3f66dc5049
third_party: ngen: avoid unnecessary copy
2025-10-15 09:08:56 -07:00
8f70dafafd
third_party: ngen: auto-SWSB: fix early A@ dependency clearing bug
2025-10-14 15:38:41 -07:00
54c4350589
ngen: workaround for HW bug with cross-pipe cmod WAR hazards
2025-10-07 10:11:44 -07:00
c0c069ef2f
ngen: fix SIMD2 dst stride
2025-09-12 08:07:45 -07:00
c793a7a35f
ngen: workaround gcc13 bitfield assignment bug
...
Assignments to *VS0 and *VS1 were being incorrectly optimized away by GCC in
some scenarios.
2025-09-08 10:47:54 -07:00
eddad2d50c
third_party: ngen: fix build error with gcc 8.5
2025-09-02 15:09:03 -07:00
fbea4cd863
third_party: ngen: fix isARF()
2025-08-28 16:38:02 -07:00
0d6b37c137
third_party, src: update nGEN
2025-08-27 17:52:51 -07:00
337b34867a
ngen: remove conflict with windows.h interface define
2025-08-19 10:14:37 -07:00
831220021b
ngen: add ELFCodeGenerator interface setter
2025-08-07 06:32:32 -07:00
6750723b59
xe: jit: ir: enable ngen interface to kernel_iface_t construction
2025-08-05 06:52:04 -07:00
c6e9be775e
Cpu: update Xbyak to v7.28 ( #3682 )
2025-07-31 12:00:12 -04:00
871e4aac01
third_party: spdlog: upgrade version to 1.15.3
2025-07-10 11:44:44 -07:00
1afce61b82
ngen: improve 64-bit emul
...
In some cases, emulation state may not have any registers so 64-bit
multiplication with an immediate could fail. In particular,
07c15ce29a421a49a46e361fc88e67aff1afbaca happened to enable just such a scenario
causing failures for:
```
benchdnn --reorder --engine=gpu --allow-enum-tags-only=false --sdt=f32 --ddt=f32 --stag=ABcd32a16b --dtag=abcd 1024x256x56x56
```
2025-07-08 07:55:50 -07:00
43cef3c738
cpu: x64: brgemm: enable mem advice support for NVL ( #3513 )
2025-07-07 21:19:27 -04:00
90dc538525
ngen: rename ELFCodeGenerator::interface
...
The windows.h header contains the following define:
```
#define interface struct
```
This adversely interacts with the ELFCodeGenerator::interface function and
produces a compilation error. To avoid this interaction, the function
is renamed.
2025-07-07 13:41:35 -07:00
b565e22dee
ngen: improve interface for injectors
2025-07-02 11:20:04 -07:00
03344bbc29
ngen: fix correctness with ZET_ENABLE_PROGRAM_DEBUGGING
2025-07-02 11:20:04 -07:00
42f34b0378
ngen: improve emulated emul with Immediate
2025-07-02 11:20:04 -07:00
66eec0e06b
third_party: level_zero: update headers to v1.13.1
2025-07-01 16:59:03 -07:00
6c2cc18b56
xbyak: allow repeated assignment of the same opmask index ( #3407 )
2025-06-24 16:03:00 +02:00
0bfca9daed
third_party: ngen: fixup region advance for emulation
2025-05-28 15:07:06 -07:00
492e96510d
cpu: x64: enable hf8 support for matmul on AVX10.2
2025-05-23 17:17:10 -04:00
1c825b46bb
build: enable Wundef
2025-05-09 16:52:00 -07:00
a21297e5d3
third_party: gtest: work around Wundef hits from gtest.h
...
The better solution is to create a single entry point to gtest.h from
dnnl_gtest_common.hpp and use a proxy header file technique, but this is
the improvement for the next time.
2025-05-09 16:52:00 -07:00
7edc35c37c
third_party: itt: fix undefined macro hit
2025-05-09 16:52:00 -07:00
924f90e983
third_party: ngen: enable bf8 emulated mov
2025-05-07 16:51:37 -07:00
b7136b91d5
ngen: address coverity hits
2025-05-06 22:37:21 -07:00
8a8dedc07c
third_party: ngen: fix locale handling
2025-05-06 07:58:09 -07:00
2804466041
fixup: xe: FP precision casting warning fix ( #3178 )
2025-04-29 10:07:07 -07:00
4b8fc9b01e
cpu: x64: upstream DMR support to main branch
...
Co-authored-by: Putta, Srinivas <srinivas.putta@intel.com >
Co-authored-by: Yair Nahum <yair.y.nahum@intel.com >
Co-authored-by: Palicki, Stefan <stefan.palicki@intel.com >
Co-authored-by: Andrey Kalinin <andrey.kalinin@intel.com >
Co-authored-by: Tomasz Czeszun <tomasz.czeszun@intel.com >
Co-authored-by: Yair Nahum <64333457+ynahum@users.noreply.github.com >
2025-04-23 22:34:55 -04:00
8605df92d6
third_party: ngen: extend emulated mov to qword types
2025-04-23 07:26:55 -07:00
f1131cb2d6
third_party: ngen: extend emul for qword src0
2025-04-23 07:26:55 -07:00
47b206c5e2
third_party: ngen: use getHardware interface
2025-04-23 07:26:55 -07:00
5f001dd373
third_party: ngen: add missing defaults
2025-04-17 11:32:28 -07:00
37bda40e31
third_party: level_zero: removed unused Python scripts
2025-04-09 14:51:26 -07:00
6d8a780a35
ngen: add cross-thread data relocations for GTPin
2025-04-09 11:14:16 -07:00
7b395787eb
ngen: workaround for SYCL + GCC 12.3 compiler bug
2025-04-02 10:01:08 -07:00
9bef39e84d
ngen: fix missing field initialization warning
2025-03-28 07:18:38 -07:00
eaaf1c0847
third_party: ngen: prepare for SYCL generator usage
2025-03-27 13:13:24 -07:00
931cc27d80
ngen: downstream nGEN
2025-03-25 17:12:32 -07:00
607a318d26
ngen: update PVC WAR bug workaround
2025-03-24 15:19:47 -07:00
34f6711818
third_party: spdlog: upgrade version to 1.15.1
2025-03-21 10:05:28 -07:00
b6fab3ecda
ngen: add getHardware() interface
2025-03-19 13:58:17 -07:00
37202682c4
ngen: improve macro fowarding to arbitrary scopes
...
Additionally, includes various tweaks required to get AsmCodeGenerator
to function when forwarded.
2025-03-19 13:58:17 -07:00
7e8504f617
third_party: ngen: fixup insn restrictions
2025-03-19 09:10:57 -07:00
8c811df174
Xbyak: upgrade xbyak version to v7.23.1 ( #2867 )
2025-03-17 18:31:22 -04:00
055d8cb522
xe: BMG/LNL product detection
2025-03-12 14:03:35 -04:00
b86e408e15
ngen: fix various warnings due to NGEN_ASM
2025-03-04 06:04:01 -08:00
6cd1beaea3
xe: enable building with out-of-tree nGEN
2025-03-04 06:04:01 -08:00
