* feat(cann): add optional support for ACL Graph execution
This commit adds support for executing ggml computational graphs using
Huawei's ACL graph mode via the USE_CANN_GRAPH flag. The support can be
enabled at compile time using the CMake option:
-DUSE_CANN_GRAPH=ON
By default, ACL graph execution is **disabled**, and the fallback path
uses node-by-node execution.
Key additions:
- CMake option to toggle graph mode
- Graph capture and execution logic using
- Tensor property matching to determine whether graph update is required
- Safe fallback and logging if the environment variable LLAMA_SET_ROWS
is unset or invalid
This prepares the backend for performance improvements in repetitive graph
execution scenarios on Ascend devices.
Signed-off-by: noemotiovon <757486878@qq.com>
* Fix review comments
Signed-off-by: noemotiovon <757486878@qq.com>
* remane USE_CANN_GRAPH to USE_ACL_GRAPH
Signed-off-by: noemotiovon <757486878@qq.com>
* fix typo
Signed-off-by: noemotiovon <757486878@qq.com>
---------
Signed-off-by: noemotiovon <757486878@qq.com>
* 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
* Disable set_rows until it's implemented
* Fix potential issue around empty queue submission
* Try synchronous submission
* Try waiting on all futures explicitly
* Add debug
* Add more debug messages
* Work on getting ssh access for debugging
* Debug on failure
* Disable other tests
* Remove extra if
* Try more locking
* maybe passes?
* test
* Some cleanups
* Restore build file
* Remove extra testing branch ci
* cmake: Add GGML_BACKEND_DIR option
This can be used by distributions to specify where to look for backends
when ggml is built with GGML_BACKEND_DL=ON.
* Fix phrasing
* Add parameter 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
- Increase tile size for k-quants, to match non-k-quants
- Choose more carefully between large and medium tiles, considering how it
interacts with split_k
- Allow larger/non-power of two split_k, and make the splits a multiple of 256
- Use split_k==3 to when >1/2 and <=2/3 of the SMs would hae been used
* vulkan: optimizations for direct convolution
- Empirically choose a better tile size. Reducing BS_K/BS_NPQ helps fill
the GPU. The new size should be amenable to using coopmat, too.
- Fix shmem bank conflicts. 16B padding should work with coopmat.
- Some explicit loop unrolling.
- Skip math/stores work for parts of the tile that are OOB.
- Apply fastdiv opt.
- Disable shuffles for NV.
* Three tiles sizes for CONV_2D, and a heuristic to choose
* reallow collectives for pre-Turing
* make SHMEM_PAD a spec constant
* fixes for intel perf - no shmem padding, placeholder shader core count
* shader variants with/without unrolling
* 0cc4m's fixes for AMD perf
Co-authored-by: 0cc4m <picard12@live.de>
---------
Co-authored-by: 0cc4m <picard12@live.de>
* Initial Q2_K Block Interleaving Implementation
* Addressed review comments and clean up of the code
* Post rebase fixes
* Initial CI/CD fixes
* Update declarations in arch-fallback.h
* Changes for GEMV Q2_K in arch-fallback.h
* Enable repacking only on AVX-512 machines
* Update comments in repack.cpp
* Address q2k comments
---------
Co-authored-by: Manogna-Sree <elisetti.manognasree@multicorewareinc.com>
The pipeline member can be cast to VkPipeline.
This is a VkPipeline_T* on 64 bit but a uint64_t on 32 bit.
Cf. VK_DEFINE_NON_DISPATCHABLE_HANDLE documentation.
This is useful for testing for regressions on GCN with CDNA hardware.
With GGML_HIP_MMQ_MFMA=Off and GGML_CUDA_FORCE_MMQ=On we can conveniently test the GCN code path on CDNA. As CDNA is just GCN renamed with MFMA added and limited use ACC registers, this provides a good alternative for regression testing when GCN hardware is not available.
llvm with the amdgcn target dose not support unrolling loops with conditional break statements, when those statements can not be resolved at compile time. Similar to other places in GGML lets simply ignore this warning.
* remove redundant code in riscv
* remove redundant code in arm
* remove redundant code in loongarch
* remove redundant code in ppc
* remove redundant code in s390
* remove redundant code in wasm
* remove redundant code in x86
* remove fallback headers
* fix x86 ggml_vec_dot_q8_0_q8_0
* SYCL: Add set_rows support for quantized types
This commit adds support for GGML_OP_SET_ROWS operation for various
quantized tensor types (Q8_0, Q5_1, Q5_0, Q4_1, Q4_0, IQ4_NL) and BF16
type in the SYCL backend.
The quantization/dequantization copy kernels were moved from cpy.cpp
to cpy.hpp to make them available for set_rows.cpp.
This addresses part of the TODOs mentioned in the code.
* Use get_global_linear_id() instead
ggml-ci
* Fix formatting
ggml-ci
* Use const for ne11 and size_t variables in set_rows_sycl_q
ggml-ci
* Increase block size for q kernel to 256
ggml-ci
* Cleanup imports
* Add float.h to cpy.hpp
This commit adds support for MFMA instructions to MMQ. CDNA1/GFX908 CDNA2/GFX90a and CDNA3/GFX942 are supported by the MFMA-enabled code path added by this commit. The code path and stream-k is only enabled on CDNA3 for now as it fails to outperform blas in all cases on the other devices.
Blas is currently only consistently outperformed on CDNA3 due to issues in the amd-provided blas libraries.
This commit also improves the awareness of MMQ towards different warp sizes and as a side effect improves the performance of all quant formats besides q4_0 and q4_1, which regress slightly, on GCN gpus.
* feat: Add s_off as a parameter in the args struct
This may not be necessary, but it more closely mirrors the CUDA kernel
Branch: GraniteFourPerf
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* perf: Parallelize mamba2 SSM_SCAN metal kernel over d_state
This is a first attempt at optimizing the metal kernel. The changes here
are:
- Launch the kernel with a thread group of size d_state
- Use simd groups and shared memory to do the summation for the y
computation
When tested with G4 tiny preview, this shows roughly a 3x speedup on
prefill and 15% speedup on decode.
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Update logic to correctly do the multi-layer parallel sum
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Correctly size the shared memory bufer and assert expected size relationships
Branch: GraniteFourPerf
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* refactor: Compute block offsets once rather than once per token
Branch: GraniteFourPerf
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Use local variable for state recursion
Branch: GraniteFourPerf
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Use a secondary simd_sum instead of a for loop
Branch: GraniteFourPerf
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add assertion and comment about relationship between simd size and num simd groups
Branch: GraniteFourPerf
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Parallelize of d_state for mamba-1
Branch: GraniteFourPerf
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Parallel sum in SSM_CONV
Branch: GraniteFourPerf
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* Revert "feat: Parallel sum in SSM_CONV"
After discussion with @compilade, the size of the parallelism here is
not worth the cost in complexity or overhead of the parallel for.
https://github.com/ggml-org/llama.cpp/pull/14743#discussion_r2223395357
This reverts commit 16bc059660.
