Linux sh52.ich-4.com 5.14.0-611.26.1.el9_7.x86_64 #1 SMP PREEMPT_DYNAMIC Thu Jan 29 05:24:47 EST 2026 x86_64
LiteSpeed
Server IP : 198.143.147.58 & Your IP : 216.73.217.21
Domains :
Cant Read [ /etc/named.conf ]
User : actualbuzz
Terminal
Auto Root
Create File
Create Folder
Localroot Suggester
Backdoor Destroyer
Readme
/
lib /
clang /
20 /
include /
Delete
Unzip
Name
Size
Permission
Date
Action
cuda_wrappers
[ DIR ]
drwxr-xr-x
2026-02-03 15:09
fuzzer
[ DIR ]
drwxr-xr-x
2026-02-03 15:08
llvm_libc_wrappers
[ DIR ]
drwxr-xr-x
2026-02-03 15:09
llvm_offload_wrappers
[ DIR ]
drwxr-xr-x
2026-02-03 15:09
openmp_wrappers
[ DIR ]
drwxr-xr-x
2026-02-03 15:09
orc
[ DIR ]
drwxr-xr-x
2026-02-03 15:08
ppc_wrappers
[ DIR ]
drwxr-xr-x
2026-02-03 15:09
profile
[ DIR ]
drwxr-xr-x
2026-02-03 15:08
sanitizer
[ DIR ]
drwxr-xr-x
2026-02-03 15:08
xray
[ DIR ]
drwxr-xr-x
2026-02-03 15:08
zos_wrappers
[ DIR ]
drwxr-xr-x
2026-02-03 15:09
__clang_cuda_builtin_vars.h
4.78
KB
-rw-r--r--
2025-07-08 23:06
__clang_cuda_cmath.h
18.06
KB
-rw-r--r--
2025-07-08 23:06
__clang_cuda_complex_builtins.h
9.36
KB
-rw-r--r--
2025-07-08 23:06
__clang_cuda_device_functions.h
57.36
KB
-rw-r--r--
2025-07-08 23:06
__clang_cuda_intrinsics.h
29.87
KB
-rw-r--r--
2025-07-08 23:06
__clang_cuda_libdevice_declares.h
21.89
KB
-rw-r--r--
2025-07-08 23:06
__clang_cuda_math.h
16.21
KB
-rw-r--r--
2025-07-08 23:06
__clang_cuda_math_forward_declares.h
8.27
KB
-rw-r--r--
2025-07-08 23:06
__clang_cuda_runtime_wrapper.h
17.6
KB
-rw-r--r--
2025-07-08 23:06
__clang_cuda_texture_intrinsics.h
31.86
KB
-rw-r--r--
2025-07-08 23:06
__clang_hip_cmath.h
26.48
KB
-rw-r--r--
2025-07-08 23:06
__clang_hip_libdevice_declares.h
19.87
KB
-rw-r--r--
2025-07-08 23:06
__clang_hip_math.h
32.78
KB
-rw-r--r--
2025-07-08 23:06
__clang_hip_runtime_wrapper.h
4.94
KB
-rw-r--r--
2025-07-08 23:06
__clang_hip_stdlib.h
1.19
KB
-rw-r--r--
2025-07-08 23:06
__stdarg___gnuc_va_list.h
467
B
-rw-r--r--
2025-07-08 23:06
__stdarg___va_copy.h
445
B
-rw-r--r--
2025-07-08 23:06
__stdarg_header_macro.h
417
B
-rw-r--r--
2025-07-08 23:06
__stdarg_va_arg.h
793
B
-rw-r--r--
2025-07-08 23:06
__stdarg_va_copy.h
451
B
-rw-r--r--
2025-07-08 23:06
__stdarg_va_list.h
448
B
-rw-r--r--
2025-07-08 23:06
__stddef_header_macro.h
417
B
-rw-r--r--
2025-07-08 23:06
__stddef_max_align_t.h
857
B
-rw-r--r--
2025-07-08 23:06
__stddef_null.h
875
B
-rw-r--r--
2025-07-08 23:06
__stddef_nullptr_t.h
958
B
-rw-r--r--
2025-07-08 23:06
__stddef_offsetof.h
708
B
-rw-r--r--
2025-07-08 23:06
__stddef_ptrdiff_t.h
717
B
-rw-r--r--
2025-07-08 23:06
__stddef_rsize_t.h
710
B
-rw-r--r--
2025-07-08 23:06
__stddef_size_t.h
708
B
-rw-r--r--
2025-07-08 23:06
__stddef_unreachable.h
735
B
-rw-r--r--
2025-07-08 23:06
__stddef_wchar_t.h
853
B
-rw-r--r--
2025-07-08 23:06
__stddef_wint_t.h
443
B
-rw-r--r--
2025-07-08 23:06
__wmmintrin_aes.h
5.15
KB
-rw-r--r--
2025-07-08 23:06
__wmmintrin_pclmul.h
1.99
KB
-rw-r--r--
2025-07-08 23:06
adcintrin.h
5.43
KB
-rw-r--r--
2025-07-08 23:06
adxintrin.h
3.37
KB
-rw-r--r--
2025-07-08 23:06
altivec.h
698.45
KB
-rw-r--r--
2025-07-08 23:06
amdgpuintrin.h
7.4
KB
-rw-r--r--
2025-07-08 23:06
ammintrin.h
7.56
KB
-rw-r--r--
2025-07-08 23:06
amxavx512intrin.h
12.67
KB
-rw-r--r--
2025-07-08 23:06
amxbf16transposeintrin.h
3.44
KB
-rw-r--r--
2025-07-08 23:06
amxcomplexintrin.h
6.81
KB
-rw-r--r--
2025-07-08 23:06
amxcomplextransposeintrin.h
11.78
KB
-rw-r--r--
2025-07-08 23:06
amxfp16intrin.h
3.25
KB
-rw-r--r--
2025-07-08 23:06
amxfp16transposeintrin.h
3.44
KB
-rw-r--r--
2025-07-08 23:06
amxfp8intrin.h
8.63
KB
-rw-r--r--
2025-07-08 23:06
amxintrin.h
19.83
KB
-rw-r--r--
2025-07-08 23:06
amxmovrsintrin.h
2.24
KB
-rw-r--r--
2025-07-08 23:06
amxmovrstransposeintrin.h
9.18
KB
-rw-r--r--
2025-07-08 23:06
amxtf32intrin.h
3.48
KB
-rw-r--r--
2025-07-08 23:06
amxtf32transposeintrin.h
3.56
KB
-rw-r--r--
2025-07-08 23:06
amxtransposeintrin.h
10.6
KB
-rw-r--r--
2025-07-08 23:06
arm64intr.h
993
B
