aboutsummaryrefslogtreecommitdiff
path: root/sysdeps
diff options
context:
space:
mode:
authorSunil K Pandey <skpgkp2@gmail.com>2022-03-07 10:47:11 -0800
committerSunil K Pandey <skpgkp2@gmail.com>2022-03-07 21:14:10 -0800
commit85ef28262a5796284f0642b21bbd61d5a8673e19 (patch)
tree6ee4efe9ebacbff58fae695def2d5ff0863b0ea1 /sysdeps
parentc8a3012174421daf9bba97398a995635ea193b97 (diff)
downloadglibc-85ef28262a5796284f0642b21bbd61d5a8673e19.tar
glibc-85ef28262a5796284f0642b21bbd61d5a8673e19.tar.gz
glibc-85ef28262a5796284f0642b21bbd61d5a8673e19.tar.bz2
glibc-85ef28262a5796284f0642b21bbd61d5a8673e19.zip
x86_64: Fix svml_d_cbrt4_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>
Diffstat (limited to 'sysdeps')
-rw-r--r--sysdeps/x86_64/fpu/multiarch/svml_d_cbrt4_core_avx2.S880
1 files changed, 439 insertions, 441 deletions
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_cbrt4_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_cbrt4_core_avx2.S
index 28f779d23c..96ecbe05c1 100644
--- a/sysdeps/x86_64/fpu/multiarch/svml_d_cbrt4_core_avx2.S
+++ b/sysdeps/x86_64/fpu/multiarch/svml_d_cbrt4_core_avx2.S
@@ -31,475 +31,473 @@
/* Offsets for data table __svml_dcbrt_data_internal
*/
-#define _dRcp 0
-#define _dCbrtHiLo 256
-#define _dA7 1024
-#define _dA6 1056
-#define _dA5 1088
-#define _dA4 1120
-#define _dA3 1152
-#define _dA2 1184
-#define _dA1 1216
-#define _dNeg65Div64 1248
-#define _dSgnf6Mask 1280
-#define _dNegOne 1312
-#define _dMantissaMask 1344
-#define _lExpHiMask 1376
-#define _lExpLoMask 1408
-#define _l1556 1440
-#define _iRcpIndexMask 1472
-#define _iAbsMask 1504
-#define _iSignMask 1536
-#define _iBias 1568
-#define _iSub 1600
-#define _iCmp 1632
+#define _dRcp 0
+#define _dCbrtHiLo 256
+#define _dA7 1024
+#define _dA6 1056
+#define _dA5 1088
+#define _dA4 1120
+#define _dA3 1152
+#define _dA2 1184
+#define _dA1 1216
+#define _dNeg65Div64 1248
+#define _dSgnf6Mask 1280
+#define _dNegOne 1312
+#define _dMantissaMask 1344
+#define _lExpHiMask 1376
+#define _lExpLoMask 1408
+#define _l1556 1440
+#define _iRcpIndexMask 1472
+#define _iAbsMask 1504
+#define _iSignMask 1536
+#define _iBias 1568
+#define _iSub 1600
+#define _iCmp 1632
#include <sysdep.h>
- .text
- .section .text.avx2,"ax",@progbits
+ .section .text.avx2, "ax", @progbits
ENTRY(_ZGVdN4v_cbrt_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
-
-/* Load 1/(1+iRcpIndex/32+1/64) reciprocal table value */
- lea __svml_dcbrt_data_internal(%rip), %rax
- vmovapd %ymm0, %ymm5
-
-/*
- * Declarations
- * Load constants
- * Get iX - high part of argument
- */
- vextractf128 $1, %ymm5, %xmm6
-
-/* Calculate CbrtIndex */
- vpsrlq $52, %ymm5, %ymm15
- vshufps $221, %xmm6, %xmm5, %xmm4
-
-/* Calculate Rcp table index */
- vandps _iRcpIndexMask+__svml_dcbrt_data_internal(%rip), %xmm4, %xmm10
- vpsrld $12, %xmm10, %xmm3
- vmovd %xmm3, %ecx
-
-/* If the exponent field is zero - go to callout to process denormals */
- vandps _iAbsMask+__svml_dcbrt_data_internal(%rip), %xmm4, %xmm7
-
-/* Compute 2^k */
- vpsrld $20, %xmm4, %xmm4
- vpsubd _iSub+__svml_dcbrt_data_internal(%rip), %xmm7, %xmm8
- vandps _lExpLoMask+__svml_dcbrt_data_internal(%rip), %ymm15, %ymm0
- vpmuludq _l1556+__svml_dcbrt_data_internal(%rip), %ymm0, %ymm6
- vpextrd $2, %xmm3, %edi
- movslq %ecx, %rcx
- vpextrd $1, %xmm3, %esi
- movslq %edi, %rdi
- vpextrd $3, %xmm3, %r8d
- movslq %esi, %rsi
- movslq %r8d, %r8
- vpcmpgtd _iCmp+__svml_dcbrt_data_internal(%rip), %xmm8, %xmm9
- vmovsd (%rax,%rcx), %xmm11
- vmovmskps %xmm9, %edx
- vmovsd (%rax,%rdi), %xmm13
- vmovhpd (%rax,%rsi), %xmm11, %xmm12
- vmovhpd (%rax,%r8), %xmm13, %xmm14
- vextractf128 $1, %ymm6, %xmm7
- vshufps $136, %xmm7, %xmm6, %xmm8
- vmovups __VUNPACK_ODD_ind1.613.0.1(%rip), %ymm7
- vextractf128 $1, %ymm0, %xmm1
- vshufps $136, %xmm1, %xmm0, %xmm9
- vpsrld $14, %xmm8, %xmm1
- vpsubd %xmm1, %xmm9, %xmm10
- vpaddd %xmm1, %xmm1, %xmm11
-
-/*
- * VAND( L, l2k, = l2k, lExpHiMask );
- * Argument reduction Z
- */
- vandpd _dMantissaMask+__svml_dcbrt_data_internal(%rip), %ymm5, %ymm9
- vinsertf128 $1, %xmm14, %ymm12, %ymm2
- vpsubd %xmm11, %xmm10, %xmm12
- vpslld $8, %xmm12, %xmm13
- vpaddd %xmm13, %xmm3, %xmm15
-
-/* Load cbrt(2^j*(1+iRcpIndex/32+1/64)) Hi & Lo values */
- vmovd %xmm15, %r9d
- vpextrd $2, %xmm15, %r11d
- movslq %r9d, %r9
- vpextrd $1, %xmm15, %r10d
- movslq %r11d, %r11
- vpextrd $3, %xmm15, %ecx
- movslq %r10d, %r10
- movslq %ecx, %rcx
- vmovsd 256(%rax,%r9), %xmm3
- vmovsd 256(%rax,%r11), %xmm0
- vandpd _dSgnf6Mask+__svml_dcbrt_data_internal(%rip), %ymm5, %ymm10
- vmovhpd 256(%rax,%r10), %xmm3, %xmm14
- vmovhpd 256(%rax,%rcx), %xmm0, %xmm3
- vorpd _dNegOne+__svml_dcbrt_data_internal(%rip), %ymm9, %ymm11
