Merge pull request #2649 from xmake-io/batchjobs

Adding parallel compilation support for c++ modules

7 files changed, 690 insertions, 169 deletions

diff --git a/tests/apis/rules/xmake.lua b/tests/apis/rules/xmake.lua
index 5a6e5d546..7d67f0dca 100644
--- a/tests/apis/rules/xmake.lua
+++ b/tests/apis/rules/xmake.lua
@@ -58,6 +58,12 @@ rule("stub2")
     after_build(function (target)
         print("rule(stub2): after_build")
     end)
+    before_build_files(function (target)
+        print("rule(stub2): before_build_files")
+    end)
+    after_build_files(function (target)
+        print("rule(stub2): after_build_files")
+    end)
 
 -- define rule: stub1
 rule("stub1")
@@ -148,3 +154,10 @@ target("test")
     add_files("src/index.md")
     add_files("src/test.c.in",  {rule = "c code"})
 
+    before_build(function (target)
+        print("target: before_build")
+    end)
+
+    after_build(function (target)
+        print("target: after_build")
+    end)
diff --git a/xmake/actions/build/build.lua b/xmake/actions/build/build.lua
index a32b8d705..3eddf1610 100644
--- a/xmake/actions/build/build.lua
+++ b/xmake/actions/build/build.lua
@@ -37,25 +37,27 @@ function _clean_target(target)
     end
 end
 
--- add builtin batch jobs
-function _add_batchjobs_builtin(batchjobs, rootjob, target)
+-- add batch jobs for rules
+function _add_batchjobs_for_rules(batchjobs, rootjob, target, suffix)
 
     -- uses the rules script?
     local job, job_leaf
     for _, r in irpairs(target:orderules()) do -- reverse rules order for batchjobs:addjob()
-        local script = r:script("build")
+        local scriptname = "build" .. (suffix and ("_" .. suffix) or "")
+        local script = r:script(scriptname)
         if script then
-            if r:extraconf("build", "batch") then
-                job, job_leaf = assert(script(target, batchjobs, {rootjob = job or rootjob}), "rule(%s):on_build(): no returned job!", r:name())
+            if r:extraconf(scriptname, "batch") then
+                job, job_leaf = assert(script(target, batchjobs, {rootjob = job or rootjob}), "rule(%s):%s(): no returned job!", r:name(), scriptname)
             else
-                job = batchjobs:addjob("rule/" .. r:name() .. "/build", function (index, total)
+                job = batchjobs:addjob("rule/" .. r:name() .. "/" .. scriptname, function (index, total)
                     script(target, {progress = (index * 100) / total})
                 end, {rootjob = job or rootjob})
             end
         else
-            local buildcmd = r:script("buildcmd")
+            scriptname = "buildcmd" .. (suffix and ("_" .. suffix) or "")
+            local buildcmd = r:script(scriptname)
             if buildcmd then
-                job = batchjobs:addjob("rule/" .. r:name() .. "/build", function (index, total)
+                job = batchjobs:addjob("rule/" .. r:name() .. "/" .. scriptname, function (index, total)
                     local batchcmds_ = batchcmds.new({target = target})
                     buildcmd(target, batchcmds_, {progress =  (index * 100) / total})
                     batchcmds_:runcmds({dryrun = option.get("dry-run")})
@@ -63,6 +65,14 @@ function _add_batchjobs_builtin(batchjobs, rootjob, target)
             end
         end
     end
+    return job, job_leaf or job
+end
+
+-- add builtin batch jobs
+function _add_batchjobs_builtin(batchjobs, rootjob, target)
+
+    -- add batchjobs for rules
+    local job, job_leaf = _add_batchjobs_for_rules(batchjobs, rootjob, target)
 
     -- uses the builtin target script
     if not job and (target:is_static() or target:is_binary() or target:is_shared() or target:is_object()) then
@@ -125,19 +135,6 @@ function _add_batchjobs_for_target(batchjobs, rootjob, target)
         if after_build then
             after_build(target, {progress = progress})
         end
-        for _, r in ipairs(target:orderules()) do
-            local after_build = r:script("build_after")
-            if after_build then
-                after_build(target, {progress = progress})
-            else
-                local after_buildcmd = r:script("buildcmd_after")
-                if after_buildcmd then
-                    local batchcmds_ = batchcmds.new({target = target})
-                    after_buildcmd(target, batchcmds_, {progress = progress})
-                    batchcmds_:runcmds({dryrun = option.get("dry-run")})
-                end
-            end
-        end
 
         -- restore environments
         if oldenvs then
@@ -146,8 +143,14 @@ function _add_batchjobs_for_target(batchjobs, rootjob, target)
 
     end, {rootjob = rootjob})
 
+    -- add batchjobs for rules/build_after
+    local rules_job_build_after, rules_job_build_after_leaf = _add_batchjobs_for_rules(batchjobs, job_build_after, target, "after")
+
     -- add batch jobs for target, @note only on_build script support batch jobs
-    local job_build, job_build_leaf = _add_batchjobs(batchjobs, job_build_after, target)
+    local job_build, job_build_leaf = _add_batchjobs(batchjobs, rules_job_build_after_leaf or job_build_after, target)
+
+    -- add batchjobs for rules/build_before
+    local rules_job_build_before, rules_job_build_before_leaf = _add_batchjobs_for_rules(batchjobs, job_build_leaf, target, "before")
 
     -- add before_build job for target
     local job_build_before = batchjobs:addjob(target:name() .. "/before_build", function (index, total)
@@ -166,20 +169,7 @@ function _add_batchjobs_for_target(batchjobs, rootjob, target)
         if before_build then
             before_build(target, {progress = progress})
         end
-        for _, r in ipairs(target:orderules()) do
-            local before_build = r:script("build_before")
-            if before_build then
-                before_build(target, {progress = progress})
-            else
-                local before_buildcmd = r:script("buildcmd_before")
-                if before_buildcmd then
-                    local batchcmds_ = batchcmds.new({target = target})
-                    before_buildcmd(target, batchcmds_, {progress = progress})
-                    batchcmds_:runcmds({dryrun = option.get("dry-run")})
-                end
-            end
-        end
-    end, {rootjob = job_build_leaf})
+    end, {rootjob = rules_job_build_before_leaf or job_build_leaf})
     return job_build_before, job_build, job_build_after
 end
 
diff --git a/xmake/rules/c++/modules/modules_support/clang.lua b/xmake/rules/c++/modules/modules_support/clang.lua
index 35b7cdbc5..866669257 100644
--- a/xmake/rules/c++/modules/modules_support/clang.lua
+++ b/xmake/rules/c++/modules/modules_support/clang.lua
@@ -33,15 +33,14 @@ import("common")
 -- -fmodule-file=foo=build/.gens/Foo/rules/modules/cache/foo.pcm
 --
 function _add_module_to_mapper(target, module, bmi)
-    local cache = common.localcache():get("mapflags") or {}
+    local mapflags = common.localcache():get("mapflags") or {}
     local modulefileflag = get_modulefileflag(target)
     local mapflag = format("%s%s=%s", modulefileflag, module, bmi)
-    if table.contains(cache, mapflag) then
+    if table.contains(mapflags, mapflag) then
         return
     end
-    table.insert(cache, mapflag)
-    common.localcache():set("mapflags", cache)
-    common.localcache():save("mapflags")
+    table.insert(mapflags, mapflag)
+    common.localcache():set("mapflags", mapflags)
 end
 
