swscale: force --enable-runtime-cpudetect and remove SWS_CPU_CAPS_*.

16 files changed, 171 insertions, 340 deletions

diff --git a/libswscale/bfin/swscale_bfin.c b/libswscale/bfin/swscale_bfin.c
index fa3c03b4ea..60b7f8310b 100644
--- a/libswscale/bfin/swscale_bfin.c
+++ b/libswscale/bfin/swscale_bfin.c
@@ -79,15 +79,13 @@ static int yuyvtoyv12_unscaled(SwsContext *c, uint8_t* src[], int srcStride[], i
 void ff_bfin_get_unscaled_swscale(SwsContext *c)
 {
     SwsFunc swScale = c->swScale;
-    if (c->flags & SWS_CPU_CAPS_BFIN)
-        if (c->dstFormat == PIX_FMT_YUV420P)
-            if (c->srcFormat == PIX_FMT_UYVY422) {
-                av_log (NULL, AV_LOG_VERBOSE, "selecting Blackfin optimized uyvytoyv12_unscaled\n");
-                c->swScale = uyvytoyv12_unscaled;
-            }
-        if (c->dstFormat == PIX_FMT_YUV420P)
-            if (c->srcFormat == PIX_FMT_YUYV422) {
-                av_log (NULL, AV_LOG_VERBOSE, "selecting Blackfin optimized yuyvtoyv12_unscaled\n");
-                c->swScale = yuyvtoyv12_unscaled;
-            }
+
+    if (c->dstFormat == PIX_FMT_YUV420P && c->srcFormat == PIX_FMT_UYVY422) {
+        av_log (NULL, AV_LOG_VERBOSE, "selecting Blackfin optimized uyvytoyv12_unscaled\n");
+        c->swScale = uyvytoyv12_unscaled;
+    }
+    if (c->dstFormat == PIX_FMT_YUV420P && c->srcFormat == PIX_FMT_YUYV422) {
+        av_log (NULL, AV_LOG_VERBOSE, "selecting Blackfin optimized yuyvtoyv12_unscaled\n");
+        c->swScale = yuyvtoyv12_unscaled;
+    }
 }
diff --git a/libswscale/colorspace-test.c b/libswscale/colorspace-test.c
index 4e7116f5dc..4ed9164323 100644
--- a/libswscale/colorspace-test.c
+++ b/libswscale/colorspace-test.c
@@ -33,31 +33,6 @@
 
 #define FUNC(s,d,n) {s,d,#n,n}
 
-static int cpu_caps;
-
-static char *args_parse(int argc, char *argv[])
-{
-    int o;
-
-    while ((o = getopt(argc, argv, "m23")) != -1) {
-        switch (o) {
-        case 'm':
-            cpu_caps |= SWS_CPU_CAPS_MMX;
-            break;
-        case '2':
-            cpu_caps |= SWS_CPU_CAPS_MMX2;
-            break;
-        case '3':
-            cpu_caps |= SWS_CPU_CAPS_3DNOW;
-            break;
-        default:
-            av_log(NULL, AV_LOG_ERROR, "Unknown option %c\n", o);
-        }
-    }
-
-    return argv[optind];
-}
-
 int main(int argc, char **argv)
 {
     int i, funcNum;
@@ -70,9 +45,7 @@ int main(int argc, char **argv)
         return -1;
 
     av_log(NULL, AV_LOG_INFO, "memory corruption test ...\n");
-    args_parse(argc, argv);
-    av_log(NULL, AV_LOG_INFO, "CPU capabilities forced to %x\n", cpu_caps);
-    sws_rgb2rgb_init(cpu_caps);
+    sws_rgb2rgb_init();
 
     for(funcNum=0; ; funcNum++) {
         struct func_info_s {
diff --git a/libswscale/options.c b/libswscale/options.c
index f80735b261..ecd0ecd53a 100644
--- a/libswscale/options.c
+++ b/libswscale/options.c
@@ -48,12 +48,6 @@ static const AVOption options[] = {
     { "spline", "natural bicubic spline", 0, FF_OPT_TYPE_CONST, {.dbl = SWS_SPLINE }, INT_MIN, INT_MAX, VE, "sws_flags" },
     { "print_info", "print info", 0, FF_OPT_TYPE_CONST, {.dbl = SWS_PRINT_INFO }, INT_MIN, INT_MAX, VE, "sws_flags" },
     { "accurate_rnd", "accurate rounding", 0, FF_OPT_TYPE_CONST, {.dbl = SWS_ACCURATE_RND }, INT_MIN, INT_MAX, VE, "sws_flags" },
-    { "mmx", "MMX SIMD acceleration", 0, FF_OPT_TYPE_CONST, {.dbl = SWS_CPU_CAPS_MMX }, INT_MIN, INT_MAX, VE, "sws_flags" },
-    { "mmx2", "MMX2 SIMD acceleration", 0, FF_OPT_TYPE_CONST, {.dbl = SWS_CPU_CAPS_MMX2 }, INT_MIN, INT_MAX, VE, "sws_flags" },
-    { "sse2", "SSE2 SIMD acceleration", 0, FF_OPT_TYPE_CONST, {.dbl = SWS_CPU_CAPS_SSE2 }, INT_MIN, INT_MAX, VE, "sws_flags" },
-    { "3dnow", "3DNOW SIMD acceleration", 0, FF_OPT_TYPE_CONST, {.dbl = SWS_CPU_CAPS_3DNOW }, INT_MIN, INT_MAX, VE, "sws_flags" },
-    { "altivec", "AltiVec SIMD acceleration", 0, FF_OPT_TYPE_CONST, {.dbl = SWS_CPU_CAPS_ALTIVEC }, INT_MIN, INT_MAX, VE, "sws_flags" },
-    { "bfin", "Blackfin SIMD acceleration", 0, FF_OPT_TYPE_CONST, {.dbl = SWS_CPU_CAPS_BFIN }, INT_MIN, INT_MAX, VE, "sws_flags" },
     { "full_chroma_int", "full chroma interpolation", 0 , FF_OPT_TYPE_CONST, {.dbl = SWS_FULL_CHR_H_INT }, INT_MIN, INT_MAX, VE, "sws_flags" },
     { "full_chroma_inp", "full chroma input", 0 , FF_OPT_TYPE_CONST, {.dbl = SWS_FULL_CHR_H_INP }, INT_MIN, INT_MAX, VE, "sws_flags" },
     { "bitexact", "", 0 , FF_OPT_TYPE_CONST, {.dbl = SWS_BITEXACT }, INT_MIN, INT_MAX, VE, "sws_flags" },
diff --git a/libswscale/ppc/yuv2rgb_altivec.c b/libswscale/ppc/yuv2rgb_altivec.c
index 0113c8ddd5..96c208a074 100644
--- a/libswscale/ppc/yuv2rgb_altivec.c
+++ b/libswscale/ppc/yuv2rgb_altivec.c
@@ -94,6 +94,7 @@ adjustment.
 #include "libswscale/rgb2rgb.h"
 #include "libswscale/swscale.h"
 #include "libswscale/swscale_internal.h"
+#include "libavutil/cpu.h"
 
