OLD | NEW |
1 ; | 1 ; |
2 ; jfdctint.asm - accurate integer FDCT (64-bit SSE2) | 2 ; jfdctint.asm - accurate integer FDCT (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 49 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
60 F_1_961 equ DESCALE(2106220350,30-CONST_BITS) ; FIX(1.961570560) | 60 F_1_961 equ DESCALE(2106220350,30-CONST_BITS) ; FIX(1.961570560) |
61 F_2_053 equ DESCALE(2204520673,30-CONST_BITS) ; FIX(2.053119869) | 61 F_2_053 equ DESCALE(2204520673,30-CONST_BITS) ; FIX(2.053119869) |
62 F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447) | 62 F_2_562 equ DESCALE(2751909506,30-CONST_BITS) ; FIX(2.562915447) |
63 F_3_072 equ DESCALE(3299298341,30-CONST_BITS) ; FIX(3.072711026) | 63 F_3_072 equ DESCALE(3299298341,30-CONST_BITS) ; FIX(3.072711026) |
64 %endif | 64 %endif |
65 | 65 |
66 ; -------------------------------------------------------------------------- | 66 ; -------------------------------------------------------------------------- |
67 SECTION SEG_CONST | 67 SECTION SEG_CONST |
68 | 68 |
69 alignz 16 | 69 alignz 16 |
70 global EXTN(jconst_fdct_islow_sse2) | 70 global EXTN(jconst_fdct_islow_sse2) PRIVATE |
71 | 71 |
72 EXTN(jconst_fdct_islow_sse2): | 72 EXTN(jconst_fdct_islow_sse2): |
73 | 73 |
74 PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541 | 74 PW_F130_F054 times 4 dw (F_0_541+F_0_765), F_0_541 |
75 PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847) | 75 PW_F054_MF130 times 4 dw F_0_541, (F_0_541-F_1_847) |
76 PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175 | 76 PW_MF078_F117 times 4 dw (F_1_175-F_1_961), F_1_175 |
77 PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390) | 77 PW_F117_F078 times 4 dw F_1_175, (F_1_175-F_0_390) |
78 PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899 | 78 PW_MF060_MF089 times 4 dw (F_0_298-F_0_899),-F_0_899 |
79 PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899) | 79 PW_MF089_F060 times 4 dw -F_0_899, (F_1_501-F_0_899) |
80 PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562 | 80 PW_MF050_MF256 times 4 dw (F_2_053-F_2_562),-F_2_562 |
(...skipping 13 matching lines...) Expand all Loading... |
94 ; GLOBAL(void) | 94 ; GLOBAL(void) |
95 ; jsimd_fdct_islow_sse2 (DCTELEM *data) | 95 ; jsimd_fdct_islow_sse2 (DCTELEM *data) |
96 ; | 96 ; |
97 | 97 |
98 ; r10 = DCTELEM *data | 98 ; r10 = DCTELEM *data |
99 | 99 |
100 %define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM] | 100 %define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM] |
101 %define WK_NUM 6 | 101 %define WK_NUM 6 |
102 | 102 |
103 align 16 | 103 align 16 |
104 global EXTN(jsimd_fdct_islow_sse2) | 104 global EXTN(jsimd_fdct_islow_sse2) PRIVATE |
105 | 105 |
106 EXTN(jsimd_fdct_islow_sse2): | 106 EXTN(jsimd_fdct_islow_sse2): |
107 push rbp | 107 push rbp |
108 mov rax,rsp ; rax = original rbp | 108 mov rax,rsp ; rax = original rbp |
109 sub rsp, byte 4 | 109 sub rsp, byte 4 |
110 and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits | 110 and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits |
111 mov [rsp],rax | 111 mov [rsp],rax |
112 mov rbp,rsp ; rbp = aligned rbp | 112 mov rbp,rsp ; rbp = aligned rbp |
113 lea rsp, [wk(0)] | 113 lea rsp, [wk(0)] |
114 collect_args | 114 collect_args |
(...skipping 498 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
613 | 613 |
614 uncollect_args | 614 uncollect_args |
615 mov rsp,rbp ; rsp <- aligned rbp | 615 mov rsp,rbp ; rsp <- aligned rbp |
616 pop rsp ; rsp <- original rbp | 616 pop rsp ; rsp <- original rbp |
617 pop rbp | 617 pop rbp |
618 ret | 618 ret |
619 | 619 |
620 ; For some reason, the OS X linker does not honor the request to align the | 620 ; For some reason, the OS X linker does not honor the request to align the |
621 ; segment unless we do this. | 621 ; segment unless we do this. |
622 align 16 | 622 align 16 |
OLD | NEW |