dotnet · TIHan · Aug 2, 2024 · Jul 2, 2024 · Jul 2, 2024 · Jul 2, 2024
diff --git a/src/coreclr/jit/compiler.hpp b/src/coreclr/jit/compiler.hpp
@@ -4245,6 +4245,10 @@ bool Compiler::fgVarIsNeverZeroInitializedInProlog(unsigned varNum)
     bool       result = varDsc->lvIsParam || lvaIsOSRLocal(varNum) || (varNum == lvaGSSecurityCookie) ||
                   (varNum == lvaInlinedPInvokeFrameVar) || (varNum == lvaStubArgumentVar) || (varNum == lvaRetAddrVar);
 
+#ifdef TARGET_ARM64
+    result = result || (varNum == lvaFfrRegister);
+#endif
+
 #if FEATURE_FIXED_OUT_ARGS
     result = result || (varNum == lvaOutgoingArgSpaceVar);
 #endif

diff --git a/src/coreclr/jit/fgdiagnostic.cpp b/src/coreclr/jit/fgdiagnostic.cpp
@@ -3428,14 +3428,15 @@ void Compiler::fgDebugCheckFlags(GenTree* tree, BasicBlock* block)
 
 #if defined(TARGET_ARM64)
                     case NI_ArmBase_Yield:
-                    case NI_Sve_PrefetchBytes:
-                    case NI_Sve_PrefetchInt16:
-                    case NI_Sve_PrefetchInt32:
-                    case NI_Sve_PrefetchInt64:
                     case NI_Sve_GatherPrefetch16Bit:
                     case NI_Sve_GatherPrefetch32Bit:
                     case NI_Sve_GatherPrefetch64Bit:
                     case NI_Sve_GatherPrefetch8Bit:
+                    case NI_Sve_PrefetchBytes:
+                    case NI_Sve_PrefetchInt16:
+                    case NI_Sve_PrefetchInt32:
+                    case NI_Sve_PrefetchInt64:
+                    case NI_Sve_SetFfr:
                     {
                         assert(tree->OperRequiresCallFlag(this));
                         expectedFlags |= GTF_GLOB_REF;

diff --git a/src/coreclr/jit/gentree.cpp b/src/coreclr/jit/gentree.cpp
@@ -26743,6 +26743,18 @@ bool GenTreeHWIntrinsic::OperIsMemoryLoad(GenTree** pAddr) const
                 addr = Op(3);
                 break;
 
+            case NI_Sve_GatherVector:
+            case NI_Sve_GatherVectorByteZeroExtend:
+            case NI_Sve_GatherVectorFirstFaulting:
+            case NI_Sve_GatherVectorInt16SignExtend:
+            case NI_Sve_GatherVectorInt16WithByteOffsetsSignExtend:
+            case NI_Sve_GatherVectorInt32SignExtend:
+            case NI_Sve_GatherVectorInt32WithByteOffsetsSignExtend:
+            case NI_Sve_GatherVectorSByteSignExtend:
+            case NI_Sve_GatherVectorUInt16WithByteOffsetsZeroExtend:
+            case NI_Sve_GatherVectorUInt16ZeroExtend:
+            case NI_Sve_GatherVectorUInt32WithByteOffsetsZeroExtend:
+            case NI_Sve_GatherVectorUInt32ZeroExtend:
             case NI_Sve_GatherVectorWithByteOffsets:
             case NI_Sve_LoadVector:
             case NI_Sve_LoadVectorNonTemporal:
@@ -26753,6 +26765,7 @@ bool GenTreeHWIntrinsic::OperIsMemoryLoad(GenTree** pAddr) const
             case NI_Sve_LoadVectorByteZeroExtendToUInt16:
             case NI_Sve_LoadVectorByteZeroExtendToUInt32:
             case NI_Sve_LoadVectorByteZeroExtendToUInt64:
+            case NI_Sve_LoadVectorFirstFaulting:
             case NI_Sve_LoadVectorInt16SignExtendToInt32:
             case NI_Sve_LoadVectorInt16SignExtendToInt64:
             case NI_Sve_LoadVectorInt16SignExtendToUInt32:
@@ -26777,20 +26790,6 @@ bool GenTreeHWIntrinsic::OperIsMemoryLoad(GenTree** pAddr) const
                 addr = Op(2);
                 break;
 
