json_view: keep GCC from cloning scan_string_vector

With the builder of accept() in the same translation unit, GCC 16's
interprocedural constant propagation cloned scan_string_vector without its
plain argument (always string_plain()). The parse loop of
json_document::parse() then called the clone, and GCC generated different
code for the whole loop: 0.7-1.2 % more instructions (twitter, citm_catalog,
instruments, tree-pretty, canada) and up to +13 % time on citm_catalog.
NLOHMANN_VIEW_NOCLONE (GCC only) keeps the function as written; the
instruction counts of parse() are then those of develop again.

Signed-off-by: Niels Lohmann <mail@nlohmann.me>

include/nlohmann/detail/view/macro_scope.hpp

@@ -40,6 +40,15 @@
     #define NLOHMANN_VIEW_NOINLINE
 #endif
 
+// GCC: a function compiled as written, without the clones in which interprocedural
+// constant propagation drops an argument that every caller passes as the same constant
+// (calling such a clone changes the code GCC generates for the parse loop that calls it)
+#if defined(__GNUC__) && !defined(__clang__)
+    #define NLOHMANN_VIEW_NOCLONE __attribute__((noclone))
+#else
+    #define NLOHMANN_VIEW_NOCLONE
+#endif
+
 #if defined(__GNUC__) || defined(__clang__)
     #define NLOHMANN_VIEW_NODISCARD __attribute__((warn_unused_result))
 #elif defined(_MSC_VER)

include/nlohmann/detail/view/macro_unscope.hpp

@@ -15,6 +15,7 @@
 #undef NLOHMANN_VIEW_UNLIKELY
 #undef NLOHMANN_VIEW_ALWAYS_INLINE
 #undef NLOHMANN_VIEW_NOINLINE
+#undef NLOHMANN_VIEW_NOCLONE
 #undef NLOHMANN_VIEW_NODISCARD
 #undef NLOHMANN_VIEW_THROW
 #undef NLOHMANN_VIEW_LITTLE_ENDIAN

include/nlohmann/detail/view/simd.hpp

@@ -289,9 +289,12 @@ compares, and the UTF-8 check covers the bytes up to it. Returns where the
 string scan stops, like scan_string_run: before ill-formed UTF-8 and for the
 last bytes of the input, the bytes are checked one sequence at a time. Out of
 line, so that no constants of the check occupy registers in the parse loop.
-On x86-64, it is compiled for SSSE3 (see cpu_has_ssse3()).
+On x86-64, it is compiled for SSSE3 (see cpu_has_ssse3()). Not cloned: with
+the calls of accept()'s builder, GCC 16 would clone it without plain (always
+string_plain()), and calling the clone made json_document::parse() run 0.7-1.2 %
+more instructions.
 */
-NLOHMANN_VIEW_SSSE3_TARGET NLOHMANN_VIEW_NOINLINE inline const unsigned char* scan_string_vector(const unsigned char* p, const unsigned char* e, const std::uint8_t* plain) noexcept
+NLOHMANN_VIEW_SSSE3_TARGET NLOHMANN_VIEW_NOINLINE NLOHMANN_VIEW_NOCLONE inline const unsigned char* scan_string_vector(const unsigned char* p, const unsigned char* e, const std::uint8_t* plain) noexcept
 {
     using lookup = utf8_lookup4<>;
     const unsigned char* block = p;

single_include/nlohmann/json_view.hpp

@@ -157,6 +157,15 @@
     #define NLOHMANN_VIEW_NOINLINE
 #endif
 
+// GCC: a function compiled as written, without the clones in which interprocedural
+// constant propagation drops an argument that every caller passes as the same constant
+// (calling such a clone changes the code GCC generates for the parse loop that calls it)
+#if defined(__GNUC__) && !defined(__clang__)
+    #define NLOHMANN_VIEW_NOCLONE __attribute__((noclone))
+#else
+    #define NLOHMANN_VIEW_NOCLONE
+#endif
+
 #if defined(__GNUC__) || defined(__clang__)
     #define NLOHMANN_VIEW_NODISCARD __attribute__((warn_unused_result))
 #elif defined(_MSC_VER)
@@ -948,9 +957,12 @@ compares, and the UTF-8 check covers the bytes up to it. Returns where the
 string scan stops, like scan_string_run: before ill-formed UTF-8 and for the
 last bytes of the input, the bytes are checked one sequence at a time. Out of
 line, so that no constants of the check occupy registers in the parse loop.
-On x86-64, it is compiled for SSSE3 (see cpu_has_ssse3()).
+On x86-64, it is compiled for SSSE3 (see cpu_has_ssse3()). Not cloned: with
+the calls of accept()'s builder, GCC 16 would clone it without plain (always
+string_plain()), and calling the clone made json_document::parse() run 0.7-1.2 %
+more instructions.
 */
-NLOHMANN_VIEW_SSSE3_TARGET NLOHMANN_VIEW_NOINLINE inline const unsigned char* scan_string_vector(const unsigned char* p, const unsigned char* e, const std::uint8_t* plain) noexcept
+NLOHMANN_VIEW_SSSE3_TARGET NLOHMANN_VIEW_NOINLINE NLOHMANN_VIEW_NOCLONE inline const unsigned char* scan_string_vector(const unsigned char* p, const unsigned char* e, const std::uint8_t* plain) noexcept
 {
     using lookup = utf8_lookup4<>;
     const unsigned char* block = p;
@@ -10927,6 +10939,7 @@ class tuple_element<N, ::nlohmann::detail::view::view_item<View>> // NOLINT(cert
 #undef NLOHMANN_VIEW_UNLIKELY
 #undef NLOHMANN_VIEW_ALWAYS_INLINE
 #undef NLOHMANN_VIEW_NOINLINE
+#undef NLOHMANN_VIEW_NOCLONE
 #undef NLOHMANN_VIEW_NODISCARD
 #undef NLOHMANN_VIEW_THROW
 #undef NLOHMANN_VIEW_LITTLE_ENDIAN

tests/src/unit-json_view_macros.cpp

@@ -25,7 +25,7 @@
 #endif
 
 // the view's own macros do not leak
-#if defined(NLOHMANN_VIEW_LIKELY) || defined(NLOHMANN_VIEW_UNLIKELY) || defined(NLOHMANN_VIEW_ALWAYS_INLINE) || defined(NLOHMANN_VIEW_NOINLINE) \
+#if defined(NLOHMANN_VIEW_LIKELY) || defined(NLOHMANN_VIEW_UNLIKELY) || defined(NLOHMANN_VIEW_ALWAYS_INLINE) || defined(NLOHMANN_VIEW_NOINLINE) || defined(NLOHMANN_VIEW_NOCLONE) \
     || defined(NLOHMANN_VIEW_NODISCARD) || defined(NLOHMANN_VIEW_THROW) || defined(NLOHMANN_VIEW_HAS_CPP_17) || defined(NLOHMANN_VIEW_LITTLE_ENDIAN) \
     || defined(NLOHMANN_VIEW_REPEAT16)
     #error "json_view.hpp leaks a macro"