[FFmpeg-cvslog] Add AVX2 capable CPU detection. Patch based on x264's AVX2 detection

[Kieran Kunhya]

ffmpeg | branch: master | Kieran Kunhya <kierank at ob-encoder.com> | Sun Oct 20 10:28:38 2013 -0500| [865b70bc5d1cf37ec6d6cb729a69dda2cca28bd5] | committer: Michael Niedermayer

Add AVX2 capable CPU detection. Patch based on x264's AVX2 detection

Signed-off-by: Michael Niedermayer <michaelni at gmx.at>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=865b70bc5d1cf37ec6d6cb729a69dda2cca28bd5
---

 configure           |    3 +++
 libavutil/cpu.c     |    3 +++
 libavutil/cpu.h     |    2 ++
 libavutil/x86/cpu.c |    7 +++++++
 libavutil/x86/cpu.h |    3 +++
 5 files changed, 18 insertions(+)

diff --git a/configure b/configure
index e382a86..dce43e8 100755
--- a/configure
+++ b/configure
@@ -315,6 +315,7 @@ Optimization options (experts only):
   --disable-sse42          disable SSE4.2 optimizations
   --disable-avx            disable AVX optimizations
   --disable-fma4           disable FMA4 optimizations
+  --disable-avx2           disable AVX2 optimizations
   --disable-armv5te        disable armv5te optimizations
   --disable-armv6          disable armv6 optimizations
   --disable-armv6t2        disable armv6t2 optimizations
@@ -1344,6 +1345,7 @@ ARCH_EXT_LIST_X86='
     amd3dnow
     amd3dnowext
     avx
+    avx2
     fma4
     i686
     mmx
@@ -1705,6 +1707,7 @@ sse4_deps="ssse3"
 sse42_deps="sse4"
 avx_deps="sse42"
 fma4_deps="avx"
+avx2_deps="avx"
 
 mmx_external_deps="yasm"
 mmx_inline_deps="inline_asm"
diff --git a/libavutil/cpu.c b/libavutil/cpu.c
index a31e195..deb2a87 100644
--- a/libavutil/cpu.c
+++ b/libavutil/cpu.c
@@ -85,6 +85,7 @@ int av_parse_cpu_flags(const char *s)
 #define CPUFLAG_AVX      (AV_CPU_FLAG_AVX      | CPUFLAG_SSE42)
 #define CPUFLAG_XOP      (AV_CPU_FLAG_XOP      | CPUFLAG_AVX)
 #define CPUFLAG_FMA4     (AV_CPU_FLAG_FMA4     | CPUFLAG_AVX)
+#define CPUFLAG_AVX2     (AV_CPU_FLAG_AVX2     | CPUFLAG_AVX)
     static const AVOption cpuflags_opts[] = {
         { "flags"   , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
 #if   ARCH_PPC
@@ -104,6 +105,7 @@ int av_parse_cpu_flags(const char *s)
         { "avx"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX          },    .unit = "flags" },
         { "xop"     , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_XOP          },    .unit = "flags" },
         { "fma4"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA4         },    .unit = "flags" },
+        { "avx2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_AVX2         },    .unit = "flags" },
         { "3dnow"   , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOW        },    .unit = "flags" },
         { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_3DNOWEXT     },    .unit = "flags" },
         { "cmov",     NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV     },    .unit = "flags" },
@@ -267,6 +269,7 @@ static const struct {
     { AV_CPU_FLAG_3DNOW,     "3dnow"      },
     { AV_CPU_FLAG_3DNOWEXT,  "3dnowext"   },
     { AV_CPU_FLAG_CMOV,      "cmov"       },
+    { AV_CPU_FLAG_AVX2,       "avx2"       },
 #endif
     { 0 }
 };
diff --git a/libavutil/cpu.h b/libavutil/cpu.h
index 50bdea7..e9185f1 100644
--- a/libavutil/cpu.h
+++ b/libavutil/cpu.h
@@ -50,6 +50,8 @@
 // #else
 // #define AV_CPU_FLAG_CMOV         0x1000 ///< supports cmov instruction
 // #endif
+#define AV_CPU_FLAG_AVX2         0x8000 ///< AVX2 functions: requires OS support even if YMM registers aren't used
+
 
 #define AV_CPU_FLAG_ALTIVEC      0x0001 ///< standard
 
diff --git a/libavutil/x86/cpu.c b/libavutil/x86/cpu.c
index 81bb15a..174515f 100644
--- a/libavutil/x86/cpu.c
+++ b/libavutil/x86/cpu.c
@@ -134,6 +134,13 @@ int ff_get_cpu_flags_x86(void)
             if ((eax & 0x6) == 0x6)
                 rval |= AV_CPU_FLAG_AVX;
         }
+    if (HAVE_AVX2 && max_std_level >= 7)
+    {
+        cpuid(7, eax, ebx, ecx, edx);
+        if (ebx&0x00000020)
+            rval |= AV_CPU_FLAG_AVX2;
+        /* TODO: BMI1/2 */
+    }
 #endif /* HAVE_AVX */
 #endif /* HAVE_SSE */
     }
diff --git a/libavutil/x86/cpu.h b/libavutil/x86/cpu.h
index 738c1ec..3724357 100644
--- a/libavutil/x86/cpu.h
+++ b/libavutil/x86/cpu.h
@@ -38,6 +38,7 @@
 #define X86_SSE42(flags)            CPUEXT(flags, SSE42)
 #define X86_AVX(flags)              CPUEXT(flags, AVX)
 #define X86_FMA4(flags)             CPUEXT(flags, FMA4)
+#define X86_AVX2(flags)             CPUEXT(flags, AVX2)
 
 #define EXTERNAL_AMD3DNOW(flags)    CPUEXT_SUFFIX(flags, _EXTERNAL, AMD3DNOW)
 #define EXTERNAL_AMD3DNOWEXT(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, AMD3DNOWEXT)
@@ -51,6 +52,7 @@
 #define EXTERNAL_SSE42(flags)       CPUEXT_SUFFIX(flags, _EXTERNAL, SSE42)
 #define EXTERNAL_AVX(flags)         CPUEXT_SUFFIX(flags, _EXTERNAL, AVX)
 #define EXTERNAL_FMA4(flags)        CPUEXT_SUFFIX(flags, _EXTERNAL, FMA4)
+#define EXTERNAL_AVX2(flags)        CPUEXT_SUFFIX(flags, _EXTERNAL, AVX2)
 
 #define INLINE_AMD3DNOW(flags)      CPUEXT_SUFFIX(flags, _INLINE, AMD3DNOW)
 #define INLINE_AMD3DNOWEXT(flags)   CPUEXT_SUFFIX(flags, _INLINE, AMD3DNOWEXT)
@@ -64,6 +66,7 @@
 #define INLINE_SSE42(flags)         CPUEXT_SUFFIX(flags, _INLINE, SSE42)
 #define INLINE_AVX(flags)           CPUEXT_SUFFIX(flags, _INLINE, AVX)
 #define INLINE_FMA4(flags)          CPUEXT_SUFFIX(flags, _INLINE, FMA4)
+#define INLINE_AVX2(flags)          CPUEXT_SUFFIX(flags, _INLINE, AVX2)
 
 void ff_cpu_cpuid(int index, int *eax, int *ebx, int *ecx, int *edx);
 void ff_cpu_xgetbv(int op, int *eax, int *edx);