diff options
| -rw-r--r-- | sysdeps/x86_64/fpu/multiarch/svml_d_expm14_core_avx2.S | 701 |
1 files changed, 350 insertions, 351 deletions
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_expm14_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_expm14_core_avx2.S index ac05a2af06..590341c243 100644 --- a/sysdeps/x86_64/fpu/multiarch/svml_d_expm14_core_avx2.S +++ b/sysdeps/x86_64/fpu/multiarch/svml_d_expm14_core_avx2.S @@ -28,381 +28,380 @@ /* Offsets for data table __svml_dexpm1_data_internal */ -#define Expm1_HA_table 0 -#define poly_coeff 2048 -#define Log2e 2176 -#define L2H 2208 -#define L2L 2240 -#define ExpAddConst 2272 -#define IndexMask 2304 -#define ExpMask 2336 -#define MOne 2368 -#define AbsMask 2400 -#define Threshold 2432 -#define L2 2464 +#define Expm1_HA_table 0 +#define poly_coeff 2048 +#define Log2e 2176 +#define L2H 2208 +#define L2L 2240 +#define ExpAddConst 2272 +#define IndexMask 2304 +#define ExpMask 2336 +#define MOne 2368 +#define AbsMask 2400 +#define Threshold 2432 +#define L2 2464 #include <sysdep.h> - .text - .section .text.avx2,"ax",@progbits + .section .text.avx2, "ax", @progbits ENTRY(_ZGVdN4v_expm1_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 - lea __svml_dexpm1_data_internal(%rip), %r8 - vmovapd %ymm0, %ymm3 - vmulpd Log2e+__svml_dexpm1_data_internal(%rip), %ymm3, %ymm4 - -/* argument reduction */ - vmovupd L2H+__svml_dexpm1_data_internal(%rip), %ymm2 - vmovupd AbsMask+__svml_dexpm1_data_internal(%rip), %ymm5 - vroundpd $0, %ymm4, %ymm8 - vaddpd ExpAddConst+__svml_dexpm1_data_internal(%rip), %ymm8, %ymm0 - vfnmadd213pd %ymm3, %ymm8, %ymm2 - -/* table lookup */ - vandps IndexMask+__svml_dexpm1_data_internal(%rip), %ymm0, %ymm9 - vandpd %ymm5, %ymm3, %ymm6 - vcmpnle_uqpd Threshold+__svml_dexpm1_data_internal(%rip), %ymm6, %ymm7 - vfnmadd231pd L2L+__svml_dexpm1_data_internal(%rip), %ymm8, %ymm2 - vandnpd %ymm3, %ymm5, %ymm1 - vmovmskpd %ymm7, %eax - vmovupd poly_coeff+64+__svml_dexpm1_data_internal(%rip), %ymm7 - vmulpd %ymm2, %ymm2, %ymm8 - vfmadd213pd poly_coeff+96+__svml_dexpm1_data_internal(%rip), %ymm2, %ymm7 - vandps ExpMask+__svml_dexpm1_data_internal(%rip), %ymm0, %ymm0 - vextractf128 $1, %ymm9, %xmm10 - vmovd %xmm9, %edx - vmovd %xmm10, %esi - vpextrd $2, %xmm9, %ecx - vpextrd $2, %xmm10, %edi - movslq %edx, %rdx - movslq %ecx, %rcx - movslq %esi, %rsi - movslq %edi, %rdi - vmovupd (%r8,%rdx), %xmm13 - vmovupd (%r8,%rcx), %xmm14 - vmovupd (%r8,%rsi), %xmm4 - vmovupd (%r8,%rdi), %xmm5 - vunpcklpd %xmm14, %xmm13, %xmm11 - vunpcklpd %xmm5, %xmm4, %xmm12 - vpsllq $41, %ymm0, %ymm10 - vunpckhpd %xmm14, %xmm13, %xmm15 - vunpckhpd %xmm5, %xmm4, %xmm13 - vinsertf128 $1, %xmm12, %ymm11, %ymm6 - -/* polynomial */ - vmovupd poly_coeff+__svml_dexpm1_data_internal(%rip), %ymm12 - -/* T-1 */ - vmovupd MOne+__svml_dexpm1_data_internal(%rip), %ymm11 - vfmadd213pd poly_coeff+32+__svml_dexpm1_data_internal(%rip), %ymm2, %ymm12 - vfmadd213pd %ymm7, %ymm8, %ymm12 - vorpd %ymm10, %ymm6, %ymm9 - vfmadd213pd %ymm2, %ymm8, %ymm12 - vaddpd %ymm11, %ymm9, %ymm2 - vinsertf128 $1, %xmm13, %ymm15, %ymm14 - -/* Th1 = (Th-1) + Tl */ - vfmadd213pd %ymm2, %ymm10, %ymm14 - -/* T = Th+Tl */ - vsubpd %ymm11, %ymm14, %ymm0 - vfmadd213pd %ymm14, %ymm12, %ymm0 - vorpd %ymm1, %ymm0, %ymm0 - testl %eax, %eax - -/* Go to special inputs processing branch */ - jne L(SPECIAL_VALUES_BRANCH) - # LOE rbx r12 r13 r14 r15 eax ymm0 ymm3 - -/* 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 + lea __svml_dexpm1_data_internal(%rip), %r8 + vmovapd %ymm0, %ymm3 + vmulpd Log2e+__svml_dexpm1_data_internal(%rip), %ymm3, %ymm4 + + /* argument reduction */ + vmovupd L2H+__svml_dexpm1_data_internal(%rip), %ymm2 + vmovupd AbsMask+__svml_dexpm1_data_internal(%rip), %ymm5 + vroundpd $0, %ymm4, %ymm8 + vaddpd ExpAddConst+__svml_dexpm1_data_internal(%rip), %ymm8, %ymm0 + vfnmadd213pd %ymm3, %ymm8, %ymm2 + + /* table lookup */ + vandps IndexMask+__svml_dexpm1_data_internal(%rip), %ymm0, %ymm9 + vandpd %ymm5, %ymm3, %ymm6 + vcmpnle_uqpd Threshold+__svml_dexpm1_data_internal(%rip), %ymm6, %ymm7 + vfnmadd231pd L2L+__svml_dexpm1_data_internal(%rip), %ymm8, %ymm2 + vandnpd %ymm3, %ymm5, %ymm1 + vmovmskpd %ymm7, %eax + vmovupd poly_coeff+64+__svml_dexpm1_data_internal(%rip), %ymm7 + vmulpd %ymm2, %ymm2, %ymm8 + vfmadd213pd poly_coeff+96+__svml_dexpm1_data_internal(%rip), %ymm2, %ymm7 + vandps ExpMask+__svml_dexpm1_data_internal(%rip), %ymm0, %ymm0 + vextractf128 $1, %ymm9, %xmm10 + vmovd %xmm9, %edx + vmovd %xmm10, %esi + vpextrd $2, %xmm9, %ecx + vpextrd $2, %xmm10, %edi + movslq %edx, %rdx + movslq %ecx, %rcx + movslq %esi, %rsi + movslq %edi, %rdi + vmovupd (%r8, %rdx), %xmm13 + vmovupd (%r8, %rcx), %xmm14 + vmovupd (%r8, %rsi), %xmm4 + vmovupd (%r8, %rdi), %xmm5 + vunpcklpd %xmm14, %xmm13, %xmm11 + vunpcklpd %xmm5, %xmm4, %xmm12 + vpsllq $41, %ymm0, %ymm10 + vunpckhpd %xmm14, %xmm13, %xmm15 + vunpckhpd %xmm5, %xmm4, %xmm13 + vinsertf128 $1, %xmm12, %ymm11, %ymm6 + + /* polynomial */ + vmovupd poly_coeff+__svml_dexpm1_data_internal(%rip), %ymm12 + + /* T-1 */ + vmovupd MOne+__svml_dexpm1_data_internal(%rip), %ymm11 + vfmadd213pd poly_coeff+32+__svml_dexpm1_data_internal(%rip), %ymm2, %ymm12 + vfmadd213pd %ymm7, %ymm8, %ymm12 + vorpd %ymm10, %ymm6, %ymm9 + vfmadd213pd %ymm2, %ymm8, %ymm12 + vaddpd %ymm11, %ymm9, %ymm2 + vinsertf128 $1, %xmm13, %ymm15, %ymm14 + + /* Th1 = (Th-1) + Tl */ + vfmadd213pd %ymm2, %ymm10, %ymm14 + + /* T = Th+Tl */ + vsubpd %ymm11, %ymm14, %ymm0 + vfmadd213pd %ymm14, %ymm12, %ymm0 + vorpd %ymm1, %ymm0, %ymm0 + testl %eax, %eax + + /* Go to special inputs processing branch */ + jne L(SPECIAL_VALUES_BRANCH) + # LOE rbx r12 r13 r14 r15 eax ymm0 ymm3 + + /* 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 %ymm3, 32(%rsp) - vmovupd %ymm0, 64(%rsp) - # LOE rbx r12 