 -- add a header unit into the mapper
@@ -50,16 +49,25 @@ end
 -- -fmodule-file=build/.gens/Foo/rules/modules/cache/foo.hpp.pcm
 --
 function _add_headerunit_to_mapper(target, bmi)
-    local cache = common.localcache():get("mapflags") or {}
+    local mapflags = common.memcache():get("mapflags") or {}
     local mapflag = format("%s%s", modulefileflag, bmi)
     if table.contains(cache, mapflag) then
         return
     end
-    table.insert(cache, mapflag)
-    common.localcache():set("mapflags", cache)
+    table.insert(mapflags, mapflag)
+    common.localcache():set("mapflags", mapflags)
+end
+
+-- flush mapflags to mapper file cache
+function _flush_mapflags_to_mapper()
     common.localcache():save("mapflags")
 end
 
+-- get mapflags from mapper
+function _get_mapflags_from_mapper()
+    return common.localcache():get("mapflags")
+end
+
 -- load module support for the current target
 function load(target)
     local cachedir = common.modules_cachedir(target)
@@ -143,13 +151,13 @@ function generate_dependencies(target, sourcebatch, opt)
                 progress.show(opt.progress, "${color.build.object}generating.cxx.module.deps %s", sourcefile)
             end
 
-            local outdir = path.translate(path.join(cachedir, path.directory(path.relative(sourcefile, projectdir))))
-            if not os.isdir(outdir) then
-                os.mkdir(outdir)
+            local outputdir = path.translate(path.join(cachedir, path.directory(path.relative(sourcefile, projectdir))))
+            if not os.isdir(outputdir) then
+                os.mkdir(outputdir)
             end
 
             -- no support of p1689 atm
-            local jsonfile = path.translate(path.join(outdir, path.filename(sourcefile) .. ".json"))
+            local jsonfile = path.translate(path.join(outputdir, path.filename(sourcefile) .. ".json"))
             common.fallback_generate_dependencies(target, jsonfile, sourcefile)
             changed = true
 
@@ -160,16 +168,45 @@ function generate_dependencies(target, sourcebatch, opt)
     return changed
 end
 
+-- generate target stl header units for batchjobs
+function generate_stl_headerunits_for_batchjobs(target, batchjobs, headerunits, opt)
+    local compinst = target:compiler("cxx")
+    local stlcachedir = common.stlmodules_cachedir(target)
+    local modulecachepathflag = get_modulecachepathflag(target)
+    local modulefileflag = get_modulefileflag(target)
+    assert(has_headerunitsupport(target), "compiler(clang): does not support c++ header units!")
+
+    -- build headerunits
+    local projectdir = os.projectdir()
+    for i, headerunit in ipairs(headerunits) do
+        local bmifile = path.join(stlcachedir, headerunit.name .. get_bmi_extension())
+        if not os.isfile(bmifile) then
+            batchjobs:addjob(headerunit.name, function (index, total)
+                depend.on_changed(function()
+                    progress.show((index * 100) / total, "${color.build.object}generating.cxx.headerunit.bmi %s", headerunit.name)
+                    local args = {modulecachepathflag .. stlcachedir, "-c", "-o", bmifile, "-x", "c++-system-header", headerunit.name}
+                    os.vrunv(compinst:program(), table.join(compinst:compflags({target = target}), args))
+                end, {dependfile = target:dependfile(bmifile), files = {headerunit.path}})
+            end, {rootjob = opt.rootjob})
+
+            -- libc++ have a builtin module mapper
+            if not target:data_set("cxx.modules.use_libc++") then
+                _add_headerunit_to_mapper(target, bmifile)
+            end
+        end
+    end
+    _flush_mapflags_to_mapper()
+end
+
 -- generate target stl header units for batchcmds
 function generate_stl_headerunits_for_batchcmds(target, batchcmds, headerunits, opt)
     local compinst = target:compiler("cxx")
 
     -- get cachedirs
     local stlcachedir = common.stlmodules_cachedir(target)
-
-    -- get headerunits flags
     local modulecachepathflag = get_modulecachepathflag(target)
     local modulefileflag = get_modulefileflag(target)
+    assert(has_headerunitsupport(target), "compiler(clang): does not support c++ header units!")
 
     -- build headerunits
     local projectdir = os.projectdir()
@@ -189,6 +226,61 @@ function generate_stl_headerunits_for_batchcmds(target, batchcmds, headerunits,
         depmtime = math.max(depmtime, os.mtime(bmifile))
     end
     batchcmds:set_depmtime(depmtime)
+    _flush_mapflags_to_mapper()
+end
+
+-- generate target user header units for batchjobs
+function generate_user_headerunits_for_batchjobs(target, batchjobs, headerunits, opt)
+    local compinst = target:compiler("cxx")
+    assert(has_headerunitsupport(target), "compiler(clang): does not support c++ header units!")
+
+    -- get cachedirs
+    local cachedir = common.modules_cachedir(target)
+    local modulecachepathflag = get_modulecachepathflag(target)
+    local emitmoduleflag = get_emitmoduleflag(target)
+    local modulefileflag = get_modulefileflag(target)
+
+    -- build headerunits
+    local objectfiles = {}
+    local flags = {}
+    local projectdir = os.projectdir()
+    for _, headerunit in ipairs(headerunits) do
+        local file = path.relative(headerunit.path, target:scriptdir())
+        local objectfile = target:objectfile(file)
+
+        local outputdir
+        if headerunit.type == ":quote" then
+            outputdir = path.join(cachedir, path.directory(path.relative(headerunit.path, projectdir)))
+        else
+            outputdir = path.join(cachedir, path.directory(headerunit.path))
+        end
+        local bmifilename = path.basename(objectfile) .. get_bmi_extension()
+        local bmifile = (outputdir and path.join(outputdir, bmifilename) or bmifilename)
+        batchjobs:addjob(headerunit.name, function (index, total)
+            depend.on_changed(function()
+                progress.show((index * 100) / total, "${color.build.object}generating.cxx.headerunit.bmi %s", headerunit.name)
+                local objectdir = path.directory(objectfile)
+                if not os.isdir(objectdir) then
+                    os.mkdir(objectdir)
+                end
+                if not os.isdir(outputdir) then
+                    os.mkdir(outputdir)
+                end
+
+                -- generate headerunit
+                local args = { modulecachepathflag .. cachedir, emitmoduleflag, "-c", "-o", bmifile}
+                if headerunit.type == ":quote" then
+                    table.join2(args, {"-I", path.directory(headerunit.path), "-x", "c++-user-header", headerunit.path})
+                elseif headerunit.type == ":angle" then
+                    table.join2(args, {"-x", "c++-system-header", headerunit.name})
+                end
+                os.vrunv(compinst:program(), table.join(compinst:compflags({target = target}), args))
+
+            end, {dependfile = target:dependfile(bmifile), files = {headerunit.path}})
+        end, {rootjob = opt.rootjob})
+        _add_headerunit_to_mapper(target, bmifile)
+    end
+    _flush_mapflags_to_mapper()
 end
 