-            case NI_Sve_GatherVector:
-            case NI_Sve_GatherVectorByteZeroExtend:
-            case NI_Sve_GatherVectorInt16SignExtend:
-            case NI_Sve_GatherVectorInt16WithByteOffsetsSignExtend:
-            case NI_Sve_GatherVectorInt32SignExtend:
-            case NI_Sve_GatherVectorInt32WithByteOffsetsSignExtend:
-            case NI_Sve_GatherVectorSByteSignExtend:
-            case NI_Sve_GatherVectorUInt16WithByteOffsetsZeroExtend:
-            case NI_Sve_GatherVectorUInt16ZeroExtend:
-            case NI_Sve_GatherVectorUInt32WithByteOffsetsZeroExtend:
-            case NI_Sve_GatherVectorUInt32ZeroExtend:
-                addr = Op(2);
-                break;
-
 #endif // TARGET_ARM64
 
             default:
@@ -26870,11 +26869,12 @@ bool GenTreeHWIntrinsic::OperIsMemoryLoad(GenTree** pAddr) const
     {
 #ifdef TARGET_ARM64
         static_assert_no_msg(
-            AreContiguous(NI_Sve_GatherVector, NI_Sve_GatherVectorByteZeroExtend, NI_Sve_GatherVectorInt16SignExtend,
-                          NI_Sve_GatherVectorInt16WithByteOffsetsSignExtend, NI_Sve_GatherVectorInt32SignExtend,
-                          NI_Sve_GatherVectorInt32WithByteOffsetsSignExtend, NI_Sve_GatherVectorSByteSignExtend,
-                          NI_Sve_GatherVectorUInt16WithByteOffsetsZeroExtend, NI_Sve_GatherVectorUInt16ZeroExtend,
-                          NI_Sve_GatherVectorUInt32WithByteOffsetsZeroExtend, NI_Sve_GatherVectorUInt32ZeroExtend));
+            AreContiguous(NI_Sve_GatherVector, NI_Sve_GatherVectorByteZeroExtend, NI_Sve_GatherVectorFirstFaulting,
+                          NI_Sve_GatherVectorInt16SignExtend, NI_Sve_GatherVectorInt16WithByteOffsetsSignExtend,
+                          NI_Sve_GatherVectorInt32SignExtend, NI_Sve_GatherVectorInt32WithByteOffsetsSignExtend,
+                          NI_Sve_GatherVectorSByteSignExtend, NI_Sve_GatherVectorUInt16WithByteOffsetsZeroExtend,
+                          NI_Sve_GatherVectorUInt16ZeroExtend, NI_Sve_GatherVectorUInt32WithByteOffsetsZeroExtend,
+                          NI_Sve_GatherVectorUInt32ZeroExtend));
         assert(varTypeIsI(addr) || (varTypeIsSIMD(addr) && ((intrinsicId >= NI_Sve_GatherVector) &&
                                                             (intrinsicId <= NI_Sve_GatherVectorUInt32ZeroExtend))));
 #else
@@ -27292,6 +27292,7 @@ bool GenTreeHWIntrinsic::OperRequiresCallFlag() const
             case NI_Sve_GatherPrefetch32Bit:
             case NI_Sve_GatherPrefetch64Bit:
             case NI_Sve_GatherPrefetch8Bit:
+            case NI_Sve_SetFfr:
             {
                 return true;
             }
@@ -27474,14 +27475,15 @@ void GenTreeHWIntrinsic::Initialize(NamedIntrinsic intrinsicId)
 
 #if defined(TARGET_ARM64)
             case NI_ArmBase_Yield:
-            case NI_Sve_PrefetchBytes:
-            case NI_Sve_PrefetchInt16:
-            case NI_Sve_PrefetchInt32:
-            case NI_Sve_PrefetchInt64:
             case NI_Sve_GatherPrefetch16Bit:
             case NI_Sve_GatherPrefetch32Bit:
             case NI_Sve_GatherPrefetch64Bit:
             case NI_Sve_GatherPrefetch8Bit:
+            case NI_Sve_PrefetchBytes:
+            case NI_Sve_PrefetchInt16:
+            case NI_Sve_PrefetchInt32:
+            case NI_Sve_PrefetchInt64:
+            case NI_Sve_SetFfr:
             {
                 // Mark as a call and global reference, much as is done for GT_KEEPALIVE
                 gtFlags |= (GTF_CALL | GTF_GLOB_REF);

