Control flow support (#107)

* control flow support

* revert the infinite loops

* bugfix

* Update CI scripts  (#111)

* Add refs to integration tests (#106)

* Add refs to integration tests

* Update .github/workflows/IntegrationTest.yml

Co-authored-by: Hong Ge <[email protected]>

* Update IntegrationTest.yml (#109)

Co-authored-by: Rik Huijzer <[email protected]>

* revise types

* Update IntegrationTest.yml

* Update BenchmarksAndMicroIntegration.yml

* only try to eval Symbol and GlobalRef

* refactor

* fix Turing branch

* Update IntegrationTest.yml

* minor update

* remove used var

* code refactor

* remove `step_in`, move copying functions

* fix perf code

* move cache mechanism into TapedFunction's constructor

* Update Project.toml

Co-authored-by: Hong Ge <[email protected]>
Co-authored-by: Rik Huijzer <[email protected]>

Author: 3 people

.github/workflows/BenchmarksAndMicroIntegration.yml

@@ -25,7 +25,7 @@ jobs:
           using Pkg
           try
             # force it to use this PR's version of the package
-            pkg"add Turing#master" # TODO: remove this when Turing is updated
+            pkg"add Turing#libtask-integration"
             Pkg.develop(PackageSpec(path="."))  # resolver may fail with main deps
             Pkg.update()
           catch err

.github/workflows/IntegrationTest.yml

@@ -15,7 +15,7 @@ jobs:
       matrix:
         package:
           - {user: TuringLang, repo: AdvancedPS.jl, ref: releases}
-          - {user: TuringLang, repo: Turing.jl, ref: hg/new-libtask2}
+          - {user: TuringLang, repo: Turing.jl, ref: libtask-integration}
 
     steps:
       - uses: actions/checkout@v2

Project.toml

@@ -3,7 +3,7 @@ uuid = "6f1fad26-d15e-5dc8-ae53-837a1d7b8c9f"
 license = "MIT"
 desc = "Tape based task copying in Turing"
 repo = "https://github.com/TuringLang/Libtask.jl.git"
-version = "0.6.5"
+version = "0.6.6"
 
 [deps]
 IRTools = "7869d1d1-7146-5819-86e3-90919afe41df"

perf/p0.jl

@@ -30,8 +30,8 @@ args = m.evaluator[2:end];
 t = @btime  Libtask.CTask(f, args...)
 # schedule(t.task) # work fine!
 # @show Libtask.result(t.tf)
-@show "Step in a tape..."
-@btime Libtask.step_in(t.tf, args)
+@show "Run a tape..."
+@btime t.tf(args...)
 
 # Case 2: SMC sampler
 
@@ -43,5 +43,5 @@ m = Turing.Core.TracedModel(gdemo(1.5, 2.), Sampler(SMC(50)), VarInfo());
 t = @btime Libtask.CTask(m.evaluator[1], m.evaluator[2:end]...);
 # schedule(t.task)
 # @show Libtask.result(t.tf.tape)
-@show "Step in a tape..."
-@btime Libtask.step_in(t.tf, m.evaluator[2:end])
+@show "Run a tape..."
+@btime t.tf(m.evaluator[2:end]...)

perf/p2.jl

@@ -58,6 +58,6 @@ args = m.evaluator[2:end]
 
 t = Libtask.CTask(f, args...)
 
-Libtask.step_in(t.tf, args)
+t.tf(args...)
 
 @show Libtask.result(t.tf)