@cpp-linter/clang-tidy-linux-arm64 20.1.0 → 20.1.8
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/20/include/__clang_cuda_intrinsics.h +8 -8
- package/lib/clang/20/include/amdgpuintrin.h +1 -1
- package/lib/clang/20/include/avx10_2_512convertintrin.h +24 -22
- package/lib/clang/20/include/avx10_2_512satcvtdsintrin.h +28 -24
- package/lib/clang/20/include/avx10_2_512satcvtintrin.h +84 -84
- package/lib/clang/20/include/avx10_2convertintrin.h +50 -62
- package/lib/clang/20/include/avx10_2minmaxintrin.h +0 -45
- package/lib/clang/20/include/avx10_2niintrin.h +0 -1666
- package/lib/clang/20/include/avx10_2satcvtdsintrin.h +52 -158
- package/lib/clang/20/include/avx10_2satcvtintrin.h +88 -200
- package/lib/clang/20/include/nvptxintrin.h +2 -1
- package/lib/clang/20/include/vecintrin.h +5 -0
- package/package.json +5 -5
package/README.md
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
# @cpp-linter/clang-tidy-linux-arm64
|
|
2
2
|
|
|
3
|
-
The Linux arm64 binary of clang-tidy from LLVM 20.1.
|
|
3
|
+
The Linux arm64 binary of clang-tidy from LLVM 20.1.8, for [`@cpp-linter/clang-tidy`](https://www.npmjs.com/package/@cpp-linter/clang-tidy).
|
|
4
4
|
|
|
5
5
|
Install `@cpp-linter/clang-tidy` instead of this package: npm picks the binary for your platform.
|
|
6
6
|
|
|
7
|
-
The binary is a static build from the release [`llvm-20.1.
|
|
7
|
+
The binary is a static build from the release [`llvm-20.1.8`](https://github.com/cpp-linter/clang-tools-static-binaries/releases/tag/llvm-20.1.8) of [cpp-linter/clang-tools-static-binaries](https://github.com/cpp-linter/clang-tools-static-binaries). LLVM is licensed under the Apache License v2.0 with LLVM Exceptions (see `LICENSE.TXT`).
|
package/bin/clang-tidy
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
1
|
[diffend] Oversized file quarantined before diffing.
|
|
2
2
|
name: package/bin/clang-tidy
|
|
3
|
-
size:
|
|
4
|
-
sha256:
|
|
3
|
+
size: 73144824 bytes
|
|
4
|
+
sha256: 4ce6994a0c5e5f7080d63215450959ab91b8d2272d95a79901cacf1c36e9a83d
|
|
@@ -515,32 +515,32 @@ __device__ inline cuuint32_t __nvvm_get_smem_pointer(void *__ptr) {
|
|
|
515
515
|
#if !defined(__CUDA_ARCH__) || __CUDA_ARCH__ >= 800
|
|
516
516
|
__device__ inline unsigned __reduce_add_sync(unsigned __mask,
|
|
517
517
|
unsigned __value) {
|
|
518
|
-
return __nvvm_redux_sync_add(
|
|
518
|
+
return __nvvm_redux_sync_add(__value, __mask);
|
|
519
519
|
}
|
|
520
520
|
__device__ inline unsigned __reduce_min_sync(unsigned __mask,
|
|
521
521
|
unsigned __value) {
|
|
522
|
-
return __nvvm_redux_sync_umin(
|
|
522
|
+
return __nvvm_redux_sync_umin(__value, __mask);
|
|
523
523
|
}
|
|
524
524
|
__device__ inline unsigned __reduce_max_sync(unsigned __mask,
|
|
525
525
|
unsigned __value) {
|
|
526
|
-
return __nvvm_redux_sync_umax(
|
|
526
|
+
return __nvvm_redux_sync_umax(__value, __mask);
|
|
527
527
|
}
|
|
528
528
|
__device__ inline int __reduce_min_sync(unsigned __mask, int __value) {
|
|
529
|
-
return __nvvm_redux_sync_min(
|
|
529
|
+
return __nvvm_redux_sync_min(__value, __mask);
