@cpp-linter/clang-tidy-darwin-x64 13.0.0 → 16.0.3
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +2 -2
- package/bin/clang-tidy +2 -2
- package/lib/clang/{13.0.0 → 16}/include/__clang_cuda_complex_builtins.h +3 -3
- package/lib/clang/{13.0.0 → 16}/include/__clang_cuda_intrinsics.h +38 -6
- package/lib/clang/{13.0.0 → 16}/include/__clang_cuda_libdevice_declares.h +6 -0
- package/lib/clang/{13.0.0 → 16}/include/__clang_cuda_math.h +1 -1
- package/lib/clang/{13.0.0 → 16}/include/__clang_cuda_runtime_wrapper.h +62 -7
- package/lib/clang/16/include/__clang_cuda_texture_intrinsics.h +742 -0
- package/lib/clang/{13.0.0 → 16}/include/__clang_hip_libdevice_declares.h +5 -0
- package/lib/clang/{13.0.0 → 16}/include/__clang_hip_math.h +10 -13
- package/lib/clang/{13.0.0 → 16}/include/__clang_hip_runtime_wrapper.h +18 -4
- package/lib/clang/16/include/__clang_hip_stdlib.h +43 -0
- package/lib/clang/{13.0.0 → 16}/include/__wmmintrin_aes.h +1 -1
- package/lib/clang/{13.0.0 → 16}/include/__wmmintrin_pclmul.h +10 -10
- package/lib/clang/{13.0.0 → 16}/include/altivec.h +691 -342
- package/lib/clang/{13.0.0 → 16}/include/ammintrin.h +4 -0
- package/lib/clang/16/include/amxfp16intrin.h +58 -0
- package/lib/clang/{13.0.0 → 16}/include/amxintrin.h +49 -18
- package/lib/clang/{13.0.0 → 16}/include/arm_acle.h +80 -73
- package/lib/clang/{13.0.0 → 16}/include/arm_fp16.h +90 -90
- package/lib/clang/{13.0.0 → 16}/include/arm_neon.h +28607 -28779
- package/lib/clang/16/include/arm_neon_sve_bridge.h +182 -0
- package/lib/clang/{13.0.0 → 16}/include/arm_sve.h +7848 -7996
- package/lib/clang/{13.0.0 → 16}/include/avx2intrin.h +189 -189
- package/lib/clang/{13.0.0 → 16}/include/avx512bf16intrin.h +21 -18
- package/lib/clang/{13.0.0 → 16}/include/avx512bwintrin.h +83 -83
- package/lib/clang/{13.0.0 → 16}/include/avx512dqintrin.h +365 -365
- package/lib/clang/{13.0.0 → 16}/include/avx512erintrin.h +102 -102
- package/lib/clang/{13.0.0 → 16}/include/avx512fintrin.h +1627 -1620
- package/lib/clang/16/include/avx512fp16intrin.h +3346 -0
- package/lib/clang/{13.0.0 → 16}/include/avx512ifmavlintrin.h +13 -27
- package/lib/clang/{13.0.0 → 16}/include/avx512vbmi2intrin.h +48 -48
- package/lib/clang/{13.0.0 → 16}/include/avx512vlbf16intrin.h +82 -41
- package/lib/clang/{13.0.0 → 16}/include/avx512vlbwintrin.h +454 -102
- package/lib/clang/{13.0.0 → 16}/include/avx512vldqintrin.h +134 -134
- package/lib/clang/16/include/avx512vlfp16intrin.h +2071 -0
- package/lib/clang/{13.0.0 → 16}/include/avx512vlintrin.h +596 -610
- package/lib/clang/{13.0.0 → 16}/include/avx512vlvbmi2intrin.h +96 -96
- package/lib/clang/{13.0.0 → 16}/include/avx512vlvnniintrin.h +24 -24
- package/lib/clang/16/include/avxifmaintrin.h +177 -0
- package/lib/clang/{13.0.0 → 16}/include/avxintrin.h +266 -224
- package/lib/clang/16/include/avxneconvertintrin.h +484 -0
- package/lib/clang/16/include/avxvnniint8intrin.h +471 -0
- package/lib/clang/{13.0.0 → 16}/include/avxvnniintrin.h +16 -16
- package/lib/clang/{13.0.0 → 16}/include/bmiintrin.h +6 -2
- package/lib/clang/{13.0.0 → 16}/include/cetintrin.h +20 -4
- package/lib/clang/16/include/cmpccxaddintrin.h +70 -0
- package/lib/clang/{13.0.0 → 16}/include/cpuid.h +14 -2
- package/lib/clang/16/include/crc32intrin.h +100 -0
- package/lib/clang/16/include/cuda_wrappers/cmath +90 -0
- package/lib/clang/{13.0.0 → 16}/include/emmintrin.h +573 -779
- package/lib/clang/{13.0.0 → 16}/include/f16cintrin.h +4 -4
- package/lib/clang/{13.0.0 → 16}/include/float.h +22 -6
- package/lib/clang/{13.0.0 → 16}/include/gfniintrin.h +54 -53
- package/lib/clang/{13.0.0 → 16}/include/hexagon_protos.h +0 -11
- package/lib/clang/{13.0.0 → 16}/include/hexagon_types.h +0 -32
- package/lib/clang/16/include/hlsl/hlsl_basic_types.h +67 -0
- package/lib/clang/16/include/hlsl/hlsl_intrinsics.h +223 -0
- package/lib/clang/16/include/hlsl.h +15 -0
- package/lib/clang/{13.0.0 → 16}/include/hresetintrin.h +2 -2
- package/lib/clang/{13.0.0 → 16}/include/hvx_hexagon_protos.h +1200 -409