 #undef PROFILE_THE_BEAST
 #undef INC_SCALING
@@ -692,7 +693,7 @@ static int altivec_uyvy_rgb32 (SwsContext *c,
 */
 SwsFunc ff_yuv2rgb_init_altivec(SwsContext *c)
 {
-    if (!(c->flags & SWS_CPU_CAPS_ALTIVEC))
+    if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC))
         return NULL;
 
     /*
diff --git a/libswscale/rgb2rgb.c b/libswscale/rgb2rgb.c
index 74cc42e3b0..e18cd51011 100644
--- a/libswscale/rgb2rgb.c
+++ b/libswscale/rgb2rgb.c
@@ -116,12 +116,11 @@ void (*yuyvtoyuv422)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t
  32-bit C version, and and&add trick by Michael Niedermayer
 */
 
-void sws_rgb2rgb_init(int flags)
+void sws_rgb2rgb_init(void)
 {
     rgb2rgb_init_c();
-#if HAVE_MMX2 || HAVE_AMD3DNOW || HAVE_MMX
-    rgb2rgb_init_x86(flags);
-#endif /* HAVE_MMX2 || HAVE_AMD3DNOW || HAVE_MMX */
+    if (HAVE_MMX)
+        rgb2rgb_init_x86();
 }
 
 void rgb32to24(const uint8_t *src, uint8_t *dst, long src_size)
diff --git a/libswscale/rgb2rgb.h b/libswscale/rgb2rgb.h
index 158d9a9a87..a8d5531cb8 100644
--- a/libswscale/rgb2rgb.h
+++ b/libswscale/rgb2rgb.h
@@ -156,8 +156,8 @@ extern void (*yuyvtoyuv422)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const u
                             long width, long height,
                             long lumStride, long chromStride, long srcStride);
 
-void sws_rgb2rgb_init(int flags);
+void sws_rgb2rgb_init(void);
 
-void rgb2rgb_init_x86(int flags);
+void rgb2rgb_init_x86(void);
 
 #endif /* SWSCALE_RGB2RGB_H */
diff --git a/libswscale/rgb2rgb_template.c b/libswscale/rgb2rgb_template.c
index 7537419d83..ea39be4f13 100644
--- a/libswscale/rgb2rgb_template.c
+++ b/libswscale/rgb2rgb_template.c
@@ -278,25 +278,6 @@ static inline void rgb16tobgr24_c(const uint8_t *src, uint8_t *dst, long src_siz
     }
 }
 
-/*
- * mm0 = 00 B3 00 B2 00 B1 00 B0
- * mm1 = 00 G3 00 G2 00 G1 00 G0
- * mm2 = 00 R3 00 R2 00 R1 00 R0
- * mm6 = FF FF FF FF FF FF FF FF
- * mm7 = 00 00 00 00 00 00 00 00
- */
-#define PACK_RGB32 \
-    "packuswb   %%mm7, %%mm0    \n\t" /* 00 00 00 00 B3 B2 B1 B0 */ \
-    "packuswb   %%mm7, %%mm1    \n\t" /* 00 00 00 00 G3 G2 G1 G0 */ \
-    "packuswb   %%mm7, %%mm2    \n\t" /* 00 00 00 00 R3 R2 R1 R0 */ \
-    "punpcklbw  %%mm1, %%mm0    \n\t" /* G3 B3 G2 B2 G1 B1 G0 B0 */ \
-    "punpcklbw  %%mm6, %%mm2    \n\t" /* FF R3 FF R2 FF R1 FF R0 */ \
-    "movq       %%mm0, %%mm3    \n\t"                               \
-    "punpcklwd  %%mm2, %%mm0    \n\t" /* FF R1 G1 B1 FF R0 G0 B0 */ \
-    "punpckhwd  %%mm2, %%mm3    \n\t" /* FF R3 G3 B3 FF R2 G2 B2 */ \
-    MOVNTQ"     %%mm0,  %0      \n\t"                               \
-    MOVNTQ"     %%mm3, 8%0      \n\t"                               \
-
 static inline void rgb15to32_c(const uint8_t *src, uint8_t *dst, long src_size)
 {
     const uint16_t *end;
diff --git a/libswscale/swscale.c b/libswscale/swscale.c
index 3943aa0b72..749a276e5a 100644
--- a/libswscale/swscale.c
+++ b/libswscale/swscale.c
@@ -62,6 +62,7 @@ untested special converters
 #include "rgb2rgb.h"
 #include "libavutil/intreadwrite.h"
 #include "libavutil/x86_cpu.h"
+#include "libavutil/cpu.h"
 #include "libavutil/avutil.h"
 #include "libavutil/mathematics.h"
 #include "libavutil/bswap.h"
@@ -70,10 +71,6 @@ untested special converters
 #undef MOVNTQ
 #undef PAVGB
 
