[RISC-V] Enable TLS field access optimization

src/coreclr/jit/codegenriscv64.cpp

@@ -1607,7 +1607,17 @@ void CodeGen::genCodeForStoreLclVar(GenTreeLclVar* lclNode)
         }
         else // store into register (i.e move into register)
         {
-            inst_Mov(targetType, targetReg, dataReg, true);
+            if (data->IsIconHandle(GTF_ICON_TLS_HDL))
+            {
+                assert(data->AsIntCon()->IconValue() == 0);
+                emitAttr attr = emitActualTypeSize(targetType);
+                // need to load the address from thread pointer reg
+                emit->emitIns_R_R(INS_mov, attr, targetReg, REG_TP);
+            }
+            else
+            {
+                inst_Mov(targetType, targetReg, dataReg, true);
+            }
             genProduceReg(lclNode);
         }
     }

src/coreclr/jit/helperexpansion.cpp

@@ -652,17 +652,21 @@ bool Compiler::fgExpandThreadLocalAccessForCall(BasicBlock** pBlock, Statement*
 #ifdef UNIX_X86_ABI
         tlsRefCall->gtFlags &= ~GTF_CALL_POP_ARGS;
 #endif // UNIX_X86_ABI
-#elif defined(TARGET_ARM64)
+#elif defined(TARGET_ARM64) || defined(TARGET_LOONGARCH64) || defined(TARGET_RISCV64)
         // Code sequence to access thread local variable on linux/arm64:
         //
         //      mrs xt, tpidr_elf0
         //      mov xd, [xt+cns]
-        tlsValue = gtNewIconHandleNode(0, GTF_ICON_TLS_HDL);
-#elif defined(TARGET_LOONGARCH64)
+        //
         // Code sequence to access thread local variable on linux/loongarch64:
         //
         //      ori, targetReg, $tp, 0
         //      load rd, targetReg, cns
+        //
+        // Code sequence to access thread local variable on linux/riscv64:
+        //
+        //      mov targetReg, $tp, 0
+        //      ld rd, targetReg(cns)
         tlsValue = gtNewIconHandleNode(0, GTF_ICON_TLS_HDL);
 #else
         assert(!"Unsupported scenario of optimizing TLS access on Linux Arm32/x86");

src/coreclr/vm/jitinterface.cpp

@@ -1414,11 +1414,11 @@ static void* GetTlsIndexObjectAddress()
     return GetThreadStaticDescriptor(p);
 }
 
-#elif defined(TARGET_ARM64) || defined(TARGET_LOONGARCH64)
+#elif defined(TARGET_ARM64) || defined(TARGET_LOONGARCH64) || defined(TARGET_RISCV64)
 
 extern "C" size_t GetThreadStaticsVariableOffset();
 
-#endif // TARGET_ARM64 || TARGET_LOONGARCH64
+#endif // TARGET_ARM64 || TARGET_LOONGARCH64 || TARGET_RISCV64
 #endif // TARGET_WINDOWS
 
 
@@ -1452,9 +1452,9 @@ void CEEInfo::getThreadLocalStaticBlocksInfo (CORINFO_THREAD_STATIC_BLOCKS_INFO*
     pInfo->tlsGetAddrFtnPtr = reinterpret_cast<void*>(&__tls_get_addr);
     pInfo->tlsIndexObject = GetTlsIndexObjectAddress();
 
-#elif defined(TARGET_ARM64) || defined(TARGET_LOONGARCH64)
+#elif defined(TARGET_ARM64) || defined(TARGET_LOONGARCH64) || defined(TARGET_RISCV64)
 
-    // For Linux arm64/loongarch64, just get the offset of thread static variable, and during execution,
+    // For Linux arm64/loongarch64/riscv64, just get the offset of thread static variable, and during execution,
     // this offset, arm64 taken from trpid_elp0 system register gives back the thread variable address.
     // this offset, loongarch64 taken from $tp register gives back the thread variable address.
     threadStaticBaseOffset = GetThreadStaticsVariableOffset();
@@ -1588,20 +1588,18 @@ void CEEInfo::getFieldInfo (CORINFO_RESOLVED_TOKEN * pResolvedToken,
                 // Optimization is disabled for linux/x86
 #elif defined(TARGET_LINUX_MUSL) && defined(TARGET_ARM64)
                 // Optimization is disabled for linux musl arm64
-#elif defined(TARGET_RISCV64)
-                // Optimization is disabled for riscv64
 #else
                 bool optimizeThreadStaticAccess = true;
 #if !defined(TARGET_OSX) && defined(TARGET_UNIX) && defined(TARGET_AMD64)
                 // For linux/x64, check if compiled coreclr as .so file and not single file.
                 // For single file, the `tls_index` might not be accurate.
                 // Do not perform this optimization in such case.
                 optimizeThreadStaticAccess = GetTlsIndexObjectAddress() != nullptr;
-#endif // TARGET_UNIX && TARGET_AMD64
+#endif // !TARGET_OSX && TARGET_UNIX && TARGET_AMD64
 
                 if (optimizeThreadStaticAccess)
                 {
-                    // For windows x64/x86/arm64, linux x64/arm64/loongarch64:
+                    // For windows x64/x86/arm64, linux x64/arm64/loongarch64/riscv64:
                     // We convert the TLS access to the optimized helper where we will store
                     // the static blocks in TLS directly and access them via inline code.
                     if ((pResult->helper == CORINFO_HELP_GETSHARED_NONGCTHREADSTATIC_BASE_NOCTOR) ||

src/coreclr/vm/riscv64/asmhelpers.S

@@ -954,3 +954,12 @@ NESTED_ENTRY OnCallCountThresholdReachedStub, _TEXT, NoHandler
 NESTED_END OnCallCountThresholdReachedStub, _TEXT
 
 #endif // FEATURE_TIERED_COMPILATION
+
+// ------------------------------------------------------------------
+// size_t GetThreadStaticsVariableOffset()
+
+// Load offset of native thread local variable `t_ThreadStatics` in TCB and return it in `a0` register.
+LEAF_ENTRY GetThreadStaticsVariableOffset, _TEXT
+    la.tls.ie   a0, t_ThreadStatics
+    EPILOG_RETURN
+LEAF_END GetThreadStaticsVariableOffset, _TEXT