]>
git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/log
Georgi Gerganov [Sat, 13 Dec 2025 20:54:14 +0000 (22:54 +0200)]
ggml : arm repack fix build (llama/0)
Jeff Bolz [Sat, 13 Dec 2025 09:12:53 +0000 (03:12 -0600)]
vulkan: support get_rows for i32 (llama/17941)
Jeff Bolz [Sat, 13 Dec 2025 09:07:49 +0000 (03:07 -0600)]
vulkan: support GGML_OP_DIAG (llama/17893)
Jeff Bolz [Sat, 13 Dec 2025 09:04:29 +0000 (03:04 -0600)]
vulkan: Multi-pass softmax for large number of cols (llama/17892)
When the number of cols is large, split each row across multiple workgroups.
There are three phases that communicate partial results through temp buffers:
(1) compute max partials
(2) take max of partials, compute sum(exp(x-max)) partials
(3) sum partials, compute scaled result
Jeff Bolz [Sat, 13 Dec 2025 07:40:04 +0000 (01:40 -0600)]
vulkan: Allow non-pow2 n_experts in topk_moe (llama/17872)
Johannes Gäßler [Fri, 12 Dec 2025 16:43:58 +0000 (17:43 +0100)]
CUDA: fix overflow in MMA kernel without stream-k (llama/17939)
Sigbjørn Skjæret [Fri, 12 Dec 2025 14:49:27 +0000 (15:49 +0100)]
cann : fix ops broken by circular padding guard (llama/17825)
ixgbe [Fri, 12 Dec 2025 14:26:03 +0000 (22:26 +0800)]
ggml-cpu : fix RISC-V Q4_0 repack select and RVV feature reporting (llama/17951)
* ggml-cpu:fix RISC-V Q4_0 repack select and RVV feature reporting
Signed-off-by: Wang Yang <redacted>
* using the name VLEN instead of CNT
* Update ggml/include/ggml-cpu.h
---------
Signed-off-by: Wang Yang <redacted>
Co-authored-by: Georgi Gerganov <redacted>
yulo [Fri, 12 Dec 2025 10:34:33 +0000 (18:34 +0800)]
HIP: enable mmf for RDNA3 (llama/17879)
* enable mmf for RDNA3
* disable mmf for some shape
* move some mmvf to mmf
* more mmfv to mmf
* 3 is good in mmvf
---------
Co-authored-by: zhang hui <redacted>
Piotr Wilkin (ilintar) [Thu, 11 Dec 2025 16:20:43 +0000 (17:20 +0100)]
SOLVE_TRI extension to more dimensions (llama/17793)
* Extended TRI
* Fix whitespace
* chore: update webui build output
* Just use cuBLAS for everything...
* Merge both versions
* Remove incorrect imports causing failures for CI
* Still failing... remove all direct cublas imports and rely on common imports from "common.cuh"
* Defines for hipBlas
* Aaaand MUSA defines...
* I hate this job...
* Stupid typo...
* Update ggml/src/ggml-cuda/solve_tri.cu
Co-authored-by: Johannes Gäßler <redacted>
---------
Co-authored-by: Johannes Gäßler <redacted>
Russ [Wed, 17 Dec 2025 09:13:38 +0000 (09:13 +0000)]
build: link whisper target against Threads::Threads for FreeBSD support (#3568)
Marcos Del Sol Vives [Sat, 13 Dec 2025 07:37:44 +0000 (08:37 +0100)]
server: allow custom temp directory for ffmpeg (#3564)
Georgi Gerganov [Sat, 13 Dec 2025 06:04:09 +0000 (08:04 +0200)]
ggml : arm repack fix build (#0)
Georgi Gerganov [Fri, 12 Dec 2025 15:56:43 +0000 (17:56 +0200)]
talk-llama : sync llama.cpp
Georgi Gerganov [Fri, 12 Dec 2025 15:55:11 +0000 (17:55 +0200)]
sync : ggml
Georgi Gerganov [Fri, 12 Dec 2025 15:54:58 +0000 (17:54 +0200)]
whisper : adjust to ggml changes (#0)
Congcong Cai [Fri, 12 Dec 2025 14:37:38 +0000 (22:37 +0800)]
cmake : set `CMAKE_RUNTIME_OUTPUT_DIRECTORY` for non standalone build (ggml/1394)
Some backend depends on CMAKE_RUNTIME_OUTPUT_DIRECTORY to create temporary file like metal backened.
Missing CMAKE_RUNTIME_OUTPUT_DIRECTORY will cause some cmake error like permission denied (try to copy file to root).
This PR wants to setup a default path for CMAKE_RUNTIME_OUTPUT_DIRECTORY when it does not exist.
