[Feature] Add ssse3 and avx2 base64 decoders

4 files changed, 987 insertions, 23 deletions

diff --git a/src/libcryptobox/base64/avx2.S b/src/libcryptobox/base64/avx2.S
new file mode 100644
index 000000000..ccc0e0d6c
--- /dev/null
+++ b/src/libcryptobox/base64/avx2.S
@@ -0,0 +1,558 @@
+/*
+ * Generated from https://github.com/aklomp/base64/blob/master/lib/arch/avx2/
+ * using gcc6 -march=core-avx2 -mtune=core-avx2 -O3 -S
+ */
+
+#include "../macro.S"
+
+#ifdef LINUX
+#define PROGBITS @progbits
+#else
+#define PROGBITS
+#endif
+.comm _base64_table_dec, 256, 5
+	.file	"avx2.c"
+	.text
+	.p2align 4,,15
+GLOBAL_HIDDEN_FN_EXT base64_decode_avx2,4,1
+base64_decode_avx2_local:
+.LFB4659:
+	.cfi_startproc
+	vmovdqa	.LC0(%rip), %ymm15
+	leaq	8(%rsp), %r10
+	.cfi_def_cfa 10, 0
+	andq	$-32, %rsp
+	vmovdqa	.LC1(%rip), %ymm6
+	vmovdqa	.LC2(%rip), %ymm14
+	pushq	-8(%r10)
+	vmovdqa	.LC3(%rip), %ymm13
+	pushq	%rbp
+	vmovdqa	.LC4(%rip), %ymm12
+	vmovdqa	.LC5(%rip), %ymm11
+	vmovdqa	.LC6(%rip), %ymm10
+	vmovdqa	.LC7(%rip), %ymm9
+	vmovdqa	.LC8(%rip), %ymm8
+	.cfi_escape 0x10,0x6,0x2,0x76,0
+	movq	%rsp, %rbp
+	pushq	%r12
+	vmovdqa	.LC9(%rip), %ymm7
+	pushq	%r10
+	.cfi_escape 0xf,0x3,0x76,0x70,0x6
+	.cfi_escape 0x10,0xc,0x2,0x76,0x78
+	xorl	%r10d, %r10d
+	pushq	%rbx
+	.cfi_escape 0x10,0x3,0x2,0x76,0x68
+.L2:
+	cmpq	$44, %rsi
+	ja	.L4
+.L26:
+	leaq	-1(%rsi), %rbx
+	testq	%rsi, %rsi
+	je	.L22
+	movzbl	(%rdi), %eax
+	leaq	_base64_table_dec(%rip), %r11
+	movzbl	(%r11,%rax), %eax
+	cmpb	$-3, %al
+	ja	.L11
+	leal	0(,%rax,4), %r8d
+	testq	%rbx, %rbx
+	je	.L22
+	movzbl	1(%rdi), %eax
+	movzbl	(%r11,%rax), %eax
+	cmpb	$-3, %al
+	ja	.L15
+.L28:
+	movl	%eax, %r9d
+	leaq	-3(%rsi), %r12
+	sall	$4, %eax
+	shrb	$4, %r9b
+	orl	%r9d, %r8d
+	leaq	1(%r10), %r9
+	movb	%r8b, (%rdx)
+	cmpq	$1, %rbx
+	je	.L16
+	movzbl	2(%rdi), %r8d
+	movzbl	(%r11,%r8), %r8d
+	cmpb	$-3, %r8b
+	ja	.L24
+	movl	%r8d, %r9d
+	leaq	2(%r10), %rbx
+	sall	$6, %r8d
+	subq	$4, %rsi
+	shrb	$2, %r9b
+	orl	%r9d, %eax
+	movb	%al, 1(%rdx)
+	testq	%r12, %r12
+	je	.L19
+	movzbl	3(%rdi), %eax
+	leaq	4(%rdi), %r9
+	movzbl	(%r11,%rax), %eax
+	cmpb	$-3, %al
+	ja	.L25
+	orl	%eax, %r8d
+	addq	$3, %r10
+	addq	$3, %rdx
+	movq	%r9, %rdi
+	movb	%r8b, -1(%rdx)
+	cmpq	$44, %rsi
+	jbe	.L26
+.L4:
+	vmovdqu	(%rdi), %ymm0
+	vpcmpgtb	%ymm6, %ymm0, %ymm1
+	vpcmpgtb	%ymm15, %ymm0, %ymm2
+	vpcmpgtb	%ymm14, %ymm0, %ymm4
+	vpcmpgtb	%ymm12, %ymm0, %ymm3
+	vpandn	%ymm1, %ymm2, %ymm2
+	vpcmpgtb	%ymm13, %ymm0, %ymm1
+	vpand	%ymm7, %ymm2, %ymm2
+	vpandn	%ymm1, %ymm4, %ymm4
+	vpcmpgtb	%ymm11, %ymm0, %ymm1
+	vpand	.LC10(%rip), %ymm4, %ymm4