- package/lib/clang/{13.0.0 → 16}/include/ia32intrin.h +17 -17
- package/lib/clang/{13.0.0 → 16}/include/immintrin.h +74 -24
- package/lib/clang/{13.0.0 → 16}/include/intrin.h +34 -35
- package/lib/clang/{13.0.0 → 16}/include/keylockerintrin.h +24 -30
- package/lib/clang/16/include/larchintrin.h +234 -0
- package/lib/clang/{13.0.0 → 16}/include/limits.h +22 -1
- package/lib/clang/{13.0.0 → 16}/include/mm_malloc.h +3 -3
- package/lib/clang/{13.0.0 → 16}/include/mmintrin.h +4 -0
- package/lib/clang/{13.0.0 → 16}/include/nmmintrin.h +4 -0
- package/lib/clang/{13.0.0 → 16}/include/opencl-c-base.h +98 -11
- package/lib/clang/{13.0.0 → 16}/include/opencl-c.h +7252 -6208
- package/lib/clang/{13.0.0 → 16}/include/openmp_wrappers/complex +10 -1
- package/lib/clang/{13.0.0 → 16}/include/openmp_wrappers/complex.h +9 -0
- package/lib/clang/{13.0.0 → 16}/include/pmmintrin.h +5 -1
- package/lib/clang/16/include/ppc_wrappers/bmi2intrin.h +134 -0
- package/lib/clang/16/include/ppc_wrappers/bmiintrin.h +165 -0
- package/lib/clang/16/include/ppc_wrappers/emmintrin.h +2268 -0
- package/lib/clang/16/include/ppc_wrappers/immintrin.h +27 -0
- package/lib/clang/{13.0.0 → 16}/include/ppc_wrappers/mm_malloc.h +12 -17
- package/lib/clang/{13.0.0 → 16}/include/ppc_wrappers/mmintrin.h +389 -386
- package/lib/clang/16/include/ppc_wrappers/pmmintrin.h +145 -0
- package/lib/clang/16/include/ppc_wrappers/smmintrin.h +663 -0
- package/lib/clang/16/include/ppc_wrappers/tmmintrin.h +453 -0
- package/lib/clang/{13.0.0/include/stdnoreturn.h → 16/include/ppc_wrappers/x86gprintrin.h} +7 -6
- package/lib/clang/16/include/ppc_wrappers/x86intrin.h +28 -0
- package/lib/clang/16/include/ppc_wrappers/xmmintrin.h +1827 -0
- package/lib/clang/16/include/prfchiintrin.h +61 -0
- package/lib/clang/{13.0.0 → 16}/include/prfchwintrin.h +5 -2
- package/lib/clang/16/include/raointintrin.h +203 -0
- package/lib/clang/16/include/rdpruintrin.h +57 -0
- package/lib/clang/{13.0.0 → 16}/include/rdseedintrin.h +3 -3
- package/lib/clang/16/include/riscv_vector.h +211 -0
- package/lib/clang/{13.0.0 → 16}/include/rtmintrin.h +1 -1
- package/lib/clang/{13.0.0 → 16}/include/smmintrin.h +260 -362
- package/lib/clang/{13.0.0 → 16}/include/stdarg.h +23 -7
- package/lib/clang/{13.0.0 → 16}/include/stdatomic.h +14 -1
- package/lib/clang/{13.0.0 → 16}/include/stdbool.h +10 -7
- package/lib/clang/{13.0.0 → 16}/include/stddef.h +9 -2
- package/lib/clang/{13.0.0 → 16}/include/stdint.h +311 -25
- package/lib/clang/16/include/stdnoreturn.h +29 -0
- package/lib/clang/{13.0.0 → 16}/include/tmmintrin.h +10 -6
- package/lib/clang/{13.0.0 → 16}/include/uintrintrin.h +8 -8
- package/lib/clang/{13.0.0 → 16}/include/unwind.h +12 -7
- package/lib/clang/{13.0.0 → 16}/include/vaesintrin.h +1 -1
- package/lib/clang/16/include/velintrin.h +71 -0
- package/lib/clang/16/include/velintrin_approx.h +120 -0
- package/lib/clang/16/include/velintrin_gen.h +1257 -0
- package/lib/clang/{13.0.0 → 16}/include/vpclmulqdqintrin.h +6 -6
- package/lib/clang/{13.0.0 → 16}/include/wasm_simd128.h +169 -26
- package/lib/clang/{13.0.0 → 16}/include/wmmintrin.h +4 -0
- package/lib/clang/16/include/x86gprintrin.h +65 -0
- package/lib/clang/{13.0.0 → 16}/include/x86intrin.h +4 -0
- package/lib/clang/{13.0.0 → 16}/include/xmmintrin.h +19 -10
- package/lib/clang/{13.0.0 → 16}/include/xopintrin.h +31 -31
- package/package.json +6 -6
- package/lib/clang/13.0.0/include/ppc_wrappers/emmintrin.h +0 -2324
- package/lib/clang/13.0.0/include/ppc_wrappers/pmmintrin.h +0 -150
- package/lib/clang/13.0.0/include/ppc_wrappers/smmintrin.h +0 -109
- package/lib/clang/13.0.0/include/ppc_wrappers/tmmintrin.h +0 -495
- package/lib/clang/13.0.0/include/ppc_wrappers/xmmintrin.h +0 -1843
- package/lib/clang/13.0.0/include/riscv_vector.h +0 -123865
- package/lib/clang/13.0.0/include/x86gprintrin.h +0 -23
