dotnet · janvorli · Mar 17, 2026 · Mar 15, 2026
diff --git a/src/coreclr/clr.featuredefines.props b/src/coreclr/clr.featuredefines.props
@@ -47,7 +47,7 @@
         <FeatureInterpreter>true</FeatureInterpreter>
     </PropertyGroup>
 
-    <PropertyGroup Condition="('$(Platform)' == 'x64' OR '$(Platform)' == 'arm64' OR '$(Platform)' == 'arm' OR '$(Platform)' == 'riscv64') AND ('$(Configuration)' == 'debug' OR '$(Configuration)' == 'checked') AND '$(TargetsAndroid)' != 'true'">
+    <PropertyGroup Condition="('$(Platform)' == 'x64' OR '$(Platform)' == 'arm64' OR '$(Platform)' == 'arm' OR '$(Platform)' == 'riscv64' OR '$(Platform)' == 'loongarch64') AND ('$(Configuration)' == 'debug' OR '$(Configuration)' == 'checked') AND '$(TargetsAndroid)' != 'true'">
         <FeatureInterpreter>true</FeatureInterpreter>
     </PropertyGroup>
 

diff --git a/src/coreclr/clrfeatures.cmake b/src/coreclr/clrfeatures.cmake
@@ -48,16 +48,17 @@ if(CLR_CMAKE_TARGET_BROWSER)
 endif()
 
 if(NOT DEFINED FEATURE_INTERPRETER)
-  if(CLR_CMAKE_TARGET_ANDROID)
-    set(FEATURE_INTERPRETER 0)
-  else()
-    if(CLR_CMAKE_TARGET_ARCH_AMD64 OR CLR_CMAKE_TARGET_ARCH_ARM64 OR CLR_CMAKE_TARGET_ARCH_ARM OR CLR_CMAKE_TARGET_ARCH_RISCV64)
-      set(FEATURE_INTERPRETER $<IF:$<CONFIG:Debug,Checked>,1,0>)
-    else(CLR_CMAKE_TARGET_ARCH_AMD64 OR CLR_CMAKE_TARGET_ARCH_ARM64 OR CLR_CMAKE_TARGET_ARCH_ARM OR CLR_CMAKE_TARGET_ARCH_RISCV64)
-      set(FEATURE_INTERPRETER 0)
-    endif(CLR_CMAKE_TARGET_ARCH_AMD64 OR CLR_CMAKE_TARGET_ARCH_ARM64 OR CLR_CMAKE_TARGET_ARCH_ARM OR CLR_CMAKE_TARGET_ARCH_RISCV64)
+  set(FEATURE_INTERPRETER 0)
+
+  if(NOT CLR_CMAKE_TARGET_ANDROID AND
+     (CLR_CMAKE_TARGET_ARCH_AMD64 OR
+      CLR_CMAKE_TARGET_ARCH_ARM64 OR
+      CLR_CMAKE_TARGET_ARCH_ARM OR
+      CLR_CMAKE_TARGET_ARCH_RISCV64 OR
+      CLR_CMAKE_TARGET_ARCH_LOONGARCH64))
+    set(FEATURE_INTERPRETER $<IF:$<CONFIG:Debug,Checked>,1,0>)
   endif()
-endif(NOT DEFINED FEATURE_INTERPRETER)
+endif()
 
 if(NOT DEFINED FEATURE_STANDALONE_GC)
   set(FEATURE_STANDALONE_GC 1)

@@ -143,28 +143,7 @@ C_FUNC(\Name):
 
 .endm
 
-// Loads the address of a thread-local variable into the target register,
-// which cannot be a0. Preserves all other registers.
-.macro INLINE_GET_TLS_VAR target, var
-    .ifc  \target, $a0
-        .error "target cannot be a0"
-    .endif
-
-    addi.d  $sp, $sp, -16
-    st.d  $a0, $sp, 0
-    st.d  $ra, $sp, 8
-
-    // This instruction is recognized and potentially patched
-    // by the linker (GD->IE/LE relaxation).
-    la.tls.desc  $a0, \var
-
-    ori  \target, $tp, 0
-    add.d  \target, \target, $a0
-
-    ld.d  $a0, $sp, 0
-    ld.d  $ra, $sp, 8
-    addi.d $sp, $sp, 16
-.endm
+#include <loongarch64/InlineTls.inc>
 
