Merge pull request #35 from FZJ-JSC/mh/sc24_fixes

Makefiles: add OUTPUT_NAME and THIS_TASK, minor updates
FZJ-JSC · Nov 15, 2024 · 9abe716 · 9abe716
2 parents 58d7ddd + 1b4d29a
commit 9abe716
Show file tree

Hide file tree

Showing 31 changed files with 165 additions and 86 deletions.
diff --git a/...ulti_GPU_Parallelization/.master/Makefile → ...i_GPU_Parallelization/.master/Makefile.in b/...ulti_GPU_Parallelization/.master/Makefile → ...i_GPU_Parallelization/.master/Makefile.in
@@ -1,4 +1,6 @@
-# Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
+# Copyright (c) 2017-2024, NVIDIA CORPORATION. All rights reserved.
+THIS_TASK := 03H-@@TASKSOL@@
+OUTPUT_NAME := jacobi.$(THIS_TASK)__$(shell date '+%Y%m%d-%H%M')
 NP ?= 4
 NVCC=nvcc
 JSC_SUBMIT_CMD ?= srun --gres=gpu:4 --ntasks-per-node 4
@@ -33,10 +35,10 @@ clean:
 	rm -f jacobi jacobi.o *.nsys-rep jacobi.*.compute-sanitizer.log
 
 sanitize: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file jacobi.%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file $(OUTPUT_NAME).%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
 
 run: jacobi
 	$(JSC_SUBMIT_CMD) -n $(NP) ./jacobi
 
 profile: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o jacobi.%q{SLURM_PROCID} ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o $(OUTPUT_NAME).%q{SLURM_PROCID} ./jacobi -niter 10
diff --git a/03-H_Multi_GPU_Parallelization/.master/copy.mk b/03-H_Multi_GPU_Parallelization/.master/copy.mk
@@ -1,13 +1,13 @@
 #!/usr/bin/make -f
-# Copyright (c) 2021, NVIDIA CORPORATION. All rights reserved.
+# Copyright (c) 2021-2024, NVIDIA CORPORATION. All rights reserved.
 TASKDIR = ../tasks/
 SOLUTIONDIR = ../solutions/
-OPT_SOLUTIONDIR = ../solutions/advanced
+OPT_SOLUTIONDIR = ../solutions/advanced/
 
 IYPNB_TEMPLATE = ../../.template.json
 
 PROCESSFILES = jacobi.cu
-COPYFILES = Makefile Instructions.ipynb Instructions.md
+COPYFILES = Instructions.ipynb Instructions.md
 
 
 TASKPROCCESFILES = $(addprefix $(TASKDIR)/,$(PROCESSFILES))
@@ -16,12 +16,19 @@ SOLUTIONPROCCESFILES = $(addprefix $(SOLUTIONDIR)/,$(PROCESSFILES))
 OPT_SOLUTIONPROCCESFILES = $(addprefix $(OPT_SOLUTIONDIR)/,$(PROCESSFILES))
 SOLUTIONCOPYFILES = $(addprefix $(SOLUTIONDIR)/,$(COPYFILES))
 OPT_SOLUTIONCOPYFILES = $(addprefix $(OPT_SOLUTIONDIR)/,$(COPYFILES))
+MAKEFILES = $(addsuffix /Makefile,$(TASKDIR) $(SOLUTIONDIR) $(OPT_SOLUTIONDIR))
 
 
 .PHONY: all task
 all: task
-task: ${TASKPROCCESFILES} ${TASKCOPYFILES} ${SOLUTIONPROCCESFILES} ${SOLUTIONCOPYFILES} ${OPT_SOLUTIONPROCCESFILES} ${OPT_SOLUTIONCOPYFILES}
-
+task: ${TASKPROCCESFILES} ${TASKCOPYFILES} ${SOLUTIONPROCCESFILES} ${SOLUTIONCOPYFILES} ${OPT_SOLUTIONPROCCESFILES} ${OPT_SOLUTIONCOPYFILES} ${MAKEFILES}
+
+$(TASKDIR)/Makefile: Makefile.in
+	sed -e 's/@@TASKSOL@@/task/' $< > $@
+$(SOLUTIONDIR)/Makefile: Makefile.in
+	sed -e 's/@@TASKSOL@@/sol/' $< > $@
+$(OPT_SOLUTIONDIR)/Makefile: Makefile.in
+	sed -e 's/@@TASKSOL@@/solopt/' $< > $@
 
