dhewm/dhewm3-sdk
0
0
Fork 0
mirror of https://github.com/dhewm/dhewm3-sdk.git synced 2024-11-24 13:31:18 +00:00
Code Issues Projects Releases Packages Wiki Activity

ID_MAYBE_INLINE for not-forced inlining

Browse source 
On Windows, ID_INLINE does __forceinline, which is bad if the function
calls alloca() and is called in a loop..
So use just __inline there so the compiler can choose not to inline
(if called in a loop).
This didn't cause actual stack overflows as far as I know, but it could
(and MSVC warns about it).

(This includes "Fix ID_MAYBE_INLINE on non-Windows platforms")
This commit is contained in:
Daniel Gibson 2020-09-06 04:36:40 +02:00
parent a0065803ff
commit 541ba91e61
4 changed files with 11 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
idlib/math/Curve.h
View file

@ -204,7 +204,7 @@ idCurve::RombergIntegral
==================== ====================
*/ */
template< class type > template< class type >
ID_INLINE float idCurve<type>::RombergIntegral( const float t0, const float t1, const int order ) const { ID_MAYBE_INLINE float idCurve<type>::RombergIntegral(const float t0, const float t1, const int order) const {
int i, j, k, m, n; int i, j, k, m, n;
float sum, delta; float sum, delta;
float *temp[2]; float *temp[2];
@ -242,7 +242,7 @@ idCurve::GetLengthBetweenKnots
==================== ====================
*/ */
template< class type > template< class type >
ID_INLINE float idCurve<type>::GetLengthBetweenKnots( const int i0, const int i1 ) const { ID_MAYBE_INLINE float idCurve<type>::GetLengthBetweenKnots(const int i0, const int i1) const {
float length = 0.0f; float length = 0.0f;
for ( int i = i0; i < i1; i++ ) { for ( int i = i0; i < i1; i++ ) {
length += RombergIntegral( times[i], times[i+1], 5 ); length += RombergIntegral( times[i], times[i+1], 5 );

2
idlib/math/Lcp.cpp
View file

@ -340,7 +340,7 @@ idLCP_Square::CalcForceDelta
modifies this->delta_f modifies this->delta_f
============ ============
*/ */
ID_INLINE void idLCP_Square::CalcForceDelta( int d, float dir ) { ID_MAYBE_INLINE void idLCP_Square::CalcForceDelta(int d, float dir) {
int i; int i;
float *ptr; float *ptr;

2
idlib/math/Matrix.h
View file

@ -2390,7 +2390,7 @@ ID_INLINE void idMatX::Clamp( float min, float max ) {
} }
} }
ID_INLINE idMatX &idMatX::SwapRows( int r1, int r2 ) { ID_MAYBE_INLINE idMatX &idMatX::SwapRows(int r1, int r2) {
float *ptr; float *ptr;
ptr = (float *) _alloca16( numColumns * sizeof( float ) ); ptr = (float *) _alloca16( numColumns * sizeof( float ) );

7
sys/platform.h
View file

@ -85,6 +85,9 @@ If you have questions concerning this license or the applicable additional terms
#define ALIGN16( x ) __declspec(align(16)) x #define ALIGN16( x ) __declspec(align(16)) x
#define PACKED #define PACKED
#define ID_INLINE __forceinline #define ID_INLINE __forceinline
// DG: alternative to forced inlining of ID_INLINE for functions that do alloca()
// and are called in a loop so inlining them might cause stack overflow
#define ID_MAYBE_INLINE __inline
#define ID_STATIC_TEMPLATE static #define ID_STATIC_TEMPLATE static
#define assertmem( x, y ) assert( _CrtIsValidPointer( x, y, true ) ) #define assertmem( x, y ) assert( _CrtIsValidPointer( x, y, true ) )
#else #else
@ -161,6 +164,10 @@ If you have questions concerning this license or the applicable additional terms
#endif #endif
#ifndef ID_MAYBE_INLINE
// for MSVC it's __inline, otherwise just inline should work
#define ID_MAYBE_INLINE inline
#endif // ID_MAYBE_INLINE
#ifdef __GNUC__ #ifdef __GNUC__
#define id_attribute(x) __attribute__(x) #define id_attribute(x) __attribute__(x)