 // Inlined version of RhpGetThread. Target cannot be a0.
 .macro INLINE_GETTHREAD target

diff --git a/src/coreclr/pal/inc/unixasmmacrosloongarch64.inc b/src/coreclr/pal/inc/unixasmmacrosloongarch64.inc
@@ -523,3 +523,11 @@ $__RedirectionStubEndFuncName
 0:
 #endif
 .endm
+
+#include <loongarch64/InlineTls.inc>
+
+// Inlined version of RhpGetThread. Target cannot be a0.
+.macro INLINE_GETTHREAD target
+        INLINE_GET_TLS_VAR \target, C_FUNC(t_CurrentThreadInfo)
+        ld.d  \target, \target, OFFSETOF__ThreadLocalInfo__m_pThread
+.endm
diff --git a/src/coreclr/runtime/loongarch64/InlineTls.inc b/src/coreclr/runtime/loongarch64/InlineTls.inc
@@ -0,0 +1,25 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+// Loads the address of a thread-local variable into the target register,
+// which cannot be a0. Preserves all other registers.
+.macro INLINE_GET_TLS_VAR target, var
+    .ifc  \target, $a0
+        .error "target cannot be a0"
+    .endif
+
+    addi.d  $sp, $sp, -16
+    st.d  $a0, $sp, 0
+    st.d  $ra, $sp, 8
+
+    // This instruction is recognized and potentially patched
+    // by the linker (GD->IE/LE relaxation).
+    la.tls.desc  $a0, \var
+
+    ori  \target, $tp, 0
+    add.d  \target, \target, $a0
+
+    ld.d  $a0, $sp, 0
+    ld.d  $ra, $sp, 8
+    addi.d $sp, $sp, 16
+.endm
diff --git a/src/coreclr/vm/callstubgenerator.cpp b/src/coreclr/vm/callstubgenerator.cpp
@@ -615,7 +615,7 @@ extern "C" void Store_Stack_4B();
 
 #endif // TARGET_ARM
 
-#ifdef TARGET_RISCV64
+#if defined(TARGET_RISCV64) || defined(TARGET_LOONGARCH64)
 
 extern "C" void Load_A0();
 extern "C" void Load_A0_A1();
@@ -783,7 +783,7 @@ extern "C" void Store_FA6();
 extern "C" void Store_FA6_FA7();
 extern "C" void Store_FA7();
 