-//#undef HAVE_MMX2
-//#define HAVE_AMD3DNOW
-//#undef HAVE_MMX
-//#undef ARCH_X86
 #define DITHER1XBPP
 
 #define isPacked(x)         (       \
@@ -1179,57 +1176,14 @@ static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uin
 
 //Note: we have C, MMX, MMX2, 3DNOW versions, there is no 3DNOW+MMX2 one
 //Plain C versions
-#if CONFIG_RUNTIME_CPUDETECT
-#  define COMPILE_C 1
-#  if   ARCH_X86
-#    define COMPILE_MMX     HAVE_MMX
-#    define COMPILE_MMX2    HAVE_MMX2
-#    define COMPILE_3DNOW   HAVE_AMD3DNOW
-#  elif ARCH_PPC
-#    define COMPILE_ALTIVEC HAVE_ALTIVEC
-#  endif
-#else /* CONFIG_RUNTIME_CPUDETECT */
-#  if   ARCH_X86
-#    if   HAVE_MMX2
-#      define COMPILE_MMX2  1
-#    elif HAVE_AMD3DNOW
-#      define COMPILE_3DNOW 1
-#    elif HAVE_MMX
-#      define COMPILE_MMX   1
-#    else
-#      define COMPILE_C     1
-#    endif
-#  elif ARCH_PPC && HAVE_ALTIVEC
-#    define COMPILE_ALTIVEC 1
-#  else
-#    define COMPILE_C       1
-#  endif
-#endif
-
-#ifndef COMPILE_C
-#  define COMPILE_C 0
-#endif
-#ifndef COMPILE_MMX
-#  define COMPILE_MMX 0
-#endif
-#ifndef COMPILE_MMX2
-#  define COMPILE_MMX2 0
-#endif
-#ifndef COMPILE_3DNOW
-#  define COMPILE_3DNOW 0
-#endif
-#ifndef COMPILE_ALTIVEC
-#  define COMPILE_ALTIVEC 0
-#endif
 
-#define COMPILE_TEMPLATE_MMX 0
 #define COMPILE_TEMPLATE_MMX2 0
 #define COMPILE_TEMPLATE_AMD3DNOW 0
 #define COMPILE_TEMPLATE_ALTIVEC 0
 
 #include "swscale_template.c"
 
-#if COMPILE_ALTIVEC
+#if HAVE_ALTIVEC
 #undef RENAME
 #undef COMPILE_TEMPLATE_ALTIVEC
 #define COMPILE_TEMPLATE_ALTIVEC 1
@@ -1237,15 +1191,11 @@ static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uin
 #include "ppc/swscale_template.c"
 #endif
 
-#if ARCH_X86
-
 //MMX versions
-#if COMPILE_MMX
+#if HAVE_MMX
 #undef RENAME
-#undef COMPILE_TEMPLATE_MMX
 #undef COMPILE_TEMPLATE_MMX2
 #undef COMPILE_TEMPLATE_AMD3DNOW
-#define COMPILE_TEMPLATE_MMX 1
 #define COMPILE_TEMPLATE_MMX2 0
 #define COMPILE_TEMPLATE_AMD3DNOW 0
 #define RENAME(a) a ## _MMX
@@ -1253,12 +1203,10 @@ static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uin
 #endif
 
 //MMX2 versions
-#if COMPILE_MMX2
+#if HAVE_MMX2
 #undef RENAME
-#undef COMPILE_TEMPLATE_MMX
 #undef COMPILE_TEMPLATE_MMX2
 #undef COMPILE_TEMPLATE_AMD3DNOW
-#define COMPILE_TEMPLATE_MMX 1
 #define COMPILE_TEMPLATE_MMX2 1
 #define COMPILE_TEMPLATE_AMD3DNOW 0
 #define RENAME(a) a ## _MMX2
@@ -1266,61 +1214,47 @@ static inline void monoblack2Y(uint8_t *dst, const uint8_t *src, long width, uin
 #endif
 
 //3DNOW versions
-#if COMPILE_3DNOW
+#if HAVE_AMD3DNOW
 #undef RENAME
-#undef COMPILE_TEMPLATE_MMX
 #undef COMPILE_TEMPLATE_MMX2
 #undef COMPILE_TEMPLATE_AMD3DNOW
-#define COMPILE_TEMPLATE_MMX 1
 #define COMPILE_TEMPLATE_MMX2 0
 #define COMPILE_TEMPLATE_AMD3DNOW 1
 #define RENAME(a) a ## _3DNow
 #include "x86/swscale_template.c"
 #endif
 
-#endif //ARCH_X86
-
 SwsFunc ff_getSwsFunc(SwsContext *c)
 {
+    int cpu_flags = av_get_cpu_flags();
+
     sws_init_swScale_c(c);
 
-#if CONFIG_RUNTIME_CPUDETECT
-#if ARCH_X86
+#if HAVE_MMX2
     // ordered per speed fastest first
-    if (c->flags & SWS_CPU_CAPS_MMX2) {
+    if (cpu_flags & AV_CPU_FLAG_MMX2) {
         sws_init_swScale_MMX2(c);
         return swScale_MMX2;
-    } else if (c->flags & SWS_CPU_CAPS_3DNOW) {
+    } else
+#endif
+#if HAVE_AMD3DNOW
+    if (cpu_flags & AV_CPU_FLAG_3DNOW) {
         sws_init_swScale_3DNow(c);
         return swScale_3DNow;
-    } else if (c->flags & SWS_CPU_CAPS_MMX) {
+    } else
+#endif
+#if HAVE_MMX
+    if (cpu_flags & AV_CPU_FLAG_MMX) {
         sws_init_swScale_MMX(c);
         return swScale_MMX;
-    }
-
-#else
-#if COMPILE_ALTIVEC
-    if (c->flags & SWS_CPU_CAPS_ALTIVEC) {
+    } else
+#endif
+#if HAVE_ALTIVEC
+    if (cpu_flags & AV_CPU_FLAG_ALTIVEC) {
         sws_init_swScale_altivec(c);
         return swScale_altivec;
-    }
-#endif
-#endif /* ARCH_X86 */
-#else //CONFIG_RUNTIME_CPUDETECT
-#if   COMPILE_TEMPLATE_MMX2
-    sws_init_swScale_MMX2(c);
-    return swScale_MMX2;
-#elif COMPILE_TEMPLATE_AMD3DNOW
-    sws_init_swScale_3DNow(c);
-    return swScale_3DNow;
-#elif COMPILE_TEMPLATE_MMX
-    sws_init_swScale_MMX(c);
-    return swScale_MMX;
-#elif COMPILE_TEMPLATE_ALTIVEC
-    sws_init_swScale_altivec(c);
-    return swScale_altivec;
+    } else
 #endif
-#endif //!CONFIG_RUNTIME_CPUDETECT
 
