0ba6d85080
x64: matmul: Enable grouped ZP for per_oc/per_ocic & rework scales
2025-10-18 16:47:57 +02:00
051b020bb1
x64: matmul: Enable AVX512 f32:int4/int8:f32 case
2025-10-18 16:47:57 +02:00
7b351ae955
src: gpu: intel: gemm: add strided batch support to group sums
2025-10-16 14:30:24 -07:00
bf6617dd1f
tests: benchdnn: utils: improve --buffer-prefix messaging
2025-10-15 18:09:06 -07:00
a4874e48ea
tests: sdpa: fix fallthrough
2025-10-15 09:08:56 -07:00
21adf4d977
tests: sdpa: fix typo
2025-10-15 09:08:56 -07:00
27e186ccd9
gtest: check returned ptr on expected md init failure
2025-10-15 10:42:00 +02:00
de84f7f345
benchdnn: properly lock before checking allocations size
2025-10-15 10:42:00 +02:00
34663d344a
benchdnn: fix reference check in init_memory
2025-10-15 10:42:00 +02:00
5331f9e068
gtest: cover mx input and output scaling in iface tests
2025-10-15 02:01:20 +02:00
76fd406dd4
benchdnn: matmul: add support for mx quantization testing
2025-10-15 02:01:20 +02:00
d5e144c943
tests: benchdnn: enable mode p on windows
2025-10-14 14:28:23 -07:00
7b95adc9fe
benchdnn: ref_post_op: eltwise: fix NAN propagation
2025-10-13 12:32:33 -07:00
15006e1167
benchdnn: graph: avoid vector copy
2025-10-13 16:15:08 +08:00
1eb43285fc
benchdnn: graph: flex_rewrite: remove deadcode after the macro SAFE
2025-10-13 16:15:08 +08:00
b18ed0017f
benchdnn: graph: displacer: return FAIL for more states
...
This should fix a nullptr dereference warning on the below line.
2025-10-13 16:15:08 +08:00
95d831e606
benchdnn: inputs: graph: replace multiply with add to pass acc check
2025-10-11 14:39:24 +08:00
105a7cd7fa
tests: prepare Eigen_threadpool implementation for new Eigen versions
2025-10-10 16:09:37 -07:00
48872b1a33
cpu: aarch64: fix segfault in eltwise_log injector ( #4063 )
...
Signed-off-by: Siddhartha Menon <siddhartha.menon@arm.com >
2025-10-10 11:40:48 +01:00
bcc0ca0084
tests: fix clang-tidy failures ( #4082 )
...
Signed-off-by: Siddhartha Menon <siddhartha.menon@arm.com >
2025-10-08 10:29:27 +01:00
aa697c9dae
benchdnn: xnorm: skip filling for empty tensors
2025-10-03 10:13:46 -07:00
d00d6d75a4
benchdnn: inputs: gnorm: introduce smoke list
2025-10-02 10:16:37 -07:00
e81b095168
fixup: benchdnn: conv: ref: update the kernel to remove excessive memory access
2025-10-02 10:16:37 -07:00
efb859c258
fixup: benchdnn: eltwise: clean up excessive memory objects
2025-10-02 10:16:37 -07:00
4431d29090
cpu: aarch64: enable brgconv bf16 on sve 128 ( #4041 )
2025-10-02 16:14:05 +01:00
d8266e1ccf
tests: benchdnn: matmul: add gs16 weights decompression cases
2025-10-01 09:17:30 -07:00
534aeb36b6
tests: benchdnn: matmul: ref: support different group sizes
2025-10-01 09:17:30 -07:00
a6598aad0c
tests: benchdnn: graph: update f32 gqa training case
2025-09-30 11:35:36 +08:00
38338d09b7
benchdnn: eltwise: clean up excessive memory objects
2025-09-29 13:28:23 -07:00
9203afa49b
benchdnn: add scratchpad memory to mapped buffers
2025-09-29 13:28:23 -07:00
733781e65c
revert: benchdnn: manually unmap scratchpad memory object
...
This reverts commit c8a746883dd3c00419f0580215543015879466e1.
2025-09-29 13:28:23 -07:00
fc59bc8720
tests: benchdnn: inputs: graph: update bf16/f16 matmul+post_ops pattern
2025-09-28 09:43:57 +08:00
8bcb8056ba
benchdnn: inputs: remove unsupported cases
2025-09-26 10:29:00 -07:00
d40bb8bb72
benchdnn: inputs: matmul: binary post-op bcast coverage for 3d cases ( #3038 )
2025-09-26 19:00:12 +02:00
d8cdbf0cee
gtests: graph: api: negative cases for host scalar
2025-09-25 11:24:51 +08:00
e9f5ef3f31
tests: benchdnn: graph: skip unsupported acc mode on cpu
2025-09-24 09:03:25 +08:00
4c8189f4dc
fixup: fixup: benchdnn: utils: memory: add device memory tracking
2025-09-22 09:05:19 -07:00
8bb071165e
benchdnn: introduce implementation summary option support
2025-09-19 14:52:01 -07:00
de710a35d1
tests: benchdnn: graph: add case for gqa training backward
2025-09-19 09:30:59 +08:00
b87172ef8f
tests: benchdnn: graph: suppport gqa training bwd validation
2025-09-19 09:30:59 +08:00
745a069f7f
tests: benchdnn: graph: add case for gqa training forward
2025-09-19 09:30:59 +08:00
561b57224b
benchdnn: matmul: add test cases for mxfp inputs
2025-09-18 21:44:56 +02:00
9ed16271fe
fixup: benchdnn: utils: memory: add device memory tracking
...
This fixes CPU SYCL scenario and starts supporting any memory kind.
2025-09-18 12:11:59 -07:00
0cad313286
fixup: cpu: postops: dt support check
2025-09-18 12:10:00 -07:00
abfeb10e1d
benchdnn: inputs: conv: adjust wei scale value
2025-09-18 12:10:00 -07:00
72dce45a0a
benchdnn: inputs: graph: test sdpa with acc mode rewrite
2025-09-18 13:10:25 +08:00
15dfd448e9
benchdnn: graph: utils: supports acc mode attribute
2025-09-18 13:10:25 +08:00
e6dcd1ccea
benchdnn: graph: rewrite op attr w/o the attr in json
2025-09-18 13:10:25 +08:00
c8a746883d
benchdnn: manually unmap scratchpad memory object
...
When mode-modifier=P is specified, it changes the scratchpad mode to
user, and automatically converts the scratchpad memory object into real
opposed to library when it'll be allocated inside the library.
2025-09-15 14:03:50 -07:00
337bf45b5c
tests: benchdnn: add capability to import buffers from files
2025-09-15 13:24:12 -07:00
