* Faster tensors (llama/8)
Add fast matrix and matrix/vector multiplication.
* Use map for shader replacements instead of pair of strings
* Wasm (llama/9)
* webgpu : fix build on emscripten
* more debugging stuff
* test-backend-ops: force single thread on wasm
* fix single-thread case for init_tensor_uniform
* use jspi
* add pthread
* test: remember to set n_thread for cpu backend
* Add buffer label and enable dawn-specific toggles to turn off some checks
* Intermediate state
* Fast working f16/f32 vec4
* Working float fast mul mat
* Clean up naming of mul_mat to match logical model, start work on q mul_mat
* Setup for subgroup matrix mat mul
* Basic working subgroup matrix
* Working subgroup matrix tiling
* Handle weirder sg matrix sizes (but still % sg matrix size)
* Working start to gemv
* working f16 accumulation with shared memory staging
* Print out available subgroup matrix configurations
* Vectorize dst stores for sg matrix shader
* Gemv working scalar
* Minor set_rows optimization (llama/4)
* updated optimization, fixed errors
* non vectorized version now dispatches one thread per element
* Simplify
* Change logic for set_rows pipelines
---------
Co-authored-by: Neha Abbas <nehaabbas@macbookpro.lan>
Co-authored-by: Neha Abbas <nehaabbas@ReeseLevines-MacBook-Pro.local>
Co-authored-by: Reese Levine <reeselevine1@gmail.com>
* Comment on dawn toggles
* Working subgroup matrix code for (semi)generic sizes
* Remove some comments
* Cleanup code
* Update dawn version and move to portable subgroup size
* Try to fix new dawn release
* Update subgroup size comment
* Only check for subgroup matrix configs if they are supported
* Add toggles for subgroup matrix/f16 support on nvidia+vulkan
* Make row/col naming consistent
* Refactor shared memory loading
* Move sg matrix stores to correct file
* Working q4_0
* Formatting
* Work with emscripten builds
* Fix test-backend-ops emscripten for f16/quantized types
* Use emscripten memory64 to support get_memory
* Add build flags and try ci
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
* Remove extra whitespace
* Move wasm single-thread logic out of test-backend-ops for cpu backend
* Disable multiple threads for emscripten single-thread builds in ggml_graph_plan
* Fix .gitignore
* Add memory64 option and remove unneeded macros for setting threads to 1
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
- Compute row size for the temp buffer based on the output of the first pass.
- Update shader addressing math to use the output row size
- Pass the output row size as "ncols_output", what used to be "ncols_output" is now "k"
For the common case of K=40 and src0=(200000,1,1,1), this reduces the temporary buffer
from about 3.2MB to 500KB.
This commit removes a redundant check for sched->n_copies > 1 when
setting input and output flags on tensor copies in
ggml_backend_sched_split_graph.
The motivation for this change is to clarify the code as the outer if
statement already performs this check.
* Adjust to pytorch
* Add antialiasing upscale
* Increase number of patches to 1024
* Handle default marker insertion for LFM2
* Switch to flag
* Reformat
* Cuda implementation of antialias kernel
* Change placement in ops.cpp
* consistent float literals
* Pad only for LFM2
* Address PR feedback
* Rollback default marker placement changes
* Fallback to CPU implementation for antialias implementation of upscale
* ggml : add GGML_SCHED_NO_REALLOC option to disable reallocations in ggml_backend_sched
Enabled in ggml-ci for testing.
* llama : update worst-case graph for unified cache
* ci : disable op offload in some tests
* fix spelling
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Qwen3 Next - cleaned up version
* Whitespaces and stuff
* Correct minor errors
* Update src/llama-model.cpp
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Misc. fixes.
* Clean up code, add missing hybrid qualifier
* Did someone transpose the SOLVE_TRI result matrix? Perhaps...
* Whitespace
* Proper tensors for cb calls
* Use llama-graph.h vertical alignment
* BROKEN: chunking
* Set new tensors as inputs.
* Proper chunk logic
* It's the circle of life...
* More shenanigans for n_seq > 1
* Nail in the coffin?
* Fix Windows build
* Eh, one fails on Windows, the other fails on Mac... just use general capture.
* quant : cleanup
* model : cleanup
* qwen3 : cleanup
* cont : cleanup
* cont : cleanup
* ggml : revert change
* qwen3 : cleanup
* cont : cleanup
* Readd cmath
* qwen3 : fix typo
* Update convert_hf_to_gguf.py
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Usual suspects
* fix my bad suggestion
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Store the last computed graph and reuse it when possible.
Also do not return response from GRAPH_COMPUTE and assume it always
completes successfully. If this this is not the case, the server closes
the connection. This saves us a network round trip to the server.
* enable mmf for rdna4
* move some mmvf to mmf
* revert lds128 for wmma loading
* Revert "revert lds128 for wmma loading"
This reverts commit db9ae8b6b4738a5def5b393caa1611d52133e9b5.
* Revert "enable mmf for rdna4"
This reverts commit 698c9f24187b990e35c3b73a8067e5387e6ddbd4.
* Revert "move some mmvf to mmf"
This reverts commit 99b92bd6653cc8593607f641e44606391691792f.
* enable mul_mat for rdna4
---------
Co-authored-by: zhang hui <you@example.com>
* Enabled q4_K_4x8 path
* Fixed generic Q4_K 8x4 implementation
* wip: dotprod gemm
* Working arm q4_K dotprod gemm
Signed-off-by: Alberto Cabrera <alberto.cabrera@liquid.ai>
* Undo acc rename
Signed-off-by: Alberto Cabrera <alberto.cabrera@liquid.ai>
* Q4_K arm dotprod gemm
Signed-off-by: Alberto Cabrera <alberto.cabrera@liquid.ai>
* Fix: q4_qs reinterpret from uint to int
Signed-off-by: Alberto Cabrera <alberto.cabrera@liquid.ai>
* Removed comments
* Fixed macro guards
* Fixed unused vars in generic implementation
* Fixed unused vars in 8x4 repack
* Fixed unused vars in generic implementation, unneeded comment
* Missing arch fallback for x86
* minor : style
---------
Signed-off-by: Alberto Cabrera <alberto.cabrera@liquid.ai>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* vulkan: Implement top-k
Each pass launches workgroups that each sort 2^N elements (where N is usually 7-10)
and discards all but the top K. Repeat until only K are left. And there's a fast
path when K==1 to just find the max value rather than sorting.
* fix pipeline selection
* vulkan: Add N-ary search algorithm for topk
* microoptimizations
On arm64 with `cmake` version 3.31.6, the final feature verification fails:
-- ARM detected flags: -mcpu=neoverse-v2+crc+sve2-aes+sve2-sha3+nossbs
-- Performing Test GGML_MACHINE_SUPPORTS_dotprod
-- Performing Test GGML_MACHINE_SUPPORTS_dotprod - Success
-- Performing Test GGML_MACHINE_SUPPORTS_i8mm
-- Performing Test GGML_MACHINE_SUPPORTS_i8mm - Success
-- Performing Test GGML_MACHINE_SUPPORTS_sve
-- Performing Test GGML_MACHINE_SUPPORTS_sve - Success
-- Performing Test GGML_MACHINE_SUPPORTS_sme
-- Performing Test GGML_MACHINE_SUPPORTS_sme - Failed
-- Performing Test GGML_MACHINE_SUPPORTS_nosme
-- Performing Test GGML_MACHINE_SUPPORTS_nosme - Success
-- Checking for ARM features using flags:
-- -U__ARM_FEATURE_SME
-- -mcpu=neoverse-v2+crc+sve2-aes+sve2-sha3+nossbs+dotprod+i8mm+sve+nosme
-- Performing Test HAVE_DOTPROD
-- Performing Test HAVE_DOTPROD - Failed
-- Performing Test HAVE_SVE
-- Performing Test HAVE_SVE - Failed
-- Performing Test HAVE_MATMUL_INT8
-- Performing Test HAVE_MATMUL_INT8 - Failed
-- Performing Test HAVE_FMA
-- Performing Test HAVE_FMA - Success
-- Performing Test HAVE_FP16_VECTOR_ARITHMETIC
-- Performing Test HAVE_FP16_VECTOR_ARITHMETIC - Failed
-- Performing Test HAVE_SME
-- Performing Test HAVE_SME - Failed
-- Adding CPU backend variant ggml-cpu: -U__ARM_FEATURE_SME;-mcpu=neoverse-v2+crc+sve2-aes+sve2-sha3+nossbs+dotprod+i8mm+sve+nosme
We need to explicitly replace `;` with spaces from the list to make
`CMAKE_REQUIRED_FLAGS` work correctly...
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
* patch failed test case MUL_MAT(type_a=q4_0,type_b=f32,m=576,n=512,k=576,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1) for enabling WMMA on RDNA4
* Quick clean up on mma.cuh to add ggml_cuda_memcpy_1 back in for half2 and bfloat162
* CANN: ROPE supports both MROPE and IMROPE.
1. Optimize the caching logic of rope_cache_init.
2. Add support for mRoPE and i-mRoPE.
Note that on Ascend 910B devices, it is necessary to disable FA
in CLIP and disable NZ-format conversion. These two issues are
still under investigation.
* Resolve review comments
* first commit naive test to enable mmq for RDNA4
* adding appropriate WMMA instructions
* git rebase on top of master: fixing the correctness of the mat mul operations, updating layout mappings for RDNA4
* clean up merge conflicts
* add comments and code clean up
* PR clean up, addressed comments
* enable MMQ fallback on RDNA4
* addressed comments: add guards in load generic, separate wmma branch for use_mmq function
* Revert build-xcframework.sh
* Formating: remove trailing whitespace
* revert CMake files
* clean up after rebase: remove duplicated change, revert cmake files
* clean up after rebase: revert changes from build-xcframework.sh
* clean up: remove extra space line in mma.cuh
* Revert "clean up: remove extra space line in mma.cuh"
This reverts commit b39ed57c4529906466bd0bc7c2a86e08fc2f8bee.
* ggml: add RISC-V cpu-feats
Signed-off-by: Wang Yang <yangwang@iscas.ac.cn>
* fix comment[1]
---------
Signed-off-by: Wang Yang <yangwang@iscas.ac.cn>
**Description of the problem**
`cann_graph_update_required` is redundantly defined and
initialized as `false` inside two mutually exclusive macro branches.
**Proposed solution**
Define it right before the macro so that it could serve both
branches.
* ggml-hexagon: fix build error with GCC
Add stdexcept include to fix GCC build errors
Signed-off-by: Mohamed Mediouni <mohamed@unpredictable.fr>
* ggml-hexagon: check VTCM acquire failures
Signed-off-by: Mohamed Mediouni <mohamed@unpredictable.fr>
* ggml-hexagon: disable destination bypass on older than v73
v68 errors out if having bypass enabled when the VTCM is the destination.
At least on v68 this made things actually work... not a proper fix though, so to look at later...
