x86_64: Fix svml_d_log1p4_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>

1 files changed, 1293 insertions, 1294 deletions

diff --git a/sysdeps/x86_64/fpu/multiarch/svml_d_log1p4_core_avx2.S b/sysdeps/x86_64/fpu/multiarch/svml_d_log1p4_core_avx2.S
index 9ae36d484b..13235793e8 100644
--- a/sysdeps/x86_64/fpu/multiarch/svml_d_log1p4_core_avx2.S
+++ b/sysdeps/x86_64/fpu/multiarch/svml_d_log1p4_core_avx2.S
@@ -19,7 +19,7 @@
 /*
  * ALGORITHM DESCRIPTION:
  *
- *    1+x = 2^k*(xh + xl) is computed in high-low parts; xh in [1,2)
+ *    1+x = 2^k*(xh + xl) is computed in high-low parts; xh in [1, 2)
  *    Get short reciprocal approximation Rcp ~ 1/xh
  *    R = (Rcp*xh - 1.0) + Rcp*xl
  *    log1p(x) = k*log(2.0) - log(Rcp) + poly(R)
@@ -30,1354 +30,1353 @@
 
 /* Offsets for data table __svml_dlog1p_data_internal
  */
-#define Log_HA_table                  	0
-#define Log_LA_table                  	8224
-#define poly_coeff                    	12352
-#define ExpMask                       	12480
-#define Two10                         	12512
-#define MinLog1p                      	12544
-#define MaxLog1p                      	12576
-#define One                           	12608
-#define SgnMask                       	12640
-#define XThreshold                    	12672
-#define XhMask                        	12704
-#define Threshold                     	12736
-#define Bias                          	12768
-#define Bias1                         	12800
-#define ExpMask0                      	12832
-#define ExpMask2                      	12864
-#define L2                            	12896
+#define Log_HA_table			0
+#define Log_LA_table			8224
+#define poly_coeff			12352
+#define ExpMask				12480
+#define Two10				12512
+#define MinLog1p			12544
+#define MaxLog1p			12576
+#define One				12608
+#define SgnMask				12640
+#define XThreshold			12672
+#define XhMask				12704
+#define Threshold			12736
+#define Bias				12768
+#define Bias1				12800
+#define ExpMask0			12832
+#define ExpMask2			12864
+#define L2				12896
 
 /* Lookup bias for data table __svml_dlog1p_data_internal.  */
-#define Table_Lookup_Bias               -0x405fe0
+#define Table_Lookup_Bias		-0x405fe0
 
 #include <sysdep.h>
 
-        .text
-	.section .text.avx2,"ax",@progbits
+	.section .text.avx2, "ax", @progbits
 ENTRY(_ZGVdN4v_log1p_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       Table_Lookup_Bias+__svml_dlog1p_data_internal(%rip), %r8
+	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	Table_Lookup_Bias+__svml_dlog1p_data_internal(%rip), %r8
 
-/* SgnMask used by all accuracies */
-        vmovupd   SgnMask+__svml_dlog1p_data_internal(%rip), %ymm12
-        vmovupd   One+__svml_dlog1p_data_internal(%rip), %ymm7
+	/* SgnMask used by all accuracies */
+	vmovupd	SgnMask+__svml_dlog1p_data_internal(%rip), %ymm12
+	vmovupd	One+__svml_dlog1p_data_internal(%rip), %ymm7
 
-/* 2^ (-10-exp(X) ) */
-        vmovupd   ExpMask2+__svml_dlog1p_data_internal(%rip), %ymm3
-        vmovapd   %ymm0, %ymm9
-        vandpd    %ymm12, %ymm9, %ymm10
-        vcmplt_oqpd XThreshold+__svml_dlog1p_data_internal(%rip), %ymm10, %ymm11
-        vaddpd    %ymm7, %ymm9, %ymm13
+	/* 2^ (-10-exp(X) ) */
+	vmovupd	ExpMask2+__svml_dlog1p_data_internal(%rip), %ymm3
+	vmovapd	%ymm0, %ymm9
+	vandpd	%ymm12, %ymm9, %ymm10
+	vcmplt_oqpd XThreshold+__svml_dlog1p_data_internal(%rip), %ymm10, %ymm11
+	vaddpd	%ymm7, %ymm9, %ymm13
 
