| OLD | NEW |
| 1 ; | 1 ; |
| 2 ; jdcolext.asm - colorspace conversion (64-bit SSE2) | 2 ; jdcolext.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 21 matching lines...) Expand all Loading... |
| 32 ; r10 = JDIMENSION out_width | 32 ; r10 = JDIMENSION out_width |
| 33 ; r11 = JSAMPIMAGE input_buf | 33 ; r11 = JSAMPIMAGE input_buf |
| 34 ; r12 = JDIMENSION input_row | 34 ; r12 = JDIMENSION input_row |
| 35 ; r13 = JSAMPARRAY output_buf | 35 ; r13 = JSAMPARRAY output_buf |
| 36 ; r14 = int num_rows | 36 ; r14 = int num_rows |
| 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 2 | 39 %define WK_NUM 2 |
| 40 | 40 |
| 41 align 16 | 41 align 16 |
| 42 global EXTN(jsimd_ycc_rgb_convert_sse2) | 42 global EXTN(jsimd_ycc_rgb_convert_sse2) PRIVATE |
| 43 | 43 |
| 44 EXTN(jsimd_ycc_rgb_convert_sse2): | 44 EXTN(jsimd_ycc_rgb_convert_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 379 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 |