- /package/lib/clang/{13.0.0 → 16}/include/__clang_cuda_builtin_vars.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/__clang_cuda_cmath.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/__clang_cuda_device_functions.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/__clang_cuda_math_forward_declares.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/__clang_hip_cmath.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/__stddef_max_align_t.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/adxintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/arm64intr.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/arm_bf16.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/arm_cde.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/arm_cmse.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/arm_mve.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/armintr.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/avx512bitalgintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/avx512cdintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/avx512ifmaintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/avx512pfintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/avx512vbmiintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/avx512vbmivlintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/avx512vlbitalgintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/avx512vlcdintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/avx512vlvp2intersectintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/avx512vnniintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/avx512vp2intersectintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/avx512vpopcntdqintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/avx512vpopcntdqvlintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/bmi2intrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/builtins.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/cet.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/cldemoteintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/clflushoptintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/clwbintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/clzerointrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/cuda_wrappers/algorithm +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/cuda_wrappers/complex +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/cuda_wrappers/new +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/enqcmdintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/fma4intrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/fmaintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/fxsrintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/hexagon_circ_brev_intrinsics.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/htmintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/htmxlintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/inttypes.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/invpcidintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/iso646.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/lwpintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/lzcntintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/mm3dnow.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/module.modulemap +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/movdirintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/msa.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/mwaitxintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/openmp_wrappers/__clang_openmp_device_functions.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/openmp_wrappers/cmath +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/openmp_wrappers/complex_cmath.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/openmp_wrappers/math.