     return swScale_c;
 }
@@ -1864,23 +1798,6 @@ static int planarCopyWrapper(SwsContext *c, const uint8_t* src[], int srcStride[
     return srcSliceH;
 }
 
-int ff_hardcodedcpuflags(void)
-{
-    int flags = 0;
-#if   COMPILE_TEMPLATE_MMX2
-    flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_MMX2;
-#elif COMPILE_TEMPLATE_AMD3DNOW
-    flags |= SWS_CPU_CAPS_MMX|SWS_CPU_CAPS_3DNOW;
-#elif COMPILE_TEMPLATE_MMX
-    flags |= SWS_CPU_CAPS_MMX;
-#elif COMPILE_TEMPLATE_ALTIVEC
-    flags |= SWS_CPU_CAPS_ALTIVEC;
-#elif ARCH_BFIN
-    flags |= SWS_CPU_CAPS_BFIN;
-#endif
-    return flags;
-}
-
 void ff_get_unscaled_swscale(SwsContext *c)
 {
     const enum PixelFormat srcFormat = c->srcFormat;
@@ -1964,8 +1881,8 @@ void ff_get_unscaled_swscale(SwsContext *c)
     if(srcFormat == PIX_FMT_UYVY422 && dstFormat == PIX_FMT_YUV422P)
         c->swScale= uyvyToYuv422Wrapper;
 
-#if COMPILE_ALTIVEC
-    if ((c->flags & SWS_CPU_CAPS_ALTIVEC) &&
+#if HAVE_ALTIVEC
+    if ((av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) &&
         !(c->flags & SWS_BITEXACT) &&
         srcFormat == PIX_FMT_YUV420P) {
         // unscaled YV12 -> packed YUV, we want speed
@@ -1995,8 +1912,7 @@ void ff_get_unscaled_swscale(SwsContext *c)
             c->swScale= planarCopyWrapper;
     }
 #if ARCH_BFIN
-    if (flags & SWS_CPU_CAPS_BFIN)
-        ff_bfin_get_unscaled_swscale (c);
+    ff_bfin_get_unscaled_swscale (c);
 #endif
 }
 
diff --git a/libswscale/swscale.h b/libswscale/swscale.h
index dd4de76b0a..1babced737 100644
--- a/libswscale/swscale.h
+++ b/libswscale/swscale.h
@@ -95,13 +95,6 @@ const char *swscale_license(void);
 #define SWS_ACCURATE_RND      0x40000
 #define SWS_BITEXACT          0x80000
 
-#define SWS_CPU_CAPS_MMX      0x80000000
-#define SWS_CPU_CAPS_MMX2     0x20000000
-#define SWS_CPU_CAPS_3DNOW    0x40000000
-#define SWS_CPU_CAPS_ALTIVEC  0x10000000
-#define SWS_CPU_CAPS_BFIN     0x01000000
-#define SWS_CPU_CAPS_SSE2     0x02000000
-
 #define SWS_MAX_REDUCE_CUTOFF 0.002
 
 #define SWS_CS_ITU709         1
diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h
index 2369546cb7..5f2ff94691 100644
--- a/libswscale/swscale_internal.h
+++ b/libswscale/swscale_internal.h
@@ -476,11 +476,6 @@ extern const AVClass sws_context_class;
 void ff_get_unscaled_swscale(SwsContext *c);
 
 /**
- * Returns the SWS_CPU_CAPS for the optimized code compiled into swscale.
- */
-int ff_hardcodedcpuflags(void);
-
-/**
  * Returns function pointer to fastest main scaler path function depending
  * on architecture and available optimizations.
  */
diff --git a/libswscale/utils.c b/libswscale/utils.c
index e9319fa67b..b49ec89be8 100644
--- a/libswscale/utils.c
+++ b/libswscale/utils.c
@@ -39,6 +39,7 @@
 #include "rgb2rgb.h"
 #include "libavutil/intreadwrite.h"
 #include "libavutil/x86_cpu.h"
+#include "libavutil/cpu.h"
 #include "libavutil/avutil.h"
 #include "libavutil/bswap.h"
 #include "libavutil/opt.h"
@@ -180,7 +181,7 @@ static double getSplineCoeff(double a, double b, double c, double d, double dist
 }
 
 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
-                      int srcW, int dstW, int filterAlign, int one, int flags,
+                      int srcW, int dstW, int filterAlign, int one, int flags, int cpu_flags,
                       SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
 {
     int i;
@@ -191,10 +192,9 @@ static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSi
     int64_t *filter2=NULL;
     const int64_t fone= 1LL<<54;
     int ret= -1;
-#if ARCH_X86
-    if (flags & SWS_CPU_CAPS_MMX)
+
+    if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX)
         __asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions)
-#endif
 
     // NOTE: the +1 is for the MMX scaler which reads over the end
     FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
@@ -411,7 +411,7 @@ static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSi
         if (min>minFilterSize) minFilterSize= min;
     }
 
-    if (flags & SWS_CPU_CAPS_ALTIVEC) {
+    if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
         // we can handle the special case 4,
         // so we don't want to go to the full 8
         if (minFilterSize < 5)
@@ -426,7 +426,7 @@ static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSi
             filterAlign = 1;
     }
 
