CodeGenPrepare: Add support for llvm.threadlocal.address address-mode sinking (#87844)

Depending on the TLSMode many thread-local accesses on x86 can be
expressed by adding a %fs: segment register to an addressing mode. Even
if there are mutliple users of a `llvm.threadlocal.address` intrinsic it
is generally not worth sharing the value in a register but instead fold
the %fs access into multiple addressing modes.

Hence this changes CodeGenPrepare to duplicate the
`llvm.threadlocal.address` intrinsic as necessary.

Introduces a new `TargetLowering::addressingModeSupportsTLS` callback
that allows targets to indicate whether TLS accesses can be part of an
addressing mode.

This is fixing a performance problem, as this folding of TLS-accesses
into multiple addressing modes happened naturally before the
introduction of the `llvm.threadlocal.address` intrinsic, but regressed
due to `SelectionDAG` keeping things in registers when accessed across
basic blocks, so CodeGenPrepare needs to duplicate to mitigate this. We
see a ~0.5% recovery in a codebase with heavy TLS usage (HHVM).

This fixes most of #87437

1 files changed, 25 insertions, 4 deletions

diff --git a/llvm/lib/CodeGen/CodeGenPrepare.cpp b/llvm/lib/CodeGen/CodeGenPrepare.cpp
index e657872..22a766f 100644
--- a/llvm/lib/CodeGen/CodeGenPrepare.cpp
+++ b/llvm/lib/CodeGen/CodeGenPrepare.cpp
@@ -5082,6 +5082,15 @@ bool AddressingModeMatcher::matchOperationAddr(User *AddrInst, unsigned Opcode,
     }
     return true;
   }
+  case Instruction::Call:
+    if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(AddrInst)) {
+      if (II->getIntrinsicID() == Intrinsic::threadlocal_address) {
+        GlobalValue &GV = cast<GlobalValue>(*II->getArgOperand(0));
+        if (TLI.addressingModeSupportsTLS(GV))
+          return matchAddr(AddrInst->getOperand(0), Depth);
+      }
+    }
+    break;
   }
   return false;
 }
@@ -5620,11 +5629,16 @@ bool CodeGenPrepare::optimizeMemoryInst(Instruction *MemoryInst, Value *Addr,
         return Modified;
     }
 
-    if (AddrMode.BaseGV) {
+    GlobalValue *BaseGV = AddrMode.BaseGV;
+    if (BaseGV != nullptr) {
       if (ResultPtr)
         return Modified;
 
-      ResultPtr = AddrMode.BaseGV;
+      if (BaseGV->isThreadLocal()) {
+        ResultPtr = Builder.CreateThreadLocalAddress(BaseGV);
+      } else {
+        ResultPtr = BaseGV;
+      }
     }
 
     // If the real base value actually came from an inttoptr, then the matcher
@@ -5789,8 +5803,15 @@ bool CodeGenPrepare::optimizeMemoryInst(Instruction *MemoryInst, Value *Addr,
     }
 
     // Add in the BaseGV if present.
-    if (AddrMode.BaseGV) {
-      Value *V = Builder.CreatePtrToInt(AddrMode.BaseGV, IntPtrTy, "sunkaddr");
+    GlobalValue *BaseGV = AddrMode.BaseGV;
+    if (BaseGV != nullptr) {
+      Value *BaseGVPtr;
+      if (BaseGV->isThreadLocal()) {
+        BaseGVPtr = Builder.CreateThreadLocalAddress(BaseGV);
+      } else {
+        BaseGVPtr = BaseGV;
+      }
+      Value *V = Builder.CreatePtrToInt(BaseGVPtr, IntPtrTy, "sunkaddr");
       if (Result)
         Result = Builder.CreateAdd(Result, V, "sunkaddr");
       else