-rw-r--r--
2025-07-08 23:06
arm_acle.h
29.97
KB
-rw-r--r--
2025-07-08 23:06
arm_bf16.h
548
B
-rw-r--r--
2025-09-17 20:10
arm_cde.h
32.67
KB
-rw-r--r--
2025-09-17 20:10
arm_cmse.h
6.21
KB
-rw-r--r--
2025-07-08 23:06
arm_fp16.h
16.94
KB
-rw-r--r--
2025-09-17 20:10
arm_mve.h
1.48
MB
-rw-r--r--
2025-09-17 20:10
arm_neon.h
2.66
MB
-rw-r--r--
2025-09-17 20:10
arm_neon_sve_bridge.h
9.48
KB
-rw-r--r--
2025-07-08 23:06
arm_sme.h
223.36
KB
-rw-r--r--
2025-09-17 20:10
arm_sve.h
2.03
MB
-rw-r--r--
2025-09-17 20:10
arm_vector_types.h
9.86
KB
-rw-r--r--
2025-09-17 20:10
armintr.h
843
B
-rw-r--r--
2025-07-08 23:06
avx10_2_512bf16intrin.h
22.87
KB
-rw-r--r--
2025-07-08 23:06
avx10_2_512convertintrin.h
12.88
KB
-rw-r--r--
2025-07-08 23:06
avx10_2_512minmaxintrin.h
7.35
KB
-rw-r--r--
2025-07-08 23:06
avx10_2_512niintrin.h
13.71
KB
-rw-r--r--
2025-07-08 23:06
avx10_2_512satcvtdsintrin.h
13.48
KB
-rw-r--r--
2025-07-08 23:06
avx10_2_512satcvtintrin.h
16.6
KB
-rw-r--r--
2025-07-08 23:06
avx10_2bf16intrin.h
43.63
KB
-rw-r--r--
2025-07-08 23:06
avx10_2convertintrin.h
22.75
KB
-rw-r--r--
2025-07-08 23:06
avx10_2copyintrin.h
2.35
KB
-rw-r--r--
2025-07-08 23:06
avx10_2minmaxintrin.h
13.53
KB
-rw-r--r--
2025-07-08 23:06
avx10_2niintrin.h
16.6
KB
-rw-r--r--
2025-07-08 23:06
avx10_2satcvtdsintrin.h
15.36
KB
-rw-r--r--
2025-07-08 23:06
avx10_2satcvtintrin.h
18.87
KB
-rw-r--r--
2025-07-08 23:06
avx2intrin.h
188.12
KB
-rw-r--r--
2025-07-08 23:06
avx512bf16intrin.h
10.58
KB
-rw-r--r--
2025-07-08 23:06
avx512bitalgintrin.h
2.56
KB
-rw-r--r--
2025-07-08 23:06
avx512bwintrin.h
76.02
KB
-rw-r--r--
2025-07-08 23:06
avx512cdintrin.h
4.23
KB
-rw-r--r--
2025-07-08 23:06
avx512dqintrin.h
58.87
KB
-rw-r--r--
2025-07-08 23:06
avx512fintrin.h
383.45
KB
-rw-r--r--
2025-07-08 23:06
avx512fp16intrin.h
156.92
KB
-rw-r--r--
2025-07-08 23:06
avx512ifmaintrin.h
2.6
KB
-rw-r--r--
2025-07-08 23:06
avx512ifmavlintrin.h
4.6
KB
-rw-r--r--
2025-07-08 23:06
avx512vbmi2intrin.h
13.18
KB
-rw-r--r--
2025-07-08 23:06
avx512vbmiintrin.h
3.83
KB
-rw-r--r--
2025-07-08 23:06
avx512vbmivlintrin.h
7.23
KB
-rw-r--r--
2025-07-08 23:06
avx512vlbf16intrin.h
19.46
KB
-rw-r--r--
2025-07-08 23:06
avx512vlbitalgintrin.h
4.52
KB
-rw-r--r--
2025-07-08 23:06
avx512vlbwintrin.h
121.56
KB
-rw-r--r--
2025-07-08 23:06
avx512vlcdintrin.h
7.95
KB
-rw-r--r--
2025-07-08 23:06
avx512vldqintrin.h
46.71
KB
-rw-r--r--
2025-07-08 23:06
avx512vlfp16intrin.h
85.51
KB
-rw-r--r--
2025-07-08 23:06
avx512vlintrin.h
322.6
KB
-rw-r--r--
2025-07-08 23:06
avx512vlvbmi2intrin.h
26.01
KB
-rw-r--r--
2025-07-08 23:06
avx512vlvnniintrin.h
13.41
KB
-rw-r--r--
2025-07-08 23:06
avx512vlvp2intersectintrin.h
4.67
KB
-rw-r--r--
2025-07-08 23:06
avx512vnniintrin.h
4.32
KB
-rw-r--r--
2025-07-08 23:06
avx512vp2intersectintrin.h
3.03
KB
-rw-r--r--
2025-07-08 23:06
avx512vpopcntdqintrin.h
2.3
KB
-rw-r--r--
2025-07-08 23:06
avx512vpopcntdqvlintrin.h
3.86
KB
-rw-r--r--
2025-07-08 23:06
avxifmaintrin.h
5.75
KB
-rw-r--r--
2025-07-08 23:06
avxintrin.h
197.12
KB
-rw-r--r--
2025-07-08 23:06
avxneconvertintrin.h
14.09
KB
-rw-r--r--
2025-07-08 23:06
avxvnniint16intrin.h
15.84
KB
-rw-r--r--
2025-07-08 23:06
avxvnniint8intrin.h
17.02
KB
-rw-r--r--
2025-07-08 23:06
avxvnniintrin.h
10.44
KB
-rw-r--r--
2025-07-08 23:06
bmi2intrin.h
7.37
KB
-rw-r--r--
2025-07-08 23:06
bmiintrin.h
20.02
KB
-rw-r--r--
2025-07-08 23:06
builtins.h
846
B
-rw-r--r--
2025-07-08 23:06
cet.h
1.49
KB
-rw-r--r--
2025-07-08 23:06
cetintrin.h
3.27
KB
-rw-r--r--
2025-07-08 23:06
cldemoteintrin.h
1.18
KB
-rw-r--r--
2025-07-08 23:06
clflushoptintrin.h
1.17
KB
-rw-r--r--
2025-07-08 23:06
clwbintrin.h
1.2
KB
-rw-r--r--
2025-07-08 23:06
clzerointrin.h
1.19
KB
-rw-r--r--
2025-07-08 23:06
cmpccxaddintrin.h
2.33
KB
-rw-r--r--
2025-07-08 23:06
cpuid.h
11.9
KB
-rw-r--r--
2025-07-08 23:06
crc32intrin.h
3.27
KB
-rw-r--r--
2025-07-08 23:06
emmintrin.h
196.89
KB
-rw-r--r--
2025-07-08 23:06
enqcmdintrin.h
2.12
KB
-rw-r--r--
2025-07-08 23:06
