Enable SSE4 support on Windows #2 (#1274)

* Initial check in Signed-off-by: Darby Johnston <[email protected]> * Changes to use CMake options for SIMD support Signed-off-by: Darby Johnston <[email protected]> * Revert change Signed-off-by: Darby Johnston <[email protected]> * Add CpuId test Signed-off-by: Darby Johnston <[email protected]> * Add CI SIMD runs Signed-off-by: Darby Johnston <[email protected]> * Update CI run names Signed-off-by: Darby Johnston <[email protected]> * Fix CI labels Signed-off-by: Darby Johnston <[email protected]> * Revert changes in favor of runtime CPU detection Signed-off-by: Darby Johnston <[email protected]> Signed-off-by: Darby Johnston <[email protected]>
AcademySoftwareFoundation · Oct 11, 2022 · 48166fc · 48166fc
1 parent df4f7c9
commit 48166fc
Show file tree

Hide file tree

Showing 9 changed files with 114 additions and 22 deletions.
diff --git a/src/lib/OpenEXR/ImfHeader.cpp b/src/lib/OpenEXR/ImfHeader.cpp
@@ -1500,9 +1500,16 @@ staticInitialize ()
         V3dAttribute::registerAttributeType ();
         V3fAttribute::registerAttributeType ();
         V3iAttribute::registerAttributeType ();
-        DwaCompressor::initializeFuncs ();
         IDManifestAttribute::registerAttributeType ();
 
+        //
+        // Register functions, for example specialized functions
+        // for different CPU architectures.
+        //
+
+        DwaCompressor::initializeFuncs ();
+        Zip::initializeFuncs ();
+
         initialized = true;
     }
 }

diff --git a/src/lib/OpenEXR/ImfSimd.h b/src/lib/OpenEXR/ImfSimd.h
@@ -13,11 +13,11 @@
 //
 
 // GCC and Visual Studio SSE2 compiler flags
-#if defined __SSE2__ || (_MSC_VER >= 1300 && (_M_IX86 || _M_X64))
+#if defined __SSE2__ || (_MSC_VER && (_M_IX86 || _M_X64))
 #    define IMF_HAVE_SSE2 1
 #endif
 
-#if defined __SSE4_1__
+#if defined __SSE4_1__ || (_MSC_VER && (_M_IX86 || _M_X64))
 #    define IMF_HAVE_SSE4_1 1
 #endif
 

diff --git a/src/lib/OpenEXR/ImfSystemSpecific.cpp b/src/lib/OpenEXR/ImfSystemSpecific.cpp
@@ -8,6 +8,9 @@
 #include "ImfSimd.h"
 #include "OpenEXRConfig.h"
 #include "OpenEXRConfigInternal.h"
+#if defined(_MSC_VER)
+#include <intrin.h>
+#endif
 
 OPENEXR_IMF_INTERNAL_NAMESPACE_SOURCE_ENTER
 
@@ -26,6 +29,20 @@ cpuid (int n, int& eax, int& ebx, int& ecx, int& edx)
         : /* Clobber */);
 }
 
+#elif defined(_MSC_VER)
+
+// Helper functions for MSVC
+void
+cpuid (int n, int& eax, int& ebx, int& ecx, int& edx)
+{
+    int cpuInfo[4] = { -1 };
+    __cpuid (cpuInfo, n);
+    eax = cpuInfo[0];
+    ebx = cpuInfo[1];
+    ecx = cpuInfo[2];
+    edx = cpuInfo[3];
+}
+
 #else // IMF_HAVE_SSE2 && __GNUC__ && !__e2k__
 
 // Helper functions for generic compiler - all disabled

diff --git a/src/lib/OpenEXR/ImfZip.cpp b/src/lib/OpenEXR/ImfZip.cpp
@@ -8,6 +8,7 @@
 #include "ImfCheckedArithmetic.h"
 #include "ImfNamespace.h"
 #include "ImfSimd.h"
+#include "ImfSystemSpecific.h"
 
 #include <math.h>
 #include <zlib.h>
@@ -109,9 +110,12 @@ Zip::compress (const char* raw, int rawSize, char* compressed)
     return outSize;
 }
 
+namespace
+{
+
 #ifdef IMF_HAVE_SSE4_1
 
-static void
+void
 reconstruct_sse41 (char* buf, size_t outSize)
 {
     static const size_t bytesPerChunk = sizeof (__m128i);
@@ -154,9 +158,9 @@ reconstruct_sse41 (char* buf, size_t outSize)
     }
 }
 
-#else
+#endif
 
-static void
+void
 reconstruct_scalar (char* buf, size_t outSize)
 {
     unsigned char* t    = (unsigned char*) buf + 1;
@@ -170,11 +174,9 @@ reconstruct_scalar (char* buf, size_t outSize)
     }
 }
 
-#endif
-
 #ifdef IMF_HAVE_SSE2
 
-static void
+void
 interleave_sse2 (const char* source, size_t outSize, char* out)
 {
     static const size_t bytesPerChunk = 2 * sizeof (__m128i);
@@ -208,9 +210,9 @@ interleave_sse2 (const char* source, size_t outSize, char* out)
     }
 }
 
-#else
+#endif
 
-static void
+void
 interleave_scalar (const char* source, size_t outSize, char* out)
 {
     const char* t1   = source;
@@ -232,7 +234,10 @@ interleave_scalar (const char* source, size_t outSize, char* out)
     }
 }
 
