[Clang][AArch64] Emit 'unimplemented' diagnostic for SME

[sdesmalen-arm]

When a function F has ZA and ZT0 state, calls another function G that only
shares ZT0 state with its caller, F will have to save ZA before the call to G,
and restore it afterwards (rather than setting up a lazy-sve).

This is not yet implemented in LLVM and does not result in a compile-time
error either. So instead of silently generating incorrect code, it's better to
emit an error saying this is not yet implemented.

[llvmbot]

@llvm/pr-subscribers-clang

Author: Sander de Smalen (sdesmalen-arm)

Changes

When a function F has ZA and ZT0 state, calls another function G that only
shares ZT0 state with its caller, F will have to save ZA before the call to G,
and restore it afterwards (rather than setting up a lazy-sve).

This is not yet implemented in LLVM and does not result in a compile-time
error either. So instead of silently generating incorrect code, it's better to
emit an error saying this is not yet implemented.

---

Full diff: https://github.com/llvm/llvm-project/pull/80295.diff

3 Files Affected:

• (modified) clang/include/clang/Basic/DiagnosticSemaKinds.td (+6)
• (modified) clang/lib/Sema/SemaChecking.cpp (+24-27)
• (modified) clang/test/Sema/aarch64-sme-func-attrs.c (+11-2)

diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td b/clang/include/clang/Basic/DiagnosticSemaKinds.td
index 7638a7e84c3c0..21a767cfc5e5c 100644
--- a/clang/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -3711,6 +3711,12 @@ def err_sme_za_call_no_za_state : Error<
   "call to a shared ZA function requires the caller to have ZA state">;
 def err_sme_zt0_call_no_zt0_state : Error<
   "call to a shared ZT0 function requires the caller to have ZT0 state">;
+def err_sme_unimplemented_za_save_restore : Error<
+  "call to a function that shares state other than 'za' from a "
+  "function that has live 'za' state requires a spill/fill of ZA, which is not yet "
+  "implemented">;
+def note_sme_use_preserves_za : Note<
+  "add '__arm_preserves(\"za\")' to the callee if it preserves ZA">;
 def err_sme_definition_using_sm_in_non_sme_target : Error<
   "function executed in streaming-SVE mode requires 'sme'">;
 def err_sme_definition_using_za_in_non_sme_target : Error<
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
index 7b877da1a928f..db9dc55c13b8d 100644
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -7513,47 +7513,44 @@ void Sema::checkCall(NamedDecl *FDecl, const FunctionProtoType *Proto,
       }
     }
 
-    // If the callee uses AArch64 SME ZA state but the caller doesn't define
-    // any, then this is an error.
-    FunctionType::ArmStateValue ArmZAState =
+    FunctionType::ArmStateValue CalleeArmZAState =
         FunctionType::getArmZAState(ExtInfo.AArch64SMEAttributes);
-    if (ArmZAState != FunctionType::ARM_None) {
+    FunctionType::ArmStateValue CalleeArmZT0State =
+        FunctionType::getArmZT0State(ExtInfo.AArch64SMEAttributes);
+    if (CalleeArmZAState != FunctionType::ARM_None ||
+        CalleeArmZT0State != FunctionType::ARM_None) {
       bool CallerHasZAState = false;
+      bool CallerHasZT0State = false;
       if (const auto *CallerFD = dyn_cast<FunctionDecl>(CurContext)) {
         auto *Attr = CallerFD->getAttr<ArmNewAttr>();
         if (Attr && Attr->isNewZA())
           CallerHasZAState = true;
-        else if (const auto *FPT =
-                     CallerFD->getType()->getAs<FunctionProtoType>())
-          CallerHasZAState = FunctionType::getArmZAState(
-                                 FPT->getExtProtoInfo().AArch64SMEAttributes) !=
-                             FunctionType::ARM_None;
-      }
-
-      if (!CallerHasZAState)
-        Diag(Loc, diag::err_sme_za_call_no_za_state);
-    }
-
-    // If the callee uses AArch64 SME ZT0 state but the caller doesn't define
-    // any, then this is an error.
-    FunctionType::ArmStateValue ArmZT0State =
-        FunctionType::getArmZT0State(ExtInfo.AArch64SMEAttributes);
-    if (ArmZT0State != FunctionType::ARM_None) {
-      bool CallerHasZT0State = false;
-      if (const auto *CallerFD = dyn_cast<FunctionDecl>(CurContext)) {
-        auto *Attr = CallerFD->getAttr<ArmNewAttr>();
         if (Attr && Attr->isNewZT0())
           CallerHasZT0State = true;
-        else if (const auto *FPT =
-                     CallerFD->getType()->getAs<FunctionProtoType>())
-          CallerHasZT0State =
+        if (const auto *FPT = CallerFD->getType()->getAs<FunctionProtoType>()) {
+          CallerHasZAState |=
+              FunctionType::getArmZAState(
+                  FPT->getExtProtoInfo().AArch64SMEAttributes) !=
+              FunctionType::ARM_None;
+          CallerHasZT0State |=
               FunctionType::getArmZT0State(
                   FPT->getExtProtoInfo().AArch64SMEAttributes) !=
               FunctionType::ARM_None;
+        }
       }
 