+	vpor	%ymm4, %ymm2, %ymm2
+	vpandn	%ymm1, %ymm3, %ymm3
+	vpcmpeqb	%ymm6, %ymm0, %ymm1
+	vpand	.LC11(%rip), %ymm3, %ymm3
+	vpand	%ymm10, %ymm1, %ymm5
+	vpcmpeqb	%ymm9, %ymm0, %ymm1
+	vpand	%ymm8, %ymm1, %ymm1
+	vpor	%ymm1, %ymm5, %ymm1
+	vpor	%ymm2, %ymm1, %ymm1
+	vpxor	%xmm2, %xmm2, %xmm2
+	vpor	%ymm3, %ymm1, %ymm1
+	vpcmpeqb	%ymm2, %ymm1, %ymm2
+	vpmovmskb	%ymm2, %eax
+	testl	%eax, %eax
+	je	.L27
+	movzbl	(%rdi), %eax
+	leaq	_base64_table_dec(%rip), %r11
+	leaq	-1(%rsi), %rbx
+	movzbl	(%r11,%rax), %eax
+	cmpb	$-3, %al
+	ja	.L11
+	leal	0(,%rax,4), %r8d
+	movzbl	1(%rdi), %eax
+	movzbl	(%r11,%rax), %eax
+	cmpb	$-3, %al
+	jbe	.L28
+.L15:
+	movl	$-1, %eax
+	vzeroupper
+	popq	%rbx
+	popq	%r10
+	.cfi_remember_state
+	.cfi_def_cfa 10, 0
+	popq	%r12
+	popq	%rbp
+	leaq	-8(%r10), %rsp
+	.cfi_def_cfa 7, 8
+	ret
+	.p2align 4,,10
+	.p2align 3
+.L27:
+	.cfi_restore_state
+	vpaddb	%ymm1, %ymm0, %ymm0
+	addq	$32, %rdi
+	addq	$24, %rdx
+	addq	$24, %r10
+	vpand	.LC12(%rip), %ymm0, %ymm3
+	vpsrld	$16, %ymm0, %ymm2
+	subq	$32, %rsi
+	vpand	.LC13(%rip), %ymm0, %ymm1
+	vpslld	$26, %ymm0, %ymm0
+	vpsrld	$2, %ymm3, %ymm3
+	vpslld	$12, %ymm1, %ymm1
+	vpor	%ymm3, %ymm2, %ymm2
+	vpor	%ymm0, %ymm1, %ymm0
+	vmovdqa	.LC15(%rip), %ymm1
+	vpor	%ymm2, %ymm0, %ymm0
+	vpshufb	.LC14(%rip), %ymm0, %ymm0
+	vpermd	%ymm0, %ymm1, %ymm0
+	vmovdqu	%ymm0, -24(%rdx)
+	jmp	.L2
+	.p2align 4,,10
+	.p2align 3
+.L19:
+	movq	%rbx, %r10
+.L22:
+	movl	$1, %eax
+.L5:
+	movq	%r10, (%rcx)
+	vzeroupper
+	popq	%rbx
+	popq	%r10
+	.cfi_remember_state
+	.cfi_def_cfa 10, 0
+	popq	%r12
+	popq	%rbp
+	leaq	-8(%r10), %rsp
+	.cfi_def_cfa 7, 8
+	ret
+	.p2align 4,,10
+	.p2align 3
+.L11:
+	.cfi_restore_state
+	xorl	%eax, %eax
+	jmp	.L5
+	.p2align 4,,10
+	.p2align 3
+.L16:
+	movq	%r9, %r10
+	movl	$1, %eax
+	jmp	.L5
+	.p2align 4,,10
+	.p2align 3
+.L24:
+	movq	%r9, %r10
+	xorl	%eax, %eax
+	cmpb	$-2, %r8b
+	jne	.L5
+	movl	$1, %eax
+	testq	%r12, %r12
+	je	.L5
+	movzbl	3(%rdi), %eax
+	cmpb	$-2, (%r11,%rax)
+	sete	%dl
+	xorl	%eax, %eax
+	cmpq	$4, %rsi
+	sete	%al
+	andl	%edx, %eax
+	jmp	.L5
+	.p2align 4,,10
+	.p2align 3
+.L25:
+	cmpb	$-2, %al
+	movq	%rbx, %r10
+	sete	%dl
+	xorl	%eax, %eax
+	testq	%rsi, %rsi
+	sete	%al
+	andl	%edx, %eax
+	jmp	.L5
+	.cfi_endproc
+.LFE4659:
+	.section	.rodata.cst32,"aM",PROGBITS,32
+	.align 16
+.LC0:
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.align 16
+.LC1:
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.align 16
+.LC2:
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.align 16
+.LC3:
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.align 16
+.LC4:
