diff --git a/libavutil/cpu_internal.h b/libavutil/cpu_internal.h
index 3c6ce6d079..2105298843 100644
--- a/libavutil/cpu_internal.h
+++ b/libavutil/cpu_internal.h
@@ -24,7 +24,17 @@
 #define CPUEXT_SUFFIX(flags, suffix, cpuext)                            \
     (HAVE_ ## cpuext ## suffix && ((flags) & AV_CPU_FLAG_ ## cpuext))
 
+#define CPUEXT_SUFFIX_FAST(flags, suffix, cpuext)                       \
+    (HAVE_ ## cpuext ## suffix && ((flags) & AV_CPU_FLAG_ ## cpuext) && \
+     !((flags) & AV_CPU_FLAG_ ## cpuext ## SLOW))
+
+#define CPUEXT_SUFFIX_SLOW(flags, suffix, cpuext)                       \
+    (HAVE_ ## cpuext ## suffix && ((flags) & AV_CPU_FLAG_ ## cpuext) && \
+     ((flags) & AV_CPU_FLAG_ ## cpuext ## SLOW))
+
 #define CPUEXT(flags, cpuext) CPUEXT_SUFFIX(flags, , cpuext)
+#define CPUEXT_FAST(flags, cpuext) CPUEXT_SUFFIX_FAST(flags, , cpuext)
+#define CPUEXT_SLOW(flags, cpuext) CPUEXT_SUFFIX_SLOW(flags, , cpuext)
 
 int ff_get_cpu_flags_aarch64(void);
 int ff_get_cpu_flags_arm(void);
diff --git a/libavutil/x86/cpu.h b/libavutil/x86/cpu.h
index bc64b1b3bd..1cea4198d0 100644
--- a/libavutil/x86/cpu.h
+++ b/libavutil/x86/cpu.h
@@ -32,11 +32,17 @@
 #define X86_MMXEXT(flags)           CPUEXT(flags, MMXEXT)
 #define X86_SSE(flags)              CPUEXT(flags, SSE)
 #define X86_SSE2(flags)             CPUEXT(flags, SSE2)
+#define X86_SSE2_FAST(flags)        CPUEXT_FAST(flags, SSE2)
+#define X86_SSE2_SLOW(flags)        CPUEXT_SLOW(flags, SSE2)
 #define X86_SSE3(flags)             CPUEXT(flags, SSE3)
+#define X86_SSE3_FAST(flags)        CPUEXT_FAST(flags, SSE3)
+#define X86_SSE3_SLOW(flags)        CPUEXT_SLOW(flags, SSE3)
 #define X86_SSSE3(flags)            CPUEXT(flags, SSSE3)
 #define X86_SSE4(flags)             CPUEXT(flags, SSE4)
 #define X86_SSE42(flags)            CPUEXT(flags, SSE42)
 #define X86_AVX(flags)              CPUEXT(flags, AVX)
+#define X86_AVX_FAST(flags)         CPUEXT_FAST(flags, AVX)
+#define X86_AVX_SLOW(flags)         CPUEXT_SLOW(flags, AVX)
 #define X86_XOP(flags)              CPUEXT(flags, XOP)
 #define X86_FMA3(flags)             CPUEXT(flags, FMA3)
 #define X86_FMA4(flags)             CPUEXT(flags, FMA4)
@@ -48,11 +54,17 @@
 #define EXTERNAL_MMXEXT(flags)      CPUEXT_SUFFIX(flags, _EXTERNAL, MMXEXT)
 #define EXTERNAL_SSE(flags)         CPUEXT_SUFFIX(flags, _EXTERNAL, SSE)
 #define EXTERNAL_SSE2(flags)        CPUEXT_SUFFIX(flags, _EXTERNAL, SSE2)