Georgi Gerganov [Thu, 11 Dec 2025 12:30:10 +0000 (14:30 +0200)]
ggml-alloc : fix reuse-parent logic for misaligned sizes (llama/17884)
nullname [Wed, 10 Dec 2025 22:45:43 +0000 (06:45 +0800)]
ggml-hexagon: fix `rope` failure at `test-backend-ops` (llama/17565)
* fix test failure
* fix: correct scaling calculations in rope_cache_init
* fix: optimize element copying in rope_hex_f32 using memcpy
* fix: optimize loop boundaries in rope_hex_f32 for better performance
* feat: add profiling macros for performance measurement in operations
Max Krasnyansky [Wed, 10 Dec 2025 20:32:23 +0000 (12:32 -0800)]
Fix race conditions in threadpool when dealing with dynamic/frequent n_threads changes (llama/17748)
* tests: update barrier test to check for race condition in active threads
* cpu: combine n_graph and n_threads into a single atomic update
* tests: add multi-graph test for test_barrier
Georgi Gerganov [Wed, 10 Dec 2025 18:53:16 +0000 (20:53 +0200)]
ggml : remove GGML_KQ_MASK_PAD constant (llama/17910)
* ggml : remove GGML_KQ_MASK_PAD constant
* cont : remove comment
Sigbjørn Skjæret [Wed, 10 Dec 2025 15:16:20 +0000 (16:16 +0100)]
cuda : add missing support check for xielu (llama/17895)
Johannes Gäßler [Wed, 10 Dec 2025 11:39:56 +0000 (12:39 +0100)]
CUDA: fix unpadded strides in MMA FA kernel (llama/17891)
Neo Zhang Jianyu [Wed, 10 Dec 2025 08:59:57 +0000 (16:59 +0800)]
fix softmax for iGPU (llama/17838)
Gabe Goodhart [Tue, 9 Dec 2025 19:30:02 +0000 (12:30 -0700)]
metal: SSM kernel improvements (llama/17876)
* feat: Add a batched version of ssm_conv
This was done using Claude Code. It found a number of optimizations around
how the threads were organized, resulting in a huge performance boost!
Branch: Mamba2SSD
Signed-off-by: Gabe Goodhart <redacted>
* feat: Optimized SSM_SCAN kernel for metal
This used Claude Code and resulted in a modest performance improvement
while maintaining correctness.
Branch: Mamba2SSD
Signed-off-by: Gabe Goodhart <redacted>
* test: Add test-backend-ops perf tests for SSM_CONV
Branch: SSMKernelImprovements
Signed-off-by: Gabe Goodhart <redacted>
* test: Real representitive tests for SSM_CONV
Branch: SSMKernelImprovements
Signed-off-by: Gabe Goodhart <redacted>
* refactor: Use function constant for ssm_conv batch size
Branch: SSMKernelImprovements
Signed-off-by: Gabe Goodhart <redacted>
* test: backend op tests for ssm_scan from granite4 1b-h
Branch: SSMKernelImprovements
Signed-off-by: Gabe Goodhart <redacted>
* style: remove commented out templates
Branch: SSMKernelImprovements
Signed-off-by: Gabe Goodhart <redacted>
* feat: float4 version of ssm_conv_batched
Branch: SSMKernelImprovements
Signed-off-by: Gabe Goodhart <redacted>
* fix: Add missing ggml_metal_cv_free
Signed-off-by: Gabe Goodhart <redacted>
Co-authored-by: Georgi Gerganov <redacted>
---------
Signed-off-by: Gabe Goodhart <redacted>
Co-authored-by: Georgi Gerganov <redacted>
Piotr Wilkin (ilintar) [Tue, 9 Dec 2025 19:28:57 +0000 (20:28 +0100)]
Add DIAG for CUDA (llama/17873)
* Add DIAG for CUDA
* Refactor parameters
Gabe Goodhart [Tue, 9 Dec 2025 16:29:07 +0000 (09:29 -0700)]
ggml : Provide macos-specific backtrace printing to avoid terminal death (llama/17869)
* fix: Provide macos-specific backtrace printing to avoid terminal death
Branch: MacOSSafeBacktrace
Signed-off-by: Gabe Goodhart <redacted>
* fix: Add GGML_BACKTRACE_LLDB env var to enable using lldb for backtrace
Branch: MacOSSafeBacktrace
Signed-off-by: Gabe Goodhart <redacted>
---------
Signed-off-by: Gabe Goodhart <redacted>
Georgi Gerganov [Tue, 9 Dec 2025 13:25:49 +0000 (15:25 +0200)]
metal : print node names for debugging (llama/17882)
Sigbjørn Skjæret [Tue, 9 Dec 2025 11:23:47 +0000 (12:23 +0100)]
ggml : allow fill node alloc inplace (llama/17870)
Chenguang Li [Tue, 9 Dec 2025 09:53:23 +0000 (17:53 +0800)]
CANN: add support for partial RoPE and Vision mode (llama/17543)
* cann: add support for partial RoPE and Vision mode
Add support for two important RoPE variants: partial rotation (rope_dims < ne0)
and Vision mode rotation.
1. Support for partial RoPE (rope_dims < ne0):
- Split tensor into head (first rope_dims dimensions) and tail portions
- Apply rotation only to head portion using RotaryPositionEmbedding operator
- Copy unrotated tail portion directly from source to destination
- Handle both contiguous and non-contiguous tensor layouts
2. Support for Vision mode (GGML_ROPE_TYPE_VISION):
- Set rope_dims = ne0 for Vision mode to rotate entire tensor
- Vision mode pairs dimension i with dimension i+n_dims (where n_dims = ne0/2)
- No tail handling needed since entire tensor is rotated
Implementation details:
- Use has_tail flag to determine execution path: head/tail splitting when
rope_dims < ne0, or full tensor rotation when rope_dims == ne0
- Support both F32 and F16 data types with intermediate F32 conversion
- Copy non-contiguous tensors to contiguous buffers before calling
RotaryPositionEmbedding operator for compatibility
- Improve cache invalidation logic to include rope_dims and indep_sects
parameters
These enhancements enable CANN backend to handle various RoPE configurations
used in modern vision-language models and models with partial rotation.
