changeset 1564:b6b7d080f1a1 libavcodec

inline -> always_inline (842 -> 690 cpu cycles for dct_quantize() difference for the dct itself should be even bigger)
author michael
date Thu, 23 Oct 2003 12:59:21 +0000
parents 820e06c6ca9b
children 1a9a63f59849
files i386/fdct_mmx.c
diffstat 1 files changed, 4 insertions(+), 4 deletions(-) [+]
line wrap: on
line diff
--- a/i386/fdct_mmx.c	Thu Oct 23 12:41:54 2003 +0000
+++ b/i386/fdct_mmx.c	Thu Oct 23 12:59:21 2003 +0000
@@ -131,7 +131,7 @@
 };
 
 
-static inline void fdct_col(const int16_t *in, int16_t *out, int offset)
+static always_inline void fdct_col(const int16_t *in, int16_t *out, int offset)
 {
     movq_m2r(*(in + offset + 1 * 8), mm0);
     movq_m2r(*(in + offset + 6 * 8), mm1);
@@ -210,15 +210,15 @@
     movq_r2m(mm3, *(out + offset + 7 * 8));
 }
 
-static inline void fdct_row(const int16_t *in, int16_t *out, const int16_t *table)
-{
+static always_inline void fdct_row(const int16_t *in, int16_t *out, const int16_t *table)
+{ 
     movd_m2r(*(in + 6), mm5);
     punpcklwd_m2r(*(in + 4), mm5);
     movq_r2r(mm5, mm2);
     psrlq_i2r(0x20, mm5);
     movq_m2r(*(in + 0), mm0);
     punpcklwd_r2r(mm2, mm5);
-    movq_r2r(mm0, mm1);
+    movq_r2r(mm0, mm1);	
     paddsw_r2r(mm5, mm0);
     psubsw_r2r(mm5, mm1);
     movq_r2r(mm0, mm2);