 ${TASKPROCCESFILES}: $(PROCESSFILES)
 	mkdir -p $(TASKDIR)/

diff --git a/03-H_Multi_GPU_Parallelization/solutions/Makefile b/03-H_Multi_GPU_Parallelization/solutions/Makefile
@@ -1,4 +1,6 @@
-# Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
+# Copyright (c) 2017-2024, NVIDIA CORPORATION. All rights reserved.
+THIS_TASK := 03H-sol
+OUTPUT_NAME := jacobi.$(THIS_TASK)__$(shell date '+%Y%m%d-%H%M')
 NP ?= 4
 NVCC=nvcc
 JSC_SUBMIT_CMD ?= srun --gres=gpu:4 --ntasks-per-node 4
@@ -33,10 +35,10 @@ clean:
 	rm -f jacobi jacobi.o *.nsys-rep jacobi.*.compute-sanitizer.log
 
 sanitize: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file jacobi.%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file $(OUTPUT_NAME).%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
 
 run: jacobi
 	$(JSC_SUBMIT_CMD) -n $(NP) ./jacobi
 
 profile: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o jacobi.%q{SLURM_PROCID} ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o $(OUTPUT_NAME).%q{SLURM_PROCID} ./jacobi -niter 10
diff --git a/03-H_Multi_GPU_Parallelization/solutions/advanced/Makefile b/03-H_Multi_GPU_Parallelization/solutions/advanced/Makefile
@@ -1,4 +1,6 @@
-# Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
+# Copyright (c) 2017-2024, NVIDIA CORPORATION. All rights reserved.
+THIS_TASK := 03H-solopt
+OUTPUT_NAME := jacobi.$(THIS_TASK)__$(shell date '+%Y%m%d-%H%M')
 NP ?= 4
 NVCC=nvcc
 JSC_SUBMIT_CMD ?= srun --gres=gpu:4 --ntasks-per-node 4
@@ -33,10 +35,10 @@ clean:
 	rm -f jacobi jacobi.o *.nsys-rep jacobi.*.compute-sanitizer.log
 
 sanitize: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file jacobi.%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file $(OUTPUT_NAME).%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
 
 run: jacobi
 	$(JSC_SUBMIT_CMD) -n $(NP) ./jacobi
 
 profile: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o jacobi.%q{SLURM_PROCID} ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o $(OUTPUT_NAME).%q{SLURM_PROCID} ./jacobi -niter 10
diff --git a/03-H_Multi_GPU_Parallelization/tasks/Makefile b/03-H_Multi_GPU_Parallelization/tasks/Makefile
@@ -1,4 +1,6 @@
-# Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
+# Copyright (c) 2017-2024, NVIDIA CORPORATION. All rights reserved.
+THIS_TASK := 03H-task
+OUTPUT_NAME := jacobi.$(THIS_TASK)__$(shell date '+%Y%m%d-%H%M')
 NP ?= 4
 NVCC=nvcc
 JSC_SUBMIT_CMD ?= srun --gres=gpu:4 --ntasks-per-node 4
@@ -33,10 +35,10 @@ clean:
 	rm -f jacobi jacobi.o *.nsys-rep jacobi.*.compute-sanitizer.log
 
 sanitize: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file jacobi.%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file $(OUTPUT_NAME).%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
 
 run: jacobi
 	$(JSC_SUBMIT_CMD) -n $(NP) ./jacobi
 
 profile: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o jacobi.%q{SLURM_PROCID} ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o $(OUTPUT_NAME).%q{SLURM_PROCID} ./jacobi -niter 10
diff --git a/06-H_Overlap_Communication_and_Computation_MPI/.master/Instructions.ipynb b/06-H_Overlap_Communication_and_Computation_MPI/.master/Instructions.ipynb
@@ -44,7 +44,9 @@
         "    target (`make profile`)\n",
         "3.  Open the recorded profile in the GUI\n",
         "    -   Either: Install Nsight Systems locally, and transfer the\n",
-        "        .qdrep/.nsys-rep file\n",
+        "        .nsys-rep file.\n",
+        "        -   *Note*: Right-click in file-browser, choose “Download” from\n",
+        "            context menu\n",
         "    -   Or: By running Xpra in your browser: In Jupyter, select “File \\>\n",
         "        New Launcher” and “Xpra Desktop”, which will open in a new tab.\n",
         "        Don’t forget to source the environment in your `xterm`.\n",
@@ -84,7 +86,7 @@
         "-   Destroy the additional cuda streams and events before ending the\n",
         "    application"
       ],
-      "id": "7563d35d-a670-47af-acef-44cee0450930"
+      "id": "21f77d33-b675-4746-9241-24837f172b29"
     }
   ],
   "nbformat": 4,