* cann: fix review comment
Johannes Gäßler [Tue, 9 Dec 2025 08:34:02 +0000 (09:34 +0100)]
CUDA: fix FP16 overflow in tile FA kernel (llama/17875)
Jay Zenith [Mon, 8 Dec 2025 13:10:12 +0000 (05:10 -0800)]
cuda : add FILL op support (llama/17851)
* cuda : add FILL op support
* cuda : add missing FILL op files
wsbagnsv1 [Mon, 8 Dec 2025 09:41:08 +0000 (10:41 +0100)]
cuda: optimize SOLVE_TRI using registers and FMAF (llama/17703)
* ggml-cuda: optimize solve_tri_f32_fast and fix stride handling
- Switch from using shared memory for the RHS/solution matrix to a register-based approach (x_low, x_high), reducing shared memory pressure and bank conflicts.
- Implement explicit `fmaf` instructions for the reduction loop.
- Update kernel arguments to pass strides in bytes rather than elements to align with standard ggml tensor arithmetic (casting to `char *` before addition).
- Remove unused `MAX_K_FAST` definition.
* Small cleanup
* Remove comments in solve_tri.cu
* Update ggml/src/ggml-cuda/solve_tri.cu
Co-authored-by: Johannes Gäßler <redacted>
* Update ggml/src/ggml-cuda/solve_tri.cu
Co-authored-by: Johannes Gäßler <redacted>
* Update ggml/src/ggml-cuda/solve_tri.cu
Co-authored-by: Johannes Gäßler <redacted>
* Use const for variables in solve_tri.cu
* Replace fmaf with more readable code
* remove last fmaf
---------
Co-authored-by: Johannes Gäßler <redacted>
ixgbe [Mon, 8 Dec 2025 08:41:34 +0000 (16:41 +0800)]
ggml-cpu: add ggml_thread_cpu_relax with Zihintpause support (llama/17784)
* ggml-cpu: add ggml_thread_cpu_relax with Zihintpause support
Signed-off-by: Wang Yang <redacted>
* cmake: enable RISC-V zihintpause extension for Spacemit builds
* readme : add ZIHINTPAUSE support for RISC-V
---------
Signed-off-by: Wang Yang <redacted>
lovedheart [Sun, 7 Dec 2025 17:40:42 +0000 (18:40 +0100)]
Vulkan: improve mul_mat_vec_iq1_m (llama/16907)
* Optimize Vulkan shader for matrix-vector multiplication
* Revert changes on compute_outputs and main
Refactor compute_outputs to handle remaining rows correctly.
* Fix trailing whitespace
Law Po Ying [Sun, 7 Dec 2025 01:18:18 +0000 (09:18 +0800)]
sycl: add missing BF16 conversion support for Intel oneAPI (llama/17780)
* sycl: add missing BF16 conversion support for Intel oneAPI
* Fix Line 645: Trailing whitespace
Jeff Bolz [Sat, 6 Dec 2025 17:46:46 +0000 (11:46 -0600)]
vulkan: perf_logger improvements (llama/17672)
* vulkan: perf_logger improvements
- Move perf_logger from device to ctx.
- Add an env var to control the frequency we dump the stats. If you set a very
large value, it just dumps when the ctx is destroyed.
- Add a fusion info string to the tracking, only log one item per fused op.
- Fix MUL_MAT_ID flops calculation.
* fix vector sizes
Vishal Singh [Sat, 6 Dec 2025 16:13:33 +0000 (21:43 +0530)]
ggml-zendnn : add ZenDNN backend for AMD CPUs (llama/17690)
* ggml-zennn: add ZenDNN backend support
* ggml-zendnn : address ZenDNN backend review fixes and suggestions
* docs : apply blockquote syntax to ZenDNN docs
---------
Co-authored-by: Manoj Kumar <redacted>
Phylliida Dev [Sat, 6 Dec 2025 14:07:02 +0000 (06:07 -0800)]
ggml : add circular tiling support to pad, for Vulkan, CUDA, and CPU (used for making seamless textures) (llama/16985)
* Feat: Added vulkan circular tiling support
* Feat: Added cpu circular
* Feat: Added cuda kernels
* Added tests
* Added tests
* Removed non-pad operations
* Removed unneded changes
* removed backend non pad tests
* Update test-backend-ops.cpp
* Fixed comment on pad test
* removed trailing whitespace
* Removed unneded test in test-backend-ops
* Removed removed test from calls
* Update ggml/src/ggml-vulkan/vulkan-shaders/pad.comp
Co-authored-by: Ruben Ortlam <redacted>
* Fixed alignment
* Formatting
Co-authored-by: Aman Gupta <redacted>
* Format pad
* Format
* Clang format
* format
* format
* don't change so much stuff
* clang format and update to bool
* fix duplicates
* don't need to fix the padding
* make circular bool
* duplicate again
* rename vulkan to wrap around
* Don't need indent
* moved to const expr
* removed unneded extra line break
* More readable method calls
* Minor wording changes
* Added final newline
* Update ggml/include/ggml.h
Co-authored-by: Georgi Gerganov <redacted>
* Update ggml/include/ggml.h
Co-authored-by: Georgi Gerganov <redacted>
* Added circular pad ext tests
* Gate non circular pad devices
* Cleaned gating of non-circular pad devices
---------
Co-authored-by: Phylliida <redacted>
Co-authored-by: Ruben Ortlam <redacted>
Co-authored-by: Aman Gupta <redacted>
Co-authored-by: Georgi Gerganov <redacted>
Johannes Gäßler [Sat, 6 Dec 2025 12:45:36 +0000 (13:45 +0100)]
HIP: fix RDNA3 FP16/BF16 matrix multiplication (llama/17817)
Sky [Sat, 6 Dec 2025 11:28:16 +0000 (19:28 +0800)]
ggml : improve error handling for search path existence checks (llama/17653)
* Improve error handling for search path existence checks
Refactor existence checks for search paths using std::error_code to handle potential errors.
