Forbid divergent execution of work-group barriers

Author: vchuravy

Project.toml

@@ -1,7 +1,7 @@
 name = "KernelAbstractions"
 uuid = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
 authors = ["Valentin Churavy <v.churavy@gmail.com> and contributors"]
-version = "0.9.33"
+version = "0.9.34"
 
 [deps]
 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"

src/KernelAbstractions.jl

@@ -284,6 +284,9 @@ end
 After a `@synchronize` statement all read and writes to global and local memory
 from each thread in the workgroup are visible in from all other threads in the
 workgroup.
+
+!!! note
+    `@synchronize()` must be encountered by all workitems of a work-group executing the kernel or by none at all.
 """
 macro synchronize()
     return quote
@@ -301,10 +304,15 @@ workgroup. `cond` is not allowed to have any visible sideffects.
 # Platform differences
   - `GPU`: This synchronization will only occur if the `cond` evaluates.
   - `CPU`: This synchronization will always occur.
+
+!!! warn
+    This variant of the `@synchronize` macro violates the requirement that `@synchronize` must be encountered
+    by all workitems of a work-group executing the kernel or by none at all.
+    Since v`0.9.34` this version of the macro is deprecated and lowers to `@synchronize()`
 """
 macro synchronize(cond)
     return quote
-        $(esc(cond)) && $__synchronize()
+        $__synchronize()
     end
 end
 

src/macros.jl

@@ -86,22 +86,24 @@ function transform_gpu!(def, constargs, force_inbounds)
         end
     end
     pushfirst!(def[:args], :__ctx__)
-    body = def[:body]
+    new_stmts = Expr[]
+    body = MacroTools.flatten(def[:body])
+    stmts = body.args
+    push!(new_stmts, Expr(:aliasscope))
+    push!(new_stmts, :(__active_lane__ = $__validindex(__ctx__)))
     if force_inbounds
-        body = quote
-            @inbounds $(body)
-        end
+        push!(new_stmts, Expr(:inbounds, true))
     end
-    body = quote
-        if $__validindex(__ctx__)
-            $(body)
-        end
-        return nothing
+    append!(new_stmts, split(emit_gpu, body.args))
+    if force_inbounds
+        push!(new_stmts, Expr(:inbounds, :pop))
     end
+    push!(new_stmts, Expr(:popaliasscope))
+    push!(new_stmts, :(return nothing))
     def[:body] = Expr(
         :let,
         Expr(:block, let_constargs...),
-        body,
+        Expr(:block, new_stmts...),
     )
     return
 end
@@ -127,7 +129,7 @@ function transform_cpu!(def, constargs, force_inbounds)
     if force_inbounds
         push!(new_stmts, Expr(:inbounds, true))
     end
-    append!(new_stmts, split(body.args))
+    append!(new_stmts, split(emit_cpu, body.args))
     if force_inbounds
         push!(new_stmts, Expr(:inbounds, :pop))
     end
@@ -167,6 +169,7 @@ end
 
 # TODO proper handling of LineInfo
 function split(
+        emit,
         stmts,
         indicies = Any[], private = Set{Symbol}(),
     )
@@ -197,7 +200,7 @@ function split(
             function recurse(expr::Expr)
                 expr = unblock(expr)
                 if is_scope_construct(expr) && any(find_sync, expr.args)
-                    new_args = unblock(split(expr.args, deepcopy(indicies), deepcopy(private)))
+                    new_args = unblock(split(emit, expr.args, deepcopy(indicies), deepcopy(private)))
                     return Expr(expr.head, new_args...)
                 else
                     return Expr(expr.head, map(recurse, expr.args)...)
@@ -246,7 +249,7 @@ function split(
     return new_stmts
 end
 
-function emit(loop)
+function emit_cpu(loop)
     idx = gensym(:I)
     for stmt in loop.indicies
         # splice index into the i = @index(Cartesian, $idx)
@@ -300,3 +303,38 @@ function emit(loop)
 
     return unblock(Expr(:block, stmts...))
 end
+
+function emit_gpu(loop)
+    stmts = Any[]
+    append!(stmts, loop.allocations)
+    for stmt in loop.private_allocations
+        if @capture(stmt, lhs_ = rhs_)
+            push!(stmts, :($lhs = $rhs))
+        else
+            error("@private $stmt not an assignment")
+        end
+    end
+
+    # don't emit empty loops
+    if !(isempty(loop.stmts) || all(s -> s isa LineNumberNode, loop.stmts))
+        body = Expr(:block, loop.stmts...)
+        body = postwalk(body) do expr
+            if @capture(expr, lhs_ = rhs_)
+                if lhs in loop.private
+                    error("Can't assign to variables marked private")
+                end
+            end
+            return expr
+        end
+        loopexpr = quote
+            if __active_lane__
+                $(loop.indicies...)
+                $(unblock(body))
+            end
+            $__synchronize()
+        end
+        push!(stmts, loopexpr)
+    end
+
+    return unblock(Expr(:block, stmts...))
+end