Forums before death by AOL, social media and spammers... "We can't have nice things"
| comp.lang.asm.x86 | Ahh, the lost art of x86 assembly | 4,675 messages |
[ << oldest | < older | list | newer > | newest >> ]
| Message 3,313 of 4,675 |
| Bernhard Schornak to John |
| Re: Hex 2 Bin (1/2) |
| 20 Mar 18 23:28:09 |
From: schornak@nospicedham.web.de Kerr-Mudd,John wrote: > I want something with far less jmps! I obviously mismatched the direction in my first post. The other way around is about *15 times* slower, because there are a lot of things to sort out to get a proper result. This one requires a bunch of lookup tables: CVTCUT 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xFF 0xFF 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xFF 0xFF 0xFF 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xFF 0xFF 0xFF 0xFF 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xFF 0xFF 0xFF 0xFF 0xFF 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0x00 0x00 0x00 0x00 0x00 0x00 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0x00 0x00 0x00 0x00 0x00 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0x00 0x00 0x00 0x00 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0x00 0x00 0x00 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0x00 0x00 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0xFF 0x00 CVTCUT() is used to mask out bytes beyond the trailing zero. CVTxx 16 consecutive byte with content xx. These LUTs are used in multiple functions, so the "extra space" (528 byte, including other characters not used here) pays off for me. /* ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ hex2D hexadecimal ASCII string -> DWORD ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ -> RCX EA source ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <- RAX dword (8 digits) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ BUFFER 0 0123456789ABCDEF "-16 byte string-" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ average latency ~ 115 clock cycles (1st call), 96 clock cycles (2nd an up) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ */ .p2align 5,,31 .globl _hex2D .def _hex2D; .scl 2; .type 32; .endef _hex2D:subq $0xF8, %rsp movdqu 0x00(%rcx), %xmm0 pxor %xmm1, %xmm1 movq _BNR(%rip), %rax # RAX = BNR pcmpeqb %xmm0, %xmm1 movq %r8, 0x68(%rsp) movq %r9, 0x70(%rsp) movq %rdx, 0x78(%rsp) pmovmskb %xmm1, %r8d movq %rbp, 0x80(%rsp) movq %rbx, 0x88(%rsp) movdqa %xmm9, 0x90(%rsp) movdqa %xmm8, 0xA0(%rsp) movdqa %xmm7, 0xB0(%rsp) bsf %r8d, %r8d # R14 = 1st zero P_0 movdqa %xmm6, 0xC0(%rsp) movdqa %xmm5, 0xD0(%rsp) movdqa %xmm4, 0xE0(%rsp) movq %rcx, 0xF0(%rsp) leaq CVTCUT(%rax), %rbx # R13 = LUT cut shlq %r8 # R08 = index 0 cmove %r8d, %eax # zero => RES = 0 je 1f movdqa CVT_60(%rax), %xmm9 # X09 = 6060...60 pand 0x00(%rbx, %r8, 8), %xmm0 # cut off trailing garbage movdqa CVT_20(%rax), %xmm4 # X04 = 2020...20 movdqa %xmm0, %xmm1 # X01 = TMP_0 movdqa CVT_30(%rax), %xmm5 # X05 = 3030...30 pcmpgtb %xmm9, %xmm1 # X01 = mask 0 movdqa CVT_39(%rax), %xmm6 # X06 = 3939...39 pand %xmm4, %xmm1 # X01 = diff 0 movdqa CVT_40(%rax), %xmm7 # X07 = 4040...40 psubb %xmm1, %xmm0 # X00 = temp res 0 movdqa CVT_46(%rax), %xmm8 # X08 = 4646...46 movdqa %xmm0, %xmm1 # X01 = TMP_0 pcmpgtb %xmm8, %xmm1 # X01 = mask > "F" pand %xmm0, %xmm1 # X01 = erase > "F" psubb %xmm1, %xmm0 # X00 = temp res 0 movdqa %xmm0, %xmm1 # X01 = TMP_0.0 movdqa %xmm0, %xmm8 # X08 = TMP_0.1 pcmpgtb %xmm6, %xmm1 # X01 = mask 0.0 > "9" pcmpgtb %xmm7, %xmm8 # X08 = mask 0.1 > "@" movdqa CVT_07(%rax), %xmm7 # X07 = 0707...07 pxor %xmm8, %xmm1 # X01 = mask 0 pand %xmm0, %xmm1 # X01 = cut 0 psubb %xmm1, %xmm0 # X00 = temp res 0 pand %xmm7, %xmm8 # X08 = diff 0 psubb %xmm8, %xmm0 # X00 = temp res 0 psubb %xmm5, %xmm0 # X00 = res - "0" movl $0x08, %ebp # RBP = words [continued in next message] --- SoupGate-Win32 v1.05 * Origin: you cannot sedate... all the things you hate (1:229/2) |
(c) 1994, bbs@darkrealms.ca