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/openmp_wrappers/new +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/pconfigintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/pkuintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/popcntintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/ptwriteintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/s390intrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/serializeintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/sgxintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/shaintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/stdalign.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/tbmintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/tgmath.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/tsxldtrkintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/vadefs.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/varargs.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/vecintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/waitpkgintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/wbnoinvdintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/xsavecintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/xsaveintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/xsaveoptintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/xsavesintrin.h +0 -0
- /package/lib/clang/{13.0.0 → 16}/include/xtestintrin.h +0 -0
|
@@ -19,7 +19,7 @@
|
|
|
19
19
|
__attribute__((__always_inline__, __nodebug__, __target__("shstk")))
|
|
20
20
|
|
|
21
21
|
static __inline__ void __DEFAULT_FN_ATTRS _incsspd(int __a) {
|
|
22
|
-
__builtin_ia32_incsspd(__a);
|
|
22
|
+
__builtin_ia32_incsspd((unsigned int)__a);
|
|
23
23
|
}
|
|
24
24
|
|
|
25
25
|
#ifdef __x86_64__
|
|
@@ -34,7 +34,7 @@ static __inline__ void __DEFAULT_FN_ATTRS _inc_ssp(unsigned int __a) {
|
|
|
34
34
|
}
|
|
35
35
|
#else /* __x86_64__ */
|
|
36
36
|
static __inline__ void __DEFAULT_FN_ATTRS _inc_ssp(unsigned int __a) {
|
|
37
|
-
__builtin_ia32_incsspd(
|
|
37
|
+
__builtin_ia32_incsspd(__a);
|
|
38
38
|
}
|
|
39
39
|
#endif /* __x86_64__ */
|
|
40
40
|
|
|
@@ -42,10 +42,26 @@ static __inline__ unsigned int __DEFAULT_FN_ATTRS _rdsspd(unsigned int __a) {
|
|
|
42
42
|
return __builtin_ia32_rdsspd(__a);
|
|
43
43
|
}
|
|
44
44
|
|
|
45
|
+
static __inline__ unsigned int __DEFAULT_FN_ATTRS _rdsspd_i32(void) {
|
|
46
|
+
#pragma clang diagnostic push
|
|
47
|
+
#pragma clang diagnostic ignored "-Wuninitialized"
|
|
48
|
+
unsigned int t;
|
|
49
|
+
return __builtin_ia32_rdsspd(t);
|
|
50
|
+
#pragma clang diagnostic pop
|
|
51
|
+
}
|
|
52
|
+
|
|
45
53
|
#ifdef __x86_64__
|
|
46
54
|
static __inline__ unsigned long long __DEFAULT_FN_ATTRS _rdsspq(unsigned long long __a) {
|
|
47
55
|
return __builtin_ia32_rdsspq(__a);
|
|
48
56
|
}
|
|
57
|
+
|
|
58
|
+
static __inline__ unsigned long long __DEFAULT_FN_ATTRS _rdsspq_i64(void) {
|
|
59
|
+
#pragma clang diagnostic push
|
|
60
|
+
#pragma clang diagnostic ignored "-Wuninitialized"
|
|
61
|
+
unsigned long long t;
|
|
62
|
+
return __builtin_ia32_rdsspq(t);
|
|
63
|
+
#pragma clang diagnostic pop
|
|
64
|
+
}
|
|
49
65
|
#endif /* __x86_64__ */
|
|
50
66
|
|
|
51
67
|
#ifdef __x86_64__
|
|
@@ -58,7 +74,7 @@ static __inline__ unsigned int __DEFAULT_FN_ATTRS _get_ssp(void) {
|
|
|
58
74
|
}
|
|
59
75
|
#endif /* __x86_64__ */
|
|
60
76
|
|
|
61
|
-
static __inline__ void __DEFAULT_FN_ATTRS _saveprevssp() {
|
|
77
|
+
static __inline__ void __DEFAULT_FN_ATTRS _saveprevssp(void) {
|
|
62
78
|
__builtin_ia32_saveprevssp();
|
|
63
79
|
}
|
|
64
80
|
|
|
@@ -86,7 +102,7 @@ static __inline__ void __DEFAULT_FN_ATTRS _wrussq(unsigned long long __a, void *
|
|
|
86
102
|
}
|
|
87
103
|
#endif /* __x86_64__ */
|
|
88
104
|
|
|
89
|
-
static __inline__ void __DEFAULT_FN_ATTRS _setssbsy() {
|
|
105
|
+
static __inline__ void __DEFAULT_FN_ATTRS _setssbsy(void) {
|
|
90
106
|
__builtin_ia32_setssbsy();
|
|
91
107
|
}
|
|
92
108
|
|
|
@@ -0,0 +1,70 @@
|
|
|
1
|
+
/*===--------------- cmpccxaddintrin.h - CMPCCXADD intrinsics--------------===
|
|
2
|
+
*
|
|
3
|
+
*
|
|
4
|
+
* Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
|
5
|
+
* See https://llvm.org/LICENSE.txt for license information.
