From b014c06a780dfafcbe2d984b7ed754753b505048 Mon Sep 17 00:00:00 2001
From: Bo Wun Cheng <bwcheng@stanford.edu>
Date: Wed, 1 Nov 2023 11:33:07 -0700
Subject: [PATCH 1/6] added spmm kernel

---
 .../sam-outputs/onyx-dot/spmm_ijk_crddrop.gv  | 55 +++++++++++++++++
 .../sam-outputs/opal-dot/masked_broadcast.gv  | 27 +++++++++
 .../opal-dot/mat_elemadd_leakyrelu_exp.gv     | 41 +++++++++++++
 .../opal-dot/matmul_ijk_crddrop_relu.gv       | 60 +++++++++++++++++++
 .../sam-outputs/opal-dot/spmm_ijk_crddrop.gv  | 55 +++++++++++++++++
 .../opal-dot/trans_masked_broadcast.gv        | 27 +++++++++
 6 files changed, 265 insertions(+)
 create mode 100644 compiler/sam-outputs/onyx-dot/spmm_ijk_crddrop.gv
 create mode 100644 compiler/sam-outputs/opal-dot/masked_broadcast.gv
 create mode 100644 compiler/sam-outputs/opal-dot/mat_elemadd_leakyrelu_exp.gv
 create mode 100644 compiler/sam-outputs/opal-dot/matmul_ijk_crddrop_relu.gv
 create mode 100644 compiler/sam-outputs/opal-dot/spmm_ijk_crddrop.gv
 create mode 100644 compiler/sam-outputs/opal-dot/trans_masked_broadcast.gv

