Merge pull request #136 from JuliaGPU/vc/features

Add support for features

Author: jpsamaroo

Project.toml

@@ -1,7 +1,7 @@
 name = "AMDGPU"
 uuid = "21141c5a-9bdb-4563-92ae-f87d6854732e"
 authors = ["Julian P Samaroo <jpsamaroo@jpsamaroo.me>"]
-version = "0.2.6"
+version = "0.2.7"
 
 [deps]
 AbstractFFTs = "621f4979-c628-5d54-868e-fcf4e3e8185c"
@@ -27,23 +27,11 @@ Adapt = "3.0"
 BinaryProvider = "0.5"
 CEnum = "0.2, 0.3, 0.4"
 GPUArrays = "6"
-GPUCompiler = "0.10, 0.11"
+GPUCompiler = "0.11.5"
 LLVM = "3"
 MacroTools = "0.5"
 Requires = "1"
 Setfield = "0.5, 0.6, 0.7"
 hsa_rocr_jll = "3.7"
 hsakmt_roct_jll = "3.7"
 julia = "1.6"
-
-[extras]
-FFTW = "7a1cc6ca-52ef-59f5-83cd-3a7055c09341"
-FillArrays = "1a297f60-69ca-5386-bcde-b61e274b549b"
-ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
-InteractiveUtils = "b77e0a4c-d291-57a0-90e8-8db25a27a240"
-Pkg = "44cfe95a-1eb2-52ea-b672-e2afdf69b78f"
-SpecialFunctions = "276daf66-3868-5448-9aa4-cd146d93841b"
-Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
-
-[targets]
-test = ["FFTW", "FillArrays", "ForwardDiff", "InteractiveUtils", "Pkg", "SpecialFunctions", "Test"]

src/agent.jl

@@ -189,15 +189,41 @@ function isas(agent::HSAAgent)
     HSA.agent_iterate_isas(agent.agent, func, isas) |> check
     isas[]
 end
-function get_first_isa_string(agent::HSAAgent)
-    isa = first(isas(agent))
+
+# TODO: PCRE regexes are not thread-safe
+const isa_regex = r"([a-z]*)-([a-z]*)-([a-z]*)--([a-z0-9]*)([a-zA-Z0-9+\-:]*)"
+function parse_isa(isa::HSA.ISA)
     len = Ref{Cuint}(0)
     getinfo(isa, HSA.ISA_INFO_NAME_LENGTH, len) |> check
-    name = repeat(" ", len[])
+    name = Vector{UInt8}(undef, len[])
     getinfo(isa, HSA.ISA_INFO_NAME, name) |> check
-    isa_name = string(rstrip(last(split(name, "-")), '\0'))
-    return isa_name
+    name = String(name)
+    m = match(isa_regex, name)
+    @assert m !== nothing "Failed to match ISA name pattern: $name"
+    m
 end
+
+function llvm_arch_features(isa::HSA.ISA)
+    m = parse_isa(isa)
+    arch = m.captures[4]
+    features = join(map(x->x[1:end-1],
+                        filter(x->!isempty(x) && (x[end]=='+'),
+                               split(m.captures[5], ':'))),
+                    ",+")
+    if !isempty(features)
+        features = '+'*features
+    end
+    if Base.libllvm_version < v"12"
+        features = replace(features, "sramecc"=>"sram-ecc")
+    end
+    return (arch, features)
+end
+architecture(isa::HSA.ISA) = llvm_arch_features(isa)[1]
+features(isa::HSA.ISA) = llvm_arch_features(isa)[2]
+
+get_first_isa_string(agent::HSAAgent) = architecture(first(isas(agent)))
+get_first_feature_string(agent::HSAAgent) = features(first(isas(agent)))
+
 function max_group_size(isa::HSA.ISA)
     size = Ref{UInt32}(0)
     getinfo(isa, HSA.ISA_INFO_WORKGROUP_MAX_SIZE, size)

src/execution.jl

@@ -298,7 +298,11 @@ function definitions change, or when different types or keyword arguments are pr
 function rocfunction(f::Core.Function, tt::Type=Tuple{}; name=nothing, device=default_device(), global_hooks=NamedTuple(), kwargs...)
     source = FunctionSpec(f, tt, true, name)
     cache = get!(()->Dict{UInt,Any}(), rocfunction_cache, device)
-    target = GCNCompilerTarget(; dev_isa=default_isa(device))
+    agent = device.device
+    isa = default_isa(device)
+    arch = architecture(isa)
+    feat = features(isa)
+    target = GCNCompilerTarget(; dev_isa=arch, features=feat)
     params = ROCCompilerParams(device, global_hooks)
     job = CompilerJob(target, source, params)
     GPUCompiler.cached_compilation(cache, job, rocfunction_compile, rocfunction_link)::HostKernel{f,tt}

src/reflection.jl

@@ -14,7 +14,10 @@ for method in (:code_typed, :code_warntype, :code_llvm, :code_native)
         function $method(io::IO, @nospecialize(func), @nospecialize(types);
                          kernel::Bool=false, device=default_device(), kwargs...)
             source = FunctionSpec(func, Base.to_tuple_type(types), kernel)
-            target = GCNCompilerTarget(; dev_isa=default_isa(device))
+            isa = default_isa(device)
+            arch = architecture(isa)
+            feat = features(isa)
+            target = GCNCompilerTarget(; dev_isa=arch,features=feat)
             params = ROCCompilerParams(device, NamedTuple())
             job = CompilerJob(target, source, params)
             GPUCompiler.$method($(args...); kwargs...)

src/runtime.jl

@@ -14,7 +14,7 @@ default_queue(::typeof(HSA_rt), device::RuntimeDevice) =
 get_device(queue::RuntimeQueue{HSAQueue}) = RuntimeDevice(queue.queue.agent)
 
 default_isa(device::RuntimeDevice{HSAAgent}) =
-    get_first_isa_string(device.device)
+    first(isas(device.device))
 
 struct RuntimeEvent{E}
     event::E

test/Project.toml

@@ -0,0 +1,15 @@
+[deps]
+Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
+FFTW = "7a1cc6ca-52ef-59f5-83cd-3a7055c09341"
+FillArrays = "1a297f60-69ca-5386-bcde-b61e274b549b"
+ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
+GPUArrays = "0c68f7d7-f131-5f86-a1c3-88cf8149b2d7"
+GPUCompiler = "61eb1bfa-7361-4325-ad38-22787b887f55"
+InteractiveUtils = "b77e0a4c-d291-57a0-90e8-8db25a27a240"
+LLVM = "929cbde3-209d-540e-8aea-75f648917ca0"
+LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+Pkg = "44cfe95a-1eb2-52ea-b672-e2afdf69b78f"
+Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
+SpecialFunctions = "276daf66-3868-5448-9aa4-cd146d93841b"
+Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
+Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"