r13 r14 r15 eax ymm0 - - 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 - -/* Range mask - * bits check - */ + vmovupd %ymm3, 32(%rsp) + vmovupd %ymm0, 64(%rsp) + # LOE rbx r12 r13 r14 r15 eax ymm0 + + 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 + + /* 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 expm1@PLT - # LOE rbx r14 r15 r12d r13d xmm0 + movl %r12d, %r14d + movsd 32(%rsp, %r14, 8), %xmm0 + call expm1@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_expm1_avx2) - .section .rodata, "a" - .align 32 + .section .rodata, "a" + .align 32 #ifdef __svml_dexpm1_data_internal_typedef typedef unsigned int VUINT32; typedef struct { - __declspec(align(32)) VUINT32 Expm1_HA_table[(1<<8)][2]; - __declspec(align(32)) VUINT32 poly_coeff[4][4][2]; - __declspec(align(32)) VUINT32 Log2e[4][2]; - __declspec(align(32)) VUINT32 L2H[4][2]; - __declspec(align(32)) VUINT32 L2L[4][2]; - __declspec(align(32)) VUINT32 ExpAddConst[4][2]; - __declspec(align(32)) VUINT32 IndexMask[4][2]; - __declspec(align(32)) VUINT32 ExpMask[4][2]; - __declspec(align(32)) VUINT32 MOne[4][2]; - __declspec(align(32)) VUINT32 AbsMask[4][2]; - __declspec(align(32)) VUINT32 Threshold[4][2]; - __declspec(align(32)) VUINT32 L2[4][2]; + __declspec(align(32)) VUINT32 Expm1_HA_table[(1<<8)][2]; + __declspec(align(32)) VUINT32 poly_coeff[4][4][2]; + __declspec(align(32)) VUINT32 Log2e[4][2]; + __declspec(align(32)) VUINT32 L2H[4][2]; + __declspec(align(32)) VUINT32 L2L[4][2]; + __declspec(align(32)) VUINT32 ExpAddConst[4][2]; + __declspec(align(32)) VUINT32 IndexMask[4][2]; + __declspec(align(32)) VUINT32 ExpMask[4][2]; + __declspec(align(32)) VUINT32 MOne[4][2]; + __declspec(align(32)) VUINT32 AbsMask[4][2]; + __declspec(align(32)) VUINT32 Threshold[4][2]; + __declspec(align(32)) VUINT32 L2[4][2]; } __svml_dexpm1_data_internal; #endif __svml_dexpm1_data_internal: - /* Expm1_HA_table */ - .quad 0x0000000000000000, 0x0000000000000000 - .quad 0x0000163da8000000, 0x3e3fb33356d84a67 - .quad 0x00002c9a40000000, 0xbe3887f9f1190835 - .quad 0x00004315e8000000, 0x3e1b9fe12f5ce3e7 - .quad 0x000059b0d0000000, 0x3e48ac2ba1d73e2a - .quad 0x0000706b28000000, 0x3e3ddf6ddc6dc404 - .quad 0x0000874518000000, 0x3e1d66f20230d7c9 - .quad 0x00009e3ec8000000, 0x3e46379c1a290f03 - .quad 0x0000b55870000000, 0xbe4833b784eb3a37 - .quad 0x0000cc9228000000, 0x3e4b923fba03db83 - .quad 0x0000e3ec30000000, 0x3e469e8d10103a17 - .quad 0x0000fb66b0000000, 0xbdb2ce50dcdf6e22 - .quad 0x00011301d0000000, 0x3df25b50a4ebbf1b - .quad 0x00012abdc0000000, 0x3e1b0c72fee4aeb5 - .quad 0x0001429ab0000000, 0xbe356d2204cbefe7 - .quad 0x00015a98c8000000, 0x3e24b1ca24901aae - .quad 0x000172b840000000, 0xbe4c15742919041c - .quad 0x00018af938000000, 0x3e2191bd3777ee17 - .quad 0x0001a35be8000000, 0x3e4b7e5ba9e5b4c8 - .quad 0x0001bbe088000000, 0xbe4fdd19632a70c7 - .quad 0x0001d48730000000, 0x3e368b9aa7805b80 - .quad 