[v1.8.x] Backport #19530

ci/build_windows.py

@@ -157,20 +157,6 @@ def windows_build(args):
     mxnet_root = get_mxnet_root()
     logging.info("Found MXNet root: {}".format(mxnet_root))
 
-    if 'GPU' in args.flavour:
-        # Get Thrust version to be shipped in Cuda 11, due to flakyness of
-        # older Thrust versions with MSVC 19 compiler
-        with remember_cwd():
-            tmpdirname = tempfile.mkdtemp()
-            os.chdir(tmpdirname)
-            r = requests.get('https://github.com/thrust/thrust/archive/1.9.8.zip', allow_redirects=True)
-            with open('thrust.zip', 'wb') as f:
-                f.write(r.content)
-            with zipfile.ZipFile('thrust.zip', 'r') as zip_ref:
-                zip_ref.extractall('.')
-            thrust_path = os.path.join(tmpdirname, "thrust-1.9.8")
-
-
     # cuda thrust / CUB + VS 2019 is flaky: try multiple times if fail
     MAXIMUM_TRY = 5
     build_try = 0
@@ -184,8 +170,7 @@ def windows_build(args):
             os.chdir(path)
             env = os.environ.copy()
             if 'GPU' in args.flavour:
-                env["CXXFLAGS"] = '/FS /MD /O2 /Ob2 /I {}'.format(thrust_path)
-                env["CUDAFLAGS"] = '-I {}'.format(thrust_path)
+                env["CXXFLAGS"] = '/FS /MD /O2 /Ob2'
             cmd = "\"{}\" && cmake -GNinja {} {}".format(args.vcvars,
                                                          CMAKE_FLAGS[args.flavour],
                                                          mxnet_root)

src/operator/contrib/multi_lamb.cu

@@ -50,10 +50,10 @@ __global__ void KernelStep1(const MultiLAMBKernelParam<DType, MPDType> kernel_pa
 
   MPDType biascorrection1, biascorrection2;
   if (bias_correction) {
-    biascorrection1 = 1.0 -
-                      static_cast<MPDType>(std::pow(beta1, kernel_params.step_count[tensor_id]));
-    biascorrection2 = 1.0 -
-                      static_cast<MPDType>(std::pow(beta2, kernel_params.step_count[tensor_id]));
+    biascorrection1 = 1.0 - static_cast<MPDType>(
+                      pow(beta1, static_cast<float>(kernel_params.step_count[tensor_id])));
+    biascorrection2 = 1.0 - static_cast<MPDType>(
+                      pow(beta2, static_cast<float>(kernel_params.step_count[tensor_id])));
   } else {
     biascorrection1 = static_cast<MPDType>(1.0);
     biascorrection2 = static_cast<MPDType>(1.0);