-#endif // TARGET_RISCV64
+#endif // TARGET_RISCV64 || TARGET_LOONGARCH64
 
 PCODE CallStubGenerator::GetStackRoutine()
 {
@@ -879,7 +879,7 @@ PCODE CallStubGenerator::GetGPRegRangeRoutine(int r1, int r2)
         (PCODE)0, (PCODE)0, (PCODE)Store_R2, (PCODE)Store_R2_R3,
         (PCODE)0, (PCODE)0, (PCODE)0, (PCODE)Store_R3
     };
-#elif defined(TARGET_RISCV64)
+#elif defined(TARGET_RISCV64) || defined(TARGET_LOONGARCH64)
     static const PCODE GPRegsLoadRoutines[] = {
         (PCODE)Load_A0, (PCODE)Load_A0_A1, (PCODE)Load_A0_A1_A2, (PCODE)Load_A0_A1_A2_A3, (PCODE)Load_A0_A1_A2_A3_A4, (PCODE)Load_A0_A1_A2_A3_A4_A5, (PCODE)Load_A0_A1_A2_A3_A4_A5_A6, (PCODE)Load_A0_A1_A2_A3_A4_A5_A6_A7,
         (PCODE)0, (PCODE)Load_A1, (PCODE)Load_A1_A2, (PCODE)Load_A1_A2_A3, (PCODE)Load_A1_A2_A3_A4, (PCODE)Load_A1_A2_A3_A4_A5, (PCODE)Load_A1_A2_A3_A4_A5_A6, (PCODE)Load_A1_A2_A3_A4_A5_A6_A7,
@@ -931,7 +931,7 @@ PCODE CallStubGenerator::GetGPRegRefRoutine(int r)
         (PCODE)Store_Ref_X0, (PCODE)Store_Ref_X1, (PCODE)Store_Ref_X2, (PCODE)Store_Ref_X3,
         (PCODE)Store_Ref_X4, (PCODE)Store_Ref_X5, (PCODE)Store_Ref_X6, (PCODE)Store_Ref_X7
     };
-#elif defined(TARGET_RISCV64)
+#elif defined(TARGET_RISCV64) || defined(TARGET_LOONGARCH64)
     static const PCODE GPRegsRefLoadRoutines[] = {
         (PCODE)Load_Ref_A0, (PCODE)Load_Ref_A1, (PCODE)Load_Ref_A2, (PCODE)Load_Ref_A3,
         (PCODE)Load_Ref_A4, (PCODE)Load_Ref_A5, (PCODE)Load_Ref_A6, (PCODE)Load_Ref_A7
@@ -1017,7 +1017,7 @@ PCODE CallStubGenerator::GetFPRegRangeRoutine(int x1, int x2)
         (PCODE)0, (PCODE)0, (PCODE)0, (PCODE)0, (PCODE)0, (PCODE)0, (PCODE)Store_D6, (PCODE)Store_D6_D7,
         (PCODE)0, (PCODE)0, (PCODE)0, (PCODE)0, (PCODE)0, (PCODE)0, (PCODE)0, (PCODE)Store_D7
     };
-#elif defined(TARGET_RISCV64)
+#elif defined(TARGET_RISCV64) || defined(TARGET_LOONGARCH64)
     static const PCODE FPRegsLoadRoutines[] = {
         (PCODE)Load_FA0, (PCODE)Load_FA0_FA1, (PCODE)Load_FA0_FA1_FA2, (PCODE)Load_FA0_FA1_FA2_FA3, (PCODE)Load_FA0_FA1_FA2_FA3_FA4, (PCODE)Load_FA0_FA1_FA2_FA3_FA4_FA5, (PCODE)Load_FA0_FA1_FA2_FA3_FA4_FA5_FA6, (PCODE)Load_FA0_FA1_FA2_FA3_FA4_FA5_FA6_FA7,
         (PCODE)0, (PCODE)Load_FA1, (PCODE)Load_FA1_FA2, (PCODE)Load_FA1_FA2_FA3, (PCODE)Load_FA1_FA2_FA3_FA4, (PCODE)Load_FA1_FA2_FA3_FA4_FA5, (PCODE)Load_FA1_FA2_FA3_FA4_FA5_FA6, (PCODE)Load_FA1_FA2_FA3_FA4_FA5_FA6_FA7,
@@ -1315,16 +1315,20 @@ extern "C" void InterpreterStubRet3Vector128();
 extern "C" void InterpreterStubRet4Vector128();
 #endif // TARGET_ARM64
 
-#if defined(TARGET_RISCV64)
+#if defined(TARGET_RISCV64) || defined(TARGET_LOONGARCH64)
+extern "C" void CallJittedMethodRetFloat(PCODE *routines, int8_t*pArgs, int8_t*pRet, int totalStackSize, PTR_PTR_Object pContinuation);
+extern "C" void CallJittedMethodRet2Float(PCODE *routines, int8_t*pArgs, int8_t*pRet, int totalStackSize, PTR_PTR_Object pContinuation);
 extern "C" void CallJittedMethodRet2I8(PCODE *routines, int8_t*pArgs, int8_t*pRet, int totalStackSize, PTR_PTR_Object pContinuation);
 extern "C" void CallJittedMethodRet2Double(PCODE *routines, int8_t*pArgs, int8_t*pRet, int totalStackSize, PTR_PTR_Object pContinuation);
 extern "C" void CallJittedMethodRetFloatInt(PCODE *routines, int8_t*pArgs, int8_t*pRet, int totalStackSize, PTR_PTR_Object pContinuation);
 extern "C" void CallJittedMethodRetIntFloat(PCODE *routines, int8_t*pArgs, int8_t*pRet, int totalStackSize, PTR_PTR_Object pContinuation);
+extern "C" void InterpreterStubRetFloat();
+extern "C" void InterpreterStubRet2Float();
 extern "C" void InterpreterStubRet2I8();
 extern "C" void InterpreterStubRet2Double();
 extern "C" void InterpreterStubRetFloatInt();
 extern "C" void InterpreterStubRetIntFloat();
-#endif // TARGET_RISCV64
+#endif // TARGET_RISCV64 || TARGET_LOONGARCH64
 
 #define INVOKE_FUNCTION_PTR(functionPtrName) LOG2((LF2_INTERPRETER, LL_INFO10000, #functionPtrName "\n")); return functionPtrName
 
