mirror of git://sourceware.org/git/glibc.git
x86_64: Fix svml_s_cbrtf8_core_avx2.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
2fb2d57f4d
commit
6075b00249
|
|
@ -31,479 +31,477 @@
|
|||
|
||||
/* Offsets for data table __svml_scbrt_data_internal
|
||||
*/
|
||||
#define _sRcp 0
|
||||
#define _sCbrtHL 128
|
||||
#define _sP2 512
|
||||
#define _sP1 544
|
||||
#define _sMantissaMask 576
|
||||
#define _sMantissaMask1 608
|
||||
#define _sExpMask 640
|
||||
#define _sExpMask1 672
|
||||
#define _iRcpIndexMask 704
|
||||
#define _iBExpMask 736
|
||||
#define _iSignMask 768
|
||||
#define _iBias 800
|
||||
#define _iOne 832
|
||||
#define _i555 864
|
||||
#define _iAbsMask 896
|
||||
#define _iSubConst 928
|
||||
#define _iCmpConst 960
|
||||
#define _sRcp 0
|
||||
#define _sCbrtHL 128
|
||||
#define _sP2 512
|
||||
#define _sP1 544
|
||||
#define _sMantissaMask 576
|
||||
#define _sMantissaMask1 608
|
||||
#define _sExpMask 640
|
||||
#define _sExpMask1 672
|
||||
#define _iRcpIndexMask 704
|
||||
#define _iBExpMask 736
|
||||
#define _iSignMask 768
|
||||
#define _iBias 800
|
||||
#define _iOne 832
|
||||
#define _i555 864
|
||||
#define _iAbsMask 896
|
||||
#define _iSubConst 928
|
||||
#define _iCmpConst 960
|
||||
|
||||
#include <sysdep.h>
|
||||
|
||||
.text
|
||||
.section .text.avx2,"ax",@progbits
|
||||
.section .text.avx2, "ax", @progbits
|
||||
ENTRY(_ZGVdN8v_cbrtf_avx2)
|
||||
pushq %rbp
|
||||
cfi_def_cfa_offset(16)
|
||||
movq %rsp, %rbp
|
||||
cfi_def_cfa(6, 16)
|
||||
cfi_offset(6, -16)
|
||||
andq $-32, %rsp
|
||||
subq $96, %rsp
|
||||
pushq %rbp
|
||||
cfi_def_cfa_offset(16)
|
||||
movq %rsp, %rbp
|
||||
cfi_def_cfa(6, 16)
|
||||
cfi_offset(6, -16)
|
||||
andq $-32, %rsp
|
||||
subq $96, %rsp
|
||||
|
||||
/* Load reciprocal value */
|
||||
lea __svml_scbrt_data_internal(%rip), %rdx
|
||||
vmovaps %ymm0, %ymm5
|
||||
/* Load reciprocal value */
|
||||
lea __svml_scbrt_data_internal(%rip), %rdx
|
||||
vmovaps %ymm0, %ymm5
|
||||
|
||||
/*
|
||||
* Load constants
|
||||
* Reciprocal index calculation
|
||||
*/
|
||||
vpsrld $16, %ymm5, %ymm3
|
||||
vpand _iRcpIndexMask+__svml_scbrt_data_internal(%rip), %ymm3, %ymm4
|
||||
vextractf128 $1, %ymm4, %xmm15
|
||||
vmovd %xmm4, %eax
|
||||
vmovd %xmm15, %r8d
|
||||
vpextrd $1, %xmm15, %r9d
|
||||
vpextrd $2, %xmm15, %r10d
|
||||
vpextrd $3, %xmm15, %r11d
|
||||
movslq %r8d, %r8
|
||||
movslq %r9d, %r9
|
||||
movslq %r10d, %r10
|
||||
movslq %r11d, %r11
|
||||
vpextrd $1, %xmm4, %ecx
|
||||
vpextrd $2, %xmm4, %esi
|
||||
vpextrd $3, %xmm4, %edi
|
||||
movslq %eax, %rax
|
||||
movslq %ecx, %rcx
|
||||
movslq %esi, %rsi
|
||||
movslq %edi, %rdi
|
||||
vmovd (%rdx,%r8), %xmm13
|
||||
vmovd (%rdx,%r9), %xmm14
|
||||
vmovd (%rdx,%r10), %xmm1
|
||||
vmovd (%rdx,%r11), %xmm0
|
||||
vpunpckldq %xmm14, %xmm13, %xmm2
|
||||
vpunpckldq %xmm0, %xmm1, %xmm13
|
||||
/*
|
||||
* Load constants
|
||||
* Reciprocal index calculation
|
||||
*/
|
||||
vpsrld $16, %ymm5, %ymm3
|
||||
vpand _iRcpIndexMask+__svml_scbrt_data_internal(%rip), %ymm3, %ymm4
|
||||
vextractf128 $1, %ymm4, %xmm15
|
||||
vmovd %xmm4, %eax
|
||||
vmovd %xmm15, %r8d
|
||||
vpextrd $1, %xmm15, %r9d
|
||||
vpextrd $2, %xmm15, %r10d
|
||||
vpextrd $3, %xmm15, %r11d
|
||||
movslq %r8d, %r8
|
||||
movslq %r9d, %r9
|
||||
movslq %r10d, %r10
|
||||
movslq %r11d, %r11
|
||||
vpextrd $1, %xmm4, %ecx
|
||||
vpextrd $2, %xmm4, %esi
|
||||
vpextrd $3, %xmm4, %edi
|
||||
movslq %eax, %rax
|
||||
movslq %ecx, %rcx
|
||||
movslq %esi, %rsi
|
||||
movslq %edi, %rdi
|
||||
vmovd (%rdx, %r8), %xmm13
|
||||
vmovd (%rdx, %r9), %xmm14
|
||||
vmovd (%rdx, %r10), %xmm1
|
||||
vmovd (%rdx, %r11), %xmm0
|
||||
vpunpckldq %xmm14, %xmm13, %xmm2
|
||||
vpunpckldq %xmm0, %xmm1, %xmm13
|
||||
|
||||
/* Get signed biased exponent */
|
||||
vpsrld $7, %ymm3, %ymm0
|
||||
vmovd (%rdx,%rax), %xmm6
|
||||
vmovd (%rdx,%rcx), %xmm7
|
||||
vmovd (%rdx,%rsi), %xmm8
|
||||
vmovd (%rdx,%rdi), %xmm9
|
||||
vpunpckldq %xmm7, %xmm6, %xmm10
|
||||
vpunpckldq %xmm9, %xmm8, %xmm11
|
||||
vpunpcklqdq %xmm11, %xmm10, %xmm12
|
||||
vpunpcklqdq %xmm13, %xmm2, %xmm6
|
||||
vandps _iAbsMask+__svml_scbrt_data_internal(%rip), %ymm5, %ymm3
|
||||
/* Get signed biased exponent */
|
||||
vpsrld $7, %ymm3, %ymm0
|
||||
vmovd (%rdx, %rax), %xmm6
|
||||
vmovd (%rdx, %rcx), %xmm7
|
||||
vmovd (%rdx, %rsi), %xmm8
|
||||
vmovd (%rdx, %rdi), %xmm9
|
||||
vpunpckldq %xmm7, %xmm6, %xmm10
|
||||
vpunpckldq %xmm9, %xmm8, %xmm11
|
||||
vpunpcklqdq %xmm11, %xmm10, %xmm12
|
||||
vpunpcklqdq %xmm13, %xmm2, %xmm6
|
||||
vandps _iAbsMask+__svml_scbrt_data_internal(%rip), %ymm5, %ymm3
|
||||
|
||||
/* Argument reduction */
|
||||
vandps _sMantissaMask+__svml_scbrt_data_internal(%rip), %ymm5, %ymm8
|
||||
vandps _sMantissaMask1+__svml_scbrt_data_internal(%rip), %ymm5, %ymm9
|
||||
vpsubd _iSubConst+__svml_scbrt_data_internal(%rip), %ymm3, %ymm7
|
||||
vorps _sExpMask+__svml_scbrt_data_internal(%rip), %ymm8, %ymm10
|
||||
vorps _sExpMask1+__svml_scbrt_data_internal(%rip), %ymm9, %ymm11
|
||||
/* Argument reduction */
|
||||
vandps _sMantissaMask+__svml_scbrt_data_internal(%rip), %ymm5, %ymm8
|
||||
vandps _sMantissaMask1+__svml_scbrt_data_internal(%rip), %ymm5, %ymm9
|
||||
vpsubd _iSubConst+__svml_scbrt_data_internal(%rip), %ymm3, %ymm7
|
||||
vorps _sExpMask+__svml_scbrt_data_internal(%rip), %ymm8, %ymm10
|
||||
vorps _sExpMask1+__svml_scbrt_data_internal(%rip), %ymm9, %ymm11
|
||||
|
||||
/* r=y-y` */
|
||||
vsubps %ymm11, %ymm10, %ymm15
|
||||
/* r=y-y` */
|
||||
vsubps %ymm11, %ymm10, %ymm15
|
||||
|
||||
/* Biased exponent-1 */
|
||||
vpand _iSignMask+__svml_scbrt_data_internal(%rip), %ymm0, %ymm8
|
||||
vpcmpgtd _iCmpConst+__svml_scbrt_data_internal(%rip), %ymm7, %ymm2
|
||||
vmovmskps %ymm2, %eax
|
||||
vinsertf128 $1, %xmm6, %ymm12, %ymm14
|
||||
/* Biased exponent-1 */
|
||||
vpand _iSignMask+__svml_scbrt_data_internal(%rip), %ymm0, %ymm8
|
||||
vpcmpgtd _iCmpConst+__svml_scbrt_data_internal(%rip), %ymm7, %ymm2
|
||||
vmovmskps %ymm2, %eax
|
||||
vinsertf128 $1, %xmm6, %ymm12, %ymm14
|
||||
|
||||
/* Get absolute biased exponent */
|
||||
vpand _iBExpMask+__svml_scbrt_data_internal(%rip), %ymm0, %ymm6
|
||||
/* Get absolute biased exponent */
|
||||
vpand _iBExpMask+__svml_scbrt_data_internal(%rip), %ymm0, %ymm6
|
||||
|
||||
/* r=(y-y`)*rcp_table(y`) */
|
||||
vmulps %ymm15, %ymm14, %ymm1
|
||||
vpsubd _iOne+__svml_scbrt_data_internal(%rip), %ymm6, %ymm10
|
||||
/* r=(y-y`)*rcp_table(y`) */
|
||||
vmulps %ymm15, %ymm14, %ymm1
|
||||
vpsubd _iOne+__svml_scbrt_data_internal(%rip), %ymm6, %ymm10
|
||||
|
||||
/*
|
||||
* Calculate exponent/3
|
||||
* i555Exp=(2^{12}-1)/3*exponent
|
||||
*/
|
||||
vpmulld _i555+__svml_scbrt_data_internal(%rip), %ymm6, %ymm3
|
||||
/*
|
||||
* Calculate exponent/3
|
||||
* i555Exp=(2^{12}-1)/3*exponent
|
||||
*/
|
||||
vpmulld _i555+__svml_scbrt_data_internal(%rip), %ymm6, %ymm3
|
||||
|
||||
/* Get K (exponent=3*k+j) */
|
||||
vpsrld $12, %ymm3, %ymm13
|
||||
/* Get K (exponent=3*k+j) */
|
||||
vpsrld $12, %ymm3, %ymm13
|
||||
|
||||
/* Get J */
|
||||
vpsubd %ymm13, %ymm10, %ymm11
|
||||
/* Get J */
|
||||
vpsubd %ymm13, %ymm10, %ymm11
|
||||
|
||||
/* Add 2/3*(bias-1)+1 to (k+1/3*(bias-1)) */
|
||||
vpaddd _iBias+__svml_scbrt_data_internal(%rip), %ymm13, %ymm7
|
||||
vpsubd %ymm13, %ymm11, %ymm12
|
||||
/* Add 2/3*(bias-1)+1 to (k+1/3*(bias-1)) */
|
||||
vpaddd _iBias+__svml_scbrt_data_internal(%rip), %ymm13, %ymm7
|
||||
vpsubd %ymm13, %ymm11, %ymm12
|
||||
|
||||
/* Attach sign to exponent */
|
||||
vpor %ymm8, %ymm7, %ymm9
|
||||
vpsubd %ymm13, %ymm12, %ymm14
|
||||
vpslld $23, %ymm9, %ymm0
|
||||
/* Attach sign to exponent */
|
||||
vpor %ymm8, %ymm7, %ymm9
|
||||
vpsubd %ymm13, %ymm12, %ymm14
|
||||
vpslld $23, %ymm9, %ymm0
|
||||
|
||||
/* Get 128*J */
|
||||
vpslld $7, %ymm14, %ymm15
|
||||
/* Get 128*J */
|
||||
vpslld $7, %ymm14, %ymm15
|
||||
|
||||
/* iCbrtIndex=4*l+128*j */
|
||||
vpaddd %ymm15, %ymm4, %ymm4
|
||||
/* iCbrtIndex=4*l+128*j */
|
||||
vpaddd %ymm15, %ymm4, %ymm4
|
||||
|
||||
/* Zero index if callout expected */
|
||||
vpandn %ymm4, %ymm2, %ymm4
|
||||
/* Zero index if callout expected */
|
||||
vpandn %ymm4, %ymm2, %ymm4
|
||||
|
||||
/* Load Cbrt table Hi & Lo values */
|
||||
vmovd %xmm4, %ecx
|
||||
vextractf128 $1, %ymm4, %xmm13
|
||||
vpextrd $1, %xmm4, %esi
|
||||
movslq %ecx, %rcx
|
||||
movslq %esi, %rsi
|
||||
vmovd %xmm13, %r9d
|
||||
vmovd 128(%rdx,%rcx), %xmm2
|
||||
vpextrd $2, %xmm4, %edi
|
||||
vpextrd $3, %xmm4, %r8d
|
||||
vmovd 128(%rdx,%rsi), %xmm3
|
||||
vpextrd $1, %xmm13, %r10d
|
||||
vpextrd $2, %xmm13, %ecx
|
||||
vpextrd $3, %xmm13, %esi
|
||||
movslq %edi, %rdi
|
||||
movslq %r8d, %r8
|
||||
movslq %r9d, %r9
|
||||
movslq %r10d, %r10
|
||||
movslq %ecx, %rcx
|
||||
movslq %esi, %rsi
|
||||
vmovd 128(%rdx,%rdi), %xmm6
|
||||
vmovd 128(%rdx,%r8), %xmm7
|
||||
vmovd 128(%rdx,%r9), %xmm11
|
||||
vmovd 128(%rdx,%r10), %xmm12
|
||||
vmovd 128(%rdx,%rcx), %xmm14
|
||||
vmovd 128(%rdx,%rsi), %xmm15
|
||||
vpunpckldq %xmm3, %xmm2, %xmm8
|
||||
vpunpckldq %xmm7, %xmm6, %xmm9
|
||||
vpunpckldq %xmm12, %xmm11, %xmm4
|
||||
vpunpckldq %xmm15, %xmm14, %xmm11
|
||||
vpunpcklqdq %xmm9, %xmm8, %xmm10
|
||||
vpunpcklqdq %xmm11, %xmm4, %xmm2
|
||||
vinsertf128 $1, %xmm2, %ymm10, %ymm3
|
||||
/* Load Cbrt table Hi & Lo values */
|
||||
vmovd %xmm4, %ecx
|
||||
vextractf128 $1, %ymm4, %xmm13
|
||||
vpextrd $1, %xmm4, %esi
|
||||
movslq %ecx, %rcx
|
||||
movslq %esi, %rsi
|
||||
vmovd %xmm13, %r9d
|
||||
vmovd 128(%rdx, %rcx), %xmm2
|
||||
vpextrd $2, %xmm4, %edi
|
||||
vpextrd $3, %xmm4, %r8d
|
||||
vmovd 128(%rdx, %rsi), %xmm3
|
||||
vpextrd $1, %xmm13, %r10d
|
||||
vpextrd $2, %xmm13, %ecx
|
||||
vpextrd $3, %xmm13, %esi
|
||||
movslq %edi, %rdi
|
||||
movslq %r8d, %r8
|
||||
movslq %r9d, %r9
|
||||
movslq %r10d, %r10
|
||||
movslq %ecx, %rcx
|
||||
movslq %esi, %rsi
|
||||
vmovd 128(%rdx, %rdi), %xmm6
|
||||
vmovd 128(%rdx, %r8), %xmm7
|
||||
vmovd 128(%rdx, %r9), %xmm11
|
||||
vmovd 128(%rdx, %r10), %xmm12
|
||||
vmovd 128(%rdx, %rcx), %xmm14
|
||||
vmovd 128(%rdx, %rsi), %xmm15
|
||||
vpunpckldq %xmm3, %xmm2, %xmm8
|
||||
vpunpckldq %xmm7, %xmm6, %xmm9
|
||||
vpunpckldq %xmm12, %xmm11, %xmm4
|
||||
vpunpckldq %xmm15, %xmm14, %xmm11
|
||||
vpunpcklqdq %xmm9, %xmm8, %xmm10
|
||||
vpunpcklqdq %xmm11, %xmm4, %xmm2
|
||||