-/* compute 1+x as high, low parts */
-        vmaxpd    %ymm9, %ymm7, %ymm15
-        vminpd    %ymm9, %ymm7, %ymm6
-        vorpd     XhMask+__svml_dlog1p_data_internal(%rip), %ymm11, %ymm14
-        vandpd    %ymm14, %ymm13, %ymm4
+	/* compute 1+x as high, low parts */
+	vmaxpd	%ymm9, %ymm7, %ymm15
+	vminpd	%ymm9, %ymm7, %ymm6
+	vorpd	XhMask+__svml_dlog1p_data_internal(%rip), %ymm11, %ymm14
+	vandpd	%ymm14, %ymm13, %ymm4
 
-/* preserve mantissa, set input exponent to 2^(-10) */
-        vandpd    ExpMask+__svml_dlog1p_data_internal(%rip), %ymm4, %ymm5
-        vorpd     Two10+__svml_dlog1p_data_internal(%rip), %ymm5, %ymm5
+	/* preserve mantissa, set input exponent to 2^(-10) */
+	vandpd	ExpMask+__svml_dlog1p_data_internal(%rip), %ymm4, %ymm5
+	vorpd	Two10+__svml_dlog1p_data_internal(%rip), %ymm5, %ymm5
 
-/* reciprocal approximation good to at least 11 bits */
-        vcvtpd2ps %ymm5, %xmm2
-        vsubpd    %ymm4, %ymm15, %ymm0
+	/* reciprocal approximation good to at least 11 bits */
+	vcvtpd2ps %ymm5, %xmm2
+	vsubpd	%ymm4, %ymm15, %ymm0
 
-/* check range */
-        vcmplt_oqpd MinLog1p+__svml_dlog1p_data_internal(%rip), %ymm9, %ymm15
-        vrcpps    %xmm2, %xmm1
-        vaddpd    %ymm0, %ymm6, %ymm6
-        vcmpnle_uqpd MaxLog1p+__svml_dlog1p_data_internal(%rip), %ymm9, %ymm0
-        vcvtps2pd %xmm1, %ymm11
+	/* check range */
+	vcmplt_oqpd MinLog1p+__svml_dlog1p_data_internal(%rip), %ymm9, %ymm15
+	vrcpps	%xmm2, %xmm1
+	vaddpd	%ymm0, %ymm6, %ymm6
+	vcmpnle_uqpd MaxLog1p+__svml_dlog1p_data_internal(%rip), %ymm9, %ymm0
+	vcvtps2pd %xmm1, %ymm11
 
-/* exponent of X needed to scale Xl */
-        vandps    ExpMask0+__svml_dlog1p_data_internal(%rip), %ymm4, %ymm10
-        vpsubq    %ymm10, %ymm3, %ymm13
+	/* exponent of X needed to scale Xl */
+	vandps	ExpMask0+__svml_dlog1p_data_internal(%rip), %ymm4, %ymm10
+	vpsubq	%ymm10, %ymm3, %ymm13
 
-/* exponent bits */
-        vpsrlq    $20, %ymm4, %ymm4
+	/* exponent bits */
+	vpsrlq	$20, %ymm4, %ymm4
 
-/* round reciprocal to nearest integer, will have 1+9 mantissa bits */
-        vroundpd  $0, %ymm11, %ymm3
+	/* round reciprocal to nearest integer, will have 1+9 mantissa bits */
+	vroundpd $0, %ymm11, %ymm3
 
-/* scale DblRcp */
-        vmulpd    %ymm13, %ymm3, %ymm2
+	/* scale DblRcp */
+	vmulpd	%ymm13, %ymm3, %ymm2
 
-/* exponent*log(2.0) */
-        vmovupd   Threshold+__svml_dlog1p_data_internal(%rip), %ymm13
-        vfmsub213pd %ymm7, %ymm3, %ymm5
+	/* exponent*log(2.0) */
+	vmovupd	Threshold+__svml_dlog1p_data_internal(%rip), %ymm13
+	vfmsub213pd %ymm7, %ymm3, %ymm5
 
