Mapping to cgra suitesparse fixes

.gitignore

@@ -5,8 +5,13 @@ __pycache__/
 # Temporary output files
 *.swp
 *.swo
+*.swn
 *.txt
 *.out
+*.log
+
+# Generated Python Package files
+*.egg-info/
 
 # Files for MacOS and IDEs
 .DS_store
@@ -17,6 +22,8 @@ cmake-build-*/
 # Generated SAM Graphs
 *.gv
 
+venv/
+
 # Generated images
 *.svg
 *.png
@@ -26,6 +33,13 @@ cmake-build-*/
 *.csv
 *.json
 
+# Tensor files
+*.mtx
+*.tns
+
+# Generated formatted tensor files
+tensor_*_mode_*
+
 # Generated folders
 build/
 logs/
@@ -38,19 +52,20 @@ compiler/benchmark/
 # Generated SAM simulator tests
 */sim/test/apps/test_*.py
 
-*.gv
-
+# Tensor files
 *.mtx
 *.tns
 
-# Temporary matrices
+# Temporary or generated tensor directories 
 tmp_mat*/
+tiles/
+synthetic/
 
-# Jupyter notebook checkpoints
-.ipynb_checkpoints/
+# Generated SAM simulator tests
+*/sim/test/apps/test_*.py
 
-# Generated formatted tensor files
-tensor_*_mode_*
+# Temporary scripts
+download_suitesparse_partial.sh
 
-# Tensor files
-tiles/
+# Network Filesystem 
+.nfs*

Makefile

@@ -31,20 +31,20 @@ endif
 
 ifeq ("$(NEVA)","ON")
 	CMD := OMP_PROC_BIND=true LD_LIBRARY_PATH=compiler/build/lib/:$(LD_LIBRARY_PATH) numactl -C 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 -m 0 compiler/build/taco-bench $(BENCHFLAGS)
-	export SUITESPARSE_PATH=/nobackup/owhsu/sparse-datasets/suitesparse/
-	export FROSTT_PATH=/nobackup/owhsu/sparse-datasets/frostt/
-	export SUITESPARSE_FORMATTED_PATH=/nobackup/owhsu/sparse-datasets/suitesparse-formatted
-	export FROSTT_FORMATTED_TACO_PATH=/nobackup/owhsu/sparse-datasets/frostt-formatted/taco-tensor
-	export FROSTT_FORMATTED_PATH=/nobackup/owhsu/sparse-datasets/frostt-formatted
-	export TACO_TENSOR_PATH=/nobackup/owhsu/sparse-datasets
+	# export SUITESPARSE_PATH=/nobackup/owhsu/sparse-datasets/suitesparse/
+	# export FROSTT_PATH=/nobackup/owhsu/sparse-datasets/frostt/
+	# export SUITESPARSE_FORMATTED_PATH=/nobackup/owhsu/sparse-datasets/suitesparse-formatted
+	# export FROSTT_FORMATTED_TACO_PATH=/nobackup/owhsu/sparse-datasets/frostt-formatted/taco-tensor
+	# export FROSTT_FORMATTED_PATH=/nobackup/owhsu/sparse-datasets/frostt-formatted
+	# export TACO_TENSOR_PATH=/nobackup/owhsu/sparse-datasets
 else ifeq ("$(LANKA)", "ON")
 	CMD := OMP_PROC_BIND=true LD_LIBRARY_PATH=compiler/build/lib/:$(LD_LIBRARY_PATH) numactl -C 0,2,4,6,8,10,24,26,28,30,32,34 -m 0 compiler/build/taco-bench $(BENCHFLAGS)
-	export SUITESPARSE_PATH=/data/scratch/changwan/florida_all
-	export FROSTT_PATH=/data/scratch/owhsu/datasets/frostt
-	export TACO_TENSOR_PATH=/data/scratch/owhsu/datasets
-	export SUITESPARSE_FORMATTED_PATH=/data/scratch/owhsu/datasets/suitesparse-formatted
-	export FROSTT_FORMATTED_TACO_PATH=/data/scratch/owhsu/datasets/frostt-formatted/taco-tensor
-	export FROSTT_FORMATTED_PATH=/data/scratch/owhsu/datasets/frostt-formatted
+	# export SUITESPARSE_PATH=/data/scratch/changwan/florida_all
+	# export FROSTT_PATH=/data/scratch/owhsu/datasets/frostt
+	# export TACO_TENSOR_PATH=/data/scratch/owhsu/datasets
+	# export SUITESPARSE_FORMATTED_PATH=/data/scratch/owhsu/datasets/suitesparse-formatted
+	# export FROSTT_FORMATTED_TACO_PATH=/data/scratch/owhsu/datasets/frostt-formatted/taco-tensor
+	# export FROSTT_FORMATTED_PATH=/data/scratch/owhsu/datasets/frostt-formatted
 else
 	CMD := LD_LIBRARY_PATH=compiler/build/lib/:$(LD_LIBRARY_PATH) compiler/build/taco-bench $(BENCHFLAGS)
 endif
@@ -57,16 +57,16 @@ guard-%:
 
