[AArch64] Fix FPMR handling when switching streaming mode

[Lukacma]

According to the documentation, the FPMR register is set to 0 when entering or exiting streaming mode. This patch models that behavior by adding FPMR as an implicit def to the instructions used for entering and exiting streaming mode.

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

Author: None (Lukacma)

Changes

According to the documentation, the FPMR register is set to 0 when entering or exiting streaming mode. This patch models that behavior by adding FPMR as an implicit def to the instructions used for entering and exiting streaming mode.

---

Full diff: https://github.com/llvm/llvm-project/pull/135827.diff

2 Files Affected:

• (modified) llvm/lib/Target/AArch64/AArch64ISelLowering.cpp (+3)
• (added) llvm/test/CodeGen/AArch64/sme-write-fpmr.ll (+23)

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 830ec6886e6bc..20dfda34fb14d 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -8880,12 +8880,15 @@ void AArch64TargetLowering::AdjustInstrPostInstrSelection(MachineInstr &MI,
         MI.removeOperand(I);
 
     // The SVE vector length can change when entering/leaving streaming mode.
+    // FPMR is set to 0 when entering/leaving streaming mode.
     if (MI.getOperand(0).getImm() == AArch64SVCR::SVCRSM ||
         MI.getOperand(0).getImm() == AArch64SVCR::SVCRSMZA) {
       MI.addOperand(MachineOperand::CreateReg(AArch64::VG, /*IsDef=*/false,
                                               /*IsImplicit=*/true));
       MI.addOperand(MachineOperand::CreateReg(AArch64::VG, /*IsDef=*/true,
                                               /*IsImplicit=*/true));
+      MI.addOperand(MachineOperand::CreateReg(AArch64::FPMR, /*IsDef=*/true,
+                                              /*IsImplicit=*/true));
     }
   }
 
diff --git a/llvm/test/CodeGen/AArch64/sme-write-fpmr.ll b/llvm/test/CodeGen/AArch64/sme-write-fpmr.ll
new file mode 100644
index 0000000000000..074d65831a584
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sme-write-fpmr.ll
@@ -0,0 +1,23 @@
+; RUN: llc -mattr=+sme -stop-after=finalize-isel < %s | FileCheck %s
+
+target triple = "aarch64"
+
+; Check that we don't define FPMR for 'smstart za' and 'smstop za'
+define void @smstart_za() "aarch64_new_za" nounwind {
+  ; CHECK-LABEL:    name: smstart_za
+  ; CHECK-NOT:        implicit-def {{[^,]*}}$fpmr
+  ret void
+}
+
+; Check that we do define FPMR for 'smstart sm' and 'smstop sm'
+define void @smstart_sm() nounwind {
+  ; CHECK-LABEL: name: smstart_sm
+  ; CHECK:          MSRpstatesvcrImm1 1, 1,
+  ; CHECK-SAME:       implicit-def {{[^,]*}}$fpmr
+  ; CHECK:          MSRpstatesvcrImm1 1, 0,
+  ; CHECK-SAME:       implicit-def {{[^,]*}}$fpmr
+  call void @require_sm()
+  ret void
+}
+
+declare void @require_sm() "aarch64_pstate_sm_enabled"

[llvm-ci]

LLVM Buildbot has detected a new failure on builder llvm-clang-aarch64-darwin running on doug-worker-5 while building llvm at step 6 "test-build-unified-tree-check-all".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/190/builds/18400

Here is the relevant piece of the build log for the reference

Step 6 (test-build-unified-tree-check-all) failure: test (failure)
******************** TEST 'LLVM :: ExecutionEngine/OrcLazy/multiple-compile-threads-basic.ll' FAILED ********************
Exit Code: 2