|
|
530
530
|
}
|
|
531
531
|
__device__ inline int __reduce_max_sync(unsigned __mask, int __value) {
|
|
532
|
-
return __nvvm_redux_sync_max(
|
|
532
|
+
return __nvvm_redux_sync_max(__value, __mask);
|
|
533
533
|
}
|
|
534
534
|
__device__ inline unsigned __reduce_or_sync(unsigned __mask, unsigned __value) {
|
|
535
|
-
return __nvvm_redux_sync_or(
|
|
535
|
+
return __nvvm_redux_sync_or(__value, __mask);
|
|
536
536
|
}
|
|
537
537
|
__device__ inline unsigned __reduce_and_sync(unsigned __mask,
|
|
538
538
|
unsigned __value) {
|
|
539
|
-
return __nvvm_redux_sync_and(
|
|
539
|
+
return __nvvm_redux_sync_and(__value, __mask);
|
|
540
540
|
}
|
|
541
541
|
__device__ inline unsigned __reduce_xor_sync(unsigned __mask,
|
|
542
542
|
unsigned __value) {
|
|
543
|
-
return __nvvm_redux_sync_xor(
|
|
543
|
+
return __nvvm_redux_sync_xor(__value, __mask);
|
|
544
544
|
}
|
|
545
545
|
|
|
546
546
|
__device__ inline void __nv_memcpy_async_shared_global_4(void *__dst,
|
|
@@ -121,7 +121,7 @@ __gpu_read_first_lane_u64(uint64_t __lane_mask, uint64_t __x) {
|
|
|
121
121
|
uint32_t __hi = (uint32_t)(__x >> 32ull);
|
|
122
122
|
uint32_t __lo = (uint32_t)(__x & 0xFFFFFFFF);
|
|
123
123
|
return ((uint64_t)__builtin_amdgcn_readfirstlane(__hi) << 32ull) |
|
|
124
|
-
((uint64_t)__builtin_amdgcn_readfirstlane(__lo));
|
|
124
|
+
((uint64_t)__builtin_amdgcn_readfirstlane(__lo) & 0xFFFFFFFF);
|
|
125
125
|
}
|
|
126
126
|
|
|
127
127
|
// Returns a bitmask of threads in the current lane for which \p x is true.
|
|
@@ -78,20 +78,20 @@ _mm512_maskz_cvtbiasph_bf8(__mmask32 __U, __m512i __A, __m512h __B) {
|
|
|
78
78
|
}
|
|
79
79
|
|
|
80
80
|
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
81
|
-
|
|
81
|
+
_mm512_cvts_biasph_bf8(__m512i __A, __m512h __B) {
|
|
82
82
|
return (__m256i)__builtin_ia32_vcvtbiasph2bf8s_512_mask(
|
|
83
83
|
(__v64qi)__A, (__v32hf)__B, (__v32qi)_mm256_undefined_si256(),
|
|
84
84
|
(__mmask32)-1);
|
|
85
85
|
}
|
|
86
86
|
|
|
87
|
-
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
87
|
+
static __inline__ __m256i __DEFAULT_FN_ATTRS512 _mm512_mask_cvts_biasph_bf8(
|
|
88
88
|
__m256i __W, __mmask32 __U, __m512i __A, __m512h __B) {
|
|
89
89
|
return (__m256i)__builtin_ia32_vcvtbiasph2bf8s_512_mask(
|
|
90
90
|
(__v64qi)__A, (__v32hf)__B, (__v32qi)(__m256i)__W, (__mmask32)__U);
|
|
91
91
|
}
|
|
92
92
|
|
|
93
93
|
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
94
|
-
|
|
94
|
+
_mm512_maskz_cvts_biasph_bf8(__mmask32 __U, __m512i __A, __m512h __B) {
|
|
95
95
|
return (__m256i)__builtin_ia32_vcvtbiasph2bf8s_512_mask(
|
|
96
96
|
(__v64qi)__A, (__v32hf)__B, (__v32qi)(__m256i)_mm256_setzero_si256(),
|
|
97
97
|
(__mmask32)__U);
|
|
@@ -118,20 +118,20 @@ _mm512_maskz_cvtbiasph_hf8(__mmask32 __U, __m512i __A, __m512h __B) {
|
|
|
118
118
|
}
|
|
119
119
|
|
|
120
120
|
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
121
|
-
|
|
121
|
+
_mm512_cvts_biasph_hf8(__m512i __A, __m512h __B) {
|
|
122
122
|
return (__m256i)__builtin_ia32_vcvtbiasph2hf8s_512_mask(
|
|
