Windows 10 ARM64 - ANGLE build

angle.patch

This is my patch for building ANGLE on Windows 10 ARM64. It's not ready for upstream because it breaks
the x86/x86_64 builds -- this isn't out of necessity but rather expedience. I just wanted to get ANGLE
built for the platform quickly.

I did have to copy a bunch of blobs out of C:\Windows\WinSXS to fill some Windows SDK directories (which
aren't present in the actual Windows SDK for some reason):

C:\Program Files (x86)\Windows Kits\10\Redist\ucrt\DLLs\arm64 
C:\Program Files (x86)\Windows Kits\10\Redist\D3D\arm64

>gclient revinfo
.: https://chromium.googlesource.com/angle/angle.git
build: https://chromium.googlesource.com/chromium/src/build.git@2f3b6e8ce9e783b2a09496d70eef2974506a41c8
buildtools: https://chromium.googlesource.com/chromium/buildtools.git@461b345a815c1c745ac0534a6a4bd52d123abe68
buildtools\clang_format\script: https://chromium.googlesource.com/chromium/llvm-project/cfe/tools/clang-format.git@0653eee0c81ea04715c635dd0885e8096ff6ba6d
buildtools\third_party\libc++\trunk: https://chromium.googlesource.com/chromium/llvm-project/libcxx.git@f56f1bba1ade4a408d403ff050d50e837bae47df
buildtools\third_party\libc++abi\trunk: https://chromium.googlesource.com/chromium/llvm-project/libcxxabi.git@05ba3281482304ae8de31123a594972a495da06d
buildtools\third_party\libunwind\trunk: https://chromium.googlesource.com/external/llvm.org/libunwind.git@fc0a910c25d5415dd72ba9451b7cba380e3cc1e7
testing: https://chromium.googlesource.com/chromium/src/testing@6dfa36ab2e5143fa2f7353e3af5d2935af2e61f7
third_party/cherry: https://android.googlesource.com/platform/external/cherry@4f8fb08d33ca5ff05a1c638f04c85bbb8d8b52cc
third_party/deqp/src: https://android.googlesource.com/platform/external/deqp@5bc346ba2d5465a2e6094e254f12b1586fd0097f
third_party/glslang/src: https://android.googlesource.com/platform/external/shaderc/glslang@2edde6665d9a56ead5ea0e55b4e64d9a803e6164
third_party/googletest/src: https://chromium.googlesource.com/external/github.com/google/googletest.git@d175c8bf823e709d570772b038757fadf63bc632
third_party/libpng/src: https://android.googlesource.com/platform/external/libpng@094e181e79a3d6c23fd005679025058b7df1ad6c
third_party/spirv-headers/src: https://android.googlesource.com/platform/external/shaderc/spirv-headers@98b01515724c428d0f0a5d01deffcce0f5f5e61c
third_party/spirv-tools/src: https://android.googlesource.com/platform/external/shaderc/spirv-tools@9996173f363729b3a97309685dbd4d78547a63a7
third_party/vulkan-validation-layers/src: https://android.googlesource.com/platform/external/vulkan-validation-layers@354ad3ba8b88136b82b712acab9b8fcb981beaff
third_party/zlib: https://chromium.googlesource.com/chromium/src/third_party/zlib@24ab14872e8e068ba08cc31cc3d43bcc6d5cb832
tools/clang: https://chromium.googlesource.com/chromium/src/tools/clang.git@e70074db10b27867e6c873adc3ac7e5f9ee0ff6e
tools/gyp: https://chromium.googlesource.com/external/gyp@5e2b3ddde7cda5eb6bc09a5546a76b00e49d888f

diff --git a/BUILD.gn b/BUILD.gn
index 345f27d95..323414d22 100644
--- a/BUILD.gn
+++ b/BUILD.gn
@@ -24,8 +24,8 @@ if (!build_with_chromium) {
     deps = [
       ":angle_shader_translator",
       ":translator_fuzzer",
-      "//samples:all",
-      "//src/tests:all",
+      #"//samples:all",
+      #"//src/tests:all",
     ]
   }
 }
diff --git a/src/common/mathutil.h b/src/common/mathutil.h
index 73b31b740..9ac117c87 100644
--- a/src/common/mathutil.h
+++ b/src/common/mathutil.h
@@ -150,7 +150,7 @@ inline bool supportsSSE2()
         return supports;
     }
 