 # ---- Run SAM python simulator stuff ----
 csv: 
-	scripts/pytest_suitesparse_with_benchmarks.sh
+	scripts/run_sam_sim/pytest_suitesparse_with_benchmarks.sh
 run-final: submodules
-	./scripts/run_suitesparse_final.sh $(realpath ./scripts/tensor_names/$(TENSOR_TXT)) 
+	./scripts/run_sam_sim/run_suitesparse_final.sh $(realpath ./scripts/tensor_names/$(TENSOR_TXT)) 
 
 run-gen: submodules
-	./scripts/run_suitesparse_generated.sh $(realpath ./scripts/tensor_names/$(TENSOR_TXT)) 
+	./scripts/run_sam_sim/run_suitesparse_generated.sh $(realpath ./scripts/tensor_names/$(TENSOR_TXT)) 
 
 tests: sam 
 	mkdir -p sam/sim/test/apps
-	python scripts/test_generating_code.py
+	python scripts/gen_sam_apps/test_generating_code.py
 
 # ---- Build taco and make sam graphs ----
 .PHONY: sam
@@ -109,18 +109,10 @@ endif
 # ---- Setup proper environment stuff ----
 suitesparse-formats: guard-SUITESPARSE_FORMATTED_PATH guard-SUITESPARSE_PATH
 	rm -rf ${SUITESPARSE_FORMATTED_PATH}/*
-	set -e && ./scripts/generate_suitesparse_formats.sh
+	set -e && ./scripts/formatting/generate_suitesparse_formats.sh
 
 frostt-formats: taco/build guard-FROSTT_FORMATTED_PATH guard-FROSTT_PATH
-	./scripts/generate_frostt_formats.sh
-
-.PHONY: env
-env:
-	export SUITESPARSE_PATH=/nobackup/owhsu/sparse-datasets/suitesparse/
-	export FROSTT_PATH=/nobackup/owhsu/sparse-datasets/frostt/
-	export SUITESPARSE_FORMATTED_PATH=/nobackup/owhsu/sparse-datasets/suitesparse-formatted
-	export FROSTT_FORMATTED_TACO_PATH=/nobackup/owhsu/sparse-datasets/frostt-formatted/taco-tensor
-	export FROSTT_FORMATTED_PATH=/nobackup/owhsu/sparse-datasets/frostt-formatted
+	./scripts/formatting/generate_frostt_formats.sh
 
 .PHONY: pydepends
 pydepends:

compiler/baseline.cpp

@@ -315,7 +315,8 @@ enum SuiteSparseOp {
     SDDMM = 4,
     MATTRANSMUL = 5,
     RESIDUAL = 6,
-    MMADD = 7
+    MMADD = 7,
+    MMMUL = 8
 };
 
 std::string opName(SuiteSparseOp op) {
@@ -341,6 +342,9 @@ std::string opName(SuiteSparseOp op) {
         case MMADD: {
             return "mmadd";
         }
+	case MMMUL: {
+	    return "mmmul"
+	}
         default:
             return "";
     }
@@ -467,6 +471,13 @@ static void bench_suitesparse(benchmark::State &state, SuiteSparseOp op, int fil
                 result(i, j) = ssTensor(i, j) + otherShifted(i, j);
                 break;
             }
+            case MMMUL: {
+                result = Tensor<int64_t>("result", ssTensor.getDimensions(), ssTensor.getFormat(), fill_value);
+
+                IndexVar i, j, k;
+                result(i, j) = ssTensor(i, j) * otherShifted(i, j);
+                break;
+            }
             case MATTRANSMUL: {
                 result = Tensor<int64_t>("result", {DIM1}, Format(Sparse), fill_value);
 
