mirror of git://sourceware.org/git/glibc.git
x86_64: Fix svml_d_tanh8_core_avx512.S code formatting
This commit contains following formatting changes 1. Instructions proceeded by a tab. 2. Instruction less than 8 characters in length have a tab between it and the first operand. 3. Instruction greater than 7 characters in length have a space between it and the first operand. 4. Tabs after `#define`d names and their value. 5. 8 space at the beginning of line replaced by tab. 6. Indent comments with code. 7. Remove redundent .text section. 8. 1 space between line content and line comment. 9. Space after all commas. Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
This commit is contained in:
Sunil K Pandey
parent
28ba5ee77f
commit
6de743a4e3
|
|
@ -28,7 +28,7 @@
|
||||||
* and to approximate tanh(.) with a polynomial on each of them.
|
* and to approximate tanh(.) with a polynomial on each of them.
|
||||||
*
|
*
|
||||||
* IEEE SPECIAL CONDITIONS:
|
* IEEE SPECIAL CONDITIONS:
|
||||||
* x = [+,-]0, r = [+,-]0
|
* x = [+, -]0, r = [+, -]0
|
||||||
* x = +Inf, r = +1
|
* x = +Inf, r = +1
|
||||||
* x = -Inf, r = -1
|
* x = -Inf, r = -1
|
||||||
* x = QNaN, r = QNaN
|
* x = QNaN, r = QNaN
|
||||||
|
|
@ -72,401 +72,399 @@
|
||||||
|
|
||||||
/* Offsets for data table __svml_dtanh_data_internal
|
/* Offsets for data table __svml_dtanh_data_internal
|
||||||
*/
|
*/
|
||||||
#define _dC 0
|
#define _dC 0
|
||||||
#define _dP0 128
|
#define _dP0 128
|
||||||
#define _dP1 256
|
#define _dP1 256
|
||||||
#define _dP2 384
|
#define _dP2 384
|
||||||
#define _dP3 512
|
#define _dP3 512
|
||||||
#define _dP4 640
|
#define _dP4 640
|
||||||
#define _dP5 768
|
#define _dP5 768
|
||||||
#define _dP6 896
|
#define _dP6 896
|
||||||
#define _dP7 1024
|
#define _dP7 1024
|
||||||
#define _dP8 1152
|
#define _dP8 1152
|
||||||
#define _dP9 1280
|
#define _dP9 1280
|
||||||
#define _dP10 1408
|
#define _dP10 1408
|
||||||
#define _dP11 1536
|
#define _dP11 1536
|
||||||
#define _dP12 1664
|
#define _dP12 1664
|
||||||
#define _dP13 1792
|
#define _dP13 1792
|
||||||
#define _dP14 1920
|
#define _dP14 1920
|
||||||
#define _dP15 2048
|
#define _dP15 2048
|
||||||
#define _dP16 2176
|
#define _dP16 2176
|
||||||
#define _dP17 2304
|
#define _dP17 2304
|
||||||
#define _iExpMantMask_UISA 2432
|
#define _iExpMantMask_UISA 2432
|
||||||
#define _iMinIdxOfsMask_UISA 2496
|
#define _iMinIdxOfsMask_UISA 2496
|
||||||
#define _iMaxIdxMask_UISA 2560
|
#define _iMaxIdxMask_UISA 2560
|
||||||
#define _dbSignMask 2624
|
#define _dbSignMask 2624
|
||||||
#define _dbAbsMask 2688
|
#define _dbAbsMask 2688
|
||||||
#define _iExpMantMask 2752
|
#define _iExpMantMask 2752
|
||||||
#define _iExpMask 2816
|
#define _iExpMask 2816
|
||||||
#define _iMinIdxOfsMask 2880
|
#define _iMinIdxOfsMask 2880
|
||||||
#define _iMaxIdxMask 2944
|
#define _iMaxIdxMask 2944
|
||||||
|
|
||||||
#include <sysdep.h>
|
#include <sysdep.h>
|
||||||
|
|
||||||
.text
|
.section .text.evex512, "ax", @progbits
|
||||||
.section .text.evex512,"ax",@progbits
|
|
||||||
ENTRY(_ZGVeN8v_tanh_skx)
|
ENTRY(_ZGVeN8v_tanh_skx)
|
||||||
pushq %rbp
|
pushq %rbp
|
||||||
cfi_def_cfa_offset(16)
|
cfi_def_cfa_offset(16)
|
||||||
movq %rsp, %rbp
|
movq %rsp, %rbp
|
||||||
cfi_def_cfa(6, 16)
|
cfi_def_cfa(6, 16)
|
||||||
cfi_offset(6, -16)
|
cfi_offset(6, -16)
|
||||||
andq $-64, %rsp
|
andq $-64, %rsp
|
||||||
subq $320, %rsp
|
subq $320, %rsp
|
||||||
vpsrlq $32, %zmm0, %zmm4
|
vpsrlq $32, %zmm0, %zmm4
|
||||||
vmovups %zmm0, (%rsp)
|
vmovups %zmm0, (%rsp)
|
||||||
vmovups __svml_dtanh_data_internal(%rip), %zmm14
|
vmovups __svml_dtanh_data_internal(%rip), %zmm14
|
||||||
vmovups _dP0+__svml_dtanh_data_internal(%rip), %zmm15
|
vmovups _dP0+__svml_dtanh_data_internal(%rip), %zmm15
|
||||||
vpmovqd %zmm4, %ymm5
|
vpmovqd %zmm4, %ymm5
|
||||||
|
|
||||||
/* Constant loading */
|
/* Constant loading */
|
||||||
vandpd _dbAbsMask+__svml_dtanh_data_internal(%rip), %zmm0, %zmm13
|
vandpd _dbAbsMask+__svml_dtanh_data_internal(%rip), %zmm0, %zmm13
|
||||||
vandpd _dbSignMask+__svml_dtanh_data_internal(%rip), %zmm0, %zmm3
|
vandpd _dbSignMask+__svml_dtanh_data_internal(%rip), %zmm0, %zmm3
|
||||||
|
|
||||||
/* Here huge arguments, INF and NaNs are filtered out to callout. */
|
/* Here huge arguments, INF and NaNs are filtered out to callout. */
|
||||||
vpand _iExpMantMask_UISA+__svml_dtanh_data_internal(%rip), %ymm5, %ymm7
|
vpand _iExpMantMask_UISA+__svml_dtanh_data_internal(%rip), %ymm5, %ymm7
|
||||||
vmovups _dP2+__svml_dtanh_data_internal(%rip), %zmm0
|
vmovups _dP2+__svml_dtanh_data_internal(%rip), %zmm0
|
||||||
vmovups _dP16+__svml_dtanh_data_internal(%rip), %zmm4
|
vmovups _dP16+__svml_dtanh_data_internal(%rip), %zmm4
|
||||||
vmovups _dP15+__svml_dtanh_data_internal(%rip), %zmm5
|
vmovups _dP15+__svml_dtanh_data_internal(%rip), %zmm5
|
||||||
vmovups %zmm3, 64(%rsp)
|
vmovups %zmm3, 64(%rsp)
|
||||||
vmovups _dP3+__svml_dtanh_data_internal(%rip), %zmm3
|
vmovups _dP3+__svml_dtanh_data_internal(%rip), %zmm3
|
||||||
vpsubd _iMinIdxOfsMask_UISA+__svml_dtanh_data_internal(%rip), %ymm7, %ymm8
|
vpsubd _iMinIdxOfsMask_UISA+__svml_dtanh_data_internal(%rip), %ymm7, %ymm8
|
||||||
|
|
||||||
/* if VMIN, VMAX is defined for I type */
|
/* if VMIN, VMAX is defined for I type */
|
||||||
vxorps %ymm9, %ymm9, %ymm9
|
vxorps %ymm9, %ymm9, %ymm9
|
||||||
vpmaxsd %ymm9, %ymm8, %ymm10
|
vpmaxsd %ymm9, %ymm8, %ymm10
|
||||||
vpminsd _iMaxIdxMask_UISA+__svml_dtanh_data_internal(%rip), %ymm10, %ymm11
|
vpminsd _iMaxIdxMask_UISA+__svml_dtanh_data_internal(%rip), %ymm10, %ymm11
|
||||||
vpsrld $19, %ymm11, %ymm12
|
vpsrld $19, %ymm11, %ymm12
|
||||||
vmovups _dP12+__svml_dtanh_data_internal(%rip), %zmm8
|
vmovups _dP12+__svml_dtanh_data_internal(%rip), %zmm8
|
||||||
vmovups _dP11+__svml_dtanh_data_internal(%rip), %zmm9
|
vmovups _dP11+__svml_dtanh_data_internal(%rip), %zmm9
|
||||||
vmovups _dP10+__svml_dtanh_data_internal(%rip), %zmm10
|
vmovups _dP10+__svml_dtanh_data_internal(%rip), %zmm10
|
||||||
vmovups _dP9+__svml_dtanh_data_internal(%rip), %zmm11
|
vmovups _dP9+__svml_dtanh_data_internal(%rip), %zmm11
|
||||||
vpmovzxdq %ymm12, %zmm2
|
vpmovzxdq %ymm12, %zmm2
|
||||||
vmovups _dP8+__svml_dtanh_data_internal(%rip), %zmm12
|
vmovups _dP8+__svml_dtanh_data_internal(%rip), %zmm12
|
||||||
vpermt2pd _dP2+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm0
|
vpermt2pd _dP2+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm0
|
||||||
vpermt2pd _dC+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm14
|
vpermt2pd _dC+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm14
|
||||||
vpermt2pd _dP16+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm4
|
vpermt2pd _dP16+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm4
|
||||||
vpermt2pd _dP15+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm5
|
vpermt2pd _dP15+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm5
|
||||||
vsubpd {rn-sae}, %zmm14, %zmm13, %zmm1
|
vsubpd {rn-sae}, %zmm14, %zmm13, %zmm1
|
||||||
vpermt2pd _dP12+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm8
|
vpermt2pd _dP12+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm8
|
||||||
vpermt2pd _dP11+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm9
|
vpermt2pd _dP11+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm9
|
||||||
vpermt2pd _dP10+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm10
|
vpermt2pd _dP10+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm10
|
||||||
vpermt2pd _dP9+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm11
|
vpermt2pd _dP9+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm11
|
||||||
vpermt2pd _dP8+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm12
|
vpermt2pd _dP8+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm12
|
||||||
vpermt2pd _dP3+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm3
|
vpermt2pd _dP3+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm3
|
||||||
vpermt2pd _dP0+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm15
|
vpermt2pd _dP0+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm15
|
||||||
vmovups %zmm0, 192(%rsp)
|
vmovups %zmm0, 192(%rsp)
|
||||||
vmovups _dP17+__svml_dtanh_data_internal(%rip), %zmm0
|
vmovups _dP17+__svml_dtanh_data_internal(%rip), %zmm0
|
||||||
vmovups _dP7+__svml_dtanh_data_internal(%rip), %zmm13
|
vmovups _dP7+__svml_dtanh_data_internal(%rip), %zmm13
|
||||||
vmovups _dP6+__svml_dtanh_data_internal(%rip), %zmm14
|
vmovups _dP6+__svml_dtanh_data_internal(%rip), %zmm14
|
||||||
vmovups %zmm3, 256(%rsp)
|
vmovups %zmm3, 256(%rsp)
|
||||||
vmovups _dP5+__svml_dtanh_data_internal(%rip), %zmm3
|
vmovups _dP5+__svml_dtanh_data_internal(%rip), %zmm3
|
||||||
vmovups %zmm15, 128(%rsp)
|
vmovups %zmm15, 128(%rsp)
|
||||||
vmovups _dP4+__svml_dtanh_data_internal(%rip), %zmm15
|
vmovups _dP4+__svml_dtanh_data_internal(%rip), %zmm15
|
||||||
vpermt2pd _dP17+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm0
|
vpermt2pd _dP17+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm0
|
||||||
vpermt2pd _dP7+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm13
|
vpermt2pd _dP7+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm13
|
||||||
vpermt2pd _dP6+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm14
|
vpermt2pd _dP6+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm14
|
||||||
vpermt2pd _dP5+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm3
|
vpermt2pd _dP5+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm3
|
||||||
vpermt2pd _dP4+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm15
|
vpermt2pd _dP4+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm15
|
||||||
vfmadd213pd {rn-sae}, %zmm4, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm4, %zmm1, %zmm0
|
||||||
vpcmpgtd _iExpMask+__svml_dtanh_data_internal(%rip), %ymm7, %ymm6
|
vpcmpgtd _iExpMask+__svml_dtanh_data_internal(%rip), %ymm7, %ymm6
|
||||||
vmovmskps %ymm6, %edx
|
vmovmskps %ymm6, %edx
|
||||||
vmovups _dP14+__svml_dtanh_data_internal(%rip), %zmm6
|
vmovups _dP14+__svml_dtanh_data_internal(%rip), %zmm6
|
||||||
vfmadd213pd {rn-sae}, %zmm5, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm5, %zmm1, %zmm0
|
||||||
vmovups _dP13+__svml_dtanh_data_internal(%rip), %zmm7
|
vmovups _dP13+__svml_dtanh_data_internal(%rip), %zmm7
|
||||||
vpermt2pd _dP14+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm6
|
vpermt2pd _dP14+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm6
|
||||||
vpermt2pd _dP13+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm7
|
vpermt2pd _dP13+64+__svml_dtanh_data_internal(%rip), %zmm2, %zmm7
|
||||||
vfmadd213pd {rn-sae}, %zmm6, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm6, %zmm1, %zmm0
|
||||||
vmovups 256(%rsp), %zmm2
|
vmovups 256(%rsp), %zmm2
|
||||||
vfmadd213pd {rn-sae}, %zmm7, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm7, %zmm1, %zmm0
|
||||||
vfmadd213pd {rn-sae}, %zmm8, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm8, %zmm1, %zmm0
|
||||||
vfmadd213pd {rn-sae}, %zmm9, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm9, %zmm1, %zmm0
|
||||||
vfmadd213pd {rn-sae}, %zmm10, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm10, %zmm1, %zmm0
|
||||||
vfmadd213pd {rn-sae}, %zmm11, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm11, %zmm1, %zmm0
|
||||||
vfmadd213pd {rn-sae}, %zmm12, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm12, %zmm1, %zmm0
|
||||||
vfmadd213pd {rn-sae}, %zmm13, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm13, %zmm1, %zmm0
|
||||||
vfmadd213pd {rn-sae}, %zmm14, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm14, %zmm1, %zmm0
|
||||||
vfmadd213pd {rn-sae}, %zmm3, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm3, %zmm1, %zmm0
|
||||||
vmovups 128(%rsp), %zmm3
|
vmovups 128(%rsp), %zmm3
|
||||||
vfmadd213pd {rn-sae}, %zmm15, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm15, %zmm1, %zmm0
|
||||||
vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
|
||||||
vmovups 192(%rsp), %zmm2
|
vmovups 192(%rsp), %zmm2
|
||||||
vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm2, %zmm1, %zmm0
|
||||||
vfmadd213pd {rn-sae}, %zmm3, %zmm1, %zmm0
|
vfmadd213pd {rn-sae}, %zmm3, %zmm1, %zmm0
|
||||||
vorpd 64(%rsp), %zmm0, %zmm0
|
vorpd 64(%rsp), %zmm0, %zmm0
|
||||||
testl %edx, %edx
|
testl %edx, %edx
|
||||||
|
|
||||||
/* Go to special inputs processing branch */
|
/* Go to special inputs processing branch */
|
||||||
jne L(SPECIAL_VALUES_BRANCH)
|
jne L(SPECIAL_VALUES_BRANCH)
|
||||||
# LOE rbx r12 r13 r14 r15 edx zmm0
|
# LOE rbx r12 r13 r14 r15 edx zmm0
|
||||||
|
|
||||||
/* Restore registers
|
/* Restore registers
|
||||||
* and exit the function
|
* and exit the function
|
||||||
*/
|
*/
|
||||||
|
|
||||||
L(EXIT):
|
L(EXIT):
|
||||||
movq %rbp, %rsp
|
movq %rbp, %rsp
|
||||||
popq %rbp
|
popq %rbp
|
||||||
cfi_def_cfa(7, 8)
|
cfi_def_cfa(7, 8)
|
||||||
cfi_restore(6)
|
cfi_restore(6)
|
||||||
ret
|
ret
|
||||||
cfi_def_cfa(6, 16)
|
cfi_def_cfa(6, 16)
|
||||||
cfi_offset(6, -16)
|
cfi_offset(6, -16)
|
||||||
|
|
||||||
/* Branch to process
|
/* Branch to process
|
||||||
* special inputs
|
* special inputs
|
||||||
*/
|
*/
|
||||||
|
|
||||||
L(SPECIAL_VALUES_BRANCH):
|
L(SPECIAL_VALUES_BRANCH):
|
||||||
vmovups (%rsp), %zmm1
|
vmovups (%rsp), %zmm1
|
||||||
vmovups %zmm0, 128(%rsp)
|
vmovups %zmm0, 128(%rsp)
|
||||||
vmovups %zmm1, 64(%rsp)
|
vmovups %zmm1, 64(%rsp)
|
||||||
# LOE rbx r12 r13 r14 r15 edx zmm0
|
# LOE rbx r12 r13 r14 r15 edx zmm0
|
||||||
|
|
||||||
xorl %eax, %eax
|
xorl %eax, %eax
|
||||||
# LOE rbx r12 r13 r14 r15 eax edx
|
# LOE rbx r12 r13 r14 r15 eax edx
|
||||||
|
|
||||||
vzeroupper
|
vzeroupper
|
||||||
movq %r12, 16(%rsp)
|
movq %r12, 16(%rsp)
|
||||||
/* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -304; DW_OP_plus) */
|
/* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -304; DW_OP_plus) */
|
||||||
.cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xd0, 0xfe, 0xff, 0xff, 0x22
|
.cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xd0, 0xfe, 0xff, 0xff, 0x22
|
||||||
movl %eax, %r12d
|
movl %eax, %r12d
|
||||||
movq %r13, 8(%rsp)
|
movq %r13, 8(%rsp)
|
||||||
/* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -312; DW_OP_plus) */
|
/* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -312; DW_OP_plus) */
|
||||||
.cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xc8, 0xfe, 0xff, 0xff, 0x22
|
.cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xc8, 0xfe, 0xff, 0xff, 0x22
|
||||||
movl %edx, %r13d
|
movl %edx, %r13d
|
||||||
movq %r14, (%rsp)
|
movq %r14, (%rsp)
|
||||||
/* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -320; DW_OP_plus) */
|
/* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -320; DW_OP_plus) */
|
||||||
.cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xc0, 0xfe, 0xff, 0xff, 0x22
|
.cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xc0, 0xfe, 0xff, 0xff, 0x22
|
||||||
# LOE rbx r15 r12d r13d
|
# LOE rbx r15 r12d r13d
|
||||||
|
|
||||||
/* Range mask
|
/* Range mask
|
||||||
* bits check
|
* bits check
|
||||||
*/
|
*/
|
||||||
|
|
||||||
L(RANGEMASK_CHECK):
|
L(RANGEMASK_CHECK):
|
||||||
btl %r12d, %r13d
|
btl %r12d, %r13d
|
||||||
|
|
||||||
/* Call scalar math function */
|
/* Call scalar math function */
|
||||||
jc L(SCALAR_MATH_CALL)
|
jc L(SCALAR_MATH_CALL)
|
||||||
# LOE rbx r15 r12d r13d
|
# LOE rbx r15 r12d r13d
|
||||||
|
|
||||||
/* Special inputs
|
/* Special inputs
|
||||||
* processing loop
|
* processing loop
|
||||||
*/
|
*/
|
||||||
|
|
||||||
L(SPECIAL_VALUES_LOOP):
|
L(SPECIAL_VALUES_LOOP):
|
||||||
incl %r12d
|
incl %r12d
|
||||||
cmpl $8, %r12d
|
cmpl $8, %r12d
|
||||||
|
|
||||||
/* Check bits in range mask */
|
/* Check bits in range mask */
|
||||||
jl L(RANGEMASK_CHECK)
|
jl L(RANGEMASK_CHECK)
|
||||||
# LOE rbx r15 r12d r13d
|
# LOE rbx r15 r12d r13d
|
||||||
|
|
||||||
movq 16(%rsp), %r12
|
movq 16(%rsp), %r12
|
||||||
cfi_restore(12)
|
cfi_restore(12)
|
||||||
movq 8(%rsp), %r13
|
movq 8(%rsp), %r13
|
||||||
cfi_restore(13)
|
cfi_restore(13)
|
||||||
movq (%rsp), %r14
|
movq (%rsp), %r14
|
||||||
cfi_restore(14)
|
cfi_restore(14)
|
||||||
vmovups 128(%rsp), %zmm0
|
vmovups 128(%rsp), %zmm0
|
||||||
|
|
||||||
/* Go to exit */
|
/* Go to exit */
|
||||||
jmp L(EXIT)
|
jmp L(EXIT)
|
||||||
/* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -304; DW_OP_plus) */
|
/* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -304; DW_OP_plus) */
|
||||||
.cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xd0, 0xfe, 0xff, 0xff, 0x22
|
.cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xd0, 0xfe, 0xff, 0xff, 0x22
|
||||||
/* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -312; DW_OP_plus) */
|
/* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -312; DW_OP_plus) */
|
||||||
.cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xc8, 0xfe, 0xff, 0xff, 0x22
|
.cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xc8, 0xfe, 0xff, 0xff, 0x22
|
||||||
/* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -320; DW_OP_plus) */
|
/* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -64; DW_OP_and; DW_OP_const4s: -320; DW_OP_plus) */
|
||||||
.cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xc0, 0xfe, 0xff, 0xff, 0x22
|
.cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xc0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xc0, 0xfe, 0xff, 0xff, 0x22
|
||||||
# LOE rbx r12 r13 r14 r15 zmm0
|
# LOE rbx r12 r13 r14 r15 zmm0
|
||||||
|
|
||||||
/* Scalar math fucntion call
|
/* Scalar math fucntion call
|
||||||
* to process special input
|
* to process special input
|
||||||
*/
|
*/
|
||||||
|
|
||||||
L(SCALAR_MATH_CALL):
|
L(SCALAR_MATH_CALL):
|
||||||
movl %r12d, %r14d
|
movl %r12d, %r14d
|
||||||
movsd 64(%rsp,%r14,8), %xmm0
|
movsd 64(%rsp, %r14, 8), %xmm0
|
||||||
call tanh@PLT
|
call tanh@PLT
|
||||||
# LOE rbx r14 r15 r12d r13d xmm0
|
# LOE rbx r14 r15 r12d r13d xmm0
|
||||||
|
|
||||||
movsd %xmm0, 128(%rsp,%r14,8)
|
movsd %xmm0, 128(%rsp, %r14, 8)
|
||||||
|
|
||||||
/* Process special inputs in loop */
|
/* Process special inputs in loop */
|
||||||
jmp L(SPECIAL_VALUES_LOOP)
|
jmp L(SPECIAL_VALUES_LOOP)
|
||||||
# LOE rbx r15 r12d r13d
|
# LOE rbx r15 r12d r13d
|
||||||
END(_ZGVeN8v_tanh_skx)
|
END(_ZGVeN8v_tanh_skx)
|
||||||
|
|
||||||
.section .rodata, "a"
|
.section .rodata, "a"
|
||||||
.align 64
|
.align 64
|
||||||
|
|
||||||
#ifdef __svml_dtanh_data_internal_typedef
|
#ifdef __svml_dtanh_data_internal_typedef
|
||||||
typedef unsigned int VUINT32;
|
typedef unsigned int VUINT32;
|
||||||
typedef struct
|
typedef struct {
|
||||||
{
|
__declspec(align(64)) VUINT32 _dC[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dC[16][2];
|
__declspec(align(64)) VUINT32 _dP0[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP0[16][2];
|
__declspec(align(64)) VUINT32 _dP1[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP1[16][2];
|
__declspec(align(64)) VUINT32 _dP2[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP2[16][2];
|
__declspec(align(64)) VUINT32 _dP3[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP3[16][2];
|
__declspec(align(64)) VUINT32 _dP4[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP4[16][2];
|
__declspec(align(64)) VUINT32 _dP5[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP5[16][2];
|
__declspec(align(64)) VUINT32 _dP6[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP6[16][2];
|
__declspec(align(64)) VUINT32 _dP7[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP7[16][2];
|
__declspec(align(64)) VUINT32 _dP8[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP8[16][2];
|
__declspec(align(64)) VUINT32 _dP9[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP9[16][2];
|
__declspec(align(64)) VUINT32 _dP10[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP10[16][2];
|
__declspec(align(64)) VUINT32 _dP11[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP11[16][2];
|
__declspec(align(64)) VUINT32 _dP12[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP12[16][2];
|
__declspec(align(64)) VUINT32 _dP13[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP13[16][2];
|
__declspec(align(64)) VUINT32 _dP14[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP14[16][2];
|
__declspec(align(64)) VUINT32 _dP15[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP15[16][2];
|
__declspec(align(64)) VUINT32 _dP16[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP16[16][2];
|
__declspec(align(64)) VUINT32 _dP17[16][2];
|
||||||
__declspec(align(64)) VUINT32 _dP17[16][2];
|
__declspec(align(64)) VUINT32 _iExpMantMask_UISA[16][1];
|
||||||
__declspec(align(64)) VUINT32 _iExpMantMask_UISA[16][1];
|
__declspec(align(64)) VUINT32 _iMinIdxOfsMask_UISA[16][1];
|
||||||
__declspec(align(64)) VUINT32 _iMinIdxOfsMask_UISA[16][1];
|
__declspec(align(64)) VUINT32 _iMaxIdxMask_UISA[16][1];
|
||||||
__declspec(align(64)) VUINT32 _iMaxIdxMask_UISA[16][1];
|
__declspec(align(64)) VUINT32 _dbSignMask[8][2];
|
||||||
__declspec(align(64)) VUINT32 _dbSignMask[8][2];
|
__declspec(align(64)) VUINT32 _dbAbsMask[8][2];
|
||||||
__declspec(align(64)) VUINT32 _dbAbsMask[8][2];
|
__declspec(align(64)) VUINT32 _iExpMantMask[16][1];
|
||||||
__declspec(align(64)) VUINT32 _iExpMantMask[16][1];
|
__declspec(align(64)) VUINT32 _iExpMask[16][1];
|
||||||
__declspec(align(64)) VUINT32 _iExpMask[16][1];
|
__declspec(align(64)) VUINT32 _iMinIdxOfsMask[16][1];
|
||||||
__declspec(align(64)) VUINT32 _iMinIdxOfsMask[16][1];
|
__declspec(align(64)) VUINT32 _iMaxIdxMask[16][1];
|
||||||
__declspec(align(64)) VUINT32 _iMaxIdxMask[16][1];
|
|
||||||
} __svml_dtanh_data_internal;
|
} __svml_dtanh_data_internal;
|
||||||
#endif
|
#endif
|
||||||
__svml_dtanh_data_internal:
|
__svml_dtanh_data_internal:
|
||||||
/*== _dC ==*/
|
/* _dC */
|
||||||
.quad 0x0000000000000000, 0x3fcc000000000000, 0x3fd4000000000000, 0x3fdc000000000000
|
.quad 0x0000000000000000, 0x3fcc000000000000, 0x3fd4000000000000, 0x3fdc000000000000
|
||||||
.quad 0x3fe4000000000000, 0x3fec000000000000, 0x3ff4000000000000, 0x3ffc000000000000
|
.quad 0x3fe4000000000000, 0x3fec000000000000, 0x3ff4000000000000, 0x3ffc000000000000
|
||||||
.quad 0x4004000000000000, 0x400c000000000000, 0x4014000000000000, 0x401c000000000000
|
.quad 0x4004000000000000, 0x400c000000000000, 0x4014000000000000, 0x401c000000000000
|
||||||
.quad 0x4024000000000000, 0x402c000000000000, 0x4034000000000000, 0x0000000000000000
|
.quad 0x4024000000000000, 0x402c000000000000, 0x4034000000000000, 0x0000000000000000
|
||||||
/*== p0 ==*/
|
/* p0 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0x0000000000000000, 0x3fcb8fd0416a7c92, 0x3fd35f98a0ea650e, 0x3fda5729ee488037
|
.quad 0x0000000000000000, 0x3fcb8fd0416a7c92, 0x3fd35f98a0ea650e, 0x3fda5729ee488037
|
||||||
.quad 0x3fe1bf47eabb8f95, 0x3fe686650b8c2015, 0x3feb2523bb6b2dee, 0x3fee1fbf97e33527
|
.quad 0x3fe1bf47eabb8f95, 0x3fe686650b8c2015, 0x3feb2523bb6b2dee, 0x3fee1fbf97e33527
|
||||||
.quad 0x3fef9258260a71c2, 0x3feff112c63a9077, 0x3fefff419668df11, 0x3feffffc832750f2
|
.quad 0x3fef9258260a71c2, 0x3feff112c63a9077, 0x3fefff419668df11, 0x3feffffc832750f2
|
||||||
.quad 0x3feffffffdc96f35, 0x3fefffffffffcf58, 0x3ff0000000000000, 0x3ff0000000000000
|
.quad 0x3feffffffdc96f35, 0x3fefffffffffcf58, 0x3ff0000000000000, 0x3ff0000000000000
|
||||||
/*== p1 ==*/
|
/* p1 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0x0000000000000000, 0x3c65e23ebcd3bcbe, 0xbc4c600bac3adf00, 0x3c6c44091785d040
|
.quad 0x0000000000000000, 0x3c65e23ebcd3bcbe, 0xbc4c600bac3adf00, 0x3c6c44091785d040
|
||||||
.quad 0x3c8221d7a6e3674b, 0x3c69f89d2cf6b85c, 0x3c73b3e9ec0b8f1c, 0xbc7f8d4b0428aada
|
.quad 0x3c8221d7a6e3674b, 0x3c69f89d2cf6b85c, 0x3c73b3e9ec0b8f1c, 0xbc7f8d4b0428aada
|
||||||
.quad 0xbc7c52d880cf43c0, 0x3c7dd36e37096480, 0x3c7b4f6380c442ca, 0xbc729755de470096
|
.quad 0xbc7c52d880cf43c0, 0x3c7dd36e37096480, 0x3c7b4f6380c442ca, 0xbc729755de470096
|
||||||
.quad 0x3c84cf852845efbd, 0x3c6fc4fb440a5378, 0xbc63981083b55870, 0x0000000000000000
|
.quad 0x3c84cf852845efbd, 0x3c6fc4fb440a5378, 0xbc63981083b55870, 0x0000000000000000
|
||||||
/*== p2 ==*/
|
/* p2 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0x3ff0000000000000, 0x3fee842ca3f08532, 0x3fed11574af58f1b, 0x3fea945b9c24e4f9
|
.quad 0x3ff0000000000000, 0x3fee842ca3f08532, 0x3fed11574af58f1b, 0x3fea945b9c24e4f9
|
||||||
.quad 0x3fe6284c3374f815, 0x3fe02500a09f8d6e, 0x3fd1f25131e3a8c0, 0x3fbd22ca1c24a139
|
.quad 0x3fe6284c3374f815, 0x3fe02500a09f8d6e, 0x3fd1f25131e3a8c0, 0x3fbd22ca1c24a139
|
||||||
.quad 0x3f9b3afe1fba5c76, 0x3f6dd37d19b22b21, 0x3f27ccec13a9ef96, 0x3ecbe6c3f33250ae
|
.quad 0x3f9b3afe1fba5c76, 0x3f6dd37d19b22b21, 0x3f27ccec13a9ef96, 0x3ecbe6c3f33250ae
|
||||||
.quad 0x3e41b4865394f75f, 0x3d8853f01bda5f28, 0x3c73953c0197ef58, 0x0000000000000000
|
.quad 0x3e41b4865394f75f, 0x3d8853f01bda5f28, 0x3c73953c0197ef58, 0x0000000000000000
|
||||||
/*== p3 ==*/
|
/* p3 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0xbbf0b3ea3fdfaa19, 0xbfca48aaeb53bc21, 0xbfd19921f4329916, 0xbfd5e0f09bef8011
|
.quad 0xbbf0b3ea3fdfaa19, 0xbfca48aaeb53bc21, 0xbfd19921f4329916, 0xbfd5e0f09bef8011
|
||||||
.quad 0xbfd893b59c35c882, 0xbfd6ba7cb7576538, 0xbfce7291743d7555, 0xbfbb6d85a01efb80
|
.quad 0xbfd893b59c35c882, 0xbfd6ba7cb7576538, 0xbfce7291743d7555, 0xbfbb6d85a01efb80
|
||||||
.quad 0xbf9addae58c7141a, 0xbf6dc59376c7aa19, 0xbf27cc5e74677410, 0xbecbe6c0e8b4cc87
|
.quad 0xbf9addae58c7141a, 0xbf6dc59376c7aa19, 0xbf27cc5e74677410, 0xbecbe6c0e8b4cc87
|
||||||
.quad 0xbe41b486526b0565, 0xbd8853f01bef63a4, 0xbc73955be519be31, 0x0000000000000000
|
.quad 0xbe41b486526b0565, 0xbd8853f01bef63a4, 0xbc73955be519be31, 0x0000000000000000
|
||||||
/*== p4 ==*/
|
/* p4 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0xbfd5555555555555, 0xbfd183afc292ba11, 0xbfcc1a4b039c9bfa, 0xbfc16e1e6d8d0be6
|
.quad 0xbfd5555555555555, 0xbfd183afc292ba11, 0xbfcc1a4b039c9bfa, 0xbfc16e1e6d8d0be6
|
||||||
.quad 0xbf92426c751e48a2, 0x3fb4f152b2bad124, 0x3fbbba40cbef72be, 0x3fb01ba038be6a3d
|
.quad 0xbf92426c751e48a2, 0x3fb4f152b2bad124, 0x3fbbba40cbef72be, 0x3fb01ba038be6a3d
|
||||||
.quad 0x3f916df44871efc8, 0x3f63c6869dfc8870, 0x3f1fb9aef915d828, 0x3ec299d1e27c6e11
|
.quad 0x3f916df44871efc8, 0x3f63c6869dfc8870, 0x3f1fb9aef915d828, 0x3ec299d1e27c6e11
|
||||||
.quad 0x3e379b5ddcca334c, 0x3d8037f57bc62c9a, 0x3c6a2d4b50a2cff7, 0x0000000000000000
|
.quad 0x3e379b5ddcca334c, 0x3d8037f57bc62c9a, 0x3c6a2d4b50a2cff7, 0x0000000000000000
|
||||||
/*== p5 ==*/
|
/* p5 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0xbce6863ee44ed636, 0x3fc04dcd0476c75e, 0x3fc43d3449a80f08, 0x3fc5c26f3699b7e7
|
.quad 0xbce6863ee44ed636, 0x3fc04dcd0476c75e, 0x3fc43d3449a80f08, 0x3fc5c26f3699b7e7
|
||||||
.quad 0x3fc1a686f6ab2533, 0x3faf203c316ce730, 0xbf89c7a02788557c, 0xbf98157e26e0d541
|
.quad 0x3fc1a686f6ab2533, 0x3faf203c316ce730, 0xbf89c7a02788557c, 0xbf98157e26e0d541
|
||||||
.quad 0xbf807b55c1c7d278, 0xbf53a18d5843190f, 0xbf0fb6bbc89b1a5b, 0xbeb299c9c684a963
|
.quad 0xbf807b55c1c7d278, 0xbf53a18d5843190f, 0xbf0fb6bbc89b1a5b, 0xbeb299c9c684a963
|
||||||
.quad 0xbe279b5dd4fb3d01, 0xbd7037f57ae72aa6, 0xbc5a2ca2bba78e86, 0x0000000000000000
|
.quad 0xbe279b5dd4fb3d01, 0xbd7037f57ae72aa6, 0xbc5a2ca2bba78e86, 0x0000000000000000
|
||||||
/*== p6 ==*/
|
/* p6 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0x3fc1111111112ab5, 0x3fb5c19efdfc08ad, 0x3fa74c98dc34fbac, 0xbf790d6a8eff0a77
|
.quad 0x3fc1111111112ab5, 0x3fb5c19efdfc08ad, 0x3fa74c98dc34fbac, 0xbf790d6a8eff0a77
|
||||||
.quad 0xbfac3c021789a786, 0xbfae2196b7326859, 0xbf93a7a011ff8c2a, 0x3f6e4709c7e8430e
|
.quad 0xbfac3c021789a786, 0xbfae2196b7326859, 0xbf93a7a011ff8c2a, 0x3f6e4709c7e8430e
|
||||||
.quad 0x3f67682afa611151, 0x3f3ef2ee77717cbf, 0x3ef95a4482f180b7, 0x3e9dc2c27da3b603
|
.quad 0x3f67682afa611151, 0x3f3ef2ee77717cbf, 0x3ef95a4482f180b7, 0x3e9dc2c27da3b603
|
||||||
.quad 0x3e12e2afd9f7433e, 0x3d59f320348679ba, 0x3c44b61d9bbcc940, 0x0000000000000000
|
.quad 0x3e12e2afd9f7433e, 0x3d59f320348679ba, 0x3c44b61d9bbcc940, 0x0000000000000000
|
||||||
/*== p7 ==*/
|
/* p7 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0xbda1ea19ddddb3b4, 0xbfb0b8df995ce4df, 0xbfb2955cf41e8164, 0xbfaf9d05c309f7c6
|
.quad 0xbda1ea19ddddb3b4, 0xbfb0b8df995ce4df, 0xbfb2955cf41e8164, 0xbfaf9d05c309f7c6
|
||||||
.quad 0xbf987d27ccff4291, 0x3f8b2ca62572b098, 0x3f8f1cf6c7f5b00a, 0x3f60379811e43dd5
|
.quad 0xbf987d27ccff4291, 0x3f8b2ca62572b098, 0x3f8f1cf6c7f5b00a, 0x3f60379811e43dd5
|
||||||
.quad 0xbf4793826f78537e, 0xbf2405695e36240f, 0xbee0e08de39ce756, 0xbe83d709ba5f714e
|
.quad 0xbf4793826f78537e, 0xbf2405695e36240f, 0xbee0e08de39ce756, 0xbe83d709ba5f714e
|
||||||
.quad 0xbdf92e3fc5ee63e0, 0xbd414cc030f2110e, 0xbc2ba022e8d82a87, 0x0000000000000000
|
.quad 0xbdf92e3fc5ee63e0, 0xbd414cc030f2110e, 0xbc2ba022e8d82a87, 0x0000000000000000
|
||||||
/*== p8 ==*/
|
/* p8 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0xbfaba1ba1990520b, 0xbf96e37bba52f6fc, 0x3ecff7df18455399, 0x3f97362834d33a4e
|
.quad 0xbfaba1ba1990520b, 0xbf96e37bba52f6fc, 0x3ecff7df18455399, 0x3f97362834d33a4e
|
||||||
.quad 0x3f9e7f8380184b45, 0x3f869543e7c420d4, 0xbf7326bd4914222a, 0xbf5fc15b0a9d98fa
|
.quad 0x3f9e7f8380184b45, 0x3f869543e7c420d4, 0xbf7326bd4914222a, 0xbf5fc15b0a9d98fa
|
||||||
.quad 0x3f14cffcfa69fbb6, 0x3f057e48e5b79d10, 0x3ec33b66d7d77264, 0x3e66ac4e578b9b10
|
.quad 0x3f14cffcfa69fbb6, 0x3f057e48e5b79d10, 0x3ec33b66d7d77264, 0x3e66ac4e578b9b10
|
||||||
.quad 0x3ddcc74b8d3d5c42, 0x3d23c589137f92b4, 0x3c107f8e2c8707a1, 0x0000000000000000
|
.quad 0x3ddcc74b8d3d5c42, 0x3d23c589137f92b4, 0x3c107f8e2c8707a1, 0x0000000000000000
|
||||||
/*== p9 ==*/
|
/* p9 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0xbe351ca7f096011f, 0x3f9eaaf3320c3851, 0x3f9cf823fe761fc1, 0x3f9022271754ff1f
|
.quad 0xbe351ca7f096011f, 0x3f9eaaf3320c3851, 0x3f9cf823fe761fc1, 0x3f9022271754ff1f
|
||||||
.quad 0xbf731fe77c9c60af, 0xbf84a6046865ec7d, 0xbf4ca3f1f2b9192b, 0x3f4c77dee0afd227
|
.quad 0xbf731fe77c9c60af, 0xbf84a6046865ec7d, 0xbf4ca3f1f2b9192b, 0x3f4c77dee0afd227
|
||||||
.quad 0x3f04055bce68597a, 0xbee2bf0cb4a71647, 0xbea31eaafe73efd5, 0xbe46abb02c4368ed
|
.quad 0x3f04055bce68597a, 0xbee2bf0cb4a71647, 0xbea31eaafe73efd5, 0xbe46abb02c4368ed
|
||||||
.quad 0xbdbcc749ca8079dd, 0xbd03c5883836b9d2, 0xbbf07a5416264aec, 0x0000000000000000
|
.quad 0xbdbcc749ca8079dd, 0xbd03c5883836b9d2, 0xbbf07a5416264aec, 0x0000000000000000
|
||||||
/*== p10 ==*/
|
/* p10 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0x3f9664f94e6ac14e, 0xbf94d3343bae39dd, 0xbf7bc748e60df843, 0xbf8c89372b43ba85
|
.quad 0x3f9664f94e6ac14e, 0xbf94d3343bae39dd, 0xbf7bc748e60df843, 0xbf8c89372b43ba85
|
||||||
.quad 0xbf8129a092de747a, 0x3f60c85b4d538746, 0x3f5be9392199ec18, 0xbf2a0c68a4489f10
|
.quad 0xbf8129a092de747a, 0x3f60c85b4d538746, 0x3f5be9392199ec18, 0xbf2a0c68a4489f10
|
||||||
.quad 0xbf00462601dc2faa, 0x3eb7b6a219dea9f4, 0x3e80cbcc8d4c5c8a, 0x3e2425bb231a5e29
|
.quad 0xbf00462601dc2faa, 0x3eb7b6a219dea9f4, 0x3e80cbcc8d4c5c8a, 0x3e2425bb231a5e29
|
||||||
.quad 0x3d9992a4beac8662, 0x3ce191ba5ed3fb67, 0x3bc892450bad44c4, 0x0000000000000000
|
.quad 0x3d9992a4beac8662, 0x3ce191ba5ed3fb67, 0x3bc892450bad44c4, 0x0000000000000000
|
||||||
/*== p11 ==*/
|
/* p11 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0xbea8c4c1fd7852fe, 0xbfccce16b1046f13, 0xbf81a16f224bb7b6, 0xbf62cbf00406bc09
|
.quad 0xbea8c4c1fd7852fe, 0xbfccce16b1046f13, 0xbf81a16f224bb7b6, 0xbf62cbf00406bc09
|
||||||
.quad 0x3f75b29bb02cf69b, 0x3f607df0f9f90c17, 0xbf4b852a6e0758d5, 0xbf0078c63d1b8445
|
.quad 0x3f75b29bb02cf69b, 0x3f607df0f9f90c17, 0xbf4b852a6e0758d5, 0xbf0078c63d1b8445
|
||||||
.quad 0x3eec12eadd55be7a, 0xbe6fa600f593181b, 0xbe5a3c935dce3f7d, 0xbe001c6d95e3ae96
|
.quad 0x3eec12eadd55be7a, 0xbe6fa600f593181b, 0xbe5a3c935dce3f7d, 0xbe001c6d95e3ae96
|
||||||
.quad 0xbd74755a00ea1fd3, 0xbcbc1c6c063bb7ac, 0xbba3be9a4460fe00, 0x0000000000000000
|
.quad 0xbd74755a00ea1fd3, 0xbcbc1c6c063bb7ac, 0xbba3be9a4460fe00, 0x0000000000000000
|
||||||
/*== p12 ==*/
|
/* p12 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0xbf822404577aa9dd, 0x403d8b07f7a82aa3, 0xbf9f44ab92fbab0a, 0x3fb2eac604473d6a
|
.quad 0xbf822404577aa9dd, 0x403d8b07f7a82aa3, 0xbf9f44ab92fbab0a, 0x3fb2eac604473d6a
|
||||||
.quad 0x3f45f87d903aaac8, 0xbf5e104671036300, 0x3f19bc98ddf0f340, 0x3f0d4304bc9246e8
|
.quad 0x3f45f87d903aaac8, 0xbf5e104671036300, 0x3f19bc98ddf0f340, 0x3f0d4304bc9246e8
|
||||||
.quad 0xbed13c415f7b9d41, 0xbe722b8d9720cdb0, 0x3e322666d739bec0, 0x3dd76a553d7e7918
|
.quad 0xbed13c415f7b9d41, 0xbe722b8d9720cdb0, 0x3e322666d739bec0, 0x3dd76a553d7e7918
|
||||||
.quad 0x3d4de0fa59416a39, 0x3c948716cf3681b4, 0x3b873f9f2d2fda99, 0x0000000000000000
|
.quad 0x3d4de0fa59416a39, 0x3c948716cf3681b4, 0x3b873f9f2d2fda99, 0x0000000000000000
|
||||||
/*== p13 ==*/
|
/* p13 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0xbefdd99a221ed573, 0x4070593a3735bab4, 0xbfccab654e44835e, 0x3fd13ed80037dbac
|
.quad 0xbefdd99a221ed573, 0x4070593a3735bab4, 0xbfccab654e44835e, 0x3fd13ed80037dbac
|
||||||
.quad 0xbf6045b9076cc487, 0x3f2085ee7e8ac170, 0x3f23524622610430, 0xbeff12a6626911b4
|
.quad 0xbf6045b9076cc487, 0x3f2085ee7e8ac170, 0x3f23524622610430, 0xbeff12a6626911b4
|
||||||
.quad 0x3eab9008bca408af, 0x3e634df71865f620, 0xbe05bb1bcf83ca73, 0xbdaf2ac143fb6762
|
.quad 0x3eab9008bca408af, 0x3e634df71865f620, 0xbe05bb1bcf83ca73, 0xbdaf2ac143fb6762
|
||||||
.quad 0xbd23eae52a3dbf57, 0xbc6b5e3e9ca0955e, 0xbb5eca68e2c1ba2e, 0x0000000000000000
|
.quad 0xbd23eae52a3dbf57, 0xbc6b5e3e9ca0955e, 0xbb5eca68e2c1ba2e, 0x0000000000000000
|
||||||
/*== p14 ==*/
|
/* p14 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0x3f6e3be689423841, 0xc0d263511f5baac1, 0x40169f73b15ebe5c, 0xc025c1dd41cd6cb5
|
.quad 0x3f6e3be689423841, 0xc0d263511f5baac1, 0x40169f73b15ebe5c, 0xc025c1dd41cd6cb5
|
||||||
.quad 0xbf58fd89fe05e0d1, 0x3f73f7af01d5af7a, 0xbf1e40bdead17e6b, 0x3ee224cd6c4513e5
|
.quad 0xbf58fd89fe05e0d1, 0x3f73f7af01d5af7a, 0xbf1e40bdead17e6b, 0x3ee224cd6c4513e5
|
||||||
.quad 0xbe24b645e68eeaa3, 0xbe4abfebfb72bc83, 0x3dd51c38f8695ed3, 0x3d8313ac38c6832b
|
.quad 0xbe24b645e68eeaa3, 0xbe4abfebfb72bc83, 0x3dd51c38f8695ed3, 0x3d8313ac38c6832b
|
||||||
.quad 0x3cf7787935626685, 0x3c401ffc49c6bc29, 0xbabf0b21acfa52ab, 0x0000000000000000
|
.quad 0x3cf7787935626685, 0x3c401ffc49c6bc29, 0xbabf0b21acfa52ab, 0x0000000000000000
|
||||||
/*== p15 ==*/
|
/* p15 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0xbf2a1306713a4f3a, 0xc1045e509116b066, 0x4041fab9250984ce, 0xc0458d090ec3de95
|
.quad 0xbf2a1306713a4f3a, 0xc1045e509116b066, 0x4041fab9250984ce, 0xc0458d090ec3de95
|
||||||
.quad 0xbf74949d60113d63, 0x3f7c9fd6200d0ade, 0x3f02cd40e0ad0a9f, 0xbe858ab8e019f311
|
.quad 0xbf74949d60113d63, 0x3f7c9fd6200d0ade, 0x3f02cd40e0ad0a9f, 0xbe858ab8e019f311
|
||||||
.quad 0xbe792fa6323b7cf8, 0x3e2df04d67876402, 0xbd95c72be95e4d2c, 0xbd55a89c30203106
|
.quad 0xbe792fa6323b7cf8, 0x3e2df04d67876402, 0xbd95c72be95e4d2c, 0xbd55a89c30203106
|
||||||
.quad 0xbccad6b3bb9eff65, 0xbc12705ccd3dd884, 0xba8e0a4c47ae75f5, 0x0000000000000000
|
.quad 0xbccad6b3bb9eff65, 0xbc12705ccd3dd884, 0xba8e0a4c47ae75f5, 0x0000000000000000
|
||||||
/*== p16 ==*/
|
/* p16 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0xbf55d7e76dc56871, 0x41528c38809c90c7, 0xc076d57fb5190b02, 0x4085f09f888f8ada
|
.quad 0xbf55d7e76dc56871, 0x41528c38809c90c7, 0xc076d57fb5190b02, 0x4085f09f888f8ada
|
||||||
.quad 0x3fa246332a2fcba5, 0xbfb29d851a896fcd, 0x3ed9065ae369b212, 0xbeb8e1ba4c98a030
|
.quad 0x3fa246332a2fcba5, 0xbfb29d851a896fcd, 0x3ed9065ae369b212, 0xbeb8e1ba4c98a030
|
||||||
.quad 0x3e6ffd0766ad4016, 0xbe0c63c29f505f5b, 0xbd7fab216b9e0e49, 0x3d2826b62056aa27
|
.quad 0x3e6ffd0766ad4016, 0xbe0c63c29f505f5b, 0xbd7fab216b9e0e49, 0x3d2826b62056aa27
|
||||||
.quad 0x3ca313e31762f523, 0x3bea37aa21895319, 0x3ae5c7f1fd871496, 0x0000000000000000
|
.quad 0x3ca313e31762f523, 0x3bea37aa21895319, 0x3ae5c7f1fd871496, 0x0000000000000000
|
||||||
/*== p17 ==*/
|
/* p17 */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0x3f35e67ab76a26e7, 0x41848ee0627d8206, 0xc0a216d618b489ec, 0x40a5b89107c8af4f
|
.quad 0x3f35e67ab76a26e7, 0x41848ee0627d8206, 0xc0a216d618b489ec, 0x40a5b89107c8af4f
|
||||||
.quad 0x3fb69d8374520eda, 0xbfbded519f981716, 0xbef02d288b5b3371, 0x3eb290981209c1a6
|
.quad 0x3fb69d8374520eda, 0xbfbded519f981716, 0xbef02d288b5b3371, 0x3eb290981209c1a6
|
||||||
.quad 0xbe567e924bf5ff6e, 0x3de3f7f7de6b0eb6, 0x3d69ed18bae3ebbc, 0xbcf7534c4f3dfa71
|
.quad 0xbe567e924bf5ff6e, 0x3de3f7f7de6b0eb6, 0x3d69ed18bae3ebbc, 0xbcf7534c4f3dfa71
|
||||||
.quad 0xbc730b73f1eaff20, 0xbbba2cff8135d462, 0xbab5a71b5f7d9035, 0x0000000000000000
|
.quad 0xbc730b73f1eaff20, 0xbbba2cff8135d462, 0xbab5a71b5f7d9035, 0x0000000000000000
|
||||||
.align 64
|
.align 64
|
||||||
.long 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000 /* _iExpMantMask_UISA */
|
.long 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000, 0x7ff80000 /* _iExpMantMask_UISA */
|
||||||
.align 64
|
.align 64
|
||||||
.long 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000 /* _iMinIdxOfsMask_UISA */
|
.long 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000, 0x3fc00000 /* _iMinIdxOfsMask_UISA */
|
||||||
.align 64
|
.align 64
|
||||||
.long 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000 /* _iMaxIdxMask_UISA */
|
.long 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000, 0x00780000 /* _iMaxIdxMask_UISA */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000 /* _dbSignMask */
|
.quad 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000, 0x8000000000000000 /* _dbSignMask */
|
||||||
.align 64
|
.align 64
|
||||||
.quad 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff /* _dbAbsMask */
|
.quad 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff /* _dbAbsMask */
|
||||||
.align 64
|
.align 64
|
||||||
.long 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000 /* _iExpMantMask */
|
.long 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000, 0x7ffe0000 /* _iExpMantMask */
|
||||||
.align 64
|
.align 64
|
||||||
.long 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000 /* _iExpMask */
|
.long 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000, 0x7fe00000 /* _iExpMask */
|
||||||
.align 64
|
.align 64
|
||||||
.long 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000 /* _iMinIdxOfsMask */
|
.long 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000, 0x3fbe0000 /* _iMinIdxOfsMask */
|
||||||
.align 64
|
.align 64
|
||||||
.long 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000 /* _iMaxIdxMask */
|
.long 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000, 0x00760000 /* _iMaxIdxMask */
|
||||||
.align 64
|
.align 64
|
||||||
.type __svml_dtanh_data_internal,@object
|
.type __svml_dtanh_data_internal, @object
|
||||||
.size __svml_dtanh_data_internal,.-__svml_dtanh_data_internal
|
.size __svml_dtanh_data_internal, .-__svml_dtanh_data_internal
|
||||||
|
|
|
||||||
Loading…
Reference in New Issue