-#if defined(ANGLE_PLATFORM_WINDOWS) && !defined(_M_ARM)
+#if defined(ANGLE_PLATFORM_WINDOWS) && !defined(_M_ARM) && !defined(_M_ARM64)
     {
         int info[4];
         __cpuid(info, 0);
@@ -889,7 +889,8 @@ inline uint32_t BitfieldReverse(uint32_t value)
 }
 
 // Count the 1 bits.
-#if defined(ANGLE_PLATFORM_WINDOWS)
+#if defined(ANGLE_PLATFORM_WINDOWS) && !defined(_M_ARM64)
+#define BITCOUNT_FOUND
 inline int BitCount(uint32_t bits)
 {
     return static_cast<int>(__popcnt(bits));
@@ -903,6 +904,7 @@ inline int BitCount(uint64_t bits)
 #endif  // defined(ANGLE_PLATFORM_WINDOWS)
 
 #if defined(ANGLE_PLATFORM_POSIX)
+#define BITCOUNT_FOUND
 inline int BitCount(uint32_t bits)
 {
     return __builtin_popcount(bits);
@@ -916,6 +918,28 @@ inline int BitCount(uint64_t bits)
 #endif  // defined(ANGLE_IS_64_BIT_CPU)
 #endif  // defined(ANGLE_PLATFORM_POSIX)
 
+#if !defined(BITCOUNT_FOUND)
+inline int BitCount(uint32_t bits)
+{
+	int c = 0;
+	while (bits) {
+		bits &= (bits - 1);
+		c++;
+	}
+	return c;
+}
+
+inline int BitCount(uint64_t bits)
+{
+	int c = 0;
+	while (bits) {
+		bits &= (bits - 1);
+		c++;
+	}
+	return c;
+}
+#endif
+
 #if defined(ANGLE_PLATFORM_WINDOWS)
 // Return the index of the least significant bit set. Indexing is such that bit 0 is the least
 // significant bit. Implemented for different bit widths on different platforms.
diff --git a/src/common/platform.h b/src/common/platform.h
index 57a2b9238..b96bc11b5 100644
--- a/src/common/platform.h
+++ b/src/common/platform.h
@@ -84,7 +84,7 @@
 #   undef far
 #endif
 
-#if defined(_MSC_VER) && !defined(_M_ARM)
+#if defined(_MSC_VER) && !defined(_M_ARM) && !defined(_M_ARM64)
 #include <intrin.h>
 #define ANGLE_USE_SSE
 #elif defined(__GNUC__) && (defined(__x86_64__) || defined(__i386__))
diff --git a/build/build_config.h b/build/build_config.h
index 6b30422e0..ce772cd37 100644
--- a/build/build_config.h
+++ b/build/build_config.h
@@ -138,7 +138,7 @@
 #define ARCH_CPU_ARMEL 1
 #define ARCH_CPU_32_BITS 1
 #define ARCH_CPU_LITTLE_ENDIAN 1
-#elif defined(__aarch64__)
+#elif defined(__aarch64__) || defined(_M_ARM64)
 #define ARCH_CPU_ARM_FAMILY 1
 #define ARCH_CPU_ARM64 1
 #define ARCH_CPU_64_BITS 1
diff --git a/build/config/win/BUILD.gn b/build/config/win/BUILD.gn
index 80bfa75f2..5038cdf22 100644
--- a/build/config/win/BUILD.gn
+++ b/build/config/win/BUILD.gn
@@ -238,6 +238,12 @@ config("sdk_link") {
       "$visual_studio_path\VC\lib\amd64",
       "$visual_studio_path\VC\atlmfc\lib\amd64",
     ]
+  } else if (current_cpu == "arm64") {
+    ldflags = [ "/MACHINE:ARM64" ]
+    lib_dirs = [
+      "$windows_sdk_path\Lib\winv6.3\um\arm64",
+      "$visual_studio_path\VC\lib\arm64",
+    ]
   } else {
     ldflags = [
       "/MACHINE:X86",
@@ -357,6 +363,9 @@ if (current_cpu == "x64") {
   # The number after the comma is the minimum required OS version.
   # 5.02 = Windows Server 2003.
   subsystem_version_suffix = ",5.02"
+} else if (current_cpu == "arm64") {
+  # 6.02 = Windows 10
+  subsystem_version_suffix = ",6.02"
 } else {
   # 5.01 = Windows XP.
   subsystem_version_suffix = ",5.01"
diff --git a/build/toolchain/win/BUILD.gn b/build/toolchain/win/BUILD.gn
index 22054716f..6f24ea351 100644
--- a/build/toolchain/win/BUILD.gn
+++ b/build/toolchain/win/BUILD.gn
@@ -391,6 +391,27 @@ if (target_cpu == "x86") {
   }
 }
 