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* refactor: Simplify shared memory sizing
Branch: GraniteFourPerf
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-Authored-By: Georgi Gerganov <ggerganov@gmail.com>
---------
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Neither "g" nor "x" are valid portPos specifiers per the official
[graphviz documents](https://graphviz.org/docs/attr-types/portPos/):
> If a compass point is used, it must have the form "n","ne","e","se","s","sw","w","nw","c","_".
I tested locally for it to fall back to default portPos specifier if an
invalid portPos is specified. As a consequence, we can remove associated
code.
* musa: apply mublas API changes
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
* musa: update musa version to 4.2.0
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
* musa: restore MUSA graph settings in CMakeLists.txt
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
* musa: disable mudnnMemcpyAsync by default
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
* musa: switch back to non-mudnn images
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
* minor changes
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
* musa: restore rc in docker image tag
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
---------
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
* CMake config: Create target only once
Fix error on repeated find_package(ggml).
For simplicity, check only for the top-level ggml::ggml.
* CMake config: Add CUDA link libs
* CMake config: Add OpenCL link libs
* CMake config: Use canonical find_dependency
Use set and append to control link lib variables.
Apply more $<LINK_ONLY...>.
* CMake config: Wire OpenMP dependency
This commit removes the inclusion of `<cstdlib>`.
The motivation for this change is that this source file does not seem to
use any functions from this header and the comment about `qsort` is a
little misleading/confusing.
* weight format to nz for 310p
* remove quant weight format to nz
* clean code
* fix
* make the conditions for converting weights to NZ format consistent
* clean code
* ggml/ggml-vulkan/test-backend-ops: adds CONV_2D for Vulkan
* ggml-vulkan: adds f32 scalar shader to compute 2D convolution directly
with gemm (no need for im2col),
* test-backend-ops: adds test_case_ref to check the validity/performance of ops
against reference implementations having different graphs, adds tests
* * Performance fixes: minimized branch divergence, uses collectives to
eliminate redundant calculation, macros removed.
* Kernel shared memory size check
* Updates test-backend-ops to support graphs for performance
measurement.
* * Apple/Win32 compile errors fixed
* Subgroup size used to determine tile size -> fixes llvmpipe errors.
* Collectives disabled by default.
* Intel support is disabled as the performance is poor.
* Conv2d enabled for Intel with disabled collectives, disabled for Apple
* test-backend-ops modifications are reverted
* Trailing spaces and missing override fixed.
* Triggering pipeline relaunch.
* Code formatted with .clang-format.
* Fix Gemma3n not executed as CUDA_GRAPH on NVGPUs
Gemma3n uses Matrix-Matrix addition as part of their input processing,
wrongly triggering CUDA_GRAPH disablement on NVGPUs even when batch-size
of 1 is used.
* Exclude `project_per_layer_input` by matching node names
This ensures that all other graphs which don't exhibit this pattern do
not have their behavior changed.
* Revert unnecessary formatting changes
* Minimal setup of webgpu backend with dawn. Just prints out the adapter and segfaults
* Initialize webgpu device
* Making progress on setting up the backend
* Finish more boilerplate/utility functions
* Organize file and work on alloc buffer
* Add webgpu_context to prepare for actually running some shaders
* Work on memset and add shader loading
* Work on memset polyfill
* Implement set_tensor as webgpu WriteBuffer, remove host_buffer stubs since webgpu doesn't support it
* Implement get_tensor and buffer_clear
* Finish rest of setup
* Start work on compute graph
* Basic mat mul working
* Work on emscripten build
* Basic WebGPU backend instructions
* Use EMSCRIPTEN flag
* Work on passing ci, implement 4d tensor multiplication
* Pass thread safety test
* Implement permuting for mul_mat and cpy
* minor cleanups
* Address feedback
* Remove division by type size in cpy op
* Fix formatting and add github action workflows for vulkan and metal (m-series) webgpu backends
* Fix name
* Fix macos dawn prefix path
Remove un-necessary templates from class definition and packing functions
Reduce deeply nested conditionals, if-else switching in mnapck function
Replace repetitive code with inline functions in Packing functions
2 ~ 7% improvement in Q8 Model
15 ~ 50% improvement in Q4 Model
Signed-off-by: Shalini Salomi Bodapati <Shalini.Salomi.Bodapati@ibm.com>
* CUDA: add set rows for f32 and f16
* Review: change kernel params, use strides from host
* Use 1-d kernel
* Review: use int64_t for blockDim.x, rename nb->s for clarity
* vulkan: support SET_ROWS
Add variants of the copy_to_quant shader that do the SET_ROWS operation.
Change these shaders to spread the work across the workgroup.
The memory access pattern is probably not great (one thread per quant block),
but should be fine for now.
* vulkan: optimize set_rows
Larger workgroups for non-quant types.
Set "norepeat" (there is manual repeat logic).
Use fastmod.
* vulkan: allow unclamped loads in coopmat2 mul_mat_id shader
* vulkan: increase coopmat2 mul_mat_id tile size
* vulkan: optimize mat_mul_id row_ids search to batch loads, and port to coopmat1 path
* vulkan: use smaller FA row size when head size is large. applies to both scalar and CM2 paths (CM1 isn't used due to shared memory limits)
* ggml : add ggml_scale_bias
* ggml_vec_mad1_f32
* add more simd
* add CUDA
* sycl
* vulkan
* cann (placeholder)
* opencl
* will this fix cpu?
* fix cuda
* suggestions from coderabbit
* fix cann compile error
* vDSP_vsmsa
* rm __ARM_FEATURE_SVE
* use memcpy for op params
* make code looks more consistent
* use scalar for __ARM_FEATURE_SVE
* add x param to ggml_vec_mad1_f32
* vulkan: allow FA split_k with smaller KV values
* vulkan: spread split_k_reduce work across more threads
k_num can get rather large. Use the whole workgroup to reduce the M/L values.
Launch a thread for each element in the HSV dimension of the output. Helps a
lot for large HSV (like deepseek).
The fused operation was grabbing the epsilon value from the wrong place.
Add an env var to disable fusion.
Add some missing checks for supported shapes/types.
Handle fused rms_norm+mul in check_results.
* vulkan: Handle updated FA dim2/3 definition
Pack mask boolean and n_head_log2 into a single dword to keep the push
constant block under the 128B limit.
* handle null mask for gqa
* allow gqa with dim3>1
* kv-cache : use ggml_set_rows
ggml-ci
* graph : separate k and v indices
ggml-ci
* cont : remove redundant ifs
ggml-ci
* kv-cache : improve find_slot impl
* kv-cache : bounds-check when accessing slot_info indices
* kv-cache : add comments
ggml-ci
* ggml : add TODOs for adding GGML_OP_SET_ROWS support in the backends
ggml-ci
* llama : initial Mamba-2 support
* ggml : SIMD ggml_ssm_scan for Mamba-2
* ggml : improve ggml_mul speed when masking recurrent states
* llama : support running Mamba-Codestral-7B-v0.1
* llama : fix Mamba-2 conv state saving
* ggml : make the ggml_mul fast broadcast path more consistently formatted
* llama : remove unused variable
* llama : add missing break
* convert_hf : prefer SentencePiece tokenizer for Mamba-2 when present
The tokenzier.json of Mamba-Codestral-7B-v0.1 otherwise requires
workarounds to work correctly.
* llama : avoid redundant state copy for Mamba 1 and 2
* metal : attempt to adapt SSM_SCAN for Mamba-2
* metal : fix SSM_SCAN pipeline scope
* metal : use log and exp instead of log1pf and expf in SSM_SCAN
* metal : remove unused arguments for SSM_SCAN
The max index is 31, so trimming the arguments is necessary.
* metal : add back n_seqs to SSM_SCAN args
Whoops, this is needed for the offset in the concatenated output.
* metal : fix SSM_SCAN state head offset
* metal : fix wrong number of tokens per sequence in SSM_SCAN
* ggml : remove unused fast broadcast path in GGML_MUL
This was initially added because states were masked with ggml_mul,
but this is no longer done and so this "optimisation" is no longer
necessary, or at least not worth the additional code complexity.
* ggml : avoid multiply by D in GGML_OP_SSM_SCAN
This makes the weight buft detection in src/llama.cpp simpler.
* convert : transpose Mamba-2 A, D and reshape SSM_NORM
This breaks existing conversions of Mamba-2 models
to avoid some reshapes.
Not sure if it's a good idea,
but it makes the graph slightly cleaner.
* llama : more appropriate SSM_SCAN and SSM_CONV buft support checks
* convert : fix flake8 lint
* metal : fix confusion between ; and ,
* metal : add missing args for nb references in ssm_scan_f32_group
* metal : single-user mamba2 inference works
* kv-cache : remove const_cast when setting inputs for s_copy
And also fix multi-user inference for recurrent models
by using cell_id instead of i as the kv cell index
when populating s_copy.
* convert : avoid AutoConfig for Mamba and Mamba2 hparams
* kv-cache : allow context shift for recurrent models
* graph : fix recurrent state copies when avoiding copies
Works, but using lambda functions might not be that clean.
* ggml : fix mamba2 ssm scan when compiled with SVE
* ggml-cpu : reorder SVE FMA for consistency with other SIMD arches
* cuda : implement ssm scan for Mamba2
There is still room for improvement, but it works!
* cuda : adapt Mamba1 ssm scan to shape changes from Mamba2
* mamba : fix mismatched new and delete size for llm_build_mamba
Subclasses of llm_graph_context cannot have extra fields,
because the called destructor is not the one from the subclass.
This otherwise would cause problems when runnning Mamba-(1|2) inference
when compiled -DGGML_SANITIZE_ADDRESS=ON
* cuda : graceful fallback for Mamba-1 models with weird embd size
* ggml : add version function to get lib version
This commit adds a function `ggml_version()` to the ggml library that
returns the version of the library as a string.
The motivation for this is that it can be useful to be able to
programmatically check the version of the ggml library being used.
Usage:
```c
printf("GGML version: %s\n", ggml_version());
```
Output:
```console
GGML version: 0.0.2219
```
* ggml : add ggml_commit()
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* CUDA: add softmax broadcast
* Pass by const ref
* Review: Use blockDims for indexing, remove designated initializers
* Add TODO for noncontigous input/output
* Add a callback that will be called just before abort. This allows apps without a console to display a message to the user and save data if needed.
* Return previous callback to allow callback chaining
* style fixes
---------
Co-authored-by: Diego Devesa <slarengh@gmail.com>
* add "align corners" mode for bilinear upscale, and allow downscaling
* add ggml_interpolate, deprecate ggml_upscale_ext, pass in align-corners as bit-flag
* test-backend-ops: replace ggml_upscale_ext with ggml_interpolate, add test cases for downscale and align-corners
This commit renames the variable `best_mad` to `best_error` in the
`make_qkx2_quants` function.
The motivation for this is that the name `best_mad` can be somewhat
confusing if mean absolute deviation (MAD) is not in use.
* Conv2D: Add CPU version
* Half decent
* Tiled approach for F32
* remove file
* Fix tests
* Support F16 operations
* add assert about size
* Review: further formatting fixes, add assert and use CPU version of fp32->fp16
* Update docker.yml
修改docker.yml文件中的内容使其停止周期性的运行该workflow，如果想要运行该workflow可以手动启动
* Remove redundant include path in CMakeLists.txt
The parent directory '..' was removed from the include directories for the ggml-cpu-feats target, to avoid unnecessary include paths.
* Enable scheduled Docker image builds
Uncomments the workflow schedule to trigger daily Docker image rebuilds at 04:12 UTC, improving automation and keeping images up to date.
* SYCL: disable faulty fp16 CPU exponent for now
* Revert "SYCL: disable faulty fp16 CPU exponent for now"
This reverts commit ed0aab1ec3.
* SYCL: disable faulty fp16 CPU exponent for now
* Fix logic of disabling exponent kernel
* implement unary REGLU/GEGLU/SWIGLU cpu ops
* relax constraints
* duplicate shape of source
* fix ggml_vec_geglu_f16
* special case gated ops
* implement unary REGLU/GEGLU/SWIGLU cuda ops
* tighten constraints again
* refactor into GGML_GLU_OP
* metal : add glu kernels
ggml-ci
* add CUDA_GLU_BLOCK_SIZE [no ci]
* more constraints and use 64bit ints
ggml-ci
* 64bit multiplication [no ci]
* implement swapped variants (cpu/cuda)
* update comment [no ci]
ggml-ci
* Vulkan: Add GLU ops and shaders
* SYCL: Implement fused kernel GEGLU, SWIGLU and REGLU for single up+gate
* ggml : implement GLU for split up/gate (#14181)
* implement GLU for split up/gate
* add tests for ggml_glu_split
* Vulkan: Implement glu_split logic and shader support
* add split to logging [no ci]
* SYCL: refactor element_size ops and add split up and gate support to gated kernels
* SYCL: switch GEGLU to use tanh approximation
---------
Co-authored-by: 0cc4m <picard12@live.de>
Co-authored-by: Akarshan <akarshan@menlo.ai>
* GGML: increase OP count in assertion
* Refactor: Optimize SYCL element-wise operations with unary function inlining
This commit refactors the SYCL element-wise operations to improve performance by:
- Inlining unary operations (sgn, abs, elu, gelu, silu, etc.) to reduce kernel launch overhead.
- Introducing helper functions `op_xxx` for each unary operation to encapsulate the logic.
- Replacing direct kernel calls with calls to these inlined functions.
- Using `__dpct_inline__` to encourage compiler inlining.
- Minor code cleanup and consistency improvements.
The changes aim to reduce kernel launch overhead and improve the overall efficiency of element-wise operations on SYCL devices.
* vulkan: Increase workgroup size for GLU, for performance (#14345)
* vulkan: Increase workgroup size for GLU, for performance
* vulkan: change GLU shaders to do one element per invocation rather than one row per workgroup
* merge fix
* metal : add support for split and swap
ggml-ci
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: 0cc4m <picard12@live.de>
Co-authored-by: Akarshan <akarshan@menlo.ai>
Co-authored-by: Jeff Bolz <jbolz@nvidia.com>
* vulkan: Add fusion support for RMS_NORM+MUL
- Add a use_count to ggml_tensor, so we can detect if an output is used more than once.