-/* Compute SignMask for all accuracies, including EP */
-        vandnpd   %ymm9, %ymm12, %ymm8
-        vorpd     %ymm0, %ymm15, %ymm7
+	/* Compute SignMask for all accuracies, including EP */
+	vandnpd	%ymm9, %ymm12, %ymm8
+	vorpd	%ymm0, %ymm15, %ymm7
 
-/*
- * prepare table index
- * table lookup
- */
-        vpsrlq    $40, %ymm3, %ymm0
+	/*
+	 * prepare table index
+	 * table lookup
+	 */
+	vpsrlq	$40, %ymm3, %ymm0
 
-/*
- * argument reduction
- * VQFMS( D, R, X, DblRcp1, One );
- */
-        vfmadd213pd %ymm5, %ymm2, %ymm6
-        vmovupd   poly_coeff+64+__svml_dlog1p_data_internal(%rip), %ymm2
-        vcmplt_oqpd %ymm3, %ymm13, %ymm3
-        vmulpd    %ymm6, %ymm6, %ymm5
-        vfmadd213pd poly_coeff+96+__svml_dlog1p_data_internal(%rip), %ymm6, %ymm2
+	/*
+	 * argument reduction
+	 * VQFMS( D, R, X, DblRcp1, One );
+	 */
+	vfmadd213pd %ymm5, %ymm2, %ymm6
+	vmovupd	poly_coeff+64+__svml_dlog1p_data_internal(%rip), %ymm2
+	vcmplt_oqpd %ymm3, %ymm13, %ymm3
+	vmulpd	%ymm6, %ymm6, %ymm5
+	vfmadd213pd poly_coeff+96+__svml_dlog1p_data_internal(%rip), %ymm6, %ymm2
 
-/* combine and get argument value range mask */
-        vmovmskpd %ymm7, %eax
-        vextractf128 $1, %ymm4, %xmm12
-        vshufps   $221, %xmm12, %xmm4, %xmm14
+	/* combine and get argument value range mask */
+	vmovmskpd %ymm7, %eax
+	vextractf128 $1, %ymm4, %xmm12
+	vshufps	$221, %xmm12, %xmm4, %xmm14
 
-/* biased exponent in DP format */
-        vcvtdq2pd %xmm14, %ymm1
-        vandpd    Bias+__svml_dlog1p_data_internal(%rip), %ymm3, %ymm14
-        vorpd     Bias1+__svml_dlog1p_data_internal(%rip), %ymm14, %ymm15
-        vsubpd    %ymm15, %ymm1, %ymm1
-        vmulpd    L2+__svml_dlog1p_data_internal(%rip), %ymm1, %ymm3
+	/* biased exponent in DP format */
+	vcvtdq2pd %xmm14, %ymm1
+	vandpd	Bias+__svml_dlog1p_data_internal(%rip), %ymm3, %ymm14
+	vorpd	Bias1+__svml_dlog1p_data_internal(%rip), %ymm14, %ymm15
+	vsubpd	%ymm15, %ymm1, %ymm1
+	vmulpd	L2+__svml_dlog1p_data_internal(%rip), %ymm1, %ymm3
 
-/* polynomial */
-        vmovupd   poly_coeff+__svml_dlog1p_data_internal(%rip), %ymm1
-        vfmadd213pd poly_coeff+32+__svml_dlog1p_data_internal(%rip), %ymm6, %ymm1
-        vfmadd213pd %ymm2, %ymm5, %ymm1
+	/* polynomial */
+	vmovupd	poly_coeff+__svml_dlog1p_data_internal(%rip), %ymm1
+	vfmadd213pd poly_coeff+32+__svml_dlog1p_data_internal(%rip), %ymm6, %ymm1
+	vfmadd213pd %ymm2, %ymm5, %ymm1
 
