From 0c742d59e58dd4c7a5a608e5d243d42319a7357a Mon Sep 17 00:00:00 2001
From: rdb <git@rdb.name>
Date: Sun, 25 Dec 2016 22:03:05 +0100
Subject: [PATCH] Fix crash due to incorrect alignment when building Eigen with
 AVX extensions Consequentially, we now use 32-byte alignment when building
 with eigen if __AVX__ is set.

---
 doc/ReleaseNotes                         |  1 +
 dtool/src/dtoolbase/deletedBufferChain.h |  4 ++++
 dtool/src/dtoolbase/dlmalloc_src.cxx     |  5 +++++
 dtool/src/dtoolbase/dtoolbase.h          |  2 +-
 dtool/src/dtoolbase/memoryHook.I         | 12 +++++++++++-
 5 files changed, 22 insertions(+), 2 deletions(-)

diff --git a/doc/ReleaseNotes b/doc/ReleaseNotes
index cd6bdf2f4c..74a0a9ea05 100644
--- a/doc/ReleaseNotes
+++ b/doc/ReleaseNotes
@@ -53,6 +53,7 @@ This issue fixes several bugs that were still found in 1.9.2.
 * Fix back-to-front sorting when gl-coordinate-system is changed
 * Now also compiles on older Linux distros (eg. CentOS 5 / manylinux1)
 * get_keyboard_map now includes keys on layouts with special characters
+* Fix crash due to incorrect alignment when compiling Eigen with AVX
 
 ------------------------  RELEASE 1.9.2  ------------------------
 
diff --git a/dtool/src/dtoolbase/deletedBufferChain.h b/dtool/src/dtoolbase/deletedBufferChain.h
index 301ee5d297..a71bc90b45 100644
--- a/dtool/src/dtoolbase/deletedBufferChain.h
+++ b/dtool/src/dtoolbase/deletedBufferChain.h
@@ -105,7 +105,11 @@ private:
 
 #elif defined(LINMATH_ALIGN)
   // With SSE2 alignment, we need all 16 bytes to preserve alignment.
+#ifdef __AVX__
+  static const size_t flag_reserved_bytes = 32;
+#else
   static const size_t flag_reserved_bytes = 16;
+#endif
 
 #else
   // Otherwise, we only need enough space for the Integer itself.
diff --git a/dtool/src/dtoolbase/dlmalloc_src.cxx b/dtool/src/dtoolbase/dlmalloc_src.cxx
index 4e21f4c914..dcbb0daa25 100644
--- a/dtool/src/dtoolbase/dlmalloc_src.cxx
+++ b/dtool/src/dtoolbase/dlmalloc_src.cxx
@@ -453,8 +453,13 @@ DEFAULT_MMAP_THRESHOLD       default: 256K
 // drose: We require 16-byte alignment of certain structures, to
 // support SSE2.  We don't strictly have to align *everything*, but
 // it's just easier to do so.
+#ifdef __AVX__
+// Eigen requires 32-byte alignment when using AVX instructions.
+#define MALLOC_ALIGNMENT ((size_t)32U)
+#else
 #define MALLOC_ALIGNMENT ((size_t)16U)
 #endif
+#endif
 
 #ifndef WIN32
 #ifdef _WIN32
diff --git a/dtool/src/dtoolbase/dtoolbase.h b/dtool/src/dtoolbase/dtoolbase.h
index 2c883f8220..27add7c84d 100644
--- a/dtool/src/dtoolbase/dtoolbase.h
+++ b/dtool/src/dtoolbase/dtoolbase.h
@@ -372,7 +372,7 @@
 // enforce alignment externally.
 #define MEMORY_HOOK_DO_ALIGN 1
 
-#elif defined(IS_OSX) || defined(_WIN64)
+#elif (defined(IS_OSX) || defined(_WIN64)) && !defined(__AVX__)
 // The OS-provided malloc implementation will do the required
 // alignment.
 #undef MEMORY_HOOK_DO_ALIGN
diff --git a/dtool/src/dtoolbase/memoryHook.I b/dtool/src/dtoolbase/memoryHook.I
index 9286a18d7a..ecccf07d65 100644
--- a/dtool/src/dtoolbase/memoryHook.I
+++ b/dtool/src/dtoolbase/memoryHook.I
@@ -55,7 +55,12 @@ get_memory_alignment() {
   // We require 16-byte alignment of certain structures, to support
   // SSE2.  We don't strictly have to align *everything*, but it's just
   // easier to do so.
+#ifdef __AVX__
+  // Eigen requires 32-byte alignment when using AVX instructions.
+  const size_t alignment_size = 32;
+#else
   const size_t alignment_size = 16;
+#endif
 #else
   // Otherwise, use word alignment.
   const size_t alignment_size = sizeof(void *);
@@ -79,7 +84,12 @@ get_header_reserved_bytes() {
 #ifdef LINMATH_ALIGN
   // If we're doing SSE2 alignment, we must reserve a full 16-byte
   // block, since anything less than that will spoil the alignment.
-  static const size_t header_reserved_bytes = 16;
+#ifdef __AVX__
+  // Eigen requires 32-byte alignment when using AVX instructions.
+  const size_t header_reserved_bytes = 32;
+#else
+  const size_t header_reserved_bytes = 16;
+#endif
 
 #elif defined(MEMORY_HOOK_DO_ALIGN)
   // If we're just aligning to words, we reserve a block as big as two