Not sure it helps with warmup, but marking a foreign function as critical can improve performance: https://docs.oracle.com/en/java/javase/22/docs/api/java.base/java/lang/foreign/Linker.Option.html#critical(boolean) .
On Mon, Apr 22, 2024 at 10:02 PM Philip Race <philip.r...@oracle.com> wrote: > No, it wasn't. I didn't even use jextracted code. > The startup cost is around initialisation of FFM - around 70 ms (IIRC) > overhead on my MacBook > Then creation of VarHandles and MethodHandles - 2-5 ms each is what I > measured, so do these lazily if you can. > And warmup cost is that it takes about 10000 iterations to get code fully > compiled. > > java -XX:+PrintFlagsFinal -version 2>&1 | grep CompileThreshold > intx CompileThreshold = > 10000 {pd product} {default} > double CompileThresholdScaling = > 1.000000 {product} {default} > uintx IncreaseFirstTierCompileThresholdAt = > 50 {product} {default} > intx Tier2CompileThreshold = > 0 {product} {default} > intx Tier3CompileThreshold = > 2000 {product} {default} > intx Tier4CompileThreshold = > 15000 {product} {default} > > -phil. > > > On 4/22/24 11:45 AM, Thiago Milczarek Sayão wrote: > > I think the startup time might be related to all static symbol lookups. > So I'm manually including just what is needed: > > jextract --output src -t com.sun.glass.wayland.extracted \ > --header-class-name GlassWayland \ > `pkg-config --libs glib-2.0 gio-2.0 libportal wayland-client` \ > `pkg-config --cflags-only-I glib-2.0 gio-2.0 libportal wayland-client` \ > glass-wayland.h \ > --include-function xdp_portal_initable_new \ > --include-function xdp_session_close \ > --include-function xdp_portal_open_file \ > --include-function xdp_portal_open_file_finish \ > --include-function g_object_unref \ > --include-function g_timeout_add \ > --include-function g_add_idle \ > --include-function g_main_loop_run \ > --include-function g_main_loop_new \ > --include-function g_main_loop_ref \ > --include-function g_main_loop_unref \ > --include-function g_main_loop_quit \ > --include-function g_settings_new \ > --include-function g_settings_get_int \ > --include-function wl_display_connect \ > --include-function wl_display_disconnect \ > --include-function wl_display_roundtrip \ > --include-function wl_display_dispatch_pending \ > --include-typedef GAsyncReadyCallback \ > --include-typedef GSourceFunc \ > --include-typedef GError > > > Em seg., 22 de abr. de 2024 às 13:24, Philip Race <philip.r...@oracle.com> > escreveu: > >> As a reminder, using FFM will require all FX *applications* to specify >> --enable-native-access on the command line >> Although this is likely coming to JNI soon too. >> >> https://docs.oracle.com/en/java/javase/21/core/restricted-methods.html >> >> But one thing to watch out for with FFM is startup + warm up time. >> I struggled a lot with that in using FFM for just one library in the >> java.desktop module. >> >> -phil >> >> On 4/22/24 9:12 AM, Nir Lisker wrote: >> >> Sorry, we bumped to Java 21 in JavaFX 22 I think since we preserve the >> N-1 rule. >> >> On Mon, Apr 22, 2024 at 6:03 PM Nir Lisker <nlis...@gmail.com> wrote: >> >>> I think that we'll be able to bump to Java 25 in JavaFX 25, like we did >>> with 21. I suggested initially to bump to Java 22 exactly for FFM as it's >>> very useful for JavaFX, but was told we shouldn't since it's not an LTS >>> version. >>> >>> I have no idea how long the work on Wayland will take including the code >>> review (a rather long process), but you should be able to request code >>> reviews with FFM and have it ready for integration by Java 25. >>> >>> On Mon, Apr 22, 2024 at 5:49 PM Thiago Milczarek Sayão < >>> thiago.sa...@gmail.com> wrote: >>> >>>> I was just experimenting, but it seems to be less work than going with >>>> JNI. >>>> If I am correct, the next Java LTS will be 25, which will be required >>>> on JavaFX 29 to be released on September/29. >>>> >>>> It's 7 years - that's really too much. >>>> >>>> Maybe it's still worthwhile to prototype using FFM and then port >>>> everything to JNI. >>>> >>>> -- Thiago. >>>> >>>> >>>> Em seg., 22 de abr. de 2024 às 11:21, Kevin Rushforth < >>>> kevin.rushfo...@oracle.com> escreveu: >>>> >>>>> Note also that we cannot use Panama in the JavaFX internals yet, since >>>>> the minimum version of the JDK is 21. >>>>> >>>>> -- Kevin >>>>> >>>>> >>>>> On 4/21/2024 10:51 AM, Thiago Milczarek Sayão wrote: >>>>> > Hi, >>>>> > >>>>> > I did a small test app to explore Wayland client and portals (for >>>>> > Robot and dialogs such as file open/save). >>>>> > >>>>> > https://github.com/tsayao/wayland-test/blob/main/wayland-test.c >>>>> > >>>>> > It seems it will work as a glass backend, but some walls will be hit >>>>> > on the way :) >>>>> > >>>>> > I have tried to use jextract (from project Panama) to work directly >>>>> > with java, but it seems it does not support wl_ types. >>>>> > >>>>> > -- Thiago. >>>>> >>>>> >> >