Andreas Beckmann

Results 49 comments of Andreas Beckmann

vector parameter passing via the stack seems to be broken: code: ``` printf("%#v2hx\n",(short2)(0x1234,0x8765)); printf("%#v2hx %#v2hx\n",(short2)(0x1234,0x8765),(short2)(0x6996,0x2552)); ``` output: ``` 0x1234,0 0x1234,0 0x8765,0 ``` disassembly: ``` 3751: 8d 83 bc e2 ff...

I could also reproduce it with pocl 1.1 - the tests just don't cover this case. I'm not convinced that this is an llvm bug. That va_arg(ap, short2) stuff is,...

From the specification quoted in https://bugs.llvm.org/show_bug.cgi?id=40491#c9 I'd conclude that the pocl printf is correctly assuming that the argument is not promoted, but the code generation (llvm?) is wrong by promoting...

There is also the other way around, i.e. a certain configuration must *not* be enabled. E.g. the zfs and nvidia modules cannot be built on PREEMPT_RT kernels (due to transitive...

A preliminary patch that adds a (yet undocumented) `BUILD_EXCLUSIVE_CONFIG` option can be found here: https://gist.github.com/anbe42/66aacc5a4ee1c91089cf11f02d677e3f

The test in question is https://github.com/pocl/pocl/blob/master/tests/regression/test_structs_as_args.cpp The [kernel](https://github.com/pocl/pocl/blob/master/tests/regression/test_structs_as_args.cpp#L101) gets a struct as argument: `kernel void test_kernel(test_struct input, global int* output)`, the struct is declared in C at [line 55](https://github.com/pocl/pocl/blob/master/tests/regression/test_structs_as_args.cpp#L55) and...

What about a patch like https://gist.github.com/anbe42/3e3c6fadf25b0619abaa8206b46efc87 ? It only adds back the `__attribute__((aligned(8)))` (and therefore the 'ignoring attributes on template argument' warning in C++) on the types and platforms where...

Enhanced version of my test program, supporting both C and C++: ``` #define CL_TARGET_OPENCL_VERSION 220 #include #include #ifdef __cplusplus #include typedef std::tuple tuple_cl_int_long; typedef std::tuple tuple_cl_int_int2; #define myalignof alignof #else...

The Debian pycuda package started failing to build when python 3.11 was added as a supported version a few days ago. We use the system boost (which should be 1.74)...

@wfpokorny : > I know a naive question, but could you provide a direct link (or method ftp?) to your updated povray_3.7.0.8-2.debian.tar.xz ? A quick search via google just now...