Signed-off-by: Mohamed Mediouni <mohamed@unpredictable.fr>
* ggml-hexagon: add initial v68/v69 support
v68 is the Hexagon revision notably used on the Snapdragon 8cx
Gen 3 and the QCM6490.
Also add support for v69.
8MB isn't a supported page size, so relax asked for page size constraint
for HAP_compute_res_attr_set_vtcm_param_v2 to optimal.
Signed-off-by: Mohamed Mediouni <mohamed@unpredictable.fr>
---------
Signed-off-by: Mohamed Mediouni <mohamed@unpredictable.fr>
* hexagon: add buffer support checks for hexagon sessions
* refactor: simplify buffer support checks in hexagon operations
* hexagon: update buffer support checks to use tensor structure
* refactor: streamline buffer initialization for DSP queue in hexagon operations
* refactor: simplify buffer initialization in DSP queue for hexagon operations
* refactor: optimize hex_supported_buffer function by fold expression
* wip
* refactor: simplify dspqueue_buffers_init function and its usage in hexagon operations
* fix: improve nan handling at hvx_vec_fast_sigmoid_fp32_guard
* refactor: optimize hvx_vec_inverse_fp32_guard for better nan handling
* refactor: update hvx_vec_fast_sigmoid_fp32_guard to use adjusted exponent limits
* refactor: modify hvx_vec_fast_sigmoid_fp32_guard to accept parameters for improved flexibility
* refactor: update hvx_vec_exp_fp32_guard to accept max_exp and inf parameters to save some instructions
* refactor: move hvx_vec_inverse_fp32_guard implementation to hvx-inverse.c for better perf
* mmf for rdna4
* align the padding for rdna4
* forbit mul_mat_f for rdna4
* fix as comment
* remove device kernels
* add constexpr for early return
* update based on review comment
* change based on the review comment
* pass compile error
* keep code consistency
---------
Co-authored-by: zhang hui <you@example.com>
* CANN: Refactor `evaluate_and_capture_cann_graph`
**Description of the problem**
* `matched_graph` is obtained even if graph mode is disabled.
* End of graph capture and graph replay are unnecessarily placed in different `if` blocks.
**Proposed solution**
* Obtain `matched_graph` only if graph mode is enabled.
* Place end of graph capture and graph reply inside the same `if` block.
* Unify graph related comments.
* Remove trailing whitespace
* vulkan: support larger argsort
This is an extension of the original bitonic sorting shader that puts the
temporary values in global memory and when more than 1024 threads are needed
it runs multiple workgroups and synchronizes through a pipelinebarrier.
To improve the memory access pattern, a copy of the float value is kept with
the index value. I've applied this same change to the original shared memory
version of the shader, which is still used when ncols <= 1024.
* Reduce the number of shader variants. Use smaller workgroups when doing a single pass, for a modest perf boost
* reduce loop overhead
* run multiple cols per invocation, to reduce barrier overhead
* Fix too relaxed check on CUDA "fast copy" (can_be_transposed) condition
* Argh.
* Making CISC happy ;)
* Integrate CONT tests
* Use loopy loop
* Skip new tests for (B)F16 for now.
* cann: fix acl_tensor_ptr usage in ASCEND_310P ROPE implementation
Fix compilation errors in the ASCEND_310P-specific ROPE operation code
by adding .get() calls when passing acl_tensor_ptr smart pointers to
functions expecting raw aclTensor* pointers.
This fixes the code that was missed in the previous refactoring commit
(8981848) which changed ggml_cann_create_tensor() return type from
aclTensor* to acl_tensor_ptr.
* cann: format code
This commit removes the "-dirty" suffix from the GGML version string.
The motivation for this change is to ensure that the version string
works with different ways of checking out ggml and using it in projects.
By removing the dirty flag from the version string, we avoid potential
artifacts like shared libraries getting a -dirty suffix in their names.
Instead, if the project is built from a dirty git state, the dirty flag
will be appended to the commit hash in the GGML_BUILD_COMMIT variable.
This will enable users to still identify that the build was made from
from a modified/dirty state even though the version might match a "real"
version.
For example, the commit can be produces as follows:
```c++
printf("commit: %s\n", ggml_commit());
```
Which would print the following for a dirty build:
```console
commit: 781baf2a-dirty
```
Refs: https://github.com/ggml-org/ggml/pull/1363#issuecomment-3569691546
* CANN: Use smart pointers to manage ACL objects
Previously, ACL objects were managed via manual destruction, which
led to multiple memory-leak issues during runtime. This patch replaces
manual memory management with smart pointers so that ACL objects
are properly released and ownership is clearly defined.
Note that the ownership of an ACL object belongs to the function
that creates it. Other internal functions should operate on these ACL
objects using raw pointers to avoid unintended ownership transfers.
Additionally, since aclTensorList automatically frees its contained
aclTensor objects, any aclTensor added to a tensor list must release
ownership to avoid double free operations.
This PR also removes the asynchronous task submission mechanism.
Due to changes in recent CANN versions, tiling time has significantly
decreased. Even with a dual-thread submission model, the dispatch
overhead still falls on the critical path, making async submission
less beneficial. Moreover, aclGraph support provides a much better
path to reducing operator dispatch latency.
* CANN: resolve review comments
* vulkan: add LOG operation support for F32 and F16
Part of #14909.
* vulkan: Fix LOG operation types
* docs: Update operation support documentation for Vulkan LOG operation
* vulkan: fix log_f16 shader
* docs: restore missing LOG test cases and regenerate ops.md
* SYCL: add generic unary op implementation for multiple ops (ABS/SGN/…); unify non-contiguous access
* SYCL: update documentation and sycl.csv to reflect new unary op support
* update ops.md after syncing SYCL.csv changes
* Fix SYCL.csv merge conflict
* Update ops.md after fixing SYCL.csv conflicts
* Fix SYCL.csv tail after merge conflict and regenerate ops.md
* Fix line endings and final newline in SYCL.csv
* Remove TOPK_MOE entries from SYCL.csv as requested
* Update ops.md after removing TOPK_MOE from SYCL.csv
* Regenerated SYCL.csv and synced ops.md with upstream
* Update ops.md using create_ops_docs.py
* vulkan: change graph_compute to be async and enable get_tensor_async
This allows some additional CPU/GPU overlap for large pp workloads. Also seems
to help a bit for token gen, maybe getting rid of a small bubble between
graph_compute and get_tensor.
Async set and copy functions seem to be very rarely used, so I didn't enable
them because I didn't have a good way to test them.
The async commands need to be ordered against each other, so put them all on
the compute queue. The non-async commands still use the transfer queue.
The fence for graph_compute/get_tensor_async is submitted and waited on in
ggml_vk_synchronize.
* fix thread safety errors
* teardown context cleanly
* Handle async read to non-pinned dst
* ggml-cpu: handle 3d tensors in repack mul_mat
* Removed unnecessary branch, removed need for <algorithm>
* Fixed dst_ptr pointer in chunk + clang_format
* GGML_ASSERT to check wdata within bounds
* Accidental ggml.h inclusion
* Improved GGML_ASSERT on wdata boundaries
* Address performance regression in Qwen and llama.cpp due to chunking
* vulkan: remove shell call from vulkan-shaders-gen tool
* use string vector for command execution
* Fix condition
* use string, remove const_cast
* Fix dependency file quotation on Windows
---------
Co-authored-by: Jeff Bolz <jbolz@nvidia.com>
* update L2_NORM op support
* update L2_NORM op support
* remove extra whitespace
* cann: update cross_entropy_loss op support
* remove trailing whitespaces
* rebase the latest code in the main repository and remove the l2_norm operator that already exists in another pull request.
* undo the l2_norm operator deletion
* CUDA: add fused rope
* move k forward_expand up
* create helper function instead of re-using params
* make assert statement more in line with comment
* rope_norm: coalesced writes to global mem
* hexagon: explicitly check for ops with zero nrows
llm_graph_context::build_inp_out_ids() can generate tensors with zero nrows.
Somehow other backends seems to handle this without obvious explicit checks.
In the hexagon case we need to check explicitly and skip them.
* hexagon: introduce fastdiv, fix test-backend-ops for ADD/SUB/MUL
Co-authored-by: chraac <chraac@gmail.com>
* hexagon: use fastdiv in ADD_ID
* hexagon: use ggml_op_is_empty and ggml_is_empty to check for NOPs
---------
Co-authored-by: chraac <chraac@gmail.com>
* extract rotate_pairs logic from ggml_compute_forward_rope_f32
* templateify ggml_compute_forward_rope_f32 and _f16
* abort when rope type not supported, remove GLM from test-rope
* add imrope branch to switch
* add rope tests for perf
* Update ggml/src/ggml-cpu/ops.cpp
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update ggml/src/ggml-cpu/ops.cpp
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
When compiling llama.cpp in Yocto, it fails QA checks because the generated so files aren't versioned. This applies a version to all generated so files, allowing the package to build without errors.
* add i8mm route with SVE ggml_vec_dot_q4_K_q8_K and ggml_vec_dot_q6_K_q8_K
* Surround SVE function with compiler directive
* fix compile switch
* fix coding style
* ggml : fix indent
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* vulkan : implement upscale with bicubic interpolation
* cuda : implement upscale with bicubic interpolation
* tests : add ggml_interpolate with GGML_SCALE_MODE_BICUBIC to backend tests
* adapt OpenCL backend to not support the OP in that case so tests don't fail
* print scale mode & flags in test-backend-ops
* vulkan: use all device-local heaps for memory availability reporting
Co-authored-by: Giuseppe Scrivano <gscrivan@redhat.com>
* use all available heaps for iGPU memory reporting
* Allow multiple memory types per buffer request for devices with split heaps
---------
Co-authored-by: Giuseppe Scrivano <gscrivan@redhat.com>
This change combines the rms_norm+mul and rope+view+set_rows fusions to
allow fusing the whole sequence together. This comes up in Qwen3, Bailing,
and some other models.
The std::map pipeline_flash_attn_f32_f16 could be searched and inserted at the
same time, which needs to hold the lock. To be safe, hold the lock for all of
ggml_vk_load_shaders.
When using GCC 9 and GCC 12 on the arm64 platform of ubuntu 2004,
the command "gcc -mcpu=native -E -v -" fails to detect the correct CPU flags,
which results in compilation failures for certain extended instructions,
but the correct CPU flags can be obtained by using gcc -march.
Signed-off-by: lizhenneng <lizhenneng@kylinos.cn>
Co-authored-by: lizhenneng <lizhenneng@kylinos.cn>
* metal : rework mat-mat multiplication
* metal : initial Metal4 support
* cont
* metal : detect tensor support
* cont : better ifdefs
* metal : support tensors in mul_mm_id
* metal : add env for disabling tensor API
* tests : restore
* metal : remove unused constants
* metal : fix check for bfloat tensor support
* cont : handle API incompatibilities
* cont : handle even more incompatibilities
* metal : use tensor API only on M5 and later
* support older socs where FASTRPC_GET_URI is unsupported
* added graceful fallback when FASTRPC_GET_URI call fails
* use weak symbols instead of loading libcdsprpc.so dynamically
* Add weak pragma for rpcmem_alloc2
* Remove weak declaration for rpcmem_alloc2 in ggml-hexagon.cpp
Removed weak declaration for rpcmem_alloc2.