diff --git a/06-H_Overlap_Communication_and_Computation_MPI/.master/Instructions.md b/06-H_Overlap_Communication_and_Computation_MPI/.master/Instructions.md
@@ -31,7 +31,8 @@ Use the Nsight System profiler to profile the starting point version non-Overlap
 1. Start by compiling and running the application with `make run`
 1. Record an Nsight Systems profile, using the appropriate Makefile target (`make profile`)
 1. Open the recorded profile in the GUI
-    - Either: Install Nsight Systems locally, and transfer the .qdrep/.nsys-rep file
+    - Either: Install Nsight Systems locally, and transfer the .nsys-rep file.
+      - *Note*: Right-click in file-browser, choose "Download" from context menu
     - Or: By running Xpra in your browser: In Jupyter, select "File > New Launcher" and "Xpra Desktop", which will open in a new tab. Don't forget to source the environment in your `xterm`.
 1. Familiarize yourself with the different rows and the traces they represent. 
     - See if you can correlate a CUDA API kernel launch call and the resulting kernel execution on the device

diff --git a/...tion_and_Computation_MPI/.master/Makefile → ...n_and_Computation_MPI/.master/Makefile.in b/...tion_and_Computation_MPI/.master/Makefile → ...n_and_Computation_MPI/.master/Makefile.in
@@ -1,4 +1,6 @@
-# Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
+# Copyright (c) 2017-2024, NVIDIA CORPORATION. All rights reserved.
+THIS_TASK := 06H-@@TASKSOL@@
+OUTPUT_NAME := jacobi.$(THIS_TASK)__$(shell date '+%Y%m%d-%H%M')
 NP ?= 1
 NVCC=nvcc
 MPICXX=mpicxx
@@ -34,10 +36,10 @@ clean:
 	rm -f jacobi jacobi_kernels.o *.nsys-rep jacobi.*.compute-sanitizer.log
 
 sanitize: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file jacobi.%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file $(OUTPUT_NAME).%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
 
 run: jacobi
 	$(JSC_SUBMIT_CMD) -n $(NP) ./jacobi
 
 profile: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o jacobi.%q{SLURM_PROCID} ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o $(OUTPUT_NAME).%q{SLURM_PROCID} ./jacobi -niter 10
diff --git a/06-H_Overlap_Communication_and_Computation_MPI/.master/copy.mk b/06-H_Overlap_Communication_and_Computation_MPI/.master/copy.mk
@@ -1,23 +1,28 @@
 #!/usr/bin/make -f
-# Copyright (c) 2021, NVIDIA CORPORATION. All rights reserved.
+# Copyright (c) 2021-2024, NVIDIA CORPORATION. All rights reserved.
 TASKDIR = ../tasks/
 SOLUTIONDIR = ../solutions/
 
 IYPNB_TEMPLATE = ../../.template.json
 
 PROCESSFILES = jacobi.cpp
-COPYFILES = Makefile Instructions.ipynb Instructions.md jacobi_kernels.cu
+COPYFILES = Instructions.ipynb Instructions.md jacobi_kernels.cu
 
 
 TASKPROCCESFILES = $(addprefix $(TASKDIR)/,$(PROCESSFILES))
 TASKCOPYFILES = $(addprefix $(TASKDIR)/,$(COPYFILES))
 SOLUTIONPROCCESFILES = $(addprefix $(SOLUTIONDIR)/,$(PROCESSFILES))
 SOLUTIONCOPYFILES = $(addprefix $(SOLUTIONDIR)/,$(COPYFILES))
+MAKEFILES = $(addsuffix /Makefile,$(TASKDIR) $(SOLUTIONDIR))
 
 .PHONY: all task clean
 all: task
-task: ${TASKPROCCESFILES} ${TASKCOPYFILES} ${SOLUTIONPROCCESFILES} ${SOLUTIONCOPYFILES}
+task: ${TASKPROCCESFILES} ${TASKCOPYFILES} ${SOLUTIONPROCCESFILES} ${SOLUTIONCOPYFILES} ${MAKEFILES}
 
+$(TASKDIR)/Makefile: Makefile.in
+	sed -e 's/@@TASKSOL@@/task/' $< > $@
+$(SOLUTIONDIR)/Makefile: Makefile.in
+	sed -e 's/@@TASKSOL@@/sol/' $< > $@
 
 ${TASKPROCCESFILES}: $(PROCESSFILES)
 	mkdir -p $(TASKDIR)/

diff --git a/06-H_Overlap_Communication_and_Computation_MPI/solutions/Instructions.ipynb b/06-H_Overlap_Communication_and_Computation_MPI/solutions/Instructions.ipynb
@@ -44,7 +44,9 @@
         "    target (`make profile`)\n",
         "3.  Open the recorded profile in the GUI\n",
         "    -   Either: Install Nsight Systems locally, and transfer the\n",
-        "        .qdrep/.nsys-rep file\n",
+        "        .nsys-rep file.\n",
+        "        -   *Note*: Right-click in file-browser, choose “Download” from\n",
+        "            context menu\n",
         "    -   Or: By running Xpra in your browser: In Jupyter, select “File \\>\n",
         "        New Launcher” and “Xpra Desktop”, which will open in a new tab.\n",
         "        Don’t forget to source the environment in your `xterm`.\n",
@@ -84,7 +86,7 @@
         "-   Destroy the additional cuda streams and events before ending the\n",
         "    application"
       ],
-      "id": "7563d35d-a670-47af-acef-44cee0450930"
+      "id": "21f77d33-b675-4746-9241-24837f172b29"
     }
   ],
   "nbformat": 4,

diff --git a/06-H_Overlap_Communication_and_Computation_MPI/solutions/Instructions.md b/06-H_Overlap_Communication_and_Computation_MPI/solutions/Instructions.md
@@ -31,7 +31,8 @@ Use the Nsight System profiler to profile the starting point version non-Overlap
 1. Start by compiling and running the application with `make run`
 1. Record an Nsight Systems profile, using the appropriate Makefile target (`make profile`)
 1. Open the recorded profile in the GUI
-    - Either: Install Nsight Systems locally, and transfer the .qdrep/.nsys-rep file
+    - Either: Install Nsight Systems locally, and transfer the .nsys-rep file.
+      - *Note*: Right-click in file-browser, choose "Download" from context menu
     - Or: By running Xpra in your browser: In Jupyter, select "File > New Launcher" and "Xpra Desktop", which will open in a new tab. Don't forget to source the environment in your `xterm`.
 1. Familiarize yourself with the different rows and the traces they represent. 
     - See if you can correlate a CUDA API kernel launch call and the resulting kernel execution on the device

diff --git a/06-H_Overlap_Communication_and_Computation_MPI/solutions/Makefile b/06-H_Overlap_Communication_and_Computation_MPI/solutions/Makefile
@@ -1,4 +1,6 @@
-# Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
+# Copyright (c) 2017-2024, NVIDIA CORPORATION. All rights reserved.
+THIS_TASK := 06H-sol
+OUTPUT_NAME := jacobi.$(THIS_TASK)__$(shell date '+%Y%m%d-%H%M')
 NP ?= 1
 NVCC=nvcc
 MPICXX=mpicxx
@@ -34,10 +36,10 @@ clean:
 	rm -f jacobi jacobi_kernels.o *.nsys-rep jacobi.*.compute-sanitizer.log
 
 sanitize: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file jacobi.%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file $(OUTPUT_NAME).%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
 