@@ -516,4 +527,5 @@ static void bench_suitesparse(benchmark::State &state, SuiteSparseOp op, int fil
     // TODO: need to fix for DCSC for this
     TACO_BENCH_ARGS(bench_suitesparse, mat_mattransmul, MATTRANSMUL);
     TACO_BENCH_ARGS(bench_suitesparse, matmul_spmm, SPMM);
+    TACO_BENCH_ARGS(bench_suitesparse, mat_elemmul, MMMUL);
 

compiler/sam-outputs/onyx-dot/mat_vecmul_ij.gv

@@ -3,6 +3,7 @@ digraph SAM {
     14 [comment="type=fiberlookup,index=i,tensor=B,mode=0,format=compressed,src=true,root=true" label="FiberLookup i: B0\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="B" mode="0" format="compressed" src="true" root="true"]
     13 [comment="type=broadcast" shape=point style=invis type="broadcast"]
     7 [comment="type=crddrop,outer=i,inner=j" label="CrdDrop i,j" color=orange shape=box style=filled type="crddrop" outer="i" inner="j"]
+    15 [comment="type=crddrop,outer=i,inner=j" label="CrdDrop i,j" color=orange shape=box style=filled type="crddrop" outer="i" inner="j"]
     1 [comment="type=fiberwrite,index=i,tensor=x,mode=0,format=compressed,segsize=2,crdsize=B0_dim,sink=true" label="FiberWrite i: x0\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="i" tensor="x" mode="0" format="compressed" segsize="2" crdsize="B0_dim" sink="true"]
     12 [comment="type=repsiggen,index=i" label="RepeatSignalGenerator i" color=cyan3 shape=box style=filled type="repsiggen" index="i"]
     11 [comment="type=repeat,index=i,tensor=c,root=true" label="Repeat i: c" color=cyan2 shape=box style=filled type="repeat" index="i" tensor="c" root="true"]
@@ -24,12 +25,16 @@ digraph SAM {
     8 -> 7 [label="crd_in-j" style=dashed type="crd" comment="in-j"]
     8 -> 5 [label="ref_out-B" style=bold type="ref" comment="out-B"]
     5 -> 4 [label="val" type="val"]
-    4 -> 3 [label="val" type="val"]
+    # 4 -> 3 [label="val" type="val"]
     3 -> 0 [label="val" type="val"]
     8 -> 6 [label="ref_out-c" style=bold type="ref" comment="out-c"]
     6 -> 4 [label="val" type="val"]
     10 -> 8 [label="ref_in-c" style=bold type="ref" comment="in-c"]
     14 -> 9 [label="ref" style=bold type="ref" comment=""]
     9 -> 8 [label="crd_in-B" style=dashed type="crd" comment="in-B"]
     9 -> 8 [label="ref_in-B" style=bold type="ref" comment="in-B"]
+
+    4 -> 15 [label="val_inner-j" type="val" comment="inner-j"]
+    15 -> 3 [label="val_inner-j" type="val" comment="inner-j"]
+    13 -> 15 [label="crd_i" style=dashed type="crd" comment="i"]
 }