diff --git a/compiler/sam-outputs/onyx-dot/spmm_ijk_crddrop.gv b/compiler/sam-outputs/onyx-dot/spmm_ijk_crddrop.gv
new file mode 100644
index 00000000..4eaef89b
--- /dev/null
+++ b/compiler/sam-outputs/onyx-dot/spmm_ijk_crddrop.gv
@@ -0,0 +1,55 @@
+digraph SAM {
+    comment="X=ss01,B=dd01,C=ss10"
+    17 [comment="type=fiberlookup,index=i,tensor=B,mode=0,format=dense,src=true,root=true" label="FiberLookup i: B0\ndense" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="B" mode="0" format="dense" src="true" root="true"]
+    16 [comment="type=broadcast" shape=point style=invis type="broadcast"]
+    2 [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"]
+    15 [comment="type=repsiggen,index=i" label="RepeatSignalGenerator i" color=cyan3 shape=box style=filled type="repsiggen" index="i"]
+    14 [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"]
+    13 [comment="type=fiberlookup,index=j,tensor=C,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: C1\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="C" mode="1" format="compressed" src="true" root="false"]
+    12 [comment="type=broadcast" shape=point style=invis type="broadcast"]
+    1 [comment="type=fiberwrite,index=j,tensor=X,mode=1,format=compressed,segsize=B0_dim+1,crdsize=B0_dim*C1_dim,sink=true" label="FiberWrite j: X1\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="j" tensor="X" mode="1" format="compressed" segsize="B0_dim+1" crdsize="B0_dim*C1_dim" sink="true"]
+    11 [comment="type=repsiggen,index=j" label="RepeatSignalGenerator j" color=cyan3 shape=box style=filled type="repsiggen" index="j"]
+    10 [comment="type=repeat,index=j,tensor=B,root=false" label="Repeat j: B" color=cyan2 shape=box style=filled type="repeat" index="j" tensor="B" root="false"]
+    8 [comment="type=fiberlookup,index=k,tensor=B,mode=1,format=dense,src=true,root=false" label="FiberLookup k: B1\ndense" color=green4 shape=box style=filled type="fiberlookup" index="k" tensor="B" mode="1" format="dense" src="true" root="false"]
+    7 [comment="type=intersect,index=k" label="intersect k" color=purple shape=box style=filled type="intersect" index="k"]
+    5 [comment="type=arrayvals,tensor=B" label="Array Vals: B" color=green2 shape=box style=filled type="arrayvals" tensor="B"]
+    4 [comment="type=mul" label="Mul" color=brown shape=box style=filled type="mul"]
+    3 [comment="type=reduce" label="Reduce" color=brown shape=box style=filled type="reduce"]
+    0 [comment="type=fiberwrite,mode=vals,tensor=X,size=1*B0_dim*C1_dim,sink=true" label="FiberWrite Vals: X" color=green3 shape=box style=filled type="fiberwrite" tensor="X" mode="vals" size="1*B0_dim*C1_dim" sink="true"]
+    6 [comment="type=arrayvals,tensor=C" label="Array Vals: C" color=green2 shape=box style=filled type="arrayvals" tensor="C"]
+    9 [comment="type=fiberlookup,index=k,tensor=C,mode=0,format=compressed,src=true,root=false" label="FiberLookup k: C0\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="k" tensor="C" mode="0" format="compressed" src="true" root="false"]
+    18 [comment="type=crddrop,outer=i,inner=j" label="CrdDrop i,j" color=orange shape=box style=filled type="crddrop" outer="i" inner="j"]
+    19 [comment="type=crddrop,outer=j,inner=k" label="CrdDrop j,k" color=orange shape=box style=filled type="crddrop" outer="j" inner="k"]
+
+    17 -> 16 [label="crd" style=dashed type="crd" comment=""]
+    # 16 -> 2 [label="crd" style=dashed type="crd"]
+    16 -> 15 [label="crd" style=dashed type="crd"]
+    15 -> 14 [label="repsig" style=dotted type="repsig"]
+    14 -> 13 [label="ref" style=bold type="ref"]
+    13 -> 12 [label="crd" style=dashed type="crd" comment=""]
+    # 12 -> 1 [label="crd" style=dashed type="crd"]
+    12 -> 11 [label="crd" style=dashed type="crd"]
+    11 -> 10 [label="repsig" style=dotted type="repsig"]
+    10 -> 8 [label="ref" style=bold type="ref"]
+    8 -> 7 [label="crd_in-B" style=dashed type="crd" comment="in-B"]
+    7 -> 5 [label="ref_out-B" style=bold type="ref" comment="out-B"]
+    5 -> 4 [label="val" type="val"]
+    # 4 -> 3 [label="val" type="val"]
+    3 -> 0 [label="val" type="val"]
+    7 -> 6 [label="ref_out-C" style=bold type="ref" comment="out-C"]
+    6 -> 4 [label="val" type="val"]
+    8 -> 7 [label="ref_in-B" style=bold type="ref" comment="in-B"]
+    13 -> 9 [label="ref" style=bold type="ref" comment=""]
+    9 -> 7 [label="crd_in-C" style=dashed type="crd" comment="in-C"]
+    9 -> 7 [label="ref_in-C" style=bold type="ref" comment="in-C"]
+    17 -> 10 [label="ref" style=bold type="ref" comment=""]
+    
+    4 -> 19 [label="val_inner-k" type="val" comment="inner-k"]
+    12 -> 19 [label="crd_outer-j" style=dashed type="crd" comment="outer-j"]
+    18 -> 1 [label="crd_inner-j" style=dashed type="crd" comment="inner-j"]
+    19 -> 3 [label="val_inner-k" type="val" comment="inner-k"]
+
+    19 -> 18 [label="crd_inner-j" style=dashed type="crd" comment="inner-j"] 
+    16 -> 18 [label="crd_outer-i" style=dashed type="crd" comment="outer-i"] 
+    18 -> 2 [label="crd_outer-i" style=dashed type="crd" comment="outer-i"] 
+}
diff --git a/compiler/sam-outputs/opal-dot/masked_broadcast.gv b/compiler/sam-outputs/opal-dot/masked_broadcast.gv
new file mode 100644
index 00000000..86b0bdba
--- /dev/null
+++ b/compiler/sam-outputs/opal-dot/masked_broadcast.gv
@@ -0,0 +1,27 @@
+digraph SAM {
+    comment="X=ss01,B=ss01,c=s0"
+    0 [comment="type=fiberlookup,index=i,tensor=B,mode=0,format=compressed,src=true,root=true" label="FiberLookup i: B\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="B" mode="0" format="compressed" src="true" root="true"]
+    1 [comment="type=repsiggen,index=i" label="RepeatSignalGenerator i" color=cyan3 shape=box style=filled type="repsiggen" index="i"]
+    2 [comment="type=repeat,index=i,tensor=c,root=true" label="Repeat i: B" color=cyan2 shape=box style=filled type="repeat" index="i" tensor="c" root="true"]
+    3 [comment="type=fiberlookup,index=i,tensor=c,mode=0,format=compressed,src=true,root=false" label="FiberLookup i: c\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="c" mode="0" format="compressed" src="true" root="false"]
+    4 [comment="type=fiberlookup,index=j,tensor=B,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: B\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="B" mode="1" format="compressed" src="true" root="false"]
+    5 [comment="type=intersect,index=j" label="intersect j" color=purple shape=box style=filled type="intersect" index="j"]
+    6 [comment="type=arrayvals,tensor=c" label="Array Vals: c" color=green2 shape=box style=filled type="arrayvals" tensor="c"]
+    7 [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"]
+    8 [comment="type=fiberwrite,index=j,tensor=X,mode=1,format=compressed,segsize=B0_dim+1,crdsize=B0_dim*B1_dim,sink=true" label="FiberWrite j: X1\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="j" tensor="X" mode="1" format="compressed" segsize="B0_dim+1" crdsize="B0_dim*B1_dim" sink="true"]
+    9 [comment="type=fiberwrite,mode=vals,tensor=X,size=1*B0_dim*B1_dim,sink=true" label="FiberWrite Vals: X" color=green3 shape=box style=filled type="fiberwrite" tensor="X" mode="vals" size="1*B0_dim*B1_dim" sink="true"]
+    10 [comment="type=broadcast" shape=point style=invis type="broadcast"]
+    0 -> 10 [label="crd" style=dashed type="crd"]
+    10 -> 1 [label="crd" style=dashed type="crd"]
+    1 -> 2 [label="repsig" style=dotted type="repsig"]
+    2 -> 3 [label="ref" style=bold type="ref"]
+    0 -> 4 [label="ref" style=bold type="ref"]
+    3 -> 5 [label="ref_in-c" style=bold type="ref" comment="in-c"]
+    3 -> 5 [label="crd_in-c" style=dashed type="crd" comment="in-c"]
+    4 -> 5 [label="ref_in-B", style=bold type="ref" comment="in-B"]
+    4 -> 5 [label="crd_in-B", style=dashed type="crd" comment="in-B"]
+    5 -> 6 [label="ref_out-c", style=bold type="ref" comment="out-c"]
+    10 -> 7 [label="crd", style=dashed type="crd"]
+    5 -> 8 [label="crd", style=dashed type="crd"]
+    6 -> 9 [label="val", type="val"]
+}
diff --git a/compiler/sam-outputs/opal-dot/mat_elemadd_leakyrelu_exp.gv b/compiler/sam-outputs/opal-dot/mat_elemadd_leakyrelu_exp.gv
new file mode 100644
index 00000000..e9a9f489
--- /dev/null
+++ b/compiler/sam-outputs/opal-dot/mat_elemadd_leakyrelu_exp.gv
@@ -0,0 +1,41 @@
+digraph SAM {
+    comment="X=ss01,B=ss01,C=ss01"
+    10 [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"]
+    9 [comment="type=union,index=i" label="union i" color=purple shape=box style=filled type="union" index="i"]
+    2 [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"]
+    7 [comment="type=fiberlookup,index=j,tensor=B,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: B1\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="B" mode="1" format="compressed" src="true" root="false"]
+    6 [comment="type=union,index=j" label="union j" color=purple shape=box style=filled type="union" index="j"]
+    1 [comment="type=fiberwrite,index=j,tensor=X,mode=1,format=compressed,segsize=B0_dim+1,crdsize=B0_dim*B1_dim,sink=true" label="FiberWrite j: X1\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="j" tensor="X" mode="1" format="compressed" segsize="B0_dim+1" crdsize="B0_dim*B1_dim" sink="true"]
+    4 [comment="type=arrayvals,tensor=B" label="Array Vals: B" color=green2 shape=box style=filled type="arrayvals" tensor="B"]
+    3 [comment="type=add,sub=0" label="Add" color=brown shape=box style=filled type="add" sub="0"]
+    0 [comment="type=fiberwrite,mode=vals,tensor=X,size=1*B0_dim*B1_dim,sink=true" label="FiberWrite Vals: X" color=green3 shape=box style=filled type="fiberwrite" tensor="X" mode="vals" size="1*B0_dim*B1_dim" sink="true"]
+    5 [comment="type=arrayvals,tensor=C" label="Array Vals: C" color=green2 shape=box style=filled type="arrayvals" tensor="C"]
+    8 [comment="type=fiberlookup,index=j,tensor=C,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: C1\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="C" mode="1" format="compressed" src="true" root="false"]
+    11 [comment="type=fiberlookup,index=i,tensor=C,mode=0,format=compressed,src=true,root=true" label="FiberLookup i: C0\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="C" mode="0" format="compressed" src="true" root="true"]
+    12 [comment="type=broadcast" shape=point style=invis type="broadcast"]
+    13 [comment="type=mul,in2=0.2" label="Mul 0.2" color=brown shape=box style=filled type="mul" in2="0.2"]
+    14 [comment="type=max" label="Max" color=brown shape=box style=filled type="max"]
+    15 [comment="type=exp" label="Exp" color=brown shape=box style=filled type="exp"]
+    10 -> 9 [label="crd_in-B" style=dashed type="crd" comment="in-B"]
+    9 -> 2 [label="crd" style=dashed type="crd"]
+    9 -> 7 [label="ref_out-B" style=bold type="ref" comment="out-B"]
+    7 -> 6 [label="crd_in-B" style=dashed type="crd" comment="in-B"]
+    6 -> 1 [label="crd" style=dashed type="crd"]
+    6 -> 4 [label="ref_out-B" style=bold type="ref" comment="out-B"]
+    4 -> 3 [label="val" type="val"]
+    3 -> 12 [label="val" type="val"]
+    12 -> 13 [label="val" type="val"]
+    13 -> 14 [label="val" type="val"]
+    12 -> 14 [label="val" type="val"]
+    14 -> 15 [label="val" type="val"]
+    15 -> 0 [label="val" type="val"]
+    6 -> 5 [label="ref_out-C" style=bold type="ref" comment="out-C"]
+    5 -> 3 [label="val" type="val"]
+    7 -> 6 [label="ref_in-B" style=bold type="ref" comment="in-B"]
+    9 -> 8 [label="ref_out-C" style=bold type="ref" comment="out-C"]
+    8 -> 6 [label="crd_in-C" style=dashed type="crd" comment="in-C"]
+    8 -> 6 [label="ref_in-C" style=bold type="ref" comment="in-C"]
+    10 -> 9 [label="ref_in-B" style=bold type="ref" comment="in-B"]
+    11 -> 9 [label="crd_in-C" style=dashed type="crd" comment="in-C"]
+    11 -> 9 [label="ref_in-C" style=bold type="ref" comment="in-C"]
+}
diff --git a/compiler/sam-outputs/opal-dot/matmul_ijk_crddrop_relu.gv b/compiler/sam-outputs/opal-dot/matmul_ijk_crddrop_relu.gv
new file mode 100644
index 00000000..5625ca41
--- /dev/null
+++ b/compiler/sam-outputs/opal-dot/matmul_ijk_crddrop_relu.gv
@@ -0,0 +1,60 @@
+digraph SAM {
+    comment="X=ss01,B=ss01,C=ss10"
+    17 [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"]
+    16 [comment="type=broadcast" shape=point style=invis type="broadcast"]
+    2 [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"]
+    15 [comment="type=repsiggen,index=i" label="RepeatSignalGenerator i" color=cyan3 shape=box style=filled type="repsiggen" index="i"]
+    14 [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"]
+    13 [comment="type=fiberlookup,index=j,tensor=C,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: C1\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="C" mode="1" format="compressed" src="true" root="false"]
