From 092309b26687a033e6d95de5cdf065b3ef8cb584 Mon Sep 17 00:00:00 2001
From: Jeffrey Walton <noloader@gmail.com>
Date: Fri, 9 Nov 2018 08:00:53 -0500
Subject: [PATCH] Fix global optimization bug for ChaCha AVX2 under VS2017 (GH
 #735) Also see https://github.com/weidai11/cryptopp/issues/649. The 649 issue
 is the one affecting AES. It appears to be the same problem.

---
 chacha-avx.cpp | 21 ++++++++++++++++++---
 rijndael.cpp   |  6 +++---
 2 files changed, 21 insertions(+), 6 deletions(-)

diff --git a/chacha-avx.cpp b/chacha-avx.cpp
index 6c875c5fe..0aba0a996 100644
--- a/chacha-avx.cpp
+++ b/chacha-avx.cpp
@@ -36,10 +36,21 @@ extern const char CHACHA_AVX_FNAME[] = __FILE__;
 # define MAYBE_CONST const
 #endif
 
-#if (CRYPTOPP_AVX2_AVAILABLE)
+// VS2017 and global optimization bug. TODO, figure out when
+// we can re-enable full optimizations for VS2017. Also see
+// https://github.com/weidai11/cryptopp/issues/649 and
+// https://github.com/weidai11/cryptopp/issues/735. The
+// 649 issue affects AES but it is the same here. The 735
+// issue is ChaCha AVX2 cut-in where it surfaced again.
+#if (_MSC_VER >= 1910) && defined(NDEBUG)
+# pragma optimize("", off)
+# pragma optimize("ts", on)
+#endif
 
 ANONYMOUS_NAMESPACE_BEGIN
 
+#if (CRYPTOPP_AVX2_AVAILABLE)
+
 template <unsigned int R>
 inline __m256i RotateLeft(const __m256i val)
 {
@@ -62,10 +73,14 @@ inline __m256i RotateLeft<16>(const __m256i val)
     return _mm256_shuffle_epi8(val, mask);
 }
 
+#endif CRYPTOPP_AVX2_AVAILABLE
+
 ANONYMOUS_NAMESPACE_END
 
 NAMESPACE_BEGIN(CryptoPP)
 
+#if (CRYPTOPP_AVX2_AVAILABLE)
+
 void ChaCha_OperateKeystream_AVX2(const word32 *state, const byte* input, byte *output, unsigned int rounds)
 {
     MAYBE_CONST __m128i* state_mm = (MAYBE_CONST __m128i*)(state);
@@ -358,6 +373,6 @@ void ChaCha_OperateKeystream_AVX2(const word32 *state, const byte* input, byte *
     }
 }
 
-NAMESPACE_END
-
 #endif  // CRYPTOPP_AVX2_AVAILABLE
+
+NAMESPACE_END
diff --git a/rijndael.cpp b/rijndael.cpp
index f00cc2772..3cc0a46be 100644
--- a/rijndael.cpp
+++ b/rijndael.cpp
@@ -88,10 +88,10 @@ being unloaded from L1 cache, until that round is finished.
 #include "misc.h"
 #include "cpu.h"
 
-// MSVC bug, still don't know how to fix it. TODO, figure out
-// when we can re-enable optimizations for MSVC. Also see
+// VS2017 and global optimization bug. TODO, figure out when
+// we can re-enable full optimizations for VS2017. Also see
 // https://github.com/weidai11/cryptopp/issues/649
-#if defined(_MSC_VER) && (_MSC_VER >= 1910)
+#if (_MSC_VER >= 1910) && defined(NDEBUG)
 # pragma optimize("", off)
 # pragma optimize("ts", on)
 #endif