AArch64: don't claim to preserve registers used by prologue code

Author: TNorthover

llvm/lib/Target/AArch64/AArch64CallingConvention.td

@@ -468,7 +468,7 @@ def CSR_Darwin_AArch64_TLS
 // CSR_Darwin_AArch64_CXX_TLS should be a subset of CSR_Darwin_AArch64_TLS.
 def CSR_Darwin_AArch64_CXX_TLS
     : CalleeSavedRegs<(add CSR_Darwin_AArch64_AAPCS,
-                           (sub (sequence "X%u", 1, 28), X15, X16, X17, X18),
+                           (sub (sequence "X%u", 1, 28), X9, X15, X16, X17, X18, X19),
                            (sequence "D%u", 0, 31))>;
 
 // CSRs that are handled by prologue, epilogue.

llvm/lib/Target/AArch64/AArch64FrameLowering.cpp

@@ -2285,7 +2285,7 @@ static void computeCalleeSaveRegisterPairs(
   // MachO's compact unwind format relies on all registers being stored in
   // pairs.
   assert((!produceCompactUnwindFrame(MF) ||
-          CC == CallingConv::PreserveMost ||
+          CC == CallingConv::PreserveMost || CC == CallingConv::CXX_FAST_TLS ||
           (Count & 1) == 0) &&
          "Odd number of callee-saved regs to spill!");
   int ByteOffset = AFI->getCalleeSavedStackSize();
@@ -2380,7 +2380,7 @@ static void computeCalleeSaveRegisterPairs(
     // MachO's compact unwind format relies on all registers being stored in
     // adjacent register pairs.
     assert((!produceCompactUnwindFrame(MF) ||
-            CC == CallingConv::PreserveMost ||
+            CC == CallingConv::PreserveMost || CC == CallingConv::CXX_FAST_TLS ||
             (RPI.isPaired() &&
              ((RPI.Reg1 == AArch64::LR && RPI.Reg2 == AArch64::FP) ||
               RPI.Reg1 + 1 == RPI.Reg2))) &&

llvm/test/CodeGen/AArch64/cxx-tlscc.ll

@@ -93,9 +93,9 @@ __tls_init.exit:
 ; CHECK-O0: stp d5, d4
 ; CHECK-O0: stp d3, d2
 ; CHECK-O0: stp d1, d0
-; CHECK-O0: stp x14, x13
-; CHECK-O0: stp x12, x11
-; CHECK-O0: stp x10, x9
+; CHECK-O0: str x14
+; CHECK-O0: stp x13, x12
+; CHECK-O0: stp x11, x10
 ; CHECK-O0: stp x8, x7
 ; CHECK-O0: stp x6, x5
 ; CHECK-O0: stp x4, x3
@@ -110,9 +110,9 @@ __tls_init.exit:
 ; CHECK-O0: ldp x4, x3
 ; CHECK-O0: ldp x6, x5
 ; CHECK-O0: ldp x8, x7
-; CHECK-O0: ldp x10, x9
-; CHECK-O0: ldp x12, x11
-; CHECK-O0: ldp x14, x13
+; CHECK-O0: ldp x11, x10
+; CHECK-O0: ldp x13, x12
+; CHECK-O0: ldr x14
 ; CHECK-O0: ldp d1, d0
 ; CHECK-O0: ldp d3, d2
 ; CHECK-O0: ldp d5, d4
@@ -220,5 +220,31 @@ entry:
   ret void
 }
 
+define cxx_fast_tlscc void @weird_prologue_regs(i32 %n) #1 {
+; CHECK-LABEL: weird_prologue_regs:
+; CHECK-NOT: str x9
+; CHECK-NOT: stp{{.*}}x9{{.*}}[
+; CHECK-NOT: str x19
+; CHECK-NOT: stp{{.*}}x19{{.*}}[
+
+; CHECK: sub x9, sp, #
+; CHECK: and sp, x9, #0x
+; CHECK: mov x19, sp
+
+; CHECK-NOT: str x9
+; CHECK-NOT: stp{{.*}}x9{{.*}}[
+; CHECK-NOT: str x19
+; CHECK-NOT: stp{{.*}}x19{{.*}}[
+
+  %p0 = alloca i32, i32 200
+  %p1 = alloca i32, align 32
+  %p2 = alloca i32, i32 %n
+  call void @callee(i32* %p0)
+  call void @callee(i32* %p1)
+  call void @callee(i32* %p2)
+  ret void
+}
+declare void @callee(i32*)
+
 attributes #0 = { nounwind "frame-pointer"="all" }
 attributes #1 = { nounwind }