vinsertf128 $1, %xmm2, %ymm10, %ymm3
|
||||
|
||||
/* sCbrtHi *= 2^k */
|
||||
vmulps %ymm3, %ymm0, %ymm2
|
||||
/* sCbrtHi *= 2^k */
|
||||
vmulps %ymm3, %ymm0, %ymm2
|
||||
|
||||
/* Polynomial: p1+r*(p2*r+r*(p3+r*p4)) */
|
||||
vmovups _sP2+__svml_scbrt_data_internal(%rip), %ymm0
|
||||
vfmadd213ps _sP1+__svml_scbrt_data_internal(%rip), %ymm1, %ymm0
|
||||
/* Polynomial: p1+r*(p2*r+r*(p3+r*p4)) */
|
||||
vmovups _sP2+__svml_scbrt_data_internal(%rip), %ymm0
|
||||
vfmadd213ps _sP1+__svml_scbrt_data_internal(%rip), %ymm1, %ymm0
|
||||
|
||||
/* T`*r */
|
||||
vmulps %ymm2, %ymm1, %ymm1
|
||||
/* T`*r */
|
||||
vmulps %ymm2, %ymm1, %ymm1
|
||||
|
||||
/* (T`*r)*P */
|
||||
vmulps %ymm1, %ymm0, %ymm0
|
||||
/* (T`*r)*P */
|
||||
vmulps %ymm1, %ymm0, %ymm0
|
||||
|
||||
/*
|
||||
* T`*r*P+D`
|
||||
* result = T`+(T`*r*P+D`)
|
||||
*/
|
||||
vaddps %ymm0, %ymm2, %ymm0
|
||||
testl %eax, %eax
|
||||
/*
|
||||
* T`*r*P+D`
|
||||
* result = T`+(T`*r*P+D`)
|
||||
*/
|
||||
vaddps %ymm0, %ymm2, %ymm0
|
||||
testl %eax, %eax
|
||||
|
||||
/* Go to special inputs processing branch */
|
||||
jne L(SPECIAL_VALUES_BRANCH)
|
||||
# LOE rbx r12 r13 r14 r15 eax ymm0 ymm5
|
||||
/* Go to special inputs processing branch */
|
||||
jne L(SPECIAL_VALUES_BRANCH)
|
||||
# LOE rbx r12 r13 r14 r15 eax ymm0 ymm5
|
||||
|
||||
/* Restore registers
|
||||
* and exit the function
|
||||
*/
|
||||
/* Restore registers
|
||||
* and exit the function
|
||||
*/
|
||||
|
||||
L(EXIT):
|
||||
movq %rbp, %rsp
|
||||
popq %rbp
|
||||
cfi_def_cfa(7, 8)
|
||||
cfi_restore(6)
|
||||
ret
|
||||
cfi_def_cfa(6, 16)
|
||||
cfi_offset(6, -16)
|
||||
movq %rbp, %rsp
|
||||
popq %rbp
|
||||
cfi_def_cfa(7, 8)
|
||||
cfi_restore(6)
|
||||
ret
|
||||
cfi_def_cfa(6, 16)
|
||||
cfi_offset(6, -16)
|
||||
|
||||
/* Branch to process
|
||||
* special inputs
|
||||
*/
|
||||
/* Branch to process
|
||||
* special inputs
|
||||
*/
|
||||
|
||||
L(SPECIAL_VALUES_BRANCH):
|
||||
vmovups %ymm5, 32(%rsp)
|
||||
vmovups %ymm0, 64(%rsp)
|
||||
# LOE rbx r12 r13 r14 r15 eax ymm0
|
||||
vmovups %ymm5, 32(%rsp)
|
||||
vmovups %ymm0, 64(%rsp)
|
||||
# LOE rbx r12 r13 r14 r15 eax ymm0
|
||||
|
||||
xorl %edx, %edx
|
||||
# LOE rbx r12 r13 r14 r15 eax edx
|
||||
xorl %edx, %edx
|
||||
# LOE rbx r12 r13 r14 r15 eax edx
|
||||
|
||||
vzeroupper
|
||||
movq %r12, 16(%rsp)
|
||||
/* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -80; DW_OP_plus) */
|
||||
.cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22
|
||||
movl %edx, %r12d
|
||||
movq %r13, 8(%rsp)
|
||||
/* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -88; DW_OP_plus) */
|
||||
.cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22
|
||||
movl %eax, %r13d
|
||||
movq %r14, (%rsp)
|
||||
/* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -96; DW_OP_plus) */
|
||||
.cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22
|
||||
# LOE rbx r15 r12d r13d
|
||||
vzeroupper
|
||||
movq %r12, 16(%rsp)
|
||||
/* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -80; DW_OP_plus) */
|
||||
.cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22
|
||||
movl %edx, %r12d
|
||||
movq %r13, 8(%rsp)
|
||||
/* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -88; DW_OP_plus) */
|
||||
.cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22
|
||||
movl %eax, %r13d
|
||||
movq %r14, (%rsp)
|
||||
/* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -96; DW_OP_plus) */
|
||||
.cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22
|
||||
# LOE rbx r15 r12d r13d
|
||||
|
||||
/* Range mask
|
||||
* bits check
|
||||
*/
|
||||
/* Range mask
|
||||
* bits check
|
||||
*/
|
||||
|
||||
L(RANGEMASK_CHECK):
|
||||
btl %r12d, %r13d
|
||||
btl %r12d, %r13d
|
||||
|
||||
/* Call scalar math function */
|
||||
jc L(SCALAR_MATH_CALL)
|
||||
# LOE rbx r15 r12d r13d
|
||||
/* Call scalar math function */
|
||||
jc L(SCALAR_MATH_CALL)
|
||||
# LOE rbx r15 r12d r13d
|
||||
|
||||
/* Special inputs
|
||||
* processing loop
|
||||
*/
|
||||
/* Special inputs
|
||||
* processing loop
|
||||
*/
|
||||
|
||||
L(SPECIAL_VALUES_LOOP):
|
||||
incl %r12d
|
||||
cmpl $8, %r12d
|
||||
incl %r12d
|
||||
cmpl $8, %r12d
|
||||
|
||||
/* Check bits in range mask */
|
||||
jl L(RANGEMASK_CHECK)
|
||||
# LOE rbx r15 r12d r13d
|
||||
/* Check bits in range mask */
|
||||
jl L(RANGEMASK_CHECK)
|
||||
# LOE rbx r15 r12d r13d
|
||||
|
||||
movq 16(%rsp), %r12
|
||||
cfi_restore(12)
|
||||
movq 8(%rsp), %r13
|
||||
cfi_restore(13)
|
||||
movq (%rsp), %r14
|
||||
cfi_restore(14)
|
||||
vmovups 64(%rsp), %ymm0
|
||||
movq 16(%rsp), %r12
|
||||
cfi_restore(12)
|
||||
movq 8(%rsp), %r13
|
||||
cfi_restore(13)
|
||||
movq (%rsp), %r14
|
||||
cfi_restore(14)
|
||||
vmovups 64(%rsp), %ymm0
|
||||
|
||||
/* Go to exit */
|
||||
jmp L(EXIT)
|
||||
/* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -80; DW_OP_plus) */
|
||||
.cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22
|
||||
/* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -88; DW_OP_plus) */
|
||||
.cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22
|
||||
/* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -96; DW_OP_plus) */
|
||||
.cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22
|
||||
# LOE rbx r12 r13 r14 r15 ymm0
|
||||
/* Go to exit */
|
||||
jmp L(EXIT)
|
||||
/* DW_CFA_expression: r12 (r12) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -80; DW_OP_plus) */
|
||||
.cfi_escape 0x10, 0x0c, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xb0, 0xff, 0xff, 0xff, 0x22
|
||||
/* DW_CFA_expression: r13 (r13) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -88; DW_OP_plus) */
|
||||
.cfi_escape 0x10, 0x0d, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa8, 0xff, 0xff, 0xff, 0x22
|
||||
/* DW_CFA_expression: r14 (r14) (DW_OP_lit8; DW_OP_minus; DW_OP_const4s: -32; DW_OP_and; DW_OP_const4s: -96; DW_OP_plus) */
|
||||
.cfi_escape 0x10, 0x0e, 0x0e, 0x38, 0x1c, 0x0d, 0xe0, 0xff, 0xff, 0xff, 0x1a, 0x0d, 0xa0, 0xff, 0xff, 0xff, 0x22
|
||||
# LOE rbx r12 r13 r14 r15 ymm0
|
||||
|
||||
/* Scalar math fucntion call
|
||||
* to process special input
|
||||
*/
|
||||
/* Scalar math fucntion call
|
||||
* to process special input
|
||||
*/
|
||||
|
||||
L(SCALAR_MATH_CALL):
|
||||
movl %r12d, %r14d
|
||||
movss 32(%rsp,%r14,4), %xmm0
|
||||
call cbrtf@PLT
|
||||
# LOE rbx r14 r15 r12d r13d xmm0
|
||||
movl %r12d, %r14d
|
||||
movss 32(%rsp, %r14, 4), %xmm0
|
||||
call cbrtf@PLT
|
||||
# LOE rbx r14 r15 r12d r13d xmm0
|
||||
|
||||
movss %xmm0, 64(%rsp,%r14,4)
|
||||
movss %xmm0, 64(%rsp, %r14, 4)
|
||||
|
||||
/* Process special inputs in loop */
|
||||
jmp L(SPECIAL_VALUES_LOOP)
|
||||
# LOE rbx r15 r12d r13d
|
||||
/* Process special inputs in loop */
|
||||
jmp L(SPECIAL_VALUES_LOOP)
|
||||
# LOE rbx r15 r12d r13d
|
||||
END(_ZGVdN8v_cbrtf_avx2)
|
||||
|
||||
.section .rodata, "a"
|
||||
.align 32
|
||||
.section .rodata, "a"
|
||||
.align 32
|
||||
|
||||
#ifdef __svml_scbrt_data_internal_typedef
|
||||
typedef unsigned int VUINT32;
|
||||
typedef struct
|
||||
{
|
||||
__declspec(align(32)) VUINT32 _sRcp[32][1];
|
||||
__declspec(align(32)) VUINT32 _sCbrtHL[96][1];
|
||||
__declspec(align(32)) VUINT32 _sP2[8][1];
|
||||
__declspec(align(32)) VUINT32 _sP1[8][1];
|
||||
__declspec(align(32)) VUINT32 _sMantissaMask[8][1];
|
||||
__declspec(align(32)) VUINT32 _sMantissaMask1[8][1];
|
||||
__declspec(align(32)) VUINT32 _sExpMask[8][1];
|
||||
__declspec(align(32)) VUINT32 _sExpMask1[8][1];
|
||||
__declspec(align(32)) VUINT32 _iRcpIndexMask[8][1];
|
||||
__declspec(align(32)) VUINT32 _iBExpMask[8][1];
|
||||
__declspec(align(32)) VUINT32 _iSignMask[8][1];
|
||||
__declspec(align(32)) VUINT32 _iBias[8][1];
|
||||
__declspec(align(32)) VUINT32 _iOne[8][1];
|
||||
__declspec(align(32)) VUINT32 _i555[8][1];
|
||||
__declspec(align(32)) VUINT32 _iAbsMask[8][1];
|
||||
__declspec(align(32)) VUINT32 _iSubConst[8][1];
|
||||
__declspec(align(32)) VUINT32 _iCmpConst[8][1];
|
||||
typedef struct {
|
||||
__declspec(align(32)) VUINT32 _sRcp[32][1];
|
||||
__declspec(align(32)) VUINT32 _sCbrtHL[96][1];
|
||||
__declspec(align(32)) VUINT32 _sP2[8][1];
|
||||
__declspec(align(32)) VUINT32 _sP1[8][1];
|
||||
__declspec(align(32)) VUINT32 _sMantissaMask[8][1];
|
||||
__declspec(align(32)) VUINT32 _sMantissaMask1[8][1];
|
||||
__declspec(align(32)) VUINT32 _sExpMask[8][1];
|
||||
__declspec(align(32)) VUINT32 _sExpMask1[8][1];
|
||||
__declspec(align(32)) VUINT32 _iRcpIndexMask[8][1];
|
||||
__declspec(align(32)) VUINT32 _iBExpMask[8][1];
|
||||
__declspec(align(32)) VUINT32 _iSignMask[8][1];
|
||||
__declspec(align(32)) VUINT32 _iBias[8][1];
|
||||
__declspec(align(32)) VUINT32 _iOne[8][1];
|
||||
__declspec(align(32)) VUINT32 _i555[8][1];
|
||||
__declspec(align(32)) VUINT32 _iAbsMask[8][1];
|
||||
__declspec(align(32)) VUINT32 _iSubConst[8][1];
|
||||
__declspec(align(32)) VUINT32 _iCmpConst[8][1];
|
||||
} __svml_scbrt_data_internal;
|
||||
#endif
|
||||
__svml_scbrt_data_internal:
|
||||
/*== _sRcp ==*/
|
||||
.long 0xBF7C0FC1 /* (1/(1+0/32+1/64)) = -.984615 */
|
||||
.long 0xBF74898D /* (1/(1+1/32+1/64)) = -.955224 */
|
||||
.long 0xBF6D7304 /* (1/(1+2/32+1/64)) = -.927536 */
|
||||
.long 0xBF66C2B4 /* (1/(1+3/32+1/64)) = -.901408 */
|
||||
.long 0xBF607038 /* (1/(1+4/32+1/64)) = -.876712 */
|
||||
.long 0xBF5A740E /* (1/(1+5/32+1/64)) = -.853333 */
|
||||
.long 0xBF54C77B /* (1/(1+6/32+1/64)) = -.831169 */
|
||||
.long 0xBF4F6475 /* (1/(1+7/32+1/64)) = -.810127 */
|
||||
.long 0xBF4A4588 /* (1/(1+8/32+1/64)) = -.790123 */
|
||||
.long 0xBF4565C8 /* (1/(1+9/32+1/64)) = -.771084 */
|
||||
.long 0xBF40C0C1 /* (1/(1+10/32+1/64)) = -.752941 */
|
||||
.long 0xBF3C5264 /* (1/(1+11/32+1/64)) = -.735632 */
|
||||
.long 0xBF381703 /* (1/(1+12/32+1/64)) = -.719101 */
|
||||
.long 0xBF340B41 /* (1/(1+13/32+1/64)) = -.703297 */
|
||||
.long 0xBF302C0B /* (1/(1+14/32+1/64)) = -.688172 */
|
||||
.long 0xBF2C7692 /* (1/(1+15/32+1/64)) = -.673684 */
|
||||
.long 0xBF28E83F /* (1/(1+16/32+1/64)) = -.659794 */
|
||||
.long 0xBF257EB5 /* (1/(1+17/32+1/64)) = -.646465 */
|
||||
.long 0xBF2237C3 /* (1/(1+18/32+1/64)) = -.633663 */
|
||||
.long 0xBF1F1166 /* (1/(1+19/32+1/64)) = -.621359 */
|
||||
.long 0xBF1C09C1 /* (1/(1+20/32+1/64)) = -.609524 */
|
||||
.long 0xBF191F1A /* (1/(1+21/32+1/64)) = -.598131 */
|
||||