@@ -1432,7 +1436,11 @@ CallStubHeader::InvokeFunctionPtr CallStubGenerator::GetInvokeFunctionPtr(CallSt
             INVOKE_FUNCTION_PTR(CallJittedMethodRetSwiftLowered);
 #endif // TARGET_APPLE
 #endif // TARGET_ARM64
-#if defined(TARGET_RISCV64)
+#if defined(TARGET_RISCV64) || defined(TARGET_LOONGARCH64)
+        case ReturnTypeFloat:
+            INVOKE_FUNCTION_PTR(CallJittedMethodRetFloat);
+        case ReturnType2Float:
+            INVOKE_FUNCTION_PTR(CallJittedMethodRet2Float);
         case ReturnType2I8:
             INVOKE_FUNCTION_PTR(CallJittedMethodRet2I8);
         case ReturnType2Double:
@@ -1441,7 +1449,7 @@ CallStubHeader::InvokeFunctionPtr CallStubGenerator::GetInvokeFunctionPtr(CallSt
             INVOKE_FUNCTION_PTR(CallJittedMethodRetFloatInt);
         case ReturnTypeIntFloat:
             INVOKE_FUNCTION_PTR(CallJittedMethodRetIntFloat);
-#endif // TARGET_RISCV64
+#endif // TARGET_RISCV64 || TARGET_LOONGARCH64
         default:
             _ASSERTE(!"Unexpected return type for interpreter stub");
             return NULL; // This should never happen, but just in case.
@@ -1547,7 +1555,11 @@ PCODE CallStubGenerator::GetInterpreterReturnTypeHandler(CallStubGenerator::Retu
         case ReturnType4Vector128:
             RETURN_TYPE_HANDLER(InterpreterStubRet4Vector128);
 #endif // TARGET_ARM64
-#if defined(TARGET_RISCV64)
+#if defined(TARGET_RISCV64) || defined(TARGET_LOONGARCH64)
+        case ReturnTypeFloat:
+            RETURN_TYPE_HANDLER(InterpreterStubRetFloat);
+        case ReturnType2Float:
+            RETURN_TYPE_HANDLER(InterpreterStubRet2Float);
         case ReturnType2I8:
             RETURN_TYPE_HANDLER(InterpreterStubRet2I8);
         case ReturnType2Double:
@@ -1556,7 +1568,7 @@ PCODE CallStubGenerator::GetInterpreterReturnTypeHandler(CallStubGenerator::Retu
             RETURN_TYPE_HANDLER(InterpreterStubRetFloatInt);
         case ReturnTypeIntFloat:
             RETURN_TYPE_HANDLER(InterpreterStubRetIntFloat);
-#endif // TARGET_RISCV64
+#endif // TARGET_RISCV64 || TARGET_LOONGARCH64
         default:
             _ASSERTE(!"Unexpected return type for interpreter stub");
             return 0; // This should never happen, but just in case.
@@ -2665,9 +2677,9 @@ CallStubGenerator::ReturnType CallStubGenerator::GetReturnType(ArgIteratorType *
                 return ReturnTypeI8;
                 break;
             case ELEMENT_TYPE_R4:
-#if defined(TARGET_ARM64) || defined(TARGET_32BIT)
+#if defined(TARGET_ARM64) || defined(TARGET_32BIT) || defined(TARGET_RISCV64) || defined(TARGET_LOONGARCH64)
                 return ReturnTypeFloat;
-#endif // TARGET_ARM64 || TARGET_32BIT
+#endif // TARGET_ARM64 || TARGET_32BIT || TARGET_RISCV64 || TARGET_LOONGARCH64
             case ELEMENT_TYPE_R8:
                 return ReturnTypeDouble;
                 break;
@@ -2821,19 +2833,21 @@ CallStubGenerator::ReturnType CallStubGenerator::GetReturnType(ArgIteratorType *
                         _ASSERTE(!"The return types should be <= 8 bytes in size");
                         break;
                 }
-#elif defined(TARGET_RISCV64)
+#elif defined(TARGET_RISCV64) || defined(TARGET_LOONGARCH64)
                 {
                     FpStructInRegistersInfo info = pArgIt->GetReturnFpStructInRegistersInfo();
-                    // RISC-V pass floating-point struct fields in FA registers
+                    // RISC-V and LoongArch64 pass floating-point struct fields in FA/F registers.
+                    // Preserve 32-bit float width where applicable to avoid reading/writing
+                    // packed 4-byte fields via 8-byte helpers.
                     if ((info.flags & FpStruct::OnlyOne) != 0)
                     {
-                        // Single field - could be float or int in single register
-                        return ReturnTypeDouble; // Use Double routine for both float and double (NaN-boxed)
+                        return (info.Size1st() == sizeof(float)) ? ReturnTypeFloat : ReturnTypeDouble;
                     }
                     else if ((info.flags & FpStruct::BothFloat) != 0)
                     {
-                        // Two float/double fields
-                        return ReturnType2Double;
+                        return (info.Size1st() == sizeof(float) && info.Size2nd() == sizeof(float))
+                            ? ReturnType2Float
+                            : ReturnType2Double;
                     }
                     else if ((info.flags & FpStruct::FloatInt) != 0)
                     {