 -- generate target user header units for batchcmds
@@ -198,8 +290,6 @@ function generate_user_headerunits_for_batchcmds(target, batchcmds, headerunits,
 
     -- get cachedirs
     local cachedir = common.modules_cachedir(target)
-
-    -- get headerunits flags
     local modulecachepathflag = get_modulecachepathflag(target)
     local emitmoduleflag = get_emitmoduleflag(target)
     local modulefileflag = get_modulefileflag(target)
@@ -213,16 +303,16 @@ function generate_user_headerunits_for_batchcmds(target, batchcmds, headerunits,
         local file = path.relative(headerunit.path, target:scriptdir())
         local objectfile = target:objectfile(file)
 
-        local outdir
+        local outputdir
         if headerunit.type == ":quote" then
-            outdir = path.join(cachedir, path.directory(path.relative(headerunit.path, projectdir)))
+            outputdir = path.join(cachedir, path.directory(path.relative(headerunit.path, projectdir)))
         else
-            outdir = path.join(cachedir, path.directory(headerunit.path))
+            outputdir = path.join(cachedir, path.directory(headerunit.path))
         end
-        batchcmds:mkdir(outdir)
+        batchcmds:mkdir(outputdir)
 
         local bmifilename = path.basename(objectfile) .. get_bmi_extension()
-        local bmifile = (outdir and path.join(outdir, bmifilename) or bmifilename)
+        local bmifile = (outputdir and path.join(outputdir, bmifilename) or bmifilename)
         batchcmds:mkdir(path.directory(objectfile))
 
         local args = { modulecachepathflag .. cachedir, emitmoduleflag, "-c", "-o", bmifile}
@@ -240,26 +330,19 @@ function generate_user_headerunits_for_batchcmds(target, batchcmds, headerunits,
         depmtime = math.max(depmtime, os.mtime(bmifile))
     end
     batchcmds:set_depmtime(depmtime)
+    _flush_mapflags_to_mapper()
 end
 
--- build module files for batchcmds
-function build_modules_for_batchcmds(target, batchcmds, objectfiles, modules, opt)
+-- build module files for batchjobs
+function build_modules_for_batchjobs(target, batchjobs, objectfiles, modules, opt)
     local compinst = target:compiler("cxx")
-
-    -- get cachedirs
     local cachedir = common.modules_cachedir(target)
-
-    -- get modules flags
     local modulecachepathflag = get_modulecachepathflag(target)
     local emitmoduleinterfaceflag = get_emitmoduleinterfaceflag(target)
 
-    -- append module mapper flags
-    local cache = common.localcache():get("mapflags") or {}
-    target:add("cxxflags", cache, {force = true})
-
     -- build modules
     local common_args = {modulecachepathflag .. cachedir}
-    local depmtime = 0
+    local provided_modules = {}
     for _, objectfile in ipairs(objectfiles) do
         local m = modules[objectfile]
         if m and m.provides then
@@ -276,16 +359,98 @@ function build_modules_for_batchcmds(target, batchcmds, objectfiles, modules, op
             end
 
             local bmifile = provide.bmi
+            local moduleinfo = table.copy(provide)
+            moduleinfo.job = batchjobs:newjob(provide.sourcefile, function (index, total)
+                depend.on_changed(function()
+                    progress.show((index * 100) / total, "${color.build.object}generating.cxx.module.bmi %s", name)
+                    local objectdir = path.directory(objectfile)
+                    if not os.isdir(objectdir) then
+                        os.mkdir(objectdir)
+                    end
+                    -- append module mapper flags first
+                    -- @note we add it at the end to ensure that the full mapflags are already stored in the mapper
+                    if not target:data("cxx.add_modules_mapflags") then
+                        local mapflags = _get_mapflags_from_mapper()
+                        if mapflags then
+                            target:add("cxxflags", mapflags, {force = true})
+                        end
+                        target:data_set("cxx.add_modules_mapflags", true)
+                    end
+                    local args = {emitmoduleinterfaceflag, "-c", "-x", "c++-module", "--precompile", provide.sourcefile, "-o", bmifile}
+                    os.vrunv(compinst:program(), table.join(compinst:compflags({target = target}), common_args, args))
+                    os.vrunv(compinst:program(), table.join(compinst:compflags({target = target}), common_args, {bmifile}, {"-c", "-o", objectfile}))
+                end, {dependfile = target:dependfile(bmifile), files = {provide.sourcefile}})
+            end)
+            if m.requires then
+                moduleinfo.deps = table.keys(m.requires)
+            end
+            moduleinfo.name = name
+            provided_modules[name] = moduleinfo
+            _add_module_to_mapper(target, name, bmifile)
+            target:add("objectfiles", objectfile)
+        end
+    end
+    _flush_mapflags_to_mapper()
+
+    -- build batchjobs for modules
+    common.build_batchjobs_for_modules(provided_modules, batchjobs, opt.rootjob)
+end
+
+-- build module files for batchcmds
+function build_modules_for_batchcmds(target, batchcmds, objectfiles, modules, opt)
+    local compinst = target:compiler("cxx")
+    local cachedir = common.modules_cachedir(target)
+    local modulecachepathflag = get_modulecachepathflag(target)
+    local emitmoduleinterfaceflag = get_emitmoduleinterfaceflag(target)
+
+    -- we need update mapper first
+    for _, objectfile in ipairs(objectfiles) do
+        local m = modules[objectfile]
+        if m and m.provides then
+            -- assume there that provides is only one, until we encounter the case
+            local length = 0
+            local name, provide
+            for k, v in pairs(m.provides) do
+                length = length + 1
+                name = k
+                provide = v
+                if length > 1 then
+                    raise("multiple provides are not supported now!")
+                end
+            end
+
+            local bmifile = provide.bmi
+            _add_module_to_mapper(target, name, bmifile)
+            target:add("objectfiles", objectfile)
+        end
+    end
+    _flush_mapflags_to_mapper()
+
+    -- append module mapper flags
+    local mapflags = _get_mapflags_from_mapper()
+    if mapflags then
+        target:add("cxxflags", mapflags, {force = true})
+    end
+
+    -- build modules
+    local depmtime = 0
+    local common_args = {modulecachepathflag .. cachedir}
+    for _, objectfile in ipairs(objectfiles) do
+        local m = modules[objectfile]
+        if m and m.provides then
+            local name, provide
+            for k, v in pairs(m.provides) do
+                name = k
+                provide = v
+                break
+            end
+            local bmifile = provide.bmi
             local args = { emitmoduleinterfaceflag, "-c", "-x", "c++-module", "--precompile", provide.sourcefile, "-o", bmifile }
             batchcmds:show_progress(opt.progress, "${color.build.object}generating.cxx.module.bmi %s", name)
             batchcmds:mkdir(path.directory(objectfile))
             batchcmds:vrunv(compinst:program(), table.join(compinst:compflags({target = target}), common_args, args))
             batchcmds:vrunv(compinst:program(), table.join(compinst:compflags({target = target}), common_args, {bmifile}, {"-c", "-o", objectfile}))
             batchcmds:add_depfiles(provide.sourcefile)
-
-            _add_module_to_mapper(target, name, bmifile)
-
-            target:add("objectfiles", objectfile)
             depmtime = math.max(depmtime, os.mtime(bmifile))
         end
     end
@@ -429,4 +594,4 @@ function has_headerunitsupport(target)
         _g.support_headerunits = support_headerunits or false
     end
     return support_headerunits or nil
-end
\ No newline at end of file
+end
diff --git a/xmake/rules/c++/modules/modules_support/common.lua b/xmake/rules/c++/modules/modules_support/common.lua
index 34d014238..31ab47a81 100644
--- a/xmake/rules/c++/modules/modules_support/common.lua
+++ b/xmake/rules/c++/modules/modules_support/common.lua
@@ -411,20 +411,26 @@ function fallback_generate_dependencies(target, jsonfile, sourcefile)
     io.writefile(jsonfile, jsondata)
 end
 