0x0001ed5020000000, 0x3e47e6c8e5c40d00 - .quad 0x0002063b88000000, 0x3e18a3358ee3bac1 - .quad 0x00021f4990000000, 0x3e37ddc962552fd3 - .quad 0x0002387a70000000, 0xbe38a9dc7993e052 - .quad 0x000251ce50000000, 0xbe135670329f5521 - .quad 0x00026b4568000000, 0xbe40ec1916d42cc6 - .quad 0x000284dfe0000000, 0x3e3f5638096cf15d - .quad 0x00029e9df8000000, 0xbe470108f69ed175 - .quad 0x0002b87fd0000000, 0x3e2b5b31ffbbd48d - .quad 0x0002d285a8000000, 0xbe31bfcf4bff6e2b - .quad 0x0002ecafa8000000, 0x3e33e2f5611ca0f4 - .quad 0x000306fe08000000, 0x3e418db8a96f46ad - .quad 0x0003217100000000, 0xbe4d993e76563187 - .quad 0x00033c08b0000000, 0x3e4320b7fa64e431 - .quad 0x000356c560000000, 0xbe1b5803cdae772e - .quad 0x000371a738000000, 0xbe28aac6ab1d7560 - .quad 0x00038cae70000000, 0xbe47d13cd3d2b1a8 - .quad 0x0003a7db38000000, 0xbe48d30048af21b7 - .quad 0x0003c32dc0000000, 0x3e489d47242000f9 - .quad 0x0003dea650000000, 0xbe4f6e5eee525f6f - .quad 0x0003fa4508000000, 0xbe4a9bff22fa047f - .quad 0x0004160a20000000, 0x3e3f72e29f84325c - .quad 0x000431f5d8000000, 0x3e350a896dc70444 - .quad 0x00044e0860000000, 0x3e18624b40c4dbd0 - .quad 0x00046a41f0000000, 0xbe4717fd446d7686 - .quad 0x000486a2b8000000, 0xbe41f6197f61f2e2 - .quad 0x0004a32af0000000, 0x3e2afa7bcce5b17a - .quad 0x0004bfdad8000000, 0xbe464eaec715e343 - .quad 0x0004dcb298000000, 0x3e3fddd0d63b36ef - .quad 0x0004f9b278000000, 0xbe362d35952cc275 - .quad 0x000516daa0000000, 0x3e467b320e0897a9 - .quad 0x0005342b58000000, 0xbe362b07e20f57c4 - .quad 0x000551a4c8000000, 0x3e42ec9076297631 - .quad 0x00056f4738000000, 0xbe34ad8259913500 - .quad 0x00058d12d8000000, 0xbe4b41c016d6a1ea - .quad 0x0005ab07e0000000, 0xbe45bd5eb539b67f - .quad 0x0005c92688000000, 0x3e42ca35b80e258e - .quad 0x0005e76f18000000, 0xbe4296f5bc8b20da - .quad 0x000605e1b8000000, 0x3e376dc08b076f59 - .quad 0x0006247eb0000000, 0x3e0d2ac258f87d03 - .quad 0x0006434638000000, 0xbe4999e701c483c7 - .quad 0x0006623880000000, 0x3e42a91124893ecf - .quad 0x00068155d8000000, 0xbe4d9ab467bf1d47 - .quad 0x0006a09e68000000, 0xbe380c4336f74d05 - .quad 0x0006c01278000000, 0xbe47a12a08944ab3 - .quad 0x0006dfb240000000, 0xbe4cd72e886ef8ea - .quad 0x0006ff7df8000000, 0x3e3519483cf87e1b - .quad 0x00071f75e8000000, 0x3e2d8bee7ba46e1e - .quad 0x00073f9a48000000, 0x3e24b02e77ab934a - .quad 0x00075feb58000000, 0xbe3bd98374091656 - .quad 0x0007806950000000, 0xbe00d1604f328fec - .quad 0x0007a11470000000, 0x3e4f580c36bea881 - .quad 0x0007c1ed00000000, 0x3e330c1327c49334 - .quad 0x0007e2f338000000, 0xbe330b19defa2fd4 - .quad 0x0008042758000000, 0xbe4e0f2f724f90cc - .quad 0x0008258998000000, 0x3e34cce128acf88b - .quad 0x0008471a48000000, 0xbe3dc385331ad094 - .quad 0x000868d998000000, 0x3e4a2497640720ed - .quad 0x00088ac7d8000000, 0x3e38a669966530bd - .quad 0x0008ace540000000, 0x3e415506dadd3e2b - .quad 0x0008cf3218000000, 0xbe34abb7410d55e3 - .quad 0x0008f1ae98000000, 