diff --git a/src/coreclr/vm/callstubgenerator.h b/src/coreclr/vm/callstubgenerator.h
@@ -124,12 +124,14 @@ class CallStubGenerator
         ReturnTypeSwiftLowered,
 #endif // TARGET_APPLE
 #endif // TARGET_ARM64
-#if defined(TARGET_RISCV64)
+#if defined(TARGET_RISCV64) || defined(TARGET_LOONGARCH64)
+        ReturnTypeFloat,
+        ReturnType2Float,
         ReturnType2I8,
         ReturnType2Double,
         ReturnTypeFloatInt,
         ReturnTypeIntFloat,
-#endif // TARGET_RISCV64
+#endif // TARGET_RISCV64 || TARGET_LOONGARCH64
     };
 
     enum class RoutineType

diff --git a/src/coreclr/vm/loongarch64/asmconstants.h b/src/coreclr/vm/loongarch64/asmconstants.h
@@ -52,6 +52,9 @@ ASMCONSTANTS_C_ASSERT(Thread__m_pFrame == offsetof(Thread, m_pFrame));
 #define               OFFSETOF__RuntimeThreadLocals__ee_alloc_context 0
 ASMCONSTANTS_C_ASSERT(OFFSETOF__RuntimeThreadLocals__ee_alloc_context == offsetof(RuntimeThreadLocals, alloc_context));
 
+#define               OFFSETOF__ThreadLocalInfo__m_pThread 0
+ASMCONSTANTS_C_ASSERT(OFFSETOF__ThreadLocalInfo__m_pThread == offsetof(ThreadLocalInfo, m_pThread));
+
 #define               OFFSETOF__ee_alloc_context__alloc_ptr 0x8
 ASMCONSTANTS_C_ASSERT(OFFSETOF__ee_alloc_context__alloc_ptr == offsetof(ee_alloc_context, m_GCAllocContext) +
                                                                offsetof(gc_alloc_context, alloc_ptr));
@@ -219,6 +222,23 @@ ASMCONSTANTS_C_ASSERT(StubPrecodeData__Target            == offsetof(StubPrecode
 #define StubPrecodeData__SecretParam 0x00
 ASMCONSTANTS_C_ASSERT(StubPrecodeData__SecretParam        == offsetof(StubPrecodeData, SecretParam))
 
+#ifdef FEATURE_INTERPRETER
+#define OFFSETOF__InterpMethod__pCallStub 0x20
+ASMCONSTANTS_C_ASSERT(OFFSETOF__InterpMethod__pCallStub == offsetof(InterpMethod, pCallStub))
+
+#define OFFSETOF__Thread__m_pInterpThreadContext 0xB20
+ASMCONSTANTS_C_ASSERT(OFFSETOF__Thread__m_pInterpThreadContext == offsetof(Thread, m_pInterpThreadContext))
+
+#define OFFSETOF__InterpThreadContext__pStackPointer 0x10
+ASMCONSTANTS_C_ASSERT(OFFSETOF__InterpThreadContext__pStackPointer == offsetof(InterpThreadContext, pStackPointer))
+
+#define OFFSETOF__CallStubHeader__Routines 0x18
+ASMCONSTANTS_C_ASSERT(OFFSETOF__CallStubHeader__Routines == offsetof(CallStubHeader, Routines))
+
+#define SIZEOF__TransitionBlock 0xA0
+ASMCONSTANTS_C_ASSERT(SIZEOF__TransitionBlock == sizeof(TransitionBlock))
+#endif // FEATURE_INTERPRETER
+
 #ifdef FEATURE_TIERED_COMPILATION
 #define CallCountingStubData__RemainingCallCountCell 0x00
 ASMCONSTANTS_C_ASSERT(CallCountingStubData__RemainingCallCountCell == offsetof(CallCountingStubData, RemainingCallCountCell))