-    if (flags & SWS_CPU_CAPS_MMX) {
+    if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
         // special case for unscaled vertical filtering
         if (minFilterSize == 1 && filterAlign == 2)
             filterAlign= 1;
@@ -516,7 +516,7 @@ fail:
     return ret;
 }
 
-#if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)
+#if HAVE_MMX2
 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
 {
     uint8_t *fragmentA;
@@ -674,7 +674,7 @@ static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *fil
 
     return fragmentPos + 1;
 }
-#endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) */
+#endif /* HAVE_MMX2 */
 
 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
 {
@@ -682,8 +682,6 @@ static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
     *v = av_pix_fmt_descriptors[format].log2_chroma_h;
 }
 
-static int update_flags_cpu(int flags);
-
 int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation)
 {
     memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
@@ -698,15 +696,12 @@ int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange
 
     c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
     c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
-    c->flags = update_flags_cpu(c->flags);
 
     ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
     //FIXME factorize
 
-#if HAVE_ALTIVEC
-    if (c->flags & SWS_CPU_CAPS_ALTIVEC)
+    if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
         ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
-#endif
     return 0;
 }
 
@@ -736,20 +731,6 @@ static int handle_jpeg(enum PixelFormat *format)
     }
 }
 
-static int update_flags_cpu(int flags)
-{
-#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_SSE2
-               |SWS_CPU_CAPS_ALTIVEC
-               |SWS_CPU_CAPS_BFIN);
-    flags |= ff_hardcodedcpuflags();
-#endif /* CONFIG_RUNTIME_CPUDETECT */
-    return flags;
-}
-
 SwsContext *sws_alloc_context(void)
 {
     SwsContext *c= av_mallocz(sizeof(SwsContext));
@@ -770,16 +751,15 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
     int srcH= c->srcH;
     int dstW= c->dstW;
     int dstH= c->dstH;
-    int flags;
+    int flags, cpu_flags;
     enum PixelFormat srcFormat= c->srcFormat;
     enum PixelFormat dstFormat= c->dstFormat;
 
-    flags= c->flags = update_flags_cpu(c->flags);
-#if ARCH_X86
-    if (flags & SWS_CPU_CAPS_MMX)
+    cpu_flags = av_get_cpu_flags();
+    flags     = c->flags;
+    if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX)
         __asm__ volatile("emms\n\t"::: "memory");
-#endif
-    if (!rgb15to16) sws_rgb2rgb_init(flags);
+    if (!rgb15to16) sws_rgb2rgb_init();
 
     unscaled = (srcW == dstW && srcH == dstH);
 
@@ -872,7 +852,7 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
         }
     }
 
-    if (flags & SWS_CPU_CAPS_MMX2) {
+    if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) {
         c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
         if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
             if (flags&SWS_PRINT_INFO)
@@ -898,7 +878,7 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
             c->chrXInc+= 20;
         }
         //we don't use the x86 asm scaler if MMX is available
-        else if (flags & SWS_CPU_CAPS_MMX) {
+        else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
             c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
             c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
         }
@@ -906,7 +886,7 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
 
     /* precalculate horizontal scaler filter coefficients */
     {
-#if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)
+#if HAVE_MMX2
 // can't downscale !!!
         if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
             c->lumMmx2FilterCodeSize = initMMX2HScaler(      dstW, c->lumXInc, NULL, NULL, NULL, 8);
@@ -938,21 +918,21 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
             mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
 #endif
         } else
-#endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) */
+#endif /* HAVE_MMX2 */
         {
             const int filterAlign=
-                (flags & SWS_CPU_CAPS_MMX) ? 4 :
-                (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
+                (HAVE_MMX     && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
+                (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
                 1;
 
             if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
                            srcW      ,       dstW, filterAlign, 1<<14,
-                           (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
+                           (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags, cpu_flags,
                            srcFilter->lumH, dstFilter->lumH, c->param) < 0)
                 goto fail;
             if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
                            c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
-                           (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
+                           (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
                            srcFilter->chrH, dstFilter->chrH, c->param) < 0)
                 goto fail;
         }
@@ -961,18 +941,18 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
     /* precalculate vertical scaler filter coefficients */
     {
         const int filterAlign=
-            (flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
-            (flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
+            (HAVE_MMX     && cpu_flags & AV_CPU_FLAG_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
+            (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
             1;
 
         if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
                        srcH      ,        dstH, filterAlign, (1<<12),
-                       (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags,
+                       (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC)  : flags, cpu_flags,
                        srcFilter->lumV, dstFilter->lumV, c->param) < 0)
             goto fail;
         if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
                        c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
-                       (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags,
+                       (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
                        srcFilter->chrV, dstFilter->chrV, c->param) < 0)
             goto fail;
 
@@ -1066,13 +1046,13 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
 #endif
                sws_format_name(dstFormat));
 
-        if      (flags & SWS_CPU_CAPS_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");
-        else if (flags & SWS_CPU_CAPS_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");
-        else if (flags & SWS_CPU_CAPS_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");
-        else if (flags & SWS_CPU_CAPS_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
+        if      (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)    av_log(c, AV_LOG_INFO, "using MMX2\n");
+        else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)   av_log(c, AV_LOG_INFO, "using 3DNOW\n");
+        else if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)     av_log(c, AV_LOG_INFO, "using MMX\n");
+        else if (HAVE_ALTIVEC  && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
         else                                   av_log(c, AV_LOG_INFO, "using C\n");
 
