[WebAssembly] Compile all TLS on Emscripten as local-exec

[quantum]

Summary

Currently, on Emscripten, dynamic linking is not supported with threads.
This means that if thread-local storage is used, it must be used in a
statically-linked executable. Hence, local-exec is the only possible model.

This diff compiles all TLS variables to use local-exec on Emscripten as a
temporary measure until dynamic linking is supported with threads.

The goal for this is to allow C++ types with constructors to be thread-local.

Currently, when clang compiles a thread_local variable with a constructor,
it generates __tls_guard variable:

@__tls_guard = internal thread_local global i8 0, align 1

As no TLS model is specified, this is treated as general-dynamic, which we do
not support (and cannot support without implementing dynamic linking support
with threads in Emscripten). As a result, any C++ constructor in thread_local
variables would not compile.

By compiling all thread_local as local-exec, __tls_guard will compile and
we can support C++ constructors with TLS without implementing dynamic linking
with threads.

Depends on D64537

[aheejin]

• Not sure what this means. The CL description says it tries to compile things into local-exec, but what this CL does look like checking if a global var is local-exec and if not report an error.
• What is supposed to happen when the OS is not emscripten? If we don't support TLS in non-emscripten OSes regardless of its mode (local-exec or not), should we print an error in that case too?
• Also, a test would be good too.

llvm/lib/Target/WebAssembly/WebAssemblyISelDAGToDAG.cpp

184 ↗

(On Diff #209970)

If this condition is to be removed once emscripten support dynamic linking with threads, rather than XXX, how about marking as TODO Remove this condition once emscripten supports dynamic linking with threads or something?

[quantum]

Change to use TODO instead of XXX

[quantum]

Undo compressing the previous diff.

[quantum]

llvm/lib/Target/WebAssembly/WebAssemblyISelDAGToDAG.cpp

184 ↗

(On Diff #209970)

Done.

[aheejin]

This comment was removed by aheejin.

[quantum]

In D64776#1586697, @aheejin wrote:

• Not sure what this means. The CL description says it tries to compile things into local-exec, but what this CL does look like checking if a global var is local-exec and if not report an error.
• What is supposed to happen when the OS is not emscripten? If we don't support TLS in non-emscripten OSes regardless of its mode (local-exec or not), should we print an error in that case too?
• Also, a test would be good too.

• On Emscripten, the code does not report an error if the variable is not local-exec and instead compiles it as local-exec.
• The local-exec implementation should still theoretically work for other embeddings. Only the idea that only local-exec is currently possible is emscripten-specific.
• A test is added.

[quantum]

Rebase onto latest D64537

[aheejin]

llvm/lib/Target/WebAssembly/WebAssemblyISelDAGToDAG.cpp

191 ↗

(On Diff #210137)

So does this mean:

1. If this is emscripten, convert all non-local-exec to local-exec
2. If this is not emscripten, error on non-local exec

Is this right?

Anyhow, it might be better to tell what global variable is causing the problem to users in the error message, something like

report_fatal_error("only -ftls-model=local-exec is supported for now in non-Emscripten OSes: " + GA->getGlobal()->getName());

llvm/test/CodeGen/WebAssembly/tls-general-dynamic.ll

1 ↗

(On Diff #210137)

Do we need a separate copy of tls.ll for this test? The file contents look the same. Can't we add a new test line or something to the existing file?

[quantum]

llvm/lib/Target/WebAssembly/WebAssemblyISelDAGToDAG.cpp

191 ↗

(On Diff #210137)

Yes, it's right. Good point regarding the message, I'll fix it.

llvm/test/CodeGen/WebAssembly/tls-general-dynamic.ll

1 ↗

(On Diff #210137)

No, the difference is that in this file, we must say

@tls = internal thread_local global i32 0

whereas in the other file, it has to say

@tls = internal thread_local(localexec) global i32 0

I thought about using grep, but @tlively suggested otherwise.

[quantum]

Use better error message.

