[Bounty $20] Codegen fails on singular boolean values

[freemo]

when a kernel has a variable that is a singular boolean that is used it fails in GPU mode. Works fine for arrays of boolean and other singular variables such as int. This issue is exposed by the following unit test:

https://github.com/Syncleus/aparapi/blob/master/src/test/java/com/aparapi/runtime/UseBooleanTest.java

[KingParmenides]

I traced this to the native JNI argument binding path: instance boolean primitives use the byte signature/accessor, while static boolean primitives already use the boolean signature/accessor.

PR: Syncleus/aparapi-native#14
Validation: git diff --check -- src/cpp/runKernel/KernelArg.cpp; focused Java codegen tests in Syncleus/aparapi for boolean arrays/local booleans passed.

BTC: 39Q34P8A7g375yqEr8buvNJkUbRgKfbKQZ

[minyanyi]

I traced this to the native boolean field lookup and opened a fix here:

Syncleus/aparapi-native#15

The existing non-static primitive boolean path looks up the Java kernel field with descriptor B and reads it as a byte, while the actual Java field descriptor is Z. That reproduces this issue as NoSuchFieldError: ...BooleanKernel.isInverse B when running UseBooleanTest in GPU mode. The PR changes that path to use Z + GetBooleanField, matching the already-correct static boolean path.

Verification performed locally:

• Reproduced the failure against current aparapi-jni 1.4.3 native binary.
• Confirmed the generated OpenCL already passes the boolean as a byte-sized argument.
• git show --check passes on the native patch.

Payout address if accepted: 0x2E4380d2e1668Ca9fA3Ef91fF776FDc140Cf3fE8

[minyanyi]

Verification update: I built the native DLL from Syncleus/aparapi-native#15 locally on Windows x64 and reran the original UseBooleanTest$BooleanKernel repro. The published 1.4.3 native DLL still throws NoSuchFieldError ... isInverse B; the DLL built from the PR completes kernel.execute(1) without that error (mode=AUTO). Added the detailed verification notes on the PR.

[ericyewang]

I traced this to the native primitive argument path rather than the Java codegen itself. Non-static primitive boolean fields are looked up with the byte field signature (B) and GetByteField, while static booleans already use the correct boolean signature (Z).

I opened a root-cause fix in aparapi-native: Syncleus/aparapi-native#16

[gshaowei6]

I traced this to the native argument reader rather than Java codegen: primitive instance booleans are looked up with JNI signature B, so the GPU path throws NoSuchFieldError: ... isInverse B for the existing UseBooleanTest shape.

I opened the native-side fix here: Syncleus/aparapi-native#17