|
|
6
|
+
* SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
|
7
|
+
*
|
|
8
|
+
*===-----------------------------------------------------------------------===
|
|
9
|
+
*/
|
|
10
|
+
#ifndef __X86GPRINTRIN_H
|
|
11
|
+
#error \
|
|
12
|
+
"Never use <cmpccxaddintrin.h> directly; include <x86gprintrin.h> instead."
|
|
13
|
+
#endif // __X86GPRINTRIN_H
|
|
14
|
+
|
|
15
|
+
#ifndef __CMPCCXADDINTRIN_H
|
|
16
|
+
#define __CMPCCXADDINTRIN_H
|
|
17
|
+
#ifdef __x86_64__
|
|
18
|
+
|
|
19
|
+
typedef enum {
|
|
20
|
+
_CMPCCX_O, /* Overflow. */
|
|
21
|
+
_CMPCCX_NO, /* No overflow. */
|
|
22
|
+
_CMPCCX_B, /* Below. */
|
|
23
|
+
_CMPCCX_NB, /* Not below. */
|
|
24
|
+
_CMPCCX_Z, /* Zero. */
|
|
25
|
+
_CMPCCX_NZ, /* Not zero. */
|
|
26
|
+
_CMPCCX_BE, /* Below or equal. */
|
|
27
|
+
_CMPCCX_NBE, /* Neither below nor equal. */
|
|
28
|
+
_CMPCCX_S, /* Sign. */
|
|
29
|
+
_CMPCCX_NS, /* No sign. */
|
|
30
|
+
_CMPCCX_P, /* Parity. */
|
|
31
|
+
_CMPCCX_NP, /* No parity. */
|
|
32
|
+
_CMPCCX_L, /* Less. */
|
|
33
|
+
_CMPCCX_NL, /* Not less. */
|
|
34
|
+
_CMPCCX_LE, /* Less or equal. */
|
|
35
|
+
_CMPCCX_NLE, /* Neither less nor equal. */
|
|
36
|
+
} _CMPCCX_ENUM;
|
|
37
|
+
|
|
38
|
+
/// Compares the value from the memory __A with the value of __B. If the
|
|
39
|
+
/// specified condition __D is met, then add the third operand __C to the
|
|
40
|
+
/// __A and write it into __A, else the value of __A is unchanged. The return
|
|
41
|
+
/// value is the original value of __A.
|
|
42
|
+
///
|
|
43
|
+
/// \headerfile <immintrin.h>
|
|
44
|
+
///
|
|
45
|
+
/// This intrinsic corresponds to the \c CMPCCXADD instructions.
|
|
46
|
+
///
|
|
47
|
+
/// \param __A
|
|
48
|
+
/// __A pointer specifying the memory address.
|
|
49
|
+
///
|
|
50
|
+
/// \param __B
|
|
51
|
+
/// A integer operand.
|
|
52
|
+
///
|
|
53
|
+
/// \param __C
|
|
54
|
+
/// A integer operand.
|
|
55
|
+
///
|
|
56
|
+
/// \param __D
|
|
57
|
+
/// The specified condition.
|
|
58
|
+
///
|
|
59
|
+
/// \returns a integer which is the original value of first operand.
