OLD | NEW |
1 ; | 1 ; |
2 ; jdmrgss2-64.asm - merged upsampling/color conversion (64-bit SSE2) | 2 ; jdmrgss2-64.asm - merged upsampling/color conversion (64-bit SSE2) |
3 ; | 3 ; |
4 ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB | 4 ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB |
5 ; Copyright 2009 D. R. Commander | 5 ; Copyright 2009 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 23 matching lines...) Expand all Loading... |
34 | 34 |
35 ; r10 = JDIMENSION output_width | 35 ; r10 = JDIMENSION output_width |
36 ; r11 = JSAMPIMAGE input_buf | 36 ; r11 = JSAMPIMAGE input_buf |
37 ; r12 = JDIMENSION in_row_group_ctr | 37 ; r12 = JDIMENSION in_row_group_ctr |
38 ; r13 = JSAMPARRAY output_buf | 38 ; r13 = JSAMPARRAY output_buf |
39 | 39 |
40 %define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM] | 40 %define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM] |
41 %define WK_NUM 3 | 41 %define WK_NUM 3 |
42 | 42 |
43 align 16 | 43 align 16 |
44 » global» EXTN(jsimd_h2v1_merged_upsample_sse2) | 44 » global» EXTN(jsimd_h2v1_merged_upsample_sse2) PRIVATE |
45 | 45 |
46 EXTN(jsimd_h2v1_merged_upsample_sse2): | 46 EXTN(jsimd_h2v1_merged_upsample_sse2): |
47 push rbp | 47 push rbp |
48 mov rax,rsp ; rax = original rbp | 48 mov rax,rsp ; rax = original rbp |
49 sub rsp, byte 4 | 49 sub rsp, byte 4 |
50 and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits | 50 and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits |
51 mov [rsp],rax | 51 mov [rsp],rax |
52 mov rbp,rsp ; rbp = aligned rbp | 52 mov rbp,rsp ; rbp = aligned rbp |
53 lea rsp, [wk(0)] | 53 lea rsp, [wk(0)] |
54 collect_args | 54 collect_args |
(...skipping 430 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
485 ; JDIMENSION in_row_group_ctr, | 485 ; JDIMENSION in_row_group_ctr, |
486 ; JSAMPARRAY output_buf); | 486 ; JSAMPARRAY output_buf); |
487 ; | 487 ; |
488 | 488 |
489 ; r10 = JDIMENSION output_width | 489 ; r10 = JDIMENSION output_width |
490 ; r11 = JSAMPIMAGE input_buf | 490 ; r11 = JSAMPIMAGE input_buf |
491 ; r12 = JDIMENSION in_row_group_ctr | 491 ; r12 = JDIMENSION in_row_group_ctr |
492 ; r13 = JSAMPARRAY output_buf | 492 ; r13 = JSAMPARRAY output_buf |
493 | 493 |
494 align 16 | 494 align 16 |
495 » global» EXTN(jsimd_h2v2_merged_upsample_sse2) | 495 » global» EXTN(jsimd_h2v2_merged_upsample_sse2) PRIVATE |
496 | 496 |
497 EXTN(jsimd_h2v2_merged_upsample_sse2): | 497 EXTN(jsimd_h2v2_merged_upsample_sse2): |
498 push rbp | 498 push rbp |
499 mov rax,rsp | 499 mov rax,rsp |
500 mov rbp,rsp | 500 mov rbp,rsp |
501 collect_args | 501 collect_args |
502 push rbx | 502 push rbx |
503 | 503 |
504 mov rax, r10 | 504 mov rax, r10 |
505 | 505 |
(...skipping 69 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
575 pop rdx | 575 pop rdx |
576 | 576 |
577 pop rbx | 577 pop rbx |
578 uncollect_args | 578 uncollect_args |
579 pop rbp | 579 pop rbp |
580 ret | 580 ret |
581 | 581 |
582 ; For some reason, the OS X linker does not honor the request to align the | 582 ; For some reason, the OS X linker does not honor the request to align the |
583 ; segment unless we do this. | 583 ; segment unless we do this. |
584 align 16 | 584 align 16 |
OLD | NEW |