* Improve cache file existence check with error code
Update fs::exists to use std::error_code for error handling.
* Simplify existence check for search paths
Simplify existence check for search paths
* Fix logging path in error message for posix_stat
* Update ggml/src/ggml-backend-reg.cpp
Co-authored-by: Aman Gupta <redacted>
* Adapt to the coding standard
---------
Co-authored-by: Aman Gupta <redacted>
Jeff Bolz [Sat, 6 Dec 2025 10:12:26 +0000 (04:12 -0600)]
vulkan: Use one row per workgroup for f32 mmv (llama/17711)
The MoE models have a mul_mat_vec with very small m (32, 64, 128) right before
the topk_moe selection. Running multiple rows per wg doesn't utilize the SMs
well. I think even for larger m, f32 is so bandwidth-limited that running
multiple rows doesn't help.
Jeff Bolz [Sat, 6 Dec 2025 07:56:45 +0000 (01:56 -0600)]
vulkan: support solve_tri with larger N/K values (llama/17781)
Split N into chunks to fit into shared memory.
If K > 128, use a larger workgroup with enough invocations.
Add perf tests matching qwen3next.
Georgi Gerganov [Sat, 6 Dec 2025 07:33:59 +0000 (09:33 +0200)]
metal : fix build(#17799)
* metal : fix build
* tests : fix context destruction
Masato Nakasaka [Sat, 6 Dec 2025 05:39:42 +0000 (14:39 +0900)]
vulkan: Replace deprecated VK_EXT_validation_features (llama/17637)
* replaced deprecated VK_EXT_validation_features
* forgot to remove old code
Masato Nakasaka [Sat, 6 Dec 2025 05:23:30 +0000 (14:23 +0900)]
vulkan: Fix mismatch in TOPK_MOE unit test (llama/17541)
* Fix shader to support 2D workgroup mapping to a single subgroup
* Set required_subgroup_size
topk_moe shader requires static WARP_SIZE and actual subgroup size to match
Jeff Bolz [Fri, 5 Dec 2025 21:08:56 +0000 (15:08 -0600)]
vulkan: add more num_blocks instantiations in rms_norm (llama/17701)
Jeff Bolz [Fri, 5 Dec 2025 21:03:19 +0000 (15:03 -0600)]
vulkan: fix top_k bug when there are ties in the input (llama/17659)
* vulkan: Reduce temporary memory usage for TOP_K
- 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.
* vulkan: fix top_k bug when there are ties in the input
I noticed by inspection a bug in the vulkan top_k shader where if the least
value in the top_k appears multiple times we could end up writing those extra
copies out rather than some larger values (if the larger values are on higher
numbered threads).
I rewrote the test verification to handle this case, where the final index set
is not necessarily the same.
* Update tests/test-backend-ops.cpp
Co-authored-by: Georgi Gerganov <redacted>
---------
Co-authored-by: Georgi Gerganov <redacted>
Acly [Fri, 5 Dec 2025 20:46:39 +0000 (21:46 +0100)]
vulkan : support conv-2d with large output size (llama/17685)
Reese Levine [Fri, 5 Dec 2025 20:25:51 +0000 (12:25 -0800)]
ggml webgpu: unary op suppport, code refactoring, ops support (llama/17764)
* Squashed commit of the following:
commit
b3c6bf4b0450d8d452b934df27a0fb7cb53cd755
Author: Abhijit Ramesh <redacted>
Date: Mon Dec 1 18:29:00 2025 -0800
ggml webgpu: fix xielu parameter passing (llama/11)
The XIELU operation was incorrectly using static_cast to convert
float parameters to uint32_t, which converted numeric values instead
of preserving IEEE 754 bit patterns. This caused incorrect values
to be interpreted by the GPU shader.
