| OLD | NEW |
| 1 ; | 1 ; |
| 2 ; jquanti.asm - sample data conversion and quantization (64-bit SSE2) | 2 ; jquanti.asm - sample data conversion and quantization (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 18 matching lines...) Expand all Loading... |
| 29 ; GLOBAL(void) | 29 ; GLOBAL(void) |
| 30 ; jsimd_convsamp_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col, | 30 ; jsimd_convsamp_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col, |
| 31 ; DCTELEM *workspace); | 31 ; DCTELEM *workspace); |
| 32 ; | 32 ; |
| 33 | 33 |
| 34 ; r10 = JSAMPARRAY sample_data | 34 ; r10 = JSAMPARRAY sample_data |
| 35 ; r11 = JDIMENSION start_col | 35 ; r11 = JDIMENSION start_col |
| 36 ; r12 = DCTELEM *workspace | 36 ; r12 = DCTELEM *workspace |
| 37 | 37 |
| 38 align 16 | 38 align 16 |
| 39 global EXTN(jsimd_convsamp_sse2) | 39 global EXTN(jsimd_convsamp_sse2) PRIVATE |
| 40 | 40 |
| 41 EXTN(jsimd_convsamp_sse2): | 41 EXTN(jsimd_convsamp_sse2): |
| 42 push rbp | 42 push rbp |
| 43 mov rax,rsp | 43 mov rax,rsp |
| 44 mov rbp,rsp | 44 mov rbp,rsp |
| 45 collect_args | 45 collect_args |
| 46 push rbx | 46 push rbx |
| 47 | 47 |
| 48 pxor xmm6,xmm6 ; xmm6=(all 0's) | 48 pxor xmm6,xmm6 ; xmm6=(all 0's) |
| 49 pcmpeqw xmm7,xmm7 | 49 pcmpeqw xmm7,xmm7 |
| (...skipping 55 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 105 | 105 |
| 106 %define RECIPROCAL(m,n,b) XMMBLOCK(DCTSIZE*0+(m),(n),(b),SIZEOF_DCTELEM) | 106 %define RECIPROCAL(m,n,b) XMMBLOCK(DCTSIZE*0+(m),(n),(b),SIZEOF_DCTELEM) |
| 107 %define CORRECTION(m,n,b) XMMBLOCK(DCTSIZE*1+(m),(n),(b),SIZEOF_DCTELEM) | 107 %define CORRECTION(m,n,b) XMMBLOCK(DCTSIZE*1+(m),(n),(b),SIZEOF_DCTELEM) |
| 108 %define SCALE(m,n,b) XMMBLOCK(DCTSIZE*2+(m),(n),(b),SIZEOF_DCTELEM) | 108 %define SCALE(m,n,b) XMMBLOCK(DCTSIZE*2+(m),(n),(b),SIZEOF_DCTELEM) |
| 109 | 109 |
| 110 ; r10 = JCOEFPTR coef_block | 110 ; r10 = JCOEFPTR coef_block |
| 111 ; r11 = DCTELEM *divisors | 111 ; r11 = DCTELEM *divisors |
| 112 ; r12 = DCTELEM *workspace | 112 ; r12 = DCTELEM *workspace |
| 113 | 113 |
| 114 align 16 | 114 align 16 |
| 115 global EXTN(jsimd_quantize_sse2) | 115 global EXTN(jsimd_quantize_sse2) PRIVATE |
| 116 | 116 |
| 117 EXTN(jsimd_quantize_sse2): | 117 EXTN(jsimd_quantize_sse2): |
| 118 push rbp | 118 push rbp |
| 119 mov rax,rsp | 119 mov rax,rsp |
| 120 mov rbp,rsp | 120 mov rbp,rsp |
| 121 collect_args | 121 collect_args |
| 122 | 122 |
| 123 mov rsi, r12 | 123 mov rsi, r12 |
| 124 mov rdx, r11 | 124 mov rdx, r11 |
| 125 mov rdi, r10 | 125 mov rdi, r10 |
| (...skipping 52 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 178 dec rax | 178 dec rax |
| 179 jnz near .quantloop | 179 jnz near .quantloop |
| 180 | 180 |
| 181 uncollect_args | 181 uncollect_args |
| 182 pop rbp | 182 pop rbp |
| 183 ret | 183 ret |
| 184 | 184 |
| 185 ; For some reason, the OS X linker does not honor the request to align the | 185 ; For some reason, the OS X linker does not honor the request to align the |
| 186 ; segment unless we do this. | 186 ; segment unless we do this. |
| 187 align 16 | 187 align 16 |
| OLD | NEW |