+#define EXTERNAL_SSE2_FAST(flags)   CPUEXT_SUFFIX_FAST(flags, _EXTERNAL, SSE2)
+#define EXTERNAL_SSE2_SLOW(flags)   CPUEXT_SUFFIX_SLOW(flags, _EXTERNAL, SSE2)
 #define EXTERNAL_SSE3(flags)        CPUEXT_SUFFIX(flags, _EXTERNAL, SSE3)
+#define EXTERNAL_SSE3_FAST(flags)   CPUEXT_SUFFIX_FAST(flags, _EXTERNAL, SSE3)
+#define EXTERNAL_SSE3_SLOW(flags)   CPUEXT_SUFFIX_SLOW(flags, _EXTERNAL, SSE3)
 #define EXTERNAL_SSSE3(flags)       CPUEXT_SUFFIX(flags, _EXTERNAL, SSSE3)
 #define EXTERNAL_SSE4(flags)        CPUEXT_SUFFIX(flags, _EXTERNAL, SSE4)
 #define EXTERNAL_SSE42(flags)       CPUEXT_SUFFIX(flags, _EXTERNAL, SSE42)
 #define EXTERNAL_AVX(flags)         CPUEXT_SUFFIX(flags, _EXTERNAL, AVX)
+#define EXTERNAL_AVX_FAST(flags)    CPUEXT_SUFFIX_FAST(flags, _EXTERNAL, AVX)
+#define EXTERNAL_AVX_SLOW(flags)    CPUEXT_SUFFIX_SLOW(flags, _EXTERNAL, AVX)
 #define EXTERNAL_XOP(flags)         CPUEXT_SUFFIX(flags, _EXTERNAL, XOP)
 #define EXTERNAL_FMA3(flags)        CPUEXT_SUFFIX(flags, _EXTERNAL, FMA3)
 #define EXTERNAL_FMA4(flags)        CPUEXT_SUFFIX(flags, _EXTERNAL, FMA4)
@@ -64,11 +76,17 @@
 #define INLINE_MMXEXT(flags)        CPUEXT_SUFFIX(flags, _INLINE, MMXEXT)
 #define INLINE_SSE(flags)           CPUEXT_SUFFIX(flags, _INLINE, SSE)
 #define INLINE_SSE2(flags)          CPUEXT_SUFFIX(flags, _INLINE, SSE2)
+#define INLINE_SSE2_FAST(flags)     CPUEXT_SUFFIX_FAST(flags, _INLINE, SSE2)
+#define INLINE_SSE2_SLOW(flags)     CPUEXT_SUFFIX_SLOW(flags, _INLINE, SSE2)
 #define INLINE_SSE3(flags)          CPUEXT_SUFFIX(flags, _INLINE, SSE3)
+#define INLINE_SSE3_FAST(flags)     CPUEXT_SUFFIX_FAST(flags, _INLINE, SSE3)
+#define INLINE_SSE3_SLOW(flags)     CPUEXT_SUFFIX_SLOW(flags, _INLINE, SSE3)
 #define INLINE_SSSE3(flags)         CPUEXT_SUFFIX(flags, _INLINE, SSSE3)
 #define INLINE_SSE4(flags)          CPUEXT_SUFFIX(flags, _INLINE, SSE4)
 #define INLINE_SSE42(flags)         CPUEXT_SUFFIX(flags, _INLINE, SSE42)
 #define INLINE_AVX(flags)           CPUEXT_SUFFIX(flags, _INLINE, AVX)
+#define INLINE_AVX_FAST(flags)      CPUEXT_SUFFIX_FAST(flags, _INLINE, AVX)
+#define INLINE_AVX_SLOW(flags)      CPUEXT_SUFFIX_SLOW(flags, _INLINE, AVX)
 #define INLINE_XOP(flags)           CPUEXT_SUFFIX(flags, _INLINE, XOP)
 #define INLINE_FMA3(flags)          CPUEXT_SUFFIX(flags, _INLINE, FMA3)
 #define INLINE_FMA4(flags)          CPUEXT_SUFFIX(flags, _INLINE, FMA4)