diff options
author | Joseph Huber <35342157+jhuber6@users.noreply.github.com> | 2023-09-26 16:09:48 -0500 |
---|---|---|
committer | GitHub <noreply@github.com> | 2023-09-26 16:09:48 -0500 |
commit | 1a5d3b6cda2c56a39bbe2a529db4d3ac3d5ffa0f (patch) | |
tree | 416123f63f14326fda52fd17e951de3389538a0c /libc/utils | |
parent | 0f88be77ea59360ae43f4e5683e23f7edce6aceb (diff) | |
download | llvm-1a5d3b6cda2c56a39bbe2a529db4d3ac3d5ffa0f.zip llvm-1a5d3b6cda2c56a39bbe2a529db4d3ac3d5ffa0f.tar.gz llvm-1a5d3b6cda2c56a39bbe2a529db4d3ac3d5ffa0f.tar.bz2 |
[libc] Scan the ports more fairly in the RPC server (#66680)
Summary:
Currently, we use the RPC server to respond to different ports which
each contain a request from some client thread wishing to do work on the
server. This scan starts at zero and continues until its checked all
ports at which point it resets. If we find an active port, we service it
and then restart the search.
This is bad for two reasons. First, it means that we will always bias
the lower ports. If a thread grabs a high port it will be stuck for a
very long time until all the other work is done. Second, it means that
the `handle_server` function can technically run indefinitely as long as
the client is always pushing new work. Because the OpenMP implementation
uses the user thread to service the kernel, this means that it could be
stalled with another asyncrhonous device's kernels.
This patch addresses this by making the server restart at the next port
over. This means we will always do a full scan of the ports before
quitting.
Diffstat (limited to 'libc/utils')
-rw-r--r-- | libc/utils/gpu/server/rpc_server.cpp | 18 |
1 files changed, 12 insertions, 6 deletions
diff --git a/libc/utils/gpu/server/rpc_server.cpp b/libc/utils/gpu/server/rpc_server.cpp index 6395a80..1c1c9f1 100644 --- a/libc/utils/gpu/server/rpc_server.cpp +++ b/libc/utils/gpu/server/rpc_server.cpp @@ -36,11 +36,12 @@ struct Server { rpc_status_t handle_server( const std::unordered_map<rpc_opcode_t, rpc_opcode_callback_ty> &callbacks, - const std::unordered_map<rpc_opcode_t, void *> &callback_data) { + const std::unordered_map<rpc_opcode_t, void *> &callback_data, + uint32_t &index) { rpc_status_t ret = RPC_STATUS_SUCCESS; std::visit( [&](auto &server) { - ret = handle_server(*server, callbacks, callback_data); + ret = handle_server(*server, callbacks, callback_data, index); }, server); return ret; @@ -51,8 +52,9 @@ private: rpc_status_t handle_server( rpc::Server<lane_size> &server, const std::unordered_map<rpc_opcode_t, rpc_opcode_callback_ty> &callbacks, - const std::unordered_map<rpc_opcode_t, void *> &callback_data) { - auto port = server.try_open(); + const std::unordered_map<rpc_opcode_t, void *> &callback_data, + uint32_t &index) { + auto port = server.try_open(index); if (!port) return RPC_STATUS_SUCCESS; @@ -203,6 +205,9 @@ private: (handler->second)(port_ref, data); } } + + // Increment the index so we start the scan after this port. + index = port->get_index() + 1; port->close(); return RPC_STATUS_CONTINUE; } @@ -333,10 +338,11 @@ rpc_status_t rpc_handle_server(uint32_t device_id) { if (!state->devices[device_id]) return RPC_STATUS_ERROR; + uint32_t index = 0; for (;;) { auto &device = *state->devices[device_id]; - rpc_status_t status = - device.server.handle_server(device.callbacks, device.callback_data); + rpc_status_t status = device.server.handle_server( + device.callbacks, device.callback_data, index); if (status != RPC_STATUS_CONTINUE) return status; } |