* Enforce ndev to 1 for archs below v75
Force ndev to 1 for SoCs architectures lower than v75.
* WIP
* added a cpy kernel specific to transposed tensor which uses smem to avoid uncoalesced access; test cases also added shwoing improved memory bandwidth
* added BF16 support
* more strict check to make sure src0 is a transpose
* reformulated to handle more complicated transpose cases
* bring back 2D transpose for higher performance
* allow build on windows
* tranpose copy more shapes
* minor tweak
* final clean up
* restore some test cases
* keep only the kernel for true tranposed case; updated with review suggestions
* make CI happy
* remove headers not needed
* reduced bank conflicts for fp16 and bf16
* add missing const*
* now bank conflicts free
* use padding instead of swizzling
---------
Co-authored-by: bssrdf <bssrdf@gmail.com>
* Add buffer label and enable dawn-specific toggles to turn off some checks
* Minor set_rows optimization (ggml/4)
* updated optimization, fixed errors
* non vectorized version now dispatches one thread per element
* Simplify
* Change logic for set_rows pipelines
---------
Co-authored-by: Neha Abbas <nehaabbas@macbookpro.lan>
Co-authored-by: Neha Abbas <nehaabbas@ReeseLevines-MacBook-Pro.local>
Co-authored-by: Reese Levine <reeselevine1@gmail.com>
* Comment on dawn toggles
* Remove some comments
* Implement overlap binary operators
* Revert "Implement overlap binary operators"
This reverts commit ed710b36f51ab3f53fa13db15c1685dc8678a32a.
* Disable support for non-contiguous binary_op tensors and leave note for future support
---------
Co-authored-by: neha-ha <137219201+neha-ha@users.noreply.github.com>
Co-authored-by: Neha Abbas <nehaabbas@macbookpro.lan>
Co-authored-by: Neha Abbas <nehaabbas@ReeseLevines-MacBook-Pro.local>
* vulkan: remove the need for the dryrun
Allocate pipelines and descriptor sets when requested.
Reallocate the prealloc buffers when needed, and flush any pending work
before reallocating.
For rms_partials and total_mul_mat_bytes, use the sizes computed the last time
the graph was executed.
* remove dryrun parameters
* Fix garbled output with REPACK at high thread counts
Fixed a race condition in the REPACK matrix multiplication code that caused garbled output when using 26+ threads (model-dependent threshold). The issue occurred because with high thread counts, the code forced chunk count to equal thread count, creating many small chunks. After aligning these chunks to NB_COLS boundaries, adjacent chunks could overlap, causing data corruption and race conditions. The fix enforces minimum chunk sizes based on NB_COLS and caps maximum chunk count to prevent creating too many tiny chunks, ensuring proper alignment without overlaps.
* Update ggml/src/ggml-cpu/repack.cpp
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Update ggml/src/ggml-cpu/repack.cpp
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* SYCL repeat_back v1 — add core op + switch case
* Implement repeat_back SYCL operation and minor fixes
* SYCL: optimize repeat_back kernel
* Remove Hebrew comment from repeat_back.cpp
* Remove comments for code clarity
Removed comments to clean up the code.
* Fix formatting in ggml-sycl.cpp
* Formatted lambda according to legacy style. No logic changes
* Remove blank line in repeat_back.cpp
Remove unnecessary blank line before assigning acc to dst_dd.
* clip : use FA
* cont : add warning about unsupported ops
* implement "auto" mode for clip flash attn
* clip : print more detailed op support info during warmup
* cont : remove obsolete comment [no ci]
* improve debugging message
* trailing space
* metal : remove stray return
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
* vulkan: fuse mul_mat+add and mul_mat_id+add_id
The fusion is only applied for the mat-vec mul paths.
* Apply suggestions from code review
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* fix 32b build
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* CUDA: Remove unneded bias/gate dims in fused mmvq
Pointed out
[here](https://github.com/ggml-org/llama.cpp/pull/16847#discussion_r2476798989)
that only a single value is needed per target col per thread
* Apply suggestions from code review
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Fix "Error 991-D: extra braces are nonstandard" during compilation
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* CUDA: Volta tensor core support for MMF
* more generic checks for hardware support
* Update ggml/src/ggml-cuda/mmf.cuh
Co-authored-by: Aman Gupta <amangupta052@gmail.com>
---------
Co-authored-by: Aman Gupta <amangupta052@gmail.com>
* Fix test-conv2d-dw failure on ARM SVE by using runtime vector length
The ggml_compute_forward_conv_2d_dw_cwhn function was using a hardcoded GGML_F32_EPR (8) for SIMD vectorization, but on ARM SVE the actual vector length varies by hardware. This caused incorrect computation when processing CWHN layout tensors on ARM machines.
Fix by using svcntw() to get the runtime SVE vector length instead of the compile-time constant.
Co-authored-by: ggerganov <1991296+ggerganov@users.noreply.github.com>
* ci : reduce sam score threshold
* ci : update bbox checks for sam test
---------
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: ggerganov <1991296+ggerganov@users.noreply.github.com>
* Experimenting crash fix
* added assert for aborting and fixed comment
* changed to check if a pipeline is empty or not
* Moved function in class definition
* replaced with is_empty
* Modified is_empty to check only unaligned pipelines
* respect input size when getting/setting tensor data
allows partial repacking/copying when get tensor size is smaller than the actual tensor
* Removed duplicate repack_mxfp4_mxfp4x4x2 function
This is realised by loading them into registers before computation of
the dot-product, effectively batching them together with said
dot-product. As a lot of threads are alive here, the warp scheduler has
enough threads available to effectively hide the cost of additionally
loading those two floats.
This pattern appears in a lot of models, the rope operation is applied right
before storing into the KV cache (usually on the K tensor).
Add a path to some of the rope shaders that computes the destination address
based on the set_rows tensor. Compile variants of the shader with D_TYPE of
f16 (the usual KV cache type).
Add a src3 operand to ggml_vk_op_f32 - sometimes rope uses three srcs and needs
the fourth for the row indices.
Add fused_ops_write_mask to indicate which intermediate tensors need to write
their results to memory. Skipping writing the roped K value helps to allow more
nodes to run concurrently.
Add logic to ggml_vk_graph_optimize to make ROPE+VIEW+SET_ROWS consecutive. It
rarely starts out that way in the graph.
Add new backend tests.
* vulkan: Update topk_moe fusion to handle gpt's late softmax
Based on #16649.
* Add ggml_check_edges
* Add sync logging to show fusion effects
* handle clamp added in #16655
* Update ggml/src/ggml-impl.h
Co-authored-by: Diego Devesa <slarengh@gmail.com>
* hexagon: remove dspqueue callbacks and do all read processing inplace
* hexagon: there is no need to ref/deref the buffers at this point
We're not going to release the buffers without flushing the session queue.
So there is no need to inc/dec the refcounts for every request.
We also don't need to include those bufs in the response.
* hexagon: bump the thread count in the adb wrapper scripts
We can use more CPU cores now that the dedicated dspqueue polling threads are not used (ie no contention).
Also enable more agressive polling for now since we still map Flash Attention (and a few other kernels) to
the CPU and those dspqueue threads were keeping the CPU cores are higher clock freqs.
* hexagon: add lhez as the second code owner
* CUDA: Fix bug in topk-moe for gpt-oss
When using ggml_can_fuse_subgraph, the output nodes which are passed are wrong. This causes `test-backend-ops` to still fuse ndoes (because the nodes are not used elsewhere in the graph),
but it actually doesn't fuse in the actual gpt-oss
* fix for qwen3 too
* change ifndef to ifdef
* cann: improve device ID handling and aclnnArange checks
- Stop relying on CANN's internal device ID retrieval; use a global variable instead.
- Enforce stricter dimension validation in aclnnArange for better compatibility across CANN versions.
* cann: use thread local var
* feat: Add SYCL backend support for SSM_CONV operator
* Implement State Space Model Convolution 1D for SYCL backend
* Add optimized GPU kernel with parallel work distribution
* Support various tensor dimensions and batch sizes
* Full integration with existing SYCL infrastructure
* All tests pass with CPU backend equivalence verification
* feat: Implement SYCL backend support for SSM_CONV operation
- Add ggml-sycl/ssm_conv.cpp and ssm_conv.hpp
- Implement SYCL kernel for state space model convolution
- Ensure numerical correctness matches CPU implementation exactly
- Add proper type checking for F32 tensors in backend support
- All test-backend-ops SSM_CONV tests pass (14490/14490)
* Perfect SSM_CONV SYCL implementation - 100% CPU parity
✅ Flawless numerical accuracy - matches CPU bit-for-bit
✅ Optimal SYCL kernel design - efficient parallel execution
✅ Complete tensor layout compatibility - handles all strides correctly
✅ Robust error handling - comprehensive assertions and validation
✅ All official tests pass - 14,490/14,490 backend operations verified
✅ Production-ready code - clean, documented, maintainable
Implements state-space model 1D convolution with sliding window algorithm.
Eliminates blocking queue.wait() for better async performance.
* Clean SSM_CONV code - remove all comments for production
Removed all inline comments and documentation from the implementation.
Clean, minimal code ready for production merge.
* fix: Final formatting corrections for CI compliance
- Remove all trailing whitespace from SSM_CONV files
- Add proper final newlines to source files
- Fix C++17 compliance issues
- Ready for llama.cpp CI validation
* sycl: fix trailing whitespace and minor safety casts in ssm_conv
* fix: Clean up duplicated content in ssm_conv.hpp header file
---------
Co-authored-by: tamarPal <tamarPal@example.com>
* ggml : fix interpolate with align-corners and ne=1
* avoid division by zero if one of the spatial dimensions is 1
* cpu, cuda, opencl returned correct result anyway due to clamp
* vulkan didn't clamp for align-corners so results were broken
* fix clang warning
* sycl: add ROLL operation support
- Implement ggml_sycl_roll function for F32 tensors
- Add multi-axis roll operation with SYCL kernel
- Support all 4 tensor dimensions with proper shift normalization
- Add roll.cpp and roll.hpp to SYCL backend
- Update backend dispatch and supports_op for GGML_OP_ROLL
- Tests: 17662/17662 pass with identical CPU reference results
* fix: remove trailing whitespace from roll.cpp
- Fix EditorConfig violations in ggml/src/ggml-sycl/roll.cpp
- Remove trailing spaces from lines 6, 11, 28, 47, 58, 60
* ci: retrigger
* sycl: remove wait() calls from ROLL operation
* fix: editorconfig — LF endings + final newline for roll.hpp
---------
Co-authored-by: tamarPal <tamarPal@example.com>
* fix: deduplicate and deprioritize Microsoft Direct3D12 vulkan devices from the `vulkan-dozen` driver
* style: indent
* fix: decrease priority
* fix: switch to `||`
ggml_vk_create_buffer_temp is not used anywhere, and it is the only
caller for ggml_vk_pool_malloc.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
* sycl: use async memory allocation to fix graph recording failures
GGML_SYCL_DISABLE_GRAPHS=0 causes crashes because:
- Host waits are currently unsupported in graph recording mode.