- Change the ggml-vulkan rms_norm shader to optionally multiply by another tensor.
- Add detection logic and basic fusion logic in ggml-vulkan.
- Add some testing support for fusion. Rather than computing one node at a time, allow
for computing the whole graph and just testing one node's results. Add rms_norm_mul tests
and enable a llama test.
* extract some common fusion logic
* fix -Winconsistent-missing-override
* move ggml_can_fuse to a common function
* build fix
* C and C++ versions of can_fuse
* move use count to the graph to avoid data races and double increments when used in multiple threads
* use hash table lookup to find node index
* change use_counts to be indexed by hash table slot
* minimize hash lookups
style fixes
* last node doesn't need single use.
fix type.
handle mul operands being swapped.
* remove redundant parameter
---------
Co-authored-by: slaren <slarengh@gmail.com>
* CUDA: add bf16 and f32 support to cublas_mul_mat_batched
* Review: add type traits and make function more generic
* Review: make check more explicit, add back comments, and fix formatting
* Review: fix formatting, remove useless type conversion, fix naming for bools
* ggml : add ggml_set_rows
Add ggml_set_rows(a, b, c) which copies rows from 'b' into 'a' using
indices from 'c'.
ref: #8366
* use I64 for indices
* ggml : add repeat impl for i64
* ggml : add ggml_is_contiguous_rows
* ggml : ggml_set_rows support broadcast
* ggml : ggml_set_rows support quantized dst
ggml-ci
* ggml : support GGML_TYPE_F32 ".from_float" trait
* ggml : ggml_set_rows update comment + better index name
* tests : add ggml_set_rows
* metal : add ggml_set_rows implementation
ggml-ci
* ggml : simplify forward_dup_f32
* ggml : fix supports_op
* tests : add comment to set_rows
* ggml : leave the repeat_i64 for a separate PR
ggml-ci
* ggml : set_rows use std::min instead of MIN
* ggml : better error message for set_rows unsupported type
* metal : perform op->type check only once
* tests : more consistent implementation + more tests
ggml-ci
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* Add support for VK_EXT_debug_utils to add labels to Vulkan objects. In step 1 compute pipelines are getting labeled.
* remove #ifdef for debug utils and add queue marker.
* Add header and namespace to use enqueue_functions extension
* Convert submit and parallel_for to use new extension in convert.cpp
* Convert submit and parallel_for to use extension in ggml-sycl.cpp
* Convert submit and parallel_for to use extension in gla.cpp
* Convert submit and parallel_for in mmq.cpp
* Convert submit and parallel_for in mmvq.cpp
* Convert submit and parallel_for in remaining files
* Convert all simple parallel_for to nd_launch from enqueue_functions
extension
* Wrapping extension in general function
Create a general function that enable the enqueue_functions extension if
it is enable in the compiler, otherwise call the general SYCL function
to launch kernels.
---------
Signed-off-by: nscipione <nicolo.scipione@codeplay.com>
* Add PowerPC feature detection and scoring
* ggml-cpu: Implement GGML_CPU_ALL_VARIANTS for PowerPC
* ggml-cpu: Delay some initializations until function is called
When using GGML_BACKEND_DL=ON, these initializations might use
instructions that are not supported by the current CPU.
---------
Co-authored-by: Diego Devesa <slarengh@gmail.com>
This commit removes the unused `ggml_context_container` structure from
the ggml library. It looks like the usage of this struct was removed in
Commit 4757fe18d56ec11bf9c07feaca6e9d5b5357e7f4 ("ggml : alloc
ggml_contexts on the heap (whisper/2525)").
The motivation for this changes is to improve code clarity/readability.
* llama : add thread safety test
* llamafile : remove global state
* llama : better LLAMA_SPLIT_MODE_NONE logic
when main_gpu < 0 GPU devices are not used
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Update oneMath commit to merged PR https://github.com/uxlfoundation/oneMath/pull/669
which adds SYCL-Graph support for recording CUDA BLAS commands.
With this change the `MUL_MAT` tests now pass on DPC++ CUDA backends with SYCL-Graph
enabled. Prior to this change, an error would be thrown.
```
$ GGML_SYCL_DISABLE_GRAPH=0 ./bin/test-backend-ops -b SYCL0 -o MUL_MAT -p type_a=f16,type_b=f32,m=16,n=1,k=256,bs=\\[1,1\\],nr=\\[2
UR CUDA ERROR:
Value: 700
Name: CUDA_ERROR_ILLEGAL_ADDRESS
Description: an illegal memory access was encountered
Function: operator()
Source Location: $HOME/dpcpp/unified-runtime/source/adapters/cuda/queue.cpp:154
Native API failed. Native API returns: 2147483646 (UR_RESULT_ERROR_UNKNOWN)
Exception caught at file:$HOME/llama.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp, line:3598, func:operator()
SYCL error: CHECK_TRY_ERROR((stream)->wait()): Meet error in this line code!
in function ggml_backend_sycl_synchronize at $HOME/llama.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp:3598
$HOME/llama.cpp/ggml/src/ggml-sycl/../ggml-sycl/common.hpp:118: SYCL error
Could not attach to process. If your uid matches the uid of the target
process, check the setting of /proc/sys/kernel/yama/ptrace_scope, or try
again as the root user. For more details, see /etc/sysctl.d/10-ptrace.conf
ptrace: Operation not permitted.
No stack.
The program is not being run.
```
* ggml-cpu: Factor out feature detection build from x86
* ggml-cpu: Add ARM feature detection and scoring
This is analogous to cpu-feats-x86.cpp. However, to detect compile-time
activation of features, we rely on GGML_USE_<FEAT> which need to be set
in cmake, instead of GGML_<FEAT> that users would set for x86.
This is because on ARM, users specify features with GGML_CPU_ARM_ARCH,
rather than with individual flags.
* ggml-cpu: Implement GGML_CPU_ALL_VARIANTS for ARM
Like x86, however to pass around arch flags within cmake, we use
GGML_INTERNAL_<FEAT> as we don't have GGML_<FEAT>.
Some features are optional, so we may need to build multiple backends
per arch version (armv8.2_1, armv8.2_2, ...), and let the scoring
function sort out which one can be used.
* ggml-cpu: Limit ARM GGML_CPU_ALL_VARIANTS to Linux for now
The other platforms will need their own specific variants.
This also fixes the bug that the the variant-building branch was always
being executed as the else-branch of GGML_NATIVE=OFF. The branch is
moved to an elseif-branch which restores the previous behavior.
This change moves the command pool/buffer tracking into a vk_command_pool
structure. There are two instances per context (for compute+transfer) and
two instances per device for operations that don't go through a context.
This should prevent separate contexts from stomping on each other.