diff --git a/src/coreclr/jit/hwintrinsic.cpp b/src/coreclr/jit/hwintrinsic.cpp
@@ -2234,6 +2234,7 @@ GenTree* Compiler::impHWIntrinsic(NamedIntrinsic        intrinsic,
 #elif defined(TARGET_ARM64)
                     case NI_Sve_GatherVector:
                     case NI_Sve_GatherVectorByteZeroExtend:
+                    case NI_Sve_GatherVectorFirstFaulting:
                     case NI_Sve_GatherVectorInt16SignExtend:
                     case NI_Sve_GatherVectorInt16WithByteOffsetsSignExtend:
                     case NI_Sve_GatherVectorInt32SignExtend:

diff --git a/src/coreclr/jit/hwintrinsiccodegenarm64.cpp b/src/coreclr/jit/hwintrinsiccodegenarm64.cpp
@@ -2049,6 +2049,33 @@ void CodeGen::genHWIntrinsic(GenTreeHWIntrinsic* node)
                 break;
             }
 
+            case NI_Sve_GatherVectorFirstFaulting:
+            {
+                if (node->GetAuxiliaryType() == TYP_UNKNOWN)
+                {
+                    if (intrin.numOperands == 3)
+                    {
+                        // We have extra argument which means there is a "use" of FFR here. Restore it back in FFR
+                        // register.
+                        assert(op3Reg != REG_NA);
+                        GetEmitter()->emitIns_R(INS_sve_wrffr, emitSize, op3Reg, opt);
+                    }
+                }
+                else
+                {
+                    // AuxilaryType is added only for numOperands == 3. If there is an extra argument, we need to
+                    // "use" FFR here. Restore it back in FFR register.
+
+                    if (intrin.numOperands == 4)
+                    {
+                        // We have extra argument which means there is a "use" of FFR here. Restore it back in FFR
+                        // register.
+                        assert(op4Reg != REG_NA);
+                        GetEmitter()->emitIns_R(INS_sve_wrffr, emitSize, op4Reg, opt);
+                    }
+                }
+                FALLTHROUGH;
+            }
             case NI_Sve_GatherVector:
             case NI_Sve_GatherVectorByteZeroExtend:
             case NI_Sve_GatherVectorInt16SignExtend:
@@ -2065,25 +2092,24 @@ void CodeGen::genHWIntrinsic(GenTreeHWIntrinsic* node)
                 {
                     // GatherVector...(Vector<T> mask, T* address, Vector<T2> indices)
 
-                    assert(intrin.numOperands == 3);
-                    emitAttr        baseSize = emitActualTypeSize(intrin.baseType);
-                    insScalableOpts sopt     = INS_SCALABLE_OPTS_NONE;
+                    emitAttr baseSize   = emitActualTypeSize(intrin.baseType);
+                    bool isLoadingBytes = ((ins == INS_sve_ld1b) || (ins == INS_sve_ld1sb) || (ins == INS_sve_ldff1b) ||
+                                           (ins == INS_sve_ldff1sb));
+                    insScalableOpts sopt = INS_SCALABLE_OPTS_NONE;
 
-                    if (baseSize == EA_8BYTE)
-                    {
-                        // Index is multiplied.
-                        sopt = (ins == INS_sve_ld1b || ins == INS_sve_ld1sb) ? INS_SCALABLE_OPTS_NONE
-                                                                             : INS_SCALABLE_OPTS_LSL_N;
-                    }
-                    else
+                    if (baseSize == EA_4BYTE)
                     {
                         // Index is sign or zero extended to 64bits, then multiplied.
-                        assert(baseSize == EA_4BYTE);
                         opt = varTypeIsUnsigned(node->GetAuxiliaryType()) ? INS_OPTS_SCALABLE_S_UXTW
                                                                           : INS_OPTS_SCALABLE_S_SXTW;
 
