Enable AVX512 embedded masking for most other intrinsics (dotnet#101886)

* Remove HW_Flag_MultiIns in favor of using HW_Flag_SpecialCodeGen

* Add a new flag HW_Flag_InvalidNodeId

* Change HW_Flag_EmbMaskingIncompatible to be HW_Flag_EmbMaskingCompatible

* Mark various compare intrinsics with HW_Flag_NoEvexSemantics

* Marking various intrinsics as EmbBroadcastCompatible, EmbMaskingCompatible, or Commutative

* Applying formatting patch

* Ensure WithLower/WithUpper are not marked as InvalidNodeId

* Ensure that instOptions are being passed down all relevant hwintrinsic code paths

* Ensure the insOpts are plumbed through for EVEX instructions

* Ensure EVEX instructions are properly annotated with EmbeddedBroadcastSupported

* Ensure that embedded broadcast/masking is displayed in the disassembly

* Applying formatting patch

* Updating the hwintrinsic tests to cover embedded broadcast/masking

* Fix some handling in the JIT related to embedded broadcast/masking

* Fixup some tests where validating embedded masking is non-trivial

* Cleanup some cases found by SPMI

* Ensure that CompareLessThan has its operands swapped back if its being converted to the AVX512 form

* Don't regress a scenario around op_Equality and TYP_MASK

* Adjusting hardware intrinsic tests to test non-zero masks

* Avoid some messiness around operand swapping

* Ensure embedded masks mark TYP_SIMD16 and TYP_SIMD32 instructions as needing EVEX

* Mark Sse2_r/Sse2_ro as AotIncompatible due to runtime/102037

src/coreclr/jit/codegen.h

@@ -959,38 +959,35 @@ class CodeGen final : public CodeGenInterface
 #ifdef FEATURE_HW_INTRINSICS
     void genHWIntrinsic(GenTreeHWIntrinsic* node);
 #if defined(TARGET_XARCH)
-    void genHWIntrinsic_R_RM(GenTreeHWIntrinsic* node,
-                             instruction         ins,
-                             emitAttr            attr,
-                             regNumber           reg,
-                             GenTree*            rmOp,
-                             insOpts             instOptions = INS_OPTS_NONE);
-    void genHWIntrinsic_R_RM_I(GenTreeHWIntrinsic* node, instruction ins, emitAttr attr, int8_t ival);
+    void genHWIntrinsic_R_RM(
+        GenTreeHWIntrinsic* node, instruction ins, emitAttr attr, regNumber reg, GenTree* rmOp, insOpts instOptions);
+    void genHWIntrinsic_R_RM_I(
+        GenTreeHWIntrinsic* node, instruction ins, emitAttr attr, int8_t ival, insOpts instOptions);
     void genHWIntrinsic_R_R_RM(GenTreeHWIntrinsic* node, instruction ins, emitAttr attr, insOpts instOptions);
-    void genHWIntrinsic_R_R_RM_I(GenTreeHWIntrinsic* node, instruction ins, emitAttr attr, int8_t ival);
-    void genHWIntrinsic_R_R_RM_R(GenTreeHWIntrinsic* node, instruction ins, emitAttr attr);
+    void genHWIntrinsic_R_R_RM_I(
+        GenTreeHWIntrinsic* node, instruction ins, emitAttr attr, int8_t ival, insOpts instOptions);
+    void genHWIntrinsic_R_R_RM_R(GenTreeHWIntrinsic* node, instruction ins, emitAttr attr, insOpts instOptions);
     void genHWIntrinsic_R_R_R_RM(instruction ins,
                                  emitAttr    attr,
                                  regNumber   targetReg,
                                  regNumber   op1Reg,
                                  regNumber   op2Reg,
                                  GenTree*    op3,
-                                 insOpts     instOptions = INS_OPTS_NONE);
-    void genHWIntrinsic_R_R_R_RM_I(GenTreeHWIntrinsic* node, instruction ins, emitAttr attr, int8_t ival);
+                                 insOpts     instOptions);
+    void genHWIntrinsic_R_R_R_RM_I(
+        GenTreeHWIntrinsic* node, instruction ins, emitAttr attr, int8_t ival, insOpts instOptions);
 