Use the same descriptor set layout for all pipelines (MAX_PARAMETER_COUNT == 8)
and move it to the vk_device. Move all the descriptor pool and set tracking to
the context - none of it is specific to pipelines anymore. It has a single vector
of pools and vector of sets, and a single counter to track requests and a single
counter to track use.
* Simplify the environment variable setting to specify the memory pool type.
* Adjust the GGML_CANN_ASYNC_MODE setting to accept yes, enable, 1, or on (case-insensitive) as valid options.
* update
* fix CI
* update
* delete whitespace
* fix according to review
* update CANN.md
* update CANN.md
* Add Reorder to Q6_K mmvq implementation
* Address PR comments: clean up comments
* Remove unused parameter after refactoring q4_k
* Adding inline to function and removing unnecessary reference to int
---------
Signed-off-by: nscipione <nicolo.scipione@codeplay.com>
* SYCL: Implement few same quantized type copy kernels
* Use memcpy for copying contiguous tensors
ggml-ci
* feat(sycl): add contiguous tensor copy support and device checks
Adds a memcpy path for contiguous tensors of the same type to optimize data transfer. Updates device support checks to recognize contiguous tensor operations, improving compatibility and performance.
* refactor: replace specific block copy functions with template
The changes replace multiple redundant block copy functions (e.g., cpy_block_q8_0_q8_0, cpy_block_q5_0_q5_0) with a single templated function cpy_blck_q_q. This reduces code duplication by using a generic template that works for any block type, improving maintainability while preserving the same functionality. The template is instantiated with specific block types (e.g., block_q8_0) where needed.
* Exclude BF16 support for COPY tensors for now
ggml-ci
* perf: adjust SYCL copy kernel block sizes for efficiency
Use ceil_div to ensure full element coverage and update nd_range parameters to better align with SYCL block sizes, improving parallelism and device utilization in copy operations.
* * ggml-vulkan: adds op CONV_TRANSPOSE_1D
* test-backend-ops: adds more spohisticated tests for CONV_TRANSPOSE_1D
* Missing barrier added to shader.
Number of additional tests reduced to 108.
* * Fixes typo in variable name.
* Removes extra whitespaces.
* Adds int64->int32 casts to prevent possible warnings.
* Problem size reduced in tests to pass tests with llvmpipe.
* supports_op condition moved from unintended position
* This is not needed by the normal use where the result is read
using `tensor_get`, but it allows perf mode of `test-backend-ops`
to properly measure performance.
Some systems report the CPU implementation as "Power11" instead of "POWER11".
The existing CMake logic uses a case-sensitive regular expression to extract
the CPU generation, which fails when the casing doesn't exactly match "POWER".
This patch provides a fix by first converting the string to uppercase before applying the regex.
Signed-off-by: root <root@rheldb2v.pperf.tadn.ibm.com>
Co-authored-by: root <root@rheldb2v.pperf.tadn.ibm.com>
* threading: support for GGML_SCHED_PRIO_LOW, update thread info on Windows to avoid throttling
We talked about adding LOW priority for GGML threads in the original threadpool PR.
It might be useful for some cases to avoid contention.
Latest Windows ARM64 releases started parking (offlining) the CPU cores
more aggresively which results in suboptimal performance with n_threads > 4.
To deal with that we now disable Power Throttling for our threads for the NORMAL
and higher priorities.
Co-authored-by: Diego Devesa <slarengh@gmail.com>
* threading: disable SetThreadInfo() calls for older Windows versions
* Update tools/llama-bench/llama-bench.cpp
Co-authored-by: Diego Devesa <slarengh@gmail.com>
---------
Co-authored-by: Diego Devesa <slarengh@gmail.com>
* 1. add "integrated" in ggml_cuda_device_info for distinguish whether it is Intergrate_gpu or discrete_gpu
2. Adjust the func:"ggml_backend_cuda_device_supports_buft" for this new feature
* Update ggml/src/ggml-cuda/ggml-cuda.cu
Adjusted code indentation
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Update ggml/src/ggml-cuda/ggml-cuda.cu
Fixed incorrect setting of variable types
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Update ggml/src/ggml-cuda/ggml-cuda.cu
Adjusted the judgment logic
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* add a host_buft assert in case of integrated_cuda_device with func:'evaluate_and_capture_cuda_graph()'
* Update ggml/src/ggml-cuda/ggml-cuda.cu
Add a defensive security assert
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Update ggml/src/ggml-cuda/ggml-cuda.cu
Adjusted the support judgment logic.
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* revoke the suggest commit changes due to it's not applicable in jetson_device
* Update ggml/src/ggml-cuda/ggml-cuda.cu
Add parentheses to enforce operator precedence
Co-authored-by: Diego Devesa <slarengh@gmail.com>
* Update ggml/src/ggml-cuda/ggml-cuda.cu
Fix ci bug: add a spaces
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
---------
Co-authored-by: yangxiao <yang_xl@tju.edu.cn>
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
Co-authored-by: yangxiao <yangxl_zz@qq.com>
Co-authored-by: Diego Devesa <slarengh@gmail.com>
* SYCL: Add mrope kernel
* feat: Optimize rope operations with vectorization
Uses `sycl::vec` to load and store two elements at a time,
significantly improving performance in `rope_norm`,
`rope_neox`, and `rope_multi`. This reduces the number of memory
accesses and leverages SIMD instructions for faster execution.
* Use ceil_div
* cmake: Define function for querying architecture
The tests and results match exactly those of ggml/src/CMakeLists.txt
* Switch arch detection over to new function
* SYCL: Add non contiguous input support to norm kernel
* refactor and add RMS_NORM non contiguous input support
ggml-ci
* restore subgroup reduction for multi-subgroup thread blocks in norm kernels
* Swap grid dims of nsamples and nrows
ggml-ci
* Revert "Swap grid dims of nsamples and nrows"
This reverts commit 43be2d657fec7f7fba54e2cd154106bc0fc45adf.
* restore not required changes
ggml-ci
* address review comments: change it to more like SYCL
* Use a common function to calculate offset
* remove wrap around logic for handling broadcasts
* remove static from calculate_offset fn and use ceil_div
* cann: add the basic FA support
* cann: update the readme
* cann: update the FlashAttention with PSEShift
* cann: update the input parameters in FA
* cann: update the alibi with max_bias
* cann: add the constrints of softcap
* cann: update the docs CANN.md
* cann: update the docs CANN.md
* cann: fix typo of CANN.md
* cann: add some comments and update the CANN.md
* cann: update the CANN.md
* cann: update the inner precise for fusedInferAttention
* cann: update the constraints of flash_attn_ext on ggml-cann.cpp
* cann: clean the whitespace
* cann: clean the whitespace
* cann: add a new endline
* opencl: Add support for multiple devices
... but limited to one platform. A platform with a GPU will be preferred.
Additionally:
* Filter out devices that lack capabilities needed by the backend
implementation (half support, OpenCL 2.0+, etc).
* Make ggml_backend_opencl_reg() thread-safe.
* fixup: fix an error in sync_with_other_backends
... when there is only one OpenCL device available.