-        if (flags & SWS_CPU_CAPS_MMX) {
+        if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
             if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
                 av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
             else {
@@ -1091,7 +1071,7 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
                     av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
             }
         } else {
-#if ARCH_X86
+#if HAVE_MMX
             av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
 #else
             if (flags & SWS_FAST_BILINEAR)
@@ -1102,31 +1082,41 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
         }
         if (isPlanarYUV(dstFormat)) {
             if (c->vLumFilterSize==1)
-                av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+                av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n",
+                       (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
             else
-                av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+                av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n",
+                       (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
         } else {
             if (c->vLumFilterSize==1 && c->vChrFilterSize==2)
                 av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
-                       "      2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+                       "      2-tap scaler for vertical chrominance scaling (BGR)\n",
+                       (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
             else if (c->vLumFilterSize==2 && c->vChrFilterSize==2)
-                av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+                av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n",
+                       (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
             else
-                av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+                av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n",
+                       (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
         }
 
         if (dstFormat==PIX_FMT_BGR24)
             av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
-                   (flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"));
+                   (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) ? "MMX2" :
+                   ((HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C"));
         else if (dstFormat==PIX_FMT_RGB32)
-            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n",
+                   (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
         else if (dstFormat==PIX_FMT_BGR565)
-            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n",
+                   (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
         else if (dstFormat==PIX_FMT_BGR555)
-            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n",
+                   (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
         else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
                  dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE)
-            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C");
+            av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n",
+                   (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? "MMX" : "C");
 
         av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
         av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
@@ -1504,7 +1494,7 @@ void sws_freeContext(SwsContext *c)
     av_freep(&c->hLumFilterPos);
     av_freep(&c->hChrFilterPos);
 
-#if ARCH_X86
+#if HAVE_MMX
 #ifdef MAP_ANONYMOUS
     if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
     if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
@@ -1517,7 +1507,7 @@ void sws_freeContext(SwsContext *c)
 #endif
     c->lumMmx2FilterCode=NULL;
     c->chrMmx2FilterCode=NULL;
-#endif /* ARCH_X86 */
+#endif /* HAVE_MMX */
 
     av_freep(&c->yuvTable);
 
@@ -1534,8 +1524,6 @@ struct SwsContext *sws_getCachedContext(struct SwsContext *context,
     if (!param)
         param = default_param;
 
-    flags = update_flags_cpu(flags);
-
     if (context &&
         (context->srcW      != srcW      ||
          context->srcH      != srcH      ||
diff --git a/libswscale/x86/rgb2rgb.c b/libswscale/x86/rgb2rgb.c
index e84bc1bcc9..cf901affe6 100644
--- a/libswscale/x86/rgb2rgb.c
+++ b/libswscale/x86/rgb2rgb.c
@@ -27,6 +27,7 @@
 
 #include "config.h"
 #include "libavutil/x86_cpu.h"
+#include "libavutil/cpu.h"
 #include "libavutil/bswap.h"
 #include "libswscale/rgb2rgb.h"
 #include "libswscale/swscale.h"
@@ -122,16 +123,16 @@ DECLARE_ASM_CONST(8, uint64_t, blue_15mask)  = 0x0000001f0000001fULL;
  32-bit C version, and and&add trick by Michael Niedermayer
 */
 
-void rgb2rgb_init_x86(int flags)
+void rgb2rgb_init_x86(void)
 {
-#if HAVE_MMX2 || HAVE_AMD3DNOW || HAVE_MMX
-    if (flags & SWS_CPU_CAPS_SSE2)
-        rgb2rgb_init_SSE2();
-    else if (flags & SWS_CPU_CAPS_MMX2)
-        rgb2rgb_init_MMX2();
-    else if (flags & SWS_CPU_CAPS_3DNOW)
-        rgb2rgb_init_3DNOW();
-    else if (flags & SWS_CPU_CAPS_MMX)
+    int cpu_flags = av_get_cpu_flags();
+
+    if (HAVE_MMX      && cpu_flags & AV_CPU_FLAG_MMX)
         rgb2rgb_init_MMX();
-#endif /* HAVE_MMX2 || HAVE_AMD3DNOW || HAVE_MMX */
+    if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW)
+        rgb2rgb_init_3DNOW();
+    if (HAVE_MMX2     && cpu_flags & AV_CPU_FLAG_MMX2)
+        rgb2rgb_init_MMX2();
+    if (HAVE_SSE      && cpu_flags & AV_CPU_FLAG_SSE2)
+        rgb2rgb_init_SSE2();
 }
diff --git a/libswscale/x86/swscale_template.c b/libswscale/x86/swscale_template.c
index d719721693..e9e093780d 100644
--- a/libswscale/x86/swscale_template.c
+++ b/libswscale/x86/swscale_template.c
@@ -2721,10 +2721,11 @@ static int RENAME(swScale)(SwsContext *c, const uint8_t* src[], int srcStride[],
     if ((dstFormat == PIX_FMT_YUVA420P) && !alpPixBuf)
         fillPlane(dst[3], dstStride[3], dstW, dstY-lastDstY, lastDstY, 255);
 
-    if (flags & SWS_CPU_CAPS_MMX2 )  __asm__ volatile("sfence":::"memory");
+    if (COMPILE_TEMPLATE_MMX2)      __asm__ volatile("sfence":::"memory");
     /* On K6 femms is faster than emms. On K7 femms is directly mapped to emms. */
-    if (flags & SWS_CPU_CAPS_3DNOW)  __asm__ volatile("femms" :::"memory");
-    else                             __asm__ volatile("emms"  :::"memory");
+    if (COMPILE_TEMPLATE_AMD3DNOW)  __asm__ volatile("femms" :::"memory");
+    else                            __asm__ volatile("emms"  :::"memory");
+
     /* store changed local vars back in the context */
     c->dstY= dstY;
     c->lumBufIndex= lumBufIndex;
diff --git a/libswscale/x86/yuv2rgb_mmx.c b/libswscale/x86/yuv2rgb_mmx.c
index ff3a93db36..23d4c42700 100644
--- a/libswscale/x86/yuv2rgb_mmx.c
+++ b/libswscale/x86/yuv2rgb_mmx.c
@@ -34,6 +34,7 @@
 #include "libswscale/swscale.h"
 #include "libswscale/swscale_internal.h"
 #include "libavutil/x86_cpu.h"
+#include "libavutil/cpu.h"
 
