[AMDGPU] modify named barrier builtins and intrinsics (#114550)

Use a local pointer type to represent the named barrier in builtin and
intrinsic. This makes the definitions more user friendly
bacause they do not need to worry about the hardware ID assignment. Also
this approach is more like the other popular GPU programming language.
Named barriers should be represented as global variables of addrspace(3)
in LLVM-IR. Compiler assigns the special LDS offsets for those variables
during AMDGPULowerModuleLDS pass. Those addresses are converted to hw
barrier ID during instruction selection. The rest of the
instruction-selection changes are primarily due to the
intrinsic-definition changes.

1 files changed, 8 insertions, 0 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUMachineFunction.cpp b/llvm/lib/Target/AMDGPU/AMDGPUMachineFunction.cpp
index 1d83d0c..f524764 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUMachineFunction.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUMachineFunction.cpp
@@ -8,6 +8,7 @@
 
 #include "AMDGPUMachineFunction.h"
 #include "AMDGPU.h"
+#include "AMDGPUMemoryUtils.h"
 #include "AMDGPUPerfHintAnalysis.h"
 #include "AMDGPUSubtarget.h"
 #include "Utils/AMDGPUBaseInfo.h"
@@ -102,6 +103,13 @@ unsigned AMDGPUMachineFunction::allocateLDSGlobal(const DataLayout &DL,
 
   unsigned Offset;
   if (GV.getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS) {
+    if (TargetExtType *TTy = AMDGPU::isNamedBarrier(GV)) {
+      std::optional<unsigned> BarAddr = getLDSAbsoluteAddress(GV);
+      if (!BarAddr)
+        llvm_unreachable("named barrier should have an assigned address");
+      Entry.first->second = BarAddr.value();
+      return BarAddr.value();
+    }
 
     std::optional<uint32_t> MaybeAbs = getLDSAbsoluteAddress(GV);
     if (MaybeAbs) {