From 4677a2aaccb551b76b4957d92ff5e21318ca39d4 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=EC=9C=A0=EC=9A=A9=ED=99=98?=
 <33824408+eric-yoo@users.noreply.github.com>
Date: Thu, 9 Apr 2020 02:50:57 +0900
Subject: [PATCH 1/2] Update README

fix PATH (append to original path)
---
 README | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/README b/README
index 7ec45ea8a..b08ca0f53 100644
--- a/README
+++ b/README
@@ -189,7 +189,7 @@ your PATH.  You probably want to modify your .bashrc file to incude the
 following (this assumes the CUDA Toolkit was installed in /usr/local/cuda):
 
 	export CUDA_INSTALL_PATH=/usr/local/cuda
-	export PATH=$CUDA_INSTALL_PATH/bin
+	export PATH=$PATH:$CUDA_INSTALL_PATH/bin
 
 
 Step 2: Build

From 3cd5b1bc879274abf3f0e73b0278f1472e2c752e Mon Sep 17 00:00:00 2001
From: cesar-avalos3 <cesar.avalos3@gmail.com>
Date: Thu, 30 Jul 2020 15:58:06 -0400
Subject: [PATCH 2/2] Add Completed CTAs output in Visualizer Log

---
 src/gpgpu-sim/shader.cc | 4 ++++
 src/gpgpu-sim/shader.h  | 2 ++
 2 files changed, 6 insertions(+)

diff --git a/src/gpgpu-sim/shader.cc b/src/gpgpu-sim/shader.cc
index 8b226b6e9..c6e7b8f67 100644
--- a/src/gpgpu-sim/shader.cc
+++ b/src/gpgpu-sim/shader.cc
@@ -748,6 +748,8 @@ void shader_core_stats::visualizer_print(gzFile visualizer_file) {
   }
   gzprintf(visualizer_file, "\n");
 
+  gzprintf(visualizer_file, "ctas_completed: %d\n", ctas_completed);
+  ctas_completed = 0;
   // warp issue breakdown
   unsigned sid = m_config->gpgpu_warp_issue_shader;
   unsigned count = 0;
@@ -2685,6 +2687,7 @@ void shader_core_ctx::register_cta_thread_exit(unsigned cta_num,
   m_cta_status[cta_num]--;
   if (!m_cta_status[cta_num]) {
     // Increment the completed CTAs
+    m_stats->ctas_completed++;
     m_gpu->inc_completed_cta();
     m_n_active_cta--;
     m_barriers.deallocate_barrier(cta_num);
@@ -2751,6 +2754,7 @@ void gpgpu_sim::shader_print_runtime_stat(FILE *fout) {
 
 void gpgpu_sim::shader_print_scheduler_stat(FILE *fout,
                                             bool print_dynamic_info) const {
+  fprintf(fout, "ctas_completed %d, ", m_shader_stats->ctas_completed);
   // Print out the stats from the sampling shader core
   const unsigned scheduler_sampling_core =
       m_shader_config->gpgpu_warp_issue_shader;
diff --git a/src/gpgpu-sim/shader.h b/src/gpgpu-sim/shader.h
index 07cd2d00d..d00177409 100644
--- a/src/gpgpu-sim/shader.h
+++ b/src/gpgpu-sim/shader.h
@@ -1668,6 +1668,7 @@ struct shader_core_stats_pod {
   unsigned *single_issue_nums;
   unsigned *dual_issue_nums;
 
+  unsigned ctas_completed;
   // memory access classification
   int gpgpu_n_mem_read_local;
   int gpgpu_n_mem_write_local;
@@ -1781,6 +1782,7 @@ class shader_core_stats : public shader_core_stats_pod {
     dual_issue_nums =
         (unsigned *)calloc(config->gpgpu_num_sched_per_core, sizeof(unsigned));
 
+    ctas_completed = 0;
     n_simt_to_mem = (long *)calloc(config->num_shader(), sizeof(long));
     n_mem_to_simt = (long *)calloc(config->num_shader(), sizeof(long));