Add initial VE support

src/GPUCompiler.jl

@@ -24,6 +24,7 @@ include("native.jl")
 include("ptx.jl")
 include("gcn.jl")
 include("spirv.jl")
+include("ve.jl")
 
 include("runtime.jl")
 

src/ve.jl

@@ -0,0 +1,31 @@
+# implementation of the GPUCompiler interfaces for generating Aurora-SX VE code
+
+## target
+
+export VECompilerTarget
+
+Base.@kwdef struct VECompilerTarget <: AbstractCompilerTarget
+end
+
+llvm_triple(::VECompilerTarget) = "ve-unkown-unkown"
+
+function llvm_machine(target::VECompilerTarget)
+ triple = llvm_triple(target)
+ t = Target(triple=triple)
+
+ cpu = "ve"
+ feat = ""
+ optlevel = LLVM.API.LLVMCodeGenLevelDefault
+ reloc = LLVM.API.LLVMRelocPIC
+ tm = TargetMachine(t, triple, cpu, feat, optlevel, reloc)
+ asm_verbosity!(tm, true)
+
+ return tm
+end
+
+# TODO: encode debug build or not in the compiler job
+# https://github.com/JuliaGPU/CUDAnative.jl/issues/368
+runtime_slug(job::CompilerJob{VECompilerTarget}) = "ve"
+
+const ve_intrinsics = () # TODO: ("vprintf", "__assertfail", "malloc", "free")
+isintrinsic(::CompilerJob{VECompilerTarget}, fn::String) = in(fn, gcn_intrinsics)

test/definitions/ve.jl

@@ -0,0 +1,50 @@
+using GPUCompiler
+
+if !@isdefined(TestRuntime)
+ include("../util.jl")
+end
+
+
+# create a VE-based test compiler, and generate reflection methods for it
+
+function ve_job(@nospecialize(func), @nospecialize(types); kernel::Bool=false, kwargs...)
+ source = FunctionSpec(func, Base.to_tuple_type(types), kernel)
+ target = VECompilerTarget()
+ params = TestCompilerParams()
+ CompilerJob(target, source, params), kwargs
+end
+
+function ve_code_typed(@nospecialize(func), @nospecialize(types); kwargs...)
+ job, kwargs = ve_job(func, types; kwargs...)
+ GPUCompiler.code_typed(job; kwargs...)
+end
+
+function ve_code_warntype(io::IO, @nospecialize(func), @nospecialize(types); kwargs...)
+ job, kwargs = ve_job(func, types; kwargs...)
+ GPUCompiler.code_warntype(io, job; kwargs...)
+end
+
+function ve_code_llvm(io::IO, @nospecialize(func), @nospecialize(types); kwargs...)
+ job, kwargs = ve_job(func, types; kwargs...)
+ GPUCompiler.code_llvm(io, job; kwargs...)
+end
+
+function ve_code_native(io::IO, @nospecialize(func), @nospecialize(types); kwargs...)
+ job, kwargs = ve_job(func, types; kwargs...)
+ GPUCompiler.code_native(io, job; kwargs...)
+end
+
+# aliases without ::IO argument
+for method in (:code_warntype, :code_llvm, :code_native)
+ ve_method = Symbol("ve_$(method)")
+ @eval begin
+ $ve_method(@nospecialize(func), @nospecialize(types); kwargs...) =
+ $ve_method(stdout, func, types; kwargs...)
+ end
+end
+
+# simulates codegen for a kernel function: validates by default
+function ve_code_execution(@nospecialize(func), @nospecialize(types); kwargs...)
+ job, kwargs = ve_job(func, types; kernel=true, kwargs...)
+ GPUCompiler.compile(:asm, job; kwargs...)
+end