[idd] clipboard: optimize write-combined reads

Add a reusable copy path for moving data from write-combined mappings
into normal cacheable memory. Dispatch to MOVNTDQA-based SSE4.1 or AVX2
loads when the processor and OS support them, with a fenced fallback.

Use the optimized path when staging inbound clipboard stream records from
the IVSHMEM mapping.

idd/LGCommon/LGCommon.vcxproj

@@ -70,6 +70,10 @@
     <ClCompile Include="CDebug.cpp" />
     <ClCompile Include="CPipeEndpoint.cpp" />
     <ClCompile Include="RefreshRate.cpp" />
+    <ClCompile Include="WCCopy.cpp" />
+    <ClCompile Include="WCCopyAVX2.cpp">
+      <EnableEnhancedInstructionSet>AdvancedVectorExtensions2</EnableEnhancedInstructionSet>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="CClipboardChannel.h" />
@@ -85,6 +89,7 @@
     <ClInclude Include="PipeMsg.h" />
     <ClInclude Include="RefreshRate.h" />
     <ClInclude Include="Seq.h" />
+    <ClInclude Include="WCCopy.h" />
   </ItemGroup>
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
   <ImportGroup Label="ExtensionTargets" />

idd/LGCommon/LGCommon.vcxproj.filters

@@ -27,6 +27,12 @@
     <ClCompile Include="RefreshRate.cpp">
       <Filter>Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="WCCopy.cpp">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="WCCopyAVX2.cpp">
+      <Filter>Source Files</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="Atomic.h">
@@ -68,5 +74,8 @@
     <ClInclude Include="Seq.h">
       <Filter>Header Files</Filter>
     </ClInclude>
+    <ClInclude Include="WCCopy.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
   </ItemGroup>
 </Project>

idd/LGCommon/WCCopy.cpp

@@ -0,0 +1,141 @@
+/**
+ * Looking Glass
+ * Copyright © 2017-2026 The Looking Glass Authors
+ * https://looking-glass.io
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, write to the Free Software Foundation, Inc., 59
+ * Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+
+#include "WCCopy.h"
+
+#include <immintrin.h>
+#include <intrin.h>
+#include <stdint.h>
+#include <string.h>
+
+namespace WCCopy
+{
+  namespace Detail
+  {
+    void CopyAVX2(void * destination, const void * source, size_t size);
+  }
+
+  namespace
+  {
+    using CopyFn = void (*)(void *, const void *, size_t);
+    static constexpr unsigned __int64 AVX_XSTATE_MASK = 0x6U;
+
+    void CopyFallback(void * destination, const void * source, size_t size)
+    {
+      _mm_mfence();
+      memcpy(destination, source, size);
+      _mm_mfence();
+    }
+
+#if defined(_M_IX86) || defined(_M_X64)
+    __declspec(noinline) void CopySSE41(
+      void * destination, const void * source, size_t size)
+    {
+      uint8_t * dst       = static_cast<uint8_t *>(destination);
+      const uint8_t * src = static_cast<const uint8_t *>(source);
+
+      _mm_mfence();
+
+      const size_t prefix =
+        (16U - (reinterpret_cast<uintptr_t>(src) & 15U)) & 15U;
+      if (prefix)
+      {
+        const size_t copy = prefix < size ? prefix : size;
+        memcpy(dst, src, copy);
+        dst  += copy;
+        src  += copy;
+        size -= copy;
+      }
+
+      while (size >= 64U)
+      {
+        // MSVC's SSE4.1 intrinsic omits const from its input pointer.
+        __m128i * input = reinterpret_cast<__m128i *>(
+          const_cast<uint8_t *>(src));
+        __m128i * output = reinterpret_cast<__m128i *>(dst);
+        const __m128i v0 = _mm_stream_load_si128(input + 0);
+        const __m128i v1 = _mm_stream_load_si128(input + 1);
+        const __m128i v2 = _mm_stream_load_si128(input + 2);
+        const __m128i v3 = _mm_stream_load_si128(input + 3);
+        _mm_storeu_si128(output + 0, v0);
+        _mm_storeu_si128(output + 1, v1);
+        _mm_storeu_si128(output + 2, v2);
+        _mm_storeu_si128(output + 3, v3);
+        dst  += 64U;
+        src  += 64U;
+        size -= 64U;
+      }
+
+      while (size >= 16U)
+      {
+        const __m128i value = _mm_stream_load_si128(
+          reinterpret_cast<__m128i *>(const_cast<uint8_t *>(src)));
+        _mm_storeu_si128(reinterpret_cast<__m128i *>(dst), value);
+        dst  += 16U;
+        src  += 16U;
+        size -= 16U;
+      }
+
+      if (size)
+        memcpy(dst, src, size);
+
+      _mm_mfence();
+    }
+
+    CopyFn SelectCopy()
+    {
+      int registers[4] = {};
+      __cpuid(registers, 0);
+      const int maximumLeaf = registers[0];
+      if (maximumLeaf < 1)
+        return &CopyFallback;
+
+      __cpuidex(registers, 1, 0);
+      const bool sse41   = (registers[2] & (1 << 19)) != 0;
+      const bool avx     = (registers[2] & (1 << 28)) != 0;
+      const bool osxsave = (registers[2] & (1 << 27)) != 0;
+
+      if (maximumLeaf >= 7 && avx && osxsave &&
+        (_xgetbv(0) & AVX_XSTATE_MASK) == AVX_XSTATE_MASK)
+      {
+        __cpuidex(registers, 7, 0);
+        if ((registers[1] & (1 << 5)) != 0)
+          return &Detail::CopyAVX2;
+      }
+
+      return sse41 ? &CopySSE41 : &CopyFallback;
+    }
+#else
+    CopyFn SelectCopy()
+    {
+      return &CopyFallback;
+    }
+#endif
+  }
+
+  void Copy(void * destination, const void * source, size_t size)
+  {
+    if (!size)
+      return;
+
+    static const CopyFn copy = SelectCopy();
+    copy(destination, source, size);
+  }
+}

