[libc-commits] [libc] [libc] Update HSA queues to use the maximum size and set the barrier bit (PR #110034)

[Joseph Huber via libc-commits]

https://github.com/jhuber6 created https://github.com/llvm/llvm-project/pull/110034

Summary:
It's safer to use the maximum size, as this prevents the runtime from
oversubscribing with multiple producers. Additionally we should set the
barrier bit to ensure that the queue entries block if multiple are
submitted (Which shouldn't happen for this tool).


>From 85015cb8a379effb0e96ee59e4f1c5550af39096 Mon Sep 17 00:00:00 2001
From: Joseph Huber <huberjn at outlook.com>
Date: Wed, 25 Sep 2024 14:53:59 -0500
Subject: [PATCH] [libc] Update HSA queues to use the maximum size and set the
 barrier bit

Summary:
It's safer to use the maximum size, as this prevents the runtime from
oversubscribing with multiple producers. Additionally we should set the
barrier bit to ensure that the queue entries block if multiple are
submitted (Which shouldn't happen for this tool).
---
 libc/utils/gpu/loader/amdgpu/amdhsa-loader.cpp | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/libc/utils/gpu/loader/amdgpu/amdhsa-loader.cpp b/libc/utils/gpu/loader/amdgpu/amdhsa-loader.cpp
index 1beef8170475a1..f6ce598cd7102a 100644
--- a/libc/utils/gpu/loader/amdgpu/amdhsa-loader.cpp
+++ b/libc/utils/gpu/loader/amdgpu/amdhsa-loader.cpp
@@ -281,6 +281,7 @@ hsa_status_t launch_kernel(hsa_agent_t dev_agent, hsa_executable_t executable,
   // Initialize the packet header and set the doorbell signal to begin execution
   // by the HSA runtime.
   uint16_t header =
+      1u << HSA_PACKET_HEADER_BARRIER |
       (HSA_PACKET_TYPE_KERNEL_DISPATCH << HSA_PACKET_HEADER_TYPE) |
       (HSA_FENCE_SCOPE_SYSTEM << HSA_PACKET_HEADER_SCACQUIRE_FENCE_SCOPE) |
       (HSA_FENCE_SCOPE_SYSTEM << HSA_PACKET_HEADER_SCRELEASE_FENCE_SCOPE);
@@ -540,11 +541,11 @@ int load(int argc, const char **argv, const char **envp, void *image,
     }
   }
 
-  // Obtain a queue with the minimum (power of two) size, used to send commands
+  // Obtain a queue with the maximum (power of two) size, used to send commands
   // to the HSA runtime and launch execution on the device.
   uint64_t queue_size;
   if (hsa_status_t err = hsa_agent_get_info(
-          dev_agent, HSA_AGENT_INFO_QUEUE_MIN_SIZE, &queue_size))
+          dev_agent, HSA_AGENT_INFO_QUEUE_MAX_SIZE, &queue_size))
     handle_error(err);
   hsa_queue_t *queue = nullptr;
   if (hsa_status_t err =