 #define DITHER1XBPP // only for MMX
 
@@ -46,57 +47,58 @@ DECLARE_ASM_CONST(8, uint64_t, pb_03) = 0x0303030303030303ULL;
 DECLARE_ASM_CONST(8, uint64_t, pb_07) = 0x0707070707070707ULL;
 
 //MMX versions
+#if HAVE_MMX
 #undef RENAME
-#undef HAVE_MMX2
-#undef HAVE_AMD3DNOW
-#define HAVE_MMX2 0
-#define HAVE_AMD3DNOW 0
+#undef COMPILE_TEMPLATE_MMX2
+#define COMPILE_TEMPLATE_MMX2 0
 #define RENAME(a) a ## _MMX
 #include "yuv2rgb_template.c"
+#endif /* HAVE_MMX */
 
 //MMX2 versions
+#if HAVE_MMX2
 #undef RENAME
-#undef HAVE_MMX2
-#define HAVE_MMX2 1
+#undef COMPILE_TEMPLATE_MMX2
+#define COMPILE_TEMPLATE_MMX2 1
 #define RENAME(a) a ## _MMX2
 #include "yuv2rgb_template.c"
+#endif /* HAVE_MMX2 */
 
 SwsFunc ff_yuv2rgb_init_mmx(SwsContext *c)
 {
-    if (c->flags & SWS_CPU_CAPS_MMX2) {
+    int cpu_flags = av_get_cpu_flags();
+
+    if (c->srcFormat != PIX_FMT_YUV420P &&
+        c->srcFormat != PIX_FMT_YUVA420P)
+        return NULL;
+
+    if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) {
         switch (c->dstFormat) {
-        case PIX_FMT_RGB32:
-            if (CONFIG_SWSCALE_ALPHA && c->srcFormat == PIX_FMT_YUVA420P) {
-                if (HAVE_7REGS) return yuva420_rgb32_MMX2;
-                break;
-            } else return yuv420_rgb32_MMX2;
-        case PIX_FMT_BGR32:
-            if (CONFIG_SWSCALE_ALPHA && c->srcFormat == PIX_FMT_YUVA420P) {
-                if (HAVE_7REGS) return yuva420_bgr32_MMX2;
-                break;
-            } else return yuv420_bgr32_MMX2;
         case PIX_FMT_RGB24:  return yuv420_rgb24_MMX2;
         case PIX_FMT_BGR24:  return yuv420_bgr24_MMX2;
-        case PIX_FMT_RGB565: return yuv420_rgb16_MMX2;
-        case PIX_FMT_RGB555: return yuv420_rgb15_MMX2;
         }
     }
-    if (c->flags & SWS_CPU_CAPS_MMX) {
+
+    if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
         switch (c->dstFormat) {
-        case PIX_FMT_RGB32:
-            if (CONFIG_SWSCALE_ALPHA && c->srcFormat == PIX_FMT_YUVA420P) {
-                if (HAVE_7REGS) return yuva420_rgb32_MMX;
-                break;
-            } else return yuv420_rgb32_MMX;
-        case PIX_FMT_BGR32:
-            if (CONFIG_SWSCALE_ALPHA && c->srcFormat == PIX_FMT_YUVA420P) {
-                if (HAVE_7REGS) return yuva420_bgr32_MMX;
-                break;
-            } else return yuv420_bgr32_MMX;
-        case PIX_FMT_RGB24:  return yuv420_rgb24_MMX;
-        case PIX_FMT_BGR24:  return yuv420_bgr24_MMX;
-        case PIX_FMT_RGB565: return yuv420_rgb16_MMX;
-        case PIX_FMT_RGB555: return yuv420_rgb15_MMX;
+            case PIX_FMT_RGB32:
+                if (CONFIG_SWSCALE_ALPHA && c->srcFormat == PIX_FMT_YUVA420P) {
+#if HAVE_7REGS
+                    return yuva420_rgb32_MMX;
+#endif
+                    break;
+                } else return yuv420_rgb32_MMX;
+            case PIX_FMT_BGR32:
+                if (CONFIG_SWSCALE_ALPHA && c->srcFormat == PIX_FMT_YUVA420P) {
+#if HAVE_7REGS
+                    return yuva420_bgr32_MMX;
+#endif
+                    break;
+                } else return yuv420_bgr32_MMX;
+            case PIX_FMT_RGB24:  return yuv420_rgb24_MMX;
+            case PIX_FMT_BGR24:  return yuv420_bgr24_MMX;
+            case PIX_FMT_RGB565: return yuv420_rgb16_MMX;
+            case PIX_FMT_RGB555: return yuv420_rgb15_MMX;
         }
     }
 
diff --git a/libswscale/x86/yuv2rgb_template.c b/libswscale/x86/yuv2rgb_template.c
index cf8f0d3cfb..5d1fa5b309 100644
--- a/libswscale/x86/yuv2rgb_template.c
+++ b/libswscale/x86/yuv2rgb_template.c
@@ -25,14 +25,7 @@
 #undef EMMS
 #undef SFENCE
 
-#if HAVE_AMD3DNOW
-/* On K6 femms is faster than emms. On K7 femms is directly mapped to emms. */
-#define EMMS   "femms"
-#else
-#define EMMS   "emms"
-#endif
-
-#if HAVE_MMX2
+#if COMPILE_TEMPLATE_MMX2
 #define MOVNTQ "movntq"
 #define SFENCE "sfence"
 #else
@@ -159,7 +152,8 @@
     }                                                             \
 
 #define YUV2RGB_ENDFUNC                          \
-    __asm__ volatile (SFENCE"\n\t"EMMS);         \
+    __asm__ volatile (SFENCE"\n\t"               \
+                    "emms    \n\t");             \
     return srcSliceH;                            \
 
 #define IF0(x)
@@ -188,6 +182,7 @@
     "paddusb "GREEN_DITHER"(%4), %%mm2\n\t"      \
     "paddusb "RED_DITHER"(%4),   %%mm1\n\t"      \
 
