Improvements

git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@887 b3059339-0415-0410-9bf9-f77b7e298cf2

1 files changed, 54 insertions, 7 deletions

diff --git a/liba52/srfftp_3dnow.h b/liba52/srfftp_3dnow.h
index 721bfa720a..10f780b1a6 100644
--- a/liba52/srfftp_3dnow.h
+++ b/liba52/srfftp_3dnow.h
@@ -33,6 +33,8 @@
 #ifndef SRFFTP_3DNOW_H__
 #define SRFFTP_3DNOW_H__
 
+static float HSQRT2_3DNOW = 0.707106781188;
+
 #ifdef HAVE_3DNOWEX
 #define TRANS_FILL_MM6_MM7_3DNOW()\
     asm(\
@@ -59,10 +61,10 @@
 #endif
 
 #ifdef HAVE_3DNOWEX
-#define PSWAP_MM(mm_base,mm_hlp) "pswapd	"##mm_base","##mm_base" \n\t"
+#define PSWAP_MM(mm_base,mm_hlp) "pswapd	"##mm_base","##mm_base"\n\t"
 #else
 #define PSWAP_MM(mm_base,mm_hlp)\
-	"movq	"##mm_base","##mm_hlp" \n\t"\
+	"movq	"##mm_base","##mm_hlp"\n\t"\
 	"psrlq $32, "##mm_base"\n\t"\
 	"punpckldq "##mm_hlp","##mm_base"\n\t"
 #endif
@@ -82,16 +84,16 @@
 	"movq	%%mm0, %%mm4\n\t"/*v.re =-wTB[0].im + wTB[k*2].im;*/\
 	PSWAP_MM("%%mm4","%%mm2")/* mm4 = v*/\
 	"movq	%6, %%mm0\n\t" /* a1 = A0;*/\
-	"movq	%%mm0, %%mm1\n\t"\
-	"pfadd	%%mm5, %%mm0\n\t" /*A0 = a1 + u;*/\
-	"pfsub	%%mm5, %%mm1\n\t" /*A1 = a1 - u;*/\
-	"movq	%%mm0, %0\n\t"\
-	"movq	%%mm1, %1\n\t"\
 	"movq	%7, %%mm2\n\t" /* a1 = A4;*/\
+	"movq	%%mm0, %%mm1\n\t"\
 	"movq	%%mm2, %%mm3\n\t"\
+	"pfadd	%%mm5, %%mm0\n\t" /*A0 = a1 + u;*/\
 	"pfadd	%%mm4, %%mm2\n\t" /*A12 = a1 + v;*/\
+	"pfsub	%%mm5, %%mm1\n\t" /*A1 = a1 - u;*/\
 	"pfsub	%%mm4, %%mm3\n\t" /*A4  = a1 - v;*/\
+	"movq	%%mm0, %0\n\t"\
 	"movq	%%mm2, %3\n\t"\
+	"movq	%%mm1, %1\n\t"\
 	"movq	%%mm3, %2"\
 	:"=m"(A0), "=m"(A8), "=m"(A4), "=m"(A12)\
 	:"m"(wTB[0]), "m"(wTB[k*2]), "0"(A0), "2"(A4)\
@@ -99,4 +101,49 @@
     asm volatile("femms":::"memory");\
 }
 
+#define TRANSHALF_16_3DNOW(A2,A6,A10,A14)\
+{\
+    asm volatile("femms":::"memory");\
+    TRANS_FILL_MM6_MM7_3DNOW()\
+    asm(\
+	"movq	%4, %%mm0\n\t"/*u.re = wTB[2].im + wTB[2].re;*/\
+	"movq	%%mm0, %%mm1\n\t"\
+	"pfmul	%%mm7, %%mm1\n\t"\
+	"pfacc	%%mm1, %%mm0\n\t"/*u.im = wTB[2].im - wTB[2].re; mm0 = u*/\
+	"movq	%5, %%mm1\n\t"  /*a.re = wTB[6].im - wTB[6].re; */\
+	"movq	%%mm1, %%mm2\n\t"\
+	"pfmul	%%mm7, %%mm1\n\t"\
+	"pfacc	%%mm2, %%mm1\n\t"/*a.im = wTB[6].im + wTB[6].re;  mm1 = a*/\
+	"movq	%%mm1, %%mm2\n\t"\
+	"pfmul	%%mm7, %%mm2\n\t"/*v.im = u.re - a.re;*/\
+	"movq	%%mm0, %%mm3\n\t"/*v.re = u.im + a.im;*/\
+	"pfadd	%%mm2, %%mm3\n\t"\
+	PSWAP_MM("%%mm3","%%mm2")/*mm3 = v*/\
+	"pfmul	%%mm6, %%mm1\n\t"/*u.re = u.re + a.re;*/\
+	"pfadd	%%mm1, %%mm0\n\t"/*u.im = u.im - a.im; mm0 = u*/\
+	"movd	%8, %%mm2\n\t"\
+	"punpckldq %8, %%mm2\n\t"\
+	"pfmul	%%mm2, %%mm3\n\t" /* v *= HSQRT2_3DNOW; */\
+	"pfmul	%%mm2, %%mm0\n\t" /* u *= HSQRT2_3DNOW; */\
+	"movq	%6, %%mm1\n\t" /* a1 = A2;*/\
+	"movq	%%mm1, %%mm2\n\t"\
+	"pfadd	%%mm0, %%mm1\n\t" /*A2 = a1 + u;*/\
+	"pfsub	%%mm0, %%mm2\n\t" /*A2 = a1 - u;*/\
+	"movq	%%mm1, %0\n\t"\
+	"movq	%%mm2, %1\n\t"\
+	"movq	%7, %%mm1\n\t" /* a1 = A6;*/\
+	"movq	%%mm1, %%mm2\n\t"\
+	"movq	%%mm3, %%mm4\n\t"\
+	"pfmul	%%mm6, %%mm4\n\t"/*A6.re  = a1.re + v.re;*/\
+	"pfadd	%%mm4, %%mm1\n\t"/*A6.im  = a1.im - v.im;*/\
+	"pfmul	%%mm7, %%mm3\n\t"/*A14.re = a1.re - v.re;*/\
+	"pfadd	%%mm3, %%mm2\n\t"/*A14.im = a1.im + v.im;*/\
+	"movq	%%mm1, %2\n\t"\
+	"movq	%%mm2, %3"\
+	:"=m"(A2), "=m"(A10), "=m"(A6), "=m"(A14)\
+	:"m"(wTB[2]), "m"(wTB[6]), "0"(A2), "2"(A6), "m"(HSQRT2_3DNOW)\
+	:"memory");\
+    asm volatile("femms":::"memory");\
+}
+
 #endif