0x3e31577362b98274 - .quad 0x0009145b08000000, 0x3e4c8ffe2c4530da - .quad 0x00093737b0000000, 0x3e29b8bc9e8a0388 - .quad 0x00095a44c8000000, 0x3e4e4290774da41b - .quad 0x00097d82a0000000, 0xbe00d8d83a30b6f8 - .quad 0x0009a0f170000000, 0x3e2940f737462137 - .quad 0x0009c49180000000, 0x3e451f8480e3e236 - .quad 0x0009e86318000000, 0x3e3e323231824ca8 - .quad 0x000a0c6678000000, 0x3e4aef2b2594d6d4 - .quad 0x000a309bf0000000, 0xbe4dae966539f470 - .quad 0x000a5503b0000000, 0x3e41f12ae45a1225 - .quad 0x000a799e10000000, 0x3e49859ac3796fd9 - .quad 0x000a9e6b58000000, 0xbe44301205e0a6de - .quad 0x000ac36bc0000000, 0xbe0606431f9234cb - .quad 0x000ae89f98000000, 0x3e35ad3ad5e8734d - .quad 0x000b0e0728000000, 0x3e38db66590842ad - .quad 0x000b33a2b8000000, 0x3e13c57ebdaff43a - .quad 0x000b597290000000, 0xbe40d536338e3bf7 - .quad 0x000b7f76f0000000, 0x3e47daf237553d84 - .quad 0x000ba5b030000000, 0x3e2420c930819679 - .quad 0x000bcc1e90000000, 0x3e12f074891ee83d - .quad 0x000bf2c258000000, 0x3e4eb8f0442046b8 - .quad 0x000c199be0000000, 0xbe43d56b1eeef9a7 - .quad 0x000c40ab60000000, 0xbd87c2c975903ef8 - .quad 0x000c67f130000000, 0xbe3a82eb4b5dec80 - .quad 0x000c8f6d98000000, 0xbe4fc8c257729a1e - .quad 0x000cb720e0000000, 0xbe48837cb757e1a1 - .quad 0x000cdf0b58000000, 0xbe4511e031dd83b5 - .quad 0x000d072d48000000, 0x3e403c4bdc687918 - .quad 0x000d2f8708000000, 0x3deb13e315bc2473 - .quad 0x000d5818e0000000, 0xbe4822dbc6d12fd3 - .quad 0x000d80e318000000, 0xbe3367c68447b063 - .quad 0x000da9e600000000, 0x3e4ed9942b84600d - .quad 0x000dd321f0000000, 0x3e480da3025b4aef - .quad 0x000dfc9730000000, 0x3e4bdcdaf5cb4656 - .quad 0x000e264618000000, 0xbe4852f6baf6c4f0 - .quad 0x000e502ee8000000, 0xbe1d30027630bb40 - .quad 0x000e7a51f8000000, 0x3e4e3a641a5aa459 - .quad 0x000ea4afa0000000, 0x3e452486cc2c7b9d - .quad 0x000ecf4830000000, 0xbe438cc07b927e77 - .quad 0x000efa1bf0000000, 0xbe39ea5d888e02de - .quad 0x000f252b38000000, 0xbe2288ad162f2d20 - .quad 0x000f507658000000, 0x3e4b722a033a7c26 - .quad 0x000f7bfdb0000000, 0xbe431a0f63b7625a - .quad 0x000fa7c180000000, 0x3e39e90d82e90a7e - .quad 0x000fd3c228000000, 0x3e4c7b8f884badd2 - /*== poly_coeff[4] ==*/ - .align 32 - .quad 0x3f81111168877F38, 0x3f81111168877F38, 0x3f81111168877F38, 0x3f81111168877F38 /* coeff5 */ - .quad 0x3fa55555C2A9C0F3, 0x3fa55555C2A9C0F3, 0x3fa55555C2A9C0F3, 0x3fa55555C2A9C0F3 /* coeff4 */ - .quad 0x3fc555555555541D, 0x3fc555555555541D, 0x3fc555555555541D, 0x3fc555555555541D /* coeff3 */ - .quad 0x3fdFFFFFFFFFFE5C, 0x3fdFFFFFFFFFFE5C, 0x3fdFFFFFFFFFFE5C, 0x3fdFFFFFFFFFFE5C /* coeff2 */ - /*== Log2e ==*/ - .align 32 - .quad 0x40671547652B82FE, 0x40671547652B82FE, 0x40671547652B82FE, 0x40671547652B82FE - /*== L2H ==*/ - .align 32 - .quad 0x3f762e42fef80000, 0x3f762e42fef80000, 0x3f762e42fef80000, 0x3f762e42fef80000 - /*== L2L ==*/ - .align 32 - .quad 0x3d41cf79abc9e3b4, 0x3d41cf79abc9e3b4, 0x3d41cf79abc9e3b4, 