::R4;
@@ -1775,8 +1781,6 @@ void ARMFrameLowering::adjustForSegmentedStacks(MachineFunction &MF) const {
   MF.push_front(PrevStackMBB);
 
   // The required stack size that is aligned to ARM constant criterion.
-  uint64_t StackSize = MFI->getStackSize();
-
   AlignedStackSize = alignToARMConstant(StackSize);
 
   // When the frame size is less than 256 we just compare the stack
diff --git a/lib/Target/X86/X86FrameLowering.cpp b/lib/Target/X86/X86FrameLowering.cpp
index 1c1b06623b..4c1374f70f 100644
--- a/lib/Target/X86/X86FrameLowering.cpp
+++ b/lib/Target/X86/X86FrameLowering.cpp
@@ -1176,6 +1176,15 @@ X86FrameLowering::adjustForSegmentedStacks(MachineFunction &MF) const {
       !STI.isTargetWin32() && !STI.isTargetWin64() && !STI.isTargetFreeBSD())
     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();
+
+  // Do not generate a prologue for functions with a stack of size zero
+  if (StackSize == 0)
+    return;
+
   MachineBasicBlock *allocMBB = MF.CreateMachineBasicBlock();
   MachineBasicBlock *checkMBB = MF.CreateMachineBasicBlock();
   X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
@@ -1200,11 +1209,6 @@ X86FrameLowering::adjustForSegmentedStacks(MachineFunction &MF) const {
   MF.push_front(allocMBB);
   MF.push_front(checkMBB);
 
-  // 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();
-
   // When the frame size is less than 256 we just compare the stack
   // boundary directly to the value of the stack pointer, per gcc.
   bool CompareStackPointer = StackSize < kSplitStackAvailable;
diff --git a/test/CodeGen/ARM/segmented-stacks.ll b/test/CodeGen/ARM/segmented-stacks.ll
index a7804b900a..9873bf3329 100644
--- a/test/CodeGen/ARM/segmented-stacks.ll
+++ b/test/CodeGen/ARM/segmented-stacks.ll
@@ -57,6 +57,8 @@ define void @test_basic() #0 {
 define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
        %addend = load i32 * %closure
        %result = add i32 %other, %addend
+       %mem = alloca i32, i32 10
+       call void @dummy_use (i32* %mem, i32 10)
        ret i32 %result
 
 ; ARM-linux:      test_nested:
@@ -68,7 +70,7 @@ define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
 ; ARM-linux-NEXT: cmp     r4, r5
 ; ARM-linux-NEXT: blo     .LBB1_2
 
-; ARM-linux:      mov     r4, #0
+; ARM-linux:      mov     r4, #56
 ; ARM-linux-NEXT: mov     r5, #0
 ; ARM-linux-NEXT: stmdb   sp!, {lr}
 ; ARM-linux-NEXT: bl      __morestack
@@ -87,7 +89,7 @@ define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
 ; ARM-android-NEXT: cmp     r4, r5
 ; ARM-android-NEXT: blo     .LBB1_2
 
-; ARM-android:      mov     r4, #0
+; ARM-android:      mov     r4, #56
 ; ARM-android-NEXT: mov     r5, #0
 ; ARM-android-NEXT: stmdb   sp!, {lr}
 ; ARM-android-NEXT: bl      __morestack
@@ -234,4 +236,14 @@ define fastcc void @test_fastcc_large() #0 {
 
 }
 
+define void @test_nostack() #0 {
+	ret void
+
+; ARM-linux-LABEL: test_nostack:
+; ARM-linux-NOT:   bl __morestack
+
+; ARM-android-LABEL: test_nostack:
+; ARM-android-NOT:   bl __morestack
+}
+
 attributes #0 = { "split-stack" }
diff --git a/test/CodeGen/Thumb/segmented-stacks.ll b/test/CodeGen/Thumb/segmented-stacks.ll
index 89043ec11d..d6e25c7792 100644
--- a/test/CodeGen/Thumb/segmented-stacks.ll
+++ b/test/CodeGen/Thumb/segmented-stacks.ll
@@ -57,6 +57,8 @@ define void @test_basic() #0 {
 define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
        %addend = load i32 * %closure
        %result = add i32 %other, %addend
+       %mem = alloca i32, i32 10
+       call void @dummy_use (i32* %mem, i32 10)
        ret i32 %result
 
 ; Thumb-android:      test_nested:
@@ -68,7 +70,7 @@ define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
 ; Thumb-android-NEXT: cmp     r4, r5
 ; Thumb-android-NEXT: blo     .LBB1_2
 
-; Thumb-android:      mov     r4, #0
+; Thumb-android:      mov     r4, #56
 ; Thumb-android-NEXT: mov     r5, #0
 ; Thumb-android-NEXT: push    {lr}
 ; Thumb-android-NEXT: bl      __morestack
@@ -88,7 +90,7 @@ define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
 ; Thumb-linux-NEXT: cmp     r4, r5
 ; Thumb-linux-NEXT: blo     .LBB1_2
 
-; Thumb-linux:      mov     r4, #0
+; Thumb-linux:      mov     r4, #56
 ; Thumb-linux-NEXT: mov     r5, #0
 ; Thumb-linux-NEXT: push    {lr}
 ; Thumb-linux-NEXT: bl      __morestack
@@ -246,4 +248,14 @@ define fastcc void @test_fastcc_large() #0 {
 