[quantum]

Rebase onto latest LLVM.

llvm/trunk/lib/Target/WebAssembly/WebAssemblyISelDAGToDAG.cpp

void WebAssemblyDAGToDAGISel::Select(SDNode *Node) {
 
   case ISD::GlobalTLSAddress: {
     const auto *GA = cast<GlobalAddressSDNode>(Node);
 
     if (!MF.getSubtarget<WebAssemblySubtarget>().hasBulkMemory())
       report_fatal_error("cannot use thread-local storage without bulk memory",
                          false);
 
+    // Currently Emscripten does not support dynamic linking with threads.
+    // Therefore, if we have thread-local storage, only the local-exec model
+    // is possible.
+    // TODO: remove this and implement proper TLS models once Emscripten
+    // supports dynamic linking with threads.
     if (GA->getGlobal()->getThreadLocalMode() !=
-        GlobalValue::LocalExecTLSModel) {
-      report_fatal_error("only -ftls-model=local-exec is supported for now",
+            GlobalValue::LocalExecTLSModel &&
+        !Subtarget->getTargetTriple().isOSEmscripten()) {
+      report_fatal_error("only -ftls-model=local-exec is supported for now on "
+                         "non-Emscripten OSes: variable " +
+                             GA->getGlobal()->getName(),
                          false);
     }
 
     MVT PtrVT = TLI->getPointerTy(CurDAG->getDataLayout());
     assert(PtrVT == MVT::i32 && "only wasm32 is supported for now");
 
     SDValue TLSBaseSym = CurDAG->getTargetExternalSymbol("__tls_base", PtrVT);
     SDValue TLSOffsetSym = CurDAG->getTargetGlobalAddress(

llvm/trunk/test/CodeGen/WebAssembly/tls-general-dynamic.ll

+; RUN: not llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -mattr=+bulk-memory 2>&1 | FileCheck %s --check-prefix=ERROR
+; RUN: not llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -mattr=+bulk-memory -fast-isel 2>&1 | FileCheck %s --check-prefix=ERROR
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -mattr=+bulk-memory --mtriple wasm32-unknown-emscripten | FileCheck %s --check-prefixes=CHECK,TLS
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -mattr=+bulk-memory --mtriple wasm32-unknown-emscripten -fast-isel | FileCheck %s --check-prefixes=CHECK,TLS
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -mattr=-bulk-memory | FileCheck %s --check-prefixes=CHECK,NO-TLS
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32-unknown-unknown"
+
+; ERROR: LLVM ERROR: only -ftls-model=local-exec is supported for now on non-Emscripten OSes: variable tls
+
+; CHECK-LABEL: address_of_tls:
+; CHECK-NEXT: .functype  address_of_tls () -> (i32)
+define i32 @address_of_tls() {
+  ; TLS-DAG: global.get __tls_base
+  ; TLS-DAG: i32.const tls
+  ; TLS-NEXT: i32.add
+  ; TLS-NEXT: return
+
+  ; NO-TLS-NEXT: i32.const tls
+  ; NO-TLS-NEXT: return
+  ret i32 ptrtoint(i32* @tls to i32)
+}
+
+; CHECK-LABEL: ptr_to_tls:
+; CHECK-NEXT: .functype ptr_to_tls () -> (i32)
+define i32* @ptr_to_tls() {
+  ; TLS-DAG: global.get __tls_base
+  ; TLS-DAG: i32.const tls
+  ; TLS-NEXT: i32.add
+  ; TLS-NEXT: return
+
+  ; NO-TLS-NEXT: i32.const tls
+  ; NO-TLS-NEXT: return
+  ret i32* @tls
+}
+
+; CHECK-LABEL: tls_load:
+; CHECK-NEXT: .functype tls_load () -> (i32)
+define i32 @tls_load() {
+  ; TLS-DAG: global.get __tls_base
+  ; TLS-DAG: i32.const tls
+  ; TLS-NEXT: i32.add
+  ; TLS-NEXT: i32.load 0
+  ; TLS-NEXT: return
+
+  ; NO-TLS-NEXT: i32.const 0
+  ; NO-TLS-NEXT: i32.load tls
+  ; NO-TLS-NEXT: return
+  %tmp = load i32, i32* @tls, align 4
+  ret i32 %tmp
+}
+
+; CHECK-LABEL: tls_store:
+; CHECK-NEXT: .functype tls_store (i32) -> ()
+define void @tls_store(i32 %x) {
+  ; TLS-DAG: global.get __tls_base
+  ; TLS-DAG: i32.const tls
+  ; TLS-NEXT: i32.add
+  ; TLS-NEXT: i32.store 0
+  ; TLS-NEXT: return
+
+  ; NO-TLS-NEXT: i32.const 0
+  ; NO-TLS-NEXT: i32.store tls
+  ; NO-TLS-NEXT: return
+  store i32 %x, i32* @tls, align 4
+  ret void
+}
+
+; CHECK-LABEL: tls_size:
+; CHECK-NEXT: .functype tls_size () -> (i32)
+define i32 @tls_size() {
+; CHECK-NEXT: global.get __tls_size
+; CHECK-NEXT: return
+  %1 = call i32 @llvm.wasm.tls.size.i32()
+  ret i32 %1
+}
+
+; CHECK: .type tls,@object
+; TLS-NEXT: .section .tbss.tls,"",@
+; NO-TLS-NEXT: .section .bss.tls,"",@
+; CHECK-NEXT: .p2align 2
+; CHECK-NEXT: tls:
+; CHECK-NEXT: .int32 0
+@tls = internal thread_local global i32 0
+
+declare i32 @llvm.wasm.tls.size.i32()