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.align 16
+.LC5:
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.align 16
+.LC6:
+	.quad	1157442765409226768
+	.quad	1157442765409226768
+	.quad	1157442765409226768
+	.quad	1157442765409226768
+	.align 16
+.LC7:
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.align 16
+.LC8:
+	.quad	1374463283923456787
+	.quad	1374463283923456787
+	.quad	1374463283923456787
+	.quad	1374463283923456787
+	.align 16
+.LC9:
+	.quad	289360691352306692
+	.quad	289360691352306692
+	.quad	289360691352306692
+	.quad	289360691352306692
+	.align 16
+.LC10:
+	.quad	-4629771061636907073
+	.quad	-4629771061636907073
+	.quad	-4629771061636907073
+	.quad	-4629771061636907073
+	.align 16
+.LC11:
+	.quad	-5063812098665367111
+	.quad	-5063812098665367111
+	.quad	-5063812098665367111
+	.quad	-5063812098665367111
+	.align 16
+.LC12:
+	.quad	17732923536900096
+	.quad	17732923536900096
+	.quad	17732923536900096
+	.quad	17732923536900096
+	.align 16
+.LC13:
+	.quad	69269232566016
+	.quad	69269232566016
+	.quad	69269232566016
+	.quad	69269232566016
+	.align 16
+.LC14:
+	.byte	3
+	.byte	2
+	.byte	1
+	.byte	7
+	.byte	6
+	.byte	5
+	.byte	11
+	.byte	10
+	.byte	9
+	.byte	15
+	.byte	14
+	.byte	13
+	.byte	-1
+	.byte	-1
+	.byte	-1
+	.byte	-1
+	.byte	3
+	.byte	2
+	.byte	1
+	.byte	7
+	.byte	6
+	.byte	5
+	.byte	11
+	.byte	10
+	.byte	9
+	.byte	15
+	.byte	14
+	.byte	13
+	.byte	-1
+	.byte	-1
+	.byte	-1
+	.byte	-1
+	.align 16
+.LC15:
+	.long	0
+	.long	1
+	.long	2
+	.long	4
+	.long	5
+	.long	6
+	.long	-1
+	.long	-1
+	.ident	"GCC: (Debian 6.2.1-5) 6.2.1 20161124"
diff --git a/src/libcryptobox/base64/base64.c b/src/libcryptobox/base64/base64.c
index c280b59fa..3a1e74ac7 100644
--- a/src/libcryptobox/base64/base64.c
+++ b/src/libcryptobox/base64/base64.c
@@ -20,6 +20,26 @@
 #include "platform_config.h"
 
 extern unsigned long cpu_config;
+const uint8_t
+base64_table_dec[256] =
+{
+	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
+	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
+	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,  62, 255, 255, 255,  63,
+	 52,  53,  54,  55,  56,  57,  58,  59,  60,  61, 255, 255, 255, 254, 255, 255,
+	255,   0,   1,   2,   3,   4,   5,   6,   7,   8,   9,  10,  11,  12,  13,  14,
+	 15,  16,  17,  18,  19,  20,  21,  22,  23,  24,  25, 255, 255, 255, 255, 255,
+	255,  26,  27,  28,  29,  30,  31,  32,  33,  34,  35,  36,  37,  38,  39,  40,
+	 41,  42,  43,  44,  45,  46,  47,  48,  49,  50,  51, 255, 255, 255, 255, 255,
+	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