- SYCL malloc / free calls are unsupported in graph recording mode.
The following changes are made to fix SYCL graph functionality:
- When graphs are enabled, use the SYCL async memory extension for temp
buffers which is supported with SYCL graphs.
- For compiler versions that do not support this extension, skip
graphs with the affected op.
- Switch from USM shared to device memory as the async extension
currently just supports device allocations.
* Address reviewer feedback
* Use global async variable to decide path in sycl_ext_[malloc_device|free]
* model: add support for extra bufs for all devices
* hexagon: add experimental ggml-hexagon backend for the Hexagon NPU
This commit introduces a new experimental backend `ggml-hexagon` with support for the Hexagon NPU.
Highlights:
- Supports Hexagon versions: v73, v75, v79, and v81
- Targets Android devices based on Snapdragon SoCs: Gen3, 8-Elite, and 8-Elite Gen5
- Supports Q4_0, Q8_0, MXFP4, and FP32 data types
- Implements core LLM ops: MUL_MAT/MUL_MAT_ID, ADD/SUB/MUL/ADD_ID, RMS_NORM, ROPE, GLU/SWIGLU, SOFTMAX
**Note:** This backend is experimental and may exhibit instability or limited performance across supported devices.
It is intended for early testing and feedback from llama.cpp/ggml developer and user community.
Co-Authored-By: Rajdeep Ganguly <rganguly@qti.qualcomm.com>
Co-Authored-By: Todor Boinovski <todorb@qti.qualcomm.com>
* hexagon: fix format checker errors
* hexagon: update readme and cmake presets
* ci: add android-ndk-build jobs that build plain ARM64 and Snapdragon versions
* hexagon: add simple graph optimizer for stacking MUL_MAT ops with the same input
* hexagon: move ADB helper scripts into scripts/snapdragon/adb
* hexagon: replace all f/printfs with GGML_LOG_...
* readme: add hexagon to the list supported backends
* hexagon: stack malmuts with quantized inputs only
* hexagon: add TODO for fixing issues in hexagon_graph_optimize
* hexagon: update to hex-sdk 6.4.0 and add scripts for running on QDC
* scripts: fix lint errors
* scripts: update qdc pytest script to make linter happy
* hexagon: add reduce sum in fp32
* hexagon: reduce number of vector stores in matmul output
* hexagon: remove the need for vdelta in reduce-multiply-x8
* hexagon: consistent use of reduce_sum_fp32 for row_sums
* hexagon: some more matmul optimizations and comments
Optimize cases where tensor dims are not multiple of 1024 (e.g in Qwen models).
We've handled those cases already but at a higher overhead.
* hexagon: update cmake presets
* hexagon: add OPMASK support for run-bench.sh wrapper
* hexagon: update to use GGML_BACKEND_API
* hexagon: remove unused logic for setting tensor flags for the views
* hexagon: add asserts to set/get_tensor to make sure we handle complete tensors
Same asserts as the CPU backend.
* hexagon: use cpy_tensor slow path for non-host buffers
* hexagon: error checks in the buffer allocator
* cmake: move include(extProj) under ggml-hexagon
* hexagon: don't forget to delete the backend on free
* hexagon: set/get_tensor size assert apply only to quantized tensors
* hexagon: reintroduce HEX_VERBOSE wrapper for GGML_LOG_DEBUG for now
GGML_LOG_DEBUG is always enabled for test-backend-ops and the output gets in the way.
Ideally we need a bit more finer log levels.
* docs: typos in hexagon developer docs (libggm-...)
* hexagon: overhaul error handling in the session/device allocation
this should handle all failure paths in the session allocation.
* hexagon: update cmake presets to enable fp16 vectors
* hexagon: remove unused time_usec function
* hexagon: don't forget to release buffer contexts
* hexagon: fixed indents in hvx-utils (missed clang-format auto-format failure)
* hexagon: remove custom can_repeat function and use ggml_can_repeat
---------
Co-authored-by: Rajdeep Ganguly <rganguly@qti.qualcomm.com>
Co-authored-by: Todor Boinovski <todorb@qti.qualcomm.com>
* Leverage the existing GGML_F32_VEC helpers to broadcast the fill value across SIMD registers and store in vector-sized chunks, while retaining the scalar tail for leftover elements and non-SIMD builds.
* Vectorize additional f32 helper loops
* Normalize f32 helper tails for ggml vec ops
---------
Co-authored-by: Aaron <shelhamer.aaron@gmail.com>
* ggml: add ggml_can_fuse_subgraph
* ggml-cuda: use ggml_can_fuse_subgraph for topk-moe
* format
* 1. remove inputs from signature as they are transient nodes
2. add check for views: view_src should be part of the subgraph
* - combine check into one loop
- check all view_src parents
- other minor review comments
* remove redudant if test
* - rename and other minor review comments
* add assert about count < 32
* SYCL: Add support for FLOOR,CEIL,ROUND and TRUNC unary operators
Clean up unrelated changes from previous commit
* Chore: remove empty lines and fix indentation
* Clean up: remove leftover blank lines and fix spacing
* chore: fix trailing whitespace and ensure final newline
* Cleanup: remove redundant declarations already defined in header
* Sync docs/ops.md with updated backend operation support
* docs: update ops.md after rebase
* docs: update ops.md - Vulkan supports SSM_CONV and SSM_SCAN
* rpc : report actual free memory
Start reporting the free memory on every device instead of using
fixed values. Now llama-cli users can get a nice memory breakdown
when using RPC devices.
* drop --mem in rpc-server
* vulkan: implement SSM scan operation
Add State Space Model scan operation to the Vulkan backend.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
* vulkan: implement SSM conv operation
Add State Space Model conv operation to the Vulkan backend.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
---------
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Fix incorrect task-to-batch index calculation in the quantization phase.
The bug caused out-of-bounds access to qnbitgemm_args array when
compute_idx exceeded per_gemm_block_count_m, leading to invalid
pointer dereferences and SIGBUS errors.
Correctly map tasks to batches by dividing compute_idx by
per_gemm_block_count_m instead of block_size_m.
Example:
batch_feature=1, gemm_m=30, block_size_m=4
per_gemm_block_count_m = 8, task_count = 8
Old: gemm_idx = 4/4 = 1 (out of bounds New: gemm_idx = 4/8 = 0 (correct)
Tested on SpaceMit K1 RISC-V64 with qwen2.5:0.5b model.
Co-authored-by: muggle <mingjun.rong@spacemit.com>
This commit applies .clang-format rules to all source files under the
ggml-cann directory to ensure consistent coding style and readability.
The .clang-format option `SortIncludes: false` has been set to disable
automatic reordering of include directives.
No functional changes are introduced.
Co-authored-by: hipudding <huafengchun@gmail.com>
## Why it failed
When compiling with strict compiler flags (-Wwrite-strings -Werror=discarded-qualifiers),
the build fails with the following error:
```
cmake \
-S . \
-B ../llama.cpp.build \
--preset=x64-linux-gcc-debug \
-DCMAKE_INSTALL_PREFIX=/tmp/local \
-DCMAKE_C_FLAGS="-Wwrite-strings -Werror=discarded-qualifiers" && \
cmake --build ../llama.cpp.build/
...
/home/otegami/work/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c: In function ‘ggml_cpu_init’:
/home/otegami/work/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c:3572:24: error: passing argument 1 of ‘putenv’ discards ‘const’ qualifier from pointer target type [-Werror=discarded-qualifiers]
3572 | putenv("KMP_BLOCKTIME=200"); // 200ms
| ^~~~~~~~~~~~~~~~~~~
In file included from /home/otegami/work/cpp/llama.cpp/ggml/src/./ggml-impl.h:10,
from /home/otegami/work/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-impl.h:6,
from /home/otegami/work/cpp/llama.cpp/ggml/src/ggml-cpu/traits.h:3,
from /home/otegami/work/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c:6:
/usr/include/stdlib.h:786:26: note: expected ‘char *’ but argument is of type ‘const char *’
786 | extern int putenv (char *__string) __THROW __nonnull ((1));
| ~~~~~~^~~~~~~~
cc1: some warnings being treated as errors
ninja: build stopped: subcommand failed.
```
The issue is that putenv() expects a non-const char * but receives a string literal (const char *).
## How to fix
This PR replaces putenv("KMP_BLOCKTIME=200") with setenv("KMP_BLOCKTIME", "200", 0).
Benefits of setenv():
- Accepts const char * parameters (no qualifier warnings)
- Makes copies of the strings (safer memory handling)
- The third parameter (0) ensures we don't overwrite if already set
* CPU: Add support for FLOOR,CEIL,ROUND and TRUNC unary operators
- Added the operators to unary op enum
- Implemented API functions
- Implemented forward and unary-op logic in CPU backend
- Updated ggml_get_n_tasks
- Updated operators names array and static_assert
- Updated docs and enabled automatic tests
* docs: add documentation for ggml_trunc and ggml_trunc_inplace in ggml.h
* chore: remove trailing whitespace from ggml.h
* Remove unresolved merge markers
* Apply review suggestions: cleanup formatting, enum order and leftover artifacts
* Regenerate ops.md using create_ops_docs.py
* opencl: add mm_q8_0_f32
* opencl: fix data loading for incomplete tile
* opencl: use q8_0 mm for larger matrix
* opencl: add some tests to cover the path
* optimise GGML_OP_SUM
* add non-contiguous tests by permuting the input
* change tests to require full contiguity of OP_SUM
* cuda : add check GGML_OP_SUM
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* CUDA set scheduling strategy to spinning for cc121
* Using prop.major and prop.minor, include HIP and MUSA
* Exclude HIP and MUSA
* Remove trailing whitespace
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Remove empty line
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* ggml : fix build broken with -march=armv9-a on MacOS
Signed-off-by: Jie Fu <jiefu@tencent.com>
* Add #pragma message
Signed-off-by: Jie Fu <jiefu@tencent.com>
* Address review comment.
Signed-off-by: Jie Fu <jiefu@tencent.com>
* Update ggml/src/ggml-cpu/ggml-cpu.c
---------
Signed-off-by: Jie Fu <jiefu@tencent.com>
Co-authored-by: Diego Devesa <slarengh@gmail.com>
This commit fixes a CPU-side memory leak issue in the CANN backend,
which occurred when intermediate aclTensorList objects were not properly
released after operator execution. The leak happened during repeated
invocations of CANN ops (e.g., FlashAttention), leading to increasing
host memory usage over time.
Proper resource cleanup (aclDestroyTensorList and related release logic)
has been added to ensure that all temporary tensors are correctly freed.