0x3d41cf79abc9e3b4 - /*== ExpAddConst ==*/ - .align 32 - .quad 0x42f80000001ff800, 0x42f80000001ff800, 0x42f80000001ff800, 0x42f80000001ff800 - /*== IndexMask ==*/ - .align 32 - .quad 0x00000000000007f0, 0x00000000000007f0, 0x00000000000007f0, 0x00000000000007f0 - /*== ExpMask ==*/ - .align 32 - .quad 0x00000000003ff800, 0x00000000003ff800, 0x00000000003ff800, 0x00000000003ff800 - /*== MOne ==*/ - .align 32 - .quad 0xbff0000000000000, 0xbff0000000000000, 0xbff0000000000000, 0xbff0000000000000 - /*== AbsMask ==*/ - .align 32 - .quad 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff - /*== Threshold ==*/ - .align 32 - .quad 0x40861DA04CBAFE43, 0x40861DA04CBAFE43, 0x40861DA04CBAFE43, 0x40861DA04CBAFE43 - /*== L2 ==*/ - .align 32 - .quad 0x3f762e42fefa39ef, 0x3f762e42fefa39ef, 0x3f762e42fefa39ef, 0x3f762e42fefa39ef - .align 32 - .type __svml_dexpm1_data_internal,@object - .size __svml_dexpm1_data_internal,.-__svml_dexpm1_data_internal + /* Expm1_HA_table */ + .quad 0x0000000000000000, 0x0000000000000000 + .quad 0x0000163da8000000, 0x3e3fb33356d84a67 + .quad 0x00002c9a40000000, 0xbe3887f9f1190835 + .quad 0x00004315e8000000, 0x3e1b9fe12f5ce3e7 + .quad 0x000059b0d0000000, 0x3e48ac2ba1d73e2a + .quad 0x0000706b28000000, 0x3e3ddf6ddc6dc404 + .quad 0x0000874518000000, 0x3e1d66f20230d7c9 + .quad 0x00009e3ec8000000, 0x3e46379c1a290f03 + .quad 0x0000b55870000000, 0xbe4833b784eb3a37 + .quad 0x0000cc9228000000, 0x3e4b923fba03db83 + .quad 0x0000e3ec30000000, 0x3e469e8d10103a17 + .quad 0x0000fb66b0000000, 0xbdb2ce50dcdf6e22 + .quad 0x00011301d0000000, 0x3df25b50a4ebbf1b + .quad 0x00012abdc0000000, 0x3e1b0c72fee4aeb5 + .quad 0x0001429ab0000000, 0xbe356d2204cbefe7 + .quad 0x00015a98c8000000, 0x3e24b1ca24901aae + .quad 0x000172b840000000, 0xbe4c15742919041c + .quad 0x00018af938000000, 0x3e2191bd3777ee17 + .quad 0x0001a35be8000000, 0x3e4b7e5ba9e5b4c8 + .quad 0x0001bbe088000000, 0xbe4fdd19632a70c7 + .quad 0x0001d48730000000, 0x3e368b9aa7805b80 + .quad 0x0001ed5020000000, 0x3e47e6c8e5c40d00 + .quad 0x0002063b88000000, 0x3e18a3358ee3bac1 + .quad 0x00021f4990000000, 0x3e37ddc962552fd3 + .quad 0x0002387a70000000, 0xbe38a9dc7993e052 + .quad 0x000251ce50000000, 0xbe135670329f5521 + .quad 0x00026b4568000000, 0xbe40ec1916d42cc6 + .quad 0x000284dfe0000000, 0x3e3f5638096cf15d + .quad 0x00029e9df8000000, 0xbe470108f69ed175 + .quad 0x0002b87fd0000000, 0x3e2b5b31ffbbd48d + .quad 0x0002d285a8000000, 0xbe31bfcf4bff6e2b + .quad 0x0002ecafa8000000, 0x3e33e2f5611ca0f4 + .quad 0x000306fe08000000, 0x3e418db8a96f46ad + .quad 0x0003217100000000, 0xbe4d993e76563187 + .quad 0x00033c08b0000000, 0x3e4320b7fa64e431 + .quad 0x000356c560000000, 0xbe1b5803cdae772e + .quad 0x000371a738000000, 0xbe28aac6ab1d7560 + .quad 0x00038cae70000000, 0xbe47d13cd3d2b1a8 + .quad 0x0003a7db38000000, 0xbe48d30048af21b7 + .quad 0x0003c32dc0000000, 0x3e489d47242000f9 + .quad 0x0003dea650000000, 0xbe4f6e5eee525f6f + .quad 0x0003fa4508000000, 0xbe4a9bff22fa047f + .quad 0x0004160a20000000, 0x3e3f72e29f84325c + .quad 0x000431f5d8000000, 0x3e350a896dc70444 + .quad 0x00044e0860000000, 0x3e18624b40c4dbd0 + .quad 0x00046a41f0000000, 0xbe4717fd446d7686 + .quad 0x000486a2b8000000, 0xbe41f6197f61f2e2 + .quad 0x0004a32af0000000, 0x3e2afa7bcce5b17a + .quad 0x0004bfdad8000000, 0xbe464eaec715e343 + .quad 0x0004dcb298000000, 0x3e3fddd0d63b36ef + .quad 0x0004f9b278000000, 0xbe362d35952cc275 + .quad 0x000516daa0000000, 0x3e467b320e0897a9 + .quad 0x0005342b58000000, 0xbe362b07e20f57c4 + .quad 0x000551a4c8000000, 0x3e42ec9076297631 + .quad 0x00056f4738000000, 0xbe34ad8259913500 + .quad 0x00058d12d8000000, 0xbe4b41c016d6a1ea + .quad 0x0005ab07e0000000, 0xbe45bd5eb539b67f + .quad 0x0005c92688000000, 0x3e42ca35b80e258e + .quad 0x0005e76f18000000, 0xbe4296f5bc8b20da + .quad 0x000605e1b8000000, 0x3e376dc08b076f59 + .quad 0x0006247eb0000000, 0x3e0d2ac258f87d03 + .quad 0x0006434638000000, 0xbe4999e701c483c7 + .quad 0x0006623880000000, 0x3e42a91124893ecf + .quad 0x00068155d8000000, 0xbe4d9ab467bf1d47 + .quad 0x0006a09e68000000, 0xbe380c4336f74d05 + .quad 0x0006c01278000000, 0xbe47a12a08944ab3 + .quad 0x0006dfb240000000, 0xbe4cd72e886ef8ea + .quad 0x0006ff7df8000000, 0x3e3519483cf87e1b + .quad 0x00071f75e8000000, 0x3e2d8bee7ba46e1e + .quad 0x00073f9a48000000, 0x3e24b02e77ab934a + .quad 0x00075feb58000000, 0xbe3bd98374091656 + .quad 0x0007806950000000, 0xbe00d1604f328fec + .quad 0x0007a11470000000, 0x3e4f580c36bea881 + .quad 0x0007c1ed00000000, 0x3e330c1327c49334 + .quad 0x0007e2f338000000, 0xbe330b19defa2fd4 + .quad 0x0008042758000000, 0xbe4e0f2f724f90cc + .quad 0x0008258998000000, 0x3e34cce128acf88b + .quad 0x0008471a48000000, 0xbe3dc385331ad094 + .quad 0x000868d998000000, 0x3e4a2497640720ed + .quad 0x00088ac7d8000000, 0x3e38a669966530bd + .quad 0x0008ace540000000, 0x3e415506dadd3e2b + .quad 0x0008cf3218000000, 0xbe34abb7410d55e3 + .quad 0x0008f1ae98000000, 0x3e31577362b98274 + .quad 0x0009145b08000000, 0x3e4c8ffe2c4530da + .quad 0x00093737b0000000, 0x3e29b8bc9e8a0388 + .quad 0x00095a44c8000000, 0x3e4e4290774da41b + .quad 0x00097d82a0000000, 0xbe00d8d83a30b6f8 + .quad 0x0009a0f170000000, 0x3e2940f737462137 + .quad 0x0009c49180000000, 0x3e451f8480e3e236 + .quad 0x0009e86318000000, 0x3e3e323231824ca8 + .quad 0x000a0c6678000000, 0x3e4aef2b2594d6d4 + .quad 0x000a309bf0000000, 0xbe4dae966539f470 + .quad 0x000a5503b0000000, 0x3e41f12ae45a1225 + .quad 0x000a799e10000000, 0x3e49859ac3796fd9 + .quad 0x000a9e6b58000000, 0xbe44301205e0a6de + .quad 0x000ac36bc0000000, 0xbe0606431f9234cb + .quad 0x000ae89f98000000, 0x3e35ad3ad5e8734d + .quad 0x000b0e0728000000, 0x3e38db66590842ad + .quad 0x000b33a2b8000000, 0x3e13c57ebdaff43a + .quad 0x000b597290000000, 0xbe40d536338e3bf7 + .quad 0x000b7f76f0000000, 0x3e47daf237553d84 + .quad 0x000ba5b030000000, 0x3e2420c930819679 + .quad 0x000bcc1e90000000, 