Add script to tune parameters

.gitignore

@@ -1 +1,2 @@
-test/Manifest.toml
+Manifest.toml
+tuning/.CondaPkg/

configs/configs.jl

@@ -3,6 +3,7 @@
 using GemmKernels
 using LinearAlgebra
 using ForwardDiff
+using Octavian
 
 struct Configuration
     name           # Human-readable name of the configuration.
@@ -64,7 +65,8 @@ function generate_inputs(cf::Configuration)
     new_b_h = cf.transpose_b ? transpose(b_h) : b_h
 
     (cf.calc_reference)(c_h, new_a_h, new_b_h, cf.alpha, cf.beta)
-    c_h, a, b, c, d
+    c_ref = CuArray(c_h)
+    c_ref, a, b, c, d
 end
 
 # Run the GEMM.
@@ -88,21 +90,21 @@ function run_baseline(cf::Configuration, a, b, c, d)
 end
 
 # Verify results.
-function verify(cf::Configuration, c_h, d)
-    cf.verify(c_h, d)
+function verify(cf::Configuration, c_ref, d)
+    cf.verify(c_ref, d)
 end
 
-function verify_default(c_h, d)
-    isapprox(c_h, Array(d))
+function verify_default(c_ref, d)
+    isapprox(c_ref, d)
 end
 
-function verify_bias(c_h, d, bias)
-    c_h .+ Array(bias) ≈ Array(d)
+function verify_bias(c_ref, d, bias)
+    c_ref .+ bias ≈ d
 end
 
-function verify_dual(c_h, d)
-    c_dual = reinterpret(ForwardDiff.Dual{Float32,Float32,1}, c_h)
-    d_dual = reinterpret(ForwardDiff.Dual{Float32,Float32,1}, Array(d))
+function verify_dual(c_ref, d)
+    c_dual = reinterpret(ForwardDiff.Dual{Float32,Float32,1}, c_ref)
+    d_dual = reinterpret(ForwardDiff.Dual{Float32,Float32,1}, d)
     isapprox(c_dual, d_dual)
 end
 
@@ -238,7 +240,7 @@ macro get_wmma_config()
                       CD_type,
                       transpose_a,
                       transpose_b,
-                      mul!,
+                      Octavian.matmul!,
                       Epilogue.Default(),
                       verify_default,
                       kernel,

test/Project.toml

@@ -5,5 +5,6 @@ Distributed = "8ba89e20-285c-5b6f-9357-94700520ee1b"
 ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
 InteractiveUtils = "b77e0a4c-d291-57a0-90e8-8db25a27a240"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+Octavian = "6fd5a793-0b7e-452c-907f-f8bfe9c57db4"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 XUnit = "3e3c03f2-1a94-11e9-2981-050a4ca824ab"

tuning/Project.toml

@@ -0,0 +1,14 @@
+[deps]
+CSV = "336ed68f-0bac-5ca0-87d4-7b16caf5d00b"
+Cthulhu = "f68482b8-f384-11e8-15f7-abe071a5a75f"
+DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
+DataStructures = "864edb3b-99cc-5e75-8d2d-829cb0a9cfe8"
+GPUCompiler = "61eb1bfa-7361-4325-ad38-22787b887f55"
+Hyperopt = "93e5fe13-2215-51db-baaf-2e9a34fb2712"
+JSON = "682c06a0-de6a-54ab-a142-c8b1cf79cde6"
+LoggingExtras = "e6f89c97-d47a-5376-807f-9c37f3926c36"
+Octavian = "6fd5a793-0b7e-452c-907f-f8bfe9c57db4"
+Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
+ProgressMeter = "92933f4c-e287-5a05-a399-4b506db050ca"
+PythonPlot = "274fc56d-3b97-40fa-a1cd-1b4a50311bf9"
+StatsBase = "2913bbd2-ae8a-5f71-8c99-4fb6c76f3a91"