-                        sopt = (ins == INS_sve_ld1b || ins == INS_sve_ld1sb) ? INS_SCALABLE_OPTS_NONE
-                                                                             : INS_SCALABLE_OPTS_MOD_N;
+                        sopt = isLoadingBytes ? INS_SCALABLE_OPTS_NONE : INS_SCALABLE_OPTS_MOD_N;
+                    }
+                    else
+                    {
+                        // Index is multiplied.
+                        assert(baseSize == EA_8BYTE);
+                        sopt = isLoadingBytes ? INS_SCALABLE_OPTS_NONE : INS_SCALABLE_OPTS_LSL_N;
                     }
 
                     GetEmitter()->emitIns_R_R_R_R(ins, emitSize, targetReg, op1Reg, op2Reg, op3Reg, opt, sopt);
@@ -2092,7 +2118,6 @@ void CodeGen::genHWIntrinsic(GenTreeHWIntrinsic* node)
                 {
                     // GatherVector...(Vector<T> mask, Vector<T2> addresses)
 
-                    assert(intrin.numOperands == 2);
                     GetEmitter()->emitIns_R_R_R_I(ins, emitSize, targetReg, op1Reg, op2Reg, 0, opt);
                 }
 

diff --git a/src/coreclr/jit/hwintrinsiclistarm64sve.h b/src/coreclr/jit/hwintrinsiclistarm64sve.h
@@ -111,6 +111,7 @@ HARDWARE_INTRINSIC(Sve,           GatherPrefetch64Bit,
 HARDWARE_INTRINSIC(Sve,           GatherPrefetch8Bit,                                               -1,     -1,      false, {INS_sve_prfb,       INS_sve_prfb,       INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid},     HW_Category_Special,               HW_Flag_Scalable|HW_Flag_BaseTypeFromFirstArg|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_HasImmediateOperand|HW_Flag_HasEnumOperand|HW_Flag_SpecialSideEffect_Other)
 HARDWARE_INTRINSIC(Sve,           GatherVector,                                                     -1,     -1,      false, {INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_sve_ld1w,       INS_sve_ld1w,       INS_sve_ld1d,       INS_sve_ld1d,       INS_sve_ld1w,       INS_sve_ld1d},    HW_Category_MemoryLoad,            HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation)
 HARDWARE_INTRINSIC(Sve,           GatherVectorByteZeroExtend,                                       -1,     -1,      false, {INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_sve_ld1b,       INS_sve_ld1b,       INS_sve_ld1b,       INS_sve_ld1b,       INS_invalid,        INS_invalid},     HW_Category_MemoryLoad,            HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation)
+HARDWARE_INTRINSIC(Sve,           GatherVectorFirstFaulting,                                        -1,     -1,      false, {INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_sve_ldff1w,     INS_sve_ldff1w,     INS_sve_ldff1d,     INS_sve_ldff1d,     INS_sve_ldff1w,     INS_sve_ldff1d},  HW_Category_MemoryLoad,            HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation|HW_Flag_SpecialSideEffectMask)
 HARDWARE_INTRINSIC(Sve,           GatherVectorInt16SignExtend,                                      -1,     -1,      false, {INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_sve_ld1sh,      INS_sve_ld1sh,      INS_sve_ld1sh,      INS_sve_ld1sh,      INS_invalid,        INS_invalid},     HW_Category_MemoryLoad,            HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation)
 HARDWARE_INTRINSIC(Sve,           GatherVectorInt16WithByteOffsetsSignExtend,                       -1,     -1,      false, {INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_sve_ld1sh,      INS_sve_ld1sh,      INS_sve_ld1sh,      INS_sve_ld1sh,      INS_invalid,        INS_invalid},     HW_Category_MemoryLoad,            HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation)
 HARDWARE_INTRINSIC(Sve,           GatherVectorInt32SignExtend,                                      -1,     -1,      false, {INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_invalid,        INS_sve_ld1sw,      INS_sve_ld1sw,      INS_invalid,        INS_invalid},     HW_Category_MemoryLoad,            HW_Flag_Scalable|HW_Flag_SpecialCodeGen|HW_Flag_ExplicitMaskedOperation|HW_Flag_LowMaskedOperation)

diff --git a/src/coreclr/jit/lowerarmarch.cpp b/src/coreclr/jit/lowerarmarch.cpp
@@ -1775,6 +1775,50 @@ GenTree* Lowering::LowerHWIntrinsic(GenTreeHWIntrinsic* node)
 
