[Feature]【Hackathon 10th Spring No.46】WIN32 ifdef compilation guards [cf]

custom_ops/gpu_ops/beam_search_softmax.cu

@@ -12,21 +12,24 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+// Hackathon 10th Spring No.46 — compilation guards
 #ifdef __NVCC__
 #include <cub/cub.cuh>
 #endif
 #ifdef __HIPCC__
 #include <hipcub/hipcub.hpp>
 namespace cub = hipcub;
 #endif
-#include <fcntl.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <sys/mman.h>
 #include <sys/stat.h>
 #include <sys/types.h>
+#ifndef _WIN32
+#include <fcntl.h>
+#include <sys/mman.h>
 #include <unistd.h>
+#endif
 #include <algorithm>
 #include "helper.h"
 #include "stdint.h"

custom_ops/gpu_ops/custom_ftok.h

@@ -12,8 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+// Hackathon 10th Spring No.46 — compilation guards
 #pragma once
 
+#ifndef _WIN32
 #include <errno.h>
 #include <stdio.h>
 #include <sys/stat.h>
@@ -35,3 +37,4 @@ inline key_t custom_ftok(const char* path, int id) {
   return static_cast<key_t>(((st.st_dev & 0x0f) << 28) |
                             ((st.st_ino & 0xff) << 20) | (id & 0xfffff));
 }
+#endif

custom_ops/gpu_ops/dequant_int8.cu

@@ -12,14 +12,17 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include <fcntl.h>
+// Hackathon 10th Spring No.46 — compilation guards
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <sys/mman.h>
 #include <sys/stat.h>
 #include <sys/types.h>
+#ifndef _WIN32
+#include <fcntl.h>
+#include <sys/mman.h>
 #include <unistd.h>
+#endif
 #include <algorithm>
 #include "helper.h"
 

custom_ops/gpu_ops/env.h

@@ -12,8 +12,12 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+// Hackathon 10th Spring No.46 — compilation guards
 #pragma once
 
+#include <cstdlib>
+#include <string>
+
 inline uint32_t get_decoder_block_shape_q() {
   static const char* decoder_block_shape_q_env =
       std::getenv("FLAGS_dec_block_shape_q");

custom_ops/gpu_ops/fused_get_rotary_embedding.cu

@@ -12,14 +12,17 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include <fcntl.h>
+// Hackathon 10th Spring No.46 — compilation guards
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <sys/mman.h>
 #include <sys/stat.h>
 #include <sys/types.h>
+#ifndef _WIN32
+#include <fcntl.h>
+#include <sys/mman.h>
 #include <unistd.h>
+#endif
 #include <algorithm>
 #include "paddle/extension.h"
 

custom_ops/gpu_ops/fused_hadamard_quant_fp8.cu

@@ -12,14 +12,17 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include <fcntl.h>
+// Hackathon 10th Spring No.46 — compilation guards
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <sys/mman.h>
 #include <sys/stat.h>
 #include <sys/types.h>
+#ifndef _WIN32
+#include <fcntl.h>
+#include <sys/mman.h>
 #include <unistd.h>
+#endif
 #include <algorithm>
 #include "helper.h"
 

custom_ops/gpu_ops/get_data_ptr_ipc.cu

@@ -12,11 +12,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+// Hackathon 10th Spring No.46 — compilation guards
 #include <cstring>
 #include "cuda_multiprocess.h"
 #include "helper.h"
 
 namespace {
+#ifndef _WIN32
 int sharedMemoryOpen2(const char *name, size_t sz, sharedMemoryInfo *info) {
   info->size = sz;
   info->shmFd = shm_open(name, O_RDWR, 0777);
@@ -31,10 +33,16 @@ int sharedMemoryOpen2(const char *name, size_t sz, sharedMemoryInfo *info) {
 
   return 0;
 }
+#endif
 }  // namespace
 
 std::vector<paddle::Tensor> GetDataPtrIpc(const paddle::Tensor &tmp_input,
                                           const std::string &shm_name) {
+#ifdef _WIN32
+  PD_THROW(
+      "GetDataPtrIpc is not supported on Windows "
+      "(POSIX shared memory required).");
+#else
   auto out_data_ptr_tensor =
       paddle::full({1}, 0, paddle::DataType::INT64, paddle::CPUPlace());
   auto out_data_ptr_tensor_ptr = out_data_ptr_tensor.data<int64_t>();
@@ -53,6 +61,7 @@ std::vector<paddle::Tensor> GetDataPtrIpc(const paddle::Tensor &tmp_input,
 
   out_data_ptr_tensor_ptr[0] = reinterpret_cast<int64_t>(ptr);
   return {out_data_ptr_tensor};
+#endif
 }
 
 PD_BUILD_STATIC_OP(get_data_ptr_ipc)

custom_ops/gpu_ops/get_output.cc

@@ -12,11 +12,14 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+// Hackathon 10th Spring No.46 — compilation guards
 #include <stdio.h>
 #include <string.h>
+#include <sys/types.h>
+#ifndef _WIN32
 #include <sys/ipc.h>
 #include <sys/msg.h>
-#include <sys/types.h>
+#endif
 #include "custom_ftok.h"
 #include "paddle/extension.h"
 