+	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
+	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
+	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
+	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
+	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
+	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
+	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
+};
 
 typedef struct base64_impl {
 	unsigned long cpu_flags;
@@ -36,15 +56,16 @@ typedef struct base64_impl {
 
 BASE64_DECLARE(ref);
 #define BASE64_REF BASE64_IMPL(0, "ref", ref)
+BASE64_DECLARE(avx2);
+#define BASE64_AVX2 BASE64_IMPL(CPUID_AVX2,"avx2", avx2)
+BASE64_DECLARE(ssse3);
+#define BASE64_SSSE3 BASE64_IMPL(CPUID_SSSE3, "avx2", ssse3)
 
 static const base64_impl_t base64_list[] = {
 		BASE64_REF,
 #if defined(BASE64_AVX2)
 		BASE64_AVX2,
 #endif
-#if defined(BASE64_AVX)
-		BASE64_AVX,
-#endif
 #if defined(BASE64_SSSE3)
 		BASE64_SSSE3,
 #endif
diff --git a/src/libcryptobox/base64/ref.c b/src/libcryptobox/base64/ref.c
index b262fa5e5..22cedee6a 100644
--- a/src/libcryptobox/base64/ref.c
+++ b/src/libcryptobox/base64/ref.c
@@ -29,26 +29,7 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 #include "config.h"
 
-const uint8_t
-base64_table_dec[] =
-{
-	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,  62, 255, 255, 255,  63,
-	 52,  53,  54,  55,  56,  57,  58,  59,  60,  61, 255, 255, 255, 254, 255, 255,
-	255,   0,   1,   2,   3,   4,   5,   6,   7,   8,   9,  10,  11,  12,  13,  14,
-	 15,  16,  17,  18,  19,  20,  21,  22,  23,  24,  25, 255, 255, 255, 255, 255,
-	255,  26,  27,  28,  29,  30,  31,  32,  33,  34,  35,  36,  37,  38,  39,  40,
-	 41,  42,  43,  44,  45,  46,  47,  48,  49,  50,  51, 255, 255, 255, 255, 255,
-	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-	255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
-};
+extern const uint8_t base64_table_dec[256];
 
 #define INNER_LOOP_64 do { \
 	while (inlen >= 13) { \
diff --git a/src/libcryptobox/base64/ssse3.S b/src/libcryptobox/base64/ssse3.S
new file mode 100644
index 000000000..5cf6da52f
--- /dev/null
+++ b/src/libcryptobox/base64/ssse3.S
@@ -0,0 +1,404 @@
+/*
+ * Generated from https://github.com/aklomp/base64/blob/master/lib/arch/ssse3/
+ * using gcc6 -march=core2 -mtune=core2 -O3 -S
+ */
+
+
+#include "../macro.S"
+
+
+#ifdef LINUX
+#define PROGBITS @progbits
+#else
+#define PROGBITS
+#endif
+
+.comm _base64_table_dec, 256, 5
+
+	.file	"ssse3.c"
+	.text
+	.p2align 4,,15
+GLOBAL_HIDDEN_FN_EXT base64_decode_ssse3,4,1
+base64_decode_ssse3_local:
+.LFB4658:
+	.cfi_startproc
+	movdqa	.LC0(%rip), %xmm11
+	pushq	%rbp
+	.cfi_def_cfa_offset 16
+	.cfi_offset 6, -16
+	xorl	%r10d, %r10d
+	movdqa	.LC1(%rip), %xmm2
+	pushq	%rbx
+	.cfi_def_cfa_offset 24
+	.cfi_offset 3, -24
+	movdqa	.LC2(%rip), %xmm10
+	movdqa	.LC3(%rip), %xmm9
+	movdqa	.LC4(%rip), %xmm8
+	movdqa	.LC5(%rip), %xmm7
+	movdqa	.LC6(%rip), %xmm6
+	movdqa	.LC7(%rip), %xmm5
+	movdqa	.LC8(%rip), %xmm4
+	movdqa	.LC9(%rip), %xmm3
+.L2:
+	cmpq	$23, %rsi
+	ja	.L4
+.L25:
+	leaq	-1(%rsi), %rbx
+	testq	%rsi, %rsi
+	je	.L21
+	movzbl	(%rdi), %eax
+	leaq	_base64_table_dec(%rip), %r9
+	movzbl	(%r9,%rax), %eax
+	cmpb	$-3, %al
+	ja	.L11
+	leal	0(,%rax,4), %r8d
+	testq	%rbx, %rbx
+	je	.L21
+	movzbl	1(%rdi), %eax
+	movzbl	(%r9,%rax), %eax
+	cmpb	$-3, %al
+	ja	.L15
+.L27:
+	leaq	-3(%rsi), %rbp
+	movl	%eax, %r11d
+	sall	$4, %eax
+	shrb	$4, %r11b
+	orl	%r11d, %r8d
+	cmpq	$1, %rbx
+	movb	%r8b, (%rdx)
+	leaq	1(%r10), %r11
+	je	.L16
+	movzbl	2(%rdi), %r8d
+	movzbl	(%r9,%r8), %r8d
+	cmpb	$-3, %r8b
+	ja	.L23
+	leaq	2(%r10), %rbx
+	movl	%r8d, %r11d
+	subq	$4, %rsi
+	shrb	$2, %r11b
+	sall	$6, %r8d
+	orl	%r11d, %eax
+	testq	%rbp, %rbp
+	movb	%al, 1(%rdx)
+	je	.L19
+	movzbl	3(%rdi), %eax
+	leaq	4(%rdi), %r11
+	movzbl	(%r9,%rax), %eax
+	cmpb	$-3, %al
+	ja	.L24
+	orl	%eax, %r8d
+	addq	$3, %r10
+	addq	$3, %rdx
+	movb	%r8b, -1(%rdx)
+	cmpq	$23, %rsi
+	movq	%r11, %rdi
+	jbe	.L25
+.L4:
+	movdqu	(%rdi), %xmm1
+	movdqa	%xmm1, %xmm0
+	movdqa	%xmm1, %xmm12
+	movdqa	%xmm1, %xmm13
+	pcmpgtb	%xmm2, %xmm12
+	movdqa	%xmm1, %xmm14
+	pcmpgtb	%xmm11, %xmm0
+	pcmpgtb	%xmm10, %xmm13
+	pcmpgtb	%xmm7, %xmm14
+	pandn	%xmm12, %xmm0
+	movdqa	%xmm1, %xmm12
+	pand	%xmm6, %xmm0
+	pcmpgtb	%xmm9, %xmm12
+	pandn	%xmm12, %xmm13
+	movdqa	%xmm1, %xmm12
+	pand	%xmm5, %xmm13
+	por	%xmm13, %xmm0
+	pcmpgtb	%xmm8, %xmm12
+	pandn	%xmm14, %xmm12
+	pand	%xmm4, %xmm12
+	por	%xmm12, %xmm0
+	movdqa	%xmm1, %xmm12
+	pcmpeqb	%xmm3, %xmm12
+	pand	.LC10(%rip), %xmm12
+	por	%xmm12, %xmm0
+	movdqa	%xmm1, %xmm12
+	pcmpeqb	%xmm2, %xmm12
+	pand	.LC11(%rip), %xmm12
+	por	%xmm12, %xmm0
+	pxor	%xmm12, %xmm12
+	pcmpeqb	%xmm0, %xmm12
+	pmovmskb	%xmm12, %eax
+	testl	%eax, %eax
+	je	.L26
+	movzbl	(%rdi), %eax
+	leaq	_base64_table_dec(%rip), %r9
+	leaq	-1(%rsi), %rbx
+	movzbl	(%r9,%rax), %eax
+	cmpb	$-3, %al
+	ja	.L11
+	leal	0(,%rax,4), %r8d
+	movzbl	1(%rdi), %eax
+	movzbl	(%r9,%rax), %eax
+	cmpb	$-3, %al
+	jbe	.L27
+.L15:
+	movl	$-1, %eax
+	popq	%rbx
+	.cfi_remember_state
+	.cfi_def_cfa_offset 16
+	popq	%rbp
+	.cfi_def_cfa_offset 8
+	ret
+	.p2align 4,,10
+	.p2align 3
+.L26:
+	.cfi_restore_state
+	paddb	%xmm1, %xmm0