+    12 [comment="type=broadcast" shape=point style=invis type="broadcast"]
+    1 [comment="type=fiberwrite,index=j,tensor=X,mode=1,format=compressed,segsize=B0_dim+1,crdsize=B0_dim*C1_dim,sink=true" label="FiberWrite j: X1\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="j" tensor="X" mode="1" format="compressed" segsize="B0_dim+1" crdsize="B0_dim*C1_dim" sink="true"]
+    11 [comment="type=repsiggen,index=j" label="RepeatSignalGenerator j" color=cyan3 shape=box style=filled type="repsiggen" index="j"]
+    10 [comment="type=repeat,index=j,tensor=B,root=false" label="Repeat j: B" color=cyan2 shape=box style=filled type="repeat" index="j" tensor="B" root="false"]
+    8 [comment="type=fiberlookup,index=k,tensor=B,mode=1,format=compressed,src=true,root=false" label="FiberLookup k: B1\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="k" tensor="B" mode="1" format="compressed" src="true" root="false"]
+    7 [comment="type=intersect,index=k" label="intersect k" color=purple shape=box style=filled type="intersect" index="k"]
+    5 [comment="type=arrayvals,tensor=B" label="Array Vals: B" color=green2 shape=box style=filled type="arrayvals" tensor="B"]
+    4 [comment="type=mul" label="Mul" color=brown shape=box style=filled type="mul"]
+    3 [comment="type=reduce" label="Reduce" color=brown shape=box style=filled type="reduce"]
+    0 [comment="type=fiberwrite,mode=vals,tensor=X,size=1*B0_dim*C1_dim,sink=true" label="FiberWrite Vals: X" color=green3 shape=box style=filled type="fiberwrite" tensor="X" mode="vals" size="1*B0_dim*C1_dim" sink="true"]
+    6 [comment="type=arrayvals,tensor=C" label="Array Vals: C" color=green2 shape=box style=filled type="arrayvals" tensor="C"]
+    9 [comment="type=fiberlookup,index=k,tensor=C,mode=0,format=compressed,src=true,root=false" label="FiberLookup k: C0\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="k" tensor="C" mode="0" format="compressed" src="true" root="false"]
+    18 [comment="type=crddrop,outer=i,inner=j" label="CrdDrop i,j" color=orange shape=box style=filled type="crddrop" outer="i" inner="j"]
+    19 [comment="type=crddrop,outer=j,inner=k" label="CrdDrop j,k" color=orange shape=box style=filled type="crddrop" outer="j" inner="k"]
+    20 [comment="type=max" label="Max 0" color=brown shape=box style=filled type="max"]
+    21 [comment="type=crddrop,outer=j,inner=val,mode=0" label="CrdDrop Compression j, val" color=orange style=filled type="crddrop" outer="j" inner="val" mode="0"]
+    22 [comment="type=crddrop,outer=i,inner=j" label="CrdDrop i,j" color=orange shape=box style=filled type="crddrop" outer="i" inner="j"]
+    17 -> 16 [label="crd" style=dashed type="crd" comment=""]
+    16 -> 15 [label="crd" style=dashed type="crd"]
+    15 -> 14 [label="repsig" style=dotted type="repsig"]
+    14 -> 13 [label="ref" style=bold type="ref"]
+    13 -> 12 [label="crd" style=dashed type="crd" comment=""]
+    12 -> 11 [label="crd" style=dashed type="crd"]
+    11 -> 10 [label="repsig" style=dotted type="repsig"]
+    10 -> 8 [label="ref" style=bold type="ref"]
+    8 -> 7 [label="crd_in-B" style=dashed type="crd" comment="in-B"]
+    7 -> 5 [label="ref_out-B" style=bold type="ref" comment="out-B"]
+    5 -> 4 [label="val" type="val"]
+    7 -> 6 [label="ref_out-C" style=bold type="ref" comment="out-C"]
+    6 -> 4 [label="val" type="val"]
+    8 -> 7 [label="ref_in-B" style=bold type="ref" comment="in-B"]
+    13 -> 9 [label="ref" style=bold type="ref" comment=""]
+    9 -> 7 [label="crd_in-C" style=dashed type="crd" comment="in-C"]
+    9 -> 7 [label="ref_in-C" style=bold type="ref" comment="in-C"]
+    17 -> 10 [label="ref" style=bold type="ref" comment=""]
+    
+    4 -> 19 [label="val_inner-k" type="val" comment="inner-k"]
+    12 -> 19 [label="crd_outer-j" style=dashed type="crd" comment="outer-j"]
+    19 -> 3 [label="val_inner-k" type="val" comment="inner-k"]
+    3 -> 20 [label="val" type="val" comment="val"]
+    20 -> 21 [label="val" type="val" comment="inner-val"]
+    18 -> 21 [label="crd_inner-j" style=dashed type="crd" comment="outer-j"]
+    21 -> 22 [label="crd_inner-j" style=dashed type="crd" comment="outer-j"]
+    18 -> 22 [label="crd_outer-i" style=dashed type="crd" comment="outer-i"] 
+
+    19 -> 18 [label="crd_inner-j" style=dashed type="crd" comment="inner-j"] 
+    16 -> 18 [label="crd_outer-i" style=dashed type="crd" comment="outer-i"]
+    21 -> 0 [label="val" type="val" comment="inner-val"]
+    
+    22 -> 2 [label="crd_outer-i" style=dashed type="crd" comment="outer-i"] 
+    22 -> 1 [label="crd_inner-j" style=dashed type="crd" comment="inner-j"]
+}
diff --git a/compiler/sam-outputs/opal-dot/spmm_ijk_crddrop.gv b/compiler/sam-outputs/opal-dot/spmm_ijk_crddrop.gv
new file mode 100644
index 00000000..4eaef89b
--- /dev/null
+++ b/compiler/sam-outputs/opal-dot/spmm_ijk_crddrop.gv
@@ -0,0 +1,55 @@
+digraph SAM {
+    comment="X=ss01,B=dd01,C=ss10"
+    17 [comment="type=fiberlookup,index=i,tensor=B,mode=0,format=dense,src=true,root=true" label="FiberLookup i: B0\ndense" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="B" mode="0" format="dense" src="true" root="true"]
+    16 [comment="type=broadcast" shape=point style=invis type="broadcast"]
+    2 [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"]
+    15 [comment="type=repsiggen,index=i" label="RepeatSignalGenerator i" color=cyan3 shape=box style=filled type="repsiggen" index="i"]
+    14 [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"]
+    13 [comment="type=fiberlookup,index=j,tensor=C,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: C1\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="C" mode="1" format="compressed" src="true" root="false"]
+    12 [comment="type=broadcast" shape=point style=invis type="broadcast"]
+    1 [comment="type=fiberwrite,index=j,tensor=X,mode=1,format=compressed,segsize=B0_dim+1,crdsize=B0_dim*C1_dim,sink=true" label="FiberWrite j: X1\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="j" tensor="X" mode="1" format="compressed" segsize="B0_dim+1" crdsize="B0_dim*C1_dim" sink="true"]
+    11 [comment="type=repsiggen,index=j" label="RepeatSignalGenerator j" color=cyan3 shape=box style=filled type="repsiggen" index="j"]
+    10 [comment="type=repeat,index=j,tensor=B,root=false" label="Repeat j: B" color=cyan2 shape=box style=filled type="repeat" index="j" tensor="B" root="false"]
+    8 [comment="type=fiberlookup,index=k,tensor=B,mode=1,format=dense,src=true,root=false" label="FiberLookup k: B1\ndense" color=green4 shape=box style=filled type="fiberlookup" index="k" tensor="B" mode="1" format="dense" src="true" root="false"]
+    7 [comment="type=intersect,index=k" label="intersect k" color=purple shape=box style=filled type="intersect" index="k"]
+    5 [comment="type=arrayvals,tensor=B" label="Array Vals: B" color=green2 shape=box style=filled type="arrayvals" tensor="B"]
+    4 [comment="type=mul" label="Mul" color=brown shape=box style=filled type="mul"]
+    3 [comment="type=reduce" label="Reduce" color=brown shape=box style=filled type="reduce"]
+    0 [comment="type=fiberwrite,mode=vals,tensor=X,size=1*B0_dim*C1_dim,sink=true" label="FiberWrite Vals: X" color=green3 shape=box style=filled type="fiberwrite" tensor="X" mode="vals" size="1*B0_dim*C1_dim" sink="true"]
+    6 [comment="type=arrayvals,tensor=C" label="Array Vals: C" color=green2 shape=box style=filled type="arrayvals" tensor="C"]
+    9 [comment="type=fiberlookup,index=k,tensor=C,mode=0,format=compressed,src=true,root=false" label="FiberLookup k: C0\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="k" tensor="C" mode="0" format="compressed" src="true" root="false"]
+    18 [comment="type=crddrop,outer=i,inner=j" label="CrdDrop i,j" color=orange shape=box style=filled type="crddrop" outer="i" inner="j"]
+    19 [comment="type=crddrop,outer=j,inner=k" label="CrdDrop j,k" color=orange shape=box style=filled type="crddrop" outer="j" inner="k"]
+
+    17 -> 16 [label="crd" style=dashed type="crd" comment=""]
+    # 16 -> 2 [label="crd" style=dashed type="crd"]
+    16 -> 15 [label="crd" style=dashed type="crd"]
+    15 -> 14 [label="repsig" style=dotted type="repsig"]
+    14 -> 13 [label="ref" style=bold type="ref"]
+    13 -> 12 [label="crd" style=dashed type="crd" comment=""]
+    # 12 -> 1 [label="crd" style=dashed type="crd"]
+    12 -> 11 [label="crd" style=dashed type="crd"]
+    11 -> 10 [label="repsig" style=dotted type="repsig"]
+    10 -> 8 [label="ref" style=bold type="ref"]
+    8 -> 7 [label="crd_in-B" style=dashed type="crd" comment="in-B"]
+    7 -> 5 [label="ref_out-B" style=bold type="ref" comment="out-B"]
+    5 -> 4 [label="val" type="val"]
+    # 4 -> 3 [label="val" type="val"]
+    3 -> 0 [label="val" type="val"]
+    7 -> 6 [label="ref_out-C" style=bold type="ref" comment="out-C"]
+    6 -> 4 [label="val" type="val"]
+    8 -> 7 [label="ref_in-B" style=bold type="ref" comment="in-B"]
+    13 -> 9 [label="ref" style=bold type="ref" comment=""]
+    9 -> 7 [label="crd_in-C" style=dashed type="crd" comment="in-C"]
+    9 -> 7 [label="ref_in-C" style=bold type="ref" comment="in-C"]
+    17 -> 10 [label="ref" style=bold type="ref" comment=""]
+    
+    4 -> 19 [label="val_inner-k" type="val" comment="inner-k"]
+    12 -> 19 [label="crd_outer-j" style=dashed type="crd" comment="outer-j"]
+    18 -> 1 [label="crd_inner-j" style=dashed type="crd" comment="inner-j"]
+    19 -> 3 [label="val_inner-k" type="val" comment="inner-k"]
+
+    19 -> 18 [label="crd_inner-j" style=dashed type="crd" comment="inner-j"] 
+    16 -> 18 [label="crd_outer-i" style=dashed type="crd" comment="outer-i"] 
+    18 -> 2 [label="crd_outer-i" style=dashed type="crd" comment="outer-i"] 
+}
diff --git a/compiler/sam-outputs/opal-dot/trans_masked_broadcast.gv b/compiler/sam-outputs/opal-dot/trans_masked_broadcast.gv
new file mode 100644
index 00000000..06d63c95
--- /dev/null
+++ b/compiler/sam-outputs/opal-dot/trans_masked_broadcast.gv
@@ -0,0 +1,27 @@
+digraph SAM {
+    comment="X=ss01,B=ss01,c=s0"
+    0 [comment="type=fiberlookup,index=i,tensor=B,mode=0,format=compressed,src=true,root=true" label="FiberLookup i: B\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="B" mode="0" format="compressed" src="true" root="true"]
+    1 [comment="type=fiberlookup,index=i,tensor=c,mode=0,format=compressed,src=true,root=true" label="FiberLookup i: c\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="c" mode="0" format="compressed" src="true" root="true"]
+    2 [comment="type=intersect,index=i" label="intersect i" color=purple shape=box style=filled type="intersect" index="i"]
+    3 [comment="type=fiberlookup,index=j,tensor=B,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: B\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="B" mode="1" format="compressed" src="true" root="false"]
+    4 [comment="type=repsiggen,index=j" label="RepeatSignalGenerator j" color=cyan3 shape=box style=filled type="repsiggen" index="j"]
+    5 [comment="type=repeat,index=j,tensor=c,root=false" label="Repeat j: B" color=cyan2 shape=box style=filled type="repeat" index="j" tensor="c" root="false"]
+    6 [comment="type=arrayvals,tensor=c" label="Array Vals: c" color=green2 shape=box style=filled type="arrayvals" tensor="c"]
+    7 [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"]
+    8 [comment="type=fiberwrite,index=j,tensor=X,mode=1,format=compressed,segsize=B0_dim+1,crdsize=B0_dim*B1_dim,sink=true" label="FiberWrite j: X1\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="j" tensor="X" mode="1" format="compressed" segsize="B0_dim+1" crdsize="B0_dim*B1_dim" sink="true"]
+    9 [comment="type=fiberwrite,mode=vals,tensor=X,size=1*B0_dim*B1_dim,sink=true" label="FiberWrite Vals: X" color=green3 shape=box style=filled type="fiberwrite" tensor="X" mode="vals" size="1*B0_dim*B1_dim" sink="true"]
+    10 [comment="type=broadcast" shape=point style=invis type="broadcast"]
+    0 -> 2 [label="ref_in-B" style=bold type="ref" comment="in-B"]
+    0 -> 2 [label="crd_in-B" style=dashed type="crd" comment="in-B"]
+    1 -> 2 [label="ref_in-c" style=bold type="ref" comment="in-c"]
+    1 -> 2 [label="crd_in-c" style=dashed type="crd" comment="in-c"]
+    2 -> 3 [label="ref_out-B" style=bold type="ref" comment="out-B"]
+    2 -> 7 [label="crd" style=dashed type="crd"]
+    3 -> 10 [label="crd" style=dashed type="crd"]
+    10 -> 4 [label="crd" style=dashed type="crd"]
+    10 -> 8 [label="crd" style=dashed type="crd"]
+    4 -> 5 [label="repsig" style=dotted type="repsig"]
+    2 -> 5 [label="ref_out-c" style=bold type="ref" comment="out-c"]
+    5 -> 6 [label="ref" style=bold type="ref"]
+    6 -> 9 [label="val" type="val"]
+}

