From 00f46120c79f841dcecf78d75e7498bf7a3fc708 Mon Sep 17 00:00:00 2001
From: Trung Nguyen <ndactrung@gmail.com>
Date: Fri, 7 Oct 2022 15:50:30 -0500
Subject: [PATCH] Removed max_cus() from Device, used device->gpu->cus()
 instead

---
 lib/gpu/lal_amoeba.cpp      | 4 ++--
 lib/gpu/lal_base_amoeba.cpp | 8 ++++----
 lib/gpu/lal_device.cpp      | 1 -
 lib/gpu/lal_device.h        | 4 +---
 lib/gpu/lal_hippo.cpp       | 4 ++--
 5 files changed, 9 insertions(+), 12 deletions(-)

diff --git a/lib/gpu/lal_amoeba.cpp b/lib/gpu/lal_amoeba.cpp
index b61d7595af..1c0aa77706 100644
--- a/lib/gpu/lal_amoeba.cpp
+++ b/lib/gpu/lal_amoeba.cpp
@@ -278,10 +278,10 @@ int AmoebaT::polar_real(const int eflag, const int vflag) {
   int nbor_pitch=this->nbor->nbor_pitch();
 
   // Compute the block size and grid size to keep all cores busy
-  const int max_cus = this->device->max_cus();
+  const int cus = this->device->gpu->cus();
   int BX=this->block_size();
   int GX=static_cast<int>(ceil(static_cast<double>(ainum)/(BX/this->_threads_per_atom)));
-  while (GX < max_cus) {
+  while (GX < cus) {
     BX /= 2;
     GX=static_cast<int>(ceil(static_cast<double>(ainum)/(BX/this->_threads_per_atom)));
   }
diff --git a/lib/gpu/lal_base_amoeba.cpp b/lib/gpu/lal_base_amoeba.cpp
index 3b2381f211..8e4e8faf83 100644
--- a/lib/gpu/lal_base_amoeba.cpp
+++ b/lib/gpu/lal_base_amoeba.cpp
@@ -727,10 +727,10 @@ int BaseAmoebaT::fphi_uind() {
     return 0;
 
   // Compute the block size and grid size to keep all cores busy
-  const int max_cus = device->max_cus();
+  const int cus = device->gpu->cus();
   int BX=block_size();
   int GX=static_cast<int>(ceil(static_cast<double>(ainum)/BX));
-  while (GX < max_cus) {
+  while (GX < cus) {
     BX /= 2;
     GX=static_cast<int>(ceil(static_cast<double>(ainum)/BX));
   }
@@ -793,10 +793,10 @@ int BaseAmoebaT::fphi_mpole() {
   int nbor_pitch=nbor->nbor_pitch();
 
   // Compute the block size and grid size to keep all cores busy
-  const int max_cus = device->max_cus();
+  const int cus = device->gpu->cus();
   int BX=block_size();
   int GX=static_cast<int>(ceil(static_cast<double>(ainum)/BX));
-  while (GX < max_cus) {
+  while (GX < cus) {
     BX /= 2;
     GX=static_cast<int>(ceil(static_cast<double>(ainum)/BX));
   }
diff --git a/lib/gpu/lal_device.cpp b/lib/gpu/lal_device.cpp
index 89ae503a97..039970a0d3 100644
--- a/lib/gpu/lal_device.cpp
+++ b/lib/gpu/lal_device.cpp
@@ -214,7 +214,6 @@ int DeviceT::init_device(MPI_Comm world, MPI_Comm replica, const int ngpu,
       }
     }
     _first_device = _last_device = best_device;
-    _max_cus = best_cus;
     type = gpu->device_type(_first_device);
 
     if (ndevices > 0) {
diff --git a/lib/gpu/lal_device.h b/lib/gpu/lal_device.h
index 7def4b7f82..74f802a096 100644
--- a/lib/gpu/lal_device.h
+++ b/lib/gpu/lal_device.h
@@ -241,8 +241,6 @@ class Device {
   inline int shuffle_avail() const { return _shuffle_avail; }
   /// For OpenCL, 0 if fast-math options disabled, 1 enabled
   inline int fast_math() const { return _fast_math; }
-  /// return the max number of CUs among the devices
-  inline int max_cus() const { return _max_cus; }
 
   /// Return the number of threads per atom for pair styles
   inline int threads_per_atom() const { return _threads_per_atom; }
@@ -326,7 +324,7 @@ class Device {
 
  private:
   std::queue<Answer<numtyp,acctyp> *> ans_queue;
-  int _init_count, _max_cus;
+  int _init_count;
   bool _device_init, _host_timer_started, _time_device;
   MPI_Comm _comm_world, _comm_replica, _comm_gpu;
   int _procs_per_gpu, _gpu_rank, _world_me, _world_size, _replica_me,
diff --git a/lib/gpu/lal_hippo.cpp b/lib/gpu/lal_hippo.cpp
index d8ef3e9a44..f20a0cfd62 100644
--- a/lib/gpu/lal_hippo.cpp
+++ b/lib/gpu/lal_hippo.cpp
@@ -619,10 +619,10 @@ int HippoT::polar_real(const int eflag, const int vflag) {
   int nbor_pitch=this->nbor->nbor_pitch();
 
   // Compute the block size and grid size to keep all cores busy
-  const int max_cus = this->device->max_cus();
+  const int cus = this->device->gpu->cus();
   int BX=this->block_size();
   int GX=static_cast<int>(ceil(static_cast<double>(ainum)/(BX/this->_threads_per_atom)));
-  while (GX < max_cus) {
+  while (GX < cus) {
     BX /= 2;
     GX=static_cast<int>(ceil(static_cast<double>(ainum)/(BX/this->_threads_per_atom)));
   }