SOL > Releases > v0.8 Gacrux

v0.8 Gacrux

Version/Date Changes

Version/Date	Changes
17.04.2026 v0.8.0 Docs	This is a feature update. It is possible that performance decreased compared to v0.7 release. We are focusing on performance improvements in upcoming releases. Highlights All `sol.config[...]` values can now also be set via env vars. The env name always starts with `SOL_CONFIG_` followed by the config key, in upper case and `::` replaced with `_`, e.g., `sol.config["dfp::debug"]=True` becomes `SOL_CONFIG_DFP_DEBUG=TRUE`. For boolean values `TRUE`, `ON` or `1` can be used. Switched to intrinsics-like code generation for compute kernels. Added support for variable pixel dimensions of conv and pooling layers. Breaking Changes The default location for the SOL cache (`.sol` folder) is now at `$HOME/.cache/sol`. Please use the env var `SOL_CWD=/path/to/cache` to move the SOL cache if required. `python3 -m sol fix-omp` has been deprecated. Use `nec-sol fix-omp` instead! `python3 -m sol-fix` will be removed in v0.9! We decided to make no variable dimensions enabled by default for performance reasons. Use `vdims=[...]` to manually enable variable dimensions. Changing backend heuristics via `sol.config[...]` has been removed. Closed Issues #1943 [PyTorch] Test v2.11.0 #1938 [TF] Test v2.21.0 #1934 [CUBLAS] Set RPATH for libcublas.so in CUBLAS-handle #1933 [LicenseServer] Change .postVERSION to .postEXPIRATIONDATE #1931 [DFP] Improve Constant Setting #1930 [DFP] Improve mask sorting #1928 [HLIR] Performance problem when using BatchNormConvFuse, if the weights are significantly larger than the input data #1927 [DFP] atomicAdd(0.0) does not need to be executed! #1926 [DFP] Wrong gradients for complex mul and div #1925 [DFP] VE Intrinsics performance regression #1923 [DFP] Inline constant computations #1921 [HLIR] Add topk #1920 [DFP] Add code path for CUDA atomics for float16/32 dtypes #1918 [CUDNNv9] Returns CUDNN_STATUS_NOT_INITIALIZED with float16/bfloat16 but works with float32 #1916 [DFP] Lookup boundary checks #1914 [Keras] Keras.predict does not throw exception when wrong input dtype is detected #1913 [License] Add .postX for every time a new license is issued, to prevent caching effects of expired license files. #1912 [DFP] move autoSqueeze from lowerIR to Planner #1911 [DFP] Unnecessary select for select(not(X), scalar(1), scalar(0)) #1910 [DFP] Minimize register usage #1909 [SQLite] Improve stability when SOL cache is on a network drive #1906 [DFP] Investigate why PyTorch is faster on PY_Issue_1740 #1901 [DFP] AllReduce performance optimization #1899 [Cache] Add XDG_CACHE_HOME to determine ~/.cache #1896 [PyTorch] Test v2.9.1 #1895 [Core] Consider to move .sol to $HOME/.cache/sol as default #1890 [Installer] Add support for credential files #1867 [HLIR] ViewDType breaks sol_tensor_ptr_XXX(...) dtype check! #1854 [PyTorch] ANEMOI parsing: torch.compile reports different DTypes than SOL #1851 [Debug] Estimated Peak Memory differs from Memory Dump #1845 [Installer] Keyring #1834 [CUDNNv8] Remove SymPad and use explicit padding if leftPadding != rightPadding #1831 [Installer] add --fix-omp option #1819 [VE] VEDA_ERROR_CANT_FREE_NON_DELAYED_VPTR in TF_Dropout using VEO mode #1812 [HLIR] Enable Conv with vdims in pixel dimensions #1797 [JIT] Use SQL lock to mutex compilation of handle/modules in parallel processes #1795 [VE] VEDA_ERROR_UNKNOWN_VEDA_ARCHITECTURE thrown in initVEDA if no VE and no VEOS is present in system #1782 [DFP] Split nested loops that underutilize the vectors #1781 [Python] SOL might use wrong Python #1780 [Installer] Project-URL is wrong, should be v0.X not v0.X.Y #1778 [HLIR] Remove returnCarry from PrefixSum, instead handle like in MaxPool #1777 [Utils] cleanup file paths #1774 [DNN] Rework Handlers #1773 [DNNL] Report DNNL Workspace Size #1771 [YAAL] Workspaces #1770 [CUDNN] Fix CUDA Graphs when using CUDNN #1769 [TF] sol.optimize example inputs support #1767 [Python] "There is a newer version of SOL available" detects post installers as NEW SOL version #1764 [DFP] Deprecate LoopStack::sort #1763 [DFP] Improve implementation of CoresSIMD -> SIMD #1757 [DNN] Can we remove sol_ctx from the constructor of DeviceMap? #1734 [OMP] Remove Cost Model #1710 [Macros] Unify SOL_HAS_FUNC and SOL_HAS_FUNC2 #1662 [VLLM] Remove support #1615 [HLIR] Implement Interpolate as Gather? #1438 [DFP] AllReduce #919 [HLIR] Merge two Reorder's with different shape sizes #472 [X86, VE, NVIDIA] Reverse PrefixSum

17.04.2026
v0.8.0
Docs

This is a feature update. It is possible that performance decreased compared to v0.7 release. We are focusing on performance improvements in upcoming releases.