|
|
60
|
+
|
|
61
|
+
#define _cmpccxadd_epi32(__A, __B, __C, __D) \
|
|
62
|
+
((int)(__builtin_ia32_cmpccxadd32((void *)(__A), (int)(__B), (int)(__C), \
|
|
63
|
+
(int)(__D))))
|
|
64
|
+
|
|
65
|
+
#define _cmpccxadd_epi64(__A, __B, __C, __D) \
|
|
66
|
+
((long long)(__builtin_ia32_cmpccxadd64((void *)(__A), (long long)(__B), \
|
|
67
|
+
(long long)(__C), (int)(__D))))
|
|
68
|
+
|
|
69
|
+
#endif // __x86_64__
|
|
70
|
+
#endif // __CMPCCXADDINTRIN_H
|
|
@@ -195,13 +195,23 @@
|
|
|
195
195
|
#define bit_PCONFIG 0x00040000
|
|
196
196
|
#define bit_IBT 0x00100000
|
|
197
197
|
#define bit_AMXBF16 0x00400000
|
|
198
|
+
#define bit_AVX512FP16 0x00800000
|
|
198
199
|
#define bit_AMXTILE 0x01000000
|
|
199
200
|
#define bit_AMXINT8 0x02000000
|
|
200
201
|
|
|
201
202
|
/* Features in %eax for leaf 7 sub-leaf 1 */
|
|
202
|
-
#define
|
|
203
|
+
#define bit_RAOINT 0x00000008
|
|
204
|
+
#define bit_AVXVNNI 0x00000010
|
|
203
205
|
#define bit_AVX512BF16 0x00000020
|
|
206
|
+
#define bit_CMPCCXADD 0x00000080
|
|
207
|
+
#define bit_AMXFP16 0x00200000
|
|
204
208
|
#define bit_HRESET 0x00400000
|
|
209
|
+
#define bit_AVXIFMA 0x00800000
|
|
210
|
+
|
|
211
|
+
/* Features in %edx for leaf 7 sub-leaf 1 */
|
|
212
|
+
#define bit_AVXVNNIINT8 0x00000010
|
|
213
|
+
#define bit_AVXNECONVERT 0x00000020
|
|
214
|
+
#define bit_PREFETCHI 0x00004000
|
|
205
215
|
|
|
206
216
|
/* Features in %eax for leaf 13 sub-leaf 1 */
|
|
207
217
|
#define bit_XSAVEOPT 0x00000001
|
|
@@ -231,6 +241,7 @@
|
|
|
231
241
|
|
|
232
242
|
/* Features in %ebx for leaf 0x80000008 */
|
|
233
243
|
#define bit_CLZERO 0x00000001
|
|
244
|
+
#define bit_RDPRU 0x00000010
|
|
234
245
|
#define bit_WBNOINVD 0x00000200
|
|
235
246
|
|
|
236
247
|
|
|
@@ -259,7 +270,8 @@
|
|
|
259
270
|
: "0"(__leaf), "2"(__count))
|
|
260
271
|
#endif
|
|
261
272
|
|
|
262
|
-
static __inline int __get_cpuid_max (unsigned int __leaf,
|
|
273
|
+
static __inline unsigned int __get_cpuid_max (unsigned int __leaf,
|
|
274
|
+
unsigned int *__sig)
|
|
263
275
|
{
|
|
264
276
|
unsigned int __eax, __ebx, __ecx, __edx;
|
|
265
277
|
#if __i386__
|
|
@@ -0,0 +1,100 @@
|
|
|
1
|
+
/*===---- crc32intrin.h - SSE4.2 Accumulate CRC32 intrinsics ---------------===
|
|
2
|
+
*
|
|
3
|
+
* Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
|
4
|
+
* See https://llvm.org/LICENSE.txt for license information.
|
|
5
|
+
* SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
|
6
|
+
*
|
|
7
|
+
*===-----------------------------------------------------------------------===
|
|
8
|
+
*/
|
|
9
|
+
|
|
10
|
+
#ifndef __CRC32INTRIN_H
|
|
11
|
+
#define __CRC32INTRIN_H
|
|
12
|
+
|
|
13
|
+
#define __DEFAULT_FN_ATTRS \
|
|
14
|
+
__attribute__((__always_inline__, __nodebug__, __target__("crc32")))
|
|
15
|
+
|
|
16
|
+
/// Adds the unsigned integer operand to the CRC-32C checksum of the
|
|
17
|
+
/// unsigned char operand.
|
|
18
|
+
///
|
|
19
|
+
/// \headerfile <x86intrin.h>
|
|
20
|
+
///
|
|
21
|
+
/// This intrinsic corresponds to the <c> CRC32B </c> instruction.
|
|
22
|
+
///
|
|
23
|
+
/// \param __C
|
|
24
|
+
/// An unsigned integer operand to add to the CRC-32C checksum of operand
|
|
25
|
+
/// \a __D.
|
|
26
|
+
/// \param __D
|
|
27
|
+
/// An unsigned 8-bit integer operand used to compute the CRC-32C checksum.
|
|
28
|
+
/// \returns The result of adding operand \a __C to the CRC-32C checksum of
|
|
29
|
+
/// operand \a __D.
|
|
30
|
+
static __inline__ unsigned int __DEFAULT_FN_ATTRS
|
|
31
|
+
_mm_crc32_u8(unsigned int __C, unsigned char __D)
|
|
32
|
+
{
|
|
33
|
+
return __builtin_ia32_crc32qi(__C, __D);
|
|
34
|
+
}
|
|
35
|
+
|
|
36
|
+
/// Adds the unsigned integer operand to the CRC-32C checksum of the
|
|
37
|
+
/// unsigned short operand.