+#if !COMPILE_TEMPLATE_MMX2
 static inline int RENAME(yuv420_rgb15)(SwsContext *c, const uint8_t *src[],
                                        int srcStride[],
                                        int srcSliceY, int srcSliceH,
@@ -243,6 +238,7 @@ static inline int RENAME(yuv420_rgb16)(SwsContext *c, const uint8_t *src[],
     YUV2RGB_OPERANDS
     YUV2RGB_ENDFUNC
 }
+#endif /* !COMPILE_TEMPLATE_MMX2 */
 
 #define RGB_PACK24(blue, red)\
     "packuswb  %%mm3,      %%mm0 \n" /* R0 R2 R4 R6 R1 R3 R5 R7 */\
@@ -259,7 +255,7 @@ static inline int RENAME(yuv420_rgb16)(SwsContext *c, const uint8_t *src[],
     "punpckhwd %%mm6,      %%mm5 \n" /* R4 G4 B4 R5 R6 G6 B6 R7 */\
     RGB_PACK24_B
 
-#if HAVE_MMX2
+#if COMPILE_TEMPLATE_MMX2
 DECLARE_ASM_CONST(8, int16_t, mask1101[4]) = {-1,-1, 0,-1};
 DECLARE_ASM_CONST(8, int16_t, mask0010[4]) = { 0, 0,-1, 0};
 DECLARE_ASM_CONST(8, int16_t, mask0110[4]) = { 0,-1,-1, 0};
@@ -366,6 +362,7 @@ static inline int RENAME(yuv420_bgr24)(SwsContext *c, const uint8_t *src[],
     MOVNTQ "   %%mm5,       16(%1)\n\t"      \
     MOVNTQ "   %%mm"alpha", 24(%1)\n\t"      \
 
+#if !COMPILE_TEMPLATE_MMX2
 static inline int RENAME(yuv420_rgb32)(SwsContext *c, const uint8_t *src[],
                                        int srcStride[],
                                        int srcSliceY, int srcSliceH,
@@ -386,12 +383,12 @@ static inline int RENAME(yuv420_rgb32)(SwsContext *c, const uint8_t *src[],
     YUV2RGB_ENDFUNC
 }
 
+#if HAVE_7REGS && CONFIG_SWSCALE_ALPHA
 static inline int RENAME(yuva420_rgb32)(SwsContext *c, const uint8_t *src[],
                                         int srcStride[],
                                         int srcSliceY, int srcSliceH,
                                         uint8_t *dst[], int dstStride[])
 {
-#if HAVE_7REGS
     int y, h_size;
 
     YUV2RGB_LOOP(4)
@@ -406,10 +403,8 @@ static inline int RENAME(yuva420_rgb32)(SwsContext *c, const uint8_t *src[],
     YUV2RGB_ENDLOOP(4)
     YUV2RGB_OPERANDS_ALPHA
     YUV2RGB_ENDFUNC
-#else
-    return 0;
-#endif
 }
+#endif
 
 static inline int RENAME(yuv420_bgr32)(SwsContext *c, const uint8_t *src[],
                                        int srcStride[],
@@ -431,12 +426,12 @@ static inline int RENAME(yuv420_bgr32)(SwsContext *c, const uint8_t *src[],
     YUV2RGB_ENDFUNC
 }
 
+#if HAVE_7REGS && CONFIG_SWSCALE_ALPHA
 static inline int RENAME(yuva420_bgr32)(SwsContext *c, const uint8_t *src[],
                                         int srcStride[],
                                         int srcSliceY, int srcSliceH,
                                         uint8_t *dst[], int dstStride[])
 {
-#if HAVE_7REGS
     int y, h_size;
 
     YUV2RGB_LOOP(4)
@@ -451,7 +446,7 @@ static inline int RENAME(yuva420_bgr32)(SwsContext *c, const uint8_t *src[],
     YUV2RGB_ENDLOOP(4)
     YUV2RGB_OPERANDS_ALPHA
     YUV2RGB_ENDFUNC
-#else
-    return 0;
-#endif
 }
+#endif
+
+#endif /* !COMPILE_TEMPLATE_MMX2 */
diff --git a/libswscale/yuv2rgb.c b/libswscale/yuv2rgb.c
index 74262c6b7f..a502f654ed 100644
--- a/libswscale/yuv2rgb.c
+++ b/libswscale/yuv2rgb.c
@@ -32,7 +32,7 @@
 #include "rgb2rgb.h"
 #include "swscale.h"
 #include "swscale_internal.h"
-#include "libavutil/x86_cpu.h"
+#include "libavutil/cpu.h"
 #include "libavutil/bswap.h"
 
 extern const uint8_t dither_4x4_16[4][8];
@@ -579,24 +579,18 @@ CLOSEYUV2RGBFUNC(1)
 SwsFunc ff_yuv2rgb_get_func_ptr(SwsContext *c)
 {
     SwsFunc t = NULL;
-#if HAVE_MMX
-     t = ff_yuv2rgb_init_mmx(c);
-#endif
-#if HAVE_VIS
-    t = ff_yuv2rgb_init_vis(c);
-#endif
-#if CONFIG_MLIB
-    t = ff_yuv2rgb_init_mlib(c);
-#endif
-#if HAVE_ALTIVEC
-    if (c->flags & SWS_CPU_CAPS_ALTIVEC)
-        t = ff_yuv2rgb_init_altivec(c);
-#endif
 
-#if ARCH_BFIN
-    if (c->flags & SWS_CPU_CAPS_BFIN)
+    if (HAVE_MMX) {
+        t = ff_yuv2rgb_init_mmx(c);
+    } else if (HAVE_VIS) {
+        t = ff_yuv2rgb_init_vis(c);
+    } else if (CONFIG_MLIB) {
+        t = ff_yuv2rgb_init_mlib(c);
+    } else if (HAVE_ALTIVEC) {
+        t = ff_yuv2rgb_init_altivec(c);
+    } else if (ARCH_BFIN) {
         t = ff_yuv2rgb_get_func_ptr_bfin(c);
-#endif
+    }
 
     if (t)
         return t;