56e219e035
all: updated Github links to uxlfoundation
2025-03-12 15:08:59 -07:00
2095256d96
fixup: benchdnn: parser: check for dangling symbol at get_substr
...
Parsing values for two global objects in a single function must be done
carefully.
2025-03-12 14:52:26 -07:00
8323d89198
benchdnn: parser: check for dangling symbol at get_substr
2025-03-11 14:46:35 -07:00
143bf7e521
benchdnn: cold-cache: introduce cold-tlb support
2025-02-27 16:19:22 -08:00
af71de057d
benchdnn: parser: fix help message
2025-02-27 16:19:22 -08:00
b47f23a081
benchdnn: parser: improve engine parsing
2025-02-21 15:21:19 -08:00
38b450996b
benchdnn: prim_ref impl filter ignores values from global-(skip-)impl
2025-02-18 13:39:30 -08:00
85a408bb35
benchdnn: introduce --summary option with failures support
2025-02-02 21:30:07 -08:00
c4e4c31780
tests: benchdnn: add support for graph execution
...
Currently limited to SYCL with L0 backend.
2025-01-17 16:44:06 -08:00
a6dfddf0d1
benchdnn: remove out-dated fast-ref-gpu option
2025-01-16 21:14:22 -08:00
337c51c703
benchdnn: rename no_host_memory and relax use restrictions
...
Renames no_host_memory to no_ref_memory to more accurately reflect its
usage. Removes restrictions on usage and enables it by default for
bench_mode_t::exec.
2024-11-19 11:38:37 -08:00
cc73f12a63
benchdnn: introduce --global-impl and --global-skip-impl support
2024-11-05 14:50:59 -08:00
3a2db9184d
benchdnn: introduce --impl support
...
`--skip-impl` knob moves into a local context. The option now respects
`--reset` option.
2024-11-05 14:50:59 -08:00
29cc8574b4
benchdnn: parser: remove a check for engine index
2024-10-21 17:19:10 -07:00
12cf2ef6aa
src, tests: fix static code analysis hits
2024-09-05 12:00:50 -07:00
d95fdcb155
benchdnn: add rounding_mode attribute testing
2024-08-28 14:31:54 +02:00
ed9f6d1b08
x64, api, benchdnn: revert dice post op and related changes
...
Reverts the following commits:
* 8bfcff59efbf5c9846d7f75a2a847e35676d582c verbose_converter: add prelu with scaleshift support
* d3ffb808d5e20068630c4052dbd0b37b7721b59c x64: add jitted prelu with scaleshift postop
* f2c62e01799eb62d68c2c0c5a348fe16d3dc6e6d x64: properly initialize postop injector constant tables
* 6ae2d9cc8c485fe4c8db720752324c871c7a302f benchdnn: add prelu postop with scaleshift coverage
* c11bab0c4970886c84eee1c2e7e98266d05dfa35 api: add dice post op
2024-08-06 13:46:23 -07:00
6ae2d9cc8c
benchdnn: add prelu postop with scaleshift coverage
2024-08-05 09:49:29 +02:00
f1ede37b97
src: cpu: dropout: add dropout attribute
2024-06-27 13:30:04 -07:00
8fdbf3ad34
benchdnn: introduce a ref-impl check knob
2024-05-09 14:48:58 -07:00
2072168f70
benchdnn: parser: replace attributes method with a consolidated one
2024-04-16 16:45:59 -07:00
7e00d5e3cb
benchdnn: introduce numerical determinism testing mode
2024-01-25 23:06:57 -08:00
a3b1a1481c
tests: benchdnn: fpmath: force -> apply_to_int
2024-01-22 14:42:02 -08:00
461cffcbd5
tests: benchdnn: align fpmath_mode with new api
2024-01-22 14:42:02 -08:00
fd06a76cf6
benchdnn: introduce attr-deterministic support
2024-01-22 11:59:09 -08:00
7beb995603
benchdnn: rename fast-ref-gpu option into fast-ref
...
Since CPU will utilize it as well
2023-12-15 09:56:19 -08:00
6b6222422a
benchdnn: doc and help additions
2023-12-12 10:33:25 -08:00
c900643a3f
benchdnn: introduce n_streams knob support
2023-12-07 11:53:44 -08:00
84dca975e9
tests: add accumulation_mode attribute support
2023-11-13 15:33:14 +01:00
2209f83d8b
benchdnn: remove out-dated dw_ksp types
2023-10-19 14:16:59 -07:00
e46ea99def
benchdnn: update parsing routines and doc to reflect actual state
2023-10-19 14:16:59 -07:00
5c70fa221e
benchdnn: introduce a stream kind knob
2023-09-28 12:15:46 -07:00
855f903f16
benchdnn: introduce cold cache
2023-08-15 23:37:12 -07:00
1244d13644
benchdnn: parser: restrict empty dims input and improve internal doc
2023-08-07 13:51:36 -07:00
d45e84f7b4
fixup: benchdnn: extend binary post-op with mask_input notation
2023-07-28 13:45:31 -07:00
04d1f1d3bd
benchdnn: extend binary post-op with mask_input notation
...
It allows to pass mask as plain integer value. E.g. `per_tensor` policy
input for 2D-spatial tensor is interchangeble with just `15`.
2023-07-26 12:03:48 -07:00
a1a921950f
benchdnn: parser: move and update post-ops parsing routine
2023-07-26 12:03:48 -07:00
6f74af80dd
benchdnn: parser: replace unsafe atof with safe-n-talking version
2023-07-26 12:03:48 -07:00
06b561c255
benchdnn: parser: replace unsafe atoi with safe-n-talking version
2023-07-26 12:03:48 -07:00
5af344e166
benchdnn: add --repeats-per-prb option
2023-05-16 12:28:39 -07:00
3d20e5f1ac
benchdnn: introduce --mode=F alias for "fast" performance mode
2023-02-24 10:27:08 -08:00
a53a77de71
benchdnn: introduce no_host_memory mode modifier
2023-02-24 10:27:08 -08:00
81498dbc4d
benchdnn: introduce parallel test object creation modifier
2023-02-24 10:27:08 -08:00
0811db669d
benchdnn: re-factor mode and modifiers
2023-02-24 10:27:08 -08:00
257915978a
benchdnn: enable profiling mode by default
2023-02-24 10:27:08 -08:00
6c71ceaba9
benchdnn: dump performance options
2023-02-24 10:27:08 -08:00
df7d26cd68
benchdnn: move max-ms-per-prb lower bound to 10
2023-02-24 10:27:08 -08:00
c0c64cdc29
benchdnn: enable matmul driver for CSR encoding
2023-02-15 13:51:12 -08:00
5a97d070d7
benchdnn: utils: introduce a ctor for prb_vdims
2023-02-03 15:11:29 -08:00
c50897365f
benchdnn: add new options to specify encoding and sparsity
2023-02-01 11:58:41 -08:00
