[AArch64][SME] Implement inline-asm clobbers for za/zt0 (#79276)

This enables specifing "za" or "zt0" to the clobber list for inline asm.
This complies with the acle SME addition to the asm extension here:
ARM-software/acle#276

Signed-off-by: chenmiao <[email protected]>
Signed-off-by: chenmiao <[email protected]>

Author: MDevereau

clang/lib/Basic/Targets/AArch64.cpp

@@ -1146,6 +1146,8 @@ TargetInfo::BuiltinVaListKind AArch64TargetInfo::getBuiltinVaListKind() const {
 }
 
 const char *const AArch64TargetInfo::GCCRegNames[] = {
+    // clang-format off
+
     // 32-bit Integer registers
     "w0", "w1", "w2", "w3", "w4", "w5", "w6", "w7", "w8", "w9", "w10", "w11",
     "w12", "w13", "w14", "w15", "w16", "w17", "w18", "w19", "w20", "w21", "w22",
@@ -1182,7 +1184,12 @@ const char *const AArch64TargetInfo::GCCRegNames[] = {
 
     // SVE predicate-as-counter registers
     "pn0",  "pn1",  "pn2",  "pn3",  "pn4",  "pn5",  "pn6",  "pn7",  "pn8",
-    "pn9",  "pn10", "pn11", "pn12", "pn13", "pn14", "pn15"
+    "pn9",  "pn10", "pn11", "pn12", "pn13", "pn14", "pn15",
+
+    // SME registers
+    "za", "zt0",
+
+    // clang-format on
 };
 
 ArrayRef<const char *> AArch64TargetInfo::getGCCRegNames() const {

clang/test/CodeGen/aarch64-inline-asm.c

@@ -95,3 +95,11 @@ void test_reduced_gpr_constraints(int var32, long var64) {
 // CHECK: [[ARG2:%.+]] = load i64, ptr
 // CHECK: call void asm sideeffect "add x0, x0, $0", "@3Ucj,~{x0}"(i64 [[ARG2]])
 }
+
+void test_sme_constraints(){
+  asm("movt zt0[3, mul vl], z0" : : : "za");
+// CHECK: call void asm sideeffect "movt zt0[3, mul vl], z0", "~{za}"()
+
+  asm("movt zt0[3, mul vl], z0" : : : "zt0");
+// CHECK: call void asm sideeffect "movt zt0[3, mul vl], z0", "~{zt0}"()
+}

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -10364,6 +10364,14 @@ AArch64TargetLowering::getRegForInlineAsmConstraint(
       parseConstraintCode(Constraint) != AArch64CC::Invalid)
     return std::make_pair(unsigned(AArch64::NZCV), &AArch64::CCRRegClass);
 
+  if (Constraint == "{za}") {
+    return std::make_pair(unsigned(AArch64::ZA), &AArch64::MPRRegClass);
+  }
+
+  if (Constraint == "{zt0}") {
+    return std::make_pair(unsigned(AArch64::ZT0), &AArch64::ZTRRegClass);
+  }
+
   // Use the default implementation in TargetLowering to convert the register
   // constraint into a member of a register class.
   std::pair<unsigned, const TargetRegisterClass *> Res;

llvm/lib/Target/AArch64/AArch64RegisterInfo.cpp

@@ -491,6 +491,10 @@ bool AArch64RegisterInfo::isAsmClobberable(const MachineFunction &MF,
         MCRegisterInfo::regsOverlap(PhysReg, AArch64::X16))
     return true;
 
+  // ZA/ZT0 registers are reserved but may be permitted in the clobber list.
+  if (PhysReg == AArch64::ZA || PhysReg == AArch64::ZT0)
+    return true;
+
   return !isReservedReg(MF, PhysReg);
 }
 

llvm/test/CodeGen/AArch64/aarch64-za-clobber.ll

@@ -0,0 +1,16 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64-none-linux-gnu -stop-after=aarch64-isel < %s -o - | FileCheck %s
+
+define void @alpha(<vscale x 4 x i32> %x) local_unnamed_addr {
+entry:
+; CHECK: INLINEASM &"movt zt0[3, mul vl], z0", 1 /* sideeffect attdialect */, 12 /* clobber */, implicit-def early-clobber $za
+  tail call void asm sideeffect "movt zt0[3, mul vl], z0", "~{za}"()
+  ret void
+}
+
+define void @beta(<vscale x 4 x i32> %x) local_unnamed_addr {
+entry:
+; CHECK: INLINEASM &"movt zt0[3, mul vl], z0", 1 /* sideeffect attdialect */, 12 /* clobber */, implicit-def early-clobber $zt0
+  tail call void asm sideeffect "movt zt0[3, mul vl], z0", "~{zt0}"()
+  ret void
+}