* Boilerplate for q6_K repack
* q6_K repack to q6_Kx8 implementation
Signed-off-by: Alberto Cabrera <alberto.cabrera@liquid.ai>
* q6_K generic gemv and gemm
* wip, gemm_q6_K 8x8
* Still WIP: loading of q8s, q6h and q6l
* first working version of q6_K gemm
* Moved q6 loads outside of sb block, Unrolled inner loop
* Replaced modulo with mask
* First implementation of GEMV
* ggml_vdotq_s32 -> vdotq_s32
* Reduce width of accumulators in q6_K gemv
* Bsums instead of calc bias. Preload scales to use vget_lane. Unroll.
* Reuse scales in GEMM (same GEMV opt)
* Added todos for bsum and different qh repack
* Arch fallback
* VSLIQ for merging qh adn ql
* Removed TODO, already tested
* Apply suggestions
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Removed unused import
---------
Signed-off-by: Alberto Cabrera <alberto.cabrera@liquid.ai>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* [CUDA] Reduce CPU-side stalls due to the CUDA command buffer being full
With pipeline parallelism, during prompt processing, the CPU-side CUDA command buffer gets full, stalling the CPU. Due to this, enough work doesn't get submitted to the GPU, causing bubbles in the GPU timeline.
Fix this by setting the CUDA environment variable CUDA_SCALE_LAUNCH_QUEUES to 4x to increase the command buffer size.
* Set the env variable in the CUDA backend registry allocation
* Add link to PR in code comment
* Remove warning logs and update documentation
* opencl: flatten `q6_K` and add `kernel_mul_mv_q6_K_f32_flat`
* opencl: clean up
* opencl: refactor q6_K mv - put loop body in `block_q_6_K_dot_y_flat`
* opencl: tweak the workgroup size a bit
* opencl: output 4 values per subgroup for `kernel_mul_mv_q6_K_f32_flat`
* opencl: proper alignment for q6_K
* opencl: boundary handling for flattened q6_K mv
* opencl: rename q6_K mv kernel file
* opencl: put flattened q6_K mv in its own file
* opencl: use lower k in file name
* opencl: use K in variable names
* ggml-cpu: Use tiled FA for prompt-processing
the FA performance is gimped on CPU on long contexts because it essentially uses a vector kernel. This PR adds a tiled FA for PP. Perf tuning for tile sizes done on a AMD EPYC single-socket 64-c machine.
* fix out of bounds for mask
* skip rows where there are all masks
* skip tile if mask is inf
* store mask in worksize
* check inf tile earlier
* Boilerplate for q5_Kx8 REPACK on ARM and fallback
Signed-off-by: Alberto Cabrera <alberto.cabrera@liquid.ai>
* Implements make_block_q5_Kx8 by extending make_block_q4_Kx8
Signed-off-by: Alberto Cabrera <alberto.cabrera@liquid.ai>
* q5_K repack gemm and gemv generics
* Gemm and Gemv ARM implementations (i8mm)
* Improved qh manipulation looking at non-repack vec_dot implementation
* Full unroll
* Apply Q5_K Gemv vand and vshl optimizations to gemm. Improve comments.
Signed-off-by: Alberto Cabrera <alberto.cabrera@liquid.ai>
* Fix wrong fallback definitions of Q5_K
Signed-off-by: Alberto Cabrera <alberto.cabrera@liquid.ai>
* Fixed comments. Reverted unnecessary formatting
Signed-off-by: Alberto Cabrera <alberto.cabrera@liquid.ai>
* Fixed typo in generic definitions
* Switching AND + Shift with Shift Insert. Better op interleaving.
* Vectorize + unroll the block scales
* Apply gemm optimizations to gemv
* Improve bias calculation
---------
Signed-off-by: Alberto Cabrera <alberto.cabrera@liquid.ai>
* mla : pass V as a view of K to the FA op
* cuda : adjust mla logic to new layout
* kv-cache : fix rope shift
* tests : remove comment
* cuda : fix reusable_cutoff
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* opencl: add `copy_to_contiguous` and utilize mm kernels
* opencl: only copy to cont for f32 and f16 tensors
* opencl: use cont mm for fallback when dst is large
* opencl: use nb local to copy-to-cont
* opencl: use local offset as well
* vulkan: Remove transfer_ctx, do everything in compute_ctx.