* Use reinterpret_cast to preserve float bit patterns when passing
through uint32_t params buffer
* Update WGSL shader parameter types from u32 to f32
* Re-enable XIELU support (was disabled due to numerical issues)
Fixes NMSE test failures for XIELU operation on WebGPU backend.
commit
5ca9b5e49ea7cddc9ab7c8b43a11a9c76a4dff4a
Author: neha-ha <redacted>
Date: Tue Nov 18 12:17:00 2025 -0800
Refactored pipelines and workgroup calculations (llama/10)
* refactored pipelines
* refactored workgroup calculation
* removed commented out block of prior maps
* Clean up ceiling division pattern
---------
Co-authored-by: Neha Abbas <redacted>
Co-authored-by: Reese Levine <redacted>
Author: James Contini <redacted>
Date: Wed Oct 29 23:13:06 2025 -0700
formatted embed wgsl and ggml-webgpu.cpp
commit
e1f6baea31645e5d96ad53664acae856f74b96f4
Author: James Contini <redacted>
Date: Wed Oct 29 23:08:37 2025 -0700
implemented REPL_Template support and removed bug in unary operators kernel
commit
8c70b8fece445cdc9a8c660dbddbf201e52da2bb
Author: James Contini <redacted>
Date: Wed Oct 15 16:14:20 2025 -0700
responded and dealt with PR comments
commit
f9282c660c10dec4487d434549bdb707a9cd9f37
Author: James Contini <redacted>
Date: Sun Oct 12 13:41:41 2025 -0700
removed unnecesarry checking if node->src[1] exists for unary operators
commit
4cf28d7dec41c29186d66152735b244c5699f9dc
Author: James Contini <redacted>
Date: Sun Oct 12 13:32:45 2025 -0700
All operators (inlcluding xielu) working
commit
74c6add1761a59d2c2ff60b60e8ad3c8300f6d3e
Author: James Contini <redacted>
Date: Fri Oct 10 13:16:48 2025 -0700
fixed autoconfig
commit
362749910be4f0120c8ffb21ceddeb7d2c088e51
Author: James Contini <redacted>
Date: Fri Oct 10 13:10:46 2025 -0700
removed vestigial files
commit
cb0858333785757804c5104e59c4981843207c16
Author: James Contini <redacted>
Date: Fri Oct 10 12:59:32 2025 -0700
abides by editor-config
commit
5360e2852a4b51197d7d67d0a5d42e908b02d7ed
Author: James Contini <redacted>
Date: Fri Oct 10 12:45:57 2025 -0700
rms_norm double declaration bug atoned
commit
7b09baa4aa53711be5a126043670cc182c78bfcd
Merge:
8a6ec843 74b8fc17
Author: James Contini <redacted>
Date: Fri Oct 10 11:50:03 2025 -0700
resolving merge conflicts
commit
8a6ec843a50ab82f8cef59b4558eb63f318ba02d
Author: James Contini <redacted>
Date: Wed Oct 8 18:06:47 2025 -0700
unary operators pass ggml tests
commit
c3ae38278a2db236adc5912c9140e4f0d63f2c19
Author: James Contini <redacted>
Date: Wed Oct 1 16:22:40 2025 -0700
neg passes backend test
commit
aa1c9b2f8877a405470ca56709c42a1fd43713de
Author: James Contini <redacted>
Date: Tue Sep 30 23:55:27 2025 -0700
neg f16xf32xip builds and runs, havent actually ran a model that uses neg kernel yet though
Co-authored-by: James Contini <redacted>
Co-authored-by: Neha Abbas <redacted>
Co-authored-by: Abhijit Ramesh <redacted>
* Remove extra code and format
* Add ops documentation (finally)
* Update ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py
Co-authored-by: Sigbjørn Skjæret <redacted>
---------
Co-authored-by: James Contini <redacted>
Co-authored-by: Neha Abbas <redacted>
Co-authored-by: Abhijit Ramesh <redacted>
Co-authored-by: Sigbjørn Skjæret <redacted>
Jeff Bolz [Fri, 5 Dec 2025 20:21:57 +0000 (14:21 -0600)]
vulkan: enable mmvq for q2_k on NVIDIA (llama/17675)
Jeff Bolz [Fri, 5 Dec 2025 20:21:04 +0000 (14:21 -0600)]
vulkan: set all memory allocations to high priority (llama/17624)
* vulkan: set all memory allocations to high priority
* gate by env var
Georgi Gerganov [Fri, 5 Dec 2025 17:39:04 +0000 (19:39 +0200)]
rpc : fix alloc size logic (llama/17116)
* rpc : fix alloc size logic
* rpc : bump version
Georgi Gerganov [Fri, 5 Dec 2025 17:38:54 +0000 (19:38 +0200)]
metal : add residency sets keep-alive heartbeat (llama/17766)
* examples : add idle
* metal : attach residency sets to queue
* idle : add link
* idle : adjust intervals
* metal : add residency sets keep-alive heartbeat
* cont : adjust default keep-alive time
Johannes Gäßler [Fri, 5 Dec 2025 12:47:52 +0000 (13:47 +0100)]
HIP : fix RDNA4 build (llama/17792)
shalinib-ibm [Fri, 5 Dec 2025 11:41:51 +0000 (17:11 +0530)]
Q4/Q8 Tiled Gemm Optimization. (llama/16999)
Johannes Gäßler [Fri, 5 Dec 2025 08:18:10 +0000 (09:18 +0100)]
CUDA: fix FA VKQ accumulator overflow (llama/17746)
Jiacheng (Jason) Chen [Fri, 5 Dec 2025 08:17:37 +0000 (03:17 -0500)]
HIP: enable WMMA-MMQ INT kernels for RDNA 3 (llama/17576)
* enabled wmma instructions for most quantizations other than q2k
* fixed the last q2_k test case failure
* address comments: fix out of bound write for RDNA4, add comments after #endif
* clean up rebase: fix ne error in half2
* fix the EditorConfig CI
Piotr Wilkin (ilintar) [Thu, 4 Dec 2025 21:19:51 +0000 (22:19 +0100)]
Add support for CUMSUM and TRI for CUDA. (llama/17584)
* Add support for CUMSUM and TRI for CUDA.