 run: jacobi
 	$(JSC_SUBMIT_CMD) -n $(NP) ./jacobi
 
 profile: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o jacobi.%q{SLURM_PROCID} ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o $(OUTPUT_NAME).%q{SLURM_PROCID} ./jacobi -niter 10
diff --git a/06-H_Overlap_Communication_and_Computation_MPI/tasks/Instructions.ipynb b/06-H_Overlap_Communication_and_Computation_MPI/tasks/Instructions.ipynb
@@ -44,7 +44,9 @@
         "    target (`make profile`)\n",
         "3.  Open the recorded profile in the GUI\n",
         "    -   Either: Install Nsight Systems locally, and transfer the\n",
-        "        .qdrep/.nsys-rep file\n",
+        "        .nsys-rep file.\n",
+        "        -   *Note*: Right-click in file-browser, choose “Download” from\n",
+        "            context menu\n",
         "    -   Or: By running Xpra in your browser: In Jupyter, select “File \\>\n",
         "        New Launcher” and “Xpra Desktop”, which will open in a new tab.\n",
         "        Don’t forget to source the environment in your `xterm`.\n",
@@ -84,7 +86,7 @@
         "-   Destroy the additional cuda streams and events before ending the\n",
         "    application"
       ],
-      "id": "7563d35d-a670-47af-acef-44cee0450930"
+      "id": "21f77d33-b675-4746-9241-24837f172b29"
     }
   ],
   "nbformat": 4,

diff --git a/06-H_Overlap_Communication_and_Computation_MPI/tasks/Instructions.md b/06-H_Overlap_Communication_and_Computation_MPI/tasks/Instructions.md
@@ -31,7 +31,8 @@ Use the Nsight System profiler to profile the starting point version non-Overlap
 1. Start by compiling and running the application with `make run`
 1. Record an Nsight Systems profile, using the appropriate Makefile target (`make profile`)
 1. Open the recorded profile in the GUI
-    - Either: Install Nsight Systems locally, and transfer the .qdrep/.nsys-rep file
+    - Either: Install Nsight Systems locally, and transfer the .nsys-rep file.
+      - *Note*: Right-click in file-browser, choose "Download" from context menu
     - Or: By running Xpra in your browser: In Jupyter, select "File > New Launcher" and "Xpra Desktop", which will open in a new tab. Don't forget to source the environment in your `xterm`.
 1. Familiarize yourself with the different rows and the traces they represent. 
     - See if you can correlate a CUDA API kernel launch call and the resulting kernel execution on the device

diff --git a/06-H_Overlap_Communication_and_Computation_MPI/tasks/Makefile b/06-H_Overlap_Communication_and_Computation_MPI/tasks/Makefile
@@ -1,4 +1,6 @@
-# Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
+# Copyright (c) 2017-2024, NVIDIA CORPORATION. All rights reserved.
+THIS_TASK := 06H-task
+OUTPUT_NAME := jacobi.$(THIS_TASK)__$(shell date '+%Y%m%d-%H%M')
 NP ?= 1
 NVCC=nvcc
 MPICXX=mpicxx
@@ -34,10 +36,10 @@ clean:
 	rm -f jacobi jacobi_kernels.o *.nsys-rep jacobi.*.compute-sanitizer.log
 
 sanitize: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file jacobi.%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file $(OUTPUT_NAME).%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
 
 run: jacobi
 	$(JSC_SUBMIT_CMD) -n $(NP) ./jacobi
 
 profile: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o jacobi.%q{SLURM_PROCID} ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o $(OUTPUT_NAME).%q{SLURM_PROCID} ./jacobi -niter 10
diff --git a/...VSHMEM/.master/Using_CUDA_Graphs/Makefile → 08-H_NCCL_NVSHMEM/.master/NCCL/Makefile.in b/...VSHMEM/.master/Using_CUDA_Graphs/Makefile → 08-H_NCCL_NVSHMEM/.master/NCCL/Makefile.in
@@ -1,4 +1,6 @@
-# Copyright (c) 2021,2022, NVIDIA CORPORATION. All rights reserved.
+# Copyright (c) 2021-2024, NVIDIA CORPORATION. All rights reserved.
+THIS_TASK := 08H-NCCL-@@TASKSOL@@
+OUTPUT_NAME := jacobi.$(THIS_TASK)__$(shell date '+%Y%m%d-%H%M')
 NP ?= 1
 NVCC=nvcc
 JSC_SUBMIT_CMD ?= srun --cpu-bind=socket --gres=gpu:4 --ntasks-per-node 4
@@ -35,10 +37,10 @@ clean:
 	rm -f jacobi jacobi_kernels.o *.nsys-rep jacobi.*.compute-sanitizer.log
 
 sanitize: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file jacobi.%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file $(OUTPUT_NAME).%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
 
 run: jacobi
 	$(JSC_SUBMIT_CMD) -n $(NP) ./jacobi
 
 profile: jacobi
-	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx --cuda-graph-trace=node -o jacobi.%q{SLURM_PROCID} ./jacobi -niter 10
+	$(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o $(OUTPUT_NAME).%q{SLURM_PROCID} ./jacobi -niter 10
diff --git a/08-H_NCCL_NVSHMEM/.master/NCCL/copy.mk b/08-H_NCCL_NVSHMEM/.master/NCCL/copy.mk
@@ -1,23 +1,28 @@
 #!/usr/bin/make -f
-# Copyright (c) 2021, NVIDIA CORPORATION. All rights reserved.
+# Copyright (c) 2021-2024, NVIDIA CORPORATION. All rights reserved.
 TASKDIR = ../../tasks/NCCL
 SOLUTIONDIR = ../../solutions/NCCL
 