--- generate dependencies
-function generate_dependencies(target, sourcebatch, opt)
-    local modules = localcache():get("modules")
-    local changed = modules_support(target).generate_dependencies(target, sourcebatch, opt)
-    if changed or modules == nil then
-        local moduleinfos = load_moduleinfos(target, sourcebatch)
-        modules = parse_dependency_data(target, moduleinfos)
-        localcache():set("modules", modules)
-        localcache():save()
+-- get module dependencies
+function get_module_dependencies(target, sourcebatch, opt)
+    local cachekey = target:name() .. "/" .. sourcebatch.rulename
+    local modules = memcache():get2("modules", cachekey)
+    if modules == nil then
+        modules = localcache():get2("modules", cachekey)
+        opt.progress = opt.progress or 0
+        local changed = modules_support(target).generate_dependencies(target, sourcebatch, opt)
+        if changed or modules == nil then
+            local moduleinfos = load_moduleinfos(target, sourcebatch)
+            modules = parse_dependency_data(target, moduleinfos)
+            localcache():set2("modules", cachekey, modules)
+            localcache():save()
+        end
+        memcache():set2("modules", cachekey, modules)
     end
     return modules
 end
 
--- generate headerunits for batchjobs, TODO
+-- generate headerunits for batchjobs
 function generate_headerunits_for_batchjobs(target, batchjobs, sourcebatch, modules, opt)
 
     -- get headerunits info
@@ -435,11 +441,11 @@ function generate_headerunits_for_batchjobs(target, batchjobs, sourcebatch, modu
     local headerunits_flags
     if stl_headerunits then
         headerunits_flags = headerunits_flags or {}
---        table.join2(headerunits_flags, modules_support(target).generate_stl_headerunits_for_batchjobs(target, batchjobs, stl_headerunits, opt))
+        table.join2(headerunits_flags, modules_support(target).generate_stl_headerunits_for_batchjobs(target, batchjobs, stl_headerunits, opt))
     end
     if headerunits then
         headerunits_flags = headerunits_flags or {}
---        table.join2(headerunits_flags, modules_support(target).generate_user_headerunits_for_batchjobs(target, batchjobs, headerunits, opt))
+        table.join2(headerunits_flags, modules_support(target).generate_user_headerunits_for_batchjobs(target, batchjobs, headerunits, opt))
     end
     return headerunits_flags
 end
@@ -463,10 +469,50 @@ function generate_headerunits_for_batchcmds(target, batchcmds, sourcebatch, modu
     end
 end
 
--- build modules for batchjobs, TODO
+-- build batch jobs for module dependencies
+function _build_batchjobs_for_moduledeps(modules, batchjobs, rootjob, jobrefs, moduleinfo)
+    local targetjob_ref = jobrefs[moduleinfo.name]
+    if targetjob_ref then
+        batchjobs:add(targetjob_ref, rootjob)
+    else
+        local modulejob = batchjobs:add(moduleinfo.job, rootjob)
+        if modulejob then
+            jobrefs[moduleinfo.name] = modulejob
+            for _, depname in ipairs(moduleinfo.deps) do
+                local dep = modules[depname]
+                if dep then -- maybe nil, e.g. `import <string>;`
+                    _build_batchjobs_for_moduledeps(modules, batchjobs, modulejob, jobrefs, dep)
+                end
+            end
+        end
+    end
+end
+
+-- build batchjobs for modules
+function build_batchjobs_for_modules(modules, batchjobs, rootjob)
+    local depset = hashset.new()
+    for _, moduleinfo in pairs(modules) do
+        assert(moduleinfo.job)
+        for _, depname in ipairs(moduleinfo.deps) do
+            depset:insert(depname)
+        end
+    end
+    local modules_root = {}
+    for _, moduleinfo in pairs(modules) do
+        if not depset:has(moduleinfo.name) then
+            table.insert(modules_root, moduleinfo)
+        end
+    end
+    local jobrefs = {}
+    for _, moduleinfo in pairs(modules_root) do
+        _build_batchjobs_for_moduledeps(modules, batchjobs, rootjob, jobrefs, moduleinfo)
+    end
+end
+
+-- build modules for batchjobs
 function build_modules_for_batchjobs(target, batchjobs, sourcebatch, modules, opt)
     local objectfiles = sort_modules_by_dependencies(sourcebatch.objectfiles, modules)
---    modules_support(target).build_modules_for_batchjobs(target, batchjobs, objectfiles, modules, opt)
+    modules_support(target).build_modules_for_batchjobs(target, batchjobs, objectfiles, modules, opt)
 end
 
 -- build modules for batchcmds
@@ -475,6 +521,7 @@ function build_modules_for_batchcmds(target, batchcmds, sourcebatch, modules, op
     modules_support(target).build_modules_for_batchcmds(target, batchcmds, objectfiles, modules, opt)
 end
 
+-- append headerunits objectfiles to link
 function append_headerunits_objectfiles(target)
     local cache = localcache():get("headerunit_objectfiles") or {}
     if target:is_binary() then
@@ -484,4 +531,4 @@ function append_headerunits_objectfiles(target)
     elseif target:is_shared() == "shared" then
         target:add("shflags", cache, {force = true})
     end
-end
\ No newline at end of file
+end
diff --git a/xmake/rules/c++/modules/modules_support/gcc.lua b/xmake/rules/c++/modules/modules_support/gcc.lua
index 99045cf54..a37346bad 100644
--- a/xmake/rules/c++/modules/modules_support/gcc.lua
+++ b/xmake/rules/c++/modules/modules_support/gcc.lua
@@ -131,15 +131,15 @@ function generate_dependencies(target, sourcebatch, opt)
                 progress.show(opt.progress, "${color.build.object}generating.cxx.module.deps %s", sourcefile)
             end
 
-            local outdir = path.translate(path.join(cachedir, path.directory(path.relative(sourcefile, projectdir))))
-            if not os.isdir(outdir) then
-                os.mkdir(outdir)
+            local outputdir = path.translate(path.join(cachedir, path.directory(path.relative(sourcefile, projectdir))))
+            if not os.isdir(outputdir) then
+                os.mkdir(outputdir)
             end
 