* Minor optimizations.
* Correct warp_prefix_inclusive_sum in float2 variant to return float2
* Optimize TRI
* Whitespace
* Fix strides.
* Implement double loop
* Whitespace
* Fix HIP compilation bugs
* Optimizations + big case performance tests
* Implement using CUB with fallback to custom kernel
* Remove error message.
* Fixes from code review
* Comment out CPU-unsupported F16/BF16 cases to fix CI
* Fine, you win :P
* Fix last cast, use NO_DEVICE_CODE and GGML_UNUSED_VARS
* Vary warp-size based on physical warp size
* Add GGML_UNUSED_VARS in tri as well
* Use constexpr and call prefix_inclusive with warp_size template param
* Update ggml/src/ggml-cuda/cumsum.cu
Co-authored-by: Johannes Gäßler <redacted>
* Apply suggestions from code review
Co-authored-by: Johannes Gäßler <redacted>
* Change to tid % warp_size
* Fix strides; hardcode mask; add ggml_lane_mask_t
* Missing renames, remove unused get_warp_mask(), explicit calls to ggml_cuda_info()
* Too hasty...
---------
Co-authored-by: Johannes Gäßler <redacted>
Gabe Goodhart [Thu, 4 Dec 2025 17:12:19 +0000 (10:12 -0700)]
metal: TRI, FILL, EXPM1, SOFTPLUS (llama/16623)
* feat(wip): Port initial TRI impl from pervious work
The kernel does not work and is not optimized, but the
code compiles and runs, so this will be the starting point
now that the core op has been merged.
Branch: ggml-cumsum-tri
Signed-off-by: Gabe Goodhart <redacted>
* fix: Remove argument for constant val override
This was added in the original draft, but later removed. With this, the
kernel now passes tests.
Branch: ggml-cumsum-tri
Signed-off-by: Gabe Goodhart <redacted>
* feat: Move the ttype conditional to templating to avoid conditional in kernel
Branch: ggml-cumsum-tri
Signed-off-by: Gabe Goodhart <redacted>
* fix: Type fixes
Signed-off-by: Gabe Goodhart <redacted>
Co-authored-by: Georgi Gerganov <redacted>
Co-authored-by: Georgi Gerganov <redacted>
* feat: Add softplus for metal
Branch: ggml-cumsum-tri
Signed-off-by: Gabe Goodhart <redacted>
* feat: Add EXPM1 for metal
Branch: ggml-cumsum-tri
Signed-off-by: Gabe Goodhart <redacted>
* feat: Add FILL for metal
Branch: ggml-cumsum-tri
Signed-off-by: Gabe Goodhart <redacted>
* refactor: Branchless version of tri using _ggml_vec_tri_cmp as a mask
Branch: ggml-cumsum-tri
Signed-off-by: Gabe Goodhart <redacted>
* fix: Remove unused arguments
Branch: ggml-cumsum-tri
Signed-off-by: Gabe Goodhart <redacted>
* refactor: Use select instead of branch for softplus non-vec
Branch: ggml-cumsum-tri
Signed-off-by: Gabe Goodhart <redacted>
---------
Signed-off-by: Gabe Goodhart <redacted>
Co-authored-by: Georgi Gerganov <redacted>
Alberto Cabrera Pérez [Thu, 4 Dec 2025 12:16:38 +0000 (12:16 +0000)]
ggml-cpu : remove asserts always evaluating to false (llama/17728)
Georgi Gerganov [Thu, 4 Dec 2025 08:34:11 +0000 (10:34 +0200)]
metal : use params per pipeline instance (llama/17739)
Adrien Gallouët [Thu, 4 Dec 2025 06:04:02 +0000 (07:04 +0100)]
build : move _WIN32_WINNT definition to headers (llama/17736)
Previously, cmake was forcing `_WIN32_WINNT=0x0A00` for MinGW builds,
This caused "macro redefined" warnings with toolchains that define the version.
This also removes the `GGML_WIN_VER` variable as it is no longer needed.
