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

4 files changed, 64 insertions, 65 deletions

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 */