llvm/trunk/test/CodeGen/WebAssembly/tls-local-exec.ll

+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -mattr=+bulk-memory | FileCheck %s --check-prefixes=CHECK,TLS
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -mattr=+bulk-memory -fast-isel | FileCheck %s --check-prefixes=CHECK,TLS
+; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -mattr=-bulk-memory | FileCheck %s --check-prefixes=CHECK,NO-TLS
+target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
+target triple = "wasm32-unknown-unknown"
+
+; CHECK-LABEL: address_of_tls:
+; CHECK-NEXT: .functype  address_of_tls () -> (i32)
+define i32 @address_of_tls() {
+  ; TLS-DAG: global.get __tls_base
+  ; TLS-DAG: i32.const tls
+  ; TLS-NEXT: i32.add
+  ; TLS-NEXT: return
+
+  ; NO-TLS-NEXT: i32.const tls
+  ; NO-TLS-NEXT: return
+  ret i32 ptrtoint(i32* @tls to i32)
+}
+
+; CHECK-LABEL: ptr_to_tls:
+; CHECK-NEXT: .functype ptr_to_tls () -> (i32)
+define i32* @ptr_to_tls() {
+  ; TLS-DAG: global.get __tls_base
+  ; TLS-DAG: i32.const tls
+  ; TLS-NEXT: i32.add
+  ; TLS-NEXT: return
+
+  ; NO-TLS-NEXT: i32.const tls
+  ; NO-TLS-NEXT: return
+  ret i32* @tls
+}
+
+; CHECK-LABEL: tls_load:
+; CHECK-NEXT: .functype tls_load () -> (i32)
+define i32 @tls_load() {
+  ; TLS-DAG: global.get __tls_base
+  ; TLS-DAG: i32.const tls
+  ; TLS-NEXT: i32.add
+  ; TLS-NEXT: i32.load 0
+  ; TLS-NEXT: return
+
+  ; NO-TLS-NEXT: i32.const 0
+  ; NO-TLS-NEXT: i32.load tls
+  ; NO-TLS-NEXT: return
+  %tmp = load i32, i32* @tls, align 4
+  ret i32 %tmp
+}
+
+; CHECK-LABEL: tls_store:
+; CHECK-NEXT: .functype tls_store (i32) -> ()
+define void @tls_store(i32 %x) {
+  ; TLS-DAG: global.get __tls_base
+  ; TLS-DAG: i32.const tls
+  ; TLS-NEXT: i32.add
+  ; TLS-NEXT: i32.store 0
+  ; TLS-NEXT: return
+
+  ; NO-TLS-NEXT: i32.const 0
+  ; NO-TLS-NEXT: i32.store tls
+  ; NO-TLS-NEXT: return
+  store i32 %x, i32* @tls, align 4
+  ret void
+}
+
+; CHECK-LABEL: tls_size:
+; CHECK-NEXT: .functype tls_size () -> (i32)
+define i32 @tls_size() {
+; CHECK-NEXT: global.get __tls_size
+; CHECK-NEXT: return
+  %1 = call i32 @llvm.wasm.tls.size.i32()
+  ret i32 %1
+}
+
+; CHECK: .type tls,@object
+; TLS-NEXT: .section .tbss.tls,"",@
+; NO-TLS-NEXT: .section .bss.tls,"",@
+; CHECK-NEXT: .p2align 2
+; CHECK-NEXT: tls:
+; CHECK-NEXT: .int32 0
+@tls = internal thread_local(localexec) global i32 0
+
+declare i32 @llvm.wasm.tls.size.i32()