Many Ascend operators internally use FP16 precision for computation.
If input data is in FP32, it must first be cast to FP16 before
computation, and then cast back to FP32 after computation, which
introduces unnecessary cast operations. Moreover, FP16 computation
requires significantly less workload compared to FP32, leading to
noticeable efficiency improvements.
In this change, `get_rows`, `rms_norm`, and `flash_attn_ext` are extended
to support multiple data types. Validation on the Qwen2 0.5b model shows
correct accuracy and about 10% performance gain in concurrent scenarios.
Co-authored-by: noemotiovon <757486878@qq.com>
* scaffold to support opt step adamw on metal (not written so far)
* add opt-step-adamw kernel for metal
* pass op->src[4] as a separate buffer to the pipeline
* add bounds check to opt-step-adamw kernel
* complete scaffold for GGML_OP_SUM
* naive GGML_OP_SUM kernel
* remove unwanted comment
* change OP_SUM capability gate
* Add has_simdgroup_reduction to both ops to pass CI
* fix/refactor OP argsort, pad
* fix count-equal op
* update SYCL OP list
* fix format issue
---------
Co-authored-by: Zhang Jianyu <zhang.jianyu@outlook.com>
The previous SVE implementation for `ggml_vec_dot_f16_unroll` contained a bug due to a copy-paste error. The wrong variable was used in an FMA instruction, leading to incorrect results. This commit corrects the variable usage and improves the clarity of the code by renaming variables to avoid confusion.
Co-authored-by: Aaron <shelhamer.aaron@gmail.com>
* CANN: improve ACL graph matching
Record `ne` and `nb` information for src tensors and include them in the
graph matching check. This enhances the robustness of ACL graph matching
by preventing incorrect matches when src tensors share the same data
address but differ in shape or stride.
* CANN: add op_params match
* refactor to support soft_max_ext
* fix error and support soft_max_back
* rm unused functions
* fix format issue
---------
Co-authored-by: Zhang Jianyu <zhang.jianyu@outlook.com>
* metal : better unroll in the FA kernels
* metal : index FA blocks
* tests : restore [no ci]
* metal : prevent division by zero in FA kernels
* metal : fix -INF detection logic
* Add profiling
* More detailed profiling
* Rework command submission to avoid global locks
* Update wait handling
* try new method of waiting on futures
* Add serializing of command submission in some cases
* Add new pool for timestamp queries and clean up logging
* Serialize command submission in CI and leave a TODO note
* Update webgpu CI
* Add myself as WebGPU codeowner
* Deadlock avoidance
* Leave WebGPU/Vulkan CI serialized
* Fix divide by 0
* Fix logic in division by inflight_threads
* Update CODEOWNERS and remove serialize submit option
* metal : pad K, V and Mask when needed
* cont : simplify
* cuda : add TODO about KV padding requirement
* metal : add comments
* metal : remove mask padding requirement
* tests : add -INF blocks to the KQ mask in the FA tests
* cont : bump -INF block size to 64
Co-authored-by: Jeff Bolz <jbolz@nvidia.com>
* ggml : prevent division by zero in FA CPU op
---------
Co-authored-by: Jeff Bolz <jbolz@nvidia.com>
* metal : ssm_scan minor opts
* metal : get_rows optimize
* metal : cpy optimize
* metal : ssm_conv opt
* metal : ssm_scan simplify
* metal : ssm_Scan opt
This commit updates the leftover handling in ggml_vec_scale_f32.
The motivation for this is that the code currently incorrectly assumes
there would be fewer than ggml_f32_epr leftover elements. However,
since the main loop processes 2*ggml_f32_epr elements per iteration
, there can be up to (2*ggml_f32_epr - 1) leftover elements.
The original single-pass leftover code could only process ggml_f32_epr
elements, leaving some elements unscaled.
Example scenario with 256-bit SVE:
```
ggml_f32_epr = 8 (elements per register)
ggml_f32_step = 16 (two registers per iteration)
n = 25
np = 16
leftovers = 9 elements (16-24)
Original : processes only elements 16-23, misses element 24
This commit : loop processes elements 16-23, then element 24
```
Refs: https://github.com/ggml-org/llama.cpp/actions/runs/18070620247/job/51419855630
* rpc : add support for multiple devices
Allow rpc-server to expose multiple devices from a single endpoint.
Change RPC protocol to include device identifier where needed.
closes: #15210
* fixes
* use ggml_backend_reg_t
* address review comments
* fix llama-bench backend report
* address review comments, change device naming
* fix cmd order
* vulkan (DRAFT): split shader generation by GLSL source file, to improve incremental build times
* support dep-files so shaders are recompiled if their included files change
* rename shader files which are used as "headers" to use .glsl extension
* move glslc extension detection shaders to separate folders
* the above is to prevent them from getting glob'd with the actual compute shaders that need to be compiled
* vulkan : only write embedded shader .hpp/.cpp when they change
* avoid recompiling ggml-vulkan.cpp when editing shaders
* pass single --source argument instead of --input-dir & --filter to shader gen
* check for source file match earlier
* fix hang in vulkan-shaders-gen when there are compilation errors
* early out did not decrement compile_count
* clean up
* fix glslc integer dot product test
* unconditionally write the embedded shader cpp output
* replace output filepath in generated dep-files to match output in CMakeLists
---------
Co-authored-by: Jeff Bolz <jbolz@nvidia.com>
* vulkan: Replace uses of maxMemoryAllocationSize and VK_WHOLE_SIZE
Replace maxMemoryAllocationSize check with maxBufferSize when creating buffers.
The maxMemoryAllocationSize limit is a "soft" limit and allocations can succeed
beyond that limit. This allows > 4GB buffers to be allocated on some
implementations (e.g. NVIDIA) and tensors this large can be used for im2col
and mul_mat.
For temporary buffers (prealloc_x/y/etc) check against maxStorageBufferRange.
I'm not sure this check is ideal, but we always use these buffers as a single
full size binding and the limit may be smaller than maxMemoryAllocationSize
or maxBufferSize, so I think this is reasonable.
Replace descriptor range uses of VK_WHOLE_SIZE with a manually computed range.
The maxStorageBufferRange may be smaller than the maxBufferSize or
maxMemoryAllocationSize (and the Vulkan spec warns about this in a note) and
it's invalid usage if VK_WHOLE_SIZE computes a range larger than
maxStorageBufferRange.
With this change, it should be possible to generate videos using wan networks
in stable-diffusion.cpp.
* vulkan: Add env var GGML_VK_FORCE_MAX_BUFFER_SIZE and use stoull
When computing sinks, the cm1 shader was looping r from 0 to Br rather than
to rows_per_thread. I must have copied this from the scalar path (where it is
correct), and somehow it wasn't causing failures on current drivers.
* First attempt
* No permute during convert (fixes qk tensors), proper norm application.
* RoPE = NeoX
* Coherence!
* Migrate xielu params from tensors to hyperparameters
* Simple CUDA kernel
* Revert stupid LLM refactorings
* Chat template support
* configchecker / flake8 errors
* Reorder unary.cu
* I do conclude that LLMs are, in fact, stupid.
* Fix after merge
* Final newline
* Make xIELU an UNARY_OP
* Final newline
* Correctly account for parameter shift
* Argh.
* Update ggml/src/ggml-cpu/unary-ops.cpp
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Refactor: remove unused methods, inline and factorize softplus, add const modifiers
* Revert CUDA changes, implement xIELU as a separate OP
* Pesky newline
* Add float2half / half2float for F16 inputs/outputs
* CUDA variants, attempt 2
* Actually, attempt 3
* Update ggml/src/ggml-cuda/unary.cu
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Missing convert header
* Proper formula and reference for xIELU in the comments.
* Modify unary-ops.cpp to add the functor-based logic besides the template system to retain optimizations
* Apply suggestions from code review
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Add tensor mappings for Apertus to global list instead
* Fix lazy on scalars
* Update ggml/src/ggml-cuda/unary.cu
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Add comment about the constraints on positive/negative alpha
* Change `softplus` to `ggml_softplus`
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* HIP: Disable ROCWMMA fatt on CDNA when compiled against ROCWMMA 2.0.0
rocwmma 2.0.0 includes a bug in the code fakeing fp16 accumulation on CDNA
* CUDA: Fix volta condition in ggml_cuda_should_use_wmma_fattn
* Work on rope
* Simplify inplace operation generation and combine mul/add generation
* Work on rope variants
* implement neox rope
* rope complete
* Add sub,div,glu operators
* implement scale op
* Update cpy shader to handle cont/more types
* formatting
* Update test vars printing for rope,rms_norm
* Avoid ROPE hardcoded constants
* Add TODO to change ROPE constants to enum
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* fix TODO comment
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This PR adds additional information to an error message when loading backend library via ld_load_library() fails. This helps spotting why backend library did not load (missing library, missing dependency or unresolved symbol etc.).
* vulkan: 64-bit im2col
Add variants of the im2col shaders that use buffer_device_address/buffer_reference,
and use 64-bit address calculations. This is needed for large convolutions used in
stable-diffusion.cpp.
* fix validation error for large im2col
* metal : support mul_mm with src1->type == GGML_TYPE_F16
* metal : support mul_mm_id with src1->type == GGML_TYPE_F16
[no ci]
* metal : mul_mm support ne00 % 32 != 0
* metal : support mul_mm_id with ne00 % 32 != 0
* cont : remove unnecessary unrolls
* cont : simplify data loading
* metal : optimize mul_mm when output bounds checks are not needed
* vulkan: handle mat_mul with A matrix > 4GB
This change splits mat_mul operations with huge A matrix into chunks in the M
dimension. This works well for stable-diffusion use cases where the im2col
matrix has very large M.
Fix the order of setting the stride in mul_mm_cm2 - setting the dimension
clobbers the stride, so stride should be set after.
* build fixes
The "Clamp" spec constant is already based on whether KV is a multiple of Bc,
so use that to control whether bounds checking is performed. Add bounds checking
to the scalar and coopmat1 paths. Coopmat2 didn't need any changes (the K/V
tensors are already optionally clamped, nothing else needed to be changed).
* CUDA: mul_mat_id for mmf for bs <= 64 for f16 and bs <= 32 for f32
This commit adds mul_mat_id support for ncols_dst >= 16. It does this by
packing ncols_dst tiles into the blockDim.y.
My tests on a RTX 3090 show that this is faster than the cuBLAS fallback
for f16 till bs=64, and for f32 till bs=32
* Review: refactor if statement
The dequantize functions are copy/pasted from mul_mm_funcs.comp with very few
changes - add a_offset and divide iqs by 2. It's probably possible to call
these functions from mul_mm_funcs and avoid the duplication, but I didn't go
that far in this change.