-/* reconstruction */
-        vfmadd213pd %ymm6, %ymm5, %ymm1
-        vextractf128 $1, %ymm0, %xmm10
-        vmovd     %xmm0, %edx
-        vmovd     %xmm10, %esi
-        movslq    %edx, %rdx
-        vpextrd   $2, %xmm0, %ecx
-        movslq    %esi, %rsi
-        vpextrd   $2, %xmm10, %edi
-        movslq    %ecx, %rcx
-        movslq    %edi, %rdi
-        vmovsd    (%r8,%rdx), %xmm4
-        vmovsd    (%r8,%rsi), %xmm11
-        vmovhpd   (%r8,%rcx), %xmm4, %xmm7
-        vmovhpd   (%r8,%rdi), %xmm11, %xmm12
-        vinsertf128 $1, %xmm12, %ymm7, %ymm0
-        vaddpd    %ymm1, %ymm0, %ymm6
-        vaddpd    %ymm6, %ymm3, %ymm0
+	/* reconstruction */
+	vfmadd213pd %ymm6, %ymm5, %ymm1
+	vextractf128 $1, %ymm0, %xmm10
+	vmovd	%xmm0, %edx
+	vmovd	%xmm10, %esi
+	movslq	%edx, %rdx
+	vpextrd	$2, %xmm0, %ecx
+	movslq	%esi, %rsi
+	vpextrd	$2, %xmm10, %edi
+	movslq	%ecx, %rcx
+	movslq	%edi, %rdi
+	vmovsd	(%r8, %rdx), %xmm4
+	vmovsd	(%r8, %rsi), %xmm11
+	vmovhpd	(%r8, %rcx), %xmm4, %xmm7
+	vmovhpd	(%r8, %rdi), %xmm11, %xmm12
+	vinsertf128 $1, %xmm12, %ymm7, %ymm0
+	vaddpd	%ymm1, %ymm0, %ymm6
+	vaddpd	%ymm6, %ymm3, %ymm0
 
-/* OR in the Sign of input argument to produce correct log1p(-0) */
-        vorpd     %ymm8, %ymm0, %ymm0
-        testl     %eax, %eax
+	/* OR in the Sign of input argument to produce correct log1p(-0) */
+	vorpd	%ymm8, %ymm0, %ymm0
+	testl	%eax, %eax
 
-/* Go to special inputs processing branch */
-        jne       L(SPECIAL_VALUES_BRANCH)
-                                # LOE rbx r12 r13 r14 r15 eax ymm0 ymm9
+	/* Go to special inputs processing branch */
+	jne	L(SPECIAL_VALUES_BRANCH)
+	# LOE rbx r12 r13 r14 r15 eax ymm0 ymm9
 
-/* 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):
-        vmovupd   %ymm9, 32(%rsp)
-        vmovupd   %ymm0, 64(%rsp)
-                                # LOE rbx r12 r13 r14 r15 eax ymm0
+	vmovupd	%ymm9, 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
+	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      $4, %r12d
+	incl	%r12d
+	cmpl	$4, %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)
-        vmovupd   64(%rsp), %ymm0
+	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
+	/* 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
-        movsd     32(%rsp,%r14,8), %xmm0
-        call      log1p@PLT
-                                # LOE rbx r14 r15 r12d r13d xmm0
+	movl	%r12d, %r14d
+	movsd	32(%rsp, %r14, 8), %xmm0
+	call	log1p@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_log1p_avx2)
 
-        .section .rodata, "a"
-        .align 32
+	.section .rodata, "a"
+	.align	32
 