-            local jsonfile = path.translate(path.join(outdir, path.filename(sourcefile) .. ".json"))
+            local jsonfile = path.translate(path.join(outputdir, path.filename(sourcefile) .. ".json"))
             if trtbdflag and depfileflag and depoutputflag then
-                local ifile = path.translate(path.join(outdir, path.filename(sourcefile) .. ".i"))
-                local dfile = path.translate(path.join(outdir, path.filename(sourcefile) .. ".d"))
+                local ifile = path.translate(path.join(outputdir, path.filename(sourcefile) .. ".i"))
+                local dfile = path.translate(path.join(outputdir, path.filename(sourcefile) .. ".d"))
                 local args = {sourcefile, "-MD", "-MT", jsonfile, "-MF", dfile, depfileflag .. jsonfile, trtbdflag, depoutputfile .. target:objectfile(sourcefile), "-o", ifile}
                 os.vrunv(compinst:program(), table.join(compinst:compflags({target = target}), common_args, args), {envs = vcvars})
             else
@@ -154,12 +154,33 @@ function generate_dependencies(target, sourcebatch, opt)
     return changed
 end
 
+-- generate target stl header units for batchjobs
+function generate_stl_headerunits_for_batchjobs(target, batchjobs, headerunits, opt)
+    local compinst = target:compiler("cxx")
+    local mapper_file = _get_module_mapper()
+    local stlcachedir = common.stlmodules_cachedir(target)
+
+    -- build headerunits
+    local projectdir = os.projectdir()
+    for _, headerunit in ipairs(headerunits) do
+        local bmifile = path.join(stlcachedir, headerunit.name .. get_bmi_extension())
+        if not os.isfile(bmifile) then
+            batchjobs:addjob(headerunit.name, function (index, total)
+                depend.on_changed(function()
+                    progress.show((index * 100) / total, "${color.build.object}generating.cxx.headerunit.bmi %s", headerunit.name)
+                    local args = {"-c", "-x", "c++-system-header", headerunit.name}
+                    os.vrunv(compinst:program(), table.join(compinst:compflags({target = target}), args))
+                end, {dependfile = target:dependfile(bmifile), files = {headerunit.path}})
+            end, {rootjob = opt.rootjob})
+        end
+        _add_module_to_mapper(mapper_file, headerunit.path, path.absolute(bmifile, projectdir))
+    end
+end
+
 -- generate target stl header units for batchcmds
 function generate_stl_headerunits_for_batchcmds(target, batchcmds, headerunits, opt)
     local compinst = target:compiler("cxx")
     local mapper_file = _get_module_mapper()
-
-    -- get cachedirs
     local stlcachedir = common.stlmodules_cachedir(target)
 
     -- build headerunits
@@ -172,21 +193,69 @@ function generate_stl_headerunits_for_batchcmds(target, batchcmds, headerunits,
             batchcmds:show_progress(opt.progress, "${color.build.object}generating.cxx.headerunit.bmi %s", headerunit.name)
             batchcmds:vrunv(compinst:program(), table.join(compinst:compflags({target = target}), args))
         end
-
         batchcmds:add_depfiles(headerunit.path)
-
         _add_module_to_mapper(mapper_file, headerunit.path, path.absolute(bmifile, projectdir))
         depmtime = math.max(depmtime, os.mtime(bmifile))
     end
     batchcmds:set_depmtime(depmtime)
 end
 
+-- generate target user header units for batchjobs
+function generate_user_headerunits_for_batchjobs(target, batchjobs, headerunits, opt)
+    local compinst = target:compiler("cxx")
+    local mapper_file = _get_module_mapper()
+    local cachedir = common.modules_cachedir(target)
+
+    -- build headerunits
+    local projectdir = os.projectdir()
+    for _, headerunit in ipairs(headerunits) do
+        local file = path.relative(headerunit.path, projectdir)
+        local objectfile = target:objectfile(file)
+        local outputdir
+        local headerunit_path
+        if headerunit.type == ":quote" then
+            outputdir = path.join(cachedir, path.directory(path.relative(headerunit.path, projectdir)))
+        else
+            outputdir = path.join(cachedir, path.directory(headerunit.path))
+        end
+        local bmifilename = path.basename(objectfile) .. get_bmi_extension()
+        local bmifile = (outputdir and path.join(outputdir, bmifilename) or bmifilename)
+        if headerunit.type == ":quote" then
+            headerunit_path = path.join(".", path.relative(headerunit.path, projectdir))
+        elseif headerunit.type == ":angle" then
+            -- if path is relative then its a subtarget path
+            headerunit_path = path.is_absolute(headerunit.path) and headerunit.path or path.join(".", headerunit.path)
+        end
+        batchjobs:addjob(headerunit.name, function (index, total)
+            depend.on_changed(function()
+                progress.show((index * 100) / total, "${color.build.object}generating.cxx.headerunit.bmi %s", headerunit.name)
+                local objectdir = path.directory(objectfile)
+                if not os.isdir(objectdir) then
+                    os.mkdir(objectdir)
+                end
+                if not os.isdir(outputdir) then
+                    os.mkdir(outputdir)
+                end
+
+                -- generate headerunit
+                local args = { "-c" }
+                if headerunit.type == ":quote" then
+                    table.join2(args, { "-I", path.directory(path.relative(headerunit.path, projectdir)), "-x", "c++-user-header", headerunit.name })
+                elseif headerunit.type == ":angle" then
+                    table.join2(args, { "-x", "c++-system-header", headerunit.name })
+                end
+                os.vrunv(compinst:program(), table.join(compinst:compflags({target = target}), args))
+
+            end, {dependfile = target:dependfile(bmifile), files = {headerunit.path}})
+        end, {rootjob = opt.rootjob})
+        _add_module_to_mapper(mapper_file, headerunit_path, path.absolute(bmifile, projectdir))
+    end
+end
+
 -- generate target user header units for batchcmds
 function generate_user_headerunits_for_batchcmds(target, batchcmds, headerunits, opt)
     local compinst = target:compiler("cxx")
     local mapper_file = _get_module_mapper()
-
-    -- get cachedirs
     local cachedir = common.modules_cachedir(target)
 
     -- build headerunits
@@ -195,17 +264,16 @@ function generate_user_headerunits_for_batchcmds(target, batchcmds, headerunits,
     for _, headerunit in ipairs(headerunits) do
         local file = path.relative(headerunit.path, projectdir)
         local objectfile = target:objectfile(file)
-
-        local outdir
+        local outputdir
         if headerunit.type == ":quote" then
-            outdir = path.join(cachedir, path.directory(path.relative(headerunit.path, projectdir)))
+            outputdir = path.join(cachedir, path.directory(path.relative(headerunit.path, projectdir)))
         else
-            outdir = path.join(cachedir, path.directory(headerunit.path))
+            outputdir = path.join(cachedir, path.directory(headerunit.path))
         end
-        batchcmds:mkdir(outdir)
+        batchcmds:mkdir(outputdir)
 
         local bmifilename = path.basename(objectfile) .. get_bmi_extension()
-        local bmifile = (outdir and path.join(outdir, bmifilename) or bmifilename)
+        local bmifile = (outputdir and path.join(outputdir, bmifilename) or bmifilename)
         batchcmds:mkdir(path.directory(objectfile))
 
         local args = { "-c" }
@@ -229,13 +297,63 @@ function generate_user_headerunits_for_batchcmds(target, batchcmds, headerunits,
     batchcmds:set_depmtime(depmtime)
 end
 
