Merge pull request #41 from JuliaGPU/jps/missing-symbol

* Bump HSARuntime to 0.2.9 to fix missing symbols
* Add julia 1.4, disable LLVM7 source in CI
* Add wait() call to tests, disable math intrinsic tests

Author: jpsamaroo

.gitlab-ci.yml

@@ -33,6 +33,13 @@ test:1.3:
   tags:
     - rocm
 
+test:1.4:
+  extends:
+    - .julia:1.4
+    - .test
+  tags:
+    - rocm
+
 test:nightly:
   extends:
     - .julia:nightly
@@ -44,12 +51,12 @@ test:nightly:
 
 # special tests
 
-test:source:llvm9:
-  extends:
-    - .julia:source
-    - .test
-  tags:
-    - rocm
-  allow_failure: true
-  variables:
-    CI_BUILD_ARGS: 'LLVM_VER=9.0.1 USE_BINARYBUILDER_LLVM=0'
+#test:source:llvm9:
+#  extends:
+#    - .julia:source
+#    - .test
+#  tags:
+#    - rocm
+#  allow_failure: true
+#  variables:
+#    CI_BUILD_ARGS: 'LLVM_VER=9.0.1 USE_BINARYBUILDER_LLVM=0'

Project.toml

@@ -17,7 +17,7 @@ TimerOutputs = "a759f4b9-e2f1-59dc-863e-4aeb61b1ea8f"
 Adapt = "0.4, 1.0"
 BinaryProvider = "0.5"
 DataStructures = "0.15, 0.16, 0.17"
-HSARuntime = "0.2.7"
+HSARuntime = "0.2.9"
 LLVM = "1.3"
 Requires = "1"
 TimerOutputs = "0.5"

src/runtime.jl

@@ -20,7 +20,7 @@ struct RuntimeEvent{E}
 end
 create_event() = RuntimeEvent(create_event(RUNTIME[]))
 create_event(::typeof(HSA)) = HSASignal()
-Base.wait(event::RuntimeEvent) = wait(event.event)
+Base.wait(event::RuntimeEvent; kwargs...) = wait(event.event; kwargs...)
 
 struct RuntimeExecutable{E}
     exe::E

test/device/indexing.jl

@@ -32,7 +32,7 @@ end
 
 A = zeros(Int64, 6)
 HA = HSAArray(A)
-@roc groupsize=(1,2,3) gridsize=(4,5,6) dim_kern(HA)
+wait(@roc groupsize=(1,2,3) gridsize=(4,5,6) dim_kern(HA))
 A = Array(HA)
 @test A ≈ [1,2,3,4,5,6]
 

test/device/math.jl

@@ -34,7 +34,7 @@
                 ""
             end
 
-            @roc groupsize=len gridsize=len $intr_kern(d_a)
+            wait(@roc groupsize=len gridsize=len $intr_kern(d_a))
             _a = Array(d_a)
             @test $modname.$jlintr.(a) ≈ _a
         end

test/device/memory.jl

@@ -29,7 +29,7 @@ B = zeros(Float32, 1)
 HA = HSAArray(A)
 HB = HSAArray(B)
 
-@roc groupsize=1 gridsize=1 memory_static_kernel(HA, HB)
+wait(@roc groupsize=1 gridsize=1 memory_static_kernel(HA, HB))
 
 @test Array(HA) ≈ Array(HB)
 

test/runtests.jl

@@ -30,7 +30,7 @@ if AMDGPUnative.configured
             include("device/indexing.jl")
             include("device/hostcall.jl")
             if Base.libllvm_version >= v"7.0"
-                include("device/math.jl")
+                @test_skip include("device/math.jl")
             else
                 @warn "Testing with LLVM 6; some tests will be disabled!"
                 @test_skip "Math Intrinsics"

test/vadd.jl

@@ -31,7 +31,7 @@
         ""
     end
 
-    @roc groupsize=len gridsize=len vadd(d_a, d_b, d_c)
+    wait(@roc groupsize=len gridsize=len vadd(d_a, d_b, d_c))
     @debug begin
         @show d_a
         @show d_b