================
@@ -4119,6 +4119,230 @@ static bool interp__builtin_ia32_gfni_mul(InterpState
&S, CodePtr OpPC,
return true;
}
+static bool interp__builtin_x86_comi(InterpState &S, CodePtr OpPC,
+ const InterpFrame *Frame,
+ const CallExpr *Call, unsigned ID) {
+ uint32_t Predicate;
+ switch (ID) {
+ case X86::BI__builtin_ia32_comieq:
+ case X86::BI__builtin_ia32_ucomieq:
+ case X86::BI__builtin_ia32_comisdeq:
+ case X86::BI__builtin_ia32_ucomisdeq:
+ Predicate = X86CmpImm::CMP_EQ_OQ;
+ break;
+ case X86::BI__builtin_ia32_comilt:
+ case X86::BI__builtin_ia32_ucomilt:
+ case X86::BI__builtin_ia32_comisdlt:
+ case X86::BI__builtin_ia32_ucomisdlt:
+ Predicate = X86CmpImm::CMP_LT_OQ;
+ break;
+ case X86::BI__builtin_ia32_comile:
+ case X86::BI__builtin_ia32_ucomile:
+ case X86::BI__builtin_ia32_comisdle:
+ case X86::BI__builtin_ia32_ucomisdle:
+ Predicate = X86CmpImm::CMP_LE_OQ;
+ break;
+ case X86::BI__builtin_ia32_comigt:
+ case X86::BI__builtin_ia32_ucomigt:
+ case X86::BI__builtin_ia32_comisdgt:
+ case X86::BI__builtin_ia32_ucomisdgt:
+ Predicate = X86CmpImm::CMP_GT_OQ;
+ break;
+ case X86::BI__builtin_ia32_comige:
+ case X86::BI__builtin_ia32_ucomige:
+ case X86::BI__builtin_ia32_comisdge:
+ case X86::BI__builtin_ia32_ucomisdge:
+ Predicate = X86CmpImm::CMP_GE_OQ;
+ break;
+ case X86::BI__builtin_ia32_comineq:
+ case X86::BI__builtin_ia32_ucomineq:
+ case X86::BI__builtin_ia32_comisdneq:
+ case X86::BI__builtin_ia32_ucomisdneq:
+ Predicate = X86CmpImm::CMP_NEQ_UQ;
+ break;
+ case X86::BI__builtin_ia32_vcomish: {
+ discard(S.Stk, *S.getContext().classify(Call->getArg(3)));
+ const APSInt Imm = popToAPSInt(S, Call->getArg(2));
+ Predicate = Imm.getZExtValue();
+ break;
+ }
+ default:
+ llvm_unreachable("unhandled x86 comi builtin");
+ }
+
+ const Pointer &VectorB = S.Stk.pop<Pointer>();
+ const Pointer &VectorA = S.Stk.pop<Pointer>();
+
+ if (VectorA.getNumElems() == 0 ||
+ VectorA.getNumElems() != VectorB.getNumElems())
+ return false;
+
+ const llvm::APFloat A = VectorA.elem<Floating>(0).getAPFloat();
+ const llvm::APFloat B = VectorB.elem<Floating>(0).getAPFloat();
+ const bool Matches = MatchesPredicate(Predicate, A.compare(B));
+ pushInteger(S, Matches, Call->getType());
+ return true;
+}
+
+static bool interp__builtin_x86_cmp(InterpState &S, CodePtr OpPC,
+ const InterpFrame *Frame,
+ const CallExpr *Call, unsigned ID) {
+ const bool HasImmArg =
+ ID == X86::BI__builtin_ia32_cmpps || ID == X86::BI__builtin_ia32_cmppd ||
+ ID == X86::BI__builtin_ia32_cmpps256 ||
+ ID == X86::BI__builtin_ia32_cmppd256 ||
