[Native Image] Bad performance of FFM API

[scf37]

Describe the Issue

I use FFM to interact with SDL in java game engine. Native version performs about 20 times slower than on the same JVM (400 -> 25 fps drop). I suspect inefficient implementation of native calls.

Using the latest version of GraalVM can resolve many issues.

• I tried with the latest version of GraalVM.

GraalVM Version

java 25 2025-09-16 LTS
Java(TM) SE Runtime Environment Oracle GraalVM 25+37.1 (build 25+37-LTS-jvmci-b01)
Java HotSpot(TM) 64-Bit Server VM Oracle GraalVM 25+37.1 (build 25+37-LTS-jvmci-b01, mixed mode, sharing)

Operating System and Version

Linux 5.15.0-139-generic #149~20.04.1-Ubuntu SMP Wed Apr 16 08:29:56 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux

Troubleshooting Confirmation

• I tried the suggestions in the troubleshooting guide.

Run Command

./app

Expected Behavior

runtime performance on par with JVM

Actual Behavior

20x slowdown

Steps to Reproduce

I'm using org.graalvm.buildtools.native Gradle plugin with default configuration and additional jvmArgs.add("--enable-preview")

Additional Context

No response

Run-Time Log Output and Error Messages

No response

[oubidar-Abderrahim]

Hi, thank you for reaching out, could you please share a small reproducer to illustrate this issue? thanks

[scf37]

Here are relevant pieces of code.

This code calls SDL3 in hot loop, renderer, frect1 and frect2 never change so native call receives the same pointers, textures can be different but they are preallocated as well. Therefore there is no load on native allocation.
This exactly snippet is timed using System.nanoTime() and I see 20x degradation.

        SDL_FRect.x(frect1, sx);
        SDL_FRect.y(frect1, sy);
        SDL_FRect.w(frect1, sw);
        SDL_FRect.h(frect1, sh);

        SDL_FRect.x(frect2, dx + x0);
        SDL_FRect.y(frect2, dy + y0);
        SDL_FRect.w(frect2, dw);
        SDL_FRect.h(frect2, dh);

        SDL_render_h.SDL_RenderTexture(renderer, b.getTexture(), frect1, frect2);

This is generated by jextract:

    public static boolean SDL_RenderTexture(MemorySegment renderer, MemorySegment texture, MemorySegment srcrect, MemorySegment dstrect) {
        var mh$ = SDL_RenderTexture.HANDLE;
        try {
            if (TRACE_DOWNCALLS) {
                traceDowncall("SDL_RenderTexture", renderer, texture, srcrect, dstrect);
            }
            return (boolean)mh$.invokeExact(renderer, texture, srcrect, dstrect);
        } catch (Throwable ex$) {
           throw new AssertionError("should not reach here", ex$);
        }
    }
    private static class SDL_RenderTexture {
        public static final FunctionDescriptor DESC = FunctionDescriptor.of(
            SDL_render_h.C_BOOL,
            SDL_render_h.C_POINTER,
            SDL_render_h.C_POINTER,
            SDL_render_h.C_POINTER,
            SDL_render_h.C_POINTER
        );

        public static final MemorySegment ADDR = SDL_render_h.findOrThrow("SDL_RenderTexture");

        public static final MethodHandle HANDLE = Linker.nativeLinker().downcallHandle(ADDR, DESC);
    }

    public static void x(MemorySegment struct, float fieldValue) {
        struct.set(x$LAYOUT, x$OFFSET, fieldValue);
    }
private static final OfFloat x$LAYOUT = (OfFloat)$LAYOUT.select(groupElement("x"));
private static final long x$OFFSET = $LAYOUT.byteOffset(groupElement("x"));
private static final GroupLayout $LAYOUT = MemoryLayout.structLayout(
        SDL_rect_h.C_FLOAT.withName("x"),
        SDL_rect_h.C_FLOAT.withName("y"),
        SDL_rect_h.C_FLOAT.withName("w"),
        SDL_rect_h.C_FLOAT.withName("h")
    ).withName("SDL_FRect");