* opencl: fix couple crashes
* fix kernel launches failed on devices which do not support
non-uniform work-groups. When non-uniform work-groups are not
supported, set `local_work_size` to NULL (= let driver choose the
work-group sizes). This patch does not cover everything - just the
cases tested by test-backend-ops.
* fix sub-buffer creation failed due to `cl_buffer_region::origin` not
being aligned to `CL_DEVICE_MEM_BASE_ADDR_ALIGN`.
* OpenCL: query non-uniform WG sizes only on OpenCL 3.0+
* musa: fix build warning (unused parameter)
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
* musa: upgrade MUSA SDK version to rc4.0.1
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
* musa: use mudnn::Unary::IDENTITY op to accelerate D2D memory copy
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
* Update ggml/src/ggml-cuda/cpy.cu
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* musa: remove MUDNN_CHECK_GEN and use CUDA_CHECK_GEN instead in MUDNN_CHECK
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
---------
Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* Remove mmap workaround on windows
After some testing I found that mmap is supported on windows and for
many GPUs on Linux. Therefore I remove the workaround for windows since
it is not necessary.
* Update llama-bench README
SYCL backend introduced a workaround that allows execution of
llama-bench also without specifying `--mmp 0` flag
This shader uses coopmat1 to do the Q*K^T multiply. The P*V multiply is more
difficult for various reasons so I haven't done it. Performance for this
shader is around 2.5x better than for the scalar shader when doing prompt
processing. Some of the benefit may be from other optimizations like staging
through shared memory, or splitting by rows.
* batched-bench : fix pp batch contents
* metal : optimize multi-sequence FA vec kernel
ggml-ci
* metal : use FA-vec kernel up to batch size 20
ggml-ci
* llama/ggml: add LLM training support
more compact progress bar
llama_save_model_to_file
llama_opt_param_filter
ggml_graph_dup force_grads
refactor ggml_opt, fix test-opt
* remove logits_all
* refactor CUDA implementation for ACC
* reset graph at beginning of opt period
* ggml-cpu: Integrate fp32=bf16xbf16 SME KleidiAI kernel
Signed-off-by: Dan Johansson <dan.johansson@arm.com>
* * code review fixes
Signed-off-by: Dan Johansson <dan.johansson@arm.com>
* * adds a comment that clarifies barrier usage
Signed-off-by: Dan Johansson <dan.johansson@arm.com>
---------
Signed-off-by: Dan Johansson <dan.johansson@arm.com>
Co-authored-by: Charles Xu <charles.xu@arm.com>
* vulkan: scalar flash attention implementation
* vulkan: always use fp32 for scalar flash attention
* vulkan: use vector loads in scalar flash attention shader
* vulkan: remove PV matrix, helps with register usage
* vulkan: reduce register usage in scalar FA, but perf may be slightly worse
* vulkan: load each Q value once. optimize O reduction. more tuning
* vulkan: support q4_0/q8_0 KV in scalar FA
* CI: increase timeout to accommodate newly-supported tests
* vulkan: for scalar FA, select between 1 and 8 rows
* vulkan: avoid using Float16 capability in scalar FA
* sycl : Implemented reorder Q4_0 mmvq
Signed-off-by: Alberto Cabrera <alberto.cabrera@codeplay.com>
* sycl : Fixed mmvq being called when reorder is disabled
* sycl : Improved comments in the quants header
Signed-off-by: Alberto Cabrera <alberto.cabrera@codeplay.com>
* Use static_assert
* safe_div -> ceil_div
* Clarify qi comment
* change the reorder tensor from init to execute OP
* dbg
* Undo changes to test-backend-ops
* Refactor changes on top of q4_0 reorder fix
* Missing Reverts
* Refactored opt_for_reorder logic to simplify code path
* Explicit inlining and unroll
* Renamed mul_mat_algo enum for consistency
---------
Signed-off-by: Alberto Cabrera <alberto.cabrera@codeplay.com>
Co-authored-by: romain.biessy <romain.biessy@codeplay.com>
This assert fired running Qwen_Qwen3-30B-A3B-Q2_K.gguf:
GGML_ASSERT(nei0 * nei1 <= 3072);
The tensor is 8 x 512. Increase this array size to accommodate.
* ggml : remove MSVC warnings pragmas
This commit removes the MSVC-specific pragmas as these are now handled
in ggml/CMakeLists.txt.
* whisper : remove MSVC warning pragmas
This commit removes the MSVC-specific pragmas. These are now handled in
the ggml/CMakeLists.txt file.
This patch upstreams llamafile's cpu matrix multiplication kernels for ppc64le using MMA builtins for BF16 data type.
This change results in 9x - 40x gains
in total speed S t/s (ie all tokens/total time), across various batch sizes tested using llama-batched-bench benchmark.
The patch is tested with Meta-Lllama-3-8B,
and Mistral-7B models (BF16 models generated by using llama-quantize from corresponding FP32 models) on an IBM POWER10 machine.
Signed-off-by: Shalini Salomi Bodapati <Shalini.Salomi.Bodapati@ibm.com>
The following scenario will cause an assertion failure in the graph
allocator:
- Build and allocate a graph containing a tensor with a non-NULL data
pointer
- Build and allocate a new graph where that data is NULL
Result:
ggml-alloc.c:819: GGML_ASSERT(talloc->buffer_id >= 0) failed
This happens during revalidation because we think that memory should
have been previously allocated based on the current graph but in
reality the previous graph was different. In this situation, we
should do a full reallocation pass.
* vulkan: Add bfloat16 support
This adds bfloat16 matrix multiply support based on VK_KHR_shader_bfloat16.
The extension is required for coopmat multiply support, but matrix-vector
multiply trivially promotes bf16 to fp32 and doesn't require the extension.
The copy/get_rows shaders also don't require the extension.
It's probably possible to fall back to non-coopmat and promote to fp32 when
the extension isn't supported, but this change doesn't do that.
The coopmat support also requires a glslc that supports the extension, which
currently requires a custom build.
* vulkan: Support bf16 tensors without the bf16 extension or coopmat support
Compile a variant of the scalar mul_mm shader that will promote the bf16
values to float, and use that when either the bf16 extension or the coopmat
extensions aren't available.
* vulkan: bfloat16 fixes (really works without bfloat16 support now)
* vulkan: fix spirv-val failure and reenable -O
Build fails with compilation error on power pc.
This patch fixes the same.
Tested with unit tests run via
--build <build_dir> && cd <build_dir> && make test
Signed-off-by: Shalini Salomi Bodapati <Shalini.Salomi.Bodapati@ibm.com>
* fix(rpc): Improve input validation and error handling
The `rpc-server` was vulnerable to Denial of Service attacks via
several RPC commands (`SET_TENSOR`, `GRAPH_COMPUTE`, etc.). Malformed
messages could trigger failed assertions (e.g., invalid `ggml_type`)
or out-of-bounds reads/writes leading to `GGML_ABORT` calls,
crashing the server process.
