From b2a0262c5939eee9ff84642329ae8bbe5262f46f Mon Sep 17 00:00:00 2001
From: Stefan Weil <sw@weilnetz.de>
Date: Tue, 27 Dec 2016 10:21:50 +0100
Subject: [PATCH 1/2] lstm: Update AVX / SSE support

* Fix compiler warning (see below)

* Use Linux code for Mingw-w64, too

* Simplify conditional code by using X86_BUILD instead of NONX86_BUILD

* Remove unneeded call of __get_cpuid_max (already called by __get_cpuid)

* Remove unneeded #undef statement

gcc report:

lstm/weightmatrix.cpp: In static member function
 'static double tesseract::WeightMatrix::DotProduct(const double*, const double*, int)':
weightmatrix.cpp:67:29: warning:
 'ecx' may be used uninitialized in this function [-Wmaybe-uninitialized]
       avx_available_ = (ecx & 0x10000000) != 0;
                             ^
lstm/weightmatrix.cpp:64:30: note: 'ecx' was declared here
       unsigned int eax, ebx, ecx, edx;
                              ^
Signed-off-by: Stefan Weil <sw@weilnetz.de>
---
 lstm/weightmatrix.cpp | 41 +++++++++++++++++++++++------------------
 1 file changed, 23 insertions(+), 18 deletions(-)

diff --git a/lstm/weightmatrix.cpp b/lstm/weightmatrix.cpp
index e0bed08ce..8c7aa4645 100644
--- a/lstm/weightmatrix.cpp
+++ b/lstm/weightmatrix.cpp
@@ -18,13 +18,18 @@
 
 #include "weightmatrix.h"
 
-#undef NONX86_BUILD
-#if !defined(__x86_64__) && !defined(__i386__) && !defined(_WIN32) || defined(ANDROID_BUILD)
-#define NONX86_BUILD 1
-#endif
+#undef X86_BUILD
+#if defined(__x86_64__) || defined(__i386__) || defined(_WIN32)
+# if !defined(ANDROID_BUILD)
+#  define X86_BUILD 1
+# endif // !ANDROID_BUILD
+#endif // x86 target
 
-#if defined(__linux__) && !defined(NONX86_BUILD)
-#include <cpuid.h>
+#if defined(X86_BUILD)
+# if defined(__linux__) || defined(__MINGW32__)
+#  include <cpuid.h>
+# elif defined(_WIN32)
+# endif
 #endif
 #include "dotproductavx.h"
 #include "dotproductsse.h"
@@ -58,18 +63,20 @@ class SIMDDetect {
   // any other available SIMD equipment.
   void TestArchitecture() {
     SVAutoLock lock(&arch_mutex_);
-    if (arch_tested_) return;
-#if defined(__linux__) && !defined(NONX86_BUILD)
-    if (__get_cpuid_max(0, NULL) >= 1) {
+    if (!arch_tested_) {
+#if defined(X86_BUILD)
+# if defined(__linux__) || defined(__MINGW32__)
       unsigned int eax, ebx, ecx, edx;
-      __get_cpuid(1, &eax, &ebx, &ecx, &edx);
-      sse_available_ = (ecx & 0x00080000) != 0;
-      avx_available_ = (ecx & 0x10000000) != 0;
+      if (__get_cpuid(1, &eax, &ebx, &ecx, &edx) != 0) {
+        sse_available_ = (ecx & 0x00080000) != 0;
+        avx_available_ = (ecx & 0x10000000) != 0;
+      }
+# endif
+      if (avx_available_) tprintf("Found AVX\n");
+      if (sse_available_) tprintf("Found SSE\n");
+#endif // X86_BUILD
+      arch_tested_ = true;
     }
-#endif
-    if (avx_available_) tprintf("Found AVX\n");
-    if (sse_available_) tprintf("Found SSE\n");
-    arch_tested_ = true;
   }
 
  private:
@@ -439,5 +446,3 @@ void WeightMatrix::MatrixDotVectorInternal(const GENERIC_2D_ARRAY<double>& w,
 }
 
 }  // namespace tesseract.
-
-#undef NONX86_BUILD

From 5cfe0c700c27dd5834de5df09c8608cdcd976d1c Mon Sep 17 00:00:00 2001
From: Stefan Weil <sw@weilnetz.de>
Date: Tue, 27 Dec 2016 11:14:15 +0100
Subject: [PATCH 2/2] lstm: Add AVX / SSE support for Windows

Signed-off-by: Stefan Weil <sw@weilnetz.de>
---
 lstm/weightmatrix.cpp | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/lstm/weightmatrix.cpp b/lstm/weightmatrix.cpp
index 8c7aa4645..84dd9b77a 100644
--- a/lstm/weightmatrix.cpp
+++ b/lstm/weightmatrix.cpp
@@ -29,6 +29,7 @@
 # if defined(__linux__) || defined(__MINGW32__)
 #  include <cpuid.h>
 # elif defined(_WIN32)
+#  include <intrin.h>
 # endif
 #endif
 #include "dotproductavx.h"
@@ -71,6 +72,14 @@ class SIMDDetect {
         sse_available_ = (ecx & 0x00080000) != 0;
         avx_available_ = (ecx & 0x10000000) != 0;
       }
+# elif defined(_WIN32)
+      int cpuInfo[4];
+      __cpuid(cpuInfo, 0);
+      if (cpuInfo[0] >= 1) {
+        __cpuid(cpuInfo, 1);
+        sse_available_ = (cpuInfo[2] & 0x00080000) != 0;
+        avx_available_ = (cpuInfo[2] & 0x10000000) != 0;
+      }
 # endif
       if (avx_available_) tprintf("Found AVX\n");
       if (sse_available_) tprintf("Found SSE\n");