qf/quakeforge
0
0
Fork 0
mirror of https://git.code.sf.net/p/quake/quakeforge synced 2025-05-31 16:51:08 +00:00
Code Issues Projects Releases Packages Wiki Activity

don't let gcc automaticly inline functions but make gcc inline bigger

Browse source 
functions when told to. also make gcc warn if it can't inline a function.

Explicitly inline several functions (including moving VectorNormalize to
mathlib.h so it /can/ be) resulting in a 5.5% speedup for spam2 (88 to 92
fps)
This commit is contained in:
Bill Currie 2003-08-11 06:05:07 +00:00
parent b88d275994
commit 7ead5a91f8
12 changed files with 62 additions and 61 deletions
Download patch file Download diff file Expand all files Collapse all files

4
configure.ac
View file

@ -1312,7 +1312,7 @@ if test "x$optimize" = xyes; then
CC_SUB=$3 CC_SUB=$3
IFS=" " IFS=" "
AC_MSG_RESULT($CCVER) AC_MSG_RESULT($CCVER)
heavy="-O3 -ffast-math -funroll-loops -fomit-frame-pointer -fexpensive-optimizations" heavy="-O2 -frename-registers -finline-limit=32000 -ffast-math -funroll-loops -fomit-frame-pointer -fexpensive-optimizations"
light="-O2" light="-O2"
AC_ARG_ENABLE(strict-aliasing, AC_ARG_ENABLE(strict-aliasing,
[ --enable-strict-aliasing enable the -fstrict-aliasing ooption of gcc]) [ --enable-strict-aliasing enable the -fstrict-aliasing ooption of gcc])
@ -1472,7 +1472,7 @@ AC_ARG_ENABLE(Werror,
dnl We want warnings, lots of warnings... dnl We want warnings, lots of warnings...
if test "x$GCC" = "xyes"; then if test "x$GCC" = "xyes"; then
if test "x$enable_Werror" $cvs_def_enabled; then if test "x$enable_Werror" $cvs_def_enabled; then
CFLAGS="$CFLAGS -Wall -Werror -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -Wmissing-declarations" CFLAGS="$CFLAGS -Wall -Werror -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -Wmissing-declarations -Winline"
else else
CFLAGS="$CFLAGS -Wall" CFLAGS="$CFLAGS -Wall"
fi fi

24
include/QF/mathlib.h
View file

@ -95,7 +95,6 @@ void _VectorMA (const vec3_t veca, float scale, const vec3_t vecb,
void _VectorScale (const vec3_t in, vec_t scale, vec3_t out); void _VectorScale (const vec3_t in, vec_t scale, vec3_t out);
void _VectorSubtract (const vec3_t veca, const vec3_t vecb, vec3_t out); void _VectorSubtract (const vec3_t veca, const vec3_t vecb, vec3_t out);
void CrossProduct (const vec3_t v1, const vec3_t v2, vec3_t cross); void CrossProduct (const vec3_t v1, const vec3_t v2, vec3_t cross);
float VectorNormalize (vec3_t v); // returns vector length
vec_t _VectorNormalize (vec3_t v); // returns vector length vec_t _VectorNormalize (vec3_t v); // returns vector length
void VectorInverse (vec3_t v); void VectorInverse (vec3_t v);
int Q_log2(int val); int Q_log2(int val);
@ -141,6 +140,7 @@ void RotatePointAroundVector (vec3_t dst, const vec3_t axis,
extern mplane_t frustum[4]; extern mplane_t frustum[4];
extern inline qboolean R_CullBox (const vec3_t mins, const vec3_t maxs); extern inline qboolean R_CullBox (const vec3_t mins, const vec3_t maxs);
extern inline qboolean R_CullSphere (const vec3_t origin, const float radius); extern inline qboolean R_CullSphere (const vec3_t origin, const float radius);
extern inline float VectorNormalize (vec3_t v); // returns vector length
#ifndef IMPLEMENT_R_Cull #ifndef IMPLEMENT_R_Cull
extern inline extern inline
#endif #endif
@ -173,4 +173,26 @@ R_CullSphere (const vec3_t origin, const float radius)
return false; return false;
} }
#ifndef IMPLEMENT_VectorNormalize
extern inline
#endif
float
VectorNormalize (vec3_t v)
{
float length;
length = DotProduct (v, v);
if (length) {
float ilength;
length = sqrt (length);
ilength = 1.0 / length;
v[0] *= ilength;
v[1] *= ilength;
v[2] *= ilength;
}
return length;
}
#endif // __mathlib_h #endif // __mathlib_h