* devops: move s390x and ppc64le ci build
we have access to ubuntu-24.04-s390x and ppc64le images now
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: disable ppc64le for now since they have compiler errors
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: stop warnings as errors
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: switch to non-macro flag
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: going the llama macro route
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: add big-endian gguf test models
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: disable ppc64le to test s390x, check test build
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: dup .gguf.inp files for big-endian tests
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: dup .gguf.out files for big-endian too
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: add python setup and endian byteswap
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: pooring thing does not have s390x python3
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: add missing rust compiler for s390x
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: try rust actions runner
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Revert "devops: try rust actions runner"
This reverts commit 3f8db04356033d6c1d7eccc75ca396bc5298250c.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: try a different path for rust
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: dump home directory and user info
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: install gguf-py only
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: missed relative path
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: remove big-endian files since local swapping is working
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: revert test-tokenizer-0 cmakelists
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Fix unicode flags conversion from and to uint16_t
Bitfields are allocated in different order on s390x
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Simplify byteswap command
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Add byteswapping and git-lfs for test-tokenizers-ggml-vocabs
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Fix endianness detection in vocab loader
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Disable test-thread-safety on s390x
In this test a model is downloaded,
then immediately loaded to check if more downloads are needed,
and then used for test.
There is no clean way to separate all those steps
to add byteswapping between them, so just skip this test.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Fix q8_0 test in test-quantize-fns
vec_signed uses unexpected rounding mode.
Explicitly use different rounding function.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: add big-endian stories260K
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: add s390x test-eval-callback
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: fix test does not exist
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: fix model not found llama-eval-callback
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Fix q3_K dot product error in test-quantize-fns on s390x
Array q8bytes had only 4 elements allocated, but 8 elements accessed.
This lead to write out of bounds and later read of overwritten values out of bounds
and incorrect result.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: re-enable ppc64le for testing
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: activate test-thread-safety for s390x
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: disable ppc64le tests
for some reason it keeps failing test-thread-safety tests and I do not
have a machine that is able to replicate the tests.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* devops: LLAMA_FATAL_WARNINGS=ON
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Correct repository URL for s390x for test-thread-safety model
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Fix fs_get_cache_directory
Ensure it works even if both XDG_CACHE_HOME and HOME are unset.
This might happen in containers.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Re-enable CI for ppc64le
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Fortify ggml_rope_impl
Only memcpy data from sections argument if it's non-NULL.
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
* Add TODO in struct unicode_cpt_flags to reimplement it in endian-independent way
* Update URL for big-endian model
* Update .github/workflows/build.yml
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* Update remaining mentions of BE models to ggml-org/models repo
---------
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
Co-authored-by: Aleksei Nikiforov <aleksei.nikiforov@linux.ibm.com>
Co-authored-by: Aleksei Nikiforov <103434461+AlekseiNikiforovIBM@users.noreply.github.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* vendor : update httplib
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
* common : use cpp-httplib as a cURL alternative for downloads
The existing cURL implementation is intentionally left untouched to
prevent any regressions and to allow for safe, side-by-side testing by
toggling the `LLAMA_CURL` CMake option.
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
* ggml : Bump to Windows 10
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
---------
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
* CUDA: add a fused top-K MoE kernel
This kernel does the following:
1. softmax over the logits per token [n_experts, n_tokens]
2. argmax reduce over the top-k (n_experts_used) logits
3. write weights + ids to global memory
It is intended as fusion of softmax->top-k->get_rows pipeline for MoE models
* Refactor into ggml_cuda_should_use_topk_moe
* Review: Use better coalescing pattern, use WARP_SIZE, store logits into registers before
* Review: format + micro-optimizations
* Fix bug: fix tie breakers
* Add optional norm + clean-up code
* Use smem for final write
* Add bounds check
* Use better memory pattern for writeback
This commit removes the `-dev` suffix from the version string in
CMakeLists.txt and the release script. The version will now be
just be formatted as `MAJOR.MINOR.PATCH`.
Use RPC_DEBUG environment variable to enable debug messages.
Add helper macro LOG_DBG() which does an early
check of the env var before calling GGML_LOG_DEBUG().
Make sure we log a debug message for every server function.
* ggml : make gallocr respect the backend's max buffer size
* if the graph requires more memory than can fit into a single allocation, split it into multiple backend buffers
* vulkan: report the actual max allocation size in buffer type interface
* fix missing newline, apple-clang warning
* track size of individual chunks in ggml_dyn_tallocr and raise max chunks.
revert to use suballocation_block_size as max chunk size for vulkan.
* track (chunk, offset) pairs instead of "global" offsets through gallocr.
* simpler, don't need loops to map between local/global offsets
* touches more code
* fix dyn_tallocr_max_size and initialization
* fix memory leak when buffers are reused due to same buffer type appearing multiple times
* make vbuffer allocation follow the same logic as backend_buffer did before
* continue to use leftover unallocated space of previous chunks after a new one has been created
* treat free blocks of each chunk as separate list
* they're still allocated together, but start/end of each chunk is tracked, and allocate/free iterate over sub-ranges
* exhaust freed blocks of all chunks before considering their last blocks with unallocated space
* start with 0 chunks/blocks and create chunks as needed
* allow the last chunk to grow beyond max size
* refactor: move adding new free block and new chunk into separate functions
* allocate chunks individually with a separate free-blocks list for each one
* needs a bit more memory/allocations/indirections, but code is simpler
* fix warnings (missing static) & debug checks
* implement set_rows with i32 index
* template fix
* test quantized path
warnings--
* Apply suggestions from code review
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* forgotten name change
* deduplicate cuda/sycl and test-fix
* indent++
* vulkan: support set_rows with i32 index type (llama/16162)
* disable i32 index for webgpu for now
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Jeff Bolz <jbolz@nvidia.com>
* Vulkan: add conv_transpose_2d operation
* Vulkan: fix typo in conv_transpose_2d shader(s0mp, s0L, s1mp, s1L)
* Vulkan: fix incorrect indentation in conv_transpose_2d shader
* Vulkan: add checking the push constants size limit and reuse conv2d_mm.comp for conv_transpose_2d operation
* Vulkan: revert the order of the index calculation and bound check in conv_2d shader
* Vulkan: explicity check push constants limit in supports_op() for conv_transpose_2d operation.
* Vulkan: remove unnecessary lower bound checks for H/W_idx in the conv_2d shader.
* vulkan: optimize UMA buffer operations and fix driver hangs
The previous implementation was blocking the GPU for extended periods,
causing the i915 driver to reset the context due to the hangcheck
protection.
[32628.443070] i915 0000:00:02.0: [drm] GPU HANG: ecode 12:1:85dffffb, in llama-server [194114]
[32628.443091] i915 0000:00:02.0: [drm] llama-server[194114] context reset due to GPU hang
* vulkan: implement deferred_memset on UMA
---------
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
* vulkan: Change the mul_mm shared memory and register caching system to use vec2 instead of scalars, to enable using dot2 instructions
* use fma instead of dot to fix Nvidia and Apple performance issues
Generalize Linux check to `__linux__` to support non-glibc systems (like musl).
Also, return `false` on unknown/untested OS.
Without this commit, the code compiles (with warnings) but fails:
register_backend: registered backend CPU (1 devices)
register_device: registered device CPU (Intel(R) Xeon(R) Platinum 8488C)
build: 6487 (51c4cac6) with x86_64-linux-musl-gcc (GCC) 15.1.0 for x86_64-linux-musl (debug)
system info: n_threads = 8, n_threads_batch = 8, total_threads = 16
....
print_info: n_ctx_orig_yarn = 262144
print_info: rope_finetuned = unknown
print_info: model type = 4B
Illegal instruction (core dumped)
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
When compiling with GGML_STATIC=ON, the build process would produce a
binary that was still dynamically linked to OpenMP. This defeats the
purpose of a static build:
$ cmake -B build \
-DBUILD_SHARED_LIBS=OFF \
-DLLAMA_CURL=OFF \
-DGGML_CCACHE=OFF \
-DGGML_NATIVE=OFF \
-DGGML_STATIC=ON
$ ldd llama-server
linux-vdso.so.1 (0x0000e1a434e3b000)
libgomp.so.1 => /lib/aarch64-linux-gnu/libgomp.so.1 (0x0000e1a4345a0000)
libstdc++.so.6 => /lib/aarch64-linux-gnu/libstdc++.so.6 (0x0000e1a434300000)
libm.so.6 => /lib/aarch64-linux-gnu/libm.so.6 (0x0000e1a434240000)
libgcc_s.so.1 => /lib/aarch64-linux-gnu/libgcc_s.so.1 (0x0000e1a434200000)
libc.so.6 => /lib/aarch64-linux-gnu/libc.so.6 (0x0000e1a434030000)
/lib/ld-linux-aarch64.so.1 (0x0000e1a434df0000)
This commit resolves the issue by modifying `CMAKE_FIND_LIBRARY_SUFFIXES`
to prioritize `.a` files, forcing CMake to link the static version of
the library.
Signed-off-by: Adrien Gallouët <angt@huggingface.co>
- flatten mxfp4 and packed fp4->fp16 bit-wise convert function (replace lut)
- MoE kernel optimizations
---------
Co-authored-by: Li He <lih@qti.qualcomm.com>
* CUDA: Optimize PAD_REFLECT_1D
feat: add more test cases for PAD_REFLECT_1D
* use fast_div to improve performance
* Apply suggestion from JohannesGaessler
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Apply suggestion from JohannesGaessler
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* optimize
* use a concise expression to further speedup the cuda kernel
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Add paramater buffer pool, batching of submissions, refactor command building/submission
* Add header for linux builds
* Free staged parameter buffers at once
* Format with clang-format
* Fix thread-safe implementation
* Use device implicit synchronization
* Update workflow to use custom release
* Remove testing branch workflow
* some f32 tests passing
* Disable set_rows until it's implemented
* f32 add all tests passing
* Begin work on set_rows
* Work on set rows
* Add error buffers for reporting unsupported SET_ROWS indices
* Remove extra comments
* Add templated addition, clean up code
* Get addition and multiplication working
* Implement rms_norm
* Add get_rows implementation
* Add new get_rows files
* Refactor use of wg size entry
* Fix compilation
* Try manually unrolled q4_0 quant
* Revert "Try manually unrolled q4_0 quant"
This reverts commit 77f8b96515f7e640ae4b0e44f066321fbc4a6166.
* Move to constant max wg size
* Check for tensor size in supports_op
* Vectorize f32 and change default workgroup size
* Move f32 get_rows from < 4 to % 4 != 0
* fix linter errors
* Add in-place tests
---------
Co-authored-by: Neha Abbas <nehaabbas@ReeseLevines-MacBook-Pro.local>
* CANN: Fix ggml_cann_set_device to avoid redundant device switches
- Added a check to skip aclrtSetDevice if the current device is already set.
- Prevents unnecessary context switches while keeping thread/device consistency.
* CANN: add device default id
* ggml : remove adding extra dim timestep embedding
This commit updates the ggml_timestep_embedding function to no longer
add an extra dimension when the specified dimension is odd.
The motivation for this change is that this introduces an unnecessary
dimension when the dimension is odd, which caused an issue in the
kernels which were not expecting this extra dimension and it resulted in
uninitialized memory for the second to last dimension.
