Adapt to nonrecursive codegen.

Ref JuliaLang/julia#25984
JuliaGPU · Mar 26, 2020 · 9d22860 · 9d22860
1 parent 234009a
commit 9d22860
Show file tree

Hide file tree

Showing 5 changed files with 67 additions and 197 deletions.
diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
@@ -53,7 +53,7 @@ julia:1.3:
 
 cuda:10.1:
   extends:
-    - .julia:1.3
+    - .julia:nightly
     - .test
   variables:
     JULIA_CUDA_VERSION: '10.1'
@@ -62,7 +62,7 @@ cuda:10.1:
 
 cuda:10.0:
   extends:
-    - .julia:1.3
+    - .julia:nightly
     - .test
   variables:
     JULIA_CUDA_VERSION: '10.0'
@@ -71,7 +71,7 @@ cuda:10.0:
 
 cuda:9.2:
   extends:
-    - .julia:1.3
+    - .julia:nightly
     - .test
   variables:
     JULIA_CUDA_VERSION: '9.2'
@@ -80,7 +80,7 @@ cuda:9.2:
 
 cuda:9.0:
   extends:
-    - .julia:1.3
+    - .julia:nightly
     - .test
   variables:
     JULIA_CUDA_VERSION: '9.0'
@@ -89,7 +89,7 @@ cuda:9.0:
 
 cuda:local:
   extends:
-    - .julia:1.3
+    - .julia:nightly
     - .test
   image: nvidia/cuda:10.1-devel-ubuntu18.04
   variables:
@@ -99,7 +99,7 @@ cuda:local:
 
 cuda:none:
   extends:
-    - .julia:1.3
+    - .julia:nightly
     - .test
   variables:
     JULIA_CUDA_USE_BINARYBUILDER: 'false'
@@ -121,7 +121,7 @@ platform:arm64:
   variables:
     JULIA_CUDA_USE_BINARYBUILDER: 'false'
   extends:
-    - .julia:1.3
+    - .julia:nightly
     - .test
   tags:
     - nvidia-arm64
@@ -131,7 +131,7 @@ platform:arm64:
 
 debug:
   extends:
-    - .julia:1.3
+    - .julia:nightly
     - .test
   tags:
     - nvidia
@@ -150,7 +150,7 @@ debug:
 
 cuarrays:
   extends:
-    - .julia:1.3
+    - .julia:nightly
     - .test
   tags:
     - nvidia
@@ -174,12 +174,12 @@ cuarrays:
 
 coverage:
   extends:
-    - .julia:1.3
+    - .julia:nightly
     - .coverage
 
 documentation:
   extends:
-    - .julia:1.3
+    - .julia:nightly
     - .documentation
   tags:
     - nvidia

diff --git a/Project.toml b/Project.toml
@@ -26,7 +26,7 @@ CUDAapi = "3.1, 4.0"
 CUDAdrv = "6.2.1"
 Cthulhu = "1.0"
 DataStructures = "0.15, 0.16, 0.17"
-LLVM = "1.2"
+LLVM = "1.3.4"
 MacroTools = "0.5"
 TimerOutputs = "0.5"
 julia = "1.3"

diff --git a/src/compiler/irgen.jl b/src/compiler/irgen.jl
@@ -48,57 +48,11 @@ Base.showerror(io::IO, err::MethodSubstitutionWarning) =
 const method_substitution_whitelist = [:hypot]
 
 function compile_method_instance(job::CompilerJob, method_instance::Core.MethodInstance, world)
-    function postprocess(ir)
-        # get rid of jfptr wrappers
-        for llvmf in functions(ir)
-            startswith(LLVM.name(llvmf), "jfptr_") && unsafe_delete!(ir, llvmf)
-        end
-
-        return
-    end
-
     # set-up the compiler interface
-    last_method_instance = nothing
-    call_stack = Vector{Core.MethodInstance}()
-    dependencies = MultiDict{Core.MethodInstance,LLVM.Function}()
-    function hook_module_setup(ref::Ptr{Cvoid})
-        ref = convert(LLVM.API.LLVMModuleRef, ref)
-        ir = LLVM.Module(ref)
-        module_setup(ir)
-    end
-    function hook_module_activation(ref::Ptr{Cvoid})
-        ref = convert(LLVM.API.LLVMModuleRef, ref)
-        ir = LLVM.Module(ref)
-        postprocess(ir)
-
-        # find the function that this module defines
-        llvmfs = filter(llvmf -> !isdeclaration(llvmf) &&
-                                 linkage(llvmf) == LLVM.API.LLVMExternalLinkage,
-                        collect(functions(ir)))
-
-        llvmf = nothing
-        if length(llvmfs) == 1
-            llvmf = first(llvmfs)
-        elseif length(llvmfs) > 1
-            llvmfs = filter!(llvmf -> startswith(LLVM.name(llvmf), "julia_"), llvmfs)
-            if length(llvmfs) == 1
-                llvmf = first(llvmfs)
-            end
-        end
-
-        @compiler_assert llvmf !== nothing job
-
-        insert!(dependencies, last_method_instance, llvmf)
-    end
-    function hook_emit_function(method_instance, code, world)
+    call_stack = [method_instance]
+    function hook_emit_function(method_instance, code)
         push!(call_stack, method_instance)
 
-        # check for recursion
-        if method_instance in call_stack[1:end-1]
-            throw(KernelError(job, "recursion is currently not supported";
-                              bt=backtrace(job, call_stack)))
-        end
-
         # check for Base functions that exist in CUDAnative too
         # FIXME: this might be too coarse
         method = method_instance.def
@@ -115,17 +69,14 @@ function compile_method_instance(job::CompilerJob, method_instance::Core.MethodI
             end
         end
     end
-    function hook_emitted_function(method, code, world)
+    function hook_emitted_function(method, code)
         @compiler_assert last(call_stack) == method job
-        last_method_instance = pop!(call_stack)
+        pop!(call_stack)
     end
-    param_kwargs = [:cached             => false,
-                    :track_allocations  => false,
+    param_kwargs = [:track_allocations  => false,
                     :code_coverage      => false,
                     :static_alloc       => false,
                     :prefer_specsig     => true,
-                    :module_setup       => hook_module_setup,
-                    :module_activation  => hook_module_activation,
                     :emit_function      => hook_emit_function,
                     :emitted_function   => hook_emitted_function]
     if LLVM.version() >= v"8.0" && VERSION >= v"1.3.0-DEV.547"
@@ -150,99 +101,60 @@ function compile_method_instance(job::CompilerJob, method_instance::Core.MethodI
     end
     params = Base.CodegenParams(;param_kwargs...)
 
-    # get the code
-    ref = ccall(:jl_get_llvmf_defn, LLVM.API.LLVMValueRef,
-                (Any, UInt, Bool, Bool, Base.CodegenParams),
-                method_instance, world, #=wrapper=#false, #=optimize=#false, params)
-    if ref == C_NULL
-        throw(InternalCompilerError(job, "the Julia compiler could not generate LLVM IR"))
+    # generate IR
+    native_code = ccall(:jl_create_native, Ptr{Cvoid},
+                        (Vector{Core.MethodInstance}, Base.CodegenParams),
+                        [method_instance], params)
+    @assert native_code != C_NULL
+    llvm_mod_ref = ccall(:jl_get_llvm_module, LLVM.API.LLVMModuleRef,
+                         (Ptr{Cvoid},), native_code)
+    @assert llvm_mod_ref != C_NULL
+    llvm_mod = LLVM.Module(llvm_mod_ref)
+
+    # get the top-level code
+    code = Core.Compiler.inf_for_methodinstance(method_instance, world, world)
+
+    # get the top-level function index
+    llvm_func_idx = Ref{Int32}(-1)
+    llvm_specfunc_idx = Ref{Int32}(-1)
+    ccall(:jl_breakpoint, Nothing, ())
+    ccall(:jl_get_function_id, Nothing,
+          (Ptr{Cvoid}, Any, Ptr{Int32}, Ptr{Int32}),
+          native_code, code, llvm_func_idx, llvm_specfunc_idx)
+    @assert llvm_func_idx[] != -1
+    @assert llvm_specfunc_idx[] != -1
+
+    # get the top-level function)
+    llvm_func_ref = ccall(:jl_get_llvm_function, LLVM.API.LLVMValueRef,
+                     (Ptr{Cvoid}, UInt32), native_code, llvm_func_idx[]-1)
+    @assert llvm_func_ref != C_NULL
+    llvm_func = LLVM.Function(llvm_func_ref)
+    llvm_specfunc_ref = ccall(:jl_get_llvm_function, LLVM.API.LLVMValueRef,
+                         (Ptr{Cvoid}, UInt32), native_code, llvm_specfunc_idx[]-1)
+    @assert llvm_specfunc_ref != C_NULL
+    llvm_specfunc = LLVM.Function(llvm_specfunc_ref)
+
+    # configure the module
+    # NOTE: NVPTX::TargetMachine's data layout doesn't match the NVPTX user guide,
+    #       so we specify it ourselves
+    if Int === Int64
+        triple!(llvm_mod, "nvptx64-nvidia-cuda")
+        datalayout!(llvm_mod, "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64")
+    else
+        triple!(llvm_mod, "nvptx-nvidia-cuda")
+        datalayout!(llvm_mod, "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64")
     end
-    llvmf = LLVM.Function(ref)
-    ir = LLVM.parent(llvmf)
-    postprocess(ir)
 
