[LoongArch64] amend the crossgen2 for LA64. (#75118)

* [LoongArch64] amend the crossgen2 for LA64.

* amend the `ComputeReturnValueTreatment` for CR.

diff --git a/src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/ArgIterator.cs b/src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/ArgIterator.cs
index 73e0a1d..70e258d 100644 (file)
--- a/src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/ArgIterator.cs
+++ b/src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/ArgIterator.cs
@@ -1364,6 +1364,7 @@ namespace ILCompiler.DependencyAnalysis.ReadyToRun
                         {
                             if (isValueType && ((floatFieldFlags & (uint)StructFloatFieldInfoFlags.STRUCT_HAS_ONE_FLOAT_MASK) != 0))
                             {
+                                Debug.Assert(cFPRegs == 1);
                                 if ((_loongarch64IdxFPReg < 8) && (_loongarch64IdxGenReg < 8))
                                 {
                                     _argLocDescForStructInRegs = new ArgLocDesc();
@@ -1381,15 +1382,23 @@ namespace ILCompiler.DependencyAnalysis.ReadyToRun
                                     _loongarch64IdxGenReg++;
                                     return argOfsInner;
                                 }
-                                else
-                                {
-                                    _loongarch64IdxFPReg = 8;
-                                }
                             }
                             else if (cFPRegs + _loongarch64IdxFPReg <= 8)
                             {
                                 // Each floating point register in the argument area is 8 bytes.
                                 int argOfsInner = _transitionBlock.OffsetOfFloatArgumentRegisters + _loongarch64IdxFPReg * 8;
+                                if (floatFieldFlags == (uint)StructFloatFieldInfoFlags.STRUCT_FLOAT_FIELD_ONLY_TWO)
+                                {
+                                    // struct with two single-float fields.
+                                    _argLocDescForStructInRegs = new ArgLocDesc();
+                                    _argLocDescForStructInRegs.m_idxFloatReg = _loongarch64IdxFPReg;
+                                    _argLocDescForStructInRegs.m_cFloatReg = 2;
+                                    Debug.Assert(cFPRegs == 2);
+                                    Debug.Assert(argSize == 8);
+
+                                    _hasArgLocDescForStructInRegs = true;
+                                    _argLocDescForStructInRegs.m_floatFlags = (uint)StructFloatFieldInfoFlags.STRUCT_FLOAT_FIELD_ONLY_TWO;
+                                }
                                 _loongarch64IdxFPReg += cFPRegs;
                                 return argOfsInner;
                             }
@@ -1418,11 +1427,6 @@ namespace ILCompiler.DependencyAnalysis.ReadyToRun
                                 _loongarch64OfsStack += 8;
                                 return argOfsInner;
                             }
-                            else
-                            {
-                                // Don't use reg slots for this. It will be passed purely on the stack arg space.
-                                _loongarch64IdxGenReg = 8;
-                            }
                         }
 
                         argOfs = _transitionBlock.OffsetOfArgs + _loongarch64OfsStack;

diff --git a/src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/TransitionBlock.cs b/src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/TransitionBlock.cs
index 8cf01a8..2228cb2 100644 (file)
--- a/src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/TransitionBlock.cs
+++ b/src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/TransitionBlock.cs
@@ -383,7 +383,12 @@ namespace ILCompiler.DependencyAnalysis.ReadyToRun
                             }
 
                             if (size <= EnregisteredReturnTypeIntegerMaxSize)
+                            {
+                                if (IsLoongArch64)
+                                    fpReturnSize = LoongArch64PassStructInRegister.GetLoongArch64PassStructInRegisterFlags(thRetType.GetRuntimeTypeHandle()) & 0xff;
                                 break;
+                            }
+
                         }
                     }
 

diff --git a/src/coreclr/vm/loongarch64/stubs.cpp b/src/coreclr/vm/loongarch64/stubs.cpp
index d868fd9..12d56e1 100644 (file)
--- a/src/coreclr/vm/loongarch64/stubs.cpp
+++ b/src/coreclr/vm/loongarch64/stubs.cpp
@@ -1791,22 +1791,26 @@ PCODE DynamicHelpers::CreateDictionaryLookupHelper(LoaderAllocator * pAllocator,
     {
         int codeSize = 0;
         int indirectionsDataSize = 0;
-        if (pLookup->sizeOffset != CORINFO_NO_SIZE_CHECK)
+        if (pLookup->testForNull || pLookup->sizeOffset != CORINFO_NO_SIZE_CHECK)
         {
-            codeSize += (pLookup->sizeOffset > 2047 ? 8 : 4);
-            indirectionsDataSize += (pLookup->sizeOffset > 2047 ? 4 : 0);
-            codeSize += 12;
+            codeSize += 4;
         }
 
