[llvm] 472d282 - [AArch64][GlobalISel] Don't localize TLS G_GLOBAL_VALUEs on Darwin.

[Amara Emerson via llvm-commits]

Author: Amara Emerson
Date: 2020-03-24T13:35:50-07:00
New Revision: 472d282046d044b409fd2c550e101e85583c54a0

URL: https://github.com/llvm/llvm-project/commit/472d282046d044b409fd2c550e101e85583c54a0
DIFF: https://github.com/llvm/llvm-project/commit/472d282046d044b409fd2c550e101e85583c54a0.diff

LOG: [AArch64][GlobalISel] Don't localize TLS G_GLOBAL_VALUEs on Darwin.

On Darwin these need to be selected into a function call for the TLS
address lookup. As a result, they can't be moved below a physreg write,
which happens in call sequences. In the long term, we should have some
mechanism in the localizer to prevent localizing into target-specific
atomic instruction sequences.

rdar://60056248

Differential Revision: https://reviews.llvm.org/D76652

Added: 
    

Modified: 
    llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
    llvm/lib/Target/AArch64/AArch64ISelLowering.h
    llvm/test/CodeGen/AArch64/GlobalISel/localizer-arm64-tti.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index ab9176a8e77a..61521517de63 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -13926,3 +13926,16 @@ void AArch64TargetLowering::finalizeLowering(MachineFunction &MF) const {
 bool AArch64TargetLowering::needsFixedCatchObjects() const {
   return false;
 }
+
+bool AArch64TargetLowering::shouldLocalize(
+    const MachineInstr &MI, const TargetTransformInfo *TTI) const {
+  if (MI.getOpcode() == TargetOpcode::G_GLOBAL_VALUE) {
+    // On Darwin, TLS global vars get selected into function calls, which
+    // we don't want localized, as they can get moved into the middle of a
+    // another call sequence.
+    const GlobalValue &GV = *MI.getOperand(1).getGlobal();
+    if (GV.isThreadLocal() && Subtarget->isTargetMachO())
+      return false;
+  }
+  return TargetLoweringBase::shouldLocalize(MI, TTI);
+}

diff  --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.h b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
index 8394c90a11be..0932fb062a79 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.h
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
@@ -859,6 +859,9 @@ class AArch64TargetLowering : public TargetLowering {
   bool shouldNormalizeToSelectSequence(LLVMContext &, EVT) const override;
 
   void finalizeLowering(MachineFunction &MF) const override;
+
+  bool shouldLocalize(const MachineInstr &MI,
+                      const TargetTransformInfo *TTI) const override;
 };
 
 namespace AArch64 {

diff  --git a/llvm/test/CodeGen/AArch64/GlobalISel/localizer-arm64-tti.ll b/llvm/test/CodeGen/AArch64/GlobalISel/localizer-arm64-tti.ll
index 88507680af74..0210a9a89ca4 100644
--- a/llvm/test/CodeGen/AArch64/GlobalISel/localizer-arm64-tti.ll
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/localizer-arm64-tti.ll
@@ -59,3 +59,46 @@ if.end:
   ret i32 0
 }
 
+ at tls_gv = common thread_local global i32 0, align 4
+
+; This test checks that we don't try to localize TLS variables on Darwin.
+; If the user happens to be inside a call sequence, we could end up rematerializing
+; below a physreg write, clobbering it (TLS accesses on Darwin need a function call).
+; For now, we check we don't localize at all. We could in theory make sure that
+; we don't localize into the middle of a call sequence instead.
+define i32 @darwin_tls() {
+  ; CHECK-LABEL: name: darwin_tls
+  ; CHECK: bb.1.entry:
+  ; CHECK:   successors: %bb.2(0x40000000), %bb.3(0x40000000)
+  ; CHECK:   [[GV:%[0-9]+]]:gpr(p0) = G_GLOBAL_VALUE @tls_gv
+  ; CHECK:   [[GV1:%[0-9]+]]:gpr(p0) = G_GLOBAL_VALUE @var2
+  ; CHECK:   [[C:%[0-9]+]]:gpr(s32) = G_CONSTANT i32 0
+  ; CHECK:   [[GV2:%[0-9]+]]:gpr(p0) = G_GLOBAL_VALUE @var1
+  ; CHECK:   [[LOAD:%[0-9]+]]:gpr(s32) = G_LOAD [[GV2]](p0) :: (dereferenceable load 4 from @var1)
+  ; CHECK:   [[C1:%[0-9]+]]:gpr(s32) = G_CONSTANT i32 1
+  ; CHECK:   [[ICMP:%[0-9]+]]:gpr(s32) = G_ICMP intpred(ne), [[LOAD]](s32), [[C1]]
+  ; CHECK:   [[TRUNC:%[0-9]+]]:gpr(s1) = G_TRUNC [[ICMP]](s32)
+  ; CHECK:   G_BRCOND [[TRUNC]](s1), %bb.3
+  ; CHECK: bb.2.if.then:
+  ; CHECK:   successors: %bb.3(0x80000000)
+  ; CHECK:   [[LOAD1:%[0-9]+]]:gpr(s32) = G_LOAD [[GV]](p0) :: (dereferenceable load 4 from @tls_gv)
+  ; CHECK:   [[GV3:%[0-9]+]]:gpr(p0) = G_GLOBAL_VALUE @var2
+  ; CHECK:   G_STORE [[LOAD1]](s32), [[GV3]](p0) :: (store 4 into @var2)
+  ; CHECK: bb.3.if.end:
+  ; CHECK:   [[C2:%[0-9]+]]:gpr(s32) = G_CONSTANT i32 0
+  ; CHECK:   $w0 = COPY [[C2]](s32)
+  ; CHECK:   RET_ReallyLR implicit $w0
+entry:
+  %0 = load i32, i32* @var1, align 4
+  %cmp = icmp eq i32 %0, 1
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:
+  %tls = load i32, i32* @tls_gv, align 4
+  store i32 %tls, i32* @var2, align 4
+  br label %if.end
+
+if.end:
+  ret i32 0
+}
+