From de93c06e78a7051cfed4a44626ac6fc599f5c24d Mon Sep 17 00:00:00 2001
From: Thomas Gibson <14180421+thomasgibson@users.noreply.github.com>
Date: Thu, 11 Aug 2022 10:32:20 -0500
Subject: [PATCH] Add clarifying comment and further clean-up

---
 src/hip/HIPStream.cpp | 8 ++++----
 src/hip/HIPStream.h   | 1 -
 src/hip/model.cmake   | 1 -
 3 files changed, 4 insertions(+), 6 deletions(-)

diff --git a/src/hip/HIPStream.cpp b/src/hip/HIPStream.cpp
index ce69172..37fce3b 100644
--- a/src/hip/HIPStream.cpp
+++ b/src/hip/HIPStream.cpp
@@ -27,8 +27,6 @@ HIPStream<T>::HIPStream(const int ARRAY_SIZE, const int device_index)
     block_count(array_size / (TBSIZE * elements_per_lane))
 {
 
-  std::cerr << "Elements per lane: " << elements_per_lane << std::endl;
-  std::cerr << "Chunks per block: " << chunks_per_block << std::endl;
   // The array size must be divisible by total number of elements
   // moved per block for kernel launches
   if (ARRAY_SIZE % (TBSIZE * elements_per_lane) != 0)
@@ -39,7 +37,6 @@ HIPStream<T>::HIPStream(const int ARRAY_SIZE, const int device_index)
        << ").";
     throw std::runtime_error(ss.str());
   }
-  std::cerr << "block count " << block_count << std::endl;
 
   // Set device
   int count;
@@ -56,7 +53,10 @@ HIPStream<T>::HIPStream(const int ARRAY_SIZE, const int device_index)
 
   array_size = ARRAY_SIZE;
 
-  // Allocate the host array for partial sums for dot kernels
+  // Allocate the host array for partial sums for dot kernels using hipHostMalloc.
+  // This creates an array on the host which is visible to the device. However, it requires
+  // synchronization (e.g. hipDeviceSynchronize) for the result to be available on the host
+  // after it has been passed through to a kernel.
   hipHostMalloc(&sums, sizeof(T) * block_count, hipHostMallocNonCoherent);
   check_error();
 
diff --git a/src/hip/HIPStream.h b/src/hip/HIPStream.h
index 7bce0b5..305e937 100644
--- a/src/hip/HIPStream.h
+++ b/src/hip/HIPStream.h
@@ -55,7 +55,6 @@ class HIPStream : public Stream<T>
     T *d_a;
     T *d_b;
     T *d_c;
-    T *d_sum;
 
 
   public:
diff --git a/src/hip/model.cmake b/src/hip/model.cmake
index 2f7d69e..19e6fd0 100644
--- a/src/hip/model.cmake
+++ b/src/hip/model.cmake
@@ -5,6 +5,5 @@ register_flag_required(CMAKE_CXX_COMPILER
 register_flag_optional(DWORDS_PER_LANE "Flag indicating the number of dwords to process per wavefront lane." 4)
 
 macro(setup)
-    # Ensure we set the proper preprocessor directives
     register_definitions(DWORDS_PER_LANE=${DWORDS_PER_LANE})
 endmacro()
\ No newline at end of file