+-- build module files for batchjobs
+function build_modules_for_batchjobs(target, batchjobs, objectfiles, modules, opt)
+    local compinst = target:compiler("cxx")
+    local mapper_file = _get_module_mapper()
+    local common_args = {"-x", "c++"}
+    local cachedir = common.modules_cachedir(target)
+
+    -- build modules
+    local projectdir = os.projectdir()
+    local provided_modules = {}
+    for _, objectfile in ipairs(objectfiles) do
+        local m = modules[objectfile]
+        if m and m.provides then
+            -- assume there that provides is only one, until we encounter the case
+            local length = 0
+            local name, provide
+            for k, v in pairs(m.provides) do
+                length = length + 1
+                name = k
+                provide = v
+                if length > 1 then
+                    raise("multiple provides are not supported now!")
+                end
+            end
+
+            local bmifile = provide.bmi
+            local moduleinfo = table.copy(provide)
+            moduleinfo.job = batchjobs:newjob(provide.sourcefile, function (index, total)
+                depend.on_changed(function()
+                    progress.show((index * 100) / total, "${color.build.object}generating.cxx.module.bmi %s", name)
+                    local objectdir = path.directory(objectfile)
+                    if not os.isdir(objectdir) then
+                        os.mkdir(objectdir)
+                    end
+                    local args = {"-o", objectfile, "-c", provide.sourcefile}
+                    os.vrunv(compinst:program(), table.join(compinst:compflags({target = target}), common_args, args))
+                end, {dependfile = target:dependfile(bmifile), files = {provide.sourcefile}})
+            end)
+            if m.requires then
+                moduleinfo.deps = table.keys(m.requires)
+            end
+            moduleinfo.name = name
+            provided_modules[name] = moduleinfo
+            _add_module_to_mapper(mapper_file, name, path.absolute(bmifile, projectdir))
+            target:add("objectfiles", objectfile)
+        end
+    end
+
+    -- build batchjobs for modules
+    common.build_batchjobs_for_modules(provided_modules, batchjobs, opt.rootjob)
+end
+
 -- build module files for batchcmds
 function build_modules_for_batchcmds(target, batchcmds, objectfiles, modules, opt)
     local compinst = target:compiler("cxx")
     local mapper_file = _get_module_mapper()
     local common_args = {"-x", "c++"}
-
-    -- get cachedirs
     local cachedir = common.modules_cachedir(target)
 
     -- build modules
@@ -336,4 +454,4 @@ function get_depoutputflag(target)
         _g.depoutputflag = depoutputflag or false
     end
     return depoutputflag or nil
-end
\ No newline at end of file
+end
diff --git a/xmake/rules/c++/modules/modules_support/msvc.lua b/xmake/rules/c++/modules/modules_support/msvc.lua
index b2b2e605a..bcc781248 100644
--- a/xmake/rules/c++/modules/modules_support/msvc.lua
+++ b/xmake/rules/c++/modules/modules_support/msvc.lua
@@ -34,16 +34,25 @@ import("common")
 -- /headerUnit:angle glm/mat4x4.hpp=Users\arthu\AppData\Local\.xmake\packages\g\glm\0.9.9+8\91454f3ee0be416cb9c7452970a2300f\include\glm\mat4x4.hpp.ifc
 --
 function _add_module_to_mapper(argument, module)
-    local cache = common.localcache():get("mapflags") or {}
+    local mapflags = common.localcache():get("mapflags") or {}
     local mapflag = format("%s %s", argument, module)
-    if table.contains(cache, mapflag) then
+    if table.contains(mapflags, mapflag) then
         return
     end
-    table.insert(cache, mapflag)
-    common.localcache():set("mapflags", cache)
+    table.insert(mapflags, mapflag)
+    common.localcache():set("mapflags", mapflags)
+end
+
+-- flush mapflags to mapper file cache
+function _flush_mapflags_to_mapper()
     common.localcache():save("mapflags")
 end
 
+-- get mapflags from mapper
+function _get_mapflags_from_mapper()
+    return common.localcache():get("mapflags")
+end
+
 -- add an objectfile to the linker args
 --
 -- e.g
@@ -150,11 +159,12 @@ function generate_dependencies(target, sourcebatch, opt)
     return changed
 end
 
--- generate target stl header units for batchcmds
-function generate_stl_headerunits_for_batchcmds(target, batchcmds, headerunits, opt)
+-- generate target stl header units for batchjobs
+function generate_stl_headerunits_for_batchjobs(target, batchjobs, headerunits, opt)
     local compinst = target:compiler("cxx")
     local toolchain = target:toolchain("msvc")
     local vcvars = toolchain:config("vcvars")
+    local stlcachedir = common.stlmodules_cachedir(target)
 
     -- get flags
     local exportheaderflag = get_exportheaderflag(target)
@@ -163,9 +173,40 @@ function generate_stl_headerunits_for_batchcmds(target, batchcmds, headerunits,
     local ifcoutputflag = get_ifcoutputflag(target)
     assert(headerunitflag and headernameflag and exportheaderflag, "compiler(msvc): does not support c++ header units!")
 
-    -- get cachedirs
+    -- build headerunits
+    local common_args = {"-TP", exportheaderflag, "-c"}
+    for _, headerunit in ipairs(headerunits) do
+        local bmifile = path.join(stlcachedir, headerunit.name .. get_bmi_extension())
+        local objectfile = bmifile .. ".obj"
+        if not os.isfile(bmifile) or not os.isfile(objectfile) then
+            batchjobs:addjob(headerunit.name, function(index, total)
+               depend.on_changed(function()
+                    progress.show((index * 100) / total, "${color.build.object}generating.cxx.headerunit.bmi %s", headerunit.name)
+                    local args = {headernameflag .. ":angle", headerunit.name, ifcoutputflag, stlcachedir, "-Fo" .. objectfile}
+                    os.vrunv(compinst:program(), table.join(compinst:compflags({target = target}), common_args, args), {envs = vcvars})
+               end, {dependfile = target:dependfile(bmifile), files = {headerunit.path}})
+            end, {rootjob = opt.rootjob})
+            _add_module_to_mapper(headerunitflag .. ":angle", headerunit.name .. "=" .. path.filename(headerunit.name) .. get_bmi_extension())
+            _add_objectfile_to_link_arguments(objectfile)
+        end
+    end
+    _flush_mapflags_to_mapper()
+end
+
+-- generate target stl header units for batchcmds
+function generate_stl_headerunits_for_batchcmds(target, batchcmds, headerunits, opt)
+    local compinst = target:compiler("cxx")
+    local toolchain = target:toolchain("msvc")
+    local vcvars = toolchain:config("vcvars")
     local stlcachedir = common.stlmodules_cachedir(target)
 
+    -- get flags
+    local exportheaderflag = get_exportheaderflag(target)
+    local headerunitflag = get_headerunitflag(target)
+    local headernameflag = get_headernameflag(target)
+    local ifcoutputflag = get_ifcoutputflag(target)
+    assert(headerunitflag and headernameflag and exportheaderflag, "compiler(msvc): does not support c++ header units!")
+
     -- build headerunits
     local common_args = {"-TP", exportheaderflag, "-c"}
     local depmtime = 0
@@ -184,13 +225,15 @@ function generate_stl_headerunits_for_batchcmds(target, batchcmds, headerunits,
         depmtime = math.max(depmtime, os.mtime(bmifile))
     end
     batchcmds:set_depmtime(depmtime)
+    _flush_mapflags_to_mapper()
 end
 
 -- generate target user header units for batchcmds
-function generate_user_headerunits_for_batchcmds(target, batchcmds, headerunits, opt)
+function generate_user_headerunits_for_batchjobs(target, batchjobs, headerunits, opt)
     local compinst = target:compiler("cxx")
     local toolchain = target:toolchain("msvc")
     local vcvars = toolchain:config("vcvars")
+    local cachedir = common.modules_cachedir(target)
 
     -- get flags
     local exportheaderflag = get_exportheaderflag(target)
@@ -199,9 +242,58 @@ function generate_user_headerunits_for_batchcmds(target, batchcmds, headerunits,
     local ifcoutputflag = get_ifcoutputflag(target)
     assert(headerunitflag and headernameflag and exportheaderflag, "compiler(msvc): does not support c++ header units!")
 
