[X86] Add "Ws" constraint and "p" modifier for symbolic address/label…

… reference (#77886)

Printing the raw symbol is useful in inline asm (e.g. getting the C++
mangled name, referencing a symbol in a custom way while ensuring it is
not optimized out even if internal). Similar constraints are available
in other targets (e.g. "S" for aarch64/riscv, "Cs" for m68k).

```
namespace ns { extern int var, a[4]; }
void foo() {
  asm(".pushsection .xxx,\"aw\"; .dc.a %p0; .popsection" :: "Ws"(&ns::var));
  asm(".reloc ., BFD_RELOC_NONE, %p0" :: "Ws"(&ns::a[3]));
}
```

Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105576

clang/lib/Basic/Targets/X86.cpp

@@ -1418,6 +1418,14 @@ bool X86TargetInfo::validateAsmConstraint(
   case 'O':
     Info.setRequiresImmediate(0, 127);
     return true;
+  case 'W':
+    switch (*++Name) {
+    default:
+      return false;
+    case 's':
+      Info.setAllowsRegister();
+      return true;
+    }
   // Register constraints.
   case 'Y': // 'Y' is the first character for several 2-character constraints.
     // Shift the pointer to the second character of the constraint.
@@ -1715,6 +1723,9 @@ std::string X86TargetInfo::convertConstraint(const char *&Constraint) const {
     return std::string("{st}");
   case 'u':                        // second from top of floating point stack.
     return std::string("{st(1)}"); // second from top of floating point stack.
+  case 'W':
+    assert(Constraint[1] == 's');
+    return '^' + std::string(Constraint++, 2);
   case 'Y':
     switch (Constraint[1]) {
     default:

clang/test/CodeGen/X86/inline-asm-constraints.c

@@ -53,3 +53,14 @@ __m512 testZMM0(void) {
 #endif
   return zmm0;
 }
+
+extern int var, arr[4];
+struct Pair { int a, b; } pair;
+
+// CHECK-LABEL: test_Ws(
+// CHECK:         call void asm sideeffect "// ${0:p} ${1:p} ${2:p}", "^Ws,^Ws,^Ws,~{dirflag},~{fpsr},~{flags}"(ptr @var, ptr getelementptr inbounds ([4 x i32], ptr @arr, i64 0, i64 3), ptr @test_Ws)
+// CHECK:         call void asm sideeffect "// $0", "^Ws,~{dirflag},~{fpsr},~{flags}"(ptr getelementptr inbounds (%struct.Pair, ptr @pair, i32 0, i32 1))
+void test_Ws(void) {
+  asm("// %p0 %p1 %p2" :: "Ws"(&var), "Ws"(&arr[3]), "Ws"(test_Ws));
+  asm("// %0" :: "Ws"(&pair.b));
+}

clang/test/Sema/inline-asm-validate-x86.c

@@ -130,3 +130,11 @@ void pr40890(void) {
   __asm__ __volatile__("\n#define BEEF abcd%0\n" : : "n"((int*)0xdeadbeeeeeef));
 #endif
 }
+
+void test_W(int i) {
+  __asm__("" : : "Wd"(test_W)); // expected-error{{invalid input constraint 'Wd' in asm}}
+
+  __asm__("" : : "Ws"(test_W(0))); // expected-error{{invalid type 'void' in asm input for constraint 'Ws'}}
+  // Codegen error
+  __asm__("" : : "Ws"(i));
+}

llvm/docs/LangRef.rst

@@ -5336,6 +5336,8 @@ X86:
   operand in a SSE register. If AVX is also enabled, can also be a 256-bit
   vector operand in an AVX register. If AVX-512 is also enabled, can also be a
   512-bit vector operand in an AVX512 register. Otherwise, an error.
+- ``Ws``: A symbolic reference with an optional constant addend or a label
+  reference.
 - ``x``: The same as ``v``, except that when AVX-512 is enabled, the ``x`` code
   only allocates into the first 16 AVX-512 registers, while the ``v`` code
   allocates into any of the 32 AVX-512 registers.
@@ -5518,6 +5520,7 @@ X86:
   the operand. (The behavior for relocatable symbol expressions is a
   target-specific behavior for this typically target-independent modifier)
 - ``H``: Print a memory reference with additional offset +8.
+- ``p``: Print a raw symbol name (without syntax-specific prefixes).
 - ``P``: Print a memory reference used as the argument of a call instruction or
   used with explicit base reg and index reg as its offset. So it can not use
   additional regs to present the memory reference. (E.g. omit ``(rip)``, even

llvm/lib/Target/X86/X86AsmPrinter.cpp

@@ -774,6 +774,14 @@ bool X86AsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
       PrintOperand(MI, OpNo, O);
       return false;
 