- vorpd _dNeg65Div64+__svml_dcbrt_data_internal(%rip), %ymm10, %ymm12
- vsubpd %ymm12, %ymm11, %ymm13
- vmulpd %ymm13, %ymm2, %ymm2
- vinsertf128 $1, %xmm3, %ymm14, %ymm0
- vpand _iSignMask+__svml_dcbrt_data_internal(%rip), %xmm4, %xmm3
- vpor _iBias+__svml_dcbrt_data_internal(%rip), %xmm3, %xmm4
- vpaddd %xmm1, %xmm4, %xmm1
- vpslld $20, %xmm1, %xmm6
-
-/* Polynomial */
- vmovupd _dA7+__svml_dcbrt_data_internal(%rip), %ymm1
- vfmadd213pd _dA6+__svml_dcbrt_data_internal(%rip), %ymm2, %ymm1
- vfmadd213pd _dA5+__svml_dcbrt_data_internal(%rip), %ymm2, %ymm1
- vfmadd213pd _dA4+__svml_dcbrt_data_internal(%rip), %ymm2, %ymm1
- vfmadd213pd _dA3+__svml_dcbrt_data_internal(%rip), %ymm2, %ymm1
- vfmadd213pd _dA2+__svml_dcbrt_data_internal(%rip), %ymm2, %ymm1
- vfmadd213pd _dA1+__svml_dcbrt_data_internal(%rip), %ymm2, %ymm1
- vpermps %ymm6, %ymm7, %ymm8
- vandps __VUNPACK_ODD_mask.613.0.1(%rip), %ymm8, %ymm14
-
-/* THi*2^k, TLo*2^k */
- vmulpd %ymm14, %ymm0, %ymm0
-
-/* THi*2^k*Z */
- vmulpd %ymm0, %ymm2, %ymm2
-
-/* Final reconstruction */
- vmulpd %ymm2, %ymm1, %ymm3
- vaddpd %ymm3, %ymm0, %ymm0
- testl %edx, %edx
-
-/* Go to special inputs processing branch */
- jne L(SPECIAL_VALUES_BRANCH)
- # LOE rbx r12 r13 r14 r15 edx ymm0 ymm5
-
-/* Restore registers
- * and exit the function
- */
+ 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 1/(1+iRcpIndex/32+1/64) reciprocal table value */
+ lea __svml_dcbrt_data_internal(%rip), %rax
+ vmovapd %ymm0, %ymm5
+
+ /*
+ * Declarations
+ * Load constants
+ * Get iX - high part of argument
+ */
+ vextractf128 $1, %ymm5, %xmm6
+
+ /* Calculate CbrtIndex */
+ vpsrlq $52, %ymm5, %ymm15
+ vshufps $221, %xmm6, %xmm5, %xmm4
+
+ /* Calculate Rcp table index */
+ vandps _iRcpIndexMask+__svml_dcbrt_data_internal(%rip), %xmm4, %xmm10
+ vpsrld $12, %xmm10, %xmm3
+ vmovd %xmm3, %ecx
+
+ /* If the exponent field is zero - go to callout to process denormals */
+ vandps _iAbsMask+__svml_dcbrt_data_internal(%rip), %xmm4, %xmm7
+
+ /* Compute 2^k */
+ vpsrld $20, %xmm4, %xmm4
+ vpsubd _iSub+__svml_dcbrt_data_internal(%rip), %xmm7, %xmm8
+ vandps _lExpLoMask+__svml_dcbrt_data_internal(%rip), %ymm15, %ymm0
+ vpmuludq _l1556+__svml_dcbrt_data_internal(%rip), %ymm0, %ymm6
+ vpextrd $2, %xmm3, %edi
+ movslq %ecx, %rcx
+ vpextrd $1, %xmm3, %esi
+ movslq %edi, %rdi
+ vpextrd $3, %xmm3, %r8d
+ movslq %esi, %rsi
+ movslq %r8d, %r8
+ vpcmpgtd _iCmp+__svml_dcbrt_data_internal(%rip), %xmm8, %xmm9
+ vmovsd (%rax, %rcx), %xmm11
+ vmovmskps %xmm9, %edx
+ vmovsd (%rax, %rdi), %xmm13
+ vmovhpd (%rax, %rsi), %xmm11, %xmm12
+ vmovhpd (%rax, %r8), %xmm13, %xmm14
+ vextractf128 $1, %ymm6, %xmm7
+ vshufps $136, %xmm7, %xmm6, %xmm8
+ vmovups __VUNPACK_ODD_ind1.613.0.1(%rip), %ymm7
+ vextractf128 $1, %ymm0, %xmm1
+ vshufps $136, %xmm1, %xmm0, %xmm9
+ vpsrld $14, %xmm8, %xmm1
+ vpsubd %xmm1, %xmm9, %xmm10
+ vpaddd %xmm1, %xmm1, %xmm11
+
+ /*
+ * VAND( L, l2k, = l2k, lExpHiMask );
+ * Argument reduction Z
+ */
+ vandpd _dMantissaMask+__svml_dcbrt_data_internal(%rip), %ymm5, %ymm9
+ vinsertf128 $1, %xmm14, %ymm12, %ymm2
+ vpsubd %xmm11, %xmm10, %xmm12
+ vpslld $8, %xmm12, %xmm13
+ vpaddd %xmm13, %xmm3, %xmm15
+
+ /* Load cbrt(2^j*(1+iRcpIndex/32+1/64)) Hi & Lo values */
+ vmovd %xmm15, %r9d
+ vpextrd $2, %xmm15, %r11d
+ movslq %r9d, %r9
+ vpextrd $1, %xmm15, %r10d
+ movslq %r11d, %r11
+ vpextrd $3, %xmm15, %ecx
+ movslq %r10d, %r10
+ movslq %ecx, %rcx
+ vmovsd 256(%rax, %r9), %xmm3
+ vmovsd 256(%rax, %r11), %xmm0
+ vandpd _dSgnf6Mask+__svml_dcbrt_data_internal(%rip), %ymm5, %ymm10
+ vmovhpd 256(%rax, %r10), %xmm3, %xmm14
+ vmovhpd 256(%rax, %rcx), %xmm0, %xmm3
+ vorpd _dNegOne+__svml_dcbrt_data_internal(%rip), %ymm9, %ymm11
+ vorpd _dNeg65Div64+__svml_dcbrt_data_internal(%rip), %ymm10, %ymm12
+ vsubpd %ymm12, %ymm11, %ymm13
+ vmulpd %ymm13, %ymm2, %ymm2
+ vinsertf128 $1, %xmm3, %ymm14, %ymm0
+ vpand _iSignMask+__svml_dcbrt_data_internal(%rip), %xmm4, %xmm3
+ vpor _iBias+__svml_dcbrt_data_internal(%rip), %xmm3, %xmm4
+ vpaddd %xmm1, %xmm4, %xmm1
+ vpslld $20, %xmm1, %xmm6
+
+ /* Polynomial */
+ vmovupd _dA7+__svml_dcbrt_data_internal(%rip), %ymm1
+ vfmadd213pd _dA6+__svml_dcbrt_data_internal(%rip), %ymm2, %ymm1
+ vfmadd213pd _dA5+__svml_dcbrt_data_internal(%rip), %ymm2, %ymm1
+ vfmadd213pd _dA4+__svml_dcbrt_data_internal(%rip), %ymm2, %ymm1
+ vfmadd213pd _dA3+__svml_dcbrt_data_internal(%rip), %ymm2, %ymm1
+ vfmadd213pd _dA2+__svml_dcbrt_data_internal(%rip), %ymm2, %ymm1
+ vfmadd213pd _dA1+__svml_dcbrt_data_internal(%rip), %ymm2, %ymm1
+ vpermps %ymm6, %ymm7, %ymm8
+ vandps __VUNPACK_ODD_mask.613.0.1(%rip), %ymm8, %ymm14
+
+ /* THi*2^k, TLo*2^k */
+ vmulpd %ymm14, %ymm0, %ymm0
+
+ /* THi*2^k*Z */
