From 44a95f819dfa89f12cd31899c4482996452cdda7 Mon Sep 17 00:00:00 2001
From: Chen Fu <1316708+chenfucn@users.noreply.github.com>
Date: Thu, 14 Mar 2024 18:13:19 +0000
Subject: [PATCH] strange compilation error

---
 cmake/onnxruntime_unittests.cmake                      |  7 +++++++
 .../contrib_ops/cuda/quantization/matmul_nbits.cc      |  2 +-
 onnxruntime/core/mickey/blk_q4/f16_gemm_sm80.h         | 10 ++++++++++
 3 files changed, 18 insertions(+), 1 deletion(-)

diff --git a/cmake/onnxruntime_unittests.cmake b/cmake/onnxruntime_unittests.cmake
index b004054c616a5..40ed0021f7f07 100644
--- a/cmake/onnxruntime_unittests.cmake
+++ b/cmake/onnxruntime_unittests.cmake
@@ -267,6 +267,13 @@ if(NOT onnxruntime_MINIMAL_BUILD AND NOT onnxruntime_REDUCED_OPS_BUILD)
     "${TEST_SRC_DIR}/optimizer/*.h"
     )
 
+  if (MSVC AND ((onnxruntime_target_platform STREQUAL "ARM64") OR (onnxruntime_target_platform STREQUAL "ARM64EC")))
+    set_source_files_properties("${TEST_SRC_DIR}/optimizer/graph_transform_test.cc" PROPERTIES COMPILE_FLAGS "/bigobj")
+    list(REMOVE_ITEM onnxruntime_test_optimizer_src
+        "${TEST_SRC_DIR}/optimizer/gpu_op_prepack_test.cc"
+    )
+endif()
+
   set(onnxruntime_test_framework_src_patterns
     "${TEST_SRC_DIR}/framework/*.cc"
     "${TEST_SRC_DIR}/framework/*.h"
diff --git a/onnxruntime/contrib_ops/cuda/quantization/matmul_nbits.cc b/onnxruntime/contrib_ops/cuda/quantization/matmul_nbits.cc
index b8cfde96200b1..a29aaf460b7a6 100644
--- a/onnxruntime/contrib_ops/cuda/quantization/matmul_nbits.cc
+++ b/onnxruntime/contrib_ops/cuda/quantization/matmul_nbits.cc
@@ -66,7 +66,7 @@ Status MatMulNBits<T>::ComputeInternal(OpKernelContext* ctx) const {
                   "Internal Error: Prepacked gemm does not support zero points of type T. Fix the prepacking logic!");
     return PrepackedGemm(
         static_cast<cudaStream_t>(ctx->GetComputeStream()->GetHandle()),
-        helper.M(), a, b, scales, zero_points, Y);
+        static_cast<int>(helper.M()), a, b, scales, zero_points, Y);
   }
 
   const auto* a_data = a->Data<T>();
diff --git a/onnxruntime/core/mickey/blk_q4/f16_gemm_sm80.h b/onnxruntime/core/mickey/blk_q4/f16_gemm_sm80.h
index 52bff7e40dbe3..66b54a2504475 100644
--- a/onnxruntime/core/mickey/blk_q4/f16_gemm_sm80.h
+++ b/onnxruntime/core/mickey/blk_q4/f16_gemm_sm80.h
@@ -11,9 +11,19 @@
 
 #pragma once
 
+// Ignore CUTLASS warning C4100: unreferenced formal parameter
+#if defined(_MSC_VER)
+#pragma warning(push)
+#pragma warning(disable : 4100)
+#endif
+
 #include "cutlass/cutlass.h"
 #include "cutlass_ext/q4gemm/device/quantb_gemm.h"
 
+#if defined(_MSC_VER)
+#pragma warning(pop)
+#endif
+
 namespace onnxruntime {
 namespace cuda {