-    void genBaseIntrinsic(GenTreeHWIntrinsic* node);
-    void genX86BaseIntrinsic(GenTreeHWIntrinsic* node);
+    void genBaseIntrinsic(GenTreeHWIntrinsic* node, insOpts instOptions);
+    void genX86BaseIntrinsic(GenTreeHWIntrinsic* node, insOpts instOptions);
     void genSSEIntrinsic(GenTreeHWIntrinsic* node, insOpts instOptions);
     void genSSE2Intrinsic(GenTreeHWIntrinsic* node, insOpts instOptions);
-    void genSSE41Intrinsic(GenTreeHWIntrinsic* node);
-    void genSSE42Intrinsic(GenTreeHWIntrinsic* node);
+    void genSSE41Intrinsic(GenTreeHWIntrinsic* node, insOpts instOptions);
+    void genSSE42Intrinsic(GenTreeHWIntrinsic* node, insOpts instOptions);
     void genAvxFamilyIntrinsic(GenTreeHWIntrinsic* node, insOpts instOptions);
-    void genAESIntrinsic(GenTreeHWIntrinsic* node);
     void genBMI1OrBMI2Intrinsic(GenTreeHWIntrinsic* node, insOpts instOptions);
     void genFMAIntrinsic(GenTreeHWIntrinsic* node, insOpts instOptions);
-    void genPermuteVar2x(GenTreeHWIntrinsic* node);
+    void genPermuteVar2x(GenTreeHWIntrinsic* node, insOpts instOptions);
     void genLZCNTIntrinsic(GenTreeHWIntrinsic* node);
-    void genPCLMULQDQIntrinsic(GenTreeHWIntrinsic* node);
     void genPOPCNTIntrinsic(GenTreeHWIntrinsic* node);
     void genXCNTIntrinsic(GenTreeHWIntrinsic* node, instruction ins);
     void genX86SerializeIntrinsic(GenTreeHWIntrinsic* node);
@@ -1003,6 +1000,8 @@ class CodeGen final : public CodeGenInterface
                                          HWIntrinsicSwitchCaseBody emitSwCase);
 
     void genNonTableDrivenHWIntrinsicsJumpTableFallback(GenTreeHWIntrinsic* node, GenTree* lastOp);
+
+    static insOpts AddEmbBroadcastMode(insOpts instOptions);
 #endif // defined(TARGET_XARCH)
 
 #ifdef TARGET_ARM64
@@ -1576,16 +1575,22 @@ class CodeGen final : public CodeGenInterface
     void inst_TT(instruction ins, emitAttr size, GenTree* op1);
     void inst_RV_TT(instruction ins, emitAttr size, regNumber op1Reg, GenTree* op2);
     void inst_RV_RV_IV(instruction ins, emitAttr size, regNumber reg1, regNumber reg2, unsigned ival);
-    void inst_RV_TT_IV(instruction ins, emitAttr attr, regNumber reg1, GenTree* rmOp, int ival);
+    void inst_RV_TT_IV(instruction ins, emitAttr attr, regNumber reg1, GenTree* rmOp, int ival, insOpts instOptions);
     void inst_RV_RV_TT(instruction ins,
                        emitAttr    size,
                        regNumber   targetReg,
                        regNumber   op1Reg,
                        GenTree*    op2,
                        bool        isRMW,
                        insOpts     instOptions);
-    void inst_RV_RV_TT_IV(
-        instruction ins, emitAttr size, regNumber targetReg, regNumber op1Reg, GenTree* op2, int8_t ival, bool isRMW);
+    void inst_RV_RV_TT_IV(instruction ins,
+                          emitAttr    size,
+                          regNumber   targetReg,
+                          regNumber   op1Reg,
+                          GenTree*    op2,
+                          int8_t      ival,
+                          bool        isRMW,
+                          insOpts     instOptions);
 #endif
 
     void inst_set_SV_var(GenTree* tree);

src/coreclr/jit/codegencommon.cpp

@@ -3856,7 +3856,7 @@ void CodeGen::genZeroInitFltRegs(const regMaskTP& initFltRegs, const regMaskTP&
                 }
 #elif defined(TARGET_XARCH)
                 // XORPS is the fastest and smallest way to initialize a XMM register to zero.
-                GetEmitter()->emitIns_SIMD_R_R_R(INS_xorps, EA_16BYTE, reg, reg, reg);
+                GetEmitter()->emitIns_SIMD_R_R_R(INS_xorps, EA_16BYTE, reg, reg, reg, INS_OPTS_NONE);
                 dblInitReg = reg;
 #elif defined(TARGET_ARM64)
                 // We will just zero out the entire vector register. This sets it to a double/float zero value
@@ -3896,7 +3896,7 @@ void CodeGen::genZeroInitFltRegs(const regMaskTP& initFltRegs, const regMaskTP&
                 }
 #elif defined(TARGET_XARCH)
                 // XORPS is the fastest and smallest way to initialize a XMM register to zero.
-                GetEmitter()->emitIns_SIMD_R_R_R(INS_xorps, EA_16BYTE, reg, reg, reg);
+                GetEmitter()->emitIns_SIMD_R_R_R(INS_xorps, EA_16BYTE, reg, reg, reg, INS_OPTS_NONE);
                 fltInitReg = reg;
 #elif defined(TARGET_ARM64)
                 // We will just zero out the entire vector register. This sets it to a double/float zero value