+ vmulpd %ymm0, %ymm2, %ymm2
+
+ /* Final reconstruction */
+ vmulpd %ymm2, %ymm1, %ymm3
+ vaddpd %ymm3, %ymm0, %ymm0
+ testl %edx, %edx
+
+ /* Go to special inputs processing branch */
+ jne L(SPECIAL_VALUES_BRANCH)
+ # LOE rbx r12 r13 r14 r15 edx ymm0 ymm5
+
+ /* 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)
-
-/* Branch to process
- * special inputs
- */
+ 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
+ */
L(SPECIAL_VALUES_BRANCH):
- vmovupd %ymm5, 32(%rsp)
- vmovupd %ymm0, 64(%rsp)
- # LOE rbx r12 r13 r14 r15 edx ymm0
-
- xorl %eax, %eax
- # 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 %eax, %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 %edx, %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
- */
+ vmovupd %ymm5, 32(%rsp)
+ vmovupd %ymm0, 64(%rsp)
+ # LOE rbx r12 r13 r14 r15 edx ymm0
+
+ xorl %eax, %eax
+ # 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 %eax, %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 %edx, %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
+ */
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 $4, %r12d
-
-/* 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)
- vmovupd 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
-
-/* Scalar math fucntion call
- * to process special input
- */
+ incl %r12d
+ cmpl $4, %r12d
+
+ /* 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)
+ vmovupd 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
+
+ /* Scalar math fucntion call
+ * to process special input
+ */
L(SCALAR_MATH_CALL):
- movl %r12d, %r14d
- movsd 32(%rsp,%r14,8), %xmm0
- call cbrt@PLT
- # LOE rbx r14 r15 r12d r13d xmm0
+ movl %r12d, %r14d
+ movsd 32(%rsp, %r14, 8), %xmm0
+ call cbrt@PLT
+ # LOE rbx r14 r15 r12d r13d xmm0
- movsd %xmm0, 64(%rsp,%r14,8)
+ movsd %xmm0, 64(%rsp, %r14, 8)
-/* 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(_ZGVdN4v_cbrt_avx2)
- .section .rodata, "a"
- .align 32
+ .section .rodata, "a"
+ .align 32
__VUNPACK_ODD_ind1.613.0.1:
.rept 3
- .long 0
+ .long 0
.endr
- .long 1
- .long 0
- .long 2
- .long 0
- .long 3
- .align 32
+ .long 1
+ .long 0
+ .long 2
+ .long 0
+ .long 3
+ .align 32
__VUNPACK_ODD_mask.613.0.1:
- .long 0
- .long -1
- .long 0
- .long -1
- .long 0
- .long -1
- .long 0
- .long -1
-
- .section .rodata, "a"
- .align 32
+ .long 0
+ .long -1
+ .long 0
+ .long -1
+ .long 0
+ .long -1
+ .long 0
+ .long -1
+
+ .section .rodata, "a"
+ .align 32
#ifdef __svml_dcbrt_data_internal_typedef
typedef unsigned int VUINT32;
-typedef struct
-{
- __declspec(align(32)) VUINT32 _dRcp[32][2];
- __declspec(align(32)) VUINT32 _dCbrtHiLo[96][2];
- __declspec(align(32)) VUINT32 _dA7[4][2];
- __declspec(align(32)) VUINT32 _dA6[4][2];
- __declspec(align(32)) VUINT32 _dA5[4][2];
- __declspec(align(32)) VUINT32 _dA4[4][2];
- __declspec(align(32)) VUINT32 _dA3[4][2];
- __declspec(align(32)) VUINT32 _dA2[4][2];
- __declspec(align(32)) VUINT32 _dA1[4][2];
- __declspec(align(32)) VUINT32 _dNeg65Div64[4][2];
- __declspec(align(32)) VUINT32 _dSgnf6Mask[4][2];
- __declspec(align(32)) VUINT32 _dNegOne[4][2];
- __declspec(align(32)) VUINT32 _dMantissaMask[4][2];
- __declspec(align(32)) VUINT32 _lExpHiMask[4][2];
- __declspec(align(32)) VUINT32 _lExpLoMask[4][2];
- __declspec(align(32)) VUINT32 _l1556[4][2];
- __declspec(align(32)) VUINT32 _iRcpIndexMask[8][1];
- __declspec(align(32)) VUINT32 _iAbsMask[8][1];
- __declspec(align(32)) VUINT32 _iSignMask[8][1];
- __declspec(align(32)) VUINT32 _iBias[8][1];
- __declspec(align(32)) VUINT32 _iSub[8][1];
- __declspec(align(32)) VUINT32 _iCmp[8][1];
+typedef struct {
+ __declspec(align(32)) VUINT32 _dRcp[32][2];
+ __declspec(align(32)) VUINT32 _dCbrtHiLo[96][2];
+ __declspec(align(32)) VUINT32 _dA7[4][2];
+ __declspec(align(32)) VUINT32 _dA6[4][2];
+ __declspec(align(32)) VUINT32 _dA5[4][2];
+ __declspec(align(32)) VUINT32 _dA4[4][2];
+ __declspec(align(32)) VUINT32 _dA3[4][2];
+ __declspec(align(32)) VUINT32 _dA2[4][2];
+ __declspec(align(32)) VUINT32 _dA1[4][2];
+ __declspec(align(32)) VUINT32 _dNeg65Div64[4][2];
+ __declspec(align(32)) VUINT32 _dSgnf6Mask[4][2];
+ __declspec(align(32)) VUINT32 _dNegOne[4][2];
+ __declspec(align(32)) VUINT32 _dMantissaMask[4][2];
+ __declspec(align(32)) VUINT32 _lExpHiMask[4][2];
+ __declspec(align(32)) VUINT32 _lExpLoMask[4][2];
+ __declspec(align(32)) VUINT32 _l1556[4][2];
+ __declspec(align(32)) VUINT32 _iRcpIndexMask[8][1];
+ __declspec(align(32)) VUINT32 _iAbsMask[8][1];
+ __declspec(align(32)) VUINT32 _iSignMask[8][1];
+ __declspec(align(32)) VUINT32 _iBias[8][1];
+ __declspec(align(32)) VUINT32 _iSub[8][1];
+ __declspec(align(32)) VUINT32 _iCmp[8][1];
} __svml_dcbrt_data_internal;
#endif