This PR introduces robust input validation and replaces `abort()`
calls with graceful error handling:
- **Type Validation:** `deserialize_tensor` now checks if the
`tensor->type` is within the valid `GGML_TYPE_COUNT` range
*before* calling `ggml_new_tensor_4d`. Returns `nullptr` on
invalid type.
- **Bounds Checks:** Replaced `GGML_ABORT` in `set_tensor`,
`set_tensor_hash`, and `get_tensor` handlers with error
logging and returning `false` when data/offset parameters
are out of buffer bounds.
- **Size Checks:** Added safe arithmetic checks (for overflow) in
`graph_compute` when calculating required message sizes based
on client-provided `n_nodes` and `n_tensors`. Returns early
if the reported sizes conflict with the actual message size or
would lead to overflow.
- **Error Propagation:**
- `create_node` now checks for `nullptr` return values from
`deserialize_tensor` and its recursive calls, propagating
`nullptr` upwards on failure. Uses `find` instead of `at`
for safer map access.
- `copy_tensor` now checks for `nullptr` from `deserialize_tensor`
and sets the response status to failure if deserialization
or bounds checks fail.
- `graph_compute` now checks for `nullptr` return from
`create_node` and returns failure status correctly. The final
return value now reflects the actual computation status.
These changes improve the RPC server's resilience
against malformed client requests, preventing crashes and ensuring
errors are handled more gracefully.
Signed-off-by: Ville Vesilehto <ville@vesilehto.fi>
* refactor(rpc): address pr comments
removed comments and unnecessary returns
Signed-off-by: Ville Vesilehto <ville@vesilehto.fi>
* refactor(rpc): ambiguous nullptr from create_node
rpc_server::create_node could previously return nullptr if the input ID
was 0 (valid) or if an internal error (deserialization, recursion
failure) occurred (invalid). This ambiguity made error handling
difficult for the caller (`graph_compute`).
This commit clarifies the meaning of nullptr:
- `graph_compute` now checks if the input 'id' was non-zero when
`create_node` returns nullptr, correctly identifying failures
versus intentional null links.
- `create_node` avoids recursive calls for zero IDs and propagates
nullptr unambiguously on failure during recursion.
Signed-off-by: Ville Vesilehto <ville@vesilehto.fi>
* refactor(rpc): initial zero check in create_node
The caller (`graph_compute`) already checks `id != 0` when handling
a `nullptr` return from `create_node`, correctly distinguishing
intentional null links from actual errors. This makes the initial
`if (id == 0)` check redundant.
Also removes the log message when a tensor ID is not found in the
provided map which was added in this branch.
Signed-off-by: Ville Vesilehto <ville@vesilehto.fi>
* fix(rpc): Handle get_alloc_size failure in server
Check the return value of `server.get_alloc_size` in the RPC server
loop. If the call fails, return early to close the connection.
Signed-off-by: Ville Vesilehto <ville@vesilehto.fi>
* refactor(rpc): input size validation in graph_compute
Removes detailed, step-by-step size calculations and overflow
checks in favor of simpler direct comparisons, assuming 64-bit
overflow is unlikely.
Signed-off-by: Ville Vesilehto <ville@vesilehto.fi>
* refactor(rpc): remove extra status code setting
Removes the explicit setting of `response.result = GGML_STATUS_FAILED`
when `create_node` returns `nullptr` within `graph_compute`.
Primary signal is the `false` return value in case of failure.
Signed-off-by: Ville Vesilehto <ville@vesilehto.fi>
* refactor(rpc): remove redundant check for tensor->type
Breaks CI on ubuntu-cpu-make. Tensor type is uint32_t, thus
the check is not needed.
Signed-off-by: Ville Vesilehto <ville@vesilehto.fi>
---------
Signed-off-by: Ville Vesilehto <ville@vesilehto.fi>
* SYCL: Add all missing unary kernels
ggml-ci
* decouple kernel launch range from data size using strided loop
* use ciel_div helper for num_blocks
ggml-ci
* clean auto imported header files
RPC_CMD_SET_TENSOR always returns an empty response and we send this 4
times per token. We can improve TG speed if we don't wait for this empty
response.
The performance impact of this change depends on the network latency.
* tune matmul for gcn
* this one is more power efficient
* Update ggml/src/ggml-vulkan/ggml-vulkan.cpp
Co-authored-by: 0cc4m <picard12@live.de>
* disable this tune for the proprietary driver
---------
Co-authored-by: 0cc4m <picard12@live.de>
Add RPC_CMD_HELLO for getting the version of the protocol implemend by
the server. Follow the semantic versioning rules at https://semver.org
Hopefully this bring better user experience when we make breaking
changes at the protocol level and avoid issues like #12465
* graph : make mla compatible with FA
* metal : add exp FA kernels for DeepSeek models
ggml-ci
* llama : minor naming updates
ggml-ci
* ggml : disable FA for DS head sizes
* tests : add FA tests for MLA shapes
ggml-ci
Submit operators using asynchronous threads to improve performance.
Use the environment variable GGML_CANN_ASYNC_MODE to control whether
asynchronous submission is enabled. It is disabled by default.
Testing shows a 10%–20% performance improvement in scenarios with
small parameter sizes, especially in quantized models.
The grouped query attention optmization doesn't require a power of two ratio,
the only thing relying on it was the modulo operation written as bitwise &.
split_k need not depend on gqa_ratio - enable it any time there's only one
workgroup in the X dimension. The shader gets the split index from the x coord,
and multiple workgroups in the X dimension (pre-split) indicates a larger
FA operation that wouldn't need splitting.
* opencl: refactor - split the kernel files
---------
Co-authored-by: Shangqing Gu <quic_shawngu@quicinc.com>
* opencl: split more kernels into separate files
* opencl: specify subgroup size instead of querying it
* opencl: refine Adreno cl compiler version parsing
* opencl: skip some kernels not used by Adreno on old compilers
* opencl: refine logic for selecting Adreno kernels
* opencl: refine Adreno cl compiler version
* opencl: cleanup preprocessor for kernels
* opencl: consider Adreno CL compiler on Windows
* opencl: add final newline for `mul_mv_f16_f16.cl`
---------
Co-authored-by: Shangqing Gu <quic_shawngu@quicinc.com>
Replace compile-time `GGML_HIP_UMA` with environment variable `GGML_CUDA_ENABLE_UNIFIED_MEMORY`. This unifies the usage on NVIDIA and AMD GPUs, and allows a single binary to be shared between integrated and dedicated GPUs.
Multiple optional memory pools are provided for CANN, including VMM,
priority queue-based, and traditional memory pools.
1.When the memory pool is available and GGML_CANN_DISABLE_VMM_POOL
is not defined, the VMM pool is selected by default.
2.Otherwise, if GGML_CANN_ENABLE_BUF_PRIO_POOL is defined,
the priority queue-based memory pool is used.
3.If neither condition is met, the default memory pool is used.
