| OLD | NEW |
| 1 ; | 1 ; |
| 2 ; jdmrgext.asm - merged upsampling/color conversion (64-bit SSE2) | 2 ; jdmrgext.asm - merged upsampling/color 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 21 matching lines...) Expand all Loading... |
| 32 | 32 |
| 33 ; r10 = JDIMENSION output_width | 33 ; r10 = JDIMENSION output_width |
| 34 ; r11 = JSAMPIMAGE input_buf | 34 ; r11 = JSAMPIMAGE input_buf |
| 35 ; r12 = JDIMENSION in_row_group_ctr | 35 ; r12 = JDIMENSION in_row_group_ctr |
| 36 ; r13 = JSAMPARRAY output_buf | 36 ; r13 = JSAMPARRAY output_buf |
| 37 | 37 |
| 38 %define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM] | 38 %define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM] |
| 39 %define WK_NUM 3 | 39 %define WK_NUM 3 |
| 40 | 40 |
| 41 align 16 | 41 align 16 |
| 42 global EXTN(jsimd_h2v1_merged_upsample_sse2) | 42 global EXTN(jsimd_h2v1_merged_upsample_sse2) PRIVATE |
| 43 | 43 |
| 44 EXTN(jsimd_h2v1_merged_upsample_sse2): | 44 EXTN(jsimd_h2v1_merged_upsample_sse2): |
| 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 |
| (...skipping 386 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 439 ; JDIMENSION in_row_group_ctr, | 439 ; JDIMENSION in_row_group_ctr, |
| 440 ; JSAMPARRAY output_buf); | 440 ; JSAMPARRAY output_buf); |
| 441 ; | 441 ; |
| 442 | 442 |
| 443 ; r10 = JDIMENSION output_width | 443 ; r10 = JDIMENSION output_width |
| 444 ; r11 = JSAMPIMAGE input_buf | 444 ; r11 = JSAMPIMAGE input_buf |
| 445 ; r12 = JDIMENSION in_row_group_ctr | 445 ; r12 = JDIMENSION in_row_group_ctr |
| 446 ; r13 = JSAMPARRAY output_buf | 446 ; r13 = JSAMPARRAY output_buf |
| 447 | 447 |
| 448 align 16 | 448 align 16 |
| 449 global EXTN(jsimd_h2v2_merged_upsample_sse2) | 449 global EXTN(jsimd_h2v2_merged_upsample_sse2) PRIVATE |
| 450 | 450 |
| 451 EXTN(jsimd_h2v2_merged_upsample_sse2): | 451 EXTN(jsimd_h2v2_merged_upsample_sse2): |
| 452 push rbp | 452 push rbp |
| 453 mov rax,rsp | 453 mov rax,rsp |
| 454 mov rbp,rsp | 454 mov rbp,rsp |
| 455 collect_args | 455 collect_args |
| 456 push rbx | 456 push rbx |
| 457 | 457 |
| 458 mov eax, r10d | 458 mov eax, r10d |
| 459 | 459 |
| (...skipping 69 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 529 pop rdx | 529 pop rdx |
| 530 | 530 |
| 531 pop rbx | 531 pop rbx |
| 532 uncollect_args | 532 uncollect_args |
| 533 pop rbp | 533 pop rbp |
| 534 ret | 534 ret |
| 535 | 535 |
| 536 ; For some reason, the OS X linker does not honor the request to align the | 536 ; For some reason, the OS X linker does not honor the request to align the |
| 537 ; segment unless we do this. | 537 ; segment unless we do this. |
| 538 align 16 | 538 align 16 |
| OLD | NEW |