f16cintrin.h
5.39
KB
-rw-r--r--
2025-07-08 23:06
float.h
6.43
KB
-rw-r--r--
2025-07-08 23:06
fma4intrin.h
6.82
KB
-rw-r--r--
2025-07-08 23:06
fmaintrin.h
28.65
KB
-rw-r--r--
2025-07-08 23:06
fxsrintrin.h
2.82
KB
-rw-r--r--
2025-07-08 23:06
gfniintrin.h
9.59
KB
-rw-r--r--
2025-07-08 23:06
gpuintrin.h
9.07
KB
-rw-r--r--
2025-07-08 23:06
hexagon_circ_brev_intrinsics.h
15.59
KB
-rw-r--r--
2025-07-08 23:06
hexagon_protos.h
374.42
KB
-rw-r--r--
2025-07-08 23:06
hexagon_types.h
130.38
KB
-rw-r--r--
2025-07-08 23:06
hresetintrin.h
1.36
KB
-rw-r--r--
2025-07-08 23:06
htmintrin.h
6.14
KB
-rw-r--r--
2025-07-08 23:06
htmxlintrin.h
9.01
KB
-rw-r--r--
2025-07-08 23:06
hvx_hexagon_protos.h
275.18
KB
-rw-r--r--
2025-07-08 23:06
ia32intrin.h
25.38
KB
-rw-r--r--
2025-07-08 23:06
immintrin.h
25.46
KB
-rw-r--r--
2025-07-08 23:06
intrin.h
18.16
KB
-rw-r--r--
2025-07-08 23:06
intrin0.h
13.76
KB
-rw-r--r--
2025-07-08 23:06
inttypes.h
2.37
KB
-rw-r--r--
2025-07-08 23:06
invpcidintrin.h
764
B
-rw-r--r--
2025-07-08 23:06
iso646.h
763
B
-rw-r--r--
2025-07-08 23:06
keylockerintrin.h
17.85
KB
-rw-r--r--
2025-07-08 23:06
larchintrin.h
8.5
KB
-rw-r--r--
2025-07-08 23:06
lasxintrin.h
142.14
KB
-rw-r--r--
2025-07-08 23:06
limits.h
3.71
KB
-rw-r--r--
2025-07-08 23:06
lsxintrin.h
134.7
KB
-rw-r--r--
2025-07-08 23:06
lwpintrin.h
5
KB
-rw-r--r--
2025-07-08 23:06
lzcntintrin.h
3.45
KB
-rw-r--r--
2025-07-08 23:06
mm3dnow.h
729
B
-rw-r--r--
2025-07-08 23:06
mm_malloc.h
1.88
KB
-rw-r--r--
2025-07-08 23:06
mmintrin.h
58.6
KB
-rw-r--r--
2025-07-08 23:06
module.modulemap
6.47
KB
-rw-r--r--
2025-07-08 23:06
movdirintrin.h
1.57
KB
-rw-r--r--
2025-07-08 23:06
movrs_avx10_2_512intrin.h
3.88
KB
-rw-r--r--
2025-07-08 23:06
movrs_avx10_2intrin.h
6.87
KB
-rw-r--r--
2025-07-08 23:06
movrsintrin.h
1.98
KB
-rw-r--r--
2025-07-08 23:06
msa.h
25.01
KB
-rw-r--r--
2025-07-08 23:06
mwaitxintrin.h
2.19
KB
-rw-r--r--
2025-07-08 23:06
nmmintrin.h
709
B
-rw-r--r--
2025-07-08 23:06
nvptxintrin.h
7.42
KB
-rw-r--r--
2025-07-08 23:06
omp-tools.h
50.86
KB
-rw-r--r--
2025-09-17 20:37
omp.h
23.36
KB
-rw-r--r--
2025-09-17 20:37
ompt-multiplex.h
51.11
KB
-rw-r--r--
2025-07-08 23:06
ompt.h
50.86
KB
-rw-r--r--
2025-09-17 20:37
ompx.h
7.38
KB
-rw-r--r--
2025-09-17 20:37
opencl-c-base.h
30.94
KB
-rw-r--r--
2025-07-08 23:06
opencl-c.h
874.98
KB
-rw-r--r--
2025-07-08 23:06
pconfigintrin.h
1.19
KB
-rw-r--r--
2025-07-08 23:06
pkuintrin.h
934
B
-rw-r--r--
2025-07-08 23:06
pmmintrin.h
11.12
KB
-rw-r--r--
2025-07-08 23:06
popcntintrin.h
1.9
KB
-rw-r--r--
2025-07-08 23:06
prfchiintrin.h
2.02
KB
-rw-r--r--
2025-07-08 23:06
prfchwintrin.h
2.06
KB
-rw-r--r--
2025-07-08 23:06
ptrauth.h
15.61
KB
-rw-r--r--
2025-07-08 23:06
ptwriteintrin.h
1.05
KB
-rw-r--r--
2025-07-08 23:06
raointintrin.h
6.59
KB
-rw-r--r--
2025-07-08 23:06
rdpruintrin.h
1.59
KB
-rw-r--r--
2025-07-08 23:06
rdseedintrin.h
2.85
KB
-rw-r--r--
2025-07-08 23:06
riscv_bitmanip.h
5.59
KB
-rw-r--r--
2025-07-08 23:06
riscv_corev_alu.h
3.94
KB
-rw-r--r--
2025-07-08 23:06
riscv_crypto.h
5.09
KB
-rw-r--r--
2025-07-08 23:06
riscv_ntlh.h
744
B
-rw-r--r--
2025-07-08 23:06
riscv_vector.h
17.11
KB
-rw-r--r--
2025-09-17 20:11
rtmintrin.h
1.25
KB
-rw-r--r--
2025-07-08 23:06
s390intrin.h
604
B
-rw-r--r--
2025-07-08 23:06
serializeintrin.h
881
B
-rw-r--r--
2025-07-08 23:06
sgxintrin.h
1.77
KB
-rw-r--r--
2025-07-08 23:06
sha512intrin.h
5.95
KB
-rw-r--r--
2025-07-08 23:06
shaintrin.h
7.37
KB
-rw-r--r--
2025-07-08 23:06
sifive_vector.h
6.88
KB
-rw-r--r--
2025-07-08 23:06
sm3intrin.h
7.29
KB
-rw-r--r--
2025-07-08 23:06
sm4evexintrin.h
1.17
KB
-rw-r--r--
2025-07-08 23:06
sm4intrin.h
8.2
KB
-rw-r--r--
2025-07-08 23:06
smmintrin.h
99.34
KB
-rw-r--r--
2025-07-08 23:06
stdalign.h
756
B
-rw-r--r--
2025-07-08 23:06
stdarg.h
2.44
KB
-rw-r--r--
2025-07-08 23:06
stdatomic.h
8.46
KB
-rw-r--r--
2025-07-08 23:06
stdbool.h
1.14
KB
-rw-r--r--
2025-07-08 23:06
stdckdint.h
1.63
KB
-rw-r--r--
2025-07-08 23:06
stddef.h