__svml_dcbrt_data_internal:
- /*== _dRcp ==*/
- .quad 0xBFEF81F81F81F820 /* (1/(1+0/32+1/64)) = -.984615 */
- .quad 0xBFEE9131ABF0B767 /* (1/(1+1/32+1/64)) = -.955224 */
- .quad 0xBFEDAE6076B981DB /* (1/(1+2/32+1/64)) = -.927536 */
- .quad 0xBFECD85689039B0B /* (1/(1+3/32+1/64)) = -.901408 */
- .quad 0xBFEC0E070381C0E0 /* (1/(1+4/32+1/64)) = -.876712 */
- .quad 0xBFEB4E81B4E81B4F /* (1/(1+5/32+1/64)) = -.853333 */
- .quad 0xBFEA98EF606A63BE /* (1/(1+6/32+1/64)) = -.831169 */
- .quad 0xBFE9EC8E951033D9 /* (1/(1+7/32+1/64)) = -.810127 */
- .quad 0xBFE948B0FCD6E9E0 /* (1/(1+8/32+1/64)) = -.790123 */
- .quad 0xBFE8ACB90F6BF3AA /* (1/(1+9/32+1/64)) = -.771084 */
- .quad 0xBFE8181818181818 /* (1/(1+10/32+1/64)) = -.752941 */
- .quad 0xBFE78A4C8178A4C8 /* (1/(1+11/32+1/64)) = -.735632 */
- .quad 0xBFE702E05C0B8170 /* (1/(1+12/32+1/64)) = -.719101 */
- .quad 0xBFE6816816816817 /* (1/(1+13/32+1/64)) = -.703297 */
- .quad 0xBFE6058160581606 /* (1/(1+14/32+1/64)) = -.688172 */
- .quad 0xBFE58ED2308158ED /* (1/(1+15/32+1/64)) = -.673684 */
- .quad 0xBFE51D07EAE2F815 /* (1/(1+16/32+1/64)) = -.659794 */
- .quad 0xBFE4AFD6A052BF5B /* (1/(1+17/32+1/64)) = -.646465 */
- .quad 0xBFE446F86562D9FB /* (1/(1+18/32+1/64)) = -.633663 */
- .quad 0xBFE3E22CBCE4A902 /* (1/(1+19/32+1/64)) = -.621359 */
- .quad 0xBFE3813813813814 /* (1/(1+20/32+1/64)) = -.609524 */
- .quad 0xBFE323E34A2B10BF /* (1/(1+21/32+1/64)) = -.598131 */
- .quad 0xBFE2C9FB4D812CA0 /* (1/(1+22/32+1/64)) = -.587156 */
- .quad 0xBFE27350B8812735 /* (1/(1+23/32+1/64)) = -.576577 */
- .quad 0xBFE21FB78121FB78 /* (1/(1+24/32+1/64)) = -.566372 */
- .quad 0xBFE1CF06ADA2811D /* (1/(1+25/32+1/64)) = -.556522 */
- .quad 0xBFE1811811811812 /* (1/(1+26/32+1/64)) = -.547009 */
- .quad 0xBFE135C81135C811 /* (1/(1+27/32+1/64)) = -.537815 */
- .quad 0xBFE0ECF56BE69C90 /* (1/(1+28/32+1/64)) = -.528926 */
- .quad 0xBFE0A6810A6810A7 /* (1/(1+29/32+1/64)) = -.520325 */
- .quad 0xBFE0624DD2F1A9FC /* (1/(1+30/32+1/64)) = -.512 */
- .quad 0xBFE0204081020408 /* (1/(1+31/32+1/64)) = -.503937 */
- /*== _dCbrtHiLo ==*/
- .align 32
- .quad 0x3FF01539221D4C97 /* HI((2^0*(1+0/32+1/64))^(1/3)) = 1.005181 */
- .quad 0x3FF03F06771A2E33 /* HI((2^0*(1+1/32+1/64))^(1/3)) = 1.015387 */
- .quad 0x3FF06800E629D671 /* HI((2^0*(1+2/32+1/64))^(1/3)) = 1.025391 */
- .quad 0x3FF090328731DEB2 /* HI((2^0*(1+3/32+1/64))^(1/3)) = 1.035204 */
- .quad 0x3FF0B7A4B1BD64AC /* HI((2^0*(1+4/32+1/64))^(1/3)) = 1.044835 */
- .quad 0x3FF0DE601024FB87 /* HI((2^0*(1+5/32+1/64))^(1/3)) = 1.054291 */
- .quad 0x3FF1046CB0597000 /* HI((2^0*(1+6/32+1/64))^(1/3)) = 1.06358 */
- .quad 0x3FF129D212A9BA9B /* HI((2^0*(1+7/32+1/64))^(1/3)) = 1.07271 */
- .quad 0x3FF14E9736CDAF38 /* HI((2^0*(1+8/32+1/64))^(1/3)) = 1.081687 */
- .quad 0x3FF172C2A772F507 /* HI((2^0*(1+9/32+1/64))^(1/3)) = 1.090518 */
- .quad 0x3FF1965A848001D3 /* HI((2^0*(1+10/32+1/64))^(1/3)) = 1.099207 */
- .quad 0x3FF1B9648C38C55D /* HI((2^0*(1+11/32+1/64))^(1/3)) = 1.107762 */
- .quad 0x3FF1DBE6236A0C45 /* HI((2^0*(1+12/32+1/64))^(1/3)) = 1.116186 */
- .quad 0x3FF1FDE45CBB1F9F /* HI((2^0*(1+13/32+1/64))^(1/3)) = 1.124485 */
- .quad 0x3FF21F63FF409042 /* HI((2^0*(1+14/32+1/64))^(1/3)) = 1.132664 */
- .quad 0x3FF240698C6746E5 /* HI((2^0*(1+15/32+1/64))^(1/3)) = 1.140726 */
- .quad 0x3FF260F9454BB99B /* HI((2^0*(1+16/32+1/64))^(1/3)) = 1.148675 */
- .quad 0x3FF281172F8E7073 /* HI((2^0*(1+17/32+1/64))^(1/3)) = 1.156516 */
- .quad 0x3FF2A0C719B4B6D0 /* HI((2^0*(1+18/32+1/64))^(1/3)) = 1.164252 */
- .quad 0x3FF2C00C9F2263EC /* HI((2^0*(1+19/32+1/64))^(1/3)) = 1.171887 */
- .quad 0x3FF2DEEB2BB7FB78 /* HI((2^0*(1+20/32+1/64))^(1/3)) = 1.179423 */
- .quad 0x3FF2FD65FF1EFBBC /* HI((2^0*(1+21/32+1/64))^(1/3)) = 1.186865 */
- .quad 0x3FF31B802FCCF6A2 /* HI((2^0*(1+22/32+1/64))^(1/3)) = 1.194214 */
- .quad 0x3FF3393CADC50708 /* HI((2^0*(1+23/32+1/64))^(1/3)) = 1.201474 */
- .quad 0x3FF3569E451E4C2A /* HI((2^0*(1+24/32+1/64))^(1/3)) = 1.208647 */
- .quad 0x3FF373A7A0554CDE /* HI((2^0*(1+25/32+1/64))^(1/3)) = 1.215736 */
- .quad 0x3FF3905B4A6D76CE /* HI((2^0*(1+26/32+1/64))^(1/3)) = 1.222743 */
- .quad 0x3FF3ACBBB0E756B6 /* HI((2^0*(1+27/32+1/64))^(1/3)) = 1.229671 */
- .quad 0x3FF3C8CB258FA340 /* HI((2^0*(1+28/32+1/64))^(1/3)) = 1.236522 */
- .quad 0x3FF3E48BE02AC0CE /* HI((2^0*(1+29/32+1/64))^(1/3)) = 1.243297 */
- .quad 0x3FF4000000000000 /* HI((2^0*(1+30/32+1/64))^(1/3)) = 1.25 */
- .quad 0x3FF41B298D47800E /* HI((2^0*(1+31/32+1/64))^(1/3)) = 1.256631 */
- .quad 0x3FF443604B34D9B2 /* HI((2^1*(1+0/32+1/64))^(1/3)) = 1.266449 */
- .quad 0x3FF4780B20906571 /* HI((2^1*(1+1/32+1/64))^(1/3)) = 1.279307 */
- .quad 0x3FF4ABAC3EE06706 /* HI((2^1*(1+2/32+1/64))^(1/3)) = 1.291912 */
