[NVPTX] Add Volta Load/Store Atomics (.relaxed, .acquire, .release) and Volatile (.mmio/.volatile) support (#98022)

This PR adds initial support for some of Volta's (sm_70) load/store
atomic and volatile/MMIO operations, hopefully without breaking any
preexisting code. Only relaxed, acquire, and release operations w/
volatile are handled.

This PR does not aim to add support for any of the following:
- syncscope support
- read atomic ops to const, param, grid param
- local memory atomics
- sequentially consistent atomics
- atomicrmw 
- ...

Author: gonzalobg

llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXInstPrinter.cpp

@@ -227,9 +227,33 @@ void NVPTXInstPrinter::printLdStCode(const MCInst *MI, int OpNum,
   if (Modifier) {
     const MCOperand &MO = MI->getOperand(OpNum);
     int Imm = (int) MO.getImm();
-    if (!strcmp(Modifier, "volatile")) {
-      if (Imm)
+    if (!strcmp(Modifier, "sem")) {
+      switch (Imm) {
+      case NVPTX::PTXLdStInstCode::NotAtomic:
+        break;
+      case NVPTX::PTXLdStInstCode::Volatile:
         O << ".volatile";
+        break;
+      case NVPTX::PTXLdStInstCode::Relaxed:
+        O << ".relaxed.sys";
+        break;
+      case NVPTX::PTXLdStInstCode::Acquire:
+        O << ".acquire.sys";
+        break;
+      case NVPTX::PTXLdStInstCode::Release:
+        O << ".release.sys";
+        break;
+      case NVPTX::PTXLdStInstCode::RelaxedMMIO:
+        O << ".mmio.relaxed.sys";
+        break;
+      default:
+        SmallString<256> Msg;
+        raw_svector_ostream OS(Msg);
+        OS << "NVPTX LdStCode Printer does not support \"" << Imm
+           << "\" sem modifier.";
+        report_fatal_error(OS.str());
+        break;
+      }
     } else if (!strcmp(Modifier, "addsp")) {
       switch (Imm) {
       case NVPTX::PTXLdStInstCode::GLOBAL:

llvm/lib/Target/NVPTX/NVPTX.h

@@ -107,6 +107,14 @@ enum LoadStore {
 };
 
 namespace PTXLdStInstCode {
+enum MemorySemantic {
+  NotAtomic = 0, // PTX calls these: "Weak"
+  Volatile = 1,
+  Relaxed = 2,
+  Acquire = 3,
+  Release = 4,
+  RelaxedMMIO = 5
+};
 enum AddressSpace {
   GENERIC = 0,
   GLOBAL = 1,