From 1cb638a04cdfe99c3fcbdd7acb5c50f592a30c1c Mon Sep 17 00:00:00 2001
From: Bo Wun Cheng <bwcheng@stanford.edu>
Date: Wed, 1 Nov 2023 11:33:38 -0700
Subject: [PATCH 2/6] remove unwanted benchmarks

---
 .../sam-outputs/opal-dot/masked_broadcast.gv  | 27 ---------
 .../opal-dot/mat_elemadd_leakyrelu_exp.gv     | 41 -------------
 .../opal-dot/matmul_ijk_crddrop_relu.gv       | 60 -------------------
 .../sam-outputs/opal-dot/spmm_ijk_crddrop.gv  | 55 -----------------
 .../opal-dot/trans_masked_broadcast.gv        | 27 ---------
 5 files changed, 210 deletions(-)
 delete mode 100644 compiler/sam-outputs/opal-dot/masked_broadcast.gv
 delete mode 100644 compiler/sam-outputs/opal-dot/mat_elemadd_leakyrelu_exp.gv
 delete mode 100644 compiler/sam-outputs/opal-dot/matmul_ijk_crddrop_relu.gv
 delete mode 100644 compiler/sam-outputs/opal-dot/spmm_ijk_crddrop.gv
 delete mode 100644 compiler/sam-outputs/opal-dot/trans_masked_broadcast.gv