             break;
         }
+        case NI_Sve_GatherVectorFirstFaulting:
+        {
+            LIR::Use use;
+            bool     foundUse = BlockRange().TryGetUse(node, &use);
+
+            if (m_ffrTrashed)
+            {
+                // Consume the FFR register value from local variable to simulate "use" of FFR,
+                // only if it was trashed. If it was not trashed, we do not have to reload the
+                // contents of the FFR register.
+
+                unsigned lclNum = comp->getFFRegisterVarNum();
+                GenTree* lclVar = comp->gtNewLclvNode(lclNum, TYP_MASK);
+                BlockRange().InsertBefore(node, lclVar);
+                LowerNode(lclVar);
+
+                if (node->GetOperandCount() == 3)
+                {
+                    assert(node->GetAuxiliaryType() != TYP_UNKNOWN);
+                    node->ResetHWIntrinsicId(intrinsicId, comp, node->Op(1), node->Op(2), node->Op(3), lclVar);
+                }
+                else
+                {
+                    assert(node->GetOperandCount() == 2);
+                    node->ResetHWIntrinsicId(intrinsicId, comp, node->Op(1), node->Op(2), lclVar);
+                }
+            }
+
+            if (foundUse)
+            {
+                unsigned   tmpNum    = comp->lvaGrabTemp(true DEBUGARG("Return value result/FFR"));
+                LclVarDsc* tmpVarDsc = comp->lvaGetDesc(tmpNum);
+                tmpVarDsc->lvType    = node->TypeGet();
+                GenTree* storeLclVar;
+                use.ReplaceWithLclVar(comp, tmpNum, &storeLclVar);
+            }
+            else
+            {
+                node->SetUnusedValue();
+            }
+
+            StoreFFRValue(node);
+            break;
+        }
         case NI_Sve_LoadVectorFirstFaulting:
         {
             LIR::Use use;
@@ -1786,7 +1830,8 @@ GenTree* Lowering::LowerHWIntrinsic(GenTreeHWIntrinsic* node)
                 // only if it was trashed. If it was not trashed, we do not have to reload the
                 // contents of the FFR register.
 
-                GenTree* lclVar = comp->gtNewLclvNode(comp->lvaFfrRegister, TYP_MASK);
+                unsigned lclNum = comp->getFFRegisterVarNum();
+                GenTree* lclVar = comp->gtNewLclvNode(lclNum, TYP_MASK);
                 BlockRange().InsertBefore(node, lclVar);
                 LowerNode(lclVar);
 
@@ -4083,8 +4128,10 @@ void Lowering::StoreFFRValue(GenTreeHWIntrinsic* node)
 #ifdef DEBUG
     switch (node->GetHWIntrinsicId())
     {
-        case NI_Sve_SetFfr:
+        case NI_Sve_GatherVectorFirstFaulting:
         case NI_Sve_LoadVectorFirstFaulting:
+        case NI_Sve_SetFfr:
+
             break;
         default:
             assert(!"Unexpected HWIntrinsicId");

diff --git a/src/libraries/Common/tests/TestUtilities/System/Buffers/BoundedMemory.Creation.cs b/src/libraries/Common/tests/TestUtilities/System/Buffers/BoundedMemory.Creation.cs
@@ -10,6 +10,9 @@ namespace System.Buffers
     /// </summary>
     public static partial class BoundedMemory
     {
+        public static bool UnixBoundsEnabled { get; set; }
+        private static readonly int SystemPageSize = Environment.SystemPageSize;
+
         /// <summary>
         /// Allocates a new <see cref="BoundedMemory{T}"/> region which is immediately preceded by
         /// or immediately followed by a poison (MEM_NOACCESS) page. If <paramref name="placement"/>
@@ -82,6 +85,10 @@ private static BoundedMemory<T> AllocateWithoutDataPopulation<T>(int elementCoun
             {
                 return AllocateWithoutDataPopulationWindows<T>(elementCount, placement);
             }
+            else if (OperatingSystem.IsBrowser())
+            {
+                return AllocateWithoutDataPopulationDefault<T>(elementCount, placement);
+            }
             else
             {
                 return AllocateWithoutDataPopulationUnix<T>(elementCount, placement);