- .quad 0x3FF4DE505DA66B8D /* HI((2^1*(1+3/32+1/64))^(1/3)) = 1.304276 */
- .quad 0x3FF51003420A5C07 /* HI((2^1*(1+4/32+1/64))^(1/3)) = 1.316409 */
- .quad 0x3FF540CFD6FD11C1 /* HI((2^1*(1+5/32+1/64))^(1/3)) = 1.328323 */
- .quad 0x3FF570C04260716B /* HI((2^1*(1+6/32+1/64))^(1/3)) = 1.340027 */
- .quad 0x3FF59FDDF7A45F38 /* HI((2^1*(1+7/32+1/64))^(1/3)) = 1.35153 */
- .quad 0x3FF5CE31C83539DF /* HI((2^1*(1+8/32+1/64))^(1/3)) = 1.36284 */
- .quad 0x3FF5FBC3F20966A4 /* HI((2^1*(1+9/32+1/64))^(1/3)) = 1.373966 */
- .quad 0x3FF6289C2C8F1B70 /* HI((2^1*(1+10/32+1/64))^(1/3)) = 1.384915 */
- .quad 0x3FF654C1B4316DCF /* HI((2^1*(1+11/32+1/64))^(1/3)) = 1.395693 */
- .quad 0x3FF6803B54A34E44 /* HI((2^1*(1+12/32+1/64))^(1/3)) = 1.406307 */
- .quad 0x3FF6AB0F72182659 /* HI((2^1*(1+13/32+1/64))^(1/3)) = 1.416763 */
- .quad 0x3FF6D544118C08BC /* HI((2^1*(1+14/32+1/64))^(1/3)) = 1.427067 */
- .quad 0x3FF6FEDEE0388D4A /* HI((2^1*(1+15/32+1/64))^(1/3)) = 1.437224 */
- .quad 0x3FF727E53A4F645E /* HI((2^1*(1+16/32+1/64))^(1/3)) = 1.44724 */
- .quad 0x3FF7505C31104114 /* HI((2^1*(1+17/32+1/64))^(1/3)) = 1.457119 */
- .quad 0x3FF77848904CD549 /* HI((2^1*(1+18/32+1/64))^(1/3)) = 1.466866 */
- .quad 0x3FF79FAEE36B2534 /* HI((2^1*(1+19/32+1/64))^(1/3)) = 1.476485 */
- .quad 0x3FF7C69379F4605B /* HI((2^1*(1+20/32+1/64))^(1/3)) = 1.48598 */
- .quad 0x3FF7ECFA6BBCA391 /* HI((2^1*(1+21/32+1/64))^(1/3)) = 1.495356 */
- .quad 0x3FF812E79CAE7EB9 /* HI((2^1*(1+22/32+1/64))^(1/3)) = 1.504615 */
- .quad 0x3FF8385EC043C71D /* HI((2^1*(1+23/32+1/64))^(1/3)) = 1.513762 */
- .quad 0x3FF85D635CB41B9D /* HI((2^1*(1+24/32+1/64))^(1/3)) = 1.5228 */
- .quad 0x3FF881F8CDE083DB /* HI((2^1*(1+25/32+1/64))^(1/3)) = 1.531731 */
- .quad 0x3FF8A6224802B8A8 /* HI((2^1*(1+26/32+1/64))^(1/3)) = 1.54056 */
- .quad 0x3FF8C9E2DA25E5E4 /* HI((2^1*(1+27/32+1/64))^(1/3)) = 1.549289 */
- .quad 0x3FF8ED3D706E1010 /* HI((2^1*(1+28/32+1/64))^(1/3)) = 1.55792 */
- .quad 0x3FF91034D632B6DF /* HI((2^1*(1+29/32+1/64))^(1/3)) = 1.566457 */
- .quad 0x3FF932CBB7F0CF2D /* HI((2^1*(1+30/32+1/64))^(1/3)) = 1.574901 */
- .quad 0x3FF95504A517BF3A /* HI((2^1*(1+31/32+1/64))^(1/3)) = 1.583256 */
- .quad 0x3FF987AF34F8BB19 /* HI((2^2*(1+0/32+1/64))^(1/3)) = 1.595626 */
- .quad 0x3FF9CA0A8337B317 /* HI((2^2*(1+1/32+1/64))^(1/3)) = 1.611826 */
- .quad 0x3FFA0B1709CC13D5 /* HI((2^2*(1+2/32+1/64))^(1/3)) = 1.627708 */
- .quad 0x3FFA4AE4CE6419ED /* HI((2^2*(1+3/32+1/64))^(1/3)) = 1.643285 */
- .quad 0x3FFA8982A5567031 /* HI((2^2*(1+4/32+1/64))^(1/3)) = 1.658572 */
- .quad 0x3FFAC6FE500AB570 /* HI((2^2*(1+5/32+1/64))^(1/3)) = 1.673582 */
- .quad 0x3FFB036497A15A17 /* HI((2^2*(1+6/32+1/64))^(1/3)) = 1.688328 */
- .quad 0x3FFB3EC164671755 /* HI((2^2*(1+7/32+1/64))^(1/3)) = 1.702821 */
- .quad 0x3FFB791FD288C46F /* HI((2^2*(1+8/32+1/64))^(1/3)) = 1.717071 */
- .quad 0x3FFBB28A44693BE4 /* HI((2^2*(1+9/32+1/64))^(1/3)) = 1.731089 */
- .quad 0x3FFBEB0A72EB6E31 /* HI((2^2*(1+10/32+1/64))^(1/3)) = 1.744883 */
- .quad 0x3FFC22A97BF5F697 /* HI((2^2*(1+11/32+1/64))^(1/3)) = 1.758462 */
- .quad 0x3FFC596FEF6AF983 /* HI((2^2*(1+12/32+1/64))^(1/3)) = 1.771835 */
- .quad 0x3FFC8F65DAC655A3 /* HI((2^2*(1+13/32+1/64))^(1/3)) = 1.785009 */
- .quad 0x3FFCC492D38CE8D9 /* HI((2^2*(1+14/32+1/64))^(1/3)) = 1.797992 */
- .quad 0x3FFCF8FE00B19367 /* HI((2^2*(1+15/32+1/64))^(1/3)) = 1.810789 */
- .quad 0x3FFD2CAE230F8709 /* HI((2^2*(1+16/32+1/64))^(1/3)) = 1.823408 */
- .quad 0x3FFD5FA99D15208F /* HI((2^2*(1+17/32+1/64))^(1/3)) = 1.835855 */
- .quad 0x3FFD91F679B6E505 /* HI((2^2*(1+18/32+1/64))^(1/3)) = 1.848135 */
- .quad 0x3FFDC39A72BF2302 /* HI((2^2*(1+19/32+1/64))^(1/3)) = 1.860255 */
- .quad 0x3FFDF49AF68C1570 /* HI((2^2*(1+20/32+1/64))^(1/3)) = 1.872218 */
- .quad 0x3FFE24FD2D4C23B8 /* HI((2^2*(1+21/32+1/64))^(1/3)) = 1.884031 */
- .quad 0x3FFE54C5FDC5EC73 /* HI((2^2*(1+22/32+1/64))^(1/3)) = 1.895697 */
- .quad 0x3FFE83FA11B81DBB /* HI((2^2*(1+23/32+1/64))^(1/3)) = 1.907221 */
- .quad 0x3FFEB29DD9DBAF25 /* HI((2^2*(1+24/32+1/64))^(1/3)) = 1.918608 */
- .quad 0x3FFEE0B59191D374 /* HI((2^2*(1+25/32+1/64))^(1/3)) = 1.929861 */
- .quad 0x3FFF0E454245E4BF /* HI((2^2*(1+26/32+1/64))^(1/3)) = 1.940984 */
- .quad 0x3FFF3B50C68A9DD3 /* HI((2^2*(1+27/32+1/64))^(1/3)) = 1.951981 */
- .quad 0x3FFF67DBCCF922DC /* HI((2^2*(1+28/32+1/64))^(1/3)) = 1.962856 */
- .quad 0x3FFF93E9DAD7A4A6 /* HI((2^2*(1+29/32+1/64))^(1/3)) = 1.973612 */
- .quad 0x3FFFBF7E4E8CC9CB /* HI((2^2*(1+30/32+1/64))^(1/3)) = 1.984251 */
- .quad 0x3FFFEA9C61E47CD3 /* HI((2^2*(1+31/32+1/64))^(1/3)) = 1.994778 */
- .align 32
- .quad 0x3F93750AD588F115, 0x3F93750AD588F115, 0x3F93750AD588F115, 0x3F93750AD588F115 /* _dA7 */
- .align 32