We had a bug where a set_tensor_async (using transfer_ctx) didn't get
submitted before the graph_compute (using compute_ctx) that came after
it. To avoid this sort of issue, just do everything in compute_ctx.
Remove transfer_cmd_pool, which was already unused.
* fix crash with perf logger
Change ggml_vk_mul_mat_vec_id_q_f16 to loop over the batch dimension and
update the indexing calculations in get_offsets.
Mat-vec is faster than mat-mat for small values of n. We don't get the same
reuse of the weights as in the non-ID path, but with this the cost is linear
in n rather than n>1 being far slower than n==1.
I've had issues loading models with llama-server:
[44039] E gguf_init_from_file: failed to open GGUF file 'mistral-7b-v0.1.Q8_0.gguf'
and I was sure it could access the file. Seems like --models-dir and
--models-presets dont interact like I thought they would but I salvaged
this snippet that helps troubleshooting
[44039] E gguf_init_from_file: failed to open GGUF file 'mistral-7b-v0.1.Q8_0.gguf' (errno No such file or directory)
* CUDA: Replace `init_offsets` with iterators in argsort
This is a QOL improvement, saving us the cost of materializing the
iterator
* Remove unnecessary include from top-k.cu
* CANN: fix an issue where get_env was not fully renamed
* ci: add cann with acl group
* ci: define use_acl_graph using GitHub Action
* ci: update cann dockerfile with acl graph
* CANN: support gated linear attn
This change adds support for the GGML_OP_GATED_LINEAR_ATTN operator.
The feature was implemented by YushengZhao. Because the previous
submission was based on an outdated codebase, this PR was rebased to
merge.
Co-authored-by: YushengZhao <yusheng.chao@outlook.com>
Co-authored-by: hipudding <huafengchun@gmail.com>
* CANN: optimize OP gla
Optimize gla for high preformance
* Remove unused comments
---------
Co-authored-by: 赵禹昇 <2501112001@cninfer02.localdomain>
Co-authored-by: YushengZhao <yusheng.chao@outlook.com>
* hexagon: disable repack buffers if host buffers are disabled, improved handling of env vars
* hexagon: add support for OP_CPY fp16/fp32 -> fp16/fp32
Factore out all hvx_copy functions into hvx-copy.h header and reduced code duplication.
Update HTP ops infra to support OP_CPY
* hexagon: cleanup and refactor hex/hvx/htp headers and helper libs
hex is basically all scalar/core platform stuff (L2, DMA, basic utils)
hvx is all hvx related utils, helpers, etc
htp is higher level stuff like Ops, etc
hvx-utils library got a nice round of cleanup and refactoring to reduce duplication
use hvx_vec_store_a where possible
* hexagon: refactor HVX sigmoid functions to hvx-sigmoid.h
Moved sigmoid and tanh vector functions from hvx-utils.h to a new header
hvx-sigmoid.h. Implemented aligned and unaligned variants for sigmoid
array processing using a macro pattern similar to hvx-copy.h. Updated
act-ops.c to use the new aligned variant hvx_sigmoid_f32_aa. Removed
unused hvx-sigmoid.c.
* hexagon: factor out hvx-sqrt.h
* hexagon: mintor update to hvx-utils.h
* hexagon: remove spurios log
* hexagon: factor out and optimize hvx_add/sub/mul
* hexagon: remove _opt variants of add/sub/mul as they simply fully aligned versions
* hexagon: refactor reduction functions to hvx-reduce.h
Moved `hvx_self_max_f32` and `hvx_self_sum_f32` from `hvx-utils.h`/`.c` to `hvx-reduce.h`.
Renamed them to `hvx_reduce_max_f32` and `hvx_reduce_sum_f32`.
Added aligned (`_a`) and unaligned (`_u`) variants and used macros to unify logic.
Updated `softmax-ops.c` to use the new functions.
