OLD | NEW |
1 ; | 1 ; |
2 ; jdclrss2-64.asm - colorspace conversion (64-bit SSE2) | 2 ; jdclrss2-64.asm - colorspace conversion (64-bit SSE2) |
3 ; | 3 ; |
4 ; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB | 4 ; Copyright 2009, 2012 Pierre Ossman <ossman@cendio.se> for Cendio AB |
5 ; Copyright 2009, 2012 D. R. Commander | 5 ; Copyright 2009, 2012 D. R. Commander |
6 ; | 6 ; |
7 ; Based on | 7 ; Based on |
8 ; x86 SIMD extension for IJG JPEG library | 8 ; x86 SIMD extension for IJG JPEG library |
9 ; Copyright (C) 1999-2006, MIYASAKA Masaru. | 9 ; Copyright (C) 1999-2006, MIYASAKA Masaru. |
10 ; For conditions of distribution and use, see copyright notice in jsimdext.inc | 10 ; For conditions of distribution and use, see copyright notice in jsimdext.inc |
(...skipping 34 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
45 push rbp | 45 push rbp |
46 mov rax,rsp ; rax = original rbp | 46 mov rax,rsp ; rax = original rbp |
47 sub rsp, byte 4 | 47 sub rsp, byte 4 |
48 and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits | 48 and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits |
49 mov [rsp],rax | 49 mov [rsp],rax |
50 mov rbp,rsp ; rbp = aligned rbp | 50 mov rbp,rsp ; rbp = aligned rbp |
51 lea rsp, [wk(0)] | 51 lea rsp, [wk(0)] |
52 collect_args | 52 collect_args |
53 push rbx | 53 push rbx |
54 | 54 |
55 » mov» rcx, r10» ; num_cols | 55 » mov» ecx, r10d» ; num_cols |
56 test rcx,rcx | 56 test rcx,rcx |
57 jz near .return | 57 jz near .return |
58 | 58 |
59 push rcx | 59 push rcx |
60 | 60 |
61 mov rdi, r11 | 61 mov rdi, r11 |
62 » mov» rcx, r12 | 62 » mov» ecx, r12d |
63 mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY] | 63 mov rsi, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY] |
64 mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY] | 64 mov rbx, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY] |
65 mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY] | 65 mov rdx, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY] |
66 lea rsi, [rsi+rcx*SIZEOF_JSAMPROW] | 66 lea rsi, [rsi+rcx*SIZEOF_JSAMPROW] |
67 lea rbx, [rbx+rcx*SIZEOF_JSAMPROW] | 67 lea rbx, [rbx+rcx*SIZEOF_JSAMPROW] |
68 lea rdx, [rdx+rcx*SIZEOF_JSAMPROW] | 68 lea rdx, [rdx+rcx*SIZEOF_JSAMPROW] |
69 | 69 |
70 pop rcx | 70 pop rcx |
71 | 71 |
72 mov rdi, r13 | 72 mov rdi, r13 |
(...skipping 359 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
432 pop rbx | 432 pop rbx |
433 uncollect_args | 433 uncollect_args |
434 mov rsp,rbp ; rsp <- aligned rbp | 434 mov rsp,rbp ; rsp <- aligned rbp |
435 pop rsp ; rsp <- original rbp | 435 pop rsp ; rsp <- original rbp |
436 pop rbp | 436 pop rbp |
437 ret | 437 ret |
438 | 438 |
439 ; For some reason, the OS X linker does not honor the request to align the | 439 ; For some reason, the OS X linker does not honor the request to align the |
440 ; segment unless we do this. | 440 ; segment unless we do this. |
441 align 16 | 441 align 16 |
OLD | NEW |