As of now, especially in the benchmark mode, hashcat will not go to create and destroy context and command-queue for each enabled device each time it switches from one hash-mode to the next.
Specifically using OpenCL with an NVIDIA device, it was not possible to complete the benchmark because clCreateContext has memory leaks that slowly consume all available GPU memory until hashcat can activate a new context and disable the device.
Avoid deprecated HIP functions
All hipCtx* features have been declared deprecated, so we have replaced them with the new ones, also fixing a critical bug on handling multiple AMD devices in the same system.
Fixed a typedef issue for clEnqueueReadBuffer().
Updated Python/hcshared.py with missing entry for new salt_dimy attribute in salt_t struct.
Fixed a bug in the autotuner when determining the starting value for kernel loops, in cases where the iteration count is N-1 and not a multiple of 1024.
Updated additional plugins to use OPTI_TYPE_REGISTER_LIMIT.
- Integrated occupancy hints from vendor APIs (CUDA, HIP) to set a
dynamic threads-per-block limit per kernel instead of using static
values.
- Added `find_tuning_function()` to identify the relevant kernel.
- Autotuner now runs in three stages: threads -> loops -> accel. The
first two stages now stop increasing when the tested kernel runtime
gets too close to the target runtime (96ms for `-w 3`), leaving
headroom for the next stage to adjust in a finer sense.
- Accel tuning now uses a capped floating-point multiplier instead of
powers of two.
- Removed workarounds for missing thread autotuning in plugins.
- Removed the hardcoded 4GiB host memory limit for accel. Added a
cross-platform `get_free_memory()` to check actual free RAM during GPU
initialization, preventing underutilization of high-end GPUs like the
4090. If needed, users can still cap memory usage with `-T` or `-n`.
- Updated enums for ROCm 6.4.x and CUDA 12.9.
- Added code to detect kernel register spilling. That's relevant so we
can keep free enough global memory on the runtime for the runtime to
handle spills efficiently.
Add hipDeviceProp_t and bindings for hipGetDeviceProperties(), hipGetDeviceProperties is required to retrieve gcnArchName[].
Add gcnArchName[] to select the correct --gpu-architecture value for a specific device when using hiprtc.
Include sm_major and sm_minor for CUDA and gcnArchName for HIP in the kernel filename hash.
Update nvrtc_options[] and hiprtc_options[] to avoid unused variables, eliminating the use of --restrict as a placeholder and preventing nvrtc from aborting.
Add check_file_suffix() and remove_file_suffix() helper functions.
Switched HIP version check from driverVersion to runtimeVersion
Fixed syntax check of HAS_VPERM macro in several kernel includes causing invalid error message for AMD GPUs on Windows
Updated AMD driver requirements
Updated docs/changes.txt with missing changes from previous commits
Fixed invalid vector data type in Murmur Hash in -a 3 mode
Fixed uninitialized variable warning in src/hashes.c
Fixed broken support for --generate-rules-func-min