+    case 'p': {
+      const MachineOperand &MO = MI->getOperand(OpNo);
+      if (MO.getType() != MachineOperand::MO_GlobalAddress)
+        return true;
+      PrintSymbolOperand(MO, O);
+      return false;
+    }
+
     case 'P': // This is the operand of a call, treat specially.
       PrintPCRelImm(MI, OpNo, O);
       return false;

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -56686,6 +56686,10 @@ X86TargetLowering::getConstraintType(StringRef Constraint) const {
     switch (Constraint[0]) {
     default:
       break;
+    case 'W':
+      if (Constraint[1] != 's')
+        break;
+      return C_Other;
     case 'Y':
       switch (Constraint[1]) {
       default:
@@ -56890,11 +56894,6 @@ void X86TargetLowering::LowerAsmOperandForConstraint(SDValue Op,
                                                      std::vector<SDValue> &Ops,
                                                      SelectionDAG &DAG) const {
   SDValue Result;
-
-  // Only support length 1 constraints for now.
-  if (Constraint.size() > 1)
-    return;
-
   char ConstraintLetter = Constraint[0];
   switch (ConstraintLetter) {
   default: break;
@@ -56976,6 +56975,26 @@ void X86TargetLowering::LowerAsmOperandForConstraint(SDValue Op,
     }
     return;
   }
+  case 'W': {
+    assert(Constraint[1] == 's');
+    // Op is a BlockAddressSDNode or a GlobalAddressSDNode with an optional
+    // offset.
+    if (const auto *BA = dyn_cast<BlockAddressSDNode>(Op)) {
+      Ops.push_back(DAG.getTargetBlockAddress(BA->getBlockAddress(),
+                                              BA->getValueType(0)));
+    } else {
+      int64_t Offset = 0;
+      if (Op->getOpcode() == ISD::ADD &&
+          isa<ConstantSDNode>(Op->getOperand(1))) {
+        Offset = cast<ConstantSDNode>(Op->getOperand(1))->getSExtValue();
+        Op = Op->getOperand(0);
+      }
+      if (const auto *GA = dyn_cast<GlobalAddressSDNode>(Op))
+        Ops.push_back(DAG.getTargetGlobalAddress(GA->getGlobal(), SDLoc(Op),
+                                                 GA->getValueType(0), Offset));
+    }
+    return;
+  }
   case 'Z': {
     // 32-bit unsigned value
     if (auto *C = dyn_cast<ConstantSDNode>(Op)) {

llvm/test/CodeGen/X86/inline-asm-Ws-constraint-error.ll

@@ -0,0 +1,14 @@
+; RUN: not llc -mtriple=x86_64 < %s 2>&1 | FileCheck %s
+
+@a = external global [4 x i32], align 16
+
+; CHECK-COUNT-2: error: invalid operand for inline asm constraint 'Ws'
+; CHECK-NOT:     error:
+define void @test(i64 %i) {
+entry:
+  %x = alloca i32, align 4
+  %ai = getelementptr inbounds [4 x i32], ptr @a, i64 0, i64 %i
+  call void asm sideeffect "", "^Ws,~{dirflag},~{fpsr},~{flags}"(ptr %x)
+  call void asm sideeffect "", "^Ws,~{dirflag},~{fpsr},~{flags}"(ptr %ai)
+  ret void
+}

llvm/test/CodeGen/X86/inline-asm-Ws-constraint.ll

@@ -0,0 +1,36 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=i686 < %s | FileCheck %s
+; RUN: llc -mtriple=x86_64 < %s | FileCheck %s
+
+@var = external dso_local global i32, align 4
+@a = external global [4 x i32], align 16
+
+define dso_local void @test() {
+; CHECK-LABEL: test:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    # var a+12 test
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    ret{{[l|q]}}
+entry:
+  %ai = getelementptr inbounds [4 x i32], ptr @a, i64 0, i64 3
+  call void asm sideeffect "// ${0:p} ${1:p} ${2:p}", "^Ws,^Ws,^Ws,~{dirflag},~{fpsr},~{flags}"(ptr @var, ptr %ai, ptr @test)
+  ret void
+}
+
+define dso_local void @test_label() {
+; CHECK-LABEL: test_label:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:  .Ltmp0: # Block address taken
+; CHECK-NEXT:  # %bb.1: # %label
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    # .Ltmp0
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    ret{{[l|q]}}
+entry:
+  br label %label
+
+label:
+  tail call void asm sideeffect "// ${0:p}", "^Ws,~{dirflag},~{fpsr},~{flags}"(ptr blockaddress(@test_label, %label))
+  ret void
+}