 #ifdef __svml_dlog1p_data_internal_typedef
 typedef unsigned int VUINT32;
 typedef struct {
-        __declspec(align(32)) VUINT32 Log_HA_table[(1<<10)+2][2];
-        __declspec(align(32)) VUINT32 Log_LA_table[(1<<9)+1][2];
-        __declspec(align(32)) VUINT32 poly_coeff[4][4][2];
-        __declspec(align(32)) VUINT32 ExpMask[4][2];
-        __declspec(align(32)) VUINT32 Two10[4][2];
-        __declspec(align(32)) VUINT32 MinLog1p[4][2];
-        __declspec(align(32)) VUINT32 MaxLog1p[4][2];
-        __declspec(align(32)) VUINT32 One[4][2];
-        __declspec(align(32)) VUINT32 SgnMask[4][2];
-        __declspec(align(32)) VUINT32 XThreshold[4][2];
-        __declspec(align(32)) VUINT32 XhMask[4][2];
-        __declspec(align(32)) VUINT32 Threshold[4][2];
-        __declspec(align(32)) VUINT32 Bias[4][2];
-        __declspec(align(32)) VUINT32 Bias1[4][2];
-        __declspec(align(32)) VUINT32 ExpMask0[4][2];
-        __declspec(align(32)) VUINT32 ExpMask2[4][2];
-        __declspec(align(32)) VUINT32 L2[4][2];
+	__declspec(align(32)) VUINT32 Log_HA_table[(1<<10)+2][2];
+	__declspec(align(32)) VUINT32 Log_LA_table[(1<<9)+1][2];
+	__declspec(align(32)) VUINT32 poly_coeff[4][4][2];
+	__declspec(align(32)) VUINT32 ExpMask[4][2];
+	__declspec(align(32)) VUINT32 Two10[4][2];
+	__declspec(align(32)) VUINT32 MinLog1p[4][2];
+	__declspec(align(32)) VUINT32 MaxLog1p[4][2];
+	__declspec(align(32)) VUINT32 One[4][2];
+	__declspec(align(32)) VUINT32 SgnMask[4][2];
+	__declspec(align(32)) VUINT32 XThreshold[4][2];
+	__declspec(align(32)) VUINT32 XhMask[4][2];
+	__declspec(align(32)) VUINT32 Threshold[4][2];
+	__declspec(align(32)) VUINT32 Bias[4][2];
+	__declspec(align(32)) VUINT32 Bias1[4][2];
+	__declspec(align(32)) VUINT32 ExpMask0[4][2];
+	__declspec(align(32)) VUINT32 ExpMask2[4][2];
+	__declspec(align(32)) VUINT32 L2[4][2];
 } __svml_dlog1p_data_internal;
 #endif
 __svml_dlog1p_data_internal:
-        /* Log_HA_table */
-        .quad 0xc086232bdd7a8300, 0xbe1ce91eef3fb100
-        .quad 0xc086232fdc7ad828, 0xbe1cefcffda73b6a
-        .quad 0xc0862333d97d2ba0, 0xbe1cef406748f1ff
-        .quad 0xc0862337d48378e0, 0xbe1cef2a9429925a
-        .quad 0xc086233bcd8fb878, 0xbe1cf138d17ebecb
-        .quad 0xc086233fc4a3e018, 0xbe1ceff2dbbbb29e
-        .quad 0xc0862343b9c1e270, 0xbe1cf1a42aae437b
-        .quad 0xc0862347acebaf68, 0xbe1cef3b152048af
-        .quad 0xc086234b9e2333f0, 0xbe1cef20e127805e
-        .quad 0xc086234f8d6a5a30, 0xbe1cf00ad6052cf4
-        .quad 0xc08623537ac30980, 0xbe1cefc4642ee597
-        .quad 0xc0862357662f2660, 0xbe1cf1f277d36e16
-        .quad 0xc086235b4fb092a0, 0xbe1ceed009e8d8e6
-        .quad 0xc086235f37492d28, 0xbe1cf1e4038cb362
-        .quad 0xc08623631cfad250, 0xbe1cf0b0873b8557
-        .quad 0xc086236700c75b98, 0xbe1cf15bb3227c0b
-        .quad 0xc086236ae2b09fe0, 0xbe1cf151ef8ca9ed
-        .quad 0xc086236ec2b87358, 0xbe1cefe1dc2cd2ed
-        .quad 0xc0862372a0e0a780, 0xbe1cf0d1eec5454f
-        .quad 0xc08623767d2b0b48, 0xbe1ceeefd570bbce
-        .quad 0xc086237a57996af0, 0xbe1cee99ae91b3a7
-        .quad 0xc086237e302d9028, 0xbe1cf0412830fbd1
-        .quad 0xc086238206e94218, 0xbe1ceee898588610
-        .quad 0xc0862385dbce4548, 0xbe1cee9a1fbcaaea
-        .quad 0xc0862389aede5bc0, 0xbe1ceed8e7cc1ad6
-        .quad 0xc086238d801b4500, 0xbe1cf10c8d059da6
-        .quad 0xc08623914f86be18, 0xbe1ceee6c63a8165
-        .quad 0xc08623951d228180, 0xbe1cf0c3592d2ff1
-        .quad 0xc0862398e8f04758, 0xbe1cf0026cc4cb1b
-        .quad 0xc086239cb2f1c538, 0xbe1cf15d48d8e670
-        .quad 0xc08623a07b28ae60, 0xbe1cef359363787c
-        .quad 0xc08623a44196b390, 0xbe1cefdf1ab2e82c
-        .quad 0xc08623a8063d8338, 0xbe1cefe43c02aa84
-        .quad 0xc08623abc91ec960, 0xbe1cf044f5ae35b7
-        .quad 0xc08623af8a3c2fb8, 0xbe1cf0b0b4001e1b
-        .quad 0xc08623b349975d98, 0xbe1cf1bae76dfbcf
-        .quad 0xc08623b70731f810, 0xbe1cef0a72e13a62
-        .quad 0xc08623bac30da1c8, 0xbe1cf184007d2b6b
-        .quad 0xc08623be7d2bfb40, 0xbe1cf16f4b239e98
-        .quad 0xc08623c2358ea2a0, 0xbe1cf0976acada87
-        .quad 0xc08623c5ec3733d0, 0xbe1cf066318a16ff
-        .quad 0xc08623c9a1274880, 0xbe1ceffaa7148798
-        .quad 0xc08623cd54607820, 0xbe1cf23ab02e9b6e
-        .quad 0xc08623d105e45800, 0xbe1cefdfef7d4fde
-        .quad 0xc08623d4b5b47b20, 0xbe1cf17fece44f2b
-        .quad 0xc08623d863d27270, 0xbe1cf18f907d0d7c
-        .quad 0xc08623dc103fccb0, 0xbe1cee61fe072c98
-        .quad 0xc08623dfbafe1668, 0xbe1cf022dd891e2f
-        .quad 0xc08623e3640eda20, 0xbe1ceecc1daf4358
-        .quad 0xc08623e70b73a028, 0xbe1cf0173c4fa380
-        .quad 0xc08623eab12deec8, 0xbe1cf16a2150c2f4
-        .quad 0xc08623ee553f4a30, 0xbe1cf1bf980b1f4b
-        .quad 0xc08623f1f7a93480, 0xbe1cef8b731663c2
-        .quad 0xc08623f5986d2dc0, 0xbe1cee9a664d7ef4
-        .quad 0xc08623f9378cb3f0, 0xbe1cf1eda2af6400
-        .quad 0xc08623fcd5094320, 0xbe1cf1923f9d68d7
-        .quad 0xc086240070e45548, 0xbe1cf0747cd3e03a
-        .quad 0xc08624040b1f6260, 0xbe1cf22ee855bd6d
-        .quad 0xc0862407a3bbe078, 0xbe1cf0d57360c00b
-        .quad 0xc086240b3abb4398, 0xbe1ceebc815cd575
-        .quad 0xc086240ed01efdd0, 0xbe1cf03bfb970951
-        .quad 0xc086241263e87f50, 0xbe1cf16e74768529
-        .quad 0xc0862415f6193658, 0xbe1cefec64b8becb
-        .quad 0xc086241986b28f30, 0xbe1cf0838d210baa
-        .quad 0xc086241d15b5f448, 0xbe1cf0ea86e75b11
-        .quad 0xc0862420a324ce28, 0xbe1cf1708d11d805
-        .quad 0xc08624242f008380, 0xbe1ceea988c5a417