Highlights

All sol.config[...] values can now also be set via env vars. The env name always starts with SOL_CONFIG_ followed by the config key, in upper case and :: replaced with _, e.g., sol.config["dfp::debug"]=True becomes SOL_CONFIG_DFP_DEBUG=TRUE. For boolean values TRUE, ON or 1 can be used.
Switched to intrinsics-like code generation for compute kernels.
Added support for variable pixel dimensions of conv and pooling layers.

Breaking Changes

The default location for the SOL cache (.sol folder) is now at $HOME/.cache/sol. Please use the env var SOL_CWD=/path/to/cache to move the SOL cache if required.
python3 -m sol fix-omp has been deprecated. Use nec-sol fix-omp instead! python3 -m sol-fix will be removed in v0.9!
We decided to make no variable dimensions enabled by default for performance reasons. Use vdims=[...] to manually enable variable dimensions.
Changing backend heuristics via sol.config[...] has been removed.

Closed Issues

#1943 [PyTorch] Test v2.11.0
#1938 [TF] Test v2.21.0
#1934 [CUBLAS] Set RPATH for libcublas.so in CUBLAS-handle
#1933 [LicenseServer] Change .postVERSION to .postEXPIRATIONDATE
#1931 [DFP] Improve Constant Setting
#1930 [DFP] Improve mask sorting
#1928 [HLIR] Performance problem when using BatchNormConvFuse, if the weights are significantly larger than the input data
#1927 [DFP] atomicAdd(0.0) does not need to be executed!
#1926 [DFP] Wrong gradients for complex mul and div
#1925 [DFP] VE Intrinsics performance regression
#1923 [DFP] Inline constant computations
#1921 [HLIR] Add topk
#1920 [DFP] Add code path for CUDA atomics for float16/32 dtypes
#1918 [CUDNNv9] Returns CUDNN_STATUS_NOT_INITIALIZED with float16/bfloat16 but works with float32
#1916 [DFP] Lookup boundary checks
#1914 [Keras] Keras.predict does not throw exception when wrong input dtype is detected
#1913 [License] Add .postX for every time a new license is issued, to prevent caching effects of expired license files.
#1912 [DFP] move autoSqueeze from lowerIR to Planner
#1911 [DFP] Unnecessary select for select(not(X), scalar(1), scalar(0))
#1910 [DFP] Minimize register usage
#1909 [SQLite] Improve stability when SOL cache is on a network drive
#1906 [DFP] Investigate why PyTorch is faster on PY_Issue_1740
#1901 [DFP] AllReduce performance optimization
#1899 [Cache] Add XDG_CACHE_HOME to determine ~/.cache
#1896 [PyTorch] Test v2.9.1
#1895 [Core] Consider to move .sol to $HOME/.cache/sol as default
#1890 [Installer] Add support for credential files
#1867 [HLIR] ViewDType breaks sol_tensor_ptr_XXX(...) dtype check!
#1854 [PyTorch] ANEMOI parsing: torch.compile reports different DTypes than SOL
#1851 [Debug] Estimated Peak Memory differs from Memory Dump
#1845 [Installer] Keyring
#1834 [CUDNNv8] Remove SymPad and use explicit padding if leftPadding != rightPadding
#1831 [Installer] add --fix-omp option
#1819 [VE] VEDA_ERROR_CANT_FREE_NON_DELAYED_VPTR in TF_Dropout using VEO mode
#1812 [HLIR] Enable Conv with vdims in pixel dimensions
#1797 [JIT] Use SQL lock to mutex compilation of handle/modules in parallel processes
#1795 [VE] VEDA_ERROR_UNKNOWN_VEDA_ARCHITECTURE thrown in initVEDA if no VE and no VEOS is present in system
#1782 [DFP] Split nested loops that underutilize the vectors
#1781 [Python] SOL might use wrong Python
#1780 [Installer] Project-URL is wrong, should be v0.X not v0.X.Y
#1778 [HLIR] Remove returnCarry from PrefixSum, instead handle like in MaxPool
#1777 [Utils] cleanup file paths
#1774 [DNN] Rework Handlers
#1773 [DNNL] Report DNNL Workspace Size
#1771 [YAAL] Workspaces
#1770 [CUDNN] Fix CUDA Graphs when using CUDNN
#1769 [TF] sol.optimize example inputs support
#1767 [Python] "There is a newer version of SOL available" detects post installers as NEW SOL version
#1764 [DFP] Deprecate LoopStack::sort
#1763 [DFP] Improve implementation of CoresSIMD -> SIMD
#1757 [DNN] Can we remove sol_ctx from the constructor of DeviceMap?
#1734 [OMP] Remove Cost Model
#1710 [Macros] Unify SOL_HAS_FUNC and SOL_HAS_FUNC2
#1662 [VLLM] Remove support
#1615 [HLIR] Implement Interpolate as Gather?
#1438 [DFP] AllReduce
#919 [HLIR] Merge two Reorder's with different shape sizes
#472 [X86, VE, NVIDIA] Reverse PrefixSum