+ ID == X86::BI__builtin_ia32_cmpss || ID == X86::BI__builtin_ia32_cmpsd;
+
+ uint32_t Predicate;
+ if (HasImmArg) {
+ const APSInt ImmAPS = popToAPSInt(S, Call->getArg(2));
+ Predicate = ImmAPS.getZExtValue();
+ } else {
+ switch (ID) {
+ case X86::BI__builtin_ia32_cmpeqss:
+ case X86::BI__builtin_ia32_cmpeqsd:
+ case X86::BI__builtin_ia32_cmpeqps:
+ case X86::BI__builtin_ia32_cmpeqpd:
+ Predicate = X86CmpImm::CMP_EQ_OQ;
+ break;
+ case X86::BI__builtin_ia32_cmpgess:
+ case X86::BI__builtin_ia32_cmpgesd:
+ case X86::BI__builtin_ia32_cmpgeps:
+ case X86::BI__builtin_ia32_cmpgepd:
+ Predicate = X86CmpImm::CMP_GE_OS;
+ break;
+ case X86::BI__builtin_ia32_cmpgtss:
+ case X86::BI__builtin_ia32_cmpgtsd:
+ case X86::BI__builtin_ia32_cmpgtps:
+ case X86::BI__builtin_ia32_cmpgtpd:
+ Predicate = X86CmpImm::CMP_GT_OS;
+ break;
+ case X86::BI__builtin_ia32_cmpltss:
+ case X86::BI__builtin_ia32_cmpltsd:
+ case X86::BI__builtin_ia32_cmpltps:
+ case X86::BI__builtin_ia32_cmpltpd:
+ Predicate = X86CmpImm::CMP_LT_OS;
+ break;
+ case X86::BI__builtin_ia32_cmpless:
+ case X86::BI__builtin_ia32_cmplesd:
+ case X86::BI__builtin_ia32_cmpleps:
+ case X86::BI__builtin_ia32_cmplepd:
+ Predicate = X86CmpImm::CMP_LE_OS;
+ break;
+ case X86::BI__builtin_ia32_cmpneqss:
+ case X86::BI__builtin_ia32_cmpneqsd:
+ case X86::BI__builtin_ia32_cmpneqps:
+ case X86::BI__builtin_ia32_cmpneqpd:
+ Predicate = X86CmpImm::CMP_NEQ_UQ;
+ break;
+ case X86::BI__builtin_ia32_cmpngess:
+ case X86::BI__builtin_ia32_cmpngesd:
+ case X86::BI__builtin_ia32_cmpngeps:
+ case X86::BI__builtin_ia32_cmpngepd:
+ Predicate = X86CmpImm::CMP_NGE_US;
+ break;
+ case X86::BI__builtin_ia32_cmpngtss:
+ case X86::BI__builtin_ia32_cmpngtsd:
+ case X86::BI__builtin_ia32_cmpngtps:
+ case X86::BI__builtin_ia32_cmpngtpd:
+ Predicate = X86CmpImm::CMP_NGT_US;
+ break;
+ case X86::BI__builtin_ia32_cmpnless:
+ case X86::BI__builtin_ia32_cmpnlesd:
+ case X86::BI__builtin_ia32_cmpnleps:
+ case X86::BI__builtin_ia32_cmpnlepd:
+ Predicate = X86CmpImm::CMP_NLE_US;
+ break;
+ case X86::BI__builtin_ia32_cmpnltss:
+ case X86::BI__builtin_ia32_cmpnltsd:
+ case X86::BI__builtin_ia32_cmpnltps:
+ case X86::BI__builtin_ia32_cmpnltpd:
+ Predicate = X86CmpImm::CMP_NLT_US;
+ break;
+ case X86::BI__builtin_ia32_cmpordss:
+ case X86::BI__builtin_ia32_cmpordsd:
+ case X86::BI__builtin_ia32_cmpordps:
+ case X86::BI__builtin_ia32_cmpordpd:
+ Predicate = X86CmpImm::CMP_ORD_Q;
+ break;
+ case X86::BI__builtin_ia32_cmpunordss:
+ case X86::BI__builtin_ia32_cmpunordsd:
+ case X86::BI__builtin_ia32_cmpunordps:
+ case X86::BI__builtin_ia32_cmpunordpd:
+ Predicate = X86CmpImm::CMP_UNORD_Q;
+ break;
+ default:
+ llvm_unreachable("unhandled x86 cmp builtin");
+ }
+ }
+
+ const Pointer &VectorB = S.Stk.pop<Pointer>();
+ const Pointer &VectorA = S.Stk.pop<Pointer>();
+ Pointer &Dst = S.Stk.peek<Pointer>();
+
+ const bool IsScalar = ID == X86::BI__builtin_ia32_cmpss ||
+ ID == X86::BI__builtin_ia32_cmpsd ||
+ ID == X86::BI__builtin_ia32_cmpeqss ||
+ ID == X86::BI__builtin_ia32_cmpeqsd ||
+ ID == X86::BI__builtin_ia32_cmpgess ||
+ ID == X86::BI__builtin_ia32_cmpgesd ||
+ ID == X86::BI__builtin_ia32_cmpgtss ||
+ ID == X86::BI__builtin_ia32_cmpgtsd ||
+ ID == X86::BI__builtin_ia32_cmpltss ||
+ ID == X86::BI__builtin_ia32_cmpltsd ||
+ ID == X86::BI__builtin_ia32_cmpless ||
+ ID == X86::BI__builtin_ia32_cmplesd ||
+ ID == X86::BI__builtin_ia32_cmpneqss ||
+ ID == X86::BI__builtin_ia32_cmpneqsd ||
+ ID == X86::BI__builtin_ia32_cmpngess ||
+ ID == X86::BI__builtin_ia32_cmpngesd ||
+ ID == X86::BI__builtin_ia32_cmpngtss ||
+ ID == X86::BI__builtin_ia32_cmpngtsd ||
+ ID == X86::BI__builtin_ia32_cmpnless ||
+ ID == X86::BI__builtin_ia32_cmpnlesd ||
+ ID == X86::BI__builtin_ia32_cmpnltss ||
+ ID == X86::BI__builtin_ia32_cmpnltsd ||
+ ID == X86::BI__builtin_ia32_cmpordss ||
+ ID == X86::BI__builtin_ia32_cmpordsd ||
+ ID == X86::BI__builtin_ia32_cmpunordss ||
+ ID == X86::BI__builtin_ia32_cmpunordsd;
+
+ const auto NumLanes = VectorA.getNumElems();
+ if (NumLanes != VectorB.getNumElems())
+ return false;
+
+ for (unsigned int i = 0; i < NumLanes; ++i) {
+ // Handle cmpss/cmpsd
+ if (IsScalar && i > 0) {
+ // Copy the upper 3 packed elements from a to the upper elements of dst
+ Dst.elem<Floating>(i) = VectorA.elem<Floating>(i);
+ continue;
+ }
+
+ llvm::APFloat AElement = VectorA.elem<Floating>(i).getAPFloat();
+ llvm::APFloat BElement = VectorB.elem<Floating>(i).getAPFloat();
+
+ auto CompareResult = AElement.compare(BElement);
+ const bool Matches = MatchesPredicate(Predicate, CompareResult);
+
+ // Create bit patterns for comparison results:
+ // True = all bits set (0xFFFFFFFF for float, 0xFFFFFFFFFFFFFFFF for
double)
+ // False = all bits zero
+ const llvm::fltSemantics &Sem = AElement.getSemantics();
+ const unsigned BitWidth = llvm::APFloat::getSizeInBits(Sem);
+ const llvm::APFloat True(Sem, llvm::APInt::getAllOnes(BitWidth));
+ const llvm::APFloat False(Sem, llvm::APInt(BitWidth, 0));
----------------
tbaederr wrote:
Only create these when needed, i.e. based on the value of `Matched`.
https://github.com/llvm/llvm-project/pull/182589
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