Command Output (stderr):
--
/Users/buildbot/buildbot-root/aarch64-darwin/build/bin/lli -jit-kind=orc-lazy -compile-threads=2 -thread-entry hello /Users/buildbot/buildbot-root/aarch64-darwin/llvm-project/llvm/test/ExecutionEngine/OrcLazy/multiple-compile-threads-basic.ll | /Users/buildbot/buildbot-root/aarch64-darwin/build/bin/FileCheck /Users/buildbot/buildbot-root/aarch64-darwin/llvm-project/llvm/test/ExecutionEngine/OrcLazy/multiple-compile-threads-basic.ll # RUN: at line 1
+ /Users/buildbot/buildbot-root/aarch64-darwin/build/bin/lli -jit-kind=orc-lazy -compile-threads=2 -thread-entry hello /Users/buildbot/buildbot-root/aarch64-darwin/llvm-project/llvm/test/ExecutionEngine/OrcLazy/multiple-compile-threads-basic.ll
+ /Users/buildbot/buildbot-root/aarch64-darwin/build/bin/FileCheck /Users/buildbot/buildbot-root/aarch64-darwin/llvm-project/llvm/test/ExecutionEngine/OrcLazy/multiple-compile-threads-basic.ll
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
 #0 0x0000000103a1259c llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (/Users/buildbot/buildbot-root/aarch64-darwin/build/bin/lli+0x100ed659c)
 #1 0x0000000103a10620 llvm::sys::RunSignalHandlers() (/Users/buildbot/buildbot-root/aarch64-darwin/build/bin/lli+0x100ed4620)
 #2 0x0000000103a12c58 SignalHandler(int, __siginfo*, void*) (/Users/buildbot/buildbot-root/aarch64-darwin/build/bin/lli+0x100ed6c58)
 #3 0x00000001853ef584 (/usr/lib/system/libsystem_platform.dylib+0x18047b584)
 #4 0x00000001853be21c (/usr/lib/system/libsystem_pthread.dylib+0x18044a21c)
 #5 0x00000001852e4ad0 (/usr/lib/libc++.1.dylib+0x180370ad0)
 #6 0x00000001035b3d14 void llvm::detail::UniqueFunctionBase<void, llvm::Expected<llvm::DenseMap<llvm::orc::SymbolStringPtr, llvm::orc::ExecutorSymbolDef, llvm::DenseMapInfo<llvm::orc::SymbolStringPtr, void>, llvm::detail::DenseMapPair<llvm::orc::SymbolStringPtr, llvm::orc::ExecutorSymbolDef>>>>::CallImpl<llvm::orc::Platform::lookupInitSymbols(llvm::orc::ExecutionSession&, llvm::DenseMap<llvm::orc::JITDylib*, llvm::orc::SymbolLookupSet, llvm::DenseMapInfo<llvm::orc::JITDylib*, void>, llvm::detail::DenseMapPair<llvm::orc::JITDylib*, llvm::orc::SymbolLookupSet>> const&)::$_45>(void*, llvm::Expected<llvm::DenseMap<llvm::orc::SymbolStringPtr, llvm::orc::ExecutorSymbolDef, llvm::DenseMapInfo<llvm::orc::SymbolStringPtr, void>, llvm::detail::DenseMapPair<llvm::orc::SymbolStringPtr, llvm::orc::ExecutorSymbolDef>>>&) (/Users/buildbot/buildbot-root/aarch64-darwin/build/bin/lli+0x100a77d14)
 #7 0x00000001035af948 llvm::orc::AsynchronousSymbolQuery::handleComplete(llvm::orc::ExecutionSession&)::RunQueryCompleteTask::run() (/Users/buildbot/buildbot-root/aarch64-darwin/build/bin/lli+0x100a73948)
 #8 0x000000010366bba8 void* std::__1::__thread_proxy[abi:un170006]<std::__1::tuple<std::__1::unique_ptr<std::__1::__thread_struct, std::__1::default_delete<std::__1::__thread_struct>>, llvm::orc::DynamicThreadPoolTaskDispatcher::dispatch(std::__1::unique_ptr<llvm::orc::Task, std::__1::default_delete<llvm::orc::Task>>)::$_0>>(void*) (/Users/buildbot/buildbot-root/aarch64-darwin/build/bin/lli+0x100b2fba8)
 #9 0x00000001853bef94 (/usr/lib/system/libsystem_pthread.dylib+0x18044af94)
#10 0x00000001853b9d34 (/usr/lib/system/libsystem_pthread.dylib+0x180445d34)
FileCheck error: '<stdin>' is empty.
FileCheck command line:  /Users/buildbot/buildbot-root/aarch64-darwin/build/bin/FileCheck /Users/buildbot/buildbot-root/aarch64-darwin/llvm-project/llvm/test/ExecutionEngine/OrcLazy/multiple-compile-threads-basic.ll

--

********************