From 0e96cb895fd3b70f0fa78b75fa4fdc0ad2538758 Mon Sep 17 00:00:00 2001
From: Sean Keely <Sean.Keely@amd.com>
Date: Sun, 16 Jan 2022 08:24:50 -0600
Subject: [PATCH] Correct queue minimum size enforcement.

Minimum queue size was not enforced at the Agent level.  Minimum
size should be one page to give unifority across all asics.

Change-Id: I26394f79458d09fbceb79fc8aaf495e2c26a8ff3


[ROCm/ROCR-Runtime commit: a6742209f7bdf07811d6d084e14dea6cc19374e3]
---
 .../runtime/hsa-runtime/core/inc/amd_gpu_agent.h             | 5 +++--
 .../runtime/hsa-runtime/core/runtime/amd_aql_queue.cpp       | 4 ++--
 .../runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp       | 5 +++++
 3 files changed, 10 insertions(+), 4 deletions(-)

diff --git a/projects/rocr-runtime/runtime/hsa-runtime/core/inc/amd_gpu_agent.h b/projects/rocr-runtime/runtime/hsa-runtime/core/inc/amd_gpu_agent.h
index 8d1544987c..ed64d5be3e 100644
--- a/projects/rocr-runtime/runtime/hsa-runtime/core/inc/amd_gpu_agent.h
+++ b/projects/rocr-runtime/runtime/hsa-runtime/core/inc/amd_gpu_agent.h
@@ -336,7 +336,8 @@ class GpuAgent : public GpuAgentInt {
   const std::function<void(void*)>& system_deallocator() const { return system_deallocator_; }
 
  protected:
-  static const uint32_t minAqlSize_ = 0x1000;   // 4KB min
+  // Sizes are in packets.
+  static const uint32_t minAqlSize_ = 0x40;     // 4KB min
   static const uint32_t maxAqlSize_ = 0x20000;  // 8MB max
 
   // @brief Create an internal queue allowing tools to be notified.
@@ -344,7 +345,7 @@ class GpuAgent : public GpuAgentInt {
     return CreateInterceptibleQueue(core::Queue::DefaultErrorHandler, nullptr);
   }
 
-  // @brief // @brief Create an internal queue, with a custom error handler, allowing tools to be
+  // @brief Create an internal queue, with a custom error handler, allowing tools to be
   // notified.
   core::Queue* CreateInterceptibleQueue(void (*callback)(hsa_status_t status, hsa_queue_t* source,
                                                          void* data),
diff --git a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_aql_queue.cpp b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_aql_queue.cpp
index ac452a34e1..2981205a76 100644
--- a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_aql_queue.cpp
+++ b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_aql_queue.cpp
@@ -671,8 +671,8 @@ void AqlQueue::AllocRegisteredRingBuffer(uint32_t queue_size_pkts) {
 #endif
   } else {
     // Allocate storage for the ring buffer.
-    ring_buf_alloc_bytes_ = AlignUp(
-        queue_size_pkts * sizeof(core::AqlPacket), 4096);
+    ring_buf_alloc_bytes_ = queue_size_pkts * sizeof(core::AqlPacket);
+    assert(IsMultipleOf(ring_buf_alloc_bytes_, 4096) && "Ring buffer sizes must be 4KiB aligned.");
 
     ring_buf_ = agent_->system_allocator()(
         ring_buf_alloc_bytes_, 0x1000,
diff --git a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp
index 1f8b70ab9e..2230b96fcf 100644
--- a/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp
+++ b/projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp
@@ -1055,6 +1055,11 @@ hsa_status_t GpuAgent::QueueCreate(size_t size, hsa_queue_type32_t queue_type,
     return HSA_STATUS_ERROR_OUT_OF_RESOURCES;
   }
 
+  // Enforce min size
+  if (size < minAqlSize_) {
+    return HSA_STATUS_ERROR_INVALID_ARGUMENT;
+  }
+
   // Allocate scratch memory
   ScratchInfo scratch = {0};
   if (private_segment_size == UINT_MAX) {