-#endif
+auto reconstruct = reconstruct_scalar;
+auto interleave = interleave_scalar;
+
+} // namespace
 
 int
 Zip::uncompress (const char* compressed, int compressedSize, char* raw)
@@ -258,22 +263,34 @@ Zip::uncompress (const char* compressed, int compressedSize, char* raw)
     //
     // Predictor.
     //
-#ifdef IMF_HAVE_SSE4_1
-    reconstruct_sse41 (_tmpBuffer, outSize);
-#else
-    reconstruct_scalar (_tmpBuffer, outSize);
-#endif
+    reconstruct (_tmpBuffer, outSize);
 
     //
     // Reorder the pixel data.
     //
-#ifdef IMF_HAVE_SSE2
-    interleave_sse2 (_tmpBuffer, outSize, raw);
-#else
-    interleave_scalar (_tmpBuffer, outSize, raw);
-#endif
+    interleave (_tmpBuffer, outSize, raw);
 
     return outSize;
 }
 
+void
+Zip::initializeFuncs ()
+{
+    CpuId cpuId;
+
+#ifdef IMF_HAVE_SSE4_1
+    if (cpuId.sse4_1)
+    {
+        reconstruct = reconstruct_sse41;
+    }
+#endif
+
+#ifdef IMF_HAVE_SSE2
+    if (cpuId.sse2) 
+    {
+        interleave = interleave_sse2;
+    }
+#endif
+}
+
 OPENEXR_IMF_INTERNAL_NAMESPACE_SOURCE_EXIT
diff --git a/src/lib/OpenEXR/ImfZip.h b/src/lib/OpenEXR/ImfZip.h
@@ -40,6 +40,8 @@ class Zip
     //
     int uncompress (const char* compressed, int compressedSize, char* raw);
 
+    static void initializeFuncs ();
+
 private:
     size_t _maxRawSize;
     char*  _tmpBuffer;

diff --git a/src/test/OpenEXRTest/CMakeLists.txt b/src/test/OpenEXRTest/CMakeLists.txt
@@ -19,6 +19,7 @@ add_executable(OpenEXRTest
   testCopyDeepTiled.cpp
   testCopyMultiPartFile.cpp
   testCopyPixels.cpp
+  testCpuId.cpp
   testCustomAttributes.cpp
   testDeepScanLineBasic.cpp
   testDeepScanLineHuge.cpp
@@ -94,6 +95,7 @@ define_openexr_tests(
  testCopyDeepTiled
  testCopyMultiPartFile
  testCopyPixels
+ testCpuId
  testCustomAttributes
  testDeepScanLineBasic
  testDeepScanLineMultipleRead

diff --git a/src/test/OpenEXRTest/main.cpp b/src/test/OpenEXRTest/main.cpp
@@ -22,6 +22,7 @@
 #include "testCopyDeepTiled.h"
 #include "testCopyMultiPartFile.h"
 #include "testCopyPixels.h"
+#include "testCpuId.h"
 #include "testCustomAttributes.h"
 #include "testDeepScanLineBasic.h"
 #include "testDeepScanLineHuge.h"
@@ -230,6 +231,7 @@ main (int argc, char* argv[])
     TEST (testB44ExpLogTable, "core");
     TEST (testDwaLookups, "core");
     TEST (testIDManifest, "core");
+    TEST (testCpuId, "core");
 
     // NB: If you add a test here, make sure to enumerate it in the
     // CMakeLists.txt so it runs as part of the test suite

diff --git a/src/test/OpenEXRTest/testCpuId.cpp b/src/test/OpenEXRTest/testCpuId.cpp
@@ -0,0 +1,39 @@
+//
+// SPDX-License-Identifier: BSD-3-Clause
+// Copyright (c) Contributors to the OpenEXR Project.
+//
+
+#include <ImfSimd.h>
+#include <ImfSystemSpecific.h>
+#include <iostream>
+
+using namespace std;
+
+void
+testCpuId (const string&)
+{
+#if defined(IMF_HAVE_SSE2)
+    std::cout << "IMF_HAVE_SSE2: " << true << "\n";
+#else
+    std::cout << "IMF_HAVE_SSE2: " << false << "\n";
+#endif
+#if defined(IMF_HAVE_SSE4_1)
+    std::cout << "IMF_HAVE_SSE4_1: " << true << "\n";
+#else
+    std::cout << "IMF_HAVE_SSE4_1: " << false << "\n";
+#endif
+#if defined(IMF_HAVE_AVX)
+    std::cout << "IMF_HAVE_AVX: " << true << "\n";
+#else
+    std::cout << "IMF_HAVE_AVX: " << false << "\n";
+#endif
+
+    Imf::CpuId cpuId;
+    std::cout << "cpuId.sse2: " << cpuId.sse2 << "\n";
+    std::cout << "cpuId.sse3: " << cpuId.sse3 << "\n";
+    std::cout << "cpuId.ssse3: " << cpuId.ssse3 << "\n";
+    std::cout << "cpuId.sse4_1: " << cpuId.sse4_1 << "\n";
+    std::cout << "cpuId.sse4_2: " << cpuId.sse4_2 << "\n";
+    std::cout << "cpuId.avx: " << cpuId.avx << "\n";
+    std::cout << "cpuId.f16c: " << cpuId.f16c << std::endl;
+}
diff --git a/src/test/OpenEXRTest/testCpuId.h b/src/test/OpenEXRTest/testCpuId.h
@@ -0,0 +1,6 @@
+//
+// SPDX-License-Identifier: BSD-3-Clause
+// Copyright (c) Contributors to the OpenEXR Project.
+//
+
+void testCpuId (const std::string&);