123
123
|
(__v64qi)__A, (__v32hf)__B, (__v32qi)_mm256_undefined_si256(),
|
|
124
124
|
(__mmask32)-1);
|
|
125
125
|
}
|
|
126
126
|
|
|
127
|
-
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
127
|
+
static __inline__ __m256i __DEFAULT_FN_ATTRS512 _mm512_mask_cvts_biasph_hf8(
|
|
128
128
|
__m256i __W, __mmask32 __U, __m512i __A, __m512h __B) {
|
|
129
129
|
return (__m256i)__builtin_ia32_vcvtbiasph2hf8s_512_mask(
|
|
130
130
|
(__v64qi)__A, (__v32hf)__B, (__v32qi)(__m256i)__W, (__mmask32)__U);
|
|
131
131
|
}
|
|
132
132
|
|
|
133
133
|
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
134
|
-
|
|
134
|
+
_mm512_maskz_cvts_biasph_hf8(__mmask32 __U, __m512i __A, __m512h __B) {
|
|
135
135
|
return (__m256i)__builtin_ia32_vcvtbiasph2hf8s_512_mask(
|
|
136
136
|
(__v64qi)__A, (__v32hf)__B, (__v32qi)(__m256i)_mm256_setzero_si256(),
|
|
137
137
|
(__mmask32)__U);
|
|
@@ -157,21 +157,21 @@ _mm512_maskz_cvt2ph_bf8(__mmask64 __U, __m512h __A, __m512h __B) {
|
|
|
157
157
|
}
|
|
158
158
|
|
|
159
159
|
static __inline__ __m512i __DEFAULT_FN_ATTRS512
|
|
160
|
-
|
|
160
|
+
_mm512_cvts_2ph_bf8(__m512h __A, __m512h __B) {
|
|
161
161
|
return (__m512i)__builtin_ia32_vcvt2ph2bf8s_512((__v32hf)(__A),
|
|
162
162
|
(__v32hf)(__B));
|
|
163
163
|
}
|
|
164
164
|
|
|
165
165
|
static __inline__ __m512i __DEFAULT_FN_ATTRS512
|
|
166
|
-
|
|
166
|
+
_mm512_mask_cvts_2ph_bf8(__m512i __W, __mmask64 __U, __m512h __A, __m512h __B) {
|
|
167
167
|
return (__m512i)__builtin_ia32_selectb_512(
|
|
168
|
-
(__mmask64)__U, (__v64qi)
|
|
168
|
+
(__mmask64)__U, (__v64qi)_mm512_cvts_2ph_bf8(__A, __B), (__v64qi)__W);
|
|
169
169
|
}
|
|
170
170
|
|
|
171
171
|
static __inline__ __m512i __DEFAULT_FN_ATTRS512
|
|
172
|
-
|
|
172
|
+
_mm512_maskz_cvts_2ph_bf8(__mmask64 __U, __m512h __A, __m512h __B) {
|
|
173
173
|
return (__m512i)__builtin_ia32_selectb_512(
|
|
174
|
-
(__mmask64)__U, (__v64qi)
|
|
174
|
+
(__mmask64)__U, (__v64qi)_mm512_cvts_2ph_bf8(__A, __B),
|
|
175
175
|
(__v64qi)(__m512i)_mm512_setzero_si512());
|
|
176
176
|
}
|
|
177
177
|
|
|
@@ -195,21 +195,21 @@ _mm512_maskz_cvt2ph_hf8(__mmask64 __U, __m512h __A, __m512h __B) {
|
|
|
195
195
|
}
|
|
196
196
|
|
|
197
197
|
static __inline__ __m512i __DEFAULT_FN_ATTRS512
|
|
198
|
-
|
|
198
|
+
_mm512_cvts_2ph_hf8(__m512h __A, __m512h __B) {
|
|
199
199
|
return (__m512i)__builtin_ia32_vcvt2ph2hf8s_512((__v32hf)(__A),
|
|
200
200
|
(__v32hf)(__B));
|
|
201
201
|
}
|
|
202
202
|
|
|
203
203
|
static __inline__ __m512i __DEFAULT_FN_ATTRS512
|
|
204
|
-
|
|
204
|
+
_mm512_mask_cvts_2ph_hf8(__m512i __W, __mmask64 __U, __m512h __A, __m512h __B) {
|
|
205
205
|
return (__m512i)__builtin_ia32_selectb_512(
|
|
206
|
-
(__mmask64)__U, (__v64qi)
|
|
206
|
+
(__mmask64)__U, (__v64qi)_mm512_cvts_2ph_hf8(__A, __B), (__v64qi)__W);
|
|
207
207
|
}
|
|
208
208
|
|
|
209
209
|
static __inline__ __m512i __DEFAULT_FN_ATTRS512
|
|
210
|
-
|
|
210
|
+
_mm512_maskz_cvts_2ph_hf8(__mmask64 __U, __m512h __A, __m512h __B) {
|
|
211
211
|
return (__m512i)__builtin_ia32_selectb_512(
|
|
212
|
-