* hexagon: refactor the rest of arithmetic functions to hvx-arith.h
Moved `hvx_sum_of_squares_f32`, `hvx_min_scalar_f32`, and `hvx_clamp_scalar_f32` from `hvx-utils.c/h` to `hvx-arith.h`. Implemented aligned/unaligned variants (`_aa`, `_au`, etc.) and used macros to reduce code duplication. Updated `hvx_min_scalar_f32` and `hvx_clamp_scalar_f32` to use `dst, src, ..., n` argument order. Updated call sites in `act-ops.c`.
Refactor Hexagon HVX arithmetic functions (min, clamp) to hvx-arith.h
Moved `hvx_min_scalar_f32` and `hvx_clamp_scalar_f32` from `hvx-utils.c/h` to `hvx-arith.h`. Implemented aligned/unaligned variants (`_aa`, `_au`, etc.) and used macros to reduce code duplication. Updated these functions to use `dst, src, ..., n` argument order and updated call sites in `act-ops.c`. `hvx_sum_of_squares_f32` remains in `hvx-utils.c` as requested.
* hexagon: refactor hvx_sum_of_squares_f32
- Modify `hvx_sum_of_squares_f32` in `ggml/src/ggml-hexagon/htp/hvx-reduce.h` to use `dst, src` signature.
- Implement `_a` (aligned) and `_u` (unaligned) variants for `hvx_sum_of_squares_f32`.
- Update `hvx_reduce_loop_body` macro to support both returning and storing results via `finalize_op`.
- Update existing reduction functions in `hvx-reduce.h` to use the updated macro.
- Update `rms_norm_htp_f32` in `ggml/src/ggml-hexagon/htp/unary-ops.c` to match the new signature.
* hexagon: use hvx_splat instead of memset
* hexagon: consistent use of f32/f16 in all function names to match the rest of GGML
* hexagon: fix hvx_copy_f16_f32 on v75 and older
* hexagon: update readme to include GGML_HEXAGON_EXPERIMENTAL
* scripts: update snapdragon/adb scripts to enable host param
* CUDA: Refactor and expose two_stage_warp_reduce_* function
* Use `two_stage_warp_reduce` also in softmax kernel, move smem out of it
Moving smem out of `__device__` function to `__global__` function
allows for explicit smem reuse, as either compiler or cuda rt seem to not
free it afterwards (`cudaFuncSetAttribute` fails when not accounting for
it once for each call to two_stage_warp_reduce)
* Update ggml/src/ggml-cuda/common.cuh
Co-authored-by: Aman Gupta <amangupta052@gmail.com>
* Use two_stage_warp_reduce in group_norm_f32
* Use two_stage_warp_reduce in rms_norm_f32
* Fix smem calculation which expects bytes
* Make `two_stage_warp_reduce` accept all values warp_reduce accepts
Also integrate it into norm_f32 function
* Use two_stage_warp_reduce in l2_norm_f32
* Use type traits for block reduction for better legibility
Also adresss other requests by @am17an such as variable renaming
* Make norm tests cover all cuda paths
* Mark columns % WARP_SIZE !=0 as supported for RMS_NORM_BACK
Unit-tests passed locally, let's see if they pass in the CI as well
* Use `enum class` for `block_reduce_method`
This is more type-safe than plain enum
* Rename variables as suggested in code review by @am17an
* Rename two_stage_warp_reduce -> block_reduce
* Fix trailing whitespace in common.cuh
* Make condition of static_assert type-dependent
This delays evaluation until the template is actually instantiated.
Otherwise, some compilers may evaluate the assert when parsing the
template, resulting in build errors as observed here:
https://github.com/ggml-org/llama.cpp/actions/runs/20960323123/job/60235530068?pr=18785
* Inline definitions
---------
Co-authored-by: Aman Gupta <amangupta052@gmail.com>
Johannes Gäßler
Alberto Cabrera Pérez
Gaurav Garg
shalinib-ibm
lhez
ccbinn
Aman Gupta
Georgi Gerganov
nullname
Neo Zhang
shaofeiqi
Aleksei Nikiforov
Jeff Bolz
Masato Nakasaka
Matthieu Coudron
Oliver Simons
Adrien Gallouët
Reese Levine
Thore Koritzius
Perry Naseck
Raul Torres
Chenguang Li
hipudding
shaofeiqi
Max Krasnyansky
Daniel Bevenius
Ruben Ortlam
yulo