|
|
38
|
+
///
|
|
39
|
+
/// \headerfile <x86intrin.h>
|
|
40
|
+
///
|
|
41
|
+
/// This intrinsic corresponds to the <c> CRC32W </c> instruction.
|
|
42
|
+
///
|
|
43
|
+
/// \param __C
|
|
44
|
+
/// An unsigned integer operand to add to the CRC-32C checksum of operand
|
|
45
|
+
/// \a __D.
|
|
46
|
+
/// \param __D
|
|
47
|
+
/// An unsigned 16-bit integer operand used to compute the CRC-32C checksum.
|
|
48
|
+
/// \returns The result of adding operand \a __C to the CRC-32C checksum of
|
|
49
|
+
/// operand \a __D.
|
|
50
|
+
static __inline__ unsigned int __DEFAULT_FN_ATTRS
|
|
51
|
+
_mm_crc32_u16(unsigned int __C, unsigned short __D)
|
|
52
|
+
{
|
|
53
|
+
return __builtin_ia32_crc32hi(__C, __D);
|
|
54
|
+
}
|
|
55
|
+
|
|
56
|
+
/// Adds the first unsigned integer operand to the CRC-32C checksum of
|
|
57
|
+
/// the second unsigned integer operand.
|
|
58
|
+
///
|
|
59
|
+
/// \headerfile <x86intrin.h>
|
|
60
|
+
///
|
|
61
|
+
/// This intrinsic corresponds to the <c> CRC32L </c> instruction.
|
|
62
|
+
///
|
|
63
|
+
/// \param __C
|
|
64
|
+
/// An unsigned integer operand to add to the CRC-32C checksum of operand
|
|
65
|
+
/// \a __D.
|
|
66
|
+
/// \param __D
|
|
67
|
+
/// An unsigned 32-bit integer operand used to compute the CRC-32C checksum.
|
|
68
|
+
/// \returns The result of adding operand \a __C to the CRC-32C checksum of
|
|
69
|
+
/// operand \a __D.
|
|
70
|
+
static __inline__ unsigned int __DEFAULT_FN_ATTRS
|
|
71
|
+
_mm_crc32_u32(unsigned int __C, unsigned int __D)
|
|
72
|
+
{
|
|
73
|
+
return __builtin_ia32_crc32si(__C, __D);
|
|
74
|
+
}
|
|
75
|
+
|
|
76
|
+
#ifdef __x86_64__
|
|
77
|
+
/// Adds the unsigned integer operand to the CRC-32C checksum of the
|
|
78
|
+
/// unsigned 64-bit integer operand.
|
|
79
|
+
///
|
|
80
|
+
/// \headerfile <x86intrin.h>
|
|
81
|
+
///
|
|
82
|
+
/// This intrinsic corresponds to the <c> CRC32Q </c> instruction.
|
|
83
|
+
///
|
|
84
|
+
/// \param __C
|
|
85
|
+
/// An unsigned integer operand to add to the CRC-32C checksum of operand
|
|
86
|
+
/// \a __D.
|
|
87
|
+
/// \param __D
|
|
88
|
+
/// An unsigned 64-bit integer operand used to compute the CRC-32C checksum.
|
|
89
|
+
/// \returns The result of adding operand \a __C to the CRC-32C checksum of
|
|
90
|
+
/// operand \a __D.
|
|
91
|
+
static __inline__ unsigned long long __DEFAULT_FN_ATTRS
|
|
92
|
+
_mm_crc32_u64(unsigned long long __C, unsigned long long __D)
|
|
93
|
+
{
|
|
94
|
+
return __builtin_ia32_crc32di(__C, __D);
|
|
95
|
+
}
|
|
96
|
+
#endif /* __x86_64__ */
|
|
97
|
+
|
|
98
|
+
#undef __DEFAULT_FN_ATTRS
|
|
99
|
+
|
|
100
|
+
#endif /* __CRC32INTRIN_H */
|
|
@@ -0,0 +1,90 @@
|
|
|
1
|
+
/*===---- cmath - CUDA wrapper for <cmath> ---------------------------------===
|
|
2
|
+
*
|
|
3
|
+
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
4
|
+
* of this software and associated documentation files (the "Software"), to deal
|
|
5
|
+
* in the Software without restriction, including without limitation the rights
|
|
6
|
+
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
7
|
+
* copies of the Software, and to permit persons to whom the Software is
|
|
8
|
+
* furnished to do so, subject to the following conditions:
|
|
9
|
+
*
|
|
10
|
+
* The above copyright notice and this permission notice shall be included in
|
|
11
|
+
* all copies or substantial portions of the Software.
