[clang] [X86][Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - Allow SSE/AVX FP MAX/MIN intrinsics to be used in constexpr (PR #171966)

[Timm Baeder via cfe-commits]

================
@@ -5575,6 +5615,62 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, 
const CallExpr *Call,
         });
   }
 
+  case clang::X86::BI__builtin_ia32_minps:
+  case clang::X86::BI__builtin_ia32_minpd:
+  case clang::X86::BI__builtin_ia32_minps256:
+  case clang::X86::BI__builtin_ia32_minpd256:
+  case clang::X86::BI__builtin_ia32_minph128:
+  case clang::X86::BI__builtin_ia32_minph256:
+    return interp__builtin_elementwise_fp_binop(
+        S, OpPC, Call,
+        [](const APFloat &A, const APFloat &B, std::optional<APSInt>) {
+          if (A.isZero() && B.isZero())
+            return B;
+          else
+            return llvm::minimum(A, B);
+        });
+
+  case clang::X86::BI__builtin_ia32_maxps:
+  case clang::X86::BI__builtin_ia32_maxpd:
+  case clang::X86::BI__builtin_ia32_maxps256:
+  case clang::X86::BI__builtin_ia32_maxpd256:
+  case clang::X86::BI__builtin_ia32_maxph128:
+  case clang::X86::BI__builtin_ia32_maxph256:
+    return interp__builtin_elementwise_fp_binop(
+        S, OpPC, Call,
+        [](const APFloat &A, const APFloat &B, std::optional<APSInt>) {
+          if (A.isZero() && B.isZero())
+            return B;
+          else
+            return llvm::maximum(A, B);
+        });
+
+  case clang::X86::BI__builtin_ia32_minps512:
+  case clang::X86::BI__builtin_ia32_minpd512:
+  case clang::X86::BI__builtin_ia32_minph512: {
+    return interp__builtin_elementwise_fp_binop(
----------------
tbaederr wrote:

How is this different from the `clang::X86::BI__builtin_ia32_minps` block above?

https://github.com/llvm/llvm-project/pull/171966
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits