media/libjpeg/simd/jcqnts2f-64.asm@6474c204b198 (annotated)
media/libjpeg/simd/jcqnts2f-64.asm
Wed, 31 Dec 2014 06:09:35 +0100
- author
- Michael Schloh von Bennewitz <michael@schloh.com>
- date
- Wed, 31 Dec 2014 06:09:35 +0100
- changeset 0
- 6474c204b198
- permissions
- -rw-r--r--
Cloned upstream origin tor-browser at tor-browser-31.3.0esr-4.5-1-build1
revision ID fc1c9ff7c1b2defdbc039f12214767608f46423f for hacking purpose.
| michael@0 | 1 | ; |
| michael@0 | 2 | ; jcqnts2f-64.asm - sample data conversion and quantization (64-bit SSE & SSE2) |
| michael@0 | 3 | ; |
| michael@0 | 4 | ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB |
| michael@0 | 5 | ; Copyright 2009 D. R. Commander |
| michael@0 | 6 | ; |
| michael@0 | 7 | ; Based on |
| michael@0 | 8 | ; x86 SIMD extension for IJG JPEG library |
| michael@0 | 9 | ; Copyright (C) 1999-2006, MIYASAKA Masaru. |
| michael@0 | 10 | ; For conditions of distribution and use, see copyright notice in jsimdext.inc |
| michael@0 | 11 | ; |
| michael@0 | 12 | ; This file should be assembled with NASM (Netwide Assembler), |
| michael@0 | 13 | ; can *not* be assembled with Microsoft's MASM or any compatible |
| michael@0 | 14 | ; assembler (including Borland's Turbo Assembler). |
| michael@0 | 15 | ; NASM is available from http://nasm.sourceforge.net/ or |
| michael@0 | 16 | ; http://sourceforge.net/project/showfiles.php?group_id=6208 |
| michael@0 | 17 | ; |
| michael@0 | 18 | ; [TAB8] |
| michael@0 | 19 | |
| michael@0 | 20 | %include "jsimdext.inc" |
| michael@0 | 21 | %include "jdct.inc" |
| michael@0 | 22 | |
| michael@0 | 23 | ; -------------------------------------------------------------------------- |
| michael@0 | 24 | SECTION SEG_TEXT |
| michael@0 | 25 | BITS 64 |
| michael@0 | 26 | ; |
| michael@0 | 27 | ; Load data into workspace, applying unsigned->signed conversion |
| michael@0 | 28 | ; |
| michael@0 | 29 | ; GLOBAL(void) |
| michael@0 | 30 | ; jsimd_convsamp_float_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col, |
| michael@0 | 31 | ; FAST_FLOAT * workspace); |
| michael@0 | 32 | ; |
| michael@0 | 33 | |
| michael@0 | 34 | ; r10 = JSAMPARRAY sample_data |
| michael@0 | 35 | ; r11 = JDIMENSION start_col |
| michael@0 | 36 | ; r12 = FAST_FLOAT * workspace |
| michael@0 | 37 | |
| michael@0 | 38 | align 16 |
| michael@0 | 39 | global EXTN(jsimd_convsamp_float_sse2) |
| michael@0 | 40 | |
| michael@0 | 41 | EXTN(jsimd_convsamp_float_sse2): |
| michael@0 | 42 | push rbp |
| michael@0 | 43 | mov rax,rsp |
| michael@0 | 44 | mov rbp,rsp |
| michael@0 | 45 | collect_args |
| michael@0 | 46 | push rbx |
| michael@0 | 47 | |
| michael@0 | 48 | pcmpeqw xmm7,xmm7 |
| michael@0 | 49 | psllw xmm7,7 |
| michael@0 | 50 | packsswb xmm7,xmm7 ; xmm7 = PB_CENTERJSAMPLE (0x808080..) |
| michael@0 | 51 | |
| michael@0 | 52 | mov rsi, r10 |
| michael@0 | 53 | mov rax, r11 |
| michael@0 | 54 | mov rdi, r12 |
| michael@0 | 55 | mov rcx, DCTSIZE/2 |
| michael@0 | 56 | .convloop: |
| michael@0 | 57 | mov rbx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) |
| michael@0 | 58 | mov rdx, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) |
| michael@0 | 59 | |
| michael@0 | 60 | movq xmm0, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE] |
| michael@0 | 61 | movq xmm1, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE] |
| michael@0 | 62 | |
| michael@0 | 63 | psubb xmm0,xmm7 ; xmm0=(01234567) |
| michael@0 | 64 | psubb xmm1,xmm7 ; xmm1=(89ABCDEF) |
| michael@0 | 65 | |
| michael@0 | 66 | punpcklbw xmm0,xmm0 ; xmm0=(*0*1*2*3*4*5*6*7) |
| michael@0 | 67 | punpcklbw xmm1,xmm1 ; xmm1=(*8*9*A*B*C*D*E*F) |
| michael@0 | 68 | |
| michael@0 | 69 | punpcklwd xmm2,xmm0 ; xmm2=(***0***1***2***3) |
| michael@0 | 70 | punpckhwd xmm0,xmm0 ; xmm0=(***4***5***6***7) |
| michael@0 | 71 | punpcklwd xmm3,xmm1 ; xmm3=(***8***9***A***B) |
| michael@0 | 72 | punpckhwd xmm1,xmm1 ; xmm1=(***C***D***E***F) |
| michael@0 | 73 | |
| michael@0 | 74 | psrad xmm2,(DWORD_BIT-BYTE_BIT) ; xmm2=(0123) |
| michael@0 | 75 | psrad xmm0,(DWORD_BIT-BYTE_BIT) ; xmm0=(4567) |
| michael@0 | 76 | cvtdq2ps xmm2,xmm2 ; xmm2=(0123) |
| michael@0 | 77 | cvtdq2ps xmm0,xmm0 ; xmm0=(4567) |
| michael@0 | 78 | psrad xmm3,(DWORD_BIT-BYTE_BIT) ; xmm3=(89AB) |
| michael@0 | 79 | psrad xmm1,(DWORD_BIT-BYTE_BIT) ; xmm1=(CDEF) |
| michael@0 | 80 | cvtdq2ps xmm3,xmm3 ; xmm3=(89AB) |
| michael@0 | 81 | cvtdq2ps xmm1,xmm1 ; xmm1=(CDEF) |
| michael@0 | 82 | |
| michael@0 | 83 | movaps XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_FAST_FLOAT)], xmm2 |
| michael@0 | 84 | movaps XMMWORD [XMMBLOCK(0,1,rdi,SIZEOF_FAST_FLOAT)], xmm0 |
| michael@0 | 85 | movaps XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_FAST_FLOAT)], xmm3 |
| michael@0 | 86 | movaps XMMWORD [XMMBLOCK(1,1,rdi,SIZEOF_FAST_FLOAT)], xmm1 |
| michael@0 | 87 | |
| michael@0 | 88 | add rsi, byte 2*SIZEOF_JSAMPROW |
| michael@0 | 89 | add rdi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT |
| michael@0 | 90 | dec rcx |
| michael@0 | 91 | jnz short .convloop |
| michael@0 | 92 | |
| michael@0 | 93 | pop rbx |
| michael@0 | 94 | uncollect_args |
| michael@0 | 95 | pop rbp |
| michael@0 | 96 | ret |
| michael@0 | 97 | |
| michael@0 | 98 | |
| michael@0 | 99 | ; -------------------------------------------------------------------------- |
| michael@0 | 100 | ; |
| michael@0 | 101 | ; Quantize/descale the coefficients, and store into coef_block |
| michael@0 | 102 | ; |
| michael@0 | 103 | ; GLOBAL(void) |
| michael@0 | 104 | ; jsimd_quantize_float_sse2 (JCOEFPTR coef_block, FAST_FLOAT * divisors, |
| michael@0 | 105 | ; FAST_FLOAT * workspace); |
| michael@0 | 106 | ; |
| michael@0 | 107 | |
| michael@0 | 108 | ; r10 = JCOEFPTR coef_block |
| michael@0 | 109 | ; r11 = FAST_FLOAT * divisors |
| michael@0 | 110 | ; r12 = FAST_FLOAT * workspace |
| michael@0 | 111 | |
| michael@0 | 112 | align 16 |
| michael@0 | 113 | global EXTN(jsimd_quantize_float_sse2) |
| michael@0 | 114 | |
| michael@0 | 115 | EXTN(jsimd_quantize_float_sse2): |
| michael@0 | 116 | push rbp |
| michael@0 | 117 | mov rax,rsp |
| michael@0 | 118 | mov rbp,rsp |
| michael@0 | 119 | collect_args |
| michael@0 | 120 | |
| michael@0 | 121 | mov rsi, r12 |
| michael@0 | 122 | mov rdx, r11 |
| michael@0 | 123 | mov rdi, r10 |
| michael@0 | 124 | mov rax, DCTSIZE2/16 |
| michael@0 | 125 | .quantloop: |
| michael@0 | 126 | movaps xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_FAST_FLOAT)] |
| michael@0 | 127 | movaps xmm1, XMMWORD [XMMBLOCK(0,1,rsi,SIZEOF_FAST_FLOAT)] |
| michael@0 | 128 | mulps xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)] |
| michael@0 | 129 | mulps xmm1, XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)] |
| michael@0 | 130 | movaps xmm2, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_FAST_FLOAT)] |
| michael@0 | 131 | movaps xmm3, XMMWORD [XMMBLOCK(1,1,rsi,SIZEOF_FAST_FLOAT)] |
| michael@0 | 132 | mulps xmm2, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)] |
| michael@0 | 133 | mulps xmm3, XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)] |
| michael@0 | 134 | |
| michael@0 | 135 | cvtps2dq xmm0,xmm0 |
| michael@0 | 136 | cvtps2dq xmm1,xmm1 |
| