1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
|
From 78c9ec9000f873abe7a15a91b87080a2e4308260 Mon Sep 17 00:00:00 2001
From: "H.J. Lu" <hjl.tools@gmail.com>
Date: Fri, 20 Aug 2021 06:42:24 -0700
Subject: [PATCH] x86-64: Optimize load of all bits set into ZMM register [BZ
#28252]
Optimize loads of all bits set into ZMM register in AVX512 SVML codes
by replacing
vpbroadcastq .L_2il0floatpacket.16(%rip), %zmmX
and
vmovups .L_2il0floatpacket.13(%rip), %zmmX
with
vpternlogd $0xff, %zmmX, %zmmX, %zmmX
This fixes BZ #28252.
---
sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core_avx512.S | 7 +------
sysdeps/x86_64/fpu/multiarch/svml_d_log8_core_avx512.S | 7 +------
sysdeps/x86_64/fpu/multiarch/svml_d_sin8_core_avx512.S | 7 +------
sysdeps/x86_64/fpu/multiarch/svml_d_sincos8_core_avx512.S | 7 +------
sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core_avx512.S | 7 +------
sysdeps/x86_64/fpu/multiarch/svml_s_expf16_core_avx512.S | 7 +------
sysdeps/x86_64/fpu/multiarch/svml_s_logf16_core_avx512.S | 7 +------
sysdeps/x86_64/fpu/multiarch/svml_s_powf16_core_avx512.S | 12 ++----------
sysdeps/x86_64/fpu/multiarch/svml_s_sincosf16_core_avx512.S | 7 +------
sysdeps/x86_64/fpu/multiarch/svml_s_sinf16_core_avx512.S | 7 +------
10 files changed, 11 insertions(+), 64 deletions(-)
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core_avx512.S
index e68fcdb..58e588a 100644
--- a/sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core_avx512.S
+++ b/sysdeps/x86_64/fpu/multiarch/svml_d_cos8_core_avx512.S
@@ -265,7 +265,7 @@ WRAPPER_IMPL_AVX512 _ZGVdN4v_cos
vmovaps %zmm0, %zmm8
/* Check for large arguments path */
- vpbroadcastq .L_2il0floatpacket.16(%rip), %zmm2
+ vpternlogd $0xff, %zmm2, %zmm2, %zmm2
/*
ARGUMENT RANGE REDUCTION:
@@ -456,8 +456,3 @@ WRAPPER_IMPL_AVX512 _ZGVdN4v_cos
jmp .LBL_2_7
#endif
END (_ZGVeN8v_cos_skx)
-
- .section .rodata, "a"
-.L_2il0floatpacket.16:
- .long 0xffffffff,0xffffffff
- .type .L_2il0floatpacket.16,@object
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_log8_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_log8_core_avx512.S
index dfa2aca..f5f117d 100644
--- a/sysdeps/x86_64/fpu/multiarch/svml_d_log8_core_avx512.S
+++ b/sysdeps/x86_64/fpu/multiarch/svml_d_log8_core_avx512.S
@@ -274,7 +274,7 @@ WRAPPER_IMPL_AVX512 _ZGVdN4v_log
/* preserve mantissa, set input exponent to 2^(-10) */
vpternlogq $248, _ExpMask(%rax), %zmm3, %zmm2
- vpbroadcastq .L_2il0floatpacket.12(%rip), %zmm1
+ vpternlogd $0xff, %zmm1, %zmm1, %zmm1
vpsrlq $32, %zmm4, %zmm6
/* reciprocal approximation good to at least 11 bits */
@@ -461,8 +461,3 @@ WRAPPER_IMPL_AVX512 _ZGVdN4v_log
jmp .LBL_2_7
#endif
END (_ZGVeN8v_log_skx)
-
- .section .rodata, "a"
-.L_2il0floatpacket.12:
- .long 0xffffffff,0xffffffff
- .type .L_2il0floatpacket.12,@object
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_sin8_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_sin8_core_avx512.S
index be8ab7c..48d251d 100644
--- a/sysdeps/x86_64/fpu/multiarch/svml_d_sin8_core_avx512.S
+++ b/sysdeps/x86_64/fpu/multiarch/svml_d_sin8_core_avx512.S
@@ -261,7 +261,7 @@ WRAPPER_IMPL_AVX512 _ZGVdN4v_sin
andq $-64, %rsp
subq $1280, %rsp
movq __svml_d_trig_data@GOTPCREL(%rip), %rax
- vpbroadcastq .L_2il0floatpacket.14(%rip), %zmm14
+ vpternlogd $0xff, %zmm1, %zmm1, %zmm14
vmovups __dAbsMask(%rax), %zmm7
vmovups __dInvPI(%rax), %zmm2
vmovups __dRShifter(%rax), %zmm1
@@ -458,8 +458,3 @@ WRAPPER_IMPL_AVX512 _ZGVdN4v_sin
jmp .LBL_2_7
#endif
END (_ZGVeN8v_sin_skx)
-
- .section .rodata, "a"
-.L_2il0floatpacket.14:
- .long 0xffffffff,0xffffffff
- .type .L_2il0floatpacket.14,@object
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_sincos8_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_d_sincos8_core_avx512.S
index 6118870..a4944a4 100644
--- a/sysdeps/x86_64/fpu/multiarch/svml_d_sincos8_core_avx512.S
+++ b/sysdeps/x86_64/fpu/multiarch/svml_d_sincos8_core_avx512.S
@@ -430,7 +430,7 @@ WRAPPER_IMPL_AVX512_fFF _ZGVdN4vl8l8_sincos
/* SinPoly = SinR*SinPoly */
vfmadd213pd %zmm5, %zmm5, %zmm4
- vpbroadcastq .L_2il0floatpacket.15(%rip), %zmm3
+ vpternlogd $0xff, %zmm3, %zmm3, %zmm3
/* Update Cos result's sign */
vxorpd %zmm2, %zmm1, %zmm1
@@ -741,8 +741,3 @@ END (_ZGVeN8vvv_sincos_knl)
ENTRY (_ZGVeN8vvv_sincos_skx)
WRAPPER_AVX512_vvv_vl8l8 _ZGVeN8vl8l8_sincos_skx
END (_ZGVeN8vvv_sincos_skx)
-
- .section .rodata, "a"
-.L_2il0floatpacket.15:
- .long 0xffffffff,0xffffffff
- .type .L_2il0floatpacket.15,@object
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core_avx512.S
index f671d60..fe8474f 100644
--- a/sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core_avx512.S
+++ b/sysdeps/x86_64/fpu/multiarch/svml_s_cosf16_core_avx512.S
@@ -278,7 +278,7 @@ WRAPPER_IMPL_AVX512 _ZGVdN8v_cosf
X = X - Y*PI1 - Y*PI2 - Y*PI3
*/
vmovaps %zmm0, %zmm6
- vmovups .L_2il0floatpacket.13(%rip), %zmm12
+ vpternlogd $0xff, %zmm12, %zmm12, %zmm12
vmovups __sRShifter(%rax), %zmm3
vmovups __sPI1_FMA(%rax), %zmm5
vmovups __sA9_FMA(%rax), %zmm9
@@ -453,8 +453,3 @@ WRAPPER_IMPL_AVX512 _ZGVdN8v_cosf
jmp .LBL_2_7
#endif
END (_ZGVeN16v_cosf_skx)
-
- .section .rodata, "a"
-.L_2il0floatpacket.13:
- .long 0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff
- .type .L_2il0floatpacket.13,@object
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_expf16_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_s_expf16_core_avx512.S
index 637bfe3..229b782 100644
--- a/sysdeps/x86_64/fpu/multiarch/svml_s_expf16_core_avx512.S
+++ b/sysdeps/x86_64/fpu/multiarch/svml_s_expf16_core_avx512.S
@@ -264,7 +264,7 @@ WRAPPER_IMPL_AVX512 _ZGVdN8v_expf
vmovaps %zmm0, %zmm7
/* compare against threshold */
- vmovups .L_2il0floatpacket.13(%rip), %zmm3
+ vpternlogd $0xff, %zmm3, %zmm3, %zmm3
vmovups __sInvLn2(%rax), %zmm4
vmovups __sShifter(%rax), %zmm1
vmovups __sLn2hi(%rax), %zmm6
@@ -440,8 +440,3 @@ WRAPPER_IMPL_AVX512 _ZGVdN8v_expf
#endif
END (_ZGVeN16v_expf_skx)
-
- .section .rodata, "a"
-.L_2il0floatpacket.13:
- .long 0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff
- .type .L_2il0floatpacket.13,@object
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_logf16_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_s_logf16_core_avx512.S
index 9d790fb..fa2aae9 100644
--- a/sysdeps/x86_64/fpu/multiarch/svml_s_logf16_core_avx512.S
+++ b/sysdeps/x86_64/fpu/multiarch/svml_s_logf16_core_avx512.S
@@ -235,7 +235,7 @@ WRAPPER_IMPL_AVX512 _ZGVdN8v_logf
andq $-64, %rsp
subq $1280, %rsp
movq __svml_slog_data@GOTPCREL(%rip), %rax
- vmovups .L_2il0floatpacket.7(%rip), %zmm6
+ vpternlogd $0xff, %zmm6, %zmm6, %zmm6
vmovups _iBrkValue(%rax), %zmm4
vmovups _sPoly_7(%rax), %zmm8
@@ -409,8 +409,3 @@ WRAPPER_IMPL_AVX512 _ZGVdN8v_logf
#endif
END (_ZGVeN16v_logf_skx)
-
- .section .rodata, "a"
-.L_2il0floatpacket.7:
- .long 0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff
- .type .L_2il0floatpacket.7,@object
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_powf16_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_s_powf16_core_avx512.S
index c5c43c4..6aea2a4 100644
--- a/sysdeps/x86_64/fpu/multiarch/svml_s_powf16_core_avx512.S
+++ b/sysdeps/x86_64/fpu/multiarch/svml_s_powf16_core_avx512.S
@@ -385,7 +385,7 @@ WRAPPER_IMPL_AVX512_ff _ZGVdN8vv_powf
vpsrlq $32, %zmm3, %zmm2
vpmovqd %zmm2, %ymm11
vcvtps2pd %ymm14, %zmm13
- vmovups .L_2il0floatpacket.23(%rip), %zmm14
+ vpternlogd $0xff, %zmm14, %zmm14, %zmm14
vmovaps %zmm14, %zmm26
vpandd _ABSMASK(%rax), %zmm1, %zmm8
vpcmpd $1, _INF(%rax), %zmm8, %k2
@@ -427,7 +427,7 @@ WRAPPER_IMPL_AVX512_ff _ZGVdN8vv_powf
vpmovqd %zmm11, %ymm5
vpxord %zmm10, %zmm10, %zmm10
vgatherdpd _Log2Rcp_lookup(%rax,%ymm4), %zmm10{%k3}
- vpbroadcastq .L_2il0floatpacket.24(%rip), %zmm4
+ vpternlogd $0xff, %zmm4, %zmm4, %zmm4
vpxord %zmm11, %zmm11, %zmm11
vcvtdq2pd %ymm7, %zmm7
vgatherdpd _Log2Rcp_lookup(%rax,%ymm5), %zmm11{%k1}
@@ -643,11 +643,3 @@ WRAPPER_IMPL_AVX512_ff _ZGVdN8vv_powf
jmp .LBL_2_7
#endif
END (_ZGVeN16vv_powf_skx)
-
- .section .rodata, "a"
-.L_2il0floatpacket.23:
- .long 0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff
- .type .L_2il0floatpacket.23,@object
-.L_2il0floatpacket.24:
- .long 0xffffffff,0xffffffff
- .type .L_2il0floatpacket.24,@object
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_sincosf16_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_s_sincosf16_core_avx512.S
index 9cf359c..a446c50 100644
--- a/sysdeps/x86_64/fpu/multiarch/svml_s_sincosf16_core_avx512.S
+++ b/sysdeps/x86_64/fpu/multiarch/svml_s_sincosf16_core_avx512.S
@@ -317,7 +317,7 @@ WRAPPER_IMPL_AVX512_fFF _ZGVdN8vvv_sincosf
/* Result sign calculations */
vpternlogd $150, %zmm0, %zmm14, %zmm1
- vmovups .L_2il0floatpacket.13(%rip), %zmm14
+ vpternlogd $0xff, %zmm14, %zmm14, %zmm14
/* Add correction term 0.5 for cos() part */
vaddps %zmm8, %zmm5, %zmm15
@@ -748,8 +748,3 @@ END (_ZGVeN16vvv_sincosf_knl)
ENTRY (_ZGVeN16vvv_sincosf_skx)
WRAPPER_AVX512_vvv_vl4l4 _ZGVeN16vl4l4_sincosf_skx
END (_ZGVeN16vvv_sincosf_skx)
-
- .section .rodata, "a"
-.L_2il0floatpacket.13:
- .long 0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff
- .type .L_2il0floatpacket.13,@object
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_sinf16_core_avx512.S b/sysdeps/x86_64/fpu/multiarch/svml_s_sinf16_core_avx512.S
index bd05109..c1b352d 100644
--- a/sysdeps/x86_64/fpu/multiarch/svml_s_sinf16_core_avx512.S
+++ b/sysdeps/x86_64/fpu/multiarch/svml_s_sinf16_core_avx512.S
@@ -280,7 +280,7 @@ WRAPPER_IMPL_AVX512 _ZGVdN8v_sinf
movq __svml_s_trig_data@GOTPCREL(%rip), %rax
/* Check for large and special values */
- vmovups .L_2il0floatpacket.11(%rip), %zmm14
+ vpternlogd $0xff, %zmm14, %zmm14, %zmm14
vmovups __sAbsMask(%rax), %zmm5
vmovups __sInvPI(%rax), %zmm1
vmovups __sRShifter(%rax), %zmm2
@@ -472,8 +472,3 @@ WRAPPER_IMPL_AVX512 _ZGVdN8v_sinf
jmp .LBL_2_7
#endif
END (_ZGVeN16v_sinf_skx)
-
- .section .rodata, "a"
-.L_2il0floatpacket.11:
- .long 0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff,0xffffffff
- .type .L_2il0floatpacket.11,@object
--
1.8.3.1
|