-    return llvmf, dependencies
+    return llvm_specfunc, llvm_mod
 end
 
 function irgen(job::CompilerJob, method_instance::Core.MethodInstance, world)
-    entry, dependencies = @timeit_debug to "emission" compile_method_instance(job, method_instance, world)
-    mod = LLVM.parent(entry)
-
-    # link in dependent modules
-    @timeit_debug to "linking" begin
-        # we disable Julia's compilation cache not to poison it with GPU-specific code.
-        # as a result, we might get multiple modules for a single method instance.
-        cache = Dict{String,String}()
-
-        for called_method_instance in keys(dependencies)
-            llvmfs = dependencies[called_method_instance]
-
-            # link the first module
-            llvmf = popfirst!(llvmfs)
-            llvmfn = LLVM.name(llvmf)
-            link!(mod, LLVM.parent(llvmf))
-
-            # process subsequent duplicate modules
-            for dup_llvmf in llvmfs
-                if Base.JLOptions().debug_level >= 2
-                    # link them too, to ensure accurate backtrace reconstruction
-                    link!(mod, LLVM.parent(dup_llvmf))
-                else
-                    # don't link them, but note the called function name in a cache
-                    dup_llvmfn = LLVM.name(dup_llvmf)
-                    cache[dup_llvmfn] = llvmfn
-                end
-            end
-        end
-
-        # resolve function declarations with cached entries
-        for llvmf in filter(isdeclaration, collect(functions(mod)))
-            llvmfn = LLVM.name(llvmf)
-            if haskey(cache, llvmfn)
-                def_llvmfn = cache[llvmfn]
-                replace_uses!(llvmf, functions(mod)[def_llvmfn])
-
-                @compiler_assert isempty(uses(llvmf)) job
-                unsafe_delete!(LLVM.parent(llvmf), llvmf)
-            end
-        end
-    end
+    entry, mod = @timeit_debug to "emission" compile_method_instance(job, method_instance, world)
 
     # clean up incompatibilities
     @timeit_debug to "clean-up" for llvmf in functions(mod)
-        llvmfn = LLVM.name(llvmf)
-
         # only occurs in debug builds
         delete!(function_attributes(llvmf), EnumAttribute("sspstrong", 0, JuliaContext()))
-
-        # rename functions
-        if !isdeclaration(llvmf)
-            # Julia disambiguates local functions by prefixing with `#\d#`.
-            # since we don't use a global function namespace, get rid of those tags.
-            if occursin(r"^julia_#\d+#", llvmfn)
-                llvmfn′ = replace(llvmfn, r"#\d+#"=>"")
-                if !haskey(functions(mod), llvmfn′)
-                    LLVM.name!(llvmf, llvmfn′)
-                    llvmfn = llvmfn′
-                end
-            end
-
-            # anonymous functions are just named `#\d`, make that somewhat more readable
-            m = match(r"_#(\d+)_", llvmfn)
-            if m !== nothing
-                llvmfn′ = replace(llvmfn, m.match=>"_anonymous$(m.captures[1])_")
-                LLVM.name!(llvmf, llvmfn′)
-                llvmfn = llvmfn′
-            end
-
-            # finally, make function names safe for ptxas
-            # (LLVM should to do this, but fails, see eg. D17738 and D19126)
-            llvmfn′ = safe_name(llvmfn)
-            if llvmfn != llvmfn′
-                LLVM.name!(llvmf, llvmfn′)
-                llvmfn = llvmfn′
-            end
-        end
     end
 
     # add the global exception indicator flag

diff --git a/test/base.jl b/test/base.jl
@@ -14,7 +14,7 @@ function post17057_parent(arr::Ptr{Int64})
 end
 
 # bug: default module activation segfaulted on NULL child function if cached=false
-params = Base.CodegenParams(cached=false)
+params = Base.CodegenParams()
 if VERSION >= v"1.1.0-DEV.762"
     _dump_function(post17057_parent, Tuple{Ptr{Int64}},
                    #=native=#false, #=wrapper=#false, #=strip=#false,
@@ -31,4 +31,4 @@ end
 
 ############################################################################################
 
-end
+end