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 +86,35 @@ 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)
+
+    # fix convergence
+    active_stmts = Any[]
+    for stmt in stmts
+        push!(active_stmts, stmt)
+        has_sync = find_sync(stmt)
+        if has_sync
+            push!(new_stmts, Expr(:if, :__active_lane__, Expr(:block, active_stmts...)))
+            empty!(active_stmts)
         end
-        return nothing
     end
+    push!(new_stmts, Expr(:if, :__active_lane__, Expr(:block, active_stmts...)))
+    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