 IYPNB_TEMPLATE = ../../../.template.json
 
 PROCESSFILES = jacobi.cpp
-COPYFILES = Makefile jacobi_kernels.cu Instructions.ipynb Instructions.md
+COPYFILES = jacobi_kernels.cu Instructions.ipynb Instructions.md
 
 
 TASKPROCCESFILES = $(addprefix $(TASKDIR)/,$(PROCESSFILES))
 TASKCOPYFILES = $(addprefix $(TASKDIR)/,$(COPYFILES))
 SOLUTIONPROCCESFILES = $(addprefix $(SOLUTIONDIR)/,$(PROCESSFILES))
 SOLUTIONCOPYFILES = $(addprefix $(SOLUTIONDIR)/,$(COPYFILES))
+MAKEFILES = $(addsuffix /Makefile,$(TASKDIR) $(SOLUTIONDIR))
 
 .PHONY: all task
 all: task
-task: ${TASKPROCCESFILES} ${TASKCOPYFILES} ${SOLUTIONPROCCESFILES} ${SOLUTIONCOPYFILES}
+task: ${TASKPROCCESFILES} ${TASKCOPYFILES} ${SOLUTIONPROCCESFILES} ${SOLUTIONCOPYFILES} ${MAKEFILES}
 
+$(TASKDIR)/Makefile: Makefile.in
+	sed -e 's/@@TASKSOL@@/task/' $< > $@
+$(SOLUTIONDIR)/Makefile: Makefile.in
+	sed -e 's/@@TASKSOL@@/sol/' $< > $@
 
 ${TASKPROCCESFILES}: $(PROCESSFILES)
 	mkdir -p $(TASKDIR)/

diff --git a/08-H_NCCL_NVSHMEM/.master/NVSHMEM/Makefile → ..._NCCL_NVSHMEM/.master/NVSHMEM/Makefile.in b/08-H_NCCL_NVSHMEM/.master/NVSHMEM/Makefile → ..._NCCL_NVSHMEM/.master/NVSHMEM/Makefile.in
@@ -1,4 +1,6 @@
-# Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
+# Copyright (c) 2017-2024, All rights reserved.
+THIS_TASK := 08H-NVSHMEM-@@TASKSOL@@
+OUTPUT_NAME := jacobi.$(THIS_TASK)__$(shell date '+%Y%m%d-%H%M')
 NP ?= 4
 NVCC=nvcc
 N_D_C_VMM=1 #Enabled to hide warning and errors only found in  NVSHMEM/2.5.0 to be fixed in next release
@@ -38,10 +40,10 @@ clean:
 	rm -f jacobi jacobi.o *.nsys-rep jacobi.*.compute-sanitizer.log
 
 sanitize: jacobi
-	CUDA_VISIBLE_DEVICES=$(C_V_D) $(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file jacobi.%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
+	CUDA_VISIBLE_DEVICES=$(C_V_D) $(JSC_SUBMIT_CMD) -n $(NP) compute-sanitizer --log-file $(OUTPUT_NAME).%q{SLURM_PROCID}.compute-sanitizer.log ./jacobi -niter 10
 
 run: jacobi
 	CUDA_VISIBLE_DEVICES=$(C_V_D) $(JSC_SUBMIT_CMD) -n $(NP) ./jacobi
 
 profile: jacobi
-	CUDA_VISIBLE_DEVICES=$(C_V_D) $(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o jacobi.%q{SLURM_PROCID} ./jacobi -niter 10
+	CUDA_VISIBLE_DEVICES=$(C_V_D) $(JSC_SUBMIT_CMD) -n $(NP) nsys profile --trace=mpi,cuda,nvtx -o $(OUTPUT_NAME).%q{SLURM_PROCID} ./jacobi -niter 10