2
libs/gamecode/engine/pr_obj.c
View file

@ -594,7 +594,7 @@ pr_class_pose_as (progs_t *pr)
PR_RunError (pr, "%s, not implemented", __FUNCTION__); PR_RunError (pr, "%s, not implemented", __FUNCTION__);
} }
static inline pr_id_t * static inline pr_id_t *
class_create_instance (progs_t *pr, pr_class_t *class) class_create_instance (progs_t *pr, pr_class_t *class)
{ {
int size = class->instance_size * sizeof (pr_type_t); int size = class->instance_size * sizeof (pr_type_t);

8
libs/models/alias/gl_mesh.c
View file

@ -76,7 +76,7 @@ int stripcount;
int strip_size; int strip_size;
static void static inline void
alloc_used (int size) alloc_used (int size)
{ {
if (size <= used_size) if (size <= used_size)
@ -88,7 +88,7 @@ alloc_used (int size)
used_size = size; used_size = size;
} }
static void static inline void
add_command (int cmd) add_command (int cmd)
{ {
if (numcommands + 1 > commands_size) { if (numcommands + 1 > commands_size) {
@ -100,7 +100,7 @@ add_command (int cmd)
commands[numcommands++] = cmd; commands[numcommands++] = cmd;
} }
static void static inline void
add_vertex (int vert) add_vertex (int vert)
{ {
if (numorder + 1 > vertexorder_size) { if (numorder + 1 > vertexorder_size) {
@ -112,7 +112,7 @@ add_vertex (int vert)
vertexorder[numorder++] = vert; vertexorder[numorder++] = vert;
} }
static void static inline void
add_strip (int vert, int tri) add_strip (int vert, int tri)
{ {
if (stripcount + 1 > strip_size) { if (stripcount + 1 > strip_size) {

6
libs/models/alias/model_alias.c
View file

@ -135,8 +135,7 @@ Mod_LoadAllSkins (int numskins, daliasskintype_t *pskintype, int *pskinindex)
void void
Mod_LoadAliasModel (model_t *mod, void *buffer, cache_allocator_t allocator) Mod_LoadAliasModel (model_t *mod, void *buffer, cache_allocator_t allocator)
{ {
byte *p; int i, j, size, version, numframes, start, end, total;
int i, j, len, size, version, numframes, start, end, total;
dtriangle_t *pintriangles; dtriangle_t *pintriangles;
daliasframetype_t *pframetype; daliasframetype_t *pframetype;
daliasskintype_t *pskintype; daliasskintype_t *pskintype;
@ -151,8 +150,7 @@ Mod_LoadAliasModel (model_t *mod, void *buffer, cache_allocator_t allocator)
extra = 1; // extra precision bytes extra = 1; // extra precision bytes
CRC_Init (&crc); CRC_Init (&crc);
for (len = qfs_filesize, p = buffer; len; len--, p++) CRC_ProcessBlock (buffer, &crc, qfs_filesize);
CRC_ProcessByte (&crc, *p);
start = Hunk_LowMark (); start = Hunk_LowMark ();

20
libs/util/mathlib.c
View file

@ -41,6 +41,7 @@ static __attribute__ ((unused)) const char rcsid[] =
#include <math.h> #include <math.h>
#define IMPLEMENT_R_Cull #define IMPLEMENT_R_Cull
#define IMPLEMENT_VectorNormalize
#include "QF/mathlib.h" #include "QF/mathlib.h"
#include "QF/qtypes.h" #include "QF/qtypes.h"
@ -439,25 +440,6 @@ _VectorLength (const vec3_t v)
return length; return length;
} }
float
VectorNormalize (vec3_t v)
{
float length;
length = DotProduct (v, v);
if (length) {
float ilength;
length = sqrt (length);
ilength = 1.0 / length;
v[0] *= ilength;
v[1] *= ilength;
v[2] *= ilength;
}
return length;
}
vec_t vec_t
_VectorNormalize (vec3_t v) _VectorNormalize (vec3_t v)
{ {