llvm/trunk/test/CodeGen/WebAssembly/tls.ll

-; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -mattr=+bulk-memory | FileCheck %s --check-prefixes=CHECK,TLS
-; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -mattr=+bulk-memory -fast-isel | FileCheck %s --check-prefixes=CHECK,TLS
-; RUN: llc < %s -asm-verbose=false -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -mattr=-bulk-memory | FileCheck %s --check-prefixes=CHECK,NO-TLS
-target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
-target triple = "wasm32-unknown-unknown"
-
-; CHECK-LABEL: address_of_tls:
-; CHECK-NEXT: .functype  address_of_tls () -> (i32)
-define i32 @address_of_tls() {
-  ; TLS-DAG: global.get __tls_base
-  ; TLS-DAG: i32.const tls
-  ; TLS-NEXT: i32.add
-  ; TLS-NEXT: return
-
-  ; NO-TLS-NEXT: i32.const tls
-  ; NO-TLS-NEXT: return
-  ret i32 ptrtoint(i32* @tls to i32)
-}
-
-; CHECK-LABEL: ptr_to_tls:
-; CHECK-NEXT: .functype ptr_to_tls () -> (i32)
-define i32* @ptr_to_tls() {
-  ; TLS-DAG: global.get __tls_base
-  ; TLS-DAG: i32.const tls
-  ; TLS-NEXT: i32.add
-  ; TLS-NEXT: return
-
-  ; NO-TLS-NEXT: i32.const tls
-  ; NO-TLS-NEXT: return
-  ret i32* @tls
-}
-
-; CHECK-LABEL: tls_load:
-; CHECK-NEXT: .functype tls_load () -> (i32)
-define i32 @tls_load() {
-  ; TLS-DAG: global.get __tls_base
-  ; TLS-DAG: i32.const tls
-  ; TLS-NEXT: i32.add
-  ; TLS-NEXT: i32.load 0
-  ; TLS-NEXT: return
-
-  ; NO-TLS-NEXT: i32.const 0
-  ; NO-TLS-NEXT: i32.load tls
-  ; NO-TLS-NEXT: return
-  %tmp = load i32, i32* @tls, align 4
-  ret i32 %tmp
-}
-
-; CHECK-LABEL: tls_store:
-; CHECK-NEXT: .functype tls_store (i32) -> ()
-define void @tls_store(i32 %x) {
-  ; TLS-DAG: global.get __tls_base
-  ; TLS-DAG: i32.const tls
-  ; TLS-NEXT: i32.add
-  ; TLS-NEXT: i32.store 0
-  ; TLS-NEXT: return
-
-  ; NO-TLS-NEXT: i32.const 0
-  ; NO-TLS-NEXT: i32.store tls
-  ; NO-TLS-NEXT: return
-  store i32 %x, i32* @tls, align 4
-  ret void
-}
-
-; CHECK-LABEL: tls_size:
-; CHECK-NEXT: .functype tls_size () -> (i32)
-define i32 @tls_size() {
-; CHECK-NEXT: global.get __tls_size
-; CHECK-NEXT: return
-  %1 = call i32 @llvm.wasm.tls.size.i32()
-  ret i32 %1
-}
-
-; CHECK: .type tls,@object
-; TLS-NEXT: .section .tbss.tls,"",@
-; NO-TLS-NEXT: .section .bss.tls,"",@
-; CHECK-NEXT: .p2align 2
-; CHECK-NEXT: tls:
-; CHECK-NEXT: .int32 0
-@tls = internal thread_local(localexec) global i32 0
-
-declare i32 @llvm.wasm.tls.size.i32()