(__mmask64)__U, (__v64qi)
|
|
212
|
+
(__mmask64)__U, (__v64qi)_mm512_cvts_2ph_hf8(__A, __B),
|
|
213
213
|
(__v64qi)(__m512i)_mm512_setzero_si512());
|
|
214
214
|
}
|
|
215
215
|
|
|
@@ -247,19 +247,20 @@ _mm512_maskz_cvtph_bf8(__mmask32 __U, __m512h __A) {
|
|
|
247
247
|
(__v32hf)__A, (__v32qi)(__m256i)_mm256_setzero_si256(), (__mmask32)__U);
|
|
248
248
|
}
|
|
249
249
|
|
|
250
|
-
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
250
|
+
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
251
|
+
_mm512_cvts_ph_bf8(__m512h __A) {
|
|
251
252
|
return (__m256i)__builtin_ia32_vcvtph2bf8s_512_mask(
|
|
252
253
|
(__v32hf)__A, (__v32qi)(__m256i)_mm256_undefined_si256(), (__mmask32)-1);
|
|
253
254
|
}
|
|
254
255
|
|
|
255
256
|
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
256
|
-
|
|
257
|
+
_mm512_mask_cvts_ph_bf8(__m256i __W, __mmask32 __U, __m512h __A) {
|
|
257
258
|
return (__m256i)__builtin_ia32_vcvtph2bf8s_512_mask(
|
|
258
259
|
(__v32hf)__A, (__v32qi)(__m256i)__W, (__mmask32)__U);
|
|
259
260
|
}
|
|
260
261
|
|
|
261
262
|
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
262
|
-
|
|
263
|
+
_mm512_maskz_cvts_ph_bf8(__mmask32 __U, __m512h __A) {
|
|
263
264
|
return (__m256i)__builtin_ia32_vcvtph2bf8s_512_mask(
|
|
264
265
|
(__v32hf)__A, (__v32qi)(__m256i)_mm256_setzero_si256(), (__mmask32)__U);
|
|
265
266
|
}
|
|
@@ -281,19 +282,20 @@ _mm512_maskz_cvtph_hf8(__mmask32 __U, __m512h __A) {
|
|
|
281
282
|
(__v32hf)__A, (__v32qi)(__m256i)_mm256_setzero_si256(), (__mmask32)__U);
|
|
282
283
|
}
|
|
283
284
|
|
|
284
|
-
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
285
|
+
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
286
|
+
_mm512_cvts_ph_hf8(__m512h __A) {
|
|
285
287
|
return (__m256i)__builtin_ia32_vcvtph2hf8s_512_mask(
|
|
286
288
|
(__v32hf)__A, (__v32qi)(__m256i)_mm256_undefined_si256(), (__mmask32)-1);
|
|
287
289
|
}
|
|
288
290
|
|
|
289
291
|
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
290
|
-
|
|
292
|
+
_mm512_mask_cvts_ph_hf8(__m256i __W, __mmask32 __U, __m512h __A) {
|
|
291
293
|
return (__m256i)__builtin_ia32_vcvtph2hf8s_512_mask(
|
|
292
294
|
(__v32hf)__A, (__v32qi)(__m256i)__W, (__mmask32)__U);
|
|
293
295
|
}
|
|
294
296
|
|
|
295
297
|
static __inline__ __m256i __DEFAULT_FN_ATTRS512
|
|
296
|
-
|
|
298
|
+
_mm512_maskz_cvts_ph_hf8(__mmask32 __U, __m512h __A) {
|
|
297
299
|
return (__m256i)__builtin_ia32_vcvtph2hf8s_512_mask(
|
|
298
300
|
(__v32hf)__A, (__v32qi)(__m256i)_mm256_setzero_si256(), (__mmask32)__U);
|
|
299
301
|
}
|
|
@@ -20,20 +20,21 @@
|
|
|
20
20
|
__min_vector_width__(512)))
|
|
21
21
|
|
|
22
22
|
// 512 bit : Double -> Int
|
|
23
|
-
static __inline__ __m256i __DEFAULT_FN_ATTRS
|
|
23
|
+
static __inline__ __m256i __DEFAULT_FN_ATTRS
|
|
24
|
+
_mm512_cvtts_pd_epi32(__m512d __A) {
|
|
24
25
|
return ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask(
|
|
25
26
|
(__v8df)__A, (__v8si)_mm256_undefined_si256(), (__mmask8)-1,
|
|
26
27
|
_MM_FROUND_CUR_DIRECTION));
|
|
27
28
|
}
|
|
28
29
|
|
|
29
30
|