idd/LGCommon/WCCopy.h

@@ -0,0 +1,30 @@
+/**
+ * Looking Glass
+ * Copyright © 2017-2026 The Looking Glass Authors
+ * https://looking-glass.io
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, write to the Free Software Foundation, Inc., 59
+ * Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+
+#pragma once
+
+#include <stddef.h>
+
+namespace WCCopy
+{
+  // Copy from write-combined memory into normal cacheable memory. The source
+  // and destination ranges must not overlap.
+  void Copy(void * destination, const void * source, size_t size);
+}

idd/LGCommon/WCCopyAVX2.cpp

@@ -0,0 +1,85 @@
+/**
+ * Looking Glass
+ * Copyright © 2017-2026 The Looking Glass Authors
+ * https://looking-glass.io
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License as published by the Free
+ * Software Foundation; either version 2 of the License, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, write to the Free Software Foundation, Inc., 59
+ * Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+
+#include "WCCopy.h"
+
+#include <immintrin.h>
+#include <stdint.h>
+#include <string.h>
+
+namespace WCCopy
+{
+  namespace Detail
+  {
+    __declspec(noinline) void CopyAVX2(
+      void * destination, const void * source, size_t size)
+    {
+      uint8_t * dst       = static_cast<uint8_t *>(destination);
+      const uint8_t * src = static_cast<const uint8_t *>(source);
+
+      _mm_mfence();
+
+      const size_t prefix =
+        (32U - (reinterpret_cast<uintptr_t>(src) & 31U)) & 31U;
+      if (prefix)
+      {
+        const size_t copy = prefix < size ? prefix : size;
+        memcpy(dst, src, copy);
+        dst  += copy;
+        src  += copy;
+        size -= copy;
+      }
+
+      while (size >= 128U)
+      {
+        const __m256i * input =
+          reinterpret_cast<const __m256i *>(src);
+        __m256i * output = reinterpret_cast<__m256i *>(dst);
+        const __m256i v0 = _mm256_stream_load_si256(input + 0);
+        const __m256i v1 = _mm256_stream_load_si256(input + 1);
+        const __m256i v2 = _mm256_stream_load_si256(input + 2);
+        const __m256i v3 = _mm256_stream_load_si256(input + 3);
+        _mm256_storeu_si256(output + 0, v0);
+        _mm256_storeu_si256(output + 1, v1);
+        _mm256_storeu_si256(output + 2, v2);
+        _mm256_storeu_si256(output + 3, v3);
+        dst  += 128U;
+        src  += 128U;
+        size -= 128U;
+      }
+
+      while (size >= 32U)
+      {
+        const __m256i value = _mm256_stream_load_si256(
+          reinterpret_cast<const __m256i *>(src));
+        _mm256_storeu_si256(reinterpret_cast<__m256i *>(dst), value);
+        dst  += 32U;
+        src  += 32U;
+        size -= 32U;
+      }
+
+      if (size)
+        memcpy(dst, src, size);
+
+      _mm_mfence();
+      _mm256_zeroupper();
+    }
+  }
+}