         for (WORD i = 0; i < pLookup->indirections; i++) {
             _ASSERTE(pLookup->offsets[i] >= 0);
+            if (i == pLookup->indirections - 1 && pLookup->sizeOffset != CORINFO_NO_SIZE_CHECK)
+            {
+                codeSize += (pLookup->sizeOffset > 2047 ? 24 : 16);
+                indirectionsDataSize += (pLookup->sizeOffset > 2047 ? 4 : 0);
+            }
+
             codeSize += (pLookup->offsets[i] > 2047 ? 8 : 4); // if( > 2047) (8 bytes) else 4 bytes for instructions.
             indirectionsDataSize += (pLookup->offsets[i] > 2047 ? 4 : 0); // 4 bytes for storing indirection offset values
         }
 
         codeSize += indirectionsDataSize ? 4 : 0; // pcaddi
 
-        if(pLookup->testForNull)
+        if (pLookup->testForNull)
         {
             codeSize += 12; // ori-beq-jr
 
@@ -1821,7 +1825,7 @@ PCODE DynamicHelpers::CreateDictionaryLookupHelper(LoaderAllocator * pAllocator,
             codeSize += 4; /* jilr */
         }
 
-        // the offset value of data.
+        // the offset value of data_label.
         uint dataOffset = codeSize;
 
         codeSize += indirectionsDataSize;
@@ -1877,9 +1881,9 @@ PCODE DynamicHelpers::CreateDictionaryLookupHelper(LoaderAllocator * pAllocator,
                 *(DWORD*)p = 0x03800210 | (((UINT32)slotOffset & 0xfff) << 10); p += 4;
                 dataOffset -= 8;
 
-                // bge $t4,$t3, // CALL HELPER:
+                // bge $t4,$t5, // CALL HELPER:
                 pBLECall = p;       // Offset filled later
-                *(DWORD*)p = 0x6400020f; p += 4;
+                *(DWORD*)p = 0x64000211; p += 4;
                 dataOffset -= 4;
             }
 
@@ -1889,10 +1893,12 @@ PCODE DynamicHelpers::CreateDictionaryLookupHelper(LoaderAllocator * pAllocator,
                 // ld.wu  $t4,$r21,0
                 *(DWORD*)p = 0x2a8002b0 | (dataOffset<<10);
                 p += 4;
-                dataOffset += 4;
                 // ldx.d  $a0,$a0,$t4
                 *(DWORD*)p = 0x380c4084;
                 p += 4;
+
+                // move to next indirection offset data
+                dataOffset = dataOffset - 8 + 4; // subtract 8 as we have moved PC by 8 and add 4 as next data is at 4 bytes from previous data
             }
             else
             {
@@ -1902,6 +1908,7 @@ PCODE DynamicHelpers::CreateDictionaryLookupHelper(LoaderAllocator * pAllocator,
                 // ld.d  $a0,$a0,pLookup->offsets[i]
                 *(DWORD*)p = 0x28c00084 | ((pLookup->offsets[i] & 0xfff)<<10);
                 p += 4;
+                dataOffset -= 4; // subtract 4 as we have moved PC by 4
             }
         }
 
@@ -1925,7 +1932,7 @@ PCODE DynamicHelpers::CreateDictionaryLookupHelper(LoaderAllocator * pAllocator,
 
             // CALL HELPER:
             if(pBLECall != NULL)
-                *(DWORD*)pBLECall |= ((UINT32)(p - pBLECall) << 10);
+                *(DWORD*)pBLECall |= ((UINT32)(p - pBLECall) << 8);
 
             // ori  $a0,$t3,0
             *(DWORD*)p = 0x038001e4;
@@ -1943,7 +1950,7 @@ PCODE DynamicHelpers::CreateDictionaryLookupHelper(LoaderAllocator * pAllocator,
             EmitHelperWithArg(p, rxOffset, pAllocator, (TADDR)pArgs, helperAddress);
         }
 
-        // datalabel:
+        // data_label:
         for (WORD i = 0; i < pLookup->indirections; i++)
         {
             if (i == pLookup->indirections - 1 && pLookup->sizeOffset != CORINFO_NO_SIZE_CHECK && pLookup->sizeOffset > 2047)