+	addq	$16, %rdi
+	addq	$12, %rdx
+	movdqa	.LC12(%rip), %xmm13
+	movdqa	%xmm0, %xmm1
+	addq	$12, %r10
+	subq	$16, %rsi
+	movdqa	.LC13(%rip), %xmm12
+	pand	%xmm0, %xmm13
+	psrld	$16, %xmm1
+	pand	%xmm0, %xmm12
+	psrld	$2, %xmm13
+	pslld	$12, %xmm12
+	por	%xmm13, %xmm1
+	pslld	$26, %xmm0
+	por	%xmm12, %xmm1
+	por	%xmm1, %xmm0
+	pshufb	.LC14(%rip), %xmm0
+	movups	%xmm0, -12(%rdx)
+	jmp	.L2
+	.p2align 4,,10
+	.p2align 3
+.L19:
+	movq	%rbx, %r10
+.L21:
+	movl	$1, %eax
+.L5:
+	popq	%rbx
+	.cfi_remember_state
+	.cfi_def_cfa_offset 16
+	movq	%r10, (%rcx)
+	popq	%rbp
+	.cfi_def_cfa_offset 8
+	ret
+	.p2align 4,,10
+	.p2align 3
+.L11:
+	.cfi_restore_state
+	xorl	%eax, %eax
+	jmp	.L5
+	.p2align 4,,10
+	.p2align 3
+.L16:
+	movq	%r11, %r10
+	movl	$1, %eax
+	jmp	.L5
+	.p2align 4,,10
+	.p2align 3
+.L23:
+	xorl	%eax, %eax
+	cmpb	$-2, %r8b
+	movq	%r11, %r10
+	jne	.L5
+	testq	%rbp, %rbp
+	movl	$1, %eax
+	je	.L5
+	movzbl	3(%rdi), %eax
+	cmpb	$-2, (%r9,%rax)
+	sete	%dl
+	xorl	%eax, %eax
+	cmpq	$4, %rsi
+	sete	%al
+	andl	%edx, %eax
+	jmp	.L5
+	.p2align 4,,10
+	.p2align 3
+.L24:
+	cmpb	$-2, %al
+	movq	%rbx, %r10
+	sete	%dl
+	xorl	%eax, %eax
+	testq	%rsi, %rsi
+	sete	%al
+	andl	%edx, %eax
+	jmp	.L5
+	.cfi_endproc
+.LFE4658:
+	.section	.rodata.cst16,"aM",PROGBITS,16
+	.align 16
+.LC0:
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.byte	57
+	.align 16
+.LC1:
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.byte	47
+	.align 16
+.LC2:
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.byte	90
+	.align 16
+.LC3:
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.byte	64
+	.align 16
+.LC4:
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.byte	122
+	.align 16
+.LC5:
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.byte	96
+	.align 16
+.LC6:
+	.quad	289360691352306692
+	.quad	289360691352306692
+	.align 16
+.LC7:
+	.quad	-4629771061636907073
+	.quad	-4629771061636907073
+	.align 16
+.LC8:
+	.quad	-5063812098665367111
+	.quad	-5063812098665367111
+	.align 16
+.LC9:
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.byte	43
+	.align 16
+.LC10:
+	.quad	1374463283923456787
+	.quad	1374463283923456787
+	.align 16
+.LC11:
+	.quad	1157442765409226768
+	.quad	1157442765409226768
+	.align 16
+.LC12:
+	.quad	17732923536900096
+	.quad	17732923536900096
+	.align 16
+.LC13:
+	.quad	69269232566016
+	.quad	69269232566016
+	.align 16
+.LC14:
+	.byte	3
+	.byte	2
+	.byte	1
+	.byte	7
+	.byte	6
+	.byte	5
+	.byte	11
+	.byte	10
+	.byte	9
+	.byte	15
+	.byte	14
+	.byte	13
+	.byte	-1
+	.byte	-1
+	.byte	-1
+	.byte	-1
+	.ident	"GCC: (Debian 6.2.1-5) 6.2.1 20161124"