Christophe Murphy

When AF_WITH_FAST_MATH is disabled, the pow function definition incorrectly casts float16 arguments to integers. Description ----------- This error occurs when AF_WITH_FAST_MATH is disabled however when it is enabled the arguments...

Incorrect results are produced when using the pow function with float16 arguments with the CUDA backend Description =========== The following definition for the pow function is used in the JIT...

Add ability to use atan2 and hypot functions with fp16 arguments. Description ----------- Cases have been added to the atan2 and hypot functions to handle fp16 arguments. Previously an exception...

Reverted an error in the interop_cuda example code where the cuda steam id was being used instead of the arrayfire stream id. Fixes: #3404 Checklist --------- - [x ] Rebased...

Rounding was being applied to x axis min and max values but this should not be done for a histogram where the values are in fact bin labels. xmin and...

Fixes for bugs uncovered by CUDA version 12.6 Description ----------- * Merge in fixes from #3576 for shuffle sync bugs * Fixed bugs in calls to ormqr routine in the...

A number of tests have started failing since CUDA 12.2 Description =========== The following tests fail with CUDA 12.6: [ FAILED ] ConvolveStrided/0.Gradient_sig81032_filt3334_s11_p11_d11, where TypeParam = float (10 ms) [...

Fix for bug where warp primitives were called by threads outside mask, triggering a warp illegal instruction exception in CUDA versions > 12.2 Description ----------- A bug has been uncovered...

Bug in call to __shfl_down_sync in reduce_blocks_by_key results in Warp Illegal Instruction Error in CUDA > 12.2 Description =========== - Built code from main - Affects CUDA backend when using...

Modifications and bug fixes to get the OneAPI backend working with Intel OneAPI version 2024.1 This has only been tested on Linux and not Windows. Description ----------- Several changes needed...