|
|
12
|
+
*
|
|
13
|
+
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
14
|
+
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
15
|
+
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
16
|
+
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
17
|
+
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
18
|
+
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
|
19
|
+
* THE SOFTWARE.
|
|
20
|
+
*
|
|
21
|
+
*===-----------------------------------------------------------------------===
|
|
22
|
+
*/
|
|
23
|
+
|
|
24
|
+
#ifndef __CLANG_CUDA_WRAPPERS_CMATH
|
|
25
|
+
#define __CLANG_CUDA_WRAPPERS_CMATH
|
|
26
|
+
|
|
27
|
+
#include_next <cmath>
|
|
28
|
+
|
|
29
|
+
#if defined(_LIBCPP_STD_VER)
|
|
30
|
+
|
|
31
|
+
// libc++ will need long double variants of these functions, but CUDA does not
|
|
32
|
+
// provide them. We'll provide their declarations, which should allow the
|
|
33
|
+
// headers to parse, but would not allow accidental use of them on a GPU.
|
|
34
|
+
|
|
35
|
+
__attribute__((device)) long double logb(long double);
|
|
36
|
+
__attribute__((device)) long double scalbn(long double, int);
|
|
37
|
+
|
|
38
|
+
namespace std {
|
|
39
|
+
|
|
40
|
+
// For __constexpr_fmin/fmax we only need device-side overloads before c++14
|
|
41
|
+
// where they are not constexpr.
|
|
42
|
+
#if _LIBCPP_STD_VER < 14
|
|
43
|
+
|
|
44
|
+
__attribute__((device))
|
|
45
|
+
inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 float __constexpr_fmax(float __x, float __y) _NOEXCEPT {
|
|
46
|
+
return __builtin_fmaxf(__x, __y);
|
|
47
|
+
}
|
|
48
|
+
|
|
49
|
+
__attribute__((device))
|
|
50
|
+
inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 double __constexpr_fmax(double __x, double __y) _NOEXCEPT {
|
|
51
|
+
return __builtin_fmax(__x, __y);
|
|
52
|
+
}
|
|
53
|
+
|
|
54
|
+
__attribute__((device))
|
|
55
|
+
inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 long double
|
|
56
|
+
__constexpr_fmax(long double __x, long double __y) _NOEXCEPT {
|
|
57
|
+
return __builtin_fmaxl(__x, __y);
|
|
58
|
+
}
|
|
59
|
+
|
|
60
|
+
template <class _Tp, class _Up, __enable_if_t<is_arithmetic<_Tp>::value && is_arithmetic<_Up>::value, int> = 0>
|
|
61
|
+
__attribute__((device))
|
|
62
|
+
_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 typename __promote<_Tp, _Up>::type
|
|
63
|
+
__constexpr_fmax(_Tp __x, _Up __y) _NOEXCEPT {
|
|
64
|
+
using __result_type = typename __promote<_Tp, _Up>::type;
|
|
65
|
+
return std::__constexpr_fmax(static_cast<__result_type>(__x), static_cast<__result_type>(__y));
|
|
66
|
+
}
|
|
67
|
+
#endif // _LIBCPP_STD_VER < 14
|
|
68
|
+
|
|
69
|
+
// For logb/scalbn templates we must always provide device overloads because
|
|
70
|
+
// libc++ implementation uses __builtin_XXX which gets translated into a libcall
|
|
71
|
+
// which we can't handle on GPU. We need to forward those to CUDA-provided
|
|
72
|
+
// implementations.
|
|
73
|
+
|
|
74
|
+
template <class _Tp>
|
|
75
|
+
__attribute__((device))
|
|
76
|
+
_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _Tp __constexpr_logb(_Tp __x) {
|
|
77
|
+
return ::logb(__x);
|
|
78
|
+
}
|
|
79
|
+
|
|
80
|
+
template <class _Tp>
|
|
81
|
+
__attribute__((device))
|
|
82
|
+
_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _Tp __constexpr_scalbn(_Tp __x, int __exp) {
|
|
83
|
+
return ::scalbn(__x, __exp);
|
|
84
|
+
}
|
|
85
|
+
|
|
86
|
+
} // namespace std//
|
|
87
|
+
|
|
88
|
+
#endif // _LIBCPP_STD_VER
|
|
89
|
+
|
|
90
|
+
#endif // include guard
|