From cd46954cc4696362c84e90a90668ae8a861a7134 Mon Sep 17 00:00:00 2001
From: Sean Keely <Sean.Keely@amd.com>
Date: Wed, 28 Feb 2018 04:39:08 -0600
Subject: [PATCH] Cleanup in blit kernel management code.

Remove unused function (FenceRelease), add comments to barrier packet settings,
correct profiling controls to work with queue wrappers.

Change-Id: I45bb26227bcc2b78edb8ad5dc497603c33234e18
---
 .../hsa-runtime/core/inc/amd_blit_kernel.h    |  4 --
 .../core/runtime/amd_blit_kernel.cpp          | 49 +------------------
 2 files changed, 2 insertions(+), 51 deletions(-)

diff --git a/runtime/hsa-runtime/core/inc/amd_blit_kernel.h b/runtime/hsa-runtime/core/inc/amd_blit_kernel.h
index 9d7090940e..b7e63d0320 100644
--- a/runtime/hsa-runtime/core/inc/amd_blit_kernel.h
+++ b/runtime/hsa-runtime/core/inc/amd_blit_kernel.h
@@ -152,10 +152,6 @@ class BlitKernel : public core::Blit {
   /// packet processor doesn't get invalid packet.
   void ReleaseWriteIndex(uint64_t write_index, uint32_t num_packet);
 
-  /// Wait until all packets are finished.
-  hsa_status_t FenceRelease(uint64_t write_index, uint32_t num_copy_packet,
-                            hsa_fence_scope_t fence);
-
   void PopulateQueue(uint64_t index, uint64_t code_handle, void* args,
                      uint32_t grid_size_x, hsa_signal_t completion_signal);
 
diff --git a/runtime/hsa-runtime/core/runtime/amd_blit_kernel.cpp b/runtime/hsa-runtime/core/runtime/amd_blit_kernel.cpp
index b9941d69e6..f1f235c217 100644
--- a/runtime/hsa-runtime/core/runtime/amd_blit_kernel.cpp
+++ b/runtime/hsa-runtime/core/runtime/amd_blit_kernel.cpp
@@ -624,6 +624,7 @@ hsa_status_t BlitKernel::SubmitLinearCopyCommand(
   uint64_t write_index_temp = write_index;
 
   // Insert barrier packets to handle dependent signals.
+  // Barrier bit keeps signal checking traffic from competing with a copy.
   const uint16_t kBarrierPacketHeader = (HSA_PACKET_TYPE_BARRIER_AND << HSA_PACKET_HEADER_TYPE) |
       (1 << HSA_PACKET_HEADER_BARRIER) |
       (HSA_FENCE_SCOPE_NONE << HSA_PACKET_HEADER_SCACQUIRE_FENCE_SCOPE) |
@@ -775,8 +776,7 @@ hsa_status_t BlitKernel::SubmitLinearFillCommand(void* ptr, uint32_t value,
 }
 
 hsa_status_t BlitKernel::EnableProfiling(bool enable) {
-  AMD_HSA_BITS_SET(queue_->amd_queue_.queue_properties,
-                   AMD_QUEUE_PROPERTIES_ENABLE_PROFILING, enable);
+  queue_->SetProfiling(enable);
   return HSA_STATUS_SUCCESS;
 }
 
@@ -799,51 +799,6 @@ void BlitKernel::ReleaseWriteIndex(uint64_t write_index, uint32_t num_packet) {
   doorbell->StoreRelease(write_index + num_packet - 1);
 }
 
-hsa_status_t BlitKernel::FenceRelease(uint64_t write_index,
-                                      uint32_t num_copy_packet,
-                                      hsa_fence_scope_t fence) {
-  // This function is not thread safe.
-
-  const uint16_t kBarrierPacketHeader = (HSA_PACKET_TYPE_BARRIER_AND << HSA_PACKET_HEADER_TYPE) |
-      (1 << HSA_PACKET_HEADER_BARRIER) |
-      (HSA_FENCE_SCOPE_NONE << HSA_PACKET_HEADER_SCACQUIRE_FENCE_SCOPE) |
-      (fence << HSA_PACKET_HEADER_SCRELEASE_FENCE_SCOPE);
-
-  hsa_barrier_and_packet_t packet = {0};
-  packet.header = kInvalidPacketHeader;
-
-  HSA::hsa_signal_store_relaxed(completion_signal_, 1);
-  packet.completion_signal = completion_signal_;
-
-  if (num_copy_packet == 0) {
-    assert(write_index == 0);
-    // Reserve write index.
-    write_index = AcquireWriteIndex(1);
-  }
-
-  // Populate queue buffer with AQL packet.
-  hsa_barrier_and_packet_t* queue_buffer =
-      reinterpret_cast<hsa_barrier_and_packet_t*>(
-          queue_->public_handle()->base_address);
-  std::atomic_thread_fence(std::memory_order_acquire);
-  queue_buffer[(write_index + num_copy_packet) & queue_bitmask_] = packet;
-  std::atomic_thread_fence(std::memory_order_release);
-  queue_buffer[(write_index + num_copy_packet) & queue_bitmask_].header =
-      kBarrierPacketHeader;
-
-  // Launch packet.
-  ReleaseWriteIndex(write_index, num_copy_packet + 1);
-
-  // Wait for the packet to finish.
-  if (HSA::hsa_signal_wait_scacquire(packet.completion_signal, HSA_SIGNAL_CONDITION_LT, 1,
-                                     uint64_t(-1), HSA_WAIT_STATE_ACTIVE) != 0) {
-    // Signal wait returned unexpected value.
-    return HSA_STATUS_ERROR;
-  }
-
-  return HSA_STATUS_SUCCESS;
-}
-
 void BlitKernel::PopulateQueue(uint64_t index, uint64_t code_handle, void* args,
                                uint32_t grid_size_x,
                                hsa_signal_t completion_signal) {