-        .quad 0xc0862427b94a7910, 0xbe1cef166a7bbca5
-        .quad 0xc086242b420411d0, 0xbe1cf0c9d9e86a38
-        .quad 0xc086242ec92eaee8, 0xbe1cef0946455411
-        .quad 0xc08624324ecbaf98, 0xbe1cefea60907739
-        .quad 0xc0862435d2dc7160, 0xbe1cf1ed0934ce42
-        .quad 0xc086243955624ff8, 0xbe1cf191ba746c7d
-        .quad 0xc086243cd65ea548, 0xbe1ceeec78cf2a7e
-        .quad 0xc086244055d2c968, 0xbe1cef345284c119
-        .quad 0xc0862443d3c012b8, 0xbe1cf24f77355219
-        .quad 0xc08624475027d5e8, 0xbe1cf05bf087e114
-        .quad 0xc086244acb0b65d0, 0xbe1cef3504a32189
-        .quad 0xc086244e446c1398, 0xbe1ceff54b2a406f
-        .quad 0xc0862451bc4b2eb8, 0xbe1cf0757d54ed4f
-        .quad 0xc086245532aa04f0, 0xbe1cf0c8099fdfd5
-        .quad 0xc0862458a789e250, 0xbe1cf0b173796a31
-        .quad 0xc086245c1aec1138, 0xbe1cf11d8734540d
-        .quad 0xc086245f8cd1da60, 0xbe1cf1916a723ceb
-        .quad 0xc0862462fd3c84d8, 0xbe1cf19a911e1da7
-        .quad 0xc08624666c2d5608, 0xbe1cf23a9ef72e4f
-        .quad 0xc0862469d9a591c0, 0xbe1cef503d947663
-        .quad 0xc086246d45a67a18, 0xbe1cf0fceeb1a0b2
-        .quad 0xc0862470b0314fa8, 0xbe1cf107e27e4fbc
-        .quad 0xc086247419475160, 0xbe1cf03dd9922331
-        .quad 0xc086247780e9bc98, 0xbe1cefce1a10e129
-        .quad 0xc086247ae719cd18, 0xbe1ceea47f73c4f6
-        .quad 0xc086247e4bd8bd10, 0xbe1ceec0ac56d100
-        .quad 0xc0862481af27c528, 0xbe1cee8a6593278a
-        .quad 0xc086248511081c70, 0xbe1cf2231dd9dec7
-        .quad 0xc0862488717af888, 0xbe1cf0b4b8ed7da8
-        .quad 0xc086248bd0818d68, 0xbe1cf1bd8d835002
-        .quad 0xc086248f2e1d0d98, 0xbe1cf259acc107f4
-        .quad 0xc08624928a4eaa20, 0xbe1cee897636b00c
-        .quad 0xc0862495e5179270, 0xbe1cee757f20c326
-        .quad 0xc08624993e78f490, 0xbe1cefafd3aa54a4
-        .quad 0xc086249c9673fd10, 0xbe1cee7298d38b97
-        .quad 0xc086249fed09d6f8, 0xbe1ceedc158d4ceb
-        .quad 0xc08624a3423babe0, 0xbe1cf2282987cb2e
-        .quad 0xc08624a6960aa400, 0xbe1cefe7381ecc4b
-        .quad 0xc08624a9e877e600, 0xbe1cef328dbbce80
-        .quad 0xc08624ad39849728, 0xbe1cefde45f3cc71
-        .quad 0xc08624b08931db58, 0xbe1cefa8b89433b9
-        .quad 0xc08624b3d780d500, 0xbe1cef6773c0b139
-        .quad 0xc08624b72472a528, 0xbe1cf031c931c11f
-        .quad 0xc08624ba70086b78, 0xbe1cf088f49275e7
-        .quad 0xc08624bdba434630, 0xbe1cf17de0eaa86d
-        .quad 0xc08624c103245238, 0xbe1cefd492f1ba75
-        .quad 0xc08624c44aacab08, 0xbe1cf1253e154466
-        .quad 0xc08624c790dd6ad0, 0xbe1cf0fb09ee6d55
-        .quad 0xc08624cad5b7aa58, 0xbe1cf1f08dd048fe
-        .quad 0xc08624ce193c8120, 0xbe1ceeca0809697f
-        .quad 0xc08624d15b6d0538, 0xbe1cef8d5662d968
-        .quad 0xc08624d49c4a4b78, 0xbe1