-    -- get cachedirs
+    -- build headerunits
+    local common_args = {"-TP", exportheaderflag, "-c"}
+    local projectdir = os.projectdir()
+    for _, headerunit in ipairs(headerunits) do
+        local file = path.relative(headerunit.path, target:scriptdir())
+        local objectfile = target:objectfile(file)
+        local outputdir
+        if headerunit.type == ":quote" then
+            outputdir = path.join(cachedir, path.directory(path.relative(headerunit.path, projectdir)))
+        else
+            -- if path is relative then its a subtarget path
+            outputdir = path.join(cachedir, path.is_absolute(headerunit.path) and path.directory(headerunit.path):sub(3) or headerunit.path)
+        end
+        local bmifilename = path.basename(objectfile) .. get_bmi_extension()
+        local bmifile = path.join(outputdir, bmifilename)
+        batchjobs:addjob(headerunit.name, function (index, total)
+            depend.on_changed(function()
+                progress.show((index * 100) / total, "${color.build.object}generating.cxx.headerunit.bmi %s", headerunit.name)
+                local objectdir = path.directory(objectfile)
+                if not os.isdir(objectdir) then
+                    os.mkdir(objectdir)
+                end
+                if not os.isdir(outputdir) then
+                    os.mkdir(outputdir)
+                end
+
+                -- generate headerunit
+                local args = {headernameflag .. headerunit.type, headerunit.path, ifcoutputflag, outputdir, "/Fo" .. objectfile}
+                os.vrunv(compinst:program(), table.join(compinst:compflags({target = target}), common_args, args), {envs = vcvars})
+
+            end, {dependfile = target:dependfile(bmifile), files = {headerunit.path}})
+        end, {rootjob = opt.rootjob})
+        _add_module_to_mapper(headerunitflag .. headerunit.type, headerunit.name .. "=" .. path.relative(bmifile, cachedir))
+        _add_objectfile_to_link_arguments(objectfile)
+    end
+    _flush_mapflags_to_mapper()
+end
+
+-- generate target user header units for batchcmds
+function generate_user_headerunits_for_batchcmds(target, batchcmds, headerunits, opt)
+    local compinst = target:compiler("cxx")
+    local toolchain = target:toolchain("msvc")
+    local vcvars = toolchain:config("vcvars")
     local cachedir = common.modules_cachedir(target)
 
+    -- get flags
+    local exportheaderflag = get_exportheaderflag(target)
+    local headerunitflag = get_headerunitflag(target)
+    local headernameflag = get_headernameflag(target)
+    local ifcoutputflag = get_ifcoutputflag(target)
+    assert(headerunitflag and headernameflag and exportheaderflag, "compiler(msvc): does not support c++ header units!")
+
     -- build headerunits
     local common_args = {"-TP", exportheaderflag, "-c"}
     local projectdir = os.projectdir()
@@ -234,15 +326,14 @@ function generate_user_headerunits_for_batchcmds(target, batchcmds, headerunits,
         depmtime = math.max(depmtime, os.mtime(bmifile))
     end
     batchcmds:set_depmtime(depmtime)
+    _flush_mapflags_to_mapper()
 end
 
--- build module files for batchcmds
-function build_modules_for_batchcmds(target, batchcmds, objectfiles, modules, opt)
+-- build module files for batchjobs
+function build_modules_for_batchjobs(target, batchjobs, objectfiles, modules, opt)
     local compinst = target:compiler("cxx")
     local toolchain = target:toolchain("msvc")
     local vcvars = toolchain:config("vcvars")
-
-    -- get cachedirs
     local cachedir = common.modules_cachedir(target)
 
     -- get flags
@@ -250,13 +341,9 @@ function build_modules_for_batchcmds(target, batchcmds, objectfiles, modules, op
     local interfaceflag = get_interfaceflag(target)
     local referenceflag = get_referenceflag(target)
 
-    -- append module mapper flags
-    local cache = common.localcache():get("mapflags") or {}
-    target:add("cxxflags", cache, {force = true})
-
     -- build modules
     local common_args = {"-TP"}
-    local depmtime = 0
+    local provided_modules = {}
     for _, objectfile in ipairs(objectfiles) do
         local m = modules[objectfile]
         if m and m.provides then
@@ -273,20 +360,106 @@ function build_modules_for_batchcmds(target, batchcmds, objectfiles, modules, op
             end
 
             local bmifile = provide.bmi
-            local args = {"-c", "-Fo" .. objectfile, interfaceflag, ifcoutputflag, bmifile, provide.sourcefile}
+            local moduleinfo = table.copy(provide)
+            moduleinfo.job = batchjobs:newjob(provide.sourcefile, function (index, total)
+                depend.on_changed(function()
+                    progress.show((index * 100) / total, "${color.build.object}generating.cxx.module.bmi %s", name)
+                    local objectdir = path.directory(objectfile)
+                    if not os.isdir(objectdir) then
+                        os.mkdir(objectdir)
+                    end
+                    -- append module mapper flags first
+                    -- @note we add it at the end to ensure that the full mapflags are already stored in the mapper
+                    if not target:data("cxx.add_modules_mapflags") then
+                        local mapflags = _get_mapflags_from_mapper()
+                        if mapflags then
+                            target:add("cxxflags", mapflags, {force = true})
+                        end
+                        target:data_set("cxx.add_modules_mapflags", true)
+                    end
+                    local args = {"-c", "-Fo" .. objectfile, interfaceflag, ifcoutputflag, bmifile, provide.sourcefile}
+                    os.vrunv(compinst:program(), table.join(compinst:compflags({target = target}), common_args, args), {envs = vcvars})
+                end, {dependfile = target:dependfile(bmifile), files = {provide.sourcefile}})
+            end)
+            if m.requires then
+                moduleinfo.deps = table.keys(m.requires)
+            end
+            moduleinfo.name = name
+            provided_modules[name] = moduleinfo
+            _add_module_to_mapper(referenceflag, name .. "=" .. path.filename(bmifile))
+            target:add("objectfiles", objectfile)
+        end
+    end
+    _flush_mapflags_to_mapper()
+
+
+    -- build batchjobs for modules
+    common.build_batchjobs_for_modules(provided_modules, batchjobs, opt.rootjob)
+end
+
+-- build module files for batchcmds
+function build_modules_for_batchcmds(target, batchcmds, objectfiles, modules, opt)
+    local compinst = target:compiler("cxx")
+    local toolchain = target:toolchain("msvc")
+    local vcvars = toolchain:config("vcvars")
+    local cachedir = common.modules_cachedir(target)
+
+    -- get flags
+    local ifcoutputflag = get_ifcoutputflag(target)
+    local interfaceflag = get_interfaceflag(target)
+    local referenceflag = get_referenceflag(target)
+
+    -- we need update mapper first
+    for _, objectfile in ipairs(objectfiles) do
+        local m = modules[objectfile]
+        if m and m.provides then
+            -- assume there that provides is only one, until we encounter the case
+            local length = 0
+            local name, provide
+            for k, v in pairs(m.provides) do
+                length = length + 1
+                name = k
+                provide = v
+                if length > 1 then
+                    raise("multiple provides are not supported now!")
+                end
+            end
+
+            local bmifile = provide.bmi
+            _add_module_to_mapper(referenceflag, name .. "=" .. path.filename(bmifile))
+            target:add("objectfiles", objectfile)
+        end
+    end
+    _flush_mapflags_to_mapper()
+
+    -- append module mapper flags
+    local mapflags = _get_mapflags_from_mapper()
+    if mapflags then
+        target:add("cxxflags", mapflags, {force = true})
+    end
+
+    -- build modules
+    local common_args = {"-TP"}
+    local depmtime = 0
+    for _, objectfile in ipairs(objectfiles) do
+        local m = modules[objectfile]
+        if m and m.provides then
+            local name, provide
+            for k, v in pairs(m.provides) do
+                name = k
+                provide = v
+                break
+            end
 