diff --git a/compiler/sam-outputs/opal-dot/masked_broadcast.gv b/compiler/sam-outputs/opal-dot/masked_broadcast.gv
deleted file mode 100644
index 86b0bdba..00000000
--- a/compiler/sam-outputs/opal-dot/masked_broadcast.gv
+++ /dev/null
@@ -1,27 +0,0 @@
-digraph SAM {
-    comment="X=ss01,B=ss01,c=s0"
-    0 [comment="type=fiberlookup,index=i,tensor=B,mode=0,format=compressed,src=true,root=true" label="FiberLookup i: B\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="B" mode="0" format="compressed" src="true" root="true"]
-    1 [comment="type=repsiggen,index=i" label="RepeatSignalGenerator i" color=cyan3 shape=box style=filled type="repsiggen" index="i"]
-    2 [comment="type=repeat,index=i,tensor=c,root=true" label="Repeat i: B" color=cyan2 shape=box style=filled type="repeat" index="i" tensor="c" root="true"]
-    3 [comment="type=fiberlookup,index=i,tensor=c,mode=0,format=compressed,src=true,root=false" label="FiberLookup i: c\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="c" mode="0" format="compressed" src="true" root="false"]
-    4 [comment="type=fiberlookup,index=j,tensor=B,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: B\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="B" mode="1" format="compressed" src="true" root="false"]
-    5 [comment="type=intersect,index=j" label="intersect j" color=purple shape=box style=filled type="intersect" index="j"]
-    6 [comment="type=arrayvals,tensor=c" label="Array Vals: c" color=green2 shape=box style=filled type="arrayvals" tensor="c"]
-    7 [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"]
-    8 [comment="type=fiberwrite,index=j,tensor=X,mode=1,format=compressed,segsize=B0_dim+1,crdsize=B0_dim*B1_dim,sink=true" label="FiberWrite j: X1\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="j" tensor="X" mode="1" format="compressed" segsize="B0_dim+1" crdsize="B0_dim*B1_dim" sink="true"]
-    9 [comment="type=fiberwrite,mode=vals,tensor=X,size=1*B0_dim*B1_dim,sink=true" label="FiberWrite Vals: X" color=green3 shape=box style=filled type="fiberwrite" tensor="X" mode="vals" size="1*B0_dim*B1_dim" sink="true"]
-    10 [comment="type=broadcast" shape=point style=invis type="broadcast"]
-    0 -> 10 [label="crd" style=dashed type="crd"]
-    10 -> 1 [label="crd" style=dashed type="crd"]
-    1 -> 2 [label="repsig" style=dotted type="repsig"]
-    2 -> 3 [label="ref" style=bold type="ref"]
-    0 -> 4 [label="ref" style=bold type="ref"]
-    3 -> 5 [label="ref_in-c" style=bold type="ref" comment="in-c"]
-    3 -> 5 [label="crd_in-c" style=dashed type="crd" comment="in-c"]
-    4 -> 5 [label="ref_in-B", style=bold type="ref" comment="in-B"]
-    4 -> 5 [label="crd_in-B", style=dashed type="crd" comment="in-B"]
-    5 -> 6 [label="ref_out-c", style=bold type="ref" comment="out-c"]
-    10 -> 7 [label="crd", style=dashed type="crd"]
-    5 -> 8 [label="crd", style=dashed type="crd"]
-    6 -> 9 [label="val", type="val"]
-}
diff --git a/compiler/sam-outputs/opal-dot/mat_elemadd_leakyrelu_exp.gv b/compiler/sam-outputs/opal-dot/mat_elemadd_leakyrelu_exp.gv
deleted file mode 100644
index e9a9f489..00000000
--- a/compiler/sam-outputs/opal-dot/mat_elemadd_leakyrelu_exp.gv
+++ /dev/null
@@ -1,41 +0,0 @@
-digraph SAM {
-    comment="X=ss01,B=ss01,C=ss01"
-    10 [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"]
-    9 [comment="type=union,index=i" label="union i" color=purple shape=box style=filled type="union" index="i"]
-    2 [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"]
-    7 [comment="type=fiberlookup,index=j,tensor=B,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: B1\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="B" mode="1" format="compressed" src="true" root="false"]
-    6 [comment="type=union,index=j" label="union j" color=purple shape=box style=filled type="union" index="j"]
-    1 [comment="type=fiberwrite,index=j,tensor=X,mode=1,format=compressed,segsize=B0_dim+1,crdsize=B0_dim*B1_dim,sink=true" label="FiberWrite j: X1\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="j" tensor="X" mode="1" format="compressed" segsize="B0_dim+1" crdsize="B0_dim*B1_dim" sink="true"]
-    4 [comment="type=arrayvals,tensor=B" label="Array Vals: B" color=green2 shape=box style=filled type="arrayvals" tensor="B"]
-    3 [comment="type=add,sub=0" label="Add" color=brown shape=box style=filled type="add" sub="0"]
-    0 [comment="type=fiberwrite,mode=vals,tensor=X,size=1*B0_dim*B1_dim,sink=true" label="FiberWrite Vals: X" color=green3 shape=box style=filled type="fiberwrite" tensor="X" mode="vals" size="1*B0_dim*B1_dim" sink="true"]
-    5 [comment="type=arrayvals,tensor=C" label="Array Vals: C" color=green2 shape=box style=filled type="arrayvals" tensor="C"]
-    8 [comment="type=fiberlookup,index=j,tensor=C,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: C1\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="C" mode="1" format="compressed" src="true" root="false"]
-    11 [comment="type=fiberlookup,index=i,tensor=C,mode=0,format=compressed,src=true,root=true" label="FiberLookup i: C0\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="C" mode="0" format="compressed" src="true" root="true"]
-    12 [comment="type=broadcast" shape=point style=invis type="broadcast"]
-    13 [comment="type=mul,in2=0.2" label="Mul 0.2" color=brown shape=box style=filled type="mul" in2="0.2"]
-    14 [comment="type=max" label="Max" color=brown shape=box style=filled type="max"]
-    15 [comment="type=exp" label="Exp" color=brown shape=box style=filled type="exp"]
-    10 -> 9 [label="crd_in-B" style=dashed type="crd" comment="in-B"]
-    9 -> 2 [label="crd" style=dashed type="crd"]
-    9 -> 7 [label="ref_out-B" style=bold type="ref" comment="out-B"]
-    7 -> 6 [label="crd_in-B" style=dashed type="crd" comment="in-B"]
-    6 -> 1 [label="crd" style=dashed type="crd"]
-    6 -> 4 [label="ref_out-B" style=bold type="ref" comment="out-B"]
-    4 -> 3 [label="val" type="val"]
-    3 -> 12 [label="val" type="val"]
-    12 -> 13 [label="val" type="val"]
-    13 -> 14 [label="val" type="val"]
-    12 -> 14 [label="val" type="val"]
-    14 -> 15 [label="val" type="val"]
-    15 -> 0 [label="val" type="val"]
-    6 -> 5 [label="ref_out-C" style=bold type="ref" comment="out-C"]
-    5 -> 3 [label="val" type="val"]
-    7 -> 6 [label="ref_in-B" style=bold type="ref" comment="in-B"]
-    9 -> 8 [label="ref_out-C" style=bold type="ref" comment="out-C"]
-    8 -> 6 [label="crd_in-C" style=dashed type="crd" comment="in-C"]
-    8 -> 6 [label="ref_in-C" style=bold type="ref" comment="in-C"]
-    10 -> 9 [label="ref_in-B" style=bold type="ref" comment="in-B"]
-    11 -> 9 [label="crd_in-C" style=dashed type="crd" comment="in-C"]
-    11 -> 9 [label="ref_in-C" style=bold type="ref" comment="in-C"]
-}
diff --git a/compiler/sam-outputs/opal-dot/matmul_ijk_crddrop_relu.gv b/compiler/sam-outputs/opal-dot/matmul_ijk_crddrop_relu.gv
deleted file mode 100644
index 5625ca41..00000000
--- a/compiler/sam-outputs/opal-dot/matmul_ijk_crddrop_relu.gv
+++ /dev/null
@@ -1,60 +0,0 @@
-digraph SAM {
-    comment="X=ss01,B=ss01,C=ss10"
-    17 [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"]
-    16 [comment="type=broadcast" shape=point style=invis type="broadcast"]
-    2 [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"]
-    15 [comment="type=repsiggen,index=i" label="RepeatSignalGenerator i" color=cyan3 shape=box style=filled type="repsiggen" index="i"]
-    14 [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"]
-    13 [comment="type=fiberlookup,index=j,tensor=C,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: C1\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="C" mode="1" format="compressed" src="true" root="false"]
-    12 [comment="type=broadcast" shape=point style=invis type="broadcast"]
-    1 [comment="type=fiberwrite,index=j,tensor=X,mode=1,format=compressed,segsize=B0_dim+1,crdsize=B0_dim*C1_dim,sink=true" label="FiberWrite j: X1\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="j" tensor="X" mode="1" format="compressed" segsize="B0_dim+1" crdsize="B0_dim*C1_dim" sink="true"]
-    11 [comment="type=repsiggen,index=j" label="RepeatSignalGenerator j" color=cyan3 shape=box style=filled type="repsiggen" index="j"]
-    10 [comment="type=repeat,index=j,tensor=B,root=false" label="Repeat j: B" color=cyan2 shape=box style=filled type="repeat" index="j" tensor="B" root="false"]
-    8 [comment="type=fiberlookup,index=k,tensor=B,mode=1,format=compressed,src=true,root=false" label="FiberLookup k: B1\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="k" tensor="B" mode="1" format="compressed" src="true" root="false"]
-    7 [comment="type=intersect,index=k" label="intersect k" color=purple shape=box style=filled type="intersect" index="k"]
-    5 [comment="type=arrayvals,tensor=B" label="Array Vals: B" color=green2 shape=box style=filled type="arrayvals" tensor="B"]
-    4 [comment="type=mul" label="Mul" color=brown shape=box style=filled type="mul"]
-    3 [comment="type=reduce" label="Reduce" color=brown shape=box style=filled type="reduce"]
-    0 [comment="type=fiberwrite,mode=vals,tensor=X,size=1*B0_dim*C1_dim,sink=true" label="FiberWrite Vals: X" color=green3 shape=box style=filled type="fiberwrite" tensor="X" mode="vals" size="1*B0_dim*C1_dim" sink="true"]
-    6 [comment="type=arrayvals,tensor=C" label="Array Vals: C" color=green2 shape=box style=filled type="arrayvals" tensor="C"]
-    9 [comment="type=fiberlookup,index=k,tensor=C,mode=0,format=compressed,src=true,root=false" label="FiberLookup k: C0\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="k" tensor="C" mode="0" format="compressed" src="true" root="false"]
-    18 [comment="type=crddrop,outer=i,inner=j" label="CrdDrop i,j" color=orange shape=box style=filled type="crddrop" outer="i" inner="j"]
-    19 [comment="type=crddrop,outer=j,inner=k" label="CrdDrop j,k" color=orange shape=box style=filled type="crddrop" outer="j" inner="k"]
-    20 [comment="type=max" label="Max 0" color=brown shape=box style=filled type="max"]
-    21 [comment="type=crddrop,outer=j,inner=val,mode=0" label="CrdDrop Compression j, val" color=orange style=filled type="crddrop" outer="j" inner="val" mode="0"]