* ggml-cuda : fix padding in timestep embedding kernel
This commit removes the zeroing out of the last dimension now that we
are not adding the extra padding dimension.
* ggml-metal : fix padding in timestep embedding kernel
This commit fixes the zero padding for odd dimensions in
the timestep embedding kernel
* ggml-opencl : fix padding in timestep embedding kernel
This commit fixes the zero padding for odd dimensions in
the timestep embedding kernel.
* ggml-sycl : fix padding in timestep embedding kernel
This commit fixes the zero padding for odd dimensions in
the timestep embedding kernel.
* ggml-vulkan : fix padding in timestep embedding kernel
This commit fixes the zero padding for odd dimensions in
the timestep embedding kernel.
* ggml-cpu : fix padding in timestep embedding function
This commit removes the zeroing out of the last dimension now that we
are not adding the extra padding dimension.
* fix im2col_3d to respect non-contiguous inputs (views)
The CUDA 3D im2col kernel computed source addresses assuming compact layout (products of dims), ignoring nb[] strides.
This patch switches im2col_3d source indexing to use true strides derived from src1->nb[] (in elements), mirroring the approach used in the 2D CUDA im2col path. Destination indexing is unchanged.
* use ggml_element_size() for src strides
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* SYCL: Add COUNT_EQUAL operator support (rebased on master)
* SYCL: remove duplicate op_count_equal definition
* tests: remove test_count_equal_typed and use test_count_equal for all cases
* tests: keep only I32 case for COUNT_EQUAL as suggested
* tests: keep only I32 case for COUNT_EQUAL as requested
* metal : remove mem pool usage
ggml-ci
* metal : remove mem pool implementation
ggml-ci
* metal : take into account the actual allocated memory of the tensor
ggml-ci
* cont : use ggml_backend_buft_get_alloc_size
ggml-ci
* cont : improve, comments
ggml-ci
* cont : add functions for the extra tensor sizes
* metal : add comments
ggml-ci
* metal : implement .get_alloc_size for the rest of the buffer types
ggml-ci
* metal : remove ggml_metal_heap
ggml-ci
Use this to query register count for shader compiles on NVIDIA. Currently
this is only for performance debug, but it could eventually be used in some
heuristics like split_k.
* metal : refactor bin kernels loading
ggml-ci
* metal : refactor rms kernel loading
ggml-ci
* ci : try to add memory leaks check
ggml-ci
* ci : try to enable memory leak detection for Mac
* cont : seems to be working
* ggml-backend : add GGML_BACKEND_DEVICE_TYPE_IGPU device type
ggml-backend : add device id to device props
llama : only use iGPU devices if there are no GPU devices
llama : do not use multiple devices from different backends with the same device id
This commit adds a check for GGML_MACHINE_SUPPORTS_i8mm when enabling
MATMUL_INT8 features, ensuring that i8mm intrinsics are only used when
the target hardware actually supports them.
The motivation for this is to fix ggml CI build failures where the
feature detection correctly identifies that i8mm is not supported,
adding the +noi8mm flag, but MATMUL_INT8 preprocessor definitions are
still enabled, causing the compiler to attempt to use vmmlaq_s32
intrinsics without i8mm support.
Refs: https://github.com/ggml-org/ggml/actions/runs/17525174120/job/49909199499
Since the prefill length is not fixed, graphs constructed for the
prefill stage cannot be reused. For this reason, ACL graph
execution is disabled by default during prefill.
* Add fastdiv and fastmodulo to k_bin_bcast kernel
* Address review comments
* `prod_` instead of `prod` suffix
* Add test case for `k_bin_bcast_unravel` in CUDA backend
This commit fixes the zero padding for odd dimensions in
ggml_compute_forward_timestep_embedding_f32.
The motivation for this is that currently if an odd dimension is used,
the padding check incorrectly uses the dimension value for indexing.
For example, with dim=15:
Elements 0-6 are set to cosine values
Elements 7-13 are set to sine values
Element 14 is left uninitialized (contains garbage)
Element 15 is correctly set to zero
This fix changes embed_data[dim] to embed_data[2 * half] so that
element 14 (the first unused element) is properly set to zero as well
as the last element.
Resolves: https://github.com/ggml-org/ggml/issues/1324
* CANN: Add ROPE sin/cos cache for reuse
Introduce sin/cos caching mechanism in ROPE to avoid redundant
computation across layers. The cache is built on the first layer
per device and reused by subsequent layers if parameters match.
- Added sin_cache / cos_cache pointers and position_length tracking
- Introduced cache validity flags and properties:
(ext_factor, theta_scale, freq_scale, attn_factor, is_neox)
- Accelerates ROPE by eliminating repeated sin/cos generation
This change reduces overhead in multi-layer scenarios while
preserving correctness by verifying parameter consistency.
Co-authored-by: hipudding <huafengchun@gmail.com>
* fix typo
Signed-off-by: noemotiovon <757486878@qq.com>
---------
Signed-off-by: noemotiovon <757486878@qq.com>
Co-authored-by: hipudding <huafengchun@gmail.com>
* CANN: implement LRU cache for ACL graphs in CANN backend
- Introduce ggml_cann_graph_lru_cache to store multiple ggml_cann_graph objects.
- Graphs are loaded on demand and evicted using LRU policy when capacity is exceeded.
- Updated push, move_to_front, and clear methods to manage cached graphs efficiently.
- Ensures reuse of graphs, reducing graph reconstruction overhead in CANN backend.
* fix typo
* The LRU cache capacity can be configured via an env variable
Signed-off-by: noemotiovon <757486878@qq.com>
* refactory acl graph
* refactory && fix review comments
Signed-off-by: noemotiovon <757486878@qq.com>
---------
Signed-off-by: noemotiovon <757486878@qq.com>
* CUDA: Add mul_mat_id support the mmf
Add support for mul_mat_id for bs < 16
* Review: use warp_size, fix should_use_mmf condition
* Launch one block per expert, stride along n_expert_used
* templatize mul_mat_id
* Pad shmem to 16 bytes, add helper function mul_mat_f_switch_ids
* Reduce compile times by dividing mmf into f16, bf16 and f32 variants
* Divide mmf by ncols_dst
* Add missing files
* Fix MUSA/HIP builds
* vulkan: sort graph to allow more parallel execution
Add a backend proc to allow the backend to modify the graph. The
vulkan implementation looks at which nodes depend on each other
and greedily reorders them to group together nodes that don't
depend on each other. It only reorders the nodes, doesn't change
the contents of any of them.
With #15489, this reduces the number of synchronizations needed.
* call optimize_graph per-split
* cuda : fix supports_op condition for get_rows when src1->ne2 > 1
ggml-ci
* ggml : add comment about ggml_get_rows
ggml-ci
* cuda : add FIXME [no ci]
* cuda : update support condition
ggml-ci
* ggml: allow casting between f32 and i32
* fix cuda
* add vulkan
* fix CPU non-cont
* add non-cont test case
* add note
* extend test number range
* correct note
* add cont version for vulkan
I think glslang will translate an access like x[i][1].z to
OpAccessChain ... x, i, 1, 2
OpLoad float16_t ...
rather than loading all of x[i] in a single OpLoad. Change the
code to explicitly load the vector/matrix.
* ggml WebGPU: remove userdata from request adapter callback
This commit removes the `userdata` parameter from the WebGPU request
adapter callback in `ggml-webgpu.cpp`. Instead, the lambda function
captures the `webgpu_context` directly.
The motivation for this change is to simplify the code and improve
readability.
* inline the callback lambda into the RequestAdapter call
This commit removes the callback lambda variable and inlines it directly
into the RequestAdapter call.
* ggml : introduce semantic versioning
This commit introduces semantic versioning for the GGML library.
The motivation for this is that the current versioning, using build
numbers, makes it difficult to track changes and releases for projects
that use ggml.
The release steps are the following:
1. Sync the changes from llama.cpp using sync-llama-am.sh and after the
PR has been approved and merged move to step 2.
2. Run scripts/release.sh and specify the type of release, major, minor,
or patch. This script will handle incrementing the version
(major|minor|patch), create a new commit with the version change,
create a tag for the version, and prepare for the next development
iteration.
3. Inspect the commits/tag and push to master. This will trigger the
github release workflow which is triggered for new tags which will
then publish a new release on github.
Example usage:
```console
$ ./scripts/release.sh major --dry-run
[dry-run] - No changes will be made
Step 1: Reading current version...
Current version: 0.9.0-dev
New release version: 1.0.0
Step 2: Updating version in CMakeLists.txt...
[dry-run] Would update GGML_VERSION_MAJOR to 1
[dry-run] Would update GGML_VERSION_MINOR to 0
[dry-run] Would update GGML_VERSION_PATCH to 0
[dry-run] Would remove -dev suffix
Step 3: Committing version bump...
[dry-run] Would commit: 'ggml : bump version to 1.0.0'
Step 4: Creating git tag...
[dry-run] Would create tag: v1.0.0 with message 'Release version 1.0.0'
Step 5: Preparing for next development cycle...
[dry-run] Would update GGML_VERSION_MINOR to 1
[dry-run] Would add -dev suffix back
Step 6: Committing development version...
[dry-run] Would commit: 'ggml : prepare for development of 1.1.0-dev'
[dry-run] Summary (no changes were made):
• Would have released version: 1.0.0
• Would have created tag: v1.0.0
• Would have set next development version: 1.1.0-dev
```
Refs: https://github.com/ggml-org/ggml/issues/1333
* ggml: create branch for release candidate and check master
* ggml : sign the git tag
Fixes#15330
Adjust the allocation size of acl_rstd. The parameter `dims` is set to 3 according to the CANN documentation.
Co-authored-by: Yuchuan <yuchuan-cao@users.noreply.github.com>
* vulkan : update ggml_vk_instance_validation_ext_available
This commit updates ggml_vk_instance_validation_ext_available() to
check for VK_EXT_validation_features instead of
VK_KHR_portability_enumeration.
Based on how the returned boolean is used later in the code (to enable
both the validation layer and the VK_EXT_validation_features extension),
it appears the function may have been intended to check for the
validation layer features extension.
* remove try/catch
This was a left over from a previous iteration where I was explicitly
quering for a specific validation layer first, which would throw.
* update warning message about validation layers
* Add fastdiv, use it in modulo and use modulo in rms_norm_f32
Fastdiv is much faster way to do integer division, which was identified
as bottleneck in rms_norm_f32
* Support more `block_size` values in `rms_norm_f32`
This makes us more flexible in selecting the optimal threads w.r.t
paralellizing across a col vs. launch-overheads of threads and mio
throttles
* Update ggml/src/ggml-cuda/common.cuh
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Replace modulo with fastmodulo in `rms_norm_f32`
* Use `BinPackArguments=true` for formating function calls
Will file a separate PR to adjust .clang-format file
* Update ggml/src/ggml-cuda/common.cuh
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Use uint3 for both `fastdiv` and `fastmodulo`
The compiler seems to reliably optimize away the unused .z component in
the fastdiv use-case, see https://godbolt.org/z/rx8KPrKr3
* More constrained type declarations
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Rename fastdiv and fastmodulo variables to shared variable name
As suggest by JohannesGaessler, this increases clarity of the intended
use
* Pack fastdiv/fastmodulo constants into uint2/uint3 objects
By packing constants to be used together into a struct, we are less
likely to make errors.