Signed-off-by: Adrien Gallouët <redacted>
Herman Semenoff [Wed, 3 Dec 2025 21:03:19 +0000 (00:03 +0300)]
ggml-cpu: remove duplicate conditional check 'iid' (llama/17650)
Johannes Gäßler [Wed, 3 Dec 2025 15:57:05 +0000 (16:57 +0100)]
CUDA: generalized (mma) FA, add Volta support (llama/17505)
* CUDA: generalized (mma) FA, add Volta support
* use struct for MMA FA kernel config
---------
Co-authored-by: Aman Gupta <aman>
Georgi Gerganov [Wed, 3 Dec 2025 12:03:40 +0000 (14:03 +0200)]
metal : fix data race in pipeline library (llama/17731)
Reese Levine [Wed, 3 Dec 2025 09:25:34 +0000 (01:25 -0800)]
ggml webgpu: add support for emscripten builds (llama/17184)
* 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 <redacted>
Co-authored-by: Neha Abbas <redacted>
Co-authored-by: Reese Levine <redacted>
* 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 <redacted>
* 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 <redacted>
Jeff Bolz [Tue, 2 Dec 2025 18:22:04 +0000 (12:22 -0600)]
vulkan: Reduce temporary memory usage for TOP_K (llama/17623)
- 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.
xiaobing318 [Tue, 2 Dec 2025 17:50:57 +0000 (01:50 +0800)]
cmake : add utf8 compilation options for msvc (llama/17682)
Adrien Gallouët [Tue, 2 Dec 2025 16:21:11 +0000 (17:21 +0100)]
ggml : use svcntb() for SVE vector length detection (llama/17474)
Signed-off-by: Adrien Gallouët <redacted>
TianHao324 [Tue, 2 Dec 2025 12:35:23 +0000 (20:35 +0800)]
CANN: Disable Ger operator of OUT_PROD on 310p device (llama/17563)
Daniel Bevenius [Tue, 2 Dec 2025 11:52:45 +0000 (12:52 +0100)]
ggml : remove redundant n_copies check when setting input/output (llama/17612)
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.
Adrien Gallouët [Tue, 2 Dec 2025 08:41:26 +0000 (09:41 +0100)]
ggml : add fallback definition for HWCAP2_SVE2 (llama/17683)
This align with other HWCAP2 feature flags
See #17528
Signed-off-by: Adrien Gallouët <redacted>
Aman Gupta [Tue, 2 Dec 2025 04:36:31 +0000 (12:36 +0800)]
ggml-cuda: reorder only relevant nodes (llama/17639)
Neo Zhang Jianyu [Tue, 2 Dec 2025 00:56:46 +0000 (08:56 +0800)]
enhance argsort for UT (llama/17573)
Co-authored-by: Neo Zhang <redacted>
Georgi Gerganov [Mon, 1 Dec 2025 10:49:53 +0000 (12:49 +0200)]
metal : add FA head size 48 (llama/17619)
Georgi Gerganov [Mon, 1 Dec 2025 10:49:33 +0000 (12:49 +0200)]
ggml : extend the GGML_SCHED_NO_REALLOC debug logic of the scheduler (llama/17617)
Aman Gupta [Mon, 1 Dec 2025 09:12:48 +0000 (17:12 +0800)]
llama-graph: avoid expand_forward for fusion (llama/17633)
Tarek Dakhran [Sun, 30 Nov 2025 20:57:31 +0000 (21:57 +0100)]
model: LFM2-VL fixes (llama/17577)
* 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
Gilad S. [Sun, 30 Nov 2025 02:00:59 +0000 (04:00 +0200)]
ggml: fix: macOS build with `-DGGML_BACKEND_DL=ON` (llama/17581)
Aman Gupta [Sun, 30 Nov 2025 00:17:55 +0000 (08:17 +0800)]
CUDA: add stream-based concurrency (llama/16991)
* CUDA: add stream-based concurrency
* HIP: fix hipStreamWaitEvent define and nodiscard warnings
* ggml-cuda: fix fusion inside stream
* ggml-cuda: fix bug w.r.t first stream launch
* ggml-cuda: format
* ggml-cuda: improve assert message
* ggml-cuda: use lambda instead of duplicating code
* ggml-cuda: add some more comments
* ggml-cuda: add more detailed comments about concurrency
* ggml-cuda: rename + remove unused var
* ggml-cuda: fix condition for stream launch
* ggml-cuda: address review comments, add destructor
* common.cuh: add is_valid for concurrent events
* common.cuh: make comment better
* update comment
Co-authored-by: Johannes Gäßler <redacted>
* update comment
Co-authored-by: Johannes Gäßler <redacted>
* common.cuh: fix lower_bound condition + remove join_node data from write_ranges
* ggml-cuda: fix overlap condition + shadowing parameter
---------
Co-authored-by: Carl Philipp Klemm <redacted>
Co-authored-by: Johannes Gäßler <redacted>
Mahekk Shaikh [Sun, 30 Nov 2025 00:16:28 +0000 (19:16 -0500)]
cuda : add error checking for cudaMemcpyAsync in argsort (llama/17599)
* cuda : add error checking for cudaMemcpyAsync in argsort (llama/12836)
* fix indentation
Acly [Sun, 30 Nov 2025 00:03:21 +0000 (01:03 +0100)]
vulkan : fix FA mask load with bounds check (coopmat2) (llama/17606)
Neo Zhang [Sat, 29 Nov 2025 12:59:44 +0000 (20:59 +0800)]