+if (target_cpu == "arm64") {
+  x86_toolchain_data = exec_script("setup_toolchain.py",
+                                   [
+                                     visual_studio_path,
+                                     windows_sdk_path,
+                                     visual_studio_runtime_dirs,
+                                     "arm64",
+                                     goma_disabled,
+                                   ],
+                                   "scope")
+
+  msvc_toolchain("arm64") {
+    environment = "environment.arm64"
+    cl = "${goma_prefix}\"${x86_toolchain_data.vc_bin_dir}/cl.exe\""
+    toolchain_args = {
+      current_cpu = "arm64"
+      is_clang = false
+    }
+  }
+}
+
 # 64-bit toolchains.
 x64_toolchain_data = exec_script("setup_toolchain.py",
                                  [
diff --git a/build/toolchain/win/setup_toolchain.py b/build/toolchain/win/setup_toolchain.py
index a1d2ea4b2..626364516 100644
--- a/build/toolchain/win/setup_toolchain.py
+++ b/build/toolchain/win/setup_toolchain.py
@@ -89,7 +89,7 @@ def _LoadToolchainEnv(cpu, sdk_dir):
   # Check if we are running in the SDK command line environment and use
   # the setup script from the SDK if so. |cpu| should be either
   # 'x86' or 'x64'.
-  assert cpu in ('x86', 'x64')
+  assert cpu in ('x86', 'x64', 'arm64')
   if bool(int(os.environ.get('DEPOT_TOOLS_WIN_TOOLCHAIN', 1))) and sdk_dir:
     # Load environment from json file.
     env = os.path.normpath(os.path.join(sdk_dir, 'bin/SetEnv.%s.json' % cpu))
@@ -142,8 +142,10 @@ def _LoadToolchainEnv(cpu, sdk_dir):
     # Chromium requires the 10.0.15063.468 SDK - previous versions don't have
     # all of the required declarations and 10.0.16299.0 has some
     # incompatibilities (crbug.com/773476).
-    args = [script_path, 'amd64_x86' if cpu == 'x86' else 'amd64',
-            '10.0.15063.0']
+    envname = cpu
+    if cpu == 'arm64':
+      envname = 'x86_arm64'
+    args = [script_path, envname, '10.0.16299.0']
     variables = _LoadEnvFromBat(args)
   return _ExtractImportantEnvironment(variables)
 
@@ -171,7 +173,7 @@ def main():
   target_cpu = sys.argv[4]
   goma_disabled = sys.argv[5]
 