- .quad 0xBF98090D6221A247, 0xBF98090D6221A247, 0xBF98090D6221A247, 0xBF98090D6221A247 /* _dA6 */
- .align 32
- .quad 0x3F9EE7113506AC12, 0x3F9EE7113506AC12, 0x3F9EE7113506AC12, 0x3F9EE7113506AC12 /* _dA5 */
- .align 32
- .quad 0xBFA511E8D2B3183B, 0xBFA511E8D2B3183B, 0xBFA511E8D2B3183B, 0xBFA511E8D2B3183B /* _dA4 */
- .align 32
- .quad 0x3FAF9ADD3C0CA458, 0x3FAF9ADD3C0CA458, 0x3FAF9ADD3C0CA458, 0x3FAF9ADD3C0CA458 /* _dA3 */
- .align 32
- .quad 0xBFBC71C71C71C71C, 0xBFBC71C71C71C71C, 0xBFBC71C71C71C71C, 0xBFBC71C71C71C71C /* _dA2 */
- .align 32
- .quad 0x3FD5555555555555, 0x3FD5555555555555, 0x3FD5555555555555, 0x3FD5555555555555 /* _dA1 */
- .align 32
- .quad 0xBFF0400000000000, 0xBFF0400000000000, 0xBFF0400000000000, 0xBFF0400000000000 /* _dNeg65Div64 */
- .align 32
- .quad 0x000FC00000000000, 0x000FC00000000000, 0x000FC00000000000, 0x000FC00000000000 /* _dSgnf6Mask */
- .align 32
- .quad 0xBFF0000000000000, 0xBFF0000000000000, 0xBFF0000000000000, 0xBFF0000000000000 /* _dNegOne */
- .align 32
- .quad 0x000FFFFFFFFFFFFF, 0x000FFFFFFFFFFFFF, 0x000FFFFFFFFFFFFF, 0x000FFFFFFFFFFFFF /* _dMantissaMask */
- .align 32
- .quad 0xFFF0000000000000, 0xFFF0000000000000, 0xFFF0000000000000, 0xFFF0000000000000 /* _lExpHiMask */
- .align 32
- .quad 0x00000000000007FF, 0x00000000000007FF, 0x00000000000007FF, 0x00000000000007FF /* _lExpLoMask */
- .align 32
- .quad 0x0000000000001556, 0x0000000000001556, 0x0000000000001556, 0x0000000000001556 /* _l1556 */
- .align 32
- .long 0x000F8000, 0x000F8000, 0x000F8000, 0x000F8000, 0x000F8000, 0x000F8000, 0x000F8000, 0x000F8000 /* _iRcpIndexMask */
- .align 32
- .long 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF /* _iAbsMask */
- .align 32
- .long 0x00000800, 0x00000800, 0x00000800, 0x00000800, 0x00000800, 0x00000800, 0x00000800, 0x00000800 /* _iSignMask */
- .align 32
- .long 0x000002AA, 0x000002AA, 0x000002AA, 0x000002AA, 0x000002AA, 0x000002AA, 0x000002AA, 0x000002AA /* _iBias */
- .align 32
- .long 0x80100000, 0x80100000, 0x80100000, 0x80100000, 0x80100000, 0x80100000, 0x80100000, 0x80100000 /* _iSub */
- .align 32
- .long 0xffdfffff, 0xffdfffff, 0xffdfffff, 0xffdfffff, 0xffdfffff, 0xffdfffff, 0xffdfffff, 0xffdfffff /* _iCmp */
- .align 32
- .type __svml_dcbrt_data_internal,@object
- .size __svml_dcbrt_data_internal,.-__svml_dcbrt_data_internal
+ /* _dRcp */
+ .quad 0xBFEF81F81F81F820 /* (1/(1+0/32+1/64)) = -.984615 */
+ .quad 0xBFEE9131ABF0B767 /* (1/(1+1/32+1/64)) = -.955224 */
+ .quad 0xBFEDAE6076B981DB /* (1/(1+2/32+1/64)) = -.927536 */
+ .quad 0xBFECD85689039B0B /* (1/(1+3/32+1/64)) = -.901408 */
+ .quad 0xBFEC0E070381C0E0 /* (1/(1+4/32+1/64)) = -.876712 */
+ .quad 0xBFEB4E81B4E81B4F /* (1/(1+5/32+1/64)) = -.853333 */
+ .quad 0xBFEA98EF606A63BE /* (1/(1+6/32+1/64)) = -.831169 */
+ .quad 0xBFE9EC8E951033D9 /* (1/(1+7/32+1/64)) = -.810127 */
+ .quad 0xBFE948B0FCD6E9E0 /* (1/(1+8/32+1/64)) = -.790123 */
+ .quad 0xBFE8ACB90F6BF3AA /* (1/(1+9/32+1/64)) = -.771084 */
+ .quad 0xBFE8181818181818 /* (1/(1+10/32+1/64)) = -.752941 */
+ .quad 0xBFE78A4C8178A4C8 /* (1/(1+11/32+1/64)) = -.735632 */
+ .quad 0xBFE702E05C0B8170 /* (1/(1+12/32+1/64)) = -.719101 */
+ .quad 0xBFE6816816816817 /* (1/(1+13/32+1/64)) = -.703297 */
+ .quad 0xBFE6058160581606 /* (1/(1+14/32+1/64)) = -.688172 */
+ .quad 0xBFE58ED2308158ED /* (1/(1+15/32+1/64)) = -.673684 */
+ .quad 0xBFE51D07EAE2F815 /* (1/(1+16/32+1/64)) = -.659794 */
+ .quad 0xBFE4AFD6A052BF5B /* (1/(1+17/32+1/64)) = -.646465 */
+ .quad 0xBFE446F86562D9FB /* (1/(1+18/32+1/64)) = -.633663 */
+ .quad 0xBFE3E22CBCE4A902 /* (1/(1+19/32+1/64)) = -.621359 */
+ .quad 0xBFE3813813813814 /* (1/(1+20/32+1/64)) = -.609524 */
+ .quad 0xBFE323E34A2B10BF /* (1/(1+21/32+1/64)) = -.598131 */
+ .quad 0xBFE2C9FB4D812CA0 /* (1/(1+22/32+1/64)) = -.587156 */
+ .quad 0xBFE27350B8812735 /* (1/(1+23/32+1/64)) = -.576577 */
+ .quad 0xBFE21FB78121FB78 /* (1/(1+24/32+1/64)) = -.566372 */
+ .quad 0xBFE1CF06ADA2811D /* (1/(1+25/32+1/64)) = -.556522 */
+ .quad 0xBFE1811811811812 /* (1/(1+26/32+1/64)) = -.547009 */
+ .quad 0xBFE135C81135C811 /* (1/(1+27/32+1/64)) = -.537815 */
+ .quad 0xBFE0ECF56BE69C90 /* (1/(1+28/32+1/64)) = -.528926 */
+ .quad 0xBFE0A6810A6810A7 /* (1/(1+29/32+1/64)) = -.520325 */
+ .quad 0xBFE0624DD2F1A9FC /* (1/(1+30/32+1/64)) = -.512 */
+ .quad 0xBFE0204081020408 /* (1/(1+31/32+1/64)) = -.503937 */
+ /* _dCbrtHiLo */
+ .align 32
+ .quad 0x3FF01539221D4C97 /* HI((2^0*(1+0/32+1/64))^(1/3)) = 1.005181 */
+ .quad 0x3FF03F06771A2E33 /* HI((2^0*(1+1/32+1/64))^(1/3)) = 1.015387 */
+ .quad 0x3FF06800E629D671 /* HI((2^0*(1+2/32+1/64))^(1/3)) = 1.025391 */
+ .quad 0x3FF090328731DEB2 /* HI((2^0*(1+3/32+1/64))^(1/3)) = 1.035204 */
