Matmul2x2

benchmarks/float/matmul2x2.bril

@@ -0,0 +1,79 @@
+@main {
+  v0: float = const 8;
+  a11: float = id v0;
+  v1: float = const 3;
+  a12: float = id v1;
+  v2: float = const 2;
+  a21: float = id v2;
+  v3: float = const 4;
+  a22: float = id v3;
+  v4: float = const 2;
+  b11: float = id v4;
+  v5: float = const 7;
+  b12: float = id v5;
+  v6: float = const 4;
+  b21: float = id v6;
+  v7: float = const 5;
+  b22: float = id v7;
+  v8: float = id a11;
+  v9: float = id b11;
+  v10: float = fmul v8 v9;
+  c111: float = id v10;
+  v11: float = id a12;
+  v12: float = id b21;
+  v13: float = fmul v11 v12;
+  c112: float = id v13;
+  v14: float = id c111;
+  v15: float = id c112;
+  v16: float = fadd v14 v15;
+  c11: float = id v16;
+  v17: float = id a11;
+  v18: float = id b12;
+  v19: float = fmul v17 v18;
+  c121: float = id v19;
+  v20: float = id a12;
+  v21: float = id b22;
+  v22: float = fmul v20 v21;
+  c122: float = id v22;
+  v23: float = id c121;
+  v24: float = id c122;
+  v25: float = fadd v23 v24;
+  c12: float = id v25;
+  v26: float = id a21;
+  v27: float = id b11;
+  v28: float = fmul v26 v27;
+  c211: float = id v28;
+  v29: float = id a22;
+  v30: float = id b21;
+  v31: float = fmul v29 v30;
+  c212: float = id v31;
+  v32: float = id c211;
+  v33: float = id c212;
+  v34: float = fadd v32 v33;
+  c21: float = id v34;
+  v35: float = id a21;
+  v36: float = id b12;
+  v37: float = fmul v35 v36;
+  c221: float = id v37;
+  v38: float = id a22;
+  v39: float = id b22;
+  v40: float = fmul v38 v39;
+  c222: float = id v40;
+  v41: float = id c221;
+  v42: float = id c222;
+  v43: float = fadd v41 v42;
+  c22: float = id v43;
+  v44: float = id c11;
+  print v44;
+  v45: int = const 0;
+  v46: float = id c12;
+  print v46;
+  v47: int = const 0;
+  v48: float = id c21;
+  print v48;
+  v49: int = const 0;
+  v50: float = id c22;
+  print v50;
+  v51: int = const 0;
+  ret;
+}

benchmarks/float/matmul2x2.out

@@ -0,0 +1,4 @@
+28.00000000000000000
+71.00000000000000000
+20.00000000000000000
+34.00000000000000000

benchmarks/float/matmul2x2.prof

@@ -0,0 +1 @@
+total_dyn_inst: 77

benchmarks/float/matmulti2x2/gen.py

@@ -0,0 +1,93 @@
+# generate 10 2x2 float matrix multiply programs in bril
+
+from jinja2 import Template
+import numpy as np
+
+NUM_EXAMPLES = 12
+
+# Define the template for Bril code
+template_str = '''
+@main {
+  {% for i in range(NUM_EXAMPLES) %}
+  a11_{{i}}: float = const {{matrices[2*i][0][0]}};
+  a12_{{i}}: float = const {{matrices[2*i][0][1]}};
+  a21_{{i}}: float = const {{matrices[2*i][1][0]}};
+  a22_{{i}}: float = const {{matrices[2*i][1][1]}};
+  b11_{{i}}: float = const {{matrices[2*i+1][0][0]}};
+  b12_{{i}}: float = const {{matrices[2*i+1][0][1]}};
+  b21_{{i}}: float = const {{matrices[2*i+1][1][0]}};
+  b22_{{i}}: float = const {{matrices[2*i+1][1][1]}};
+
+  c11_{{i}}_part1: float = fmul a11_{{i}} b11_{{i}};
+  c11_{{i}}_part2: float = fmul a12_{{i}} b21_{{i}};
+  c11_{{i}}: float = fadd c11_{{i}}_part1 c11_{{i}}_part2;
+
+  c12_{{i}}_part1: float = fmul a11_{{i}} b12_{{i}};
+  c12_{{i}}_part2: float = fmul a12_{{i}} b22_{{i}};
+  c12_{{i}}: float = fadd c12_{{i}}_part1 c12_{{i}}_part2;
+
+  c21_{{i}}_part1: float = fmul a21_{{i}} b11_{{i}};
+  c21_{{i}}_part2: float = fmul a22_{{i}} b21_{{i}};
+  c21_{{i}}: float = fadd c21_{{i}}_part1 c21_{{i}}_part2;
+
+  c22_{{i}}_part1: float = fmul a21_{{i}} b12_{{i}};
+  c22_{{i}}_part2: float = fmul a22_{{i}} b22_{{i}};
+  c22_{{i}}: float = fadd c22_{{i}}_part1 c22_{{i}}_part2;
+
+  print c11_{{i}};
+  print c12_{{i}};
+  print c21_{{i}};
+  print c22_{{i}};
+  {% endfor %}
+  ret;
+}
+'''
+
+# Define the hardcoded matrices
+# [[8,3]  * [[2,7]  =  [[28,71]
+#  [2,4]]    [4,5]]      20,34]]
+matrices = [
+    [[8,3],[2,4]],
+    [[2,7],[4,5]],
+    [[0,1],[2,3]],
+    [[0,1],[2,3]],
+    [[1, 2], [3, 4]],
+    [[5, 6], [7, 8]],
+    [[9, 10], [11, 12]],
+    [[13, 14], [15, 16]],
+    [[17, 18], [19, 20]],
+    [[21, 22], [23, 24]],
+    [[25, 26], [27, 28]],
+    [[29, 30], [31, 32]],
+    [[33, 34], [35, 36]],
+    [[37, 38], [39, 40]],
+    [[41, 42], [43, 44]],
+    [[45, 46], [47, 48]],
+    [[49, 50], [51, 52]],
+    [[53, 54], [55, 56]],
+    [[57, 58], [59, 60]],
+    [[61, 62], [63, 64]],
+    [[65, 66], [67, 68]],
+    [[69, 70], [71, 72]],
+    [[73, 74], [75, 76]],
+    [[77, 78], [79, 80]]
+]
+
+# Generate reference results using NumPy and write to matmulti2x2.ref
+# TODO write to filename first argument to program
+with open("matmulti2x2.ref", "w") as ref_file:
+    for i in range(NUM_EXAMPLES):
+        mat1 = np.array(matrices[2*i])
+        mat2 = np.array(matrices[2*i + 1])
+        result = np.matmul(mat1, mat2)
+        for row in result:
+            for element in row:
+                ref_file.write(f"{element:.17f}\n")
+
+# Create a Jinja2 template and render it
+template = Template(template_str)
+rendered_str = template.render(matrices=matrices)
+
+# Write the rendered Bril code to matmulti2x2.bril
+with open("matmulti2x2.bril", "w") as bril_file:
+    bril_file.write(rendered_str)