@@ -36,6 +39,9 @@ void GetOutput(const paddle::Tensor& x,
                int64_t rank_id,
                bool wait_flag,
                int msg_queue_id) {
+#ifdef _WIN32
+  PD_THROW("GetOutput is not supported on Windows (POSIX IPC required).");
+#else
   if (rank_id > 0) {
     return;
   }
@@ -81,6 +87,7 @@ void GetOutput(const paddle::Tensor& x,
 #endif
 
   return;
+#endif
 }
 
 void GetOutputStatic(const paddle::Tensor& x, int64_t rank_id, bool wait_flag) {

custom_ops/gpu_ops/get_output_ep.cc

@@ -12,11 +12,14 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+// Hackathon 10th Spring No.46 — compilation guards
 #include <stdio.h>
 #include <string.h>
+#include <sys/types.h>
+#ifndef _WIN32
 #include <sys/ipc.h>
 #include <sys/msg.h>
-#include <sys/types.h>
+#endif
 #include "custom_ftok.h"
 #include "msg_utils.h"
 #include "paddle/extension.h"
@@ -29,6 +32,10 @@
 void GetOutputKVSignal(const paddle::Tensor& x,
                        int64_t rank_id,
                        bool wait_flag) {
+#ifdef _WIN32
+  PD_THROW(
+      "GetOutputKVSignal is not supported on Windows (POSIX IPC required).");
+#else
   int msg_queue_id = 1024;
   if (const char* msg_que_str_tmp = std::getenv("INFERENCE_MSG_QUEUE_ID")) {
     std::string msg_que_str(msg_que_str_tmp);
@@ -57,12 +64,16 @@ void GetOutputKVSignal(const paddle::Tensor& x,
     out_data[i] = msg_rcv.mtext[i];
   }
   return;
+#endif
 }
 
 void GetOutputEp(const paddle::Tensor& x,
                  int64_t rank_id,
                  bool wait_flag,
                  int msg_queue_id) {
+#ifdef _WIN32
+  PD_THROW("GetOutputEp is not supported on Windows (POSIX IPC required).");
+#else
   static struct msgdata msg_rcv;
   if (const char* inference_msg_queue_id_env_p =
           std::getenv("INFERENCE_MSG_QUEUE_ID")) {
@@ -108,6 +119,7 @@ void GetOutputEp(const paddle::Tensor& x,
 #endif
 
   return;
+#endif
 }
 
 void GetOutputEPStatic(const paddle::Tensor& x,

custom_ops/gpu_ops/get_output_msg_with_topk.cc

@@ -12,11 +12,14 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+// Hackathon 10th Spring No.46 — compilation guards
 #include <stdio.h>
 #include <string.h>
+#include <sys/types.h>
+#ifndef _WIN32
 #include <sys/ipc.h>
 #include <sys/msg.h>
-#include <sys/types.h>
+#endif
 #include "custom_ftok.h"
 #include "paddle/extension.h"
 
@@ -40,6 +43,9 @@ void GetOutputTopK(const paddle::Tensor& x,
                    int k,
                    int64_t rank_id,
                    bool wait_flag) {
+#ifdef _WIN32
+  PD_THROW("GetOutputTopK is not supported on Windows (POSIX IPC required).");
+#else
   static struct msgdata msg_rcv;
   int msg_queue_id = 1;
 
@@ -101,6 +107,7 @@ void GetOutputTopK(const paddle::Tensor& x,
     ranks_data[i] = (int64_t)msg_rcv.mtext_ranks[i];
   }
   return;
+#endif
 }
 
 PD_BUILD_STATIC_OP(get_output_topk)

custom_ops/gpu_ops/helper.h

@@ -12,22 +12,25 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+// Hackathon 10th Spring No.46 — compilation guards
 #pragma once
 
 #include <cuda_fp8.h>
 
 #ifndef PADDLE_WITH_COREX
 #include "glog/logging.h"
 #endif
-#include <fcntl.h>
 #include <nvml.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <sys/mman.h>
 #include <sys/stat.h>
 #include <sys/types.h>
+#ifndef _WIN32
+#include <fcntl.h>
+#include <sys/mman.h>
 #include <unistd.h>
+#endif
 #include <cassert>
 #include <cstdlib>
 #include <cstring>

custom_ops/gpu_ops/msg_utils.h

@@ -12,18 +12,21 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+// Hackathon 10th Spring No.46 — compilation guards
 #pragma once
 
-#include <fcntl.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <sys/mman.h>
 #include <sys/stat.h>
 #include <sys/types.h>
+#ifndef _WIN32
+#include <fcntl.h>
 #include <sys/ipc.h>
+#include <sys/mman.h>
 #include <sys/msg.h>
 #include <unistd.h>
+#endif
 #include "paddle/extension.h"
 
 #define MAX_BSZ 512

custom_ops/gpu_ops/remote_cache_kv_ipc.cc

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+// Hackathon 10th Spring No.46 — compilation guards
 #include "remote_cache_kv_ipc.h"
 
 RemoteCacheKvIpc::save_cache_kv_complete_signal_layerwise_meta_data
@@ -24,6 +25,11 @@ bool RemoteCacheKvIpc::kv_complete_signal_shmem_opened = false;
 RemoteCacheKvIpc::save_cache_kv_complete_signal_layerwise_meta_data
 RemoteCacheKvIpc::open_shm_and_get_complete_signal_meta_data(
     const int rank_id, const int device_id, const bool keep_pd_step_flag) {
+#ifdef _WIN32
+  PD_THROW(
+      "open_shm_and_get_complete_signal_meta_data is not supported on "
+      "Windows (POSIX shared memory required).");
+#else
   if (RemoteCacheKvIpc::kv_complete_signal_shmem_opened) {
     if (keep_pd_step_flag) {
       return RemoteCacheKvIpc::kv_complete_signal_meta_data;
@@ -103,6 +109,7 @@ RemoteCacheKvIpc::open_shm_and_get_complete_signal_meta_data(
   RemoteCacheKvIpc::kv_complete_signal_identity_ptr = identity_ptr;
   RemoteCacheKvIpc::kv_complete_signal_shmem_opened = true;
   return meta_data;
+#endif
 }
 
 void CUDART_CB