+ .quad 0x3FF0B7A4B1BD64AC /* HI((2^0*(1+4/32+1/64))^(1/3)) = 1.044835 */
+ .quad 0x3FF0DE601024FB87 /* HI((2^0*(1+5/32+1/64))^(1/3)) = 1.054291 */
+ .quad 0x3FF1046CB0597000 /* HI((2^0*(1+6/32+1/64))^(1/3)) = 1.06358 */
+ .quad 0x3FF129D212A9BA9B /* HI((2^0*(1+7/32+1/64))^(1/3)) = 1.07271 */
+ .quad 0x3FF14E9736CDAF38 /* HI((2^0*(1+8/32+1/64))^(1/3)) = 1.081687 */
+ .quad 0x3FF172C2A772F507 /* HI((2^0*(1+9/32+1/64))^(1/3)) = 1.090518 */
+ .quad 0x3FF1965A848001D3 /* HI((2^0*(1+10/32+1/64))^(1/3)) = 1.099207 */
+ .quad 0x3FF1B9648C38C55D /* HI((2^0*(1+11/32+1/64))^(1/3)) = 1.107762 */
+ .quad 0x3FF1DBE6236A0C45 /* HI((2^0*(1+12/32+1/64))^(1/3)) = 1.116186 */
+ .quad 0x3FF1FDE45CBB1F9F /* HI((2^0*(1+13/32+1/64))^(1/3)) = 1.124485 */
+ .quad 0x3FF21F63FF409042 /* HI((2^0*(1+14/32+1/64))^(1/3)) = 1.132664 */
+ .quad 0x3FF240698C6746E5 /* HI((2^0*(1+15/32+1/64))^(1/3)) = 1.140726 */
+ .quad 0x3FF260F9454BB99B /* HI((2^0*(1+16/32+1/64))^(1/3)) = 1.148675 */
+ .quad 0x3FF281172F8E7073 /* HI((2^0*(1+17/32+1/64))^(1/3)) = 1.156516 */
+ .quad 0x3FF2A0C719B4B6D0 /* HI((2^0*(1+18/32+1/64))^(1/3)) = 1.164252 */
+ .quad 0x3FF2C00C9F2263EC /* HI((2^0*(1+19/32+1/64))^(1/3)) = 1.171887 */
+ .quad 0x3FF2DEEB2BB7FB78 /* HI((2^0*(1+20/32+1/64))^(1/3)) = 1.179423 */
+ .quad 0x3FF2FD65FF1EFBBC /* HI((2^0*(1+21/32+1/64))^(1/3)) = 1.186865 */
+ .quad 0x3FF31B802FCCF6A2 /* HI((2^0*(1+22/32+1/64))^(1/3)) = 1.194214 */
+ .quad 0x3FF3393CADC50708 /* HI((2^0*(1+23/32+1/64))^(1/3)) = 1.201474 */
+ .quad 0x3FF3569E451E4C2A /* HI((2^0*(1+24/32+1/64))^(1/3)) = 1.208647 */
+ .quad 0x3FF373A7A0554CDE /* HI((2^0*(1+25/32+1/64))^(1/3)) = 1.215736 */
+ .quad 0x3FF3905B4A6D76CE /* HI((2^0*(1+26/32+1/64))^(1/3)) = 1.222743 */
+ .quad 0x3FF3ACBBB0E756B6 /* HI((2^0*(1+27/32+1/64))^(1/3)) = 1.229671 */
+ .quad 0x3FF3C8CB258FA340 /* HI((2^0*(1+28/32+1/64))^(1/3)) = 1.236522 */
+ .quad 0x3FF3E48BE02AC0CE /* HI((2^0*(1+29/32+1/64))^(1/3)) = 1.243297 */
+ .quad 0x3FF4000000000000 /* HI((2^0*(1+30/32+1/64))^(1/3)) = 1.25 */
+ .quad 0x3FF41B298D47800E /* HI((2^0*(1+31/32+1/64))^(1/3)) = 1.256631 */
+ .quad 0x3FF443604B34D9B2 /* HI((2^1*(1+0/32+1/64))^(1/3)) = 1.266449 */
+ .quad 0x3FF4780B20906571 /* HI((2^1*(1+1/32+1/64))^(1/3)) = 1.279307 */
+ .quad 0x3FF4ABAC3EE06706 /* HI((2^1*(1+2/32+1/64))^(1/3)) = 1.291912 */
+ .quad 0x3FF4DE505DA66B8D /* HI((2^1*(1+3/32+1/64))^(1/3)) = 1.304276 */
+ .quad 0x3FF51003420A5C07 /* HI((2^1*(1+4/32+1/64))^(1/3)) = 1.316409 */
+ .quad 0x3FF540CFD6FD11C1 /* HI((2^1*(1+5/32+1/64))^(1/3)) = 1.328323 */
+ .quad 0x3FF570C04260716B /* HI((2^1*(1+6/32+1/64))^(1/3)) = 1.340027 */
+ .quad 0x3FF59FDDF7A45F38 /* HI((2^1*(1+7/32+1/64))^(1/3)) = 1.35153 */
+ .quad 0x3FF5CE31C83539DF /* HI((2^1*(1+8/32+1/64))^(1/3)) = 1.36284 */
+ .quad 0x3FF5FBC3F20966A4 /* HI((2^1*(1+9/32+1/64))^(1/3)) = 1.373966 */
+ .quad 0x3FF6289C2C8F1B70 /* HI((2^1*(1+10/32+1/64))^(1/3)) = 1.384915 */
+ .quad 0x3FF654C1B4316DCF /* HI((2^1*(1+11/32+1/64))^(1/3)) = 1.395693 */
+ .quad 0x3FF6803B54A34E44 /* HI((2^1*(1+12/32+1/64))^(1/3)) = 1.406307 */
+ .quad 0x3FF6AB0F72182659 /* HI((2^1*(1+13/32+1/64))^(1/3)) = 1.416763 */
+ .quad 0x3FF6D544118C08BC /* HI((2^1*(1+14/32+1/64))^(1/3)) = 1.427067 */
+ .quad 0x3FF6FEDEE0388D4A /* HI((2^1*(1+15/32+1/64))^(1/3)) = 1.437224 */
+ .quad 0x3FF727E53A4F645E /* HI((2^1*(1+16/32+1/64))^(1/3)) = 1.44724 */
+ .quad 0x3FF7505C31104114 /* HI((2^1*(1+17/32+1/64))^(1/3)) = 1.457119 */
+ .quad 0x3FF77848904CD549 /* HI((2^1*(1+18/32+1/64))^(1/3)) = 1.466866 */
+ .quad 0x3FF79FAEE36B2534 /* HI((2^1*(1+19/32+1/64))^(1/3)) = 1.476485 */
+ .quad 0x3FF7C69379F4605B /* HI((2^1*(1+20/32+1/64))^(1/3)) = 1.48598 */
+ .quad 0x3FF7ECFA6BBCA391 /* HI((2^1*(1+21/32+1/64))^(1/3)) = 1.495356 */
+ .quad 0x3FF812E79CAE7EB9 /* HI((2^1*(1+22/32+1/64))^(1/3)) = 1.504615 */
+ .quad 0x3FF8385EC043C71D /* HI((2^1*(1+23/32+1/64))^(1/3)) = 1.513762 */
+ .quad 0x3FF85D635CB41B9D /* HI((2^1*(1+24/32+1/64))^(1/3)) = 1.5228 */
+ .quad 0x3FF881F8CDE083DB /* HI((2^1*(1+25/32+1/64))^(1/3)) = 1.531731 */
+ .quad 0x3FF8A6224802B8A8 /* HI((2^1*(1+26/32+1/64))^(1/3)) = 1.54056 */
+ .quad 0x3FF8C9E2DA25E5E4 /* HI((2^1*(1+27/32+1/64))^(1/3)) = 1.549289 */
+ .quad 0x3FF8ED3D706E1010 /* HI((2^1*(1+28/32+1/64))^(1/3)) = 1.55792 */
+ .quad 0x3FF91034D632B6DF /* HI((2^1*(1+29/32+1/64))^(1/3)) = 1.566457 */
+ .quad 0x3FF932CBB7F0CF2D /* HI((2^1*(1+30/32+1/64))^(1/3)) = 1.574901 */
+ .quad 0x3FF95504A517BF3A /* HI((2^1*(1+31/32+1/64))^(1/3)) = 1.583256 */
+ .quad 0x3FF987AF34F8BB19 /* HI((2^2*(1+0/32+1/64))^(1/3)) = 1.595626 */
+ .quad 0x3FF9CA0A8337B317 /* HI((2^2*(1+1/32+1/64))^(1/3)) = 1.611826 */
+ .quad 0x3FFA0B1709CC13D5 /* HI((2^2*(1+2/32+1/64))^(1/3)) = 1.627708 */