static __inline__ __m256i __DEFAULT_FN_ATTRS
|
|
30
|
-
|
|
31
|
+
_mm512_mask_cvtts_pd_epi32(__m256i __W, __mmask8 __U, __m512d __A) {
|
|
31
32
|
return ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask(
|
|
32
33
|
(__v8df)__A, (__v8si)__W, __U, _MM_FROUND_CUR_DIRECTION));
|
|
33
34
|
}
|
|
34
35
|
|
|
35
36
|
static __inline__ __m256i __DEFAULT_FN_ATTRS
|
|
36
|
-
|
|
37
|
+
_mm512_maskz_cvtts_pd_epi32(__mmask8 __U, __m512d __A) {
|
|
37
38
|
return ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask(
|
|
38
39
|
(__v8df)__A, (__v8si)_mm256_setzero_si256(), __U,
|
|
39
40
|
_MM_FROUND_CUR_DIRECTION));
|
|
@@ -55,20 +56,21 @@ _mm512_maskz_cvttspd_epi32(__mmask8 __U, __m512d __A) {
|
|
|
55
56
|
(const int)(__R)))
|
|
56
57
|
|
|
57
58
|
// 512 bit : Double -> uInt
|
|
58
|
-
static __inline__ __m256i __DEFAULT_FN_ATTRS
|
|
59
|
+
static __inline__ __m256i __DEFAULT_FN_ATTRS
|
|
60
|
+
_mm512_cvtts_pd_epu32(__m512d __A) {
|
|
59
61
|
return ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask(
|
|
60
62
|
(__v8df)__A, (__v8si)_mm256_undefined_si256(), (__mmask8)-1,
|
|
61
63
|
_MM_FROUND_CUR_DIRECTION));
|
|
62
64
|
}
|
|
63
65
|
|
|
64
66
|
static __inline__ __m256i __DEFAULT_FN_ATTRS
|
|
65
|
-
|
|
67
|
+
_mm512_mask_cvtts_pd_epu32(__m256i __W, __mmask8 __U, __m512d __A) {
|
|
66
68
|
return ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask(
|
|
67
69
|
(__v8df)__A, (__v8si)__W, __U, _MM_FROUND_CUR_DIRECTION));
|
|
68
70
|
}
|
|
69
71
|
|
|
70
72
|
static __inline__ __m256i __DEFAULT_FN_ATTRS
|
|
71
|
-
|
|
73
|
+
_mm512_maskz_cvtts_pd_epu32(__mmask8 __U, __m512d __A) {
|
|
72
74
|
return ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask(
|
|
73
75
|
(__v8df)__A, (__v8si)_mm256_setzero_si256(), __U,
|
|
74
76
|
_MM_FROUND_CUR_DIRECTION));
|
|
@@ -91,18 +93,19 @@ _mm512_maskz_cvttspd_epu32(__mmask8 __U, __m512d __A) {
|
|
|
91
93
|
|
|
92
94
|
// 512 bit : Double -> Long
|
|
93
95
|
|
|
94
|
-
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
96
|
+
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
97
|
+
_mm512_cvtts_pd_epi64(__m512d __A) {
|
|
95
98
|
return ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask(
|
|
96
99
|
(__v8df)__A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
|
|
97
100
|
_MM_FROUND_CUR_DIRECTION));
|
|
98
101
|
}
|
|
99
102
|
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
100
|
-
|
|
103
|
+
_mm512_mask_cvtts_pd_epi64(__m512i __W, __mmask8 __U, __m512d __A) {
|
|
101
104
|
return ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask(
|
|
102
105
|
(__v8df)__A, (__v8di)__W, __U, _MM_FROUND_CUR_DIRECTION));
|
|
103
106
|
}
|
|
104
107
|
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
105
|
-
|
|
108
|
+
_mm512_maskz_cvtts_pd_epi64(__mmask8 __U, __m512d __A) {
|
|
106
109
|
return ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask(
|
|
107
110
|
(__v8df)__A, (__v8di)_mm512_setzero_si512(), __U,
|
|
108
111
|
_MM_FROUND_CUR_DIRECTION));
|
|
@@ -125,20 +128,21 @@ _mm512_maskz_cvttspd_epi64(__mmask8 __U, __m512d __A) {
