From 494a6294bfc56bd5eee64c7a9186c7934330f544 Mon Sep 17 00:00:00 2001
From: Nick Kurshev <nickols_k@mail.ru>
Date: Sun, 28 Oct 2001 17:42:52 +0000
Subject: 32to24: MMX, MMX2, 3DNOW optimization

Originally committed as revision 2518 to svn://svn.mplayerhq.hu/mplayer/trunk/postproc
---
 postproc/rgb2rgb.c          | 43 ++++++++++++++++++++++++++++++++++++++++++-
 postproc/rgb2rgb_template.c | 43 ++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 84 insertions(+), 2 deletions(-)

(limited to 'postproc')

diff --git a/postproc/rgb2rgb.c b/postproc/rgb2rgb.c
index 925e586df5..60d3d13e12 100644
--- a/postproc/rgb2rgb.c
+++ b/postproc/rgb2rgb.c
@@ -51,13 +51,52 @@ void rgb24to32(uint8_t *src,uint8_t *dst,uint32_t src_size)
   }
 }
 
-/* TODO: MMX optimization */
 void rgb32to24(uint8_t *src,uint8_t *dst,uint32_t src_size)
 {
   uint8_t *dest = dst;
   uint8_t *s = src;
   uint8_t *end;
+#ifdef HAVE_MMX
+  const uint64_t mask24l = 0x0000000000FFFFFFULL;
+  const uint64_t mask24h = 0x0000FFFFFF000000ULL;
+  uint8_t *mm_end;
+#endif
   end = s + src_size;
+#ifdef HAVE_MMX
+  __asm __volatile(PREFETCH" %0\n\t"::"m"(*s):"memory");
+  mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2));
+  __asm __volatile(
+    "movq %0, %%mm7\n\t"
+    "movq %1, %%mm6\n\t"
+    ::"m"(mask24l),"m"(mask24h):"memory");
+  if(mm_end == end) mm_end -= MMREG_SIZE*2;
+  while(s < mm_end)
+  {
+    __asm __volatile(
+	PREFETCH" 32%1\n\t"
+	"movq	%1, %%mm0\n\t"
+	"movq	8%1, %%mm1\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"psrlq	$8, %%mm2\n\t"
+	"psrlq	$8, %%mm3\n\t"
+	"pand	%%mm7, %%mm0\n\t"
+	"pand	%%mm7, %%mm1\n\t"
+	"pand	%%mm6, %%mm2\n\t"
+	"pand	%%mm6, %%mm3\n\t"
+	"por	%%mm2, %%mm0\n\t"
+	"por	%%mm3, %%mm1\n\t"
+	MOVNTQ"	%%mm0, %0\n\t"
+	MOVNTQ"	%%mm1, 6%0"
+	:"=m"(*dest)
+	:"m"(*s)
+	:"memory");
+    dest += 12;
+    s += 16;
+  }
+  __asm __volatile(SFENCE:::"memory");
+  __asm __volatile(EMMS:::"memory");
+#endif
   while(s < end)
   {
     *dest++ = *s++;
@@ -67,6 +106,8 @@ void rgb32to24(uint8_t *src,uint8_t *dst,uint32_t src_size)
   }
 }
 
+/* TODO: 3DNOW, MMX2 optimization */
+
 /* Original by Strepto/Astral
  ported to gcc & bugfixed : A'rpi */
 void rgb15to16(uint8_t *src,uint8_t *dst,uint32_t src_size)
diff --git a/postproc/rgb2rgb_template.c b/postproc/rgb2rgb_template.c
index 925e586df5..60d3d13e12 100644
--- a/postproc/rgb2rgb_template.c
+++ b/postproc/rgb2rgb_template.c
@@ -51,13 +51,52 @@ void rgb24to32(uint8_t *src,uint8_t *dst,uint32_t src_size)
   }
 }
 
-/* TODO: MMX optimization */
 void rgb32to24(uint8_t *src,uint8_t *dst,uint32_t src_size)
 {
   uint8_t *dest = dst;
   uint8_t *s = src;
   uint8_t *end;
+#ifdef HAVE_MMX
+  const uint64_t mask24l = 0x0000000000FFFFFFULL;
+  const uint64_t mask24h = 0x0000FFFFFF000000ULL;
+  uint8_t *mm_end;
+#endif
   end = s + src_size;
+#ifdef HAVE_MMX
+  __asm __volatile(PREFETCH" %0\n\t"::"m"(*s):"memory");
+  mm_end = (uint8_t*)((((unsigned long)end)/(MMREG_SIZE*2))*(MMREG_SIZE*2));
+  __asm __volatile(
+    "movq %0, %%mm7\n\t"
+    "movq %1, %%mm6\n\t"
+    ::"m"(mask24l),"m"(mask24h):"memory");
+  if(mm_end == end) mm_end -= MMREG_SIZE*2;
+  while(s < mm_end)
+  {
+    __asm __volatile(
+	PREFETCH" 32%1\n\t"
+	"movq	%1, %%mm0\n\t"
+	"movq	8%1, %%mm1\n\t"
+	"movq	%%mm0, %%mm2\n\t"
+	"movq	%%mm1, %%mm3\n\t"
+	"psrlq	$8, %%mm2\n\t"
+	"psrlq	$8, %%mm3\n\t"
+	"pand	%%mm7, %%mm0\n\t"
+	"pand	%%mm7, %%mm1\n\t"
+	"pand	%%mm6, %%mm2\n\t"
+	"pand	%%mm6, %%mm3\n\t"
+	"por	%%mm2, %%mm0\n\t"
+	"por	%%mm3, %%mm1\n\t"
+	MOVNTQ"	%%mm0, %0\n\t"
+	MOVNTQ"	%%mm1, 6%0"
+	:"=m"(*dest)
+	:"m"(*s)
+	:"memory");
+    dest += 12;
+    s += 16;
+  }
+  __asm __volatile(SFENCE:::"memory");
+  __asm __volatile(EMMS:::"memory");
+#endif
   while(s < end)
   {
     *dest++ = *s++;
@@ -67,6 +106,8 @@ void rgb32to24(uint8_t *src,uint8_t *dst,uint32_t src_size)
   }
 }
 
+/* TODO: 3DNOW, MMX2 optimization */
+
 /* Original by Strepto/Astral
  ported to gcc & bugfixed : A'rpi */
 void rgb15to16(uint8_t *src,uint8_t *dst,uint32_t src_size)
-- 
cgit v1.2.3