 }
 
+define void @test_nostack() #0 {
+	ret void
+
+; Thumb-android-LABEL: test_nostack:
+; Thumb-android-NOT:   bl __morestack
+
+; Thumb-linux-LABEL: test_nostack:
+; Thumb-linux-NOT:   bl __morestack
+}
+
 attributes #0 = { "split-stack" }
diff --git a/test/CodeGen/X86/segmented-stacks.ll b/test/CodeGen/X86/segmented-stacks.ll
index 8089f20568..9dab3cd8d6 100644
--- a/test/CodeGen/X86/segmented-stacks.ll
+++ b/test/CodeGen/X86/segmented-stacks.ll
@@ -107,13 +107,15 @@ define void @test_basic() #0 {
 define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
        %addend = load i32 * %closure
        %result = add i32 %other, %addend
+       %mem = alloca i32, i32 10
+       call void @dummy_use (i32* %mem, i32 10)
        ret i32 %result
 
 ; X32-Linux:       cmpl %gs:48, %esp
 ; X32-Linux-NEXT:  ja      .LBB1_2
 
 ; X32-Linux:       pushl $4
-; X32-Linux-NEXT:  pushl $0
+; X32-Linux-NEXT:  pushl $60
 ; X32-Linux-NEXT:  calll __morestack
 ; X32-Linux-NEXT:  ret
 
@@ -121,7 +123,7 @@ define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
 ; X64-Linux-NEXT:  ja      .LBB1_2
 
 ; X64-Linux:       movq %r10, %rax
-; X64-Linux-NEXT:  movabsq $0, %r10
+; X64-Linux-NEXT:  movabsq $56, %r10
 ; X64-Linux-NEXT:  movabsq $0, %r11
 ; X64-Linux-NEXT:  callq __morestack
 ; X64-Linux-NEXT:  ret
@@ -132,7 +134,7 @@ define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
 ; X32-Darwin-NEXT: ja      LBB1_2
 
 ; X32-Darwin:      pushl $4
-; X32-Darwin-NEXT: pushl $0
+; X32-Darwin-NEXT: pushl $60
 ; X32-Darwin-NEXT: calll ___morestack
 ; X32-Darwin-NEXT: ret
 
@@ -140,7 +142,7 @@ define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
 ; X64-Darwin-NEXT: ja      LBB1_2
 
 ; X64-Darwin:      movq %r10, %rax
-; X64-Darwin-NEXT: movabsq $0, %r10
+; X64-Darwin-NEXT: movabsq $56, %r10
 ; X64-Darwin-NEXT: movabsq $0, %r11
 ; X64-Darwin-NEXT: callq ___morestack
 ; X64-Darwin-NEXT: ret
@@ -150,7 +152,7 @@ define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
 ; X32-MinGW-NEXT:  ja      LBB1_2
 
 ; X32-MinGW:       pushl $4
-; X32-MinGW-NEXT:  pushl $0
+; X32-MinGW-NEXT:  pushl $52
 ; X32-MinGW-NEXT:  calll ___morestack
 ; X32-MinGW-NEXT:  ret
 
@@ -159,7 +161,7 @@ define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
 ; X64-MinGW-NEXT:  ja      .LBB1_2
 
 ; X64-MinGW:       movq %r10, %rax
-; X64-MinGW-NEXT:  movabsq $0, %r10
+; X64-MinGW-NEXT:  movabsq $88, %r10
 ; X64-MinGW-NEXT:  movabsq $32, %r11
 ; X64-MinGW-NEXT:  callq __morestack
 ; X64-MinGW-NEXT:  retq
@@ -169,7 +171,7 @@ define i32 @test_nested(i32 * nest %closure, i32 %other) #0 {
 ; X64-FreeBSD-NEXT:  ja      .LBB1_2
 
 ; X64-FreeBSD:       movq %r10, %rax
-; X64-FreeBSD-NEXT:  movabsq $0, %r10
+; X64-FreeBSD-NEXT:  movabsq $56, %r10
 ; X64-FreeBSD-NEXT:  movabsq $0, %r11
 ; X64-FreeBSD-NEXT:  callq __morestack
 ; X64-FreeBSD-NEXT:  ret
@@ -435,4 +437,29 @@ define fastcc void @test_fastcc_large_with_ecx_arg(i32 %a) #0 {
 
 }
 
+define void @test_nostack() #0 {
+	ret void
+
+; X32-Linux-LABEL: test_nostack:
+; X32-Linux-NOT:   calll __morestack
+
+; X64-Linux-LABEL: test_nostack:
+; X32-Linux-NOT:   callq __morestack
+
+; X32-Darwin-LABEL: test_nostack:
+; X32-Darwin-NOT:   calll __morestack
+
+; X64-Darwin-LABEL: test_nostack:
+; X64-Darwin-NOT:   callq __morestack
+
+; X32-MinGW-LABEL: test_nostack:
+; X32-MinGW-NOT:   calll __morestack
+
+; X64-MinGW-LABEL: test_nostack:
+; X64-MinGW-NOT:   callq __morestack
+
+; X64-FreeBSD-LABEL: test_nostack:
+; X64-FreeBSD-NOT:   callq __morestack
+}
+
 attributes #0 = { "split-stack" }