|
|
|
125
128
|
|
|
126
129
|
// 512 bit : Double -> ULong
|
|
127
130
|
|
|
128
|
-
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
131
|
+
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
132
|
+
_mm512_cvtts_pd_epu64(__m512d __A) {
|
|
129
133
|
return ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask(
|
|
130
134
|
(__v8df)__A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
|
|
131
135
|
_MM_FROUND_CUR_DIRECTION));
|
|
132
136
|
}
|
|
133
137
|
|
|
134
138
|
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
135
|
-
|
|
139
|
+
_mm512_mask_cvtts_pd_epu64(__m512i __W, __mmask8 __U, __m512d __A) {
|
|
136
140
|
return ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask(
|
|
137
141
|
(__v8df)__A, (__v8di)__W, __U, _MM_FROUND_CUR_DIRECTION));
|
|
138
142
|
}
|
|
139
143
|
|
|
140
144
|
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
141
|
-
|
|
145
|
+
_mm512_maskz_cvtts_pd_epu64(__mmask8 __U, __m512d __A) {
|
|
142
146
|
return ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask(
|
|
143
147
|
(__v8df)__A, (__v8di)_mm512_setzero_si512(), __U,
|
|
144
148
|
_MM_FROUND_CUR_DIRECTION));
|
|
@@ -160,20 +164,20 @@ _mm512_maskz_cvttspd_epu64(__mmask8 __U, __m512d __A) {
|
|
|
160
164
|
(const int)(__R)))
|
|
161
165
|
|
|
162
166
|
// 512 bit: Float -> int
|
|
163
|
-
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
167
|
+
static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvtts_ps_epi32(__m512 __A) {
|
|
164
168
|
return ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask(
|
|
165
169
|
(__v16sf)(__A), (__v16si)_mm512_undefined_epi32(), (__mmask16)-1,
|
|
166
170
|
_MM_FROUND_CUR_DIRECTION));
|
|
167
171
|
}
|
|
168
172
|
|
|
169
173
|
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
170
|
-
|
|
174
|
+
_mm512_mask_cvtts_ps_epi32(__m512i __W, __mmask16 __U, __m512 __A) {
|
|
171
175
|
return ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask(
|
|
172
176
|
(__v16sf)(__A), (__v16si)(__W), __U, _MM_FROUND_CUR_DIRECTION));
|
|
173
177
|
}
|
|
174
178
|
|
|
175
179
|
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
176
|
-
|
|
180
|
+
_mm512_maskz_cvtts_ps_epi32(__mmask16 __U, __m512 __A) {
|
|
177
181
|
return ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask(
|
|
178
182
|
(__v16sf)(__A), (__v16si)_mm512_setzero_si512(), __U,
|
|
179
183
|
_MM_FROUND_CUR_DIRECTION));
|
|
@@ -195,20 +199,20 @@ _mm512_maskz_cvttsps_epi32(__mmask16 __U, __m512 __A) {
|
|
|
195
199
|
(__mmask16)(__U), (const int)(__R)))
|
|
196
200
|
|
|
197
201
|
// 512 bit: Float -> uint
|
|
198
|
-
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
202
|
+
static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvtts_ps_epu32(__m512 __A) {
|
|
199
203
|
return ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask(
|
|
200
204
|
(__v16sf)(__A), (__v16si)_mm512_undefined_epi32(), (__mmask16)-1,
|
|
201
205
|
_MM_FROUND_CUR_DIRECTION));
|
|
202
206
|
}
|
|
203
207
|
|
|
204
208
|
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
205
|
-
|
|
209
|
+
