From 0bc3de19ffe296254f214dc7615e624d8e401bcb Mon Sep 17 00:00:00 2001
From: James Almer <jamrial@gmail.com>
Date: Sat, 22 Feb 2014 02:47:02 -0300
Subject: [PATCH] x86: add detection for Bit Manipulation Instruction sets

Based on x264 code

Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
---
 libavutil/cpu.c     |  8 ++++++++
 libavutil/cpu.h     |  2 ++
 libavutil/x86/cpu.c | 22 ++++++++++++++--------
 3 files changed, 24 insertions(+), 8 deletions(-)

diff --git a/libavutil/cpu.c b/libavutil/cpu.c
index 74de61e0b7..256bd237d5 100644
--- a/libavutil/cpu.c
+++ b/libavutil/cpu.c
@@ -94,6 +94,8 @@ int av_parse_cpu_flags(const char *s)
 #define CPUFLAG_FMA3     (AV_CPU_FLAG_FMA3     | CPUFLAG_AVX)
 #define CPUFLAG_FMA4     (AV_CPU_FLAG_FMA4     | CPUFLAG_AVX)
 #define CPUFLAG_AVX2     (AV_CPU_FLAG_AVX2     | CPUFLAG_AVX)
+#define CPUFLAG_BMI1     (AV_CPU_FLAG_BMI1)
+#define CPUFLAG_BMI2     (AV_CPU_FLAG_BMI2     | CPUFLAG_BMI1)
     static const AVOption cpuflags_opts[] = {
         { "flags"   , NULL, 0, AV_OPT_TYPE_FLAGS, { .i64 = 0 }, INT64_MIN, INT64_MAX, .unit = "flags" },
 #if   ARCH_PPC
@@ -115,6 +117,8 @@ int av_parse_cpu_flags(const char *s)
         { "fma3"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_FMA3         },    .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" },
+        { "bmi1"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI1         },    .unit = "flags" },
+        { "bmi2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = CPUFLAG_BMI2         },    .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" },
@@ -171,6 +175,8 @@ int av_parse_cpu_caps(unsigned *flags, const char *s)
         { "fma3"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA3     },    .unit = "flags" },
         { "fma4"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_FMA4     },    .unit = "flags" },
         { "avx2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_AVX2     },    .unit = "flags" },
+        { "bmi1"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI1     },    .unit = "flags" },
+        { "bmi2"    , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_BMI2     },    .unit = "flags" },
         { "3dnow"   , NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOW    },    .unit = "flags" },
         { "3dnowext", NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_3DNOWEXT },    .unit = "flags" },
         { "cmov",     NULL, 0, AV_OPT_TYPE_CONST, { .i64 = AV_CPU_FLAG_CMOV     },    .unit = "flags" },
@@ -288,6 +294,8 @@ static const struct {
     { AV_CPU_FLAG_3DNOWEXT,  "3dnowext"   },
     { AV_CPU_FLAG_CMOV,      "cmov"       },
     { AV_CPU_FLAG_AVX2,      "avx2"       },
+    { AV_CPU_FLAG_BMI1,      "bmi1"       },
+    { AV_CPU_FLAG_BMI2,      "bmi2"       },
 #endif
     { 0 }
 };
diff --git a/libavutil/cpu.h b/libavutil/cpu.h
index 1d0293fed3..0ad400fefb 100644
--- a/libavutil/cpu.h
+++ b/libavutil/cpu.h
@@ -52,6 +52,8 @@
 // #endif
 #define AV_CPU_FLAG_AVX2         0x8000 ///< AVX2 functions: requires OS support even if YMM registers aren't used
 #define AV_CPU_FLAG_FMA3        0x10000 ///< Haswell FMA3 functions
+#define AV_CPU_FLAG_BMI1        0x20000 ///< Bit Manipulation Instruction Set 1
+#define AV_CPU_FLAG_BMI2        0x40000 ///< Bit Manipulation Instruction Set 2
 
 #define AV_CPU_FLAG_ALTIVEC      0x0001 ///< standard
 
diff --git a/libavutil/x86/cpu.c b/libavutil/x86/cpu.c
index 333b0f805f..e0b4c89062 100644
--- a/libavutil/x86/cpu.c
+++ b/libavutil/x86/cpu.c
@@ -137,16 +137,22 @@ int ff_get_cpu_flags_x86(void)
                     rval |= AV_CPU_FLAG_FMA3;
             }
         }
-#if HAVE_AVX2
-    if (max_std_level >= 7) {
-        cpuid(7, eax, ebx, ecx, edx);
-        if (ebx&0x00000020)
-            rval |= AV_CPU_FLAG_AVX2;
-        /* TODO: BMI1/2 */
-    }
-#endif /* HAVE_AVX2 */
 #endif /* HAVE_AVX */
 #endif /* HAVE_SSE */
+    if (max_std_level >= 7) {
+        cpuid(7, eax, ebx, ecx, edx);
+#if HAVE_AVX2
+        if (ebx&0x00000020)
+            rval |= AV_CPU_FLAG_AVX2;
+#endif /* HAVE_AVX2 */
+        /* BMI1/2 don't need OS support */
+        if (ebx&0x00000008)
+        {
+            rval |= AV_CPU_FLAG_BMI1;
+            if (ebx&0x00000100)
+                rval |= AV_CPU_FLAG_BMI2;
+        }
+    }
     }
 
     cpuid(0x80000000, max_ext_level, ebx, ecx, edx);