[libc] Fix the implementation of exit on the GPU

[jhuber6]

Summary

The RPC calls all have delays associated with them. Currently the exit
function does an async send and immediately exits the GPU. This can have
the effect that the RPC server never sees the exit call and we continue.
This patch changes that to first sync with the server before continuing
to perform its exit. There is still a hazard here, where the kernel can
complete before the RPC call reads back its response, but this is simply
multi-threaded hazards. This change ensures that the server *will*
always exit some time after the GPU exits.

[JonChesterfield]

libc/docs/gpu/support.rst

102

Copy paste error?

libc/src/__support/OSUtil/gpu/quick_exit.cpp

22

I expected the status to go in the first call but it probably doesn't matter. Maybe leave a comment in the client similar to the one in the server?

libc/utils/gpu/server/rpc_server.cpp

106

Sync is a bit overloaded. The idea here is for the client to wait until it's sure something on the server is listening to it before it sends the exit code into the void?

[jhuber6]

Addressing comments

libc/docs/gpu/support.rst

 =============  =========  ============
 Function Name  Available  RPC Required
 =============  =========  ============
 abs            |check|
 atoi           |check|
 atof           |check|
 atol           |check|
 atoll          |check|
+exit           |check|    |check|
 labs           |check|
 llabs          |check|
 strtod         |check|
 strtof         |check|
 strtol         |check|
 strtold        |check|
 strtoll        |check|
 strtoul        |check|
 strtoull       |check|
 =============  =========  ============
 
 stdio.h
 --------
 
 =============  =========  ============
 Function Name  Available  RPC Required
 =============  =========  ============
 puts           |check|    |check|
 fputs          |check|    |check|
 =============  =========  ============

libc/src/__support/OSUtil/gpu/quick_exit.cpp

 #include "quick_exit.h"
 
 #include "src/__support/RPC/rpc_client.h"
 #include "src/__support/macros/properties/architectures.h"
 
 namespace __llvm_libc {
 
 void quick_exit(int status) {
+  // We want to first make sure the server is listening before we exit.
   rpc::Client::Port port = rpc::client.open<RPC_EXIT>();
+  port.send_and_recv([](rpc::Buffer *) {}, [](rpc::Buffer *) {});
   port.send([&](rpc::Buffer *buffer) {
     reinterpret_cast<uint32_t *>(buffer->data)[0] = status;
   });
   port.close();
 
 #if defined(LIBC_TARGET_ARCH_IS_NVPTX)
   LIBC_INLINE_ASM("exit;" ::: "memory");
 #elif defined(LIBC_TARGET_ARCH_IS_AMDGPU)

libc/utils/gpu/server/rpc_server.cpp

case RPC_WRITE_TO_STDOUT: {
       });
       for (uint64_t i = 0; i < rpc::MAX_LANE_SIZE; ++i) {
         if (strs[i])
           delete[] reinterpret_cast<uint8_t *>(strs[i]);
       }
       break;
     }
     case RPC_EXIT: {
+      // Send a response to the client to signal that we are ready to exit.
+      port->recv_and_send([](rpc::Buffer *) {});
       port->recv([](rpc::Buffer *buffer) {
         int status = 0;
         std::memcpy(&status, buffer->data, sizeof(int));
         exit(status);
       });
       break;
     }
     // TODO: Move handling of these  test cases to the loader implementation.