_mm512_mask_cvtts_ps_epu32(__m512i __W, __mmask16 __U, __m512 __A) {
|
|
206
210
|
return ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask(
|
|
207
211
|
(__v16sf)(__A), (__v16si)(__W), __U, _MM_FROUND_CUR_DIRECTION));
|
|
208
212
|
}
|
|
209
213
|
|
|
210
214
|
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
211
|
-
|
|
215
|
+
_mm512_maskz_cvtts_ps_epu32(__mmask16 __U, __m512 __A) {
|
|
212
216
|
return ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask(
|
|
213
217
|
(__v16sf)(__A), (__v16si)_mm512_setzero_si512(), __U,
|
|
214
218
|
_MM_FROUND_CUR_DIRECTION));
|
|
@@ -230,20 +234,20 @@ _mm512_maskz_cvttsps_epu32(__mmask16 __U, __m512 __A) {
|
|
|
230
234
|
(__mmask16)(__U), (const int)(__R)))
|
|
231
235
|
|
|
232
236
|
// 512 bit : float -> long
|
|
233
|
-
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
237
|
+
static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvtts_ps_epi64(__m256 __A) {
|
|
234
238
|
return ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask(
|
|
235
239
|
(__v8sf)__A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
|
|
236
240
|
_MM_FROUND_CUR_DIRECTION));
|
|
237
241
|
}
|
|
238
242
|
|
|
239
243
|
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
240
|
-
|
|
244
|
+
_mm512_mask_cvtts_ps_epi64(__m512i __W, __mmask8 __U, __m256 __A) {
|
|
241
245
|
return ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask(
|
|
242
246
|
(__v8sf)__A, (__v8di)__W, __U, _MM_FROUND_CUR_DIRECTION));
|
|
243
247
|
}
|
|
244
248
|
|
|
245
249
|
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
246
|
-
|
|
250
|
+
_mm512_maskz_cvtts_ps_epi64(__mmask8 __U, __m256 __A) {
|
|
247
251
|
return ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask(
|
|
248
252
|
(__v8sf)__A, (__v8di)_mm512_setzero_si512(), __U,
|
|
249
253
|
_MM_FROUND_CUR_DIRECTION));
|
|
@@ -265,20 +269,20 @@ _mm512_maskz_cvttsps_epi64(__mmask8 __U, __m256 __A) {
|
|
|
265
269
|
(const int)(__R)))
|
|
266
270
|
|
|
267
271
|
// 512 bit : float -> ulong
|
|
268
|
-
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
272
|
+
static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvtts_ps_epu64(__m256 __A) {
|
|
269
273
|
return ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask(
|
|
270
274
|
(__v8sf)__A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
|
|
271
275
|
_MM_FROUND_CUR_DIRECTION));
|
|
272
276
|
}
|
|
273
277
|
|
|
274
278
|
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
275
|
-
|
|
279
|
+
_mm512_mask_cvtts_ps_epu64(__m512i __W, __mmask8 __U, __m256 __A) {
|
|
276
280
|
return ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask(
|
|
277
281
|
(__v8sf)__A, (__v8di)__W, __U, _MM_FROUND_CUR_DIRECTION));
|
|
278
282
|
}
|
|
279
283
|
|
|
280
284
|
static __inline__ __m512i __DEFAULT_FN_ATTRS
|
|
281
|
-
|
|
285
|
+
_mm512_maskz_cvtts_ps_epu64(__mmask8 __U, __m256 __A) {
|
|
282
286
|
return ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask(
|
|
283
287
|
(__v8sf)__A, (__v8di)_mm512_setzero_si512(), __U,
|
|
284
288
|
_MM_FROUND_CUR_DIRECTION));
|