4.93
KB
-rw-r--r--
2025-07-08 23:06
stdint.h
30.33
KB
-rw-r--r--
2025-07-08 23:06
stdnoreturn.h
1.29
KB
-rw-r--r--
2025-07-08 23:06
tbmintrin.h
3.42
KB
-rw-r--r--
2025-07-08 23:06
tgmath.h
29.68
KB
-rw-r--r--
2025-07-08 23:06
tmmintrin.h
31.18
KB
-rw-r--r--
2025-07-08 23:06
tsxldtrkintrin.h
1.97
KB
-rw-r--r--
2025-07-08 23:06
uintrintrin.h
4.96
KB
-rw-r--r--
2025-07-08 23:06
unwind.h
11.21
KB
-rw-r--r--
2025-07-08 23:06
usermsrintrin.h
1.54
KB
-rw-r--r--
2025-07-08 23:06
vadefs.h
1.39
KB
-rw-r--r--
2025-07-08 23:06
vaesintrin.h
2.61
KB
-rw-r--r--
2025-07-08 23:06
varargs.h
584
B
-rw-r--r--
2025-07-08 23:06
vecintrin.h
419.23
KB
-rw-r--r--
2025-07-08 23:06
velintrin.h
2.1
KB
-rw-r--r--
2025-07-08 23:06
velintrin_approx.h
3.54
KB
-rw-r--r--
2025-07-08 23:06
velintrin_gen.h
69.06
KB
-rw-r--r--
2025-07-08 23:06
vpclmulqdqintrin.h
1.06
KB
-rw-r--r--
2025-07-08 23:06
waitpkgintrin.h
1.33
KB
-rw-r--r--
2025-07-08 23:06
wasm_simd128.h
81.71
KB
-rw-r--r--
2025-07-08 23:06
wbnoinvdintrin.h
749
B
-rw-r--r--
2025-07-08 23:06
wmmintrin.h
659
B
-rw-r--r--
2025-07-08 23:06
x86gprintrin.h
2.22
KB
-rw-r--r--
2025-07-08 23:06
x86intrin.h
1.38
KB
-rw-r--r--
2025-07-08 23:06
xmmintrin.h
117.43
KB
-rw-r--r--
2025-07-08 23:06
xopintrin.h
19.96
KB
-rw-r--r--
2025-07-08 23:06
xsavecintrin.h
2.51
KB
-rw-r--r--
2025-07-08 23:06
xsaveintrin.h
1.64
KB
-rw-r--r--
2025-07-08 23:06
xsaveoptintrin.h
1
KB
-rw-r--r--
2025-07-08 23:06
xsavesintrin.h
1.24
KB
-rw-r--r--
2025-07-08 23:06
xtestintrin.h
873
B
-rw-r--r--
2025-07-08 23:06
yvals_core.h
687
B
-rw-r--r--
2025-07-08 23:06
Save
Rename
/*===--------------- amxintrin.h - AMX intrinsics -*- C/C++ -*---------------=== * * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. * See https://llvm.org/LICENSE.txt for license information. * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception * *===------------------------------------------------------------------------=== */ #ifndef __IMMINTRIN_H #error "Never use <amxintrin.h> directly; include <immintrin.h> instead." #endif /* __IMMINTRIN_H */ #ifndef __AMXINTRIN_H #define __AMXINTRIN_H #ifdef __x86_64__ /* Define the default attributes for the functions in this file. */ #define __DEFAULT_FN_ATTRS_TILE \ __attribute__((__always_inline__, __nodebug__, __target__("amx-tile"))) #define __DEFAULT_FN_ATTRS_INT8 \ __attribute__((__always_inline__, __nodebug__, __target__("amx-int8"))) #define __DEFAULT_FN_ATTRS_BF16 \ __attribute__((__always_inline__, __nodebug__, __target__("amx-bf16"))) /// Load tile configuration from a 64-byte memory location specified by /// "mem_addr". The tile configuration includes the tile type palette, the /// number of bytes per row, and the number of rows. If the specified /// palette_id is zero, that signifies the init state for both the tile /// config and the tile data, and the tiles are zeroed. Any invalid /// configurations will result in #GP fault. /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> LDTILECFG </c> instruction. /// /// \param __config /// A pointer to 512-bits configuration static __inline__ void __DEFAULT_FN_ATTRS_TILE _tile_loadconfig(const void *__config) { __builtin_ia32_tile_loadconfig(__config); } /// Stores the current tile configuration to a 64-byte memory location /// specified by "mem_addr". The tile configuration includes the tile type /// palette, the number of bytes per row, and the number of rows. If tiles /// are not configured, all zeroes will be stored to memory. /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> STTILECFG </c> instruction. /// /// \param __config /// A pointer to 512-bits configuration static __inline__ void __DEFAULT_FN_ATTRS_TILE _tile_storeconfig(void *__config) { __builtin_ia32_tile_storeconfig(__config); } /// Release the tile configuration to return to the init state, which /// releases all storage it currently holds. /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TILERELEASE </c> instruction. static __inline__ void __DEFAULT_FN_ATTRS_TILE _tile_release(void) { __builtin_ia32_tilerelease(); } /// Load tile rows from memory specifieid by "base" address and "stride" into /// destination tile "dst" using the tile configuration previously configured /// via "_tile_loadconfig". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TILELOADD </c> instruction. /// /// \param dst /// A destination tile. Max size is 1024 Bytes. /// \param base /// A pointer to base address. /// \param stride /// The stride between the rows' data to be loaded in memory. #define _tile_loadd(dst, base, stride) \ __builtin_ia32_tileloadd64((dst), ((const void *)(base)), \ (__SIZE_TYPE__)(stride)) /// Load tile rows from memory specifieid by "base" address and "stride" into /// destination tile "dst" using the tile configuration previously configured /// via "_tile_loadconfig". This intrinsic provides a hint to the implementation /// that the data will likely not be reused in the near future and the data /// caching can be optimized accordingly. /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TILELOADDT1 </c> instruction. /// /// \param dst /// A destination tile. Max size is 1024 Bytes. /// \param base /// A pointer to base address. /// \param stride /// The stride between the rows' data to be loaded in memory. #define _tile_stream_loadd(dst, base, stride) \ __builtin_ia32_tileloaddt164((dst), ((const void *)(base)), \ (__SIZE_TYPE__)(stride)) /// Store the tile specified by "src" to memory specifieid by "base" address and /// "stride" using the tile configuration previously configured via /// "_tile_loadconfig". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TILESTORED </c> instruction. /// /// \param dst /// A destination tile. Max size is 1024 Bytes. /// \param base /// A pointer to base address. /// \param stride /// The stride between the rows' data to be stored in memory. #define _tile_stored(dst, base, stride) \ __builtin_ia32_tilestored64((dst), ((void *)(base)), (__SIZE_TYPE__)(stride)) /// Zero the tile specified by "tdest". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TILEZERO </c> instruction. /// /// \param tile /// The destination tile to be zero. Max size is 1024 Bytes. #define _tile_zero(tile) __builtin_ia32_tilezero((tile)) /// Compute dot-product of bytes in tiles with a source/destination accumulator. /// Multiply groups of 4 adjacent pairs of signed 8-bit integers in src0 with /// corresponding signed 8-bit integers in src1, producing 4 intermediate 32-bit /// results. Sum these 4 results with the corresponding 32-bit integer in "dst", /// and store the 32-bit result back to tile "dst". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TDPBSSD </c> instruction. /// /// \param dst /// The destination tile. Max size is 1024 Bytes. /// \param src0 /// The 1st source tile. Max size is 1024 Bytes. /// \param src1 /// The 2nd source tile. Max size is 1024 Bytes. #define _tile_dpbssd(dst, src0, src1) \ __builtin_ia32_tdpbssd((dst), (src0), (src1)) /// Compute dot-product of bytes in tiles with a source/destination accumulator. /// Multiply groups of 4 adjacent pairs of signed 8-bit integers in src0 with /// corresponding unsigned 8-bit integers in src1, producing 4 intermediate /// 32-bit results. Sum these 4 results with the corresponding 32-bit integer /// in "dst", and store the 32-bit result back to tile "dst". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TDPBSUD </c> instruction. /// /// \param dst /// The destination tile. Max size is 1024 Bytes. /// \param src0 /// The 1st source tile. Max size is 1024 Bytes. /// \param src1 /// The 2nd source tile. Max size is 1024 Bytes. #define _tile_dpbsud(dst, src0, src1) \ __builtin_ia32_tdpbsud((dst), (src0), (src1)) /// Compute dot-product of bytes in tiles with a source/destination accumulator. /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in src0 with /// corresponding signed 8-bit integers in src1, producing 4 intermediate 32-bit /// results. Sum these 4 results with the corresponding 32-bit integer in "dst", /// and store the 32-bit result back to tile "dst". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TDPBUSD </c> instruction. /// /// \param dst /// The destination tile. Max size is 1024 Bytes. /// \param src0 /// The 1st source tile. Max size is 1024 Bytes. /// \param src1 /// The 2nd source tile. Max size is 1024 Bytes. #define _tile_dpbusd(dst, src0, src1) \ __builtin_ia32_tdpbusd((dst), (src0), (src1)) /// Compute dot-product of bytes in tiles with a source/destination accumulator. /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in src0 with /// corresponding unsigned 8-bit integers in src1, producing 4 intermediate /// 32-bit results. Sum these 4 results with the corresponding 32-bit integer in /// "dst", and store the 32-bit result back to tile "dst". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TDPBUUD </c> instruction. /// /// \param dst /// The destination tile. Max size is 1024 Bytes. /// \param src0 /// The 1st source tile. Max size is 1024 Bytes. /// \param src1 /// The 2nd source tile. Max size is 1024 Bytes. #define _tile_dpbuud(dst, src0, src1) \ __builtin_ia32_tdpbuud((dst), (src0), (src1)) /// Compute dot-product of BF16 (16-bit) floating-point pairs in tiles src0 and /// src1, accumulating the intermediate single-precision (32-bit) floating-point /// elements with elements in "dst", and store the 32-bit result back to tile /// "dst". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TDPBF16PS </c> instruction. /// /// \param dst /// The destination tile. Max size is 1024 Bytes. /// \param src0 /// The 1st source tile. Max size is 1024 Bytes. /// \param src1 /// The 2nd source tile. Max size is 1024 Bytes. #define _tile_dpbf16ps(dst, src0, src1) \ __builtin_ia32_tdpbf16ps((dst), (src0), (src1)) /// AMX tile register size can be configured, the maximum size is 16x64=1024 /// bytes. Since there is no 2D type in llvm IR, we use vector type to /// represent 2D tile and the fixed size is maximum amx tile register size. typedef int _tile1024i __attribute__((__vector_size__(1024), __aligned__(64))); typedef int _tile1024i_1024a __attribute__((__vector_size__(1024), __aligned__(1024))); /// This is internal intrinsic. C/C++ user should avoid calling it directly. static __inline__ _tile1024i __DEFAULT_FN_ATTRS_TILE _tile_loadd_internal(unsigned short m, unsigned short n, const void *base, __SIZE_TYPE__ stride) { return __builtin_ia32_tileloadd64_internal(m, n, base, (__SIZE_TYPE__)(stride)); } /// This is internal intrinsic. C/C++ user should avoid calling it directly. static __inline__ _tile1024i __DEFAULT_FN_ATTRS_TILE _tile_loaddt1_internal(unsigned short m, unsigned short n, const void *base, __SIZE_TYPE__ stride) { return __builtin_ia32_tileloaddt164_internal(m, n, base, (__SIZE_TYPE__)(stride)); } /// This is internal intrinsic. C/C++ user should avoid calling it directly. static __inline__ _tile1024i __DEFAULT_FN_ATTRS_INT8 _tile_dpbssd_internal(unsigned short m, unsigned short n, unsigned short k, _tile1024i dst, _tile1024i src1, _tile1024i src2) { return __builtin_ia32_tdpbssd_internal(m, n, k, dst, src1, src2); } /// This is internal intrinsic. C/C++ user should avoid calling it directly. static __inline__ _tile1024i __DEFAULT_FN_ATTRS_INT8 _tile_dpbsud_internal(unsigned short m, unsigned short n, unsigned short k, _tile1024i dst, _tile1024i src1, _tile1024i src2) { return __builtin_ia32_tdpbsud_internal(m, n, k, dst, src1, src2); } /// This is internal intrinsic. C/C++ user should avoid calling it directly. static __inline__ _tile1024i __DEFAULT_FN_ATTRS_INT8 _tile_dpbusd_internal(unsigned short m, unsigned short n, unsigned short k, _tile1024i dst, _tile1024i src1, _tile1024i src2) { return __builtin_ia32_tdpbusd_internal(m, n, k, dst, src1, src2); } /// This is internal