0x3e12f074891ee83d + .quad 0x000bf2c258000000, 0x3e4eb8f0442046b8 + .quad 0x000c199be0000000, 0xbe43d56b1eeef9a7 + .quad 0x000c40ab60000000, 0xbd87c2c975903ef8 + .quad 0x000c67f130000000, 0xbe3a82eb4b5dec80 + .quad 0x000c8f6d98000000, 0xbe4fc8c257729a1e + .quad 0x000cb720e0000000, 0xbe48837cb757e1a1 + .quad 0x000cdf0b58000000, 0xbe4511e031dd83b5 + .quad 0x000d072d48000000, 0x3e403c4bdc687918 + .quad 0x000d2f8708000000, 0x3deb13e315bc2473 + .quad 0x000d5818e0000000, 0xbe4822dbc6d12fd3 + .quad 0x000d80e318000000, 0xbe3367c68447b063 + .quad 0x000da9e600000000, 0x3e4ed9942b84600d + .quad 0x000dd321f0000000, 0x3e480da3025b4aef + .quad 0x000dfc9730000000, 0x3e4bdcdaf5cb4656 + .quad 0x000e264618000000, 0xbe4852f6baf6c4f0 + .quad 0x000e502ee8000000, 0xbe1d30027630bb40 + .quad 0x000e7a51f8000000, 0x3e4e3a641a5aa459 + .quad 0x000ea4afa0000000, 0x3e452486cc2c7b9d + .quad 0x000ecf4830000000, 0xbe438cc07b927e77 + .quad 0x000efa1bf0000000, 0xbe39ea5d888e02de + .quad 0x000f252b38000000, 0xbe2288ad162f2d20 + .quad 0x000f507658000000, 0x3e4b722a033a7c26 + .quad 0x000f7bfdb0000000, 0xbe431a0f63b7625a + .quad 0x000fa7c180000000, 0x3e39e90d82e90a7e + .quad 0x000fd3c228000000, 0x3e4c7b8f884badd2 + /* poly_coeff[4] */ + .align 32 + .quad 0x3f81111168877F38, 0x3f81111168877F38, 0x3f81111168877F38, 0x3f81111168877F38 /* coeff5 */ + .quad 0x3fa55555C2A9C0F3, 0x3fa55555C2A9C0F3, 0x3fa55555C2A9C0F3, 0x3fa55555C2A9C0F3 /* coeff4 */ + .quad 0x3fc555555555541D, 0x3fc555555555541D, 0x3fc555555555541D, 0x3fc555555555541D /* coeff3 */ + .quad 0x3fdFFFFFFFFFFE5C, 0x3fdFFFFFFFFFFE5C, 0x3fdFFFFFFFFFFE5C, 0x3fdFFFFFFFFFFE5C /* coeff2 */ + /* Log2e */ + .align 32 + .quad 0x40671547652B82FE, 0x40671547652B82FE, 0x40671547652B82FE, 0x40671547652B82FE + /* L2H */ + .align 32 + .quad 0x3f762e42fef80000, 0x3f762e42fef80000, 0x3f762e42fef80000, 0x3f762e42fef80000 + /* L2L */ + .align 32 + .quad 0x3d41cf79abc9e3b4, 0x3d41cf79abc9e3b4, 0x3d41cf79abc9e3b4, 0x3d41cf79abc9e3b4 + /* ExpAddConst */ + .align 32 + .quad 0x42f80000001ff800, 0x42f80000001ff800, 0x42f80000001ff800, 0x42f80000001ff800 + /* IndexMask */ + .align 32 + .quad 0x00000000000007f0, 0x00000000000007f0, 0x00000000000007f0, 0x00000000000007f0 + /* ExpMask */ + .align 32 + .quad 0x00000000003ff800, 0x00000000003ff800, 0x00000000003ff800, 0x00000000003ff800 + /* MOne */ + .align 32 + .quad 0xbff0000000000000, 0xbff0000000000000, 0xbff0000000000000, 0xbff0000000000000 + /* AbsMask */ + .align 32 + .quad 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff, 0x7fffffffffffffff + /* Threshold */ + .align 32 + .quad 0x40861DA04CBAFE43, 0x40861DA04CBAFE43, 0x40861DA04CBAFE43, 0x40861DA04CBAFE43 + /* L2 */ + .align 32 + .quad 0x3f762e42fefa39ef, 0x3f762e42fefa39ef, 0x3f762e42fefa39ef, 0x3f762e42fefa39ef + .align 32 + .type __svml_dexpm1_data_internal, @object + .size __svml_dexpm1_data_internal, .-__svml_dexpm1_data_internal |