Allow use of qnumeric.h without needing Windows intrinsics

Add a preprocessor variable Q_NUMERIC_NO_INTRINSICS to suppress
inclusion of <intrin.h> and fall back to the same behavior as other
platforms

Fixes: QTBUG-131757
Pick-to: 6.8
Change-Id: I8bcf912c0499897a8daf67951993cee22a86c590
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This commit is contained in:
David C. Partridge 2024-12-03 12:49:03 +00:00 committed by David Partridge
parent d89cef439f
commit 16c5398aa1

View File

@ -23,9 +23,17 @@
# undef max
#endif
#if defined(Q_CC_MSVC)
//
// SIMDe (SIMD Everywhere) can't be used if intrin.h has been included as many definitions
// conflict. Defining Q_NUMERIC_NO_INTRINSICS allows SIMDe users to use Qt, at the cost of
// falling back to the prior implementations of qMulOverflow and qAddOverflow.
//
#if defined(Q_CC_MSVC) && !defined(Q_NUMERIC_NO_INTRINSICS)
# include <intrin.h>
# include <float.h>
# if defined(Q_PROCESSOR_X86) || defined(Q_PROCESSOR_X86_64)
# define Q_HAVE_ADDCARRY
# endif
# if defined(Q_PROCESSOR_X86_64) || defined(Q_PROCESSOR_ARM_64)
# define Q_INTRINSIC_MUL_OVERFLOW64
# define Q_UMULH(v1, v2) __umulh(v1, v2);
@ -223,7 +231,7 @@ template <> inline bool qMulOverflow(int64_t v1, int64_t v2, int64_t *r)
# endif // OS_INTEGRITY ARM64
# endif // Q_INTRINSIC_MUL_OVERFLOW64
# if defined(Q_CC_MSVC) && defined(Q_PROCESSOR_X86)
# if defined(Q_HAVE_ADDCARRY) && defined(Q_PROCESSOR_X86)
// We can use intrinsics for the unsigned operations with MSVC
template <> inline bool qAddOverflow(unsigned v1, unsigned v2, unsigned *r)
{ return _addcarry_u32(0, v1, v2, r); }
@ -242,7 +250,8 @@ template <> inline bool qAddOverflow(quint64 v1, quint64 v2, quint64 *r)
return carry;
# endif // !x86-64
}
# endif // MSVC X86
# endif // HAVE ADDCARRY
#undef Q_HAVE_ADDCARRY
#endif // !GCC
// Implementations for addition, subtraction or multiplication by a