[vlc-commits] packetizer/startcode_helper: enhance with AVX2

[Lyndon Brown]

vlc | branch: master | Lyndon Brown <jnqnfe at gmail.com> | Fri Jan 25 06:05:54 2019 +0000| [c29fcde3887e4fbccc6af4a860a59febfe837f4d] | committer: Jean-Baptiste Kempf

packetizer/startcode_helper: enhance with AVX2

Signed-off-by: Jean-Baptiste Kempf <jb at videolan.org>

> http://git.videolan.org/gitweb.cgi/vlc.git/?a=commit;h=c29fcde3887e4fbccc6af4a860a59febfe837f4d
---

 modules/packetizer/startcode_helper.h | 94 +++++++++++++++++++++++++++++++++--
 1 file changed, 89 insertions(+), 5 deletions(-)

diff --git a/modules/packetizer/startcode_helper.h b/modules/packetizer/startcode_helper.h
index 2b61e5cf98..3391c68db3 100644
--- a/modules/packetizer/startcode_helper.h
+++ b/modules/packetizer/startcode_helper.h
@@ -1,7 +1,7 @@
 /*****************************************************************************
  * startcode_helper.h: Startcodes helpers
  *****************************************************************************
- * Copyright (C) 2016 VideoLAN Authors
+ * Copyright (C) 2016, 2019 VideoLAN Authors
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms of the GNU Lesser General Public License as published by
@@ -22,6 +22,9 @@
 
 #include <vlc_cpu.h>
 
+#if !defined(CAN_COMPILE_AVX2) && defined(HAVE_AVX2_INTRINSICS)
+   #include <immintrin.h>
+#endif
 #if !defined(CAN_COMPILE_SSE2) && defined(HAVE_SSE2_INTRINSICS)
    #include <emmintrin.h>
 #endif
@@ -44,15 +47,91 @@
         }\
     }
 
+#if defined(CAN_COMPILE_AVX2) || defined(HAVE_AVX2_INTRINSICS)
+
+__attribute__ ((__target__ ("avx2")))
+static inline const uint8_t * startcode_FindAnnexB_AVX2( const uint8_t *p, const uint8_t *end )
+{
+    end -= 3;
+
+    /* First align to 32 */
+    const uint8_t *alignedend = p + 32 - ((intptr_t)p & 31);
+    for (; p < alignedend && p <= end; p++) {
+        if (p[0] == 0 && p[1] == 0 && p[2] == 1)
+            return p;
+    }
+
+    if( p == end )
+        return NULL;
+
+    alignedend = end - ((intptr_t) end & 31);
+    if( alignedend > p )
+    {
+#ifdef CAN_COMPILE_AVX2
+        asm volatile(
+            "vpxor   %%ymm1, %%ymm1\n"
+            ::: "ymm1"
+        );
+#else
+        __m256i zeros = _mm256_set1_epi8( 0x00 );
+#endif
+        for( ; p < alignedend; p += 32)
+        {
+            uint32_t match;
+#ifdef CAN_COMPILE_AVX2
+            asm volatile(
+                "vmovdqa   0(%[v]),   %%ymm0\n"
+                "vpcmpeqb   %%ymm1,   %%ymm0\n"
+                "vpmovmskb  %%ymm0,   %[match]\n"
+                : [match]"=r"(match)
+                : [v]"r"(p)
+                : "ymm0"
+            );
+#else
+            __m256i v = _mm256_load_si256((__m256i*)p);
+            __m256i res = _mm256_cmpeq_epi8( zeros, v );
+            match = _mm256_movemask_epi8( res ); /* mask will be in reversed match order */
+#endif
+            if( match & 0x0000000F )
+                TRY_MATCH(p, 0);
+            if( match & 0x000000F0 )
+                TRY_MATCH(p, 4);
+            if( match & 0x00000F00 )
+                TRY_MATCH(p, 8);
+            if( match & 0x0000F000 )
+                TRY_MATCH(p, 12);
+            if( match & 0x000F0000 )
+                TRY_MATCH(p, 16);
+            if( match & 0x00F00000 )
+                TRY_MATCH(p, 20);
+            if( match & 0x0F000000 )
+                TRY_MATCH(p, 24);
+            if( match & 0xF0000000 )
+                TRY_MATCH(p, 28);
+        }
+    }
+
+    for (; p <= end; p++) {
+        if (p[0] == 0 && p[1] == 0 && p[2] == 1)
+            return p;
+    }
+
+    return NULL;
+}
+
+#endif
+
 #if defined(CAN_COMPILE_SSE2) || defined(HAVE_SSE2_INTRINSICS)
 
 __attribute__ ((__target__ ("sse2")))
 static inline const uint8_t * startcode_FindAnnexB_SSE2( const uint8_t *p, const uint8_t *end )
 {
+    end -= 3;
+
     /* First align to 16 */
     /* Skipping this step and doing unaligned loads isn't faster */
     const uint8_t *alignedend = p + 16 - ((intptr_t)p & 15);
-    for (end -= 3; p < alignedend && p <= end; p++) {
+    for (; p < alignedend && p <= end; p++) {
         if (p[0] == 0 && p[1] == 0 && p[2] == 1)
             return p;
     }
@@ -140,13 +219,18 @@ static inline const uint8_t * startcode_FindAnnexB_Bits( const uint8_t *p, const
 }
 #undef TRY_MATCH
 
-#if defined(CAN_COMPILE_SSE2) || defined(HAVE_SSE2_INTRINSICS)
+#if defined(CAN_COMPILE_AVX2) || defined(HAVE_AVX2_INTRINSICS) || defined(CAN_COMPILE_SSE2) || defined(HAVE_SSE2_INTRINSICS)
 static inline const uint8_t * startcode_FindAnnexB( const uint8_t *p, const uint8_t *end )
 {
+#if defined(CAN_COMPILE_AVX2) || defined(HAVE_AVX2_INTRINSICS)
+    if (vlc_CPU_AVX2())
+        return startcode_FindAnnexB_AVX2(p, end);
+#endif
+#if defined(CAN_COMPILE_SSE2) || defined(HAVE_SSE2_INTRINSICS)
     if (vlc_CPU_SSE2())
         return startcode_FindAnnexB_SSE2(p, end);
-    else
-        return startcode_FindAnnexB_Bits(p, end);
+#endif
+    return startcode_FindAnnexB_Bits(p, end);
 }
 #else
     #define startcode_FindAnnexB startcode_FindAnnexB_Bits