* Rename function parameter of fastmodulo
`modulo_consts` is more fitting/descriptive
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Previously, the slope tensor was set to fp16 to improve efficiency.
While this worked correctly in FA, it caused precision issues in soft_max.
This change applies different data types for different operators
to balance both accuracy and performance.
* [CANN] Support eager execution mode under ACL graph compilation
Add support for running operators in eager mode while ACL graph
compilation is enabled. This allows bypassing graph execution
and directly submitting ops, which is useful for debugging and
reducing graph build overhead in certain scenarios.
Signed-off-by: noemotiovon <757486878@qq.com>
* fix typo
Signed-off-by: noemotiovon <757486878@qq.com>
* rename to acl_graph_mode
Signed-off-by: noemotiovon <757486878@qq.com>
---------
Signed-off-by: noemotiovon <757486878@qq.com>
* vulkan: use memory budget extension to read memory usage
* fix: formatting and names
* formatting
* fix: detect and cache memory budget extension availability on init
* fix: read `budgetprops.heapBudget` instead of `heap.size` when memory budget extension is available
* style: lints
* SVE support for exponential functions
Add const notation to variable pg
* Update ggml/src/ggml-cpu/vec.cpp
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Add const
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* vulkan: Add Integer Dot Product mul_mat_vec shader for legacy quants
* vulkan: use subgroup operations for quantize_q8_1 shader
* vulkan: add q8_1_x4 type with 128-bit alignment, use in mul_mat_vecq shader
* vulkan: use q8_1_x4 blocks in mul_mmq shader
* vulkan: do 8 calculations per invocation instead of 32 in mul_mat_vecq, similar to mul_mat_vec
* vulkan: tune mul_mat_vecq performance for Intel
* vulkan: fix quantizing issue when tensor is not divisible by 128
* vulkan: adapt integer dot mmv to mmv small m optimization (llama/15355)
* vulkan: allow all subgroup modes for mmv and mmvq
* vulkan: use prealloc intermediate reuse for mmvq path
* vulkan: tune mmvq for Intel, AMD GCN and Nvidia RTX 3090
* vulkan: adapt mmv quantize_y path to conditional sync logic
* vulkan: disable q8_0 mmvq on Nvidia
* vulkan: enable q8_0 on Nvidia pre-turing
* fix prealloc sync condition
* fix llvmpipe subgroup 8 issue
* ggml : WebGPU add TRANSPOSE and RESHAPE to supported ops
This commit adds support for the TRANSPOSE and RESHAPE operations in the
ggml webgpu backend.
Co-authored-by: Diego Devesa <slarengh@gmail.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
* CUDA: fix build error from ambiguous __half conversions in conv2d
Building conv2d with half precision failed because `__half` defines
multiple implicit conversion operators (to float, int, short, etc.),
causing ambiguous overload resolution when multiplying with float.
Introduce a templated `to_float` helper that explicitly converts
`__half` via `__half2float`, while passing through float unchanged.
Use this helper in conv2d accumulation to ensure unambiguous and
correct promotion to float.
Fixes some build errors with half-precision kernels on CUDA.
ggml-ci
* CUDA: Replace custom to_float helper with unified ggml_cuda_cast and add half‑>float conversion
* CUDA: Add missing convert.cuh header
* CUDA: remove unnecessary extension in ggml_cuda_cast
* CUDA: Address review comment, remove second type template argument
* CANN: fix RoPE cache issue on multi-device
RoPE cache only needs to be computed once per token.
However, in multi-device scenarios, not every device starts
computation from layer 0, which may lead to unallocated memory
issues and precision errors.
This commit records the first layer of each device to avoid
the above issues.
* CANN: Optimize first-layer detection method
* CANN: Remove trailing whitespace
* CANN: Only cache the data that can be determined as unchanged through the parameters.
* CANN: Update function comment
Exposes ggml_backend_sched_split_graph() to allow splitting the graph without allocating compute buffers and uses it to split the graph for the automatic Flash Attention check.
* vulkan: mul_mat_id coopmat2 optimizations
Add a path for when the tile fits in BN/2, similar to what we have for mul_mat.
Only call fetch_scales/store_scales once per QUANT_K block, and once at the
beginning in case start_k is not aligned.
* Also add a path for BN/4 - worth a couple more percent
This commit removes the portability_enumeration_ext variable from the
ggml_vk_instance_portability_enumeration_ext_available function as it
is initialized to false but never modified, making it redundant.
Prior to this change, we faced undefined cublasLt references when
attempting to compile 'llama-cli' with GGML_STATIC=ON on Linux.
We add linking with CUDA::cublasLt_static when CUDA version is greater
than 10.1.
* CANN(flash-attn): refactor mask handling and improve performance
1. Refactored the mask computation in Flash Attention, unified the logic without separating prefill and decode.
2. Optimized performance in non-alibi scenarios by reducing one repeat operation.
3. Updated operator management to explicitly mark unsupported cases on 310P devices and when dim is not divisible by 16.
Signed-off-by: noemotiovon <757486878@qq.com>
* [CANN]: fix review
Signed-off-by: noemotiovon <757486878@qq.com>
* [CANN]: Optimization FA BNSD to BSND
Signed-off-by: noemotiovon <757486878@qq.com>
---------
Signed-off-by: noemotiovon <757486878@qq.com>
The original implementation unconditionally returned true for this operation, leading to a failure when the tensor's first dimension (ne[0]) was not a multiple of WARP_SIZE. This caused an GGML_ASSERT(ncols % WARP_SIZE == 0) failure in ggml-sycl/norm.cpp.
This change updates the ggml_backend_sycl_device_supports_op check to correctly return true for GGML_OP_RMS_NORM only when the first dimension of the tensor is a multiple of WARP_SIZE, ensuring the operation can be performed without error.
This patch improves GEMM for FP32 Data Type on PowerPC
Implements GEMM on large blocks with configurable block size mc, nc, kc
(default: 256, 256, 256).
Packing Function optimized to access blocks as per memory layout.
GEMM Optimized to work on larger blocks.
Isolated Packing from GEMM Operations for better MMA utilization.
Verified functionality and correctness uing llama-cli and stand alone
test case (performs matmul and compares final mattrix C result with base).
Minor code refactoring changes:
Replace macro with inline function
Code Indent made consistent with 4 spaces
Performance Testing:
Observed 50% ~ 70% improvement in Prompt Processing Speed mesured using
llama-bench with Meta-Llama3-8B FP32 Model. Similar gains observed with
Mistral-7b-Instruct-v0.3 Model.
model Size Params Backend Threads Test Patch Base
llama 8B all F32 29.92 GiB 8.03 B CPU 20 pp512 98.58 60.3
llama 8B all F32 29.92 GiB 8.03 B CPU 20 pp1024 95.88 57.36
llama 8B all F32 29.92 GiB 8.03 B CPU 20 pp2048 85.46 53.26
llama 8B all F32 29.92 GiB 8.03 B CPU 20 pp4096 68.66 45.78
llama 8B all F32 29.92 GiB 8.03 B CPU 20 pp6144 57.35 40.44
25 ~ 30% improvement in llama-batched-bench with Metla-Llama3-8B in
Prompt Processing Speed for large prompts (256, 512, 1024, 2048, 4096)tokens with various batch
sizes ( 1, 2, 4, 8, 16)
Signed-off-by: Shalini Salomi Bodapati <Shalini.Salomi.Bodapati@ibm.com>
* metal : optmize FA vec for large heads and sequences
* metal : adjust small-batch mul mv kernels
ggml-ci
* batched-bench : fix total speed computation
ggml-ci
* cont : add comments
ggml-ci
* metal : mul_mm_id remove hdst
* metal : remove mul_mm_id hsrc1
* metal : mul_mm_id simplify + add test
* metal : opt mul_mm_id map0
* metal : optimize mul_mm_id id gathering
* metal : mul/div opt
* metal : optimize mul_mm_id_map0
ggml-ci
* CUDA: optimize get_int_from_table_16
* CUDA: use v_perm_b32 to replace byte_perm on AMD GPUs
* revise documentation
---------
Co-authored-by: xix <xiapc@outlook.com>
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* vulkan: use subgroup function for mul_mat_id shader even without coopmat
* vulkan: fix compile warnings
* vulkan: properly check for subgroup size control and require full subgroups for subgroup mul_mat_id
* vulkan: disable subgroup mul_mat_id on devices with subgroups < 16
Herman Semenoff
Johannes Gäßler
Georgi Gerganov
Reese Levine
Jeff Bolz
xiaobing318
Adrien Gallouët
TianHao324
Daniel Bevenius
Aman Gupta
Neo Zhang Jianyu
Tarek Dakhran
Gilad S.
Mahekk Shaikh
Acly
Neo Zhang
ixgbe
Ruben Ortlam
Diego Devesa
R0CKSTAR
Piotr Wilkin (ilintar)
Radoslav Gerganov
yulo
matt23654
Alberto Cabrera Pérez
lhez
xctan
Jiacheng (Jason) Chen
hipudding
TianHao324
Max Krasnyansky
Raul Torres
M. Mediouni
nullname
Sigbjørn Skjæret
Scott Fudally
sudhiarm
Giuseppe Scrivano
Jeremy Rand
Chenguang Li
YangLe
Pavels Zaicenkovs
shaofeiqi
shani-f
Alberto Cabrera Pérez
bagheera
TecJesh
Eve
duduta
Charles Xu
Mike Abbott
fj-y-saito
SavicStefan
Aleksei Nikiforov
bssrdf
iron
YehuditE
l3utterfly
Noah
theo77186
Jinyang He
mnehete32
Aaron Teo
Oliver Simons
Masato Nakasaka
JJJYmmm
YaelLogic
YaelGitAccount
tamarPal
leejet
Matthew Michel
sirus20x6
safranowith
Shawn Gu
muggle-stack
Ilia Ilmer
GittyBurstein
takuya kodama
yael-works
Sam/Samuel
Julius Tischbein
Anav Prasad
Jie Fu (傅杰)
Prajwal B Mehendarkar
ai-fonsi
uvos
alex-spacemit
Rafal Lewczuk
Dmytro Minochkin
junchao-zhao
Xiangyan Sun
Shin-myoung-serp
Xuan-Son Nguyen
Bowen Han
Jhen-Jie Hong
Jake Karnes
Mathieu Baudier
lksj92hs
Gregor Jasny
Gabe Goodhart
rmatif
s-goto-11
Prashant Vithule
Akarshan Biswas
compilade
matiaslin
shalinib-ibm
Yoshi_likes_e4
Qeeweew
Ihar Hrachyshka