This could lead to stack overflow in B_storage.
Also do not require specific type for query_norm_scale,
update batch sizes for attention tensors,
more verbose Mat shape/type checks.
PiperOrigin-RevId: 824987689
Group M=4..7 into same config. Add configs for power of two sizes.
Allow odd mc to enable a single range for odd M.
io.cc: warning fix(cast).
IsBlock -> !IsOneMC
benchmark_helper: best for verbosity 3, all configs for 4
ops_test: remove unused includes
PiperOrigin-RevId: 824475104
matmul-inl: support CView=StridedView or RowPtrs; rename to C_MC_NC
matmul.cc: Allow 1 more rep for MC/NC to allow half-sized tiles, which helps.
PiperOrigin-RevId: 807291701
Lift DecompressA out of main autotuner to prevent interference
Also use kMaxNR / kNR constants instead of extra args
Fix: only require vector alignment, not cache alignment
PiperOrigin-RevId: 804333769
remove MatMul f32 special case (smaller code),
types: Add u32/u64 for use by Activations
move renamed ParallelismStrategy to threading_context so can pass ctx
ensure worker index is unique across clusters
matmul.h: const member functions for renamed policy classes (easier to call)
PiperOrigin-RevId: 802848086
Add a special case for A=F32,B=BF16, used when there is no native bf16 dot product.
dot-inl: ensure bf16,f32 and f32,bf16 both get promoted to float before f64 summation
matmul.cc: update autotuning to reflect actual A size
matmul_test: add all combinations of bf16/f32, report all results, not just first difference, check non-vector-aligned K
PiperOrigin-RevId: 800487817
Optimizations
- Better load-balancing in attention threading
(Previously, clusters were limited by #heads)
- Add MulByConstTo to avoid zero-init
- Parallel activations
Cleanup
- Prepare for RowPtr in A or B
- Pass through thread_id to ops
- Avoid warning in bench_matmul
PiperOrigin-RevId: 773723423
Detect PaliGemma models from layer names
Remove unused allocator arg from CreateInvTimescale
matmul: only warn once about dim divisibility
Print config also in tests if --verbosity 2
PiperOrigin-RevId: 766605131
Only the weights; binding MatMul output worsens batch=1 prefill.
Update gemma_batch_bench to use --decode_qbatch.
Fix/remove prefill_activations in gemma-inl.h.
Refactor:
use BasePageBytes directly when binding
Move BindB/C to .cc by de-templatizing
Remove MatOwners::AllocateFor because it is weights-specific (binding or not)
Disband MatOwners, replace with vector
PiperOrigin-RevId: 759610477
use new ThreadingContext2 instead of monostate/init in each frontend
Add ThreadingArgs(replaces AppArgs)
backprop: use Packed() accessor and MakePacked factory and row-based access to allow for stride
compress_weights: remove, moving to py-only exporter instead
Move MatPtr to mat.h and revise interface:
- Generic MatOwner
- rename accessors to Packed*
- support stride/row accessors, fix RowPtr stride
Add TypeBits(Type)
Move GenerateMat to test_util-inl for sharing between matmul test/bench
Move internal init to gemma.cc to avoid duplication
Rename GemmaEnv model_ to gemma_ for disambiguating vs upcoming ModelStorage
Remove --compressed_weights, use --weights instead.
tensor_index: add ExtentsFromInfo and TensorIndexLLM/Img
Allocator: use normal unique_ptr for AllocBytes so users can call directly
threading: use -> because AlignedPtr no longer assumes arrays
PiperOrigin-RevId: 745918637
Gemma receives a MatMulEnv arg, with comment on lifetime
Split threading into topology so the latter can be used in allocator
Add AllocClasses() for non-POD (ThreadPool)
Support binding pool to NUMA node
Update threading_test with latency measurements
Also update Highway version.
PiperOrigin-RevId: 736904748
Adds Stride to ConstMat, to support decompression of C output for test
matmul_test: add line numbers to output
Also ignore "N is not a multiple of nc" when N==nc
PiperOrigin-RevId: 731096662
Jan Wassenberg
Marie White