count_nnz_tiling.py

@@ -0,0 +1,52 @@
+import glob
+
+
+def count_nonzeros(matrix_values_file):
+    with open(matrix_values_file, 'r') as values_file:
+        matrix_values = [float(val) for val in values_file.readlines()]
+
+    nonzeros = sum(1 for val in matrix_values if val != 0)
+
+    return nonzeros
+
+
+tile_dirs = glob.glob("SPARSE_TESTS/MAT_TMP_DIR/tile*")
+num_tiles = len(tile_dirs)
+limit = 900
+print("there are ", num_tiles, "tiles")
+
+sparsity_B = 0
+sparsity_C = 0
+# tilesize=int(sys.argv[1])**2
+tot_num_nonzeros = 0
+for tile_num in range(0, num_tiles):
+    tot_num_nonzeros = 0
+
+    tensor_C_values_file = f'SPARSE_TESTS/MAT_TMP_DIR/tile{tile_num}/tensor_C_mode_vals'
+
+    num_nonzeros = count_nonzeros(tensor_C_values_file)
+    tot_num_nonzeros += num_nonzeros
+    if num_nonzeros >= limit:
+        print("num_nonzeros: ", num_nonzeros)
+        print("error! too many nonzeros in INPUT matrices")
+        raise Exception
+
+    tensor_C_values_file = f'SPARSE_TESTS/MAT_TMP_DIR/tile{tile_num}/tensor_B_mode_vals'
+
+    num_nonzeros = count_nonzeros(tensor_C_values_file)
+    tot_num_nonzeros += num_nonzeros
+    if num_nonzeros >= limit:
+        print("num_nonzeros: ", num_nonzeros)
+        print("error! too many nonzeros in INPUT matrices")
+        raise Exception
+
+    if tot_num_nonzeros >= limit:
+        print("tot_num_nonzeros: ", tot_num_nonzeros)
+        print("error! too many nonzeros in matrices")
+        raise Exception
+
+sparsity_B /= num_tiles
+sparsity_C /= num_tiles
+
+print("sparsity_B: ", sparsity_B)
+print("sparsity_C: ", sparsity_C)

environment.yml

@@ -1,4 +1,4 @@
-name: aha
+name: sam
 channels:
   - pytorch
   - anaconda
@@ -201,4 +201,4 @@ dependencies:
     - statsmodels==0.13.2
     - threadpoolctl==3.1.0
     - traitlets==5.3.0
-prefix: /home/oliviahsu/miniconda3/envs/aha
+prefix: ~/Anaconda3/envs/sam

find_max_tilesize.py

@@ -0,0 +1,78 @@
+import os
+
+
+def write_to_line(file_path, line_number, new_content):
+    with open(file_path, "r") as file:
+        lines = file.readlines()
+
+    if line_number > len(lines) or line_number < 1:
+        # Line number is out of range
+        return
+
+    lines[line_number - 1] = new_content + "\n"
+
+    with open(file_path, "w") as file:
+        file.writelines(lines)
+
+
+def check_keyword_in_output(command, keyword):
+    # Run the command and redirect the output to a file
+    os.system(f"{command} > output.txt")
+
+    # Read the contents of the file
+    with open("output.txt", "r") as file:
+        output = file.read()
+
+    # Check if the keyword is present in the output
+    if keyword in output:
+        # Optionally, you can delete the output file
+        os.remove("output.txt")
+        return True
+    else:
+        # Optionally, you can delete the output file
+        os.remove("output.txt")
+        return False
+
+
+tile_size = 300
+step = 10
+
+for _ in range(20):
+    print("********************")
+    print("tile size: ", tile_size)
+    print("step: ", step)
+
+    yaml_file = "sam/sim/src/tiling/memory_config_onyx.yaml"
+    mem_tile_line = f"Mem_tile_size: {tile_size}"
+    print(mem_tile_line)
+    write_to_line(yaml_file, 19, mem_tile_line)
+
+    run_setup_script = "python3 setup_tiling_mat.py > temp.txt"
+    os.system(run_setup_script)
+    print(run_setup_script)
+
+    run_tile_pairing = "python3 tile_pairing.py > temp.txt"
+    os.system(run_tile_pairing)
+    print(run_tile_pairing)
+
+    run_count = "python3 count_nnz_tiling.py"
+    print(run_count)
+
+    if (check_keyword_in_output(run_count, "error")) is False:
+        tile_size += step
+        step *= 2
+    else:
+        print("****************Tile broken!")
+        tile_size -= step
+        step //= 2
+
+    if tile_size == 450:
+        break
+
+    if step == 0:
+        if _ >= 15:
+            step = 10
+        else:
+            break
+
+print("max tile size: ", tile_size)