+ .quad 0x3FFA4AE4CE6419ED /* HI((2^2*(1+3/32+1/64))^(1/3)) = 1.643285 */
+ .quad 0x3FFA8982A5567031 /* HI((2^2*(1+4/32+1/64))^(1/3)) = 1.658572 */
+ .quad 0x3FFAC6FE500AB570 /* HI((2^2*(1+5/32+1/64))^(1/3)) = 1.673582 */
+ .quad 0x3FFB036497A15A17 /* HI((2^2*(1+6/32+1/64))^(1/3)) = 1.688328 */
+ .quad 0x3FFB3EC164671755 /* HI((2^2*(1+7/32+1/64))^(1/3)) = 1.702821 */
+ .quad 0x3FFB791FD288C46F /* HI((2^2*(1+8/32+1/64))^(1/3)) = 1.717071 */
+ .quad 0x3FFBB28A44693BE4 /* HI((2^2*(1+9/32+1/64))^(1/3)) = 1.731089 */
+ .quad 0x3FFBEB0A72EB6E31 /* HI((2^2*(1+10/32+1/64))^(1/3)) = 1.744883 */
+ .quad 0x3FFC22A97BF5F697 /* HI((2^2*(1+11/32+1/64))^(1/3)) = 1.758462 */
+ .quad 0x3FFC596FEF6AF983 /* HI((2^2*(1+12/32+1/64))^(1/3)) = 1.771835 */
+ .quad 0x3FFC8F65DAC655A3 /* HI((2^2*(1+13/32+1/64))^(1/3)) = 1.785009 */
+ .quad 0x3FFCC492D38CE8D9 /* HI((2^2*(1+14/32+1/64))^(1/3)) = 1.797992 */
+ .quad 0x3FFCF8FE00B19367 /* HI((2^2*(1+15/32+1/64))^(1/3)) = 1.810789 */
+ .quad 0x3FFD2CAE230F8709 /* HI((2^2*(1+16/32+1/64))^(1/3)) = 1.823408 */
+ .quad 0x3FFD5FA99D15208F /* HI((2^2*(1+17/32+1/64))^(1/3)) = 1.835855 */
+ .quad 0x3FFD91F679B6E505 /* HI((2^2*(1+18/32+1/64))^(1/3)) = 1.848135 */
+ .quad 0x3FFDC39A72BF2302 /* HI((2^2*(1+19/32+1/64))^(1/3)) = 1.860255 */
+ .quad 0x3FFDF49AF68C1570 /* HI((2^2*(1+20/32+1/64))^(1/3)) = 1.872218 */
+ .quad 0x3FFE24FD2D4C23B8 /* HI((2^2*(1+21/32+1/64))^(1/3)) = 1.884031 */
+ .quad 0x3FFE54C5FDC5EC73 /* HI((2^2*(1+22/32+1/64))^(1/3)) = 1.895697 */
+ .quad 0x3FFE83FA11B81DBB /* HI((2^2*(1+23/32+1/64))^(1/3)) = 1.907221 */
+ .quad 0x3FFEB29DD9DBAF25 /* HI((2^2*(1+24/32+1/64))^(1/3)) = 1.918608 */
+ .quad 0x3FFEE0B59191D374 /* HI((2^2*(1+25/32+1/64))^(1/3)) = 1.929861 */
+ .quad 0x3FFF0E454245E4BF /* HI((2^2*(1+26/32+1/64))^(1/3)) = 1.940984 */
+ .quad 0x3FFF3B50C68A9DD3 /* HI((2^2*(1+27/32+1/64))^(1/3)) = 1.951981 */
+ .quad 0x3FFF67DBCCF922DC /* HI((2^2*(1+28/32+1/64))^(1/3)) = 1.962856 */
+ .quad 0x3FFF93E9DAD7A4A6 /* HI((2^2*(1+29/32+1/64))^(1/3)) = 1.973612 */
+ .quad 0x3FFFBF7E4E8CC9CB /* HI((2^2*(1+30/32+1/64))^(1/3)) = 1.984251 */
+ .quad 0x3FFFEA9C61E47CD3 /* HI((2^2*(1+31/32+1/64))^(1/3)) = 1.994778 */
+ .align 32
+ .quad 0x3F93750AD588F115, 0x3F93750AD588F115, 0x3F93750AD588F115, 0x3F93750AD588F115 /* _dA7 */
+ .align 32
+ .quad 0xBF98090D6221A247, 0xBF98090D6221A247, 0xBF98090D6221A247, 0xBF98090D6221A247 /* _dA6 */
+ .align 32
+ .quad 0x3F9EE7113506AC12, 0x3F9EE7113506AC12, 0x3F9EE7113506AC12, 0x3F9EE7113506AC12 /* _dA5 */
+ .align 32
+ .quad 0xBFA511E8D2B3183B, 0xBFA511E8D2B3183B, 0xBFA511E8D2B3183B, 0xBFA511E8D2B3183B /* _dA4 */
+ .align 32
+ .quad 0x3FAF9ADD3C0CA458, 0x3FAF9ADD3C0CA458, 0x3FAF9ADD3C0CA458, 0x3FAF9ADD3C0CA458 /* _dA3 */
+ .align 32
+ .quad 0xBFBC71C71C71C71C, 0xBFBC71C71C71C71C, 0xBFBC71C71C71C71C, 0xBFBC71C71C71C71C /* _dA2 */
+ .align 32
+ .quad 0x3FD5555555555555, 0x3FD5555555555555, 0x3FD5555555555555, 0x3FD5555555555555 /* _dA1 */
+ .align 32
+ .quad 0xBFF0400000000000, 0xBFF0400000000000, 0xBFF0400000000000, 0xBFF0400000000000 /* _dNeg65Div64 */
+ .align 32
+ .quad 0x000FC00000000000, 0x000FC00000000000, 0x000FC00000000000, 0x000FC00000000000 /* _dSgnf6Mask */
+ .align 32
+ .quad 0xBFF0000000000000, 0xBFF0000000000000, 0xBFF0000000000000, 0xBFF0000000000000 /* _dNegOne */
+ .align 32
+ .quad 0x000FFFFFFFFFFFFF, 0x000FFFFFFFFFFFFF, 0x000FFFFFFFFFFFFF, 0x000FFFFFFFFFFFFF /* _dMantissaMask */
+ .align 32
+ .quad 0xFFF0000000000000, 0xFFF0000000000000, 0xFFF0000000000000, 0xFFF0000000000000 /* _lExpHiMask */
+ .align 32
+ .quad 0x00000000000007FF, 0x00000000000007FF, 0x00000000000007FF, 0x00000000000007FF /* _lExpLoMask */
+ .align 32
+ .quad 0x0000000000001556, 0x0000000000001556, 0x0000000000001556, 0x0000000000001556 /* _l1556 */
+ .align 32
+ .long 0x000F8000, 0x000F8000, 0x000F8000, 0x000F8000, 0x000F8000, 0x000F8000, 0x000F8000, 0x000F8000 /* _iRcpIndexMask */
+ .align 32
+ .long 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF /* _iAbsMask */
+ .align 32
+ .long 0x00000800, 0x00000800, 0x00000800, 0x00000800, 0x00000800, 0x00000800, 0x00000800, 0x00000800 /* _iSignMask */
+ .align 32
+ .long 0x000002AA, 0x000002AA, 0x000002AA, 0x000002AA, 0x000002AA, 0x000002AA, 0x000002AA, 0x000002AA /* _iBias */
+ .align 32
+ .long 0x80100000, 0x80100000, 0x80100000, 0x80100000, 0x80100000, 0x80100000, 0x80100000, 0x80100000 /* _iSub */
+ .align 32
+ .long 0xffdfffff, 0xffdfffff, 0xffdfffff, 0xffdfffff, 0xffdfffff, 0xffdfffff, 0xffdfffff, 0xffdfffff /* _iCmp */
+ .align 32
+ .type __svml_dcbrt_data_internal, @object
+ .size __svml_dcbrt_data_internal, .-__svml_dcbrt_data_internal
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-17 07:15:59 +0000