-    22 [comment="type=crddrop,outer=i,inner=j" label="CrdDrop i,j" color=orange shape=box style=filled type="crddrop" outer="i" inner="j"]
-    17 -> 16 [label="crd" style=dashed type="crd" comment=""]
-    16 -> 15 [label="crd" style=dashed type="crd"]
-    15 -> 14 [label="repsig" style=dotted type="repsig"]
-    14 -> 13 [label="ref" style=bold type="ref"]
-    13 -> 12 [label="crd" style=dashed type="crd" comment=""]
-    12 -> 11 [label="crd" style=dashed type="crd"]
-    11 -> 10 [label="repsig" style=dotted type="repsig"]
-    10 -> 8 [label="ref" style=bold type="ref"]
-    8 -> 7 [label="crd_in-B" style=dashed type="crd" comment="in-B"]
-    7 -> 5 [label="ref_out-B" style=bold type="ref" comment="out-B"]
-    5 -> 4 [label="val" type="val"]
-    7 -> 6 [label="ref_out-C" style=bold type="ref" comment="out-C"]
-    6 -> 4 [label="val" type="val"]
-    8 -> 7 [label="ref_in-B" style=bold type="ref" comment="in-B"]
-    13 -> 9 [label="ref" style=bold type="ref" comment=""]
-    9 -> 7 [label="crd_in-C" style=dashed type="crd" comment="in-C"]
-    9 -> 7 [label="ref_in-C" style=bold type="ref" comment="in-C"]
-    17 -> 10 [label="ref" style=bold type="ref" comment=""]
-    
-    4 -> 19 [label="val_inner-k" type="val" comment="inner-k"]
-    12 -> 19 [label="crd_outer-j" style=dashed type="crd" comment="outer-j"]
-    19 -> 3 [label="val_inner-k" type="val" comment="inner-k"]
-    3 -> 20 [label="val" type="val" comment="val"]
-    20 -> 21 [label="val" type="val" comment="inner-val"]
-    18 -> 21 [label="crd_inner-j" style=dashed type="crd" comment="outer-j"]
-    21 -> 22 [label="crd_inner-j" style=dashed type="crd" comment="outer-j"]
-    18 -> 22 [label="crd_outer-i" style=dashed type="crd" comment="outer-i"] 
-
-    19 -> 18 [label="crd_inner-j" style=dashed type="crd" comment="inner-j"] 
-    16 -> 18 [label="crd_outer-i" style=dashed type="crd" comment="outer-i"]
-    21 -> 0 [label="val" type="val" comment="inner-val"]
-    
-    22 -> 2 [label="crd_outer-i" style=dashed type="crd" comment="outer-i"] 
-    22 -> 1 [label="crd_inner-j" style=dashed type="crd" comment="inner-j"]
-}
diff --git a/compiler/sam-outputs/opal-dot/spmm_ijk_crddrop.gv b/compiler/sam-outputs/opal-dot/spmm_ijk_crddrop.gv
deleted file mode 100644
index 4eaef89b..00000000
--- a/compiler/sam-outputs/opal-dot/spmm_ijk_crddrop.gv
+++ /dev/null
@@ -1,55 +0,0 @@
-digraph SAM {
-    comment="X=ss01,B=dd01,C=ss10"
-    17 [comment="type=fiberlookup,index=i,tensor=B,mode=0,format=dense,src=true,root=true" label="FiberLookup i: B0\ndense" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="B" mode="0" format="dense" src="true" root="true"]
-    16 [comment="type=broadcast" shape=point style=invis type="broadcast"]
-    2 [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"]
-    15 [comment="type=repsiggen,index=i" label="RepeatSignalGenerator i" color=cyan3 shape=box style=filled type="repsiggen" index="i"]
-    14 [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"]
-    13 [comment="type=fiberlookup,index=j,tensor=C,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: C1\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="C" mode="1" format="compressed" src="true" root="false"]
-    12 [comment="type=broadcast" shape=point style=invis type="broadcast"]
-    1 [comment="type=fiberwrite,index=j,tensor=X,mode=1,format=compressed,segsize=B0_dim+1,crdsize=B0_dim*C1_dim,sink=true" label="FiberWrite j: X1\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="j" tensor="X" mode="1" format="compressed" segsize="B0_dim+1" crdsize="B0_dim*C1_dim" sink="true"]
-    11 [comment="type=repsiggen,index=j" label="RepeatSignalGenerator j" color=cyan3 shape=box style=filled type="repsiggen" index="j"]
-    10 [comment="type=repeat,index=j,tensor=B,root=false" label="Repeat j: B" color=cyan2 shape=box style=filled type="repeat" index="j" tensor="B" root="false"]
-    8 [comment="type=fiberlookup,index=k,tensor=B,mode=1,format=dense,src=true,root=false" label="FiberLookup k: B1\ndense" color=green4 shape=box style=filled type="fiberlookup" index="k" tensor="B" mode="1" format="dense" src="true" root="false"]
-    7 [comment="type=intersect,index=k" label="intersect k" color=purple shape=box style=filled type="intersect" index="k"]
-    5 [comment="type=arrayvals,tensor=B" label="Array Vals: B" color=green2 shape=box style=filled type="arrayvals" tensor="B"]
-    4 [comment="type=mul" label="Mul" color=brown shape=box style=filled type="mul"]
-    3 [comment="type=reduce" label="Reduce" color=brown shape=box style=filled type="reduce"]
-    0 [comment="type=fiberwrite,mode=vals,tensor=X,size=1*B0_dim*C1_dim,sink=true" label="FiberWrite Vals: X" color=green3 shape=box style=filled type="fiberwrite" tensor="X" mode="vals" size="1*B0_dim*C1_dim" sink="true"]
-    6 [comment="type=arrayvals,tensor=C" label="Array Vals: C" color=green2 shape=box style=filled type="arrayvals" tensor="C"]
-    9 [comment="type=fiberlookup,index=k,tensor=C,mode=0,format=compressed,src=true,root=false" label="FiberLookup k: C0\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="k" tensor="C" mode="0" format="compressed" src="true" root="false"]
-    18 [comment="type=crddrop,outer=i,inner=j" label="CrdDrop i,j" color=orange shape=box style=filled type="crddrop" outer="i" inner="j"]
-    19 [comment="type=crddrop,outer=j,inner=k" label="CrdDrop j,k" color=orange shape=box style=filled type="crddrop" outer="j" inner="k"]
-
-    17 -> 16 [label="crd" style=dashed type="crd" comment=""]
-    # 16 -> 2 [label="crd" style=dashed type="crd"]
-    16 -> 15 [label="crd" style=dashed type="crd"]
-    15 -> 14 [label="repsig" style=dotted type="repsig"]
-    14 -> 13 [label="ref" style=bold type="ref"]
-    13 -> 12 [label="crd" style=dashed type="crd" comment=""]
-    # 12 -> 1 [label="crd" style=dashed type="crd"]
-    12 -> 11 [label="crd" style=dashed type="crd"]
-    11 -> 10 [label="repsig" style=dotted type="repsig"]
-    10 -> 8 [label="ref" style=bold type="ref"]
-    8 -> 7 [label="crd_in-B" style=dashed type="crd" comment="in-B"]
-    7 -> 5 [label="ref_out-B" style=bold type="ref" comment="out-B"]
-    5 -> 4 [label="val" type="val"]
-    # 4 -> 3 [label="val" type="val"]
-    3 -> 0 [label="val" type="val"]
-    7 -> 6 [label="ref_out-C" style=bold type="ref" comment="out-C"]
-    6 -> 4 [label="val" type="val"]
-    8 -> 7 [label="ref_in-B" style=bold type="ref" comment="in-B"]
-    13 -> 9 [label="ref" style=bold type="ref" comment=""]
-    9 -> 7 [label="crd_in-C" style=dashed type="crd" comment="in-C"]
-    9 -> 7 [label="ref_in-C" style=bold type="ref" comment="in-C"]
-    17 -> 10 [label="ref" style=bold type="ref" comment=""]
-    
-    4 -> 19 [label="val_inner-k" type="val" comment="inner-k"]
-    12 -> 19 [label="crd_outer-j" style=dashed type="crd" comment="outer-j"]
-    18 -> 1 [label="crd_inner-j" style=dashed type="crd" comment="inner-j"]
-    19 -> 3 [label="val_inner-k" type="val" comment="inner-k"]
-
-    19 -> 18 [label="crd_inner-j" style=dashed type="crd" comment="inner-j"] 
-    16 -> 18 [label="crd_outer-i" style=dashed type="crd" comment="outer-i"] 
-    18 -> 2 [label="crd_outer-i" style=dashed type="crd" comment="outer-i"] 
-}
diff --git a/compiler/sam-outputs/opal-dot/trans_masked_broadcast.gv b/compiler/sam-outputs/opal-dot/trans_masked_broadcast.gv
deleted file mode 100644
index 06d63c95..00000000
--- a/compiler/sam-outputs/opal-dot/trans_masked_broadcast.gv
+++ /dev/null
@@ -1,27 +0,0 @@
-digraph SAM {
-    comment="X=ss01,B=ss01,c=s0"
-    0 [comment="type=fiberlookup,index=i,tensor=B,mode=0,format=compressed,src=true,root=true" label="FiberLookup i: B\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="B" mode="0" format="compressed" src="true" root="true"]
-    1 [comment="type=fiberlookup,index=i,tensor=c,mode=0,format=compressed,src=true,root=true" label="FiberLookup i: c\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="c" mode="0" format="compressed" src="true" root="true"]
-    2 [comment="type=intersect,index=i" label="intersect i" color=purple shape=box style=filled type="intersect" index="i"]
-    3 [comment="type=fiberlookup,index=j,tensor=B,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: B\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="B" mode="1" format="compressed" src="true" root="false"]
-    4 [comment="type=repsiggen,index=j" label="RepeatSignalGenerator j" color=cyan3 shape=box style=filled type="repsiggen" index="j"]
-    5 [comment="type=repeat,index=j,tensor=c,root=false" label="Repeat j: B" color=cyan2 shape=box style=filled type="repeat" index="j" tensor="c" root="false"]
-    6 [comment="type=arrayvals,tensor=c" label="Array Vals: c" color=green2 shape=box style=filled type="arrayvals" tensor="c"]
-    7 [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"]
-    8 [comment="type=fiberwrite,index=j,tensor=X,mode=1,format=compressed,segsize=B0_dim+1,crdsize=B0_dim*B1_dim,sink=true" label="FiberWrite j: X1\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="j" tensor="X" mode="1" format="compressed" segsize="B0_dim+1" crdsize="B0_dim*B1_dim" sink="true"]
-    9 [comment="type=fiberwrite,mode=vals,tensor=X,size=1*B0_dim*B1_dim,sink=true" label="FiberWrite Vals: X" color=green3 shape=box style=filled type="fiberwrite" tensor="X" mode="vals" size="1*B0_dim*B1_dim" sink="true"]
-    10 [comment="type=broadcast" shape=point style=invis type="broadcast"]
-    0 -> 2 [label="ref_in-B" style=bold type="ref" comment="in-B"]
-    0 -> 2 [label="crd_in-B" style=dashed type="crd" comment="in-B"]
-    1 -> 2 [label="ref_in-c" style=bold type="ref" comment="in-c"]
-    1 -> 2 [label="crd_in-c" style=dashed type="crd" comment="in-c"]
-    2 -> 3 [label="ref_out-B" style=bold type="ref" comment="out-B"]
-    2 -> 7 [label="crd" style=dashed type="crd"]
-    3 -> 10 [label="crd" style=dashed type="crd"]
-    10 -> 4 [label="crd" style=dashed type="crd"]
-    10 -> 8 [label="crd" style=dashed type="crd"]
-    4 -> 5 [label="repsig" style=dotted type="repsig"]
-    2 -> 5 [label="ref_out-c" style=bold type="ref" comment="out-c"]
-    5 -> 6 [label="ref" style=bold type="ref"]
-    6 -> 9 [label="val" type="val"]
-}

