Added halide mlp

npbench/benchmarks/deep_learning/mlp/mlp_halide.py

@@ -0,0 +1,107 @@
+import numpy as np
+import halide as hl
+
+
+def mlp_params():
+    input = hl.ImageParam(hl.Float(32), 2, "input")
+    w1 = hl.ImageParam(hl.Float(32), 2, "w1")
+    b1 = hl.ImageParam(hl.Float(32), 1, "b1")
+    w2 = hl.ImageParam(hl.Float(32), 2, "w2")
+    b2 = hl.ImageParam(hl.Float(32), 1, "b2")
+    w3 = hl.ImageParam(hl.Float(32), 2, "w3")
+    b3 = hl.ImageParam(hl.Float(32), 1, "b3")
+    return input, w1, b1, w2, b2, w3, b3
+
+
+# 3-layer MLP
+def mlp(input, w1, b1, w2, b2, w3, b3):
+    N = 8
+    C = 3
+    H1 = 30000
+    H2 = 10000
+    H3 = 1000
+
+    n = hl.Var("n")
+
+    h1 = hl.Var("h1")
+    r1 = hl.RDom([(0, C)])
+    layer1 = hl.Func("layer1")
+    layer1[h1, n] = b1[h1]
+    layer1[h1, n] += input[r1.x, n] * w1[r1.x, h1]
+
+    relu1 = hl.Func("relu1")
+    relu1[h1, n] = hl.max(0.0, layer1[h1, n])
+
+    # Layer 2
+
+    h2 = hl.Var("h2")
+    r2 = hl.RDom([(0, H1)])
+    layer2 = hl.Func()
+    layer2[h2, n] = b2[h2]
+    layer2[h2, n] += layer1[r2.x, n] * w2[r2.x, h2]
+
+    relu2 = hl.Func("relu2")
+    relu2[h2, n] = hl.max(0.0, layer2[h2, n])
+
+    # Layer 3
+
+    h3 = hl.Var("h3")
+    r3 = hl.RDom([(0, H2)])
+    layer3 = hl.Func("layer3")
+    layer3[h3, n] = b3[h3]
+    layer3[h3, n] += layer2[r3.x, n] * w3[r3.x, h3]
+
+    # Softmax
+
+    a = hl.RDom([(0, H3)])
+    maxi = hl.Func("maxi")
+    maxi[n] = hl.maximum(layer3[a.x, n])
+
+    expo = hl.Func("expo")
+    expo[h3, n] = hl.exp(layer3[h3, n] - maxi[n])
+
+    b = hl.RDom([(0, H3)])
+    norm = hl.Func("norm")
+    norm[n] = 0.0
+    norm[n] += expo[b.x, n]
+
+    output = hl.Func("output")
+    output[h3, n] = expo[h3, n] / norm[n]
+
+
+    input.dim(0).set_bounds(0, C).set_stride(1)
+    input.dim(1).set_bounds(0, N).set_stride(C)
+
+    w1.dim(0).set_bounds(0, C).set_stride(1)
+    w1.dim(1).set_bounds(0, H1).set_stride(C)
+    b1.dim(0).set_bounds(0, H1).set_stride(1)
+
+    w2.dim(0).set_bounds(0, H1).set_stride(1)
+    w2.dim(1).set_bounds(0, H2).set_stride(H1)
+    b2.dim(0).set_bounds(0, H2).set_stride(1)
+
+    w3.dim(0).set_bounds(0, H2).set_stride(1)
+    w3.dim(1).set_bounds(0, H3).set_stride(H2)
+    b3.dim(0).set_bounds(0, H3).set_stride(1)
+
+    # Estimates
+
+    input.dim(0).set_estimate(0, C)
+    input.dim(1).set_estimate(0, N)
+
+    w1.dim(0).set_estimate(0, C)
+    w1.dim(1).set_estimate(0, H1)
+    b1.dim(0).set_estimate(0, H1)
+
+    w2.dim(0).set_estimate(0, H1)
+    w2.dim(1).set_estimate(0, H2)
+    b2.dim(0).set_estimate(0, H2)
+
+    w3.dim(0).set_estimate(0, H2)
+    w3.dim(1).set_estimate(0, H3)
+    b3.dim(0).set_estimate(0, H3)
+
+    output.set_estimate(h3, 0, H3)
+    output.set_estimate(n, 0, N)
+
+    return output