-  cpus = ('x86', 'x64')
+  cpus = ('x86', 'x64', 'arm64')
   assert target_cpu in cpus
   vc_bin_dir = ''
   include = ''
diff --git a/build/vs_toolchain.py b/build/vs_toolchain.py
index 114eefc10..780c8d9e0 100755
--- a/build/vs_toolchain.py
+++ b/build/vs_toolchain.py
@@ -92,9 +92,10 @@ def SetEnvironmentAndGetRuntimeDllDirs():
     # directory ensures that they are available when needed.
     bitness = platform.architecture()[0]
     # When running 64-bit python the x64 DLLs will be in System32
+    arm64_path = r'C:\Program Files (x86)\Microsoft Visual Studio\2017\Professional\VC\Redist\MSVC\14.13.26020\onecore\arm64\Microsoft.VC141.CRT'
     x64_path = 'System32' if bitness == '64bit' else 'Sysnative'
     x64_path = os.path.join(r'C:\Windows', x64_path)
-    vs_runtime_dll_dirs = [x64_path, r'C:\Windows\SysWOW64']
+    vs_runtime_dll_dirs = [arm64_path, r'C:\Windows\SysWOW64']
 
   return vs_runtime_dll_dirs
 
@@ -204,14 +205,14 @@ def _CopyUCRTRuntime(target_dir, source_dir, target_cpu, dll_pattern, suffix):
                      'C:\\Program Files (x86)\\Windows Kits\\10'))
   ucrt_dll_dirs = os.path.join(win_sdk_dir, 'Redist', 'ucrt', 'DLLs',
                                target_cpu)
-  ucrt_files = glob.glob(os.path.join(ucrt_dll_dirs, 'api-ms-win-*.dll'))
+  ucrt_files = glob.glob(os.path.join(ucrt_dll_dirs, 'api-ms-win-*.dll')) + glob.glob(os.path.join(ucrt_dll_dirs, 'ucrtbase.dll'))
   assert len(ucrt_files) > 0
   for ucrt_src_file in ucrt_files:
     file_part = os.path.basename(ucrt_src_file)
     ucrt_dst_file = os.path.join(target_dir, file_part)
     _CopyRuntimeImpl(ucrt_dst_file, ucrt_src_file, False)
-  _CopyRuntimeImpl(os.path.join(target_dir, 'ucrtbase' + suffix),
-                    os.path.join(source_dir, 'ucrtbase' + suffix))
+  #_CopyRuntimeImpl(os.path.join(target_dir, 'ucrtbase' + suffix),
+  #                  os.path.join(source_dir, 'ucrtbase' + suffix))
 
 
 def FindVCToolsRoot():
@@ -249,6 +250,7 @@ def _CopyPGORuntime(target_dir, target_cpu):
     # There's no version of pgosweep.exe in HostX64/x86, so we use the copy
     # from HostX86/x86.
     pgo_x86_runtime_dir = os.path.join(pgo_runtime_root, 'HostX86', 'x86')
+    pgo_arm64_runtime_dir = os.path.join(pgo_runtime_root, 'arm64')
     pgo_x64_runtime_dir = os.path.join(pgo_runtime_root, 'HostX64', 'x64')
   else:
     raise Exception('Unexpected toolchain version: %s.' % env_version)
@@ -262,6 +264,8 @@ def _CopyPGORuntime(target_dir, target_cpu):
       source = os.path.join(pgo_x86_runtime_dir, runtime)
     elif target_cpu == 'x64':
       source = os.path.join(pgo_x64_runtime_dir, runtime)
+    elif target_cpu == 'arm64':
+      source = os.path.join(pgo_arm64_runtime_dir, runtime)
     else:
       raise NotImplementedError("Unexpected target_cpu value: " + target_cpu)
     if not os.path.exists(source):
@@ -292,7 +296,7 @@ def CopyDlls(target_dir, configuration, target_cpu):
     return
 
   x64_runtime, x86_runtime = vs_runtime_dll_dirs
-  runtime_dir = x64_runtime if target_cpu == 'x64' else x86_runtime
+  runtime_dir = x64_runtime if (target_cpu == 'x64' or target_cpu == 'arm64') else x86_runtime
   _CopyRuntime(target_dir, runtime_dir, target_cpu, debug=False)
   if configuration == 'Debug':
     _CopyRuntime(target_dir, runtime_dir, target_cpu, debug=True)