sycl : support to malloc memory on device more than 4GB, update the doc and script (llama/17566)
Co-authored-by: Neo Zhang Jianyu <redacted>
ixgbe [Sat, 29 Nov 2025 12:56:31 +0000 (20:56 +0800)]
ggml: replace hwcap with riscv_hwprobe for RVV detection (llama/17567)
Signed-off-by: Wang Yang <redacted>
Ruben Ortlam [Sat, 29 Nov 2025 08:37:22 +0000 (09:37 +0100)]
Vulkan: MMVQ Integer Dot K-Quant and MUL_MAT_ID support (llama/16900)
* vulkan: split mul_mmq_funcs for mul_mat_vecq use
* add mxfp4 mmvq
* add q2_k mmvq
* add q3_k mmvq
* add q4_k and q5_k mmvq
* add q6_k mmvq
* handle 4x4 quants per mmvq thread
* enable MUL_MAT_ID mmvq support
* enable subgroup optimizations for mul_mat_vec_id shaders
* device tuning
* request prealloc_y sync after quantization
* fix indentation
* fix llvmpipe test failures
* fix mul_mat_id mmvq condition
* fix unused variable warning
Jeff Bolz [Sat, 29 Nov 2025 07:39:57 +0000 (01:39 -0600)]
vulkan: improve topk perf for large k, fix overflow in unit tests (llama/17582)
Diego Devesa [Fri, 28 Nov 2025 15:33:23 +0000 (07:33 -0800)]
ggml : add GGML_SCHED_NO_REALLOC option to disable reallocations in ggml_backend_sched (llama/17276)
* 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 <redacted>
R0CKSTAR [Fri, 28 Nov 2025 13:08:29 +0000 (21:08 +0800)]
enable fp16/fast_fp16/bf16_mma on PH1 (llama/17551)
* [MUSA] enable fp16/fast_fp16/bf16_mma on PH1
Signed-off-by: Xiaodong Ye <redacted>
* Update ggml/src/ggml-cuda/fattn-vec.cuh
Co-authored-by: Johannes Gäßler <redacted>
* Update ggml/src/ggml-cuda/fattn-vec.cuh
Co-authored-by: Johannes Gäßler <redacted>
* Update ggml/src/ggml-cuda/fattn-tile.cuh
Co-authored-by: Johannes Gäßler <redacted>
* Address review comments
Signed-off-by: Xiaodong Ye <redacted>
---------
Signed-off-by: Xiaodong Ye <redacted>
Co-authored-by: Johannes Gäßler <redacted>
Aman Gupta [Fri, 28 Nov 2025 12:34:51 +0000 (20:34 +0800)]
ggml-cuda: add stricter checking for fusion (llama/17568)
* ggml-cuda: make conditions for fusion more explicit
* ggml-cuda: remove size check as std::equal already does it
Piotr Wilkin (ilintar) [Fri, 28 Nov 2025 11:02:56 +0000 (12:02 +0100)]
model : Qwen3 Next (llama/16095)
* Qwen3 Next - cleaned up version
* Whitespaces and stuff
* Correct minor errors
* Update src/llama-model.cpp
Co-authored-by: Sigbjørn Skjæret <redacted>
* 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 <redacted>
* Usual suspects
* fix my bad suggestion
---------
Co-authored-by: Sigbjørn Skjæret <redacted>
Co-authored-by: Georgi Gerganov <redacted>
Johannes Gäßler [Fri, 28 Nov 2025 09:29:09 +0000 (10:29 +0100)]
CUDA: no FP16 arithmetic for vector FA kernel (llama/17558)
Jeff Bolz [Fri, 28 Nov 2025 09:07:29 +0000 (03:07 -0600)]
vulkan: Implement GGML_OP_TRI (llama/17503)
* vulkan: Implement GGML_OP_TRI
* check types match
Radoslav Gerganov [Fri, 28 Nov 2025 08:33:51 +0000 (10:33 +0200)]
rpc : cache and reuse compute graphs (llama/15405)
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.
yulo [Fri, 28 Nov 2025 07:24:30 +0000 (15:24 +0800)]
HIP: enable mul_mat_f for RDNA4 (llama/17437)
* 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 <redacted>
Piotr Wilkin (ilintar) [Fri, 28 Nov 2025 04:15:32 +0000 (05:15 +0100)]
SOLVE_TRI CUDA kernel for small matrices (llama/17457)
Neo Zhang Jianyu [Fri, 28 Nov 2025 00:50:56 +0000 (08:50 +0800)]
refactor pad_reflect_1d to make the UT case pass (llama/17204)
Co-authored-by: Zhang Jianyu <redacted>
Jeff Bolz [Thu, 27 Nov 2025 14:48:00 +0000 (08:48 -0600)]
vulkan: Implement SOLVE_TRI (llama/17486)
* vulkan: Implement SOLVE_TRI
* load B matrix through shared memory
* use FLOAT_TYPE
matt23654 [Thu, 27 Nov 2025 11:35:35 +0000 (11:35 +0000)]
cuda : fix UMA detection on discrete GPUs. (llama/17537)
Alberto Cabrera Pérez [Thu, 27 Nov 2025 11:25:14 +0000 (11:25 +0000)]
ggml-cpu: aarm64: q4_K repack gemm and gemv implementations (dotprod only) (llama/17494)
* 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 <redacted>
* Undo acc rename
Signed-off-by: Alberto Cabrera <redacted>
* Q4_K arm dotprod gemm
Signed-off-by: Alberto Cabrera <redacted>
* Fix: q4_qs reinterpret from uint to int
Signed-off-by: Alberto Cabrera <redacted>
* 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 <redacted>
Co-authored-by: Georgi Gerganov <redacted>