From acd05d0bfa932d24577a1cf4b11973cab4594d86 Mon Sep 17 00:00:00 2001
From: Bo Wun Cheng <bwcheng@stanford.edu>
Date: Thu, 2 Nov 2023 00:50:35 -0700
Subject: [PATCH 3/6] added masked_broadcast and trans_masked_broadcast

---
 .../sam-outputs/onyx-dot/masked_broadcast.gv  | 27 +++++++++++++++++++
 .../onyx-dot/trans_masked_broadcast.gv        | 27 +++++++++++++++++++
 2 files changed, 54 insertions(+)
 create mode 100644 compiler/sam-outputs/onyx-dot/masked_broadcast.gv
 create mode 100644 compiler/sam-outputs/onyx-dot/trans_masked_broadcast.gv

diff --git a/compiler/sam-outputs/onyx-dot/masked_broadcast.gv b/compiler/sam-outputs/onyx-dot/masked_broadcast.gv
new file mode 100644
index 00000000..86b0bdba
--- /dev/null
+++ b/compiler/sam-outputs/onyx-dot/masked_broadcast.gv
@@ -0,0 +1,27 @@
+digraph SAM {
+    comment="X=ss01,B=ss01,c=s0"
+    0 [comment="type=fiberlookup,index=i,tensor=B,mode=0,format=compressed,src=true,root=true" label="FiberLookup i: B\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="B" mode="0" format="compressed" src="true" root="true"]
+    1 [comment="type=repsiggen,index=i" label="RepeatSignalGenerator i" color=cyan3 shape=box style=filled type="repsiggen" index="i"]
+    2 [comment="type=repeat,index=i,tensor=c,root=true" label="Repeat i: B" color=cyan2 shape=box style=filled type="repeat" index="i" tensor="c" root="true"]
+    3 [comment="type=fiberlookup,index=i,tensor=c,mode=0,format=compressed,src=true,root=false" label="FiberLookup i: c\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="c" mode="0" format="compressed" src="true" root="false"]
+    4 [comment="type=fiberlookup,index=j,tensor=B,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: B\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="B" mode="1" format="compressed" src="true" root="false"]
+    5 [comment="type=intersect,index=j" label="intersect j" color=purple shape=box style=filled type="intersect" index="j"]
+    6 [comment="type=arrayvals,tensor=c" label="Array Vals: c" color=green2 shape=box style=filled type="arrayvals" tensor="c"]
+    7 [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"]
+    8 [comment="type=fiberwrite,index=j,tensor=X,mode=1,format=compressed,segsize=B0_dim+1,crdsize=B0_dim*B1_dim,sink=true" label="FiberWrite j: X1\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="j" tensor="X" mode="1" format="compressed" segsize="B0_dim+1" crdsize="B0_dim*B1_dim" sink="true"]
+    9 [comment="type=fiberwrite,mode=vals,tensor=X,size=1*B0_dim*B1_dim,sink=true" label="FiberWrite Vals: X" color=green3 shape=box style=filled type="fiberwrite" tensor="X" mode="vals" size="1*B0_dim*B1_dim" sink="true"]
+    10 [comment="type=broadcast" shape=point style=invis type="broadcast"]
+    0 -> 10 [label="crd" style=dashed type="crd"]
+    10 -> 1 [label="crd" style=dashed type="crd"]
+    1 -> 2 [label="repsig" style=dotted type="repsig"]
+    2 -> 3 [label="ref" style=bold type="ref"]
+    0 -> 4 [label="ref" style=bold type="ref"]
+    3 -> 5 [label="ref_in-c" style=bold type="ref" comment="in-c"]
+    3 -> 5 [label="crd_in-c" style=dashed type="crd" comment="in-c"]
+    4 -> 5 [label="ref_in-B", style=bold type="ref" comment="in-B"]
+    4 -> 5 [label="crd_in-B", style=dashed type="crd" comment="in-B"]
+    5 -> 6 [label="ref_out-c", style=bold type="ref" comment="out-c"]
+    10 -> 7 [label="crd", style=dashed type="crd"]
+    5 -> 8 [label="crd", style=dashed type="crd"]
+    6 -> 9 [label="val", type="val"]
+}
diff --git a/compiler/sam-outputs/onyx-dot/trans_masked_broadcast.gv b/compiler/sam-outputs/onyx-dot/trans_masked_broadcast.gv
new file mode 100644
index 00000000..06d63c95
--- /dev/null
+++ b/compiler/sam-outputs/onyx-dot/trans_masked_broadcast.gv
@@ -0,0 +1,27 @@
+digraph SAM {
+    comment="X=ss01,B=ss01,c=s0"
+    0 [comment="type=fiberlookup,index=i,tensor=B,mode=0,format=compressed,src=true,root=true" label="FiberLookup i: B\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="B" mode="0" format="compressed" src="true" root="true"]
+    1 [comment="type=fiberlookup,index=i,tensor=c,mode=0,format=compressed,src=true,root=true" label="FiberLookup i: c\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="i" tensor="c" mode="0" format="compressed" src="true" root="true"]
+    2 [comment="type=intersect,index=i" label="intersect i" color=purple shape=box style=filled type="intersect" index="i"]
+    3 [comment="type=fiberlookup,index=j,tensor=B,mode=1,format=compressed,src=true,root=false" label="FiberLookup j: B\ncompressed" color=green4 shape=box style=filled type="fiberlookup" index="j" tensor="B" mode="1" format="compressed" src="true" root="false"]
+    4 [comment="type=repsiggen,index=j" label="RepeatSignalGenerator j" color=cyan3 shape=box style=filled type="repsiggen" index="j"]
+    5 [comment="type=repeat,index=j,tensor=c,root=false" label="Repeat j: B" color=cyan2 shape=box style=filled type="repeat" index="j" tensor="c" root="false"]
+    6 [comment="type=arrayvals,tensor=c" label="Array Vals: c" color=green2 shape=box style=filled type="arrayvals" tensor="c"]
+    7 [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"]
+    8 [comment="type=fiberwrite,index=j,tensor=X,mode=1,format=compressed,segsize=B0_dim+1,crdsize=B0_dim*B1_dim,sink=true" label="FiberWrite j: X1\ncompressed" color=green3 shape=box style=filled type="fiberwrite" index="j" tensor="X" mode="1" format="compressed" segsize="B0_dim+1" crdsize="B0_dim*B1_dim" sink="true"]
+    9 [comment="type=fiberwrite,mode=vals,tensor=X,size=1*B0_dim*B1_dim,sink=true" label="FiberWrite Vals: X" color=green3 shape=box style=filled type="fiberwrite" tensor="X" mode="vals" size="1*B0_dim*B1_dim" sink="true"]
+    10 [comment="type=broadcast" shape=point style=invis type="broadcast"]
+    0 -> 2 [label="ref_in-B" style=bold type="ref" comment="in-B"]
+    0 -> 2 [label="crd_in-B" style=dashed type="crd" comment="in-B"]
+    1 -> 2 [label="ref_in-c" style=bold type="ref" comment="in-c"]
+    1 -> 2 [label="crd_in-c" style=dashed type="crd" comment="in-c"]
+    2 -> 3 [label="ref_out-B" style=bold type="ref" comment="out-B"]
+    2 -> 7 [label="crd" style=dashed type="crd"]
+    3 -> 10 [label="crd" style=dashed type="crd"]
+    10 -> 4 [label="crd" style=dashed type="crd"]
+    10 -> 8 [label="crd" style=dashed type="crd"]
+    4 -> 5 [label="repsig" style=dotted type="repsig"]
+    2 -> 5 [label="ref_out-c" style=bold type="ref" comment="out-c"]
+    5 -> 6 [label="ref" style=bold type="ref"]
+    6 -> 9 [label="val" type="val"]
+}

From ef31fa313bf1a45bc0b23741e5c6fbd38b89b7fc Mon Sep 17 00:00:00 2001
From: Bo Wun Cheng <bwcheng@stanford.edu>
Date: Wed, 15 Nov 2023 00:13:35 -0800
Subject: [PATCH 4/6] update configuration code for pe and reduce to account
 for the new config register heirarchy in reduce_pe_cluster. also modified the
 signal name for connection from crddrop to reduce and reduce to arrayval to
 match the new signal name of reduce in cluster, matmul_ijk_crddrop passed