-      if (!CallerHasZT0State)
+      if (CalleeArmZAState != FunctionType::ARM_None && !CallerHasZAState)
+        Diag(Loc, diag::err_sme_za_call_no_za_state);
+
+      if (CalleeArmZT0State != FunctionType::ARM_None && !CallerHasZT0State)
         Diag(Loc, diag::err_sme_zt0_call_no_zt0_state);
+
+      if (CallerHasZAState &&
+          CalleeArmZAState == FunctionType::ARM_None &&
+          CalleeArmZT0State != FunctionType::ARM_None) {
+        Diag(Loc, diag::err_sme_unimplemented_za_save_restore);
+        Diag(Loc, diag::note_sme_use_preserves_za);
+      }
     }
   }
 
diff --git a/clang/test/Sema/aarch64-sme-func-attrs.c b/clang/test/Sema/aarch64-sme-func-attrs.c
index 97409ae7d6040..2bf1886951f1f 100644
--- a/clang/test/Sema/aarch64-sme-func-attrs.c
+++ b/clang/test/Sema/aarch64-sme-func-attrs.c
@@ -1,5 +1,5 @@
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -fsyntax-only -verify %s
-// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -fsyntax-only -verify=expected-cpp -x c++ %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -fsyntax-only -verify %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme2 -fsyntax-only -verify=expected-cpp -x c++ %s
 
 // Valid attributes
 
@@ -445,3 +445,12 @@ void conflicting_state_attrs_preserves_out_zt0(void) __arm_preserves("zt0") __ar
 // expected-cpp-error@+2 {{conflicting attributes for state 'zt0'}}
 // expected-error@+1 {{conflicting attributes for state 'zt0'}}
 void conflicting_state_attrs_preserves_inout_zt0(void) __arm_preserves("zt0") __arm_inout("zt0");
+
+// Test that we get a diagnostic for unimplemented case.
+void unimplemented_spill_fill_za(void (*share_zt0_only)(void) __arm_inout("zt0")) __arm_inout("za", "zt0") {
+  // expected-cpp-error@+4 {{call to a function that shares state other than 'za' from a function that has live 'za' state requires a spill/fill of ZA, which is not yet implemented}}
+  // expected-cpp-note@+3 {{add '__arm_preserves("za")' to the callee if it preserves ZA}}
+  // expected-error@+2 {{call to a function that shares state other than 'za' from a function that has live 'za' state requires a spill/fill of ZA, which is not yet implemented}}
+  // expected-note@+1 {{add '__arm_preserves("za")' to the callee if it preserves ZA}}
+  share_zt0_only();
+}

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[dtemirbulatov]

LGTM.