Fixed the Windows build. (#5596)

Fixed the Windows build.

Fixes applied:
- Remove some more ops that don't build on Windows.
- Remove the use of symlinks that didn't work correctly and replace with
`shutil.copytree()`.
- Small fixes to make the C++ code compile.

Tested with Python 3.9 and CUDA 12.1.

---------

Co-authored-by: Costin Eseanu <costineseanu@gmail.com>
Co-authored-by: Logan Adams <114770087+loadams@users.noreply.github.com>

build_win.bat

@@ -1,22 +1,15 @@
 @echo off
 
+set DISTUTILS_USE_SDK=1
+
 set DS_BUILD_AIO=0
 set DS_BUILD_CUTLASS_OPS=0
 set DS_BUILD_EVOFORMER_ATTN=0
 set DS_BUILD_FP_QUANTIZER=0
+set DS_BUILD_INFERENCE_CORE_OPS=0
 set DS_BUILD_RAGGED_DEVICE_OPS=0
 set DS_BUILD_SPARSE_ATTN=0
-
+set DS_BUILD_TRANSFORMER_INFERENCE=0
-echo Administrative permissions required. Detecting permissions...
-
-net session >nul 2>&1
-if %errorLevel% == 0 (
-    echo Success: Administrative permissions confirmed.
-) else (
-    echo Failure: Current permissions inadequate.
-    goto end
-)
-
 
 python setup.py bdist_wheel
 

csrc/includes/simd.h

@@ -12,6 +12,7 @@
 
 #define TILE (128 * 1024 * 1024)
 #if defined(__AVX512__) or defined(__AVX256__)
+#include <immintrin.h>
 
 template <typename T>
 inline T readAs(const void* src)

csrc/quantization/pt_binding.cpp

@@ -241,7 +241,7 @@ std::vector<at::Tensor> quantized_reduction(at::Tensor& input_vals,
                               .device(at::kCUDA)
                               .requires_grad(false);
 
-    std::vector<long int> sz(input_vals.sizes().begin(), input_vals.sizes().end());
+    std::vector<int64_t> sz(input_vals.sizes().begin(), input_vals.sizes().end());
     sz[sz.size() - 1] = sz.back() / devices_per_node;  // num of GPU per nodes
     const int elems_per_in_tensor = at::numel(input_vals) / devices_per_node;
     auto output = torch::empty(sz, output_options);

setup.py

@@ -19,6 +19,7 @@ The wheel will be located at: dist/*.whl
 """
 
 import os
+import shutil
 import sys
 import subprocess
 from setuptools import setup, find_packages
@@ -207,21 +208,10 @@ else:
     git_hash = "unknown"
     git_branch = "unknown"
 
-
-def create_dir_symlink(src, dest):
-    if not os.path.islink(dest):
-        if os.path.exists(dest):
-            os.remove(dest)
-        assert not os.path.exists(dest)
-        os.symlink(src, dest)
-
-
 if sys.platform == "win32":
-    # This creates a symbolic links on Windows.
+    shutil.copytree('.\\csrc', '.\\deepspeed\\ops')
-    # It needs Administrator privilege to create symlinks on Windows.
+    shutil.copytree('.\\op_builder', '.\\deepspeed\\ops')
-    create_dir_symlink('.\\deepspeed\\ops\\csrc', '..\\..\\csrc')
+    shutil.copytree('.\\accelerator', '.\\deepspeed\\accelerator')
-    create_dir_symlink('.\\deepspeed\\ops\\op_builder', '..\\..\\op_builder')
-    create_dir_symlink('.\\deepspeed\\accelerator', '..\\accelerator')
     egg_info.manifest_maker.template = 'MANIFEST_win.in'
 
 # Parse the DeepSpeed version string from version.txt.