---
 sam/onyx/hw_nodes/compute_node.py | 13 +++++++++++--
 sam/onyx/hw_nodes/merge_node.py   |  2 +-
 sam/onyx/hw_nodes/reduce_node.py  | 18 +++++++++++++++---
 3 files changed, 27 insertions(+), 6 deletions(-)

diff --git a/sam/onyx/hw_nodes/compute_node.py b/sam/onyx/hw_nodes/compute_node.py
index 2fb7620f..f6c651a2 100644
--- a/sam/onyx/hw_nodes/compute_node.py
+++ b/sam/onyx/hw_nodes/compute_node.py
@@ -151,6 +151,12 @@ def configure(self, attributes):
         comment = attributes['comment'].strip('"')
         print(c_op)
         op_code = 0
+        # configuring via sam, it is a sparse app
+        use_dense = False
+        # mapping to pe only, configuring only the pe, ignore the reduce
+        pe_only = True
+        # data I/O should interface with other primitive outside of the cluster
+        pe_in_external = True
         if c_op == 'mul':
             op_code = 1
         elif c_op == 'add' and 'sub=1' not in comment:
@@ -158,6 +164,9 @@ def configure(self, attributes):
         elif c_op == 'add' and 'sub=1' in comment:
             op_code = 2
         cfg_kwargs = {
-            'op': op_code
+            'op': op_code,
+            'use_dense': use_dense,
+            'pe_only': pe_only,
+            'pe_in_external': pe_in_external
         }
-        return op_code, cfg_kwargs
+        return (op_code, use_dense, pe_only, pe_in_external), cfg_kwargs
diff --git a/sam/onyx/hw_nodes/merge_node.py b/sam/onyx/hw_nodes/merge_node.py
index 9b1f1ae7..c0b3d158 100644
--- a/sam/onyx/hw_nodes/merge_node.py
+++ b/sam/onyx/hw_nodes/merge_node.py
@@ -68,7 +68,7 @@ def connect(self, other, edge, kwargs=None):
             other_red = other.get_name()
             new_conns = {
                 f'merge_to_reduce_inner': [
-                    ([(merge, f"cmrg_coord_out_{0}"), (other_red, f"data_in")], 17),
+                    ([(merge, f"cmrg_coord_out_{0}"), (other_red, f"reduce_data_in")], 17),
                 ]
             }
 
diff --git a/sam/onyx/hw_nodes/reduce_node.py b/sam/onyx/hw_nodes/reduce_node.py
index 5caac658..4b2e679f 100644
--- a/sam/onyx/hw_nodes/reduce_node.py
+++ b/sam/onyx/hw_nodes/reduce_node.py
@@ -39,7 +39,7 @@ def connect(self, other, edge, kwargs=None):
             new_conns = {
                 'reduce_to_wr_scan': [
                     # send output to rd scanner
-                    ([(red, "data_out"), (wr_scan, "data_in")], 17),
+                    ([(red, "reduce_data_out"), (wr_scan, "data_in")], 17),
                     # ([(red, "eos_out"), (wr_scan, "eos_in_0")], 1),
                     # ([(wr_scan, "ready_out_0"), (red, "ready_in")], 1),
                     # ([(red, "valid_out"), (wr_scan, "valid_in_0")], 1),
@@ -102,7 +102,19 @@ def connect(self, other, edge, kwargs=None):
     def configure(self, attributes):
         # TODO
         stop_lvl = 2
+        # configuring via sam, it is a sparse app
+        use_dense = False
+        # configuring both the pe and the reduce
+        pe_only = False
+        # data I/O to and from the PE should be internal with the reduce
+        pe_in_external = False
+        # op is set to integer add for the PE TODO: make this configurable in the sam graph 
+        op = 0
         cfg_kwargs = {
-            'stop_lvl': stop_lvl
+            'stop_lvl': stop_lvl,
+            'use_dense': use_dense,
+            'pe_only': pe_only,
+            'pe_in_external': pe_in_external,
+            'op': op
         }
-        return stop_lvl, cfg_kwargs
+        return (stop_lvl, use_dense, pe_only, pe_in_external, op), cfg_kwargs

From d0793dde6b547fd2d1a09a613960c4c0f5d0bc62 Mon Sep 17 00:00:00 2001
From: Bo Wun Cheng <bwcheng@stanford.edu>
Date: Wed, 15 Nov 2023 10:28:06 -0800
Subject: [PATCH 5/6] update port name for connection from reduce to reduce,
 also updated the reduce node configuration to use dense mode

---
 sam/onyx/hw_nodes/compute_node.py |  4 ++--
 sam/onyx/hw_nodes/reduce_node.py  | 16 ++++++++--------
 2 files changed, 10 insertions(+), 10 deletions(-)

diff --git a/sam/onyx/hw_nodes/compute_node.py b/sam/onyx/hw_nodes/compute_node.py
index f6c651a2..2482f399 100644
--- a/sam/onyx/hw_nodes/compute_node.py
+++ b/sam/onyx/hw_nodes/compute_node.py
@@ -81,7 +81,7 @@ def connect(self, other, edge, kwargs=None):
             pe = self.get_name()
             new_conns = {
                 f'pe_to_reduce': [
-                    ([(pe, "res"), (other_red, f"data_in")], 17),
+                    ([(pe, "res"), (other_red, f"reduce_data_in")], 17),
                 ]
             }
             return new_conns
@@ -156,7 +156,7 @@ def configure(self, attributes):
         # mapping to pe only, configuring only the pe, ignore the reduce
         pe_only = True
         # data I/O should interface with other primitive outside of the cluster
-        pe_in_external = True
+        pe_in_external = 1
         if c_op == 'mul':
             op_code = 1
         elif c_op == 'add' and 'sub=1' not in comment:
diff --git a/sam/onyx/hw_nodes/reduce_node.py b/sam/onyx/hw_nodes/reduce_node.py
index 4b2e679f..aa101d12 100644
--- a/sam/onyx/hw_nodes/reduce_node.py
+++ b/sam/onyx/hw_nodes/reduce_node.py
@@ -53,7 +53,7 @@ def connect(self, other, edge, kwargs=None):
             new_conns = {
                 'reduce_to_reduce': [
                     # send output to rd scanner
-                    ([(red, "data_out"), (other_red, "data_in")], 17),
+                    ([(red, "reduce_data_out"), (other_red, "reduce_data_in")], 17),
                     # ([(red, "eos_out"), (wr_scan, "eos_in_0")], 1),
                     # ([(wr_scan, "ready_out_0"), (red, "ready_in")], 1),
                     # ([(red, "valid_out"), (wr_scan, "valid_in_0")], 1),
@@ -72,7 +72,7 @@ def connect(self, other, edge, kwargs=None):
             new_conns = {
                 f'reduce_to_pe_{other_conn}': [
                     # send output to rd scanner
-                    ([(red, "data_out"), (pe, f"data{other_conn}")], 17),
+                    ([(red, "reduce_data_out"), (pe, f"data{other_conn}")], 17),
                     # ([(red, "eos_out"), (wr_scan, "eos_in_0")], 1),
                     # ([(wr_scan, "ready_out_0"), (red, "ready_in")], 1),
                     # ([(red, "valid_out"), (wr_scan, "valid_in_0")], 1),
@@ -102,19 +102,19 @@ def connect(self, other, edge, kwargs=None):
     def configure(self, attributes):
         # TODO
         stop_lvl = 2
-        # configuring via sam, it is a sparse app
-        use_dense = False
-        # configuring both the pe and the reduce
+        # bypassing the fifos in the pe, get result in a single cycle
         pe_only = False
+        # configuring both the pe and the reduce
+        pe_connected_to_reduce = True
         # data I/O to and from the PE should be internal with the reduce
-        pe_in_external = False
+        pe_in_external = 0
         # op is set to integer add for the PE TODO: make this configurable in the sam graph 
         op = 0
         cfg_kwargs = {
             'stop_lvl': stop_lvl,
-            'use_dense': use_dense,
+            'pe_connected_to_reduce': pe_connected_to_reduce,
             'pe_only': pe_only,
             'pe_in_external': pe_in_external,
             'op': op
         }
-        return (stop_lvl, use_dense, pe_only, pe_in_external, op), cfg_kwargs
+        return (stop_lvl, pe_connected_to_reduce, pe_only, pe_in_external, op), cfg_kwargs

From b39b2b429157e98bc58bcd9d1c62ac84af11d50a Mon Sep 17 00:00:00 2001
From: Bo Wun Cheng <bwcheng@stanford.edu>
Date: Wed, 15 Nov 2023 14:24:29 -0800
Subject: [PATCH 6/6] fix code style

---
 sam/onyx/hw_nodes/reduce_node.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/sam/onyx/hw_nodes/reduce_node.py b/sam/onyx/hw_nodes/reduce_node.py
index aa101d12..6ed41cf7 100644
--- a/sam/onyx/hw_nodes/reduce_node.py
+++ b/sam/onyx/hw_nodes/reduce_node.py
@@ -108,7 +108,7 @@ def configure(self, attributes):
         pe_connected_to_reduce = True
         # data I/O to and from the PE should be internal with the reduce
         pe_in_external = 0
-        # op is set to integer add for the PE TODO: make this configurable in the sam graph 
+        # op is set to integer add for the PE TODO: make this configurable in the sam graph
         op = 0
         cfg_kwargs = {
             'stop_lvl': stop_lvl,