The current usage of the SYCL-Graph extension checks for
the `sycl_ext_oneapi_graph` device aspect. However, it is also
possible to support `sycl_ext_oneapi_limied_graph` devices that
don't support update
* SYCL: Add fp16 support to some elementwise OP kernels
* remove comment
ggml-ci
* Use static_cast directly
* remove not needed cast from tanh
* Use static cast and remove unneeded castings
* Adjust device_support_op for unary OPs
* Use cast_data and typed_data struct to deduplicate casting code
* [CANN] Support ELU and CONV_TRANSPOSE_1D
* [CANN]Modification review comments
* [CANN]Modification review comments
* [CANN]name adjustment
* [CANN]remove lambda used in template
* [CANN]Use std::func instead of template
* [CANN]Modify the code according to the review comments
---------
Signed-off-by: noemotiovon <noemotiovon@gmail.com>
q4_k and q5_k had a lot of redundant global loads where the same 16B of
scale information is repeatedly loaded and decoded during each loop iteration.
This change restructures the loops to more explicitly iterate over whole
blocks in the outer loop (with unrolled inner loop) and to copy/decode the
scale data into shared memory once at the start of each outer loop. The copy
is pipelined so the scale load from global memory is relatively cheap.
This improves q4_k/q5_k model prompt processing performance by around 5-7%.
I briefly tried applying this to q6_k and q4_0, and it didn't help for q6_k
and hurt for q4_0.
The big "else" path in mul_mm_cm2.comp that had all the clamped/unclamped
variants isn't used as often as it originally was (e.g. due to the padded_N
change), so I trimmed it down to offset some of the new complexity of the
semi-manual loop unrolling.
* ggml : FA supports F32 V
* graph : cast KV to F16 when the KV cache is not used
ggml-ci
* server : add test that exercises embeddings with FA enabled
ggml-ci
* add bf16 support
* use convert_from_bf16_cuda instead of convert_unary_cuda for f32
* revert 7ec5085
* move functionality into convert_unary with constexpr
* cpu: refactor SIMD mappings and vectorized op functions into separate files
* Fix warning for ggml_float to float
* Fix warnings
* cpu: move all the operations (except mul_mat) to a separate c++ file
* fix whitespace
* Update ggml/src/ggml-cpu/vec.h
Co-authored-by: Diego Devesa <slarengh@gmail.com>
* Fix PR comments - use GGML_UNUSED, use cassert in ops.cpp
* Reverse the order of import for ops.h and vec.h, to match what was present in ggml-cpu.c previously
---------
Co-authored-by: Diego Devesa <slarengh@gmail.com>
nem1 must be a multiple of GGML_KQ_MASK_PAD, and GGML_KQ_MASK_PAD is a multiple
of the number of rows in the matrix. The KV dim is a multiple of the number of
columns for the aligned shader.
There seems to be a bubble waking up from waitForFences, which costs a few
percent performance and also increased variance in performance. This change
inserts an "almost_ready" fence when the graph is about 80% complete and we
waitForFences for the almost_ready fence and then spin (with _mm_pauses) waiting
for the final fence to be signaled.
* Prefer vector flash decoding kernel for Gemma models
Vector flash decoding kernel was not being picked for models with head dimension 256. Gemma models are in this category.
Removing this limit improves e2e performance by upto 12% in gen phase throughput for Gemm models.
* Update ggml/src/ggml-cuda/fattn.cu
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
---------
Co-authored-by: Johannes Gäßler <johannesg@5d6.de>
* CUDA: Simplify and improve CUDA graphs through use of indirect copy pointers
Previously there was complexity in the CUDA graphs implementation due
frequently changing parameters to copy kernels associated with K and V
cache pointers. This patch simplifies by using indirection to avoid
such parameters frequently changing, avoiding the need for frequent
graph updates.
Fixes#12152
* Addressed comments
* fix HIP builds
* properly sync to stream
* removed ggml_cuda_cpy_fn_ptrs
* move stream sync before free
* guard to only use indirection with graphs
* style fixes
* check for errors
---------
Co-authored-by: slaren <slarengh@gmail.com>
When using group query attention, we have one workgroup per KV batch and this
can be very few workgroups (e.g. just 8 in some models). Enable split_k to
spread the work across SMs. This helps a lot when the KV cache is large.
When adjacent batches of Q share the same batches of K/V, batch them into
the same workgroup. For example, when:
dst(128,32,1,1) = FA(q(128,1,32,1), k(128,16640,8,1), v(128,16640,8,1))
previously we would run 32 workgroups computing 1 result each, now we will
run 8 workgroups computing 4 results each.
This doesn't directly translate to better performance (at least when you have
>=32 SMs), but in a subsequent change I'll enable split_k which will scale much
better with 4x fewer workgroups.
* Rename oneMKL Interface to oneMath
* Use oneMath for Intel vendor
* Rename occurences to mkl
* clang-format
* Silence verbose warnings
* Set oneMath HIP_TARGETS
* Fix silence warnings
* Remove step to build oneMath from build instructions
* Use fixed oneMath version
* Remove INTEL_CPU
* Fold CMake oneDNN conditions
* Use Intel oneMKL for Intel devices
* Improve CMake message
* Link against MKL::MKL_SYCL::BLAS only
* Move oneMath documentation to Nvidia and AMD sections
Reese Levine
rmatif
lhez
Diego Devesa
Chenguang Li
Georgi Gerganov
Romain Biessy
Christian Kastner
Jeff Bolz
Johannes Gäßler
leejet
lhez
Srihari-mcw
diannao
Ruben Ortlam
hipudding
uvos
Kai Pastor
Sigbjørn Skjæret
Aman Gupta
xctan
Akarshan Biswas
Alberto Cabrera Pérez
Erik Scholz
deepsek
R0CKSTAR
Aaron Teo
Gabe Goodhart
Oliver Simons
Chris Rohlf
Daniel Bevenius
Donghyeon Jeong
lixing-star
chen fan
Charles Xu
Ervin Áron Tasnádi
Peter0x44
Oliver Simons
Neo Zhang Jianyu
Anton Mitkov
shalinib-ibm
Yavor Ivanov
Acly
Tarek Dakhran
Slobodan Josic
compilade
Xuan-Son Nguyen
Miaoqian Lin
Eve
luyhcsu
Nicolò Scipione
zhouwg
Eric Zhang
Björn Ganster
xiaobing318
Xinpeng Dou
Radoslav Gerganov
bandoti
uvos
Markus Tavenrath
fanyang
Anton Mitkov
Ewan Crawford
Isaac McFadyen
xctan
Masato Nakasaka
rmatif
Atharva Dubey
Daniel Tang
Max Krasnyansky
Shawn yang
Yibo Cai
Vineel Abhinav
leo-pony
Bizhao Shi
Judd
Henry Linjamäki
Svetlozar Georgiev
Gilad S.
Łukasz Ślusarczyk
Yibo Cai
Dan Johansson
David Huang
Jared Tweed
R0CKSTAR
Justin Santa Barbara
Jesse Gross
Aaron Teo
Ville Vesilehto
SXX
Jeffrey Morgan
Alan Gray
kimminsu
cmdr2
Ewan Crawford
Prajwal B Mehendarkar
Piotr Kubaj
Ronny Brendel
Gaurav Garg
a3sh
Junil Kim
Wagner Bruna
Sandro Hanea