.long 0xBF164FDA /* (1/(1+22/32+1/64)) = -.587156 */
|
||||
.long 0xBF139A86 /* (1/(1+23/32+1/64)) = -.576577 */
|
||||
.long 0xBF10FDBC /* (1/(1+24/32+1/64)) = -.566372 */
|
||||
.long 0xBF0E7835 /* (1/(1+25/32+1/64)) = -.556522 */
|
||||
.long 0xBF0C08C1 /* (1/(1+26/32+1/64)) = -.547009 */
|
||||
.long 0xBF09AE41 /* (1/(1+27/32+1/64)) = -.537815 */
|
||||
.long 0xBF0767AB /* (1/(1+28/32+1/64)) = -.528926 */
|
||||
.long 0xBF053408 /* (1/(1+29/32+1/64)) = -.520325 */
|
||||
.long 0xBF03126F /* (1/(1+30/32+1/64)) = -.512 */
|
||||
.long 0xBF010204 /* (1/(1+31/32+1/64)) = -.503937 */
|
||||
/*== _sCbrtHL ==*/
|
||||
.align 32
|
||||
.long 0x3F80A9C9 /* HI((2^0*(1+0/32+1/64))^(1/3)) = 1.005181 */
|
||||
.long 0x3F81F833 /* HI((2^0*(1+1/32+1/64))^(1/3)) = 1.015387 */
|
||||
.long 0x3F834007 /* HI((2^0*(1+2/32+1/64))^(1/3)) = 1.025391 */
|
||||
.long 0x3F848194 /* HI((2^0*(1+3/32+1/64))^(1/3)) = 1.035204 */
|
||||
.long 0x3F85BD25 /* HI((2^0*(1+4/32+1/64))^(1/3)) = 1.044835 */
|
||||
.long 0x3F86F300 /* HI((2^0*(1+5/32+1/64))^(1/3)) = 1.054291 */
|
||||
.long 0x3F882365 /* HI((2^0*(1+6/32+1/64))^(1/3)) = 1.06358 */
|
||||
.long 0x3F894E90 /* HI((2^0*(1+7/32+1/64))^(1/3)) = 1.07271 */
|
||||
.long 0x3F8A74B9 /* HI((2^0*(1+8/32+1/64))^(1/3)) = 1.081687 */
|
||||
.long 0x3F8B9615 /* HI((2^0*(1+9/32+1/64))^(1/3)) = 1.090518 */
|
||||
.long 0x3F8CB2D4 /* HI((2^0*(1+10/32+1/64))^(1/3)) = 1.099207 */
|
||||
.long 0x3F8DCB24 /* HI((2^0*(1+11/32+1/64))^(1/3)) = 1.107762 */
|
||||
.long 0x3F8EDF31 /* HI((2^0*(1+12/32+1/64))^(1/3)) = 1.116186 */
|
||||
.long 0x3F8FEF22 /* HI((2^0*(1+13/32+1/64))^(1/3)) = 1.124485 */
|
||||
.long 0x3F90FB1F /* HI((2^0*(1+14/32+1/64))^(1/3)) = 1.132664 */
|
||||
.long 0x3F92034C /* HI((2^0*(1+15/32+1/64))^(1/3)) = 1.140726 */
|
||||
.long 0x3F9307CA /* HI((2^0*(1+16/32+1/64))^(1/3)) = 1.148675 */
|
||||
.long 0x3F9408B9 /* HI((2^0*(1+17/32+1/64))^(1/3)) = 1.156516 */
|
||||
.long 0x3F950638 /* HI((2^0*(1+18/32+1/64))^(1/3)) = 1.164252 */
|
||||
.long 0x3F960064 /* HI((2^0*(1+19/32+1/64))^(1/3)) = 1.171887 */
|
||||
.long 0x3F96F759 /* HI((2^0*(1+20/32+1/64))^(1/3)) = 1.179423 */
|
||||
.long 0x3F97EB2F /* HI((2^0*(1+21/32+1/64))^(1/3)) = 1.186865 */
|
||||
.long 0x3F98DC01 /* HI((2^0*(1+22/32+1/64))^(1/3)) = 1.194214 */
|
||||
.long 0x3F99C9E5 /* HI((2^0*(1+23/32+1/64))^(1/3)) = 1.201474 */
|
||||
.long 0x3F9AB4F2 /* HI((2^0*(1+24/32+1/64))^(1/3)) = 1.208647 */
|
||||
.long 0x3F9B9D3D /* HI((2^0*(1+25/32+1/64))^(1/3)) = 1.215736 */
|
||||
.long 0x3F9C82DA /* HI((2^0*(1+26/32+1/64))^(1/3)) = 1.222743 */
|
||||
.long 0x3F9D65DD /* HI((2^0*(1+27/32+1/64))^(1/3)) = 1.229671 */
|
||||
.long 0x3F9E4659 /* HI((2^0*(1+28/32+1/64))^(1/3)) = 1.236522 */
|
||||
.long 0x3F9F245F /* HI((2^0*(1+29/32+1/64))^(1/3)) = 1.243297 */
|
||||
.long 0x3FA00000 /* HI((2^0*(1+30/32+1/64))^(1/3)) = 1.25 */
|
||||
.long 0x3FA0D94C /* HI((2^0*(1+31/32+1/64))^(1/3)) = 1.256631 */
|
||||
.long 0x3FA21B02 /* HI((2^1*(1+0/32+1/64))^(1/3)) = 1.266449 */
|
||||
.long 0x3FA3C059 /* HI((2^1*(1+1/32+1/64))^(1/3)) = 1.279307 */
|
||||
.long 0x3FA55D61 /* HI((2^1*(1+2/32+1/64))^(1/3)) = 1.291912 */
|
||||
.long 0x3FA6F282 /* HI((2^1*(1+3/32+1/64))^(1/3)) = 1.304276 */
|
||||
.long 0x3FA8801A /* HI((2^1*(1+4/32+1/64))^(1/3)) = 1.316409 */
|
||||
.long 0x3FAA067E /* HI((2^1*(1+5/32+1/64))^(1/3)) = 1.328323 */
|
||||
.long 0x3FAB8602 /* HI((2^1*(1+6/32+1/64))^(1/3)) = 1.340027 */
|
||||
.long 0x3FACFEEF /* HI((2^1*(1+7/32+1/64))^(1/3)) = 1.35153 */
|
||||
.long 0x3FAE718E /* HI((2^1*(1+8/32+1/64))^(1/3)) = 1.36284 */
|
||||
.long 0x3FAFDE1F /* HI((2^1*(1+9/32+1/64))^(1/3)) = 1.373966 */
|
||||
.long 0x3FB144E1 /* HI((2^1*(1+10/32+1/64))^(1/3)) = 1.384915 */
|
||||
.long 0x3FB2A60D /* HI((2^1*(1+11/32+1/64))^(1/3)) = 1.395692 */
|
||||
.long 0x3FB401DA /* HI((2^1*(1+12/32+1/64))^(1/3)) = 1.406307 */
|
||||
.long 0x3FB5587B /* HI((2^1*(1+13/32+1/64))^(1/3)) = 1.416763 */
|
||||
.long 0x3FB6AA20 /* HI((2^1*(1+14/32+1/64))^(1/3)) = 1.427067 */
|
||||
.long 0x3FB7F6F7 /* HI((2^1*(1+15/32+1/64))^(1/3)) = 1.437224 */
|
||||
.long 0x3FB93F29 /* HI((2^1*(1+16/32+1/64))^(1/3)) = 1.44724 */
|
||||
.long 0x3FBA82E1 /* HI((2^1*(1+17/32+1/64))^(1/3)) = 1.457119 */
|
||||
.long 0x3FBBC244 /* HI((2^1*(1+18/32+1/64))^(1/3)) = 1.466866 */
|
||||
.long 0x3FBCFD77 /* HI((2^1*(1+19/32+1/64))^(1/3)) = 1.476485 */
|
||||
.long 0x3FBE349B /* HI((2^1*(1+20/32+1/64))^(1/3)) = 1.48598 */
|
||||
.long 0x3FBF67D3 /* HI((2^1*(1+21/32+1/64))^(1/3)) = 1.495356 */
|
||||
.long 0x3FC0973C /* HI((2^1*(1+22/32+1/64))^(1/3)) = 1.504615 */
|
||||
.long 0x3FC1C2F6 /* HI((2^1*(1+23/32+1/64))^(1/3)) = 1.513762 */
|
||||
.long 0x3FC2EB1A /* HI((2^1*(1+24/32+1/64))^(1/3)) = 1.5228 */
|
||||
.long 0x3FC40FC6 /* HI((2^1*(1+25/32+1/64))^(1/3)) = 1.531731 */
|
||||
.long 0x3FC53112 /* HI((2^1*(1+26/32+1/64))^(1/3)) = 1.54056 */
|
||||
.long 0x3FC64F16 /* HI((2^1*(1+27/32+1/64))^(1/3)) = 1.549289 */
|
||||
.long 0x3FC769EB /* HI((2^1*(1+28/32+1/64))^(1/3)) = 1.55792 */
|
||||
.long 0x3FC881A6 /* HI((2^1*(1+29/32+1/64))^(1/3)) = 1.566457 */
|
||||
.long 0x3FC9965D /* HI((2^1*(1+30/32+1/64))^(1/3)) = 1.574901 */
|
||||
.long 0x3FCAA825 /* HI((2^1*(1+31/32+1/64))^(1/3)) = 1.583256 */
|
||||
.long 0x3FCC3D79 /* HI((2^2*(1+0/32+1/64))^(1/3)) = 1.595626 */
|
||||
.long 0x3FCE5054 /* HI((2^2*(1+1/32+1/64))^(1/3)) = 1.611826 */
|
||||
.long 0x3FD058B8 /* HI((2^2*(1+2/32+1/64))^(1/3)) = 1.627707 */
|
||||
.long 0x3FD25726 /* HI((2^2*(1+3/32+1/64))^(1/3)) = 1.643285 */
|
||||
.long 0x3FD44C15 /* HI((2^2*(1+4/32+1/64))^(1/3)) = 1.658572 */
|
||||
.long 0x3FD637F2 /* HI((2^2*(1+5/32+1/64))^(1/3)) = 1.673582 */
|
||||
.long 0x3FD81B24 /* HI((2^2*(1+6/32+1/64))^(1/3)) = 1.688328 */
|
||||
.long 0x3FD9F60B /* HI((2^2*(1+7/32+1/64))^(1/3)) = 1.702821 */
|
||||
.long 0x3FDBC8FE /* HI((2^2*(1+8/32+1/64))^(1/3)) = 1.717071 */
|
||||
.long 0x3FDD9452 /* HI((2^2*(1+9/32+1/64))^(1/3)) = 1.731089 */
|
||||
.long 0x3FDF5853 /* HI((2^2*(1+10/32+1/64))^(1/3)) = 1.744883 */
|
||||
.long 0x3FE1154B /* HI((2^2*(1+11/32+1/64))^(1/3)) = 1.758462 */
|
||||
.long 0x3FE2CB7F /* HI((2^2*(1+12/32+1/64))^(1/3)) = 1.771835 */
|
||||
.long 0x3FE47B2E /* HI((2^2*(1+13/32+1/64))^(1/3)) = 1.785009 */
|
||||
.long 0x3FE62496 /* HI((2^2*(1+14/32+1/64))^(1/3)) = 1.797992 */
|
||||
.long 0x3FE7C7F0 /* HI((2^2*(1+15/32+1/64))^(1/3)) = 1.810789 */
|
||||
.long 0x3FE96571 /* HI((2^2*(1+16/32+1/64))^(1/3)) = 1.823408 */
|
||||
.long 0x3FEAFD4C /* HI((2^2*(1+17/32+1/64))^(1/3)) = 1.835855 */
|
||||
.long 0x3FEC8FB3 /* HI((2^2*(1+18/32+1/64))^(1/3)) = 1.848135 */
|
||||
.long 0x3FEE1CD3 /* HI((2^2*(1+19/32+1/64))^(1/3)) = 1.860255 */
|
||||
.long 0x3FEFA4D7 /* HI((2^2*(1+20/32+1/64))^(1/3)) = 1.872218 */
|
||||
.long 0x3FF127E9 /* HI((2^2*(1+21/32+1/64))^(1/3)) = 1.88403 */
|
||||
.long 0x3FF2A62F /* HI((2^2*(1+22/32+1/64))^(1/3)) = 1.895697 */
|
||||
.long 0x3FF41FD0 /* HI((2^2*(1+23/32+1/64))^(1/3)) = 1.907221 */
|
||||
.long 0x3FF594EE /* HI((2^2*(1+24/32+1/64))^(1/3)) = 1.918607 */
|
||||
.long 0x3FF705AC /* HI((2^2*(1+25/32+1/64))^(1/3)) = 1.929861 */
|
||||
.long 0x3FF8722A /* HI((2^2*(1+26/32+1/64))^(1/3)) = 1.940984 */
|
||||
.long 0x3FF9DA86 /* HI((2^2*(1+27/32+1/64))^(1/3)) = 1.951981 */
|
||||
.long 0x3FFB3EDE /* HI((2^2*(1+28/32+1/64))^(1/3)) = 1.962856 */
|
||||
.long 0x3FFC9F4E /* HI((2^2*(1+29/32+1/64))^(1/3)) = 1.973612 */
|
||||
.long 0x3FFDFBF2 /* HI((2^2*(1+30/32+1/64))^(1/3)) = 1.984251 */
|
||||
.long 0x3FFF54E3 /* HI((2^2*(1+31/32+1/64))^(1/3)) = 1.994778 */
|
||||
.align 32
|
||||
.long 0xBDE3A962, 0xBDE3A962, 0xBDE3A962, 0xBDE3A962, 0xBDE3A962, 0xBDE3A962, 0xBDE3A962, 0xBDE3A962 /* _sP2 */
|
||||
.align 32
|
||||
.long 0x3EAAAC91, 0x3EAAAC91, 0x3EAAAC91, 0x3EAAAC91, 0x3EAAAC91, 0x3EAAAC91, 0x3EAAAC91, 0x3EAAAC91 /* _sP1 */
|
||||
.align 32
|
||||
.long 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff /* _sMantissaMask (EXP_MSK3) */
|
||||
.align 32
|
||||
.long 0x007e0000, 0x007e0000, 0x007e0000, 0x007e0000, 0x007e0000, 0x007e0000, 0x007e0000, 0x007e0000 /* _sMantissaMask1 (SIG_MASK) */
|
||||
.align 32
|
||||
.long 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000 /* _sExpMask (EXP_MASK) */
|
||||
.align 32
|
||||
.long 0xBF820000, 0xBF820000, 0xBF820000, 0xBF820000, 0xBF820000, 0xBF820000, 0xBF820000, 0xBF820000 /* _sExpMask1 (EXP_MASK2) */
|
||||
.align 32
|
||||
.long 0x0000007c, 0x0000007c, 0x0000007c, 0x0000007c, 0x0000007c, 0x0000007c, 0x0000007c, 0x0000007c /* _iRcpIndexMask */
|
||||
.align 32
|
||||
.long 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff /* _iBExpMask */
|
||||
.align 32
|
||||
.long 0x00000100, 0x00000100, 0x00000100, 0x00000100, 0x00000100, 0x00000100, 0x00000100, 0x00000100 /* _iSignMask */
|
||||
.align 32
|
||||
.long 0x00000055, 0x00000055, 0x00000055, 0x00000055, 0x00000055, 0x00000055, 0x00000055, 0x00000055 /* _iBias */
|
||||
.align 32
|
||||
.long 0x00000001, 0x00000001, 0x00000001, 0x00000001, 0x00000001, 0x00000001, 0x00000001, 0x00000001 /* _iOne */
|
||||
.align 32
|
||||
.long 0x00000555, 0x00000555, 0x00000555, 0x00000555, 0x00000555, 0x00000555, 0x00000555, 0x00000555 /* _i555 */
|
||||
.align 32
|
||||
.long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff /* _iAbsMask */
|
||||
.align 32
|
||||
.long 0x80800000, 0x80800000, 0x80800000, 0x80800000, 0x80800000, 0x80800000, 0x80800000, 0x80800000 /* _iSubConst */
|
||||
.align 32
|
||||
.long 0xFEFFFFFF, 0xFEFFFFFF, 0xFEFFFFFF, 0xFEFFFFFF, 0xFEFFFFFF, 0xFEFFFFFF, 0xFEFFFFFF, 0xFEFFFFFF /* _iCmpConst */
|
||||
.align 32
|
||||
.type __svml_scbrt_data_internal,@object
|
||||
.size __svml_scbrt_data_internal,.-__svml_scbrt_data_internal
|
||||
/* _sRcp */
|
||||
.long 0xBF7C0FC1 /* (1/(1+0/32+1/64)) = -.984615 */
|
||||
.long 0xBF74898D /* (1/(1+1/32+1/64)) = -.955224 */
|
||||
.long 0xBF6D7304 /* (1/(1+2/32+1/64)) = -.927536 */
|
||||
.long 0xBF66C2B4 /* (1/(1+3/32+1/64)) = -.901408 */
|
||||
.long 0xBF607038 /* (1/(1+4/32+1/64)) = -.876712 */
|
||||
.long 0xBF5A740E /* (1/(1+5/32+1/64)) = -.853333 */
|
||||
.long 0xBF54C77B /* (1/(1+6/32+1/64)) = -.831169 */
|
||||
.long 0xBF4F6475 /* (1/(1+7/32+1/64)) = -.810127 */
|
||||
.long 0xBF4A4588 /* (1/(1+8/32+1/64)) = -.790123 */
|
||||
.long 0xBF4565C8 /* (1/(1+9/32+1/64)) = -.771084 */
|
||||
.long 0xBF40C0C1 /* (1/(1+10/32+1/64)) = -.752941 */
|
||||
.long 0xBF3C5264 /* (1/(1+11/32+1/64)) = -.735632 */
|
||||
.long 0xBF381703 /* (1/(1+12/32+1/64)) = -.719101 */
|
||||
.long 0xBF340B41 /* (1/(1+13/32+1/64)) = -.703297 */
|
||||
.long 0xBF302C0B /* (1/(1+14/32+1/64)) = -.688172 */
|
||||
.long 0xBF2C7692 /* (1/(1+15/32+1/64)) = -.673684 */
|
||||
.long 0xBF28E83F /* (1/(1+16/32+1/64)) = -.659794 */
|
||||
.long 0xBF257EB5 /* (1/(1+17/32+1/64)) = -.646465 */
|
||||
.long 0xBF2237C3 /* (1/(1+18/32+1/64)) = -.633663 */
|
||||
.long 0xBF1F1166 /* (1/(1+19/32+1/64)) = -.621359 */
|
||||
.long 0xBF1C09C1 /* (1/(1+20/32+1/64)) = -.609524 */
|
||||
.long 0xBF191F1A /* (1/(1+21/32+1/64)) = -.598131 */
|
||||
.long 0xBF164FDA /* (1/(1+22/32+1/64)) = -.587156 */
|
||||
.long 0xBF139A86 /* (1/(1+23/32+1/64)) = -.576577 */
|
||||
.long 0xBF10FDBC /* (1/(1+24/32+1/64)) = -.566372 */
|
||||
.long 0xBF0E7835 /* (1/(1+25/32+1/64)) = -.556522 */
|
||||
.long 0xBF0C08C1 /* (1/(1+26/32+1/64)) = -.547009 */
|
||||
.long 0xBF09AE41 /* (1/(1+27/32+1/64)) = -.537815 */
|
||||
.long 0xBF0767AB /* (1/(1+28/32+1/64)) = -.528926 */
|
||||
.long 0xBF053408 /* (1/(1+29/32+1/64)) = -.520325 */
|
||||
.long 0xBF03126F /* (1/(1+30/32+1/64)) = -.512 */
|
||||
.long 0xBF010204 /* (1/(1+31/32+1/64)) = -.503937 */
|
||||
/* _sCbrtHL */
|
||||
.align 32
|
||||
.long 0x3F80A9C9 /* HI((2^0*(1+0/32+1/64))^(1/3)) = 1.005181 */
|
||||
.long 0x3F81F833 /* HI((2^0*(1+1/32+1/64))^(1/3)) = 1.015387 */
|
||||
.long 0x3F834007 /* HI((2^0*(1+2/32+1/64))^(1/3)) = 1.025391 */
|
||||
.long 0x3F848194 /* HI((2^0*(1+3/32+1/64))^(1/3)) = 1.035204 */
|
||||
.long 0x3F85BD25 /* HI((2^0*(1+4/32+1/64))^(1/3)) = 1.044835 */
|
||||
.long 0x3F86F300 /* HI((2^0*(1+5/32+1/64))^(1/3)) = 1.054291 */
|
||||
.long 0x3F882365 /* HI((2^0*(1+6/32+1/64))^(1/3)) = 1.06358 */
|
||||
.long 0x3F894E90 /* HI((2^0*(1+7/32+1/64))^(1/3)) = 1.07271 */
|
||||
.long 0x3F8A74B9 /* HI((2^0*(1+8/32+1/64))^(1/3)) = 1.081687 */
|
||||
.long 0x3F8B9615 /* HI((2^0*(1+9/32+1/64))^(1/3)) = 1.090518 */
|
||||
.long 0x3F8CB2D4 /* HI((2^0*(1+10/32+1/64))^(1/3)) = 1.099207 */
|
||||
.long 0x3F8DCB24 /* HI((2^0*(1+11/32+1/64))^(1/3)) = 1.107762 */
|
||||
.long 0x3F8EDF31 /* HI((2^0*(1+12/32+1/64))^(1/3)) = 1.116186 */
|
||||
.long 0x3F8FEF22 /* HI((2^0*(1+13/32+1/64))^(1/3)) = 1.124485 */
|
||||
.long 0x3F90FB1F /* HI((2^0*(1+14/32+1/64))^(1/3)) = 1.132664 */
|
||||
.long 0x3F92034C /* HI((2^0*(1+15/32+1/64))^(1/3)) = 1.140726 */
|
||||
.long 0x3F9307CA /* HI((2^0*(1+16/32+1/64))^(1/3)) = 1.148675 */
|
||||
.long 0x3F9408B9 /* HI((2^0*(1+17/32+1/64))^(1/3)) = 1.156516 */
|
||||
.long 0x3F950638 /* HI((2^0*(1+18/32+1/64))^(1/3)) = 1.164252 */
|
||||
.long 0x3F960064 /* HI((2^0*(1+19/32+1/64))^(1/3)) = 1.171887 */
|
||||
.long 0x3F96F759 /* HI((2^0*(1+20/32+1/64))^(1/3)) = 1.179423 */
|
||||
.long 0x3F97EB2F /* HI((2^0*(1+21/32+1/64))^(1/3)) = 1.186865 */
|
||||
.long 0x3F98DC01 /* HI((2^0*(1+22/32+1/64))^(1/3)) = 1.194214 */
|
||||
.long 0x3F99C9E5 /* HI((2^0*(1+23/32+1/64))^(1/3)) = 1.201474 */
|
||||
.long 0x3F9AB4F2 /* HI((2^0*(1+24/32+1/64))^(1/3)) = 1.208647 */
|
||||
.long 0x3F9B9D3D /* HI((2^0*(1+25/32+1/64))^(1/3)) = 1.215736 */
|
||||
.long 0x3F9C82DA /* HI((2^0*(1+26/32+1/64))^(1/3)) = 1.222743 */
|
||||
.long 0x3F9D65DD /* HI((2^0*(1+27/32+1/64))^(1/3)) = 1.229671 */
|
||||
.long 0x3F9E4659 /* HI((2^0*(1+28/32+1/64))^(1/3)) = 1.236522 */
|
||||
.long 0x3F9F245F /* HI((2^0*(1+29/32+1/64))^(1/3)) = 1.243297 */
|
||||
.long 0x3FA00000 /* HI((2^0*(1+30/32+1/64))^(1/3)) = 1.25 */
|
||||
.long 0x3FA0D94C /* HI((2^0*(1+31/32+1/64))^(1/3)) = 1.256631 */
|
||||
.long 0x3FA21B02 /* HI((2^1*(1+0/32+1/64))^(1/3)) = 1.266449 */
|
||||
.long 0x3FA3C059 /* HI((2^1*(1+1/32+1/64))^(1/3)) = 1.279307 */
|
||||
.long 0x3FA55D61 /* HI((2^1*(1+2/32+1/64))^(1/3)) = 1.291912 */
|
||||
.long 0x3FA6F282 /* HI((2^1*(1+3/32+1/64))^(1/3)) = 1.304276 */
|
||||
.long 0x3FA8801A /* HI((2^1*(1+4/32+1/64))^(1/3)) = 1.316409 */
|
||||
.long 0x3FAA067E /* HI((2^1*(1+5/32+1/64))^(1/3)) = 1.328323 */
|
||||
.long 0x3FAB8602 /* HI((2^1*(1+6/32+1/64))^(1/3)) = 1.340027 */
|
||||
.long 0x3FACFEEF /* HI((2^1*(1+7/32+1/64))^(1/3)) = 1.35153 */
|
||||
.long 0x3FAE718E /* HI((2^1*(1+8/32+1/64))^(1/3)) = 1.36284 */
|
||||
.long 0x3FAFDE1F /* HI((2^1*(1+9/32+1/64))^(1/3)) = 1.373966 */
|
||||
.long 0x3FB144E1 /* HI((2^1*(1+10/32+1/64))^(1/3)) = 1.384915 */
|
||||
.long 0x3FB2A60D /* HI((2^1*(1+11/32+1/64))^(1/3)) = 1.395692 */
|
||||
.long 0x3FB401DA /* HI((2^1*(1+12/32+1/64))^(1/3)) = 1.406307 */
|
||||
.long 0x3FB5587B /* HI((2^1*(1+13/32+1/64))^(1/3)) = 1.416763 */
|
||||
.long 0x3FB6AA20 /* HI((2^1*(1+14/32+1/64))^(1/3)) = 1.427067 */
|
||||
.long 0x3FB7F6F7 /* HI((2^1*(1+15/32+1/64))^(1/3)) = 1.437224 */
|
||||
.long 0x3FB93F29 /* HI((2^1*(1+16/32+1/64))^(1/3)) = 1.44724 */
|
||||
.long 0x3FBA82E1 /* HI((2^1*(1+17/32+1/64))^(1/3)) = 1.457119 */
|
||||
.long 0x3FBBC244 /* HI((2^1*(1+18/32+1/64))^(1/3)) = 1.466866 */
|
||||
.long 0x3FBCFD77 /* HI((2^1*(1+19/32+1/64))^(1/3)) = 1.476485 */
|
||||
.long 0x3FBE349B /* HI((2^1*(1+20/32+1/64))^(1/3)) = 1.48598 */
|
||||
.long 0x3FBF67D3 /* HI((2^1*(1+21/32+1/64))^(1/3)) = 1.495356 */
|
||||
.long 0x3FC0973C /* HI((2^1*(1+22/32+1/64))^(1/3)) = 1.504615 */
|
||||
.long 0x3FC1C2F6 /* HI((2^1*(1+23/32+1/64))^(1/3)) = 1.513762 */
|
||||
.long 0x3FC2EB1A /* HI((2^1*(1+24/32+1/64))^(1/3)) = 1.5228 */
|
||||
.long 0x3FC40FC6 /* HI((2^1*(1+25/32+1/64))^(1/3)) = 1.531731 */
|
||||
.long 0x3FC53112 /* HI((2^1*(1+26/32+1/64))^(1/3)) = 1.54056 */
|
||||
.long 0x3FC64F16 /* HI((2^1*(1+27/32+1/64))^(1/3)) = 1.549289 */
|
||||
.long 0x3FC769EB /* HI((2^1*(1+28/32+1/64))^(1/3)) = 1.55792 */
|
||||
.long 0x3FC881A6 /* HI((2^1*(1+29/32+1/64))^(1/3)) = 1.566457 */
|
||||
.long 0x3FC9965D /* HI((2^1*(1+30/32+1/64))^(1/3)) = 1.574901 */
|
||||
.long 0x3FCAA825 /* HI((2^1*(1+31/32+1/64))^(1/3)) = 1.583256 */
|
||||
.long 0x3FCC3D79 /* HI((2^2*(1+0/32+1/64))^(1/3)) = 1.595626 */
|
||||
.long 0x3FCE5054 /* HI((2^2*(1+1/32+1/64))^(1/3)) = 1.611826 */
|
||||
.long 0x3FD058B8 /* HI((2^2*(1+2/32+1/64))^(1/3)) = 1.627707 */
|
||||
.long 0x3FD25726 /* HI((2^2*(1+3/32+1/64))^(1/3)) = 1.643285 */
|
||||
.long 0x3FD44C15 /* HI((2^2*(1+4/32+1/64))^(1/3)) = 1.658572 */
|
||||
.long 0x3FD637F2 /* HI((2^2*(1+5/32+1/64))^(1/3)) = 1.673582 */
|
||||
.long 0x3FD81B24 /* HI((2^2*(1+6/32+1/64))^(1/3)) = 1.688328 */
|
||||
.long 0x3FD9F60B /* HI((2^2*(1+7/32+1/64))^(1/3)) = 1.702821 */
|
||||
.long 0x3FDBC8FE /* HI((2^2*(1+8/32+1/64))^(1/3)) = 1.717071 */
|
||||
.long 0x3FDD9452 /* HI((2^2*(1+9/32+1/64))^(1/3)) = 1.731089 */
|
||||
.long 0x3FDF5853 /* HI((2^2*(1+10/32+1/64))^(1/3)) = 1.744883 */
|
||||
.long 0x3FE1154B /* HI((2^2*(1+11/32+1/64))^(1/3)) = 1.758462 */
|
||||
.long 0x3FE2CB7F /* HI((2^2*(1+12/32+1/64))^(1/3)) = 1.771835 */
|
||||
.long 0x3FE47B2E /* HI((2^2*(1+13/32+1/64))^(1/3)) = 1.785009 */
|
||||
.long 0x3FE62496 /* HI((2^2*(1+14/32+1/64))^(1/3)) = 1.797992 */
|
||||
.long 0x3FE7C7F0 /* HI((2^2*(1+15/32+1/64))^(1/3)) = 1.810789 */
|
||||
.long 0x3FE96571 /* HI((2^2*(1+16/32+1/64))^(1/3)) = 1.823408 */
|
||||
.long 0x3FEAFD4C /* HI((2^2*(1+17/32+1/64))^(1/3)) = 1.835855 */
|
||||
.long 0x3FEC8FB3 /* HI((2^2*(1+18/32+1/64))^(1/3)) = 1.848135 */
|
||||
.long 0x3FEE1CD3 /* HI((2^2*(1+19/32+1/64))^(1/3)) = 1.860255 */
|
||||
.long 0x3FEFA4D7 /* HI((2^2*(1+20/32+1/64))^(1/3)) = 1.872218 */
|
||||
.long 0x3FF127E9 /* HI((2^2*(1+21/32+1/64))^(1/3)) = 1.88403 */
|
||||
.long 0x3FF2A62F /* HI((2^2*(1+22/32+1/64))^(1/3)) = 1.895697 */
|
||||
.long 0x3FF41FD0 /* HI((2^2*(1+23/32+1/64))^(1/3)) = 1.907221 */
|
||||
.long 0x3FF594EE /* HI((2^2*(1+24/32+1/64))^(1/3)) = 1.918607 */
|
||||
.long 0x3FF705AC /* HI((2^2*(1+25/32+1/64))^(1/3)) = 1.929861 */
|
||||
.long 0x3FF8722A /* HI((2^2*(1+26/32+1/64))^(1/3)) = 1.940984 */
|
||||
.long 0x3FF9DA86 /* HI((2^2*(1+27/32+1/64))^(1/3)) = 1.951981 */
|
||||
.long 0x3FFB3EDE /* HI((2^2*(1+28/32+1/64))^(1/3)) = 1.962856 */
|
||||
.long 0x3FFC9F4E /* HI((2^2*(1+29/32+1/64))^(1/3)) = 1.973612 */
|
||||
.long 0x3FFDFBF2 /* HI((2^2*(1+30/32+1/64))^(1/3)) = 1.984251 */
|
||||
.long 0x3FFF54E3 /* HI((2^2*(1+31/32+1/64))^(1/3)) = 1.994778 */
|
||||
.align 32
|
||||
.long 0xBDE3A962, 0xBDE3A962, 0xBDE3A962, 0xBDE3A962, 0xBDE3A962, 0xBDE3A962, 0xBDE3A962, 0xBDE3A962 /* _sP2 */
|
||||
.align 32
|
||||
.long 0x3EAAAC91, 0x3EAAAC91, 0x3EAAAC91, 0x3EAAAC91, 0x3EAAAC91, 0x3EAAAC91, 0x3EAAAC91, 0x3EAAAC91 /* _sP1 */
|
||||
.align 32
|
||||
.long 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff, 0x007fffff /* _sMantissaMask (EXP_MSK3) */
|
||||
.align 32
|
||||
.long 0x007e0000, 0x007e0000, 0x007e0000, 0x007e0000, 0x007e0000, 0x007e0000, 0x007e0000, 0x007e0000 /* _sMantissaMask1 (SIG_MASK) */
|
||||
.align 32
|
||||
.long 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000 /* _sExpMask (EXP_MASK) */
|
||||
.align 32
|
||||
.long 0xBF820000, 0xBF820000, 0xBF820000, 0xBF820000, 0xBF820000, 0xBF820000, 0xBF820000, 0xBF820000 /* _sExpMask1 (EXP_MASK2) */
|
||||
.align 32
|
||||
.long 0x0000007c, 0x0000007c, 0x0000007c, 0x0000007c, 0x0000007c, 0x0000007c, 0x0000007c, 0x0000007c /* _iRcpIndexMask */
|
||||
.align 32
|
||||
.long 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff, 0x000000ff /* _iBExpMask */
|
||||
.align 32
|
||||
.long 0x00000100, 0x00000100, 0x00000100, 0x00000100, 0x00000100, 0x00000100, 0x00000100, 0x00000100 /* _iSignMask */
|
||||
.align 32
|
||||
.long 0x00000055, 0x00000055, 0x00000055, 0x00000055, 0x00000055, 0x00000055, 0x00000055, 0x00000055 /* _iBias */
|
||||
.align 32
|
||||
.long 0x00000001, 0x00000001, 0x00000001, 0x00000001, 0x00000001, 0x00000001, 0x00000001, 0x00000001 /* _iOne */
|
||||
.align 32
|
||||
.long 0x00000555, 0x00000555, 0x00000555, 0x00000555, 0x00000555, 0x00000555, 0x00000555, 0x00000555 /* _i555 */
|
||||
.align 32
|
||||
.long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff /* _iAbsMask */
|
||||
.align 32
|
||||
.long 0x80800000, 0x80800000, 0x80800000, 0x80800000, 0x80800000, 0x80800000, 0x80800000, 0x80800000 /* _iSubConst */
|
||||
.align 32
|
||||
.long 0xFEFFFFFF, 0xFEFFFFFF, 0xFEFFFFFF, 0xFEFFFFFF, 0xFEFFFFFF, 0xFEFFFFFF, 0xFEFFFFFF, 0xFEFFFFFF /* _iCmpConst */
|
||||
.align 32
|
||||
.type __svml_scbrt_data_internal, @object
|
||||
.size __svml_scbrt_data_internal, .-__svml_scbrt_data_internal
|
||||
|
|
|
|||
Loading…
Reference in New Issue