From: Steinar H. Gunderson <sgunderson@bigfoot.com>
Date: Tue, 20 Sep 2016 12:17:36 +0000 (+0200)
Subject: Explicitly declare both versions of add_to_frame_fastpath_core, to prevent GCC from... 
X-Git-Tag: 0.5.2~2
X-Git-Url: https://git.sesse.net/?p=bmusb;a=commitdiff_plain;h=69b4f1db4fc485f28d6265a6adacb9a6baaa095f

Explicitly declare both versions of add_to_frame_fastpath_core, to prevent GCC from inlining the wrong version.
---

diff --git a/bmusb.cpp b/bmusb.cpp
index 080c0bb..86779b4 100644
--- a/bmusb.cpp
+++ b/bmusb.cpp
@@ -536,6 +536,10 @@ const uint8_t *add_to_frame_fastpath(FrameAllocator::Frame *current_frame, const
 
 #else  // defined(HAS_MULTIVERSIONING)
 
+__attribute__((target("sse4.1")))
+const uint8_t *add_to_frame_fastpath_core(FrameAllocator::Frame *current_frame, const uint8_t *aligned_start, const uint8_t *limit, const char sync_char);
+
+__attribute__((target("avx2")))
 const uint8_t *add_to_frame_fastpath_core(FrameAllocator::Frame *current_frame, const uint8_t *aligned_start, const uint8_t *limit, const char sync_char);
 
 // Does a memcpy and memchr in one to reduce processing time.