intrinsic. C/C++ user should avoid calling it directly. static __inline__ _tile1024i __DEFAULT_FN_ATTRS_INT8 _tile_dpbuud_internal(unsigned short m, unsigned short n, unsigned short k, _tile1024i dst, _tile1024i src1, _tile1024i src2) { return __builtin_ia32_tdpbuud_internal(m, n, k, dst, src1, src2); } /// This is internal intrinsic. C/C++ user should avoid calling it directly. static __inline__ void __DEFAULT_FN_ATTRS_TILE _tile_stored_internal(unsigned short m, unsigned short n, void *base, __SIZE_TYPE__ stride, _tile1024i tile) { return __builtin_ia32_tilestored64_internal(m, n, base, (__SIZE_TYPE__)(stride), tile); } /// This is internal intrinsic. C/C++ user should avoid calling it directly. static __inline__ _tile1024i __DEFAULT_FN_ATTRS_BF16 _tile_dpbf16ps_internal(unsigned short m, unsigned short n, unsigned short k, _tile1024i dst, _tile1024i src1, _tile1024i src2) { return __builtin_ia32_tdpbf16ps_internal(m, n, k, dst, src1, src2); } /// This struct pack the shape and tile data together for user. We suggest /// initializing the struct as early as possible, because compiler depends /// on the shape information to do configure. The constant value is preferred /// for optimization by compiler. typedef struct __tile1024i_str { const unsigned short row; const unsigned short col; _tile1024i tile; } __tile1024i; /// Load tile rows from memory specifieid by "base" address and "stride" into /// destination tile "dst". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TILELOADD </c> instruction. /// /// \param dst /// A destination tile. Max size is 1024 Bytes. /// \param base /// A pointer to base address. /// \param stride /// The stride between the rows' data to be loaded in memory. __DEFAULT_FN_ATTRS_TILE static __inline__ void __tile_loadd(__tile1024i *dst, const void *base, __SIZE_TYPE__ stride) { dst->tile = _tile_loadd_internal(dst->row, dst->col, base, stride); } /// Load tile rows from memory specifieid by "base" address and "stride" into /// destination tile "dst". This intrinsic provides a hint to the implementation /// that the data will likely not be reused in the near future and the data /// caching can be optimized accordingly. /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TILELOADDT1 </c> instruction. /// /// \param dst /// A destination tile. Max size is 1024 Bytes. /// \param base /// A pointer to base address. /// \param stride /// The stride between the rows' data to be loaded in memory. __DEFAULT_FN_ATTRS_TILE static __inline__ void __tile_stream_loadd(__tile1024i *dst, const void *base, __SIZE_TYPE__ stride) { dst->tile = _tile_loaddt1_internal(dst->row, dst->col, base, stride); } /// Compute dot-product of bytes in tiles with a source/destination accumulator. /// Multiply groups of 4 adjacent pairs of signed 8-bit integers in src0 with /// corresponding signed 8-bit integers in src1, producing 4 intermediate 32-bit /// results. Sum these 4 results with the corresponding 32-bit integer in "dst", /// and store the 32-bit result back to tile "dst". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TDPBSSD </c> instruction. /// /// \param dst /// The destination tile. Max size is 1024 Bytes. /// \param src0 /// The 1st source tile. Max size is 1024 Bytes. /// \param src1 /// The 2nd source tile. Max size is 1024 Bytes. __DEFAULT_FN_ATTRS_INT8 static __inline__ void __tile_dpbssd(__tile1024i *dst, __tile1024i src0, __tile1024i src1) { dst->tile = _tile_dpbssd_internal(src0.row, src1.col, src0.col, dst->tile, src0.tile, src1.tile); } /// Compute dot-product of bytes in tiles with a source/destination accumulator. /// Multiply groups of 4 adjacent pairs of signed 8-bit integers in src0 with /// corresponding unsigned 8-bit integers in src1, producing 4 intermediate /// 32-bit results. Sum these 4 results with the corresponding 32-bit integer /// in "dst", and store the 32-bit result back to tile "dst". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TDPBSUD </c> instruction. /// /// \param dst /// The destination tile. Max size is 1024 Bytes. /// \param src0 /// The 1st source tile. Max size is 1024 Bytes. /// \param src1 /// The 2nd source tile. Max size is 1024 Bytes. __DEFAULT_FN_ATTRS_INT8 static __inline__ void __tile_dpbsud(__tile1024i *dst, __tile1024i src0, __tile1024i src1) { dst->tile = _tile_dpbsud_internal(src0.row, src1.col, src0.col, dst->tile, src0.tile, src1.tile); } /// Compute dot-product of bytes in tiles with a source/destination accumulator. /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in src0 with /// corresponding signed 8-bit integers in src1, producing 4 intermediate 32-bit /// results. Sum these 4 results with the corresponding 32-bit integer in "dst", /// and store the 32-bit result back to tile "dst". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TDPBUSD </c> instruction. /// /// \param dst /// The destination tile. Max size is 1024 Bytes. /// \param src0 /// The 1st source tile. Max size is 1024 Bytes. /// \param src1 /// The 2nd source tile. Max size is 1024 Bytes. __DEFAULT_FN_ATTRS_INT8 static __inline__ void __tile_dpbusd(__tile1024i *dst, __tile1024i src0, __tile1024i src1) { dst->tile = _tile_dpbusd_internal(src0.row, src1.col, src0.col, dst->tile, src0.tile, src1.tile); } /// Compute dot-product of bytes in tiles with a source/destination accumulator. /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in src0 with /// corresponding unsigned 8-bit integers in src1, producing 4 intermediate /// 32-bit results. Sum these 4 results with the corresponding 32-bit integer in /// "dst", and store the 32-bit result back to tile "dst". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TDPBUUD </c> instruction. /// /// \param dst /// The destination tile. Max size is 1024 Bytes. /// \param src0 /// The 1st source tile. Max size is 1024 Bytes. /// \param src1 /// The 2nd source tile. Max size is 1024 Bytes. __DEFAULT_FN_ATTRS_INT8 static __inline__ void __tile_dpbuud(__tile1024i *dst, __tile1024i src0, __tile1024i src1) { dst->tile = _tile_dpbuud_internal(src0.row, src1.col, src0.col, dst->tile, src0.tile, src1.tile); } /// Store the tile specified by "src" to memory specifieid by "base" address and /// "stride". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TILESTORED </c> instruction. /// /// \param base /// A pointer to base address. /// \param stride /// The stride between the rows' data to be stored in memory. __DEFAULT_FN_ATTRS_TILE static __inline__ void __tile_stored(void *base, __SIZE_TYPE__ stride, __tile1024i src) { _tile_stored_internal(src.row, src.col, base, stride, src.tile); } /// Zero the tile specified by "dst". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TILEZERO </c> instruction. /// /// \param dst /// The destination tile to be zero. Max size is 1024 Bytes. __DEFAULT_FN_ATTRS_TILE static __inline__ void __tile_zero(__tile1024i *dst) { dst->tile = __builtin_ia32_tilezero_internal(dst->row, dst->col); } /// Compute dot-product of BF16 (16-bit) floating-point pairs in tiles src0 and /// src1, accumulating the intermediate single-precision (32-bit) floating-point /// elements with elements in "dst", and store the 32-bit result back to tile /// "dst". /// /// \headerfile <immintrin.h> /// /// This intrinsic corresponds to the <c> TDPBF16PS </c> instruction. /// /// \param dst /// The destination tile. Max size is 1024 Bytes. /// \param src0 /// The 1st source tile. Max size is 1024 Bytes. /// \param src1 /// The 2nd source tile. Max size is 1024 Bytes. __DEFAULT_FN_ATTRS_BF16 static __inline__ void __tile_dpbf16ps(__tile1024i *dst, __tile1024i src0, __tile1024i src1) { dst->tile = _tile_dpbf16ps_internal(src0.row, src1.col, src0.col, dst->tile, src0.tile, src1.tile); } #undef __DEFAULT_FN_ATTRS_TILE #undef __DEFAULT_FN_ATTRS_INT8 #undef __DEFAULT_FN_ATTRS_BF16 #endif /* __x86_64__ */ #endif /* __AMXINTRIN_H */