+            local bmifile = provide.bmi
+            local args = {"-c", "-Fo" .. objectfile, interfaceflag, ifcoutputflag, bmifile, provide.sourcefile}
             batchcmds:show_progress(opt.progress, "${color.build.object}generating.cxx.module.bmi %s", name)
             batchcmds:mkdir(path.directory(objectfile))
             batchcmds:vrunv(compinst:program(), table.join(compinst:compflags({target = target}), common_args, args), {envs = vcvars})
             batchcmds:add_depfiles(provide.sourcefile)
-
-            _add_module_to_mapper(referenceflag, name .. "=" .. path.filename(bmifile))
-
-            target:add("objectfiles", objectfile)
             depmtime = math.max(depmtime, os.mtime(bmifile))
         end
     end
-
     batchcmds:set_depmtime(depmtime)
 end
 
@@ -430,4 +603,4 @@ function get_scandependenciesflag(target)
         _g.scandependenciesflag = scandependenciesflag or false
     end
     return scandependenciesflag or nil
-end
\ No newline at end of file
+end
diff --git a/xmake/rules/c++/modules/xmake.lua b/xmake/rules/c++/modules/xmake.lua
index 4117004cc..23f9a0f77 100644
--- a/xmake/rules/c++/modules/xmake.lua
+++ b/xmake/rules/c++/modules/xmake.lua
@@ -53,55 +53,70 @@ rule("c++.build.modules.builder")
     set_sourcekinds("cxx")
     set_extensions(".mpp", ".mxx", ".cppm", ".ixx")
 
-    -- TODO parallel build support to accelerate `xmake build` to build modules
-    --[[
+    -- generate headerunits
+    -- parallel build support to accelerate `xmake build` to build headerunits
+    before_build(function(target, batchjobs, opt)
+        local job
+        if target:data("cxx.has_modules") then
+            import("modules_support.common")
+            local sourcebatch = target:sourcebatches()["c++.build.modules.builder"]
+            common.patch_sourcebatch(target, sourcebatch, opt)
+            local modules = common.get_module_dependencies(target, sourcebatch, opt)
+            batchjobs:group_enter(target:name() .. "/generate_headerunits")
+            common.generate_headerunits_for_batchjobs(target, batchjobs, sourcebatch, modules, opt)
+            job = batchjobs:group_leave()
+        end
+        return job or opt.rootjob
+    end, {batch = true})
+
+    -- build modules
+    -- parallel build support to accelerate `xmake build` to build modules
     before_build_files(function(target, batchjobs, sourcebatch, opt)
-        if not target:data("cxx.has_modules") then
+        if target:data("cxx.has_modules") then
+            import("modules_support.common")
+            common.patch_sourcebatch(target, sourcebatch, opt)
+            local modules = common.get_module_dependencies(target, sourcebatch, opt)
+            common.build_modules_for_batchjobs(target, batchjobs, sourcebatch, modules, opt)
+        else
+            -- avoid duplicate linking of object files of non-module programs
             sourcebatch.objectfiles = {}
-            return
         end
+    end, {batch = true})
 
-        -- patch sourcebatch
-        import("modules_support.common")
-        common.patch_sourcebatch(target, sourcebatch, opt)
-
-        -- generate dependencies
-        local modules = common.generate_dependencies(target, sourcebatch, opt)
+    -- serial compilation only, usually used to support project generator
+    before_buildcmd_files(function(target, batchcmds, sourcebatch, opt)
+        if target:data("cxx.has_modules") then
+            import("modules_support.common")
 
-        -- generate headerunits
-        local headerunits_flags = common.generate_headerunits_for_batchjobs(target, batchjobs, sourcebatch, modules, opt)
-        if headerunits_flags then
-            target:add("cxxflags", headerunits_flags, {force = true, expand = false})
-        end
+            -- patch sourcebatch
+            common.patch_sourcebatch(target, sourcebatch, opt)
 
-        -- build modules
-        common.build_modules_for_batchjobs(target, batchjobs, sourcebatch, modules, opt)
-    end, {batch = true})]]
+            -- generate headerunits
+            local modules = common.get_module_dependencies(target, sourcebatch, opt)
+            common.generate_headerunits_for_batchcmds(target, batchcmds, sourcebatch, modules, opt)
 
-    -- serial compilation only, usually used to support project generator
-    before_buildcmd_files(function(target, batchcmds, sourcebatch, opt)
-        if not target:data("cxx.has_modules") then
+            -- build modules
+            common.build_modules_for_batchcmds(target, batchcmds, sourcebatch, modules, opt)
+        else
+            -- avoid duplicate linking of object files of non-module programs
             sourcebatch.objectfiles = {}
-            return
         end
+    end)
 
-        -- patch sourcebatch
+    before_link(function (target)
         import("modules_support.common")
-        common.patch_sourcebatch(target, sourcebatch, opt)
-
-        -- generate dependencies
-        local modules = common.generate_dependencies(target, sourcebatch, opt)
-
-        -- generate headerunits
-        common.generate_headerunits_for_batchcmds(target, batchcmds, sourcebatch, modules, opt)
-
-        -- build modules
-        common.build_modules_for_batchcmds(target, batchcmds, sourcebatch, modules, opt)
+        common.append_headerunits_objectfiles(target)
     end)
 
-    before_link(function(target)
+    after_clean(function (target)
+        import("core.base.option")
         import("modules_support.common")
-        common.append_headerunits_objectfiles(target)
+        os.tryrm(common.modules_cachedir(target))
+        if option.get("all") then
+            os.tryrm(common.stlmodules_cachedir(target))
+            common.localcache():clear()
+            common.localcache():save()
+        end
     end)
 
 -- install modules