29
libs/util/zone.c
View file

@ -563,7 +563,6 @@ typedef struct cache_system_s {
cache_system_t cache_head; cache_system_t cache_head;
int cache_writelock; int cache_writelock;
static void *Cache_RealCheck (cache_user_t *c);
static cache_system_t *Cache_TryAlloc (int size, qboolean nobottom); static cache_system_t *Cache_TryAlloc (int size, qboolean nobottom);
static void Cache_RealFree (cache_user_t *c); static void Cache_RealFree (cache_user_t *c);
static void Cache_Profile (void); static void Cache_Profile (void);
@ -652,7 +651,7 @@ Cache_FreeHigh (int new_high_hunk)
#endif #endif
} }
static void static inline void
Cache_UnlinkLRU (cache_system_t * cs) Cache_UnlinkLRU (cache_system_t * cs)
{ {
if (!cs->lru_next || !cs->lru_prev) if (!cs->lru_next || !cs->lru_prev)
@ -664,7 +663,7 @@ Cache_UnlinkLRU (cache_system_t * cs)
cs->lru_prev = cs->lru_next = NULL; cs->lru_prev = cs->lru_next = NULL;
} }
static void static inline void
Cache_MakeLRU (cache_system_t * cs) Cache_MakeLRU (cache_system_t * cs)
{ {
if (cs->lru_next || cs->lru_prev) if (cs->lru_next || cs->lru_prev)
@ -882,18 +881,7 @@ Cache_RealFree (cache_user_t *c)
#endif #endif
} }
void * static inline void *
Cache_Check (cache_user_t *c)
{
void *mem;
CACHE_WRITE_LOCK;
mem = Cache_RealCheck (c);
CACHE_WRITE_UNLOCK;
return mem;
}
static void *
Cache_RealCheck (cache_user_t *c) Cache_RealCheck (cache_user_t *c)
{ {
cache_system_t *cs; cache_system_t *cs;
@ -910,6 +898,17 @@ Cache_RealCheck (cache_user_t *c)
return c->data; return c->data;
} }
void *
Cache_Check (cache_user_t *c)
{
void *mem;
CACHE_WRITE_LOCK;
mem = Cache_RealCheck (c);
CACHE_WRITE_UNLOCK;
return mem;
}
void * void *
Cache_Alloc (cache_user_t *c, int size, const char *name) Cache_Alloc (cache_user_t *c, int size, const char *name)
{ {

4
libs/video/renderer/gl/gl_lightmap.c
View file

@ -105,7 +105,7 @@ R_RecursiveLightUpdate (mnode_t *node)
surf->cached_dlight = true; surf->cached_dlight = true;
} }
static void static inline void
R_AddDynamicLights_1 (msurface_t *surf) R_AddDynamicLights_1 (msurface_t *surf)
{ {
float dist; float dist;
@ -170,7 +170,7 @@ R_AddDynamicLights_1 (msurface_t *surf)
} }
} }
static void static inline void
R_AddDynamicLights_3 (msurface_t *surf) R_AddDynamicLights_3 (msurface_t *surf)
{ {
float dist; float dist;

16
libs/video/renderer/gl/gl_mod_alias.c
View file

@ -91,7 +91,7 @@ float r_avertexnormal_dots[SHADEDOT_QUANT][256] =
vec3_t shadevector; vec3_t shadevector;
static void static inline void
GL_DrawAliasFrameTri (vert_order_t *vo) GL_DrawAliasFrameTri (vert_order_t *vo)
{ {
float color[4]; float color[4];
@ -116,7 +116,7 @@ GL_DrawAliasFrameTri (vert_order_t *vo)
qfglEnd(); qfglEnd();
} }
static void static inline void
GL_DrawAliasFrameTri_fb (vert_order_t *vo) GL_DrawAliasFrameTri_fb (vert_order_t *vo)
{ {
int count; int count;
@ -139,7 +139,7 @@ GL_DrawAliasFrameTri_fb (vert_order_t *vo)
qfglEnd(); qfglEnd();
} }
static void static inline void
GL_DrawAliasFrameTriMulti (vert_order_t *vo) GL_DrawAliasFrameTriMulti (vert_order_t *vo)
{ {
float color[4]; float color[4];
@ -168,7 +168,7 @@ GL_DrawAliasFrameTriMulti (vert_order_t *vo)
qfglEnd (); qfglEnd ();
} }
static void static inline void
GL_DrawAliasFrame (vert_order_t *vo) GL_DrawAliasFrame (vert_order_t *vo)
{ {
float color[4]; float color[4];
@ -208,7 +208,7 @@ GL_DrawAliasFrame (vert_order_t *vo)
} }
} }
static void static inline void
GL_DrawAliasFrame_fb (vert_order_t *vo) GL_DrawAliasFrame_fb (vert_order_t *vo)
{ {
int count; int count;
@ -243,7 +243,7 @@ GL_DrawAliasFrame_fb (vert_order_t *vo)
} }
} }
static void static inline void
GL_DrawAliasFrameMulti (vert_order_t *vo) GL_DrawAliasFrameMulti (vert_order_t *vo)
{ {
float color[4]; float color[4];
@ -339,7 +339,7 @@ GL_DrawAliasShadow (aliashdr_t *paliashdr, vert_order_t *vo)
} }
} }
static vert_order_t * static inline vert_order_t *
GL_GetAliasFrameVerts16 (int frame, aliashdr_t *paliashdr, entity_t *e) GL_GetAliasFrameVerts16 (int frame, aliashdr_t *paliashdr, entity_t *e)
{ {
float interval; float interval;
@ -445,7 +445,7 @@ GL_GetAliasFrameVerts16 (int frame, aliashdr_t *paliashdr, entity_t *e)
return vo; return vo;
} }
static vert_order_t * static inline vert_order_t *
GL_GetAliasFrameVerts (int frame, aliashdr_t *paliashdr, entity_t *e) GL_GetAliasFrameVerts (int frame, aliashdr_t *paliashdr, entity_t *e)
{ {
float interval; float interval;

4
libs/video/renderer/gl/gl_rsurf.c
View file

@ -144,7 +144,7 @@ R_RenderFullbrights (void)
qfglBlendFunc (GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA); qfglBlendFunc (GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
} }
void inline void
R_RenderBrushPoly (msurface_t *fa) R_RenderBrushPoly (msurface_t *fa)
{ {
float *v; float *v;
@ -233,7 +233,7 @@ R_DrawWaterSurfaces (void)
} }
} }
static void static inline void
DrawTextureChains (void) DrawTextureChains (void)
{ {
int i; int i;

4
libs/video/renderer/gl/gl_skin.c
View file

@ -94,7 +94,7 @@ Skin_Set_Translate (int top, int bottom, void *_dest)
} }
} }
static void static inline void
build_skin_8 (byte * original, int tinwidth, int tinheight, build_skin_8 (byte * original, int tinwidth, int tinheight,
unsigned int scaled_width, unsigned int scaled_height, unsigned int scaled_width, unsigned int scaled_height,
int inwidth, qboolean alpha) int inwidth, qboolean alpha)
@ -121,7 +121,7 @@ build_skin_8 (byte * original, int tinwidth, int tinheight,
alpha); alpha);
} }
static void static inline void
build_skin_32 (byte * original, int tinwidth, int tinheight, build_skin_32 (byte * original, int tinwidth, int tinheight,
unsigned int scaled_width, unsigned int scaled_height, unsigned int scaled_width, unsigned int scaled_height,
int inwidth, qboolean alpha) int inwidth, qboolean alpha)

2
libs/video/renderer/r_light.c
View file

@ -185,7 +185,7 @@ loc0:
} }
} }
static void static inline void
real_mark_surfaces (float dist, msurface_t *surf, const vec3_t lightorigin, real_mark_surfaces (float dist, msurface_t *surf, const vec3_t lightorigin,
dlight_t *light, int bit) dlight_t *light, int bit)
{ {