[FFmpeg-devel] [PATCH][0.5] backport MMX runtime detection for libswscale and libpostproc

[Reinhard Tartler]

targeted for the 0.5 branch:

$ svn diff

Property changes on: .
___________________________________________________________________
Modified: svn:mergeinfo
   Merged /trunk:r18380

Index: configure
===================================================================
--- configure	(revision 21596)
+++ configure	(working copy)
@@ -103,6 +103,7 @@
   echo "  --disable-golomb         disable Golomb code"
   echo "  --disable-mdct           disable MDCT code"
   echo "  --disable-rdft           disable RDFT code"
+  echo "  --enable-runtime-cpudetect detect cpu capabilities at runtime (bigger binary)"
   echo "  --enable-hardcoded-tables use hardcoded tables instead of runtime generation"
   echo "  --enable-memalign-hack   emulate memalign, interferes with memory debuggers"
   echo "  --enable-beos-netserver  enable BeOS netserver"
@@ -805,6 +806,7 @@
     postproc
     powerpc_perf
     rdft
+    runtime_cpudetect
     shared
     small
     static
@@ -2268,6 +2270,7 @@
     echo "version string suffix     $extra_version"
 fi
 echo "big-endian                ${bigendian-no}"
+echo "runtime cpu detection     ${runtime_cpudetect-no}"
 if enabled x86; then
     echo "yasm                      ${yasm-no}"
     echo "MMX enabled               ${mmx-no}"
Index: libavcodec/ppc/check_altivec.c
===================================================================
--- libavcodec/ppc/check_altivec.c	(revision 21596)
+++ libavcodec/ppc/check_altivec.c	(working copy)
@@ -63,7 +63,7 @@
 
     if (err == 0) return has_vu != 0;
     return 0;
-#elif defined(RUNTIME_CPUDETECT)
+#elif CONFIG_RUNTIME_CPUDETECT
     int proc_ver;
     // Support of mfspr PVR emulation added in Linux 2.6.17.
     __asm__ volatile("mfspr %0, 287" : "=r" (proc_ver));
Index: libpostproc/postprocess.c
===================================================================
--- libpostproc/postprocess.c	(revision 21596)
+++ libpostproc/postprocess.c	(working copy)
@@ -554,7 +554,7 @@
 
 //Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one
 //Plain C versions
-#if !(HAVE_MMX || HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)
+#if !(HAVE_MMX || HAVE_ALTIVEC) || CONFIG_RUNTIME_CPUDETECT
 #define COMPILE_C
 #endif
 
@@ -564,15 +564,15 @@
 
 #if ARCH_X86
 
-#if (HAVE_MMX && !HAVE_AMD3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
+#if (HAVE_MMX && !HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT
 #define COMPILE_MMX
 #endif
 
-#if HAVE_MMX2 || defined (RUNTIME_CPUDETECT)
+#if HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT
 #define COMPILE_MMX2
 #endif
 
-#if (HAVE_AMD3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
+#if (HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT
 #define COMPILE_3DNOW
 #endif
 #endif /* ARCH_X86 */
@@ -645,7 +645,7 @@
     // Using ifs here as they are faster than function pointers although the
     // difference would not be measurable here but it is much better because
     // someone might exchange the CPU whithout restarting MPlayer ;)
-#ifdef RUNTIME_CPUDETECT
+#if CONFIG_RUNTIME_CPUDETECT
 #if ARCH_X86
     // ordered per speed fastest first
     if(c->cpuCaps & PP_CPU_CAPS_MMX2)
@@ -664,7 +664,7 @@
 #endif
             postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
 #endif
-#else //RUNTIME_CPUDETECT
+#else //CONFIG_RUNTIME_CPUDETECT
 #if   HAVE_MMX2
             postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
 #elif HAVE_AMD3DNOW
@@ -676,7 +676,7 @@
 #else
             postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
 #endif
-#endif //!RUNTIME_CPUDETECT
+#endif //!CONFIG_RUNTIME_CPUDETECT
 }
 
 //static void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
Index: libswscale/swscale.c
===================================================================
--- libswscale/swscale.c	(revision 21596)
+++ libswscale/swscale.c	(working copy)
@@ -955,27 +955,27 @@
 
 //Note: we have C, X86, MMX, MMX2, 3DNOW versions, there is no 3DNOW+MMX2 one
 //Plain C versions
-#if ((!HAVE_MMX || !CONFIG_GPL) && !HAVE_ALTIVEC) || defined (RUNTIME_CPUDETECT)
+#if ((!HAVE_MMX || !CONFIG_GPL) && !HAVE_ALTIVEC) || CONFIG_RUNTIME_CPUDETECT
 #define COMPILE_C
 #endif
 
 #if ARCH_PPC
-#if HAVE_ALTIVEC || defined (RUNTIME_CPUDETECT)
+#if HAVE_ALTIVEC || CONFIG_RUNTIME_CPUDETECT
 #define COMPILE_ALTIVEC
 #endif
 #endif //ARCH_PPC
 
 #if ARCH_X86
 
-#if ((HAVE_MMX && !HAVE_AMD3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
+#if ((HAVE_MMX && !HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
 #define COMPILE_MMX
 #endif
 
-#if (HAVE_MMX2 || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
+#if (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
 #define COMPILE_MMX2
 #endif
 
-#if ((HAVE_AMD3DNOW && !HAVE_MMX2) || defined (RUNTIME_CPUDETECT)) && CONFIG_GPL
+#if ((HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT) && CONFIG_GPL
 #define COMPILE_3DNOW
 #endif
 #endif //ARCH_X86
@@ -1636,7 +1636,7 @@
 
 static SwsFunc getSwsFunc(int flags){
 
-#if defined(RUNTIME_CPUDETECT)
+#if CONFIG_RUNTIME_CPUDETECT
 #if ARCH_X86 && CONFIG_GPL
     // ordered per speed fastest first
     if (flags & SWS_CPU_CAPS_MMX2)
@@ -1657,7 +1657,7 @@
 #endif
     return swScale_C;
 #endif /* ARCH_X86 && CONFIG_GPL */
-#else //RUNTIME_CPUDETECT
+#else //CONFIG_RUNTIME_CPUDETECT
 #if   HAVE_MMX2
     return swScale_MMX2;
 #elif HAVE_AMD3DNOW
@@ -1669,7 +1669,7 @@
 #else
     return swScale_C;
 #endif
-#endif //!RUNTIME_CPUDETECT
+#endif //!CONFIG_RUNTIME_CPUDETECT
 }
 
 static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
@@ -2193,7 +2193,7 @@
         __asm__ volatile("emms\n\t"::: "memory");
 #endif
 
-#if !defined(RUNTIME_CPUDETECT) //ensure that the flags match the compiled variant if cpudetect is off
+#if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off
     flags &= ~(SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2|SWS_CPU_CAPS_3DNOW|SWS_CPU_CAPS_ALTIVEC|SWS_CPU_CAPS_BFIN);
 #if   HAVE_MMX2
     flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
@@ -2206,7 +2206,7 @@
 #elif ARCH_BFIN
     flags |= SWS_CPU_CAPS_BFIN;
 #endif
-#endif /* RUNTIME_CPUDETECT */
+#endif /* CONFIG_RUNTIME_CPUDETECT */
     if (clip_table[512] != 255) globalInit();
     if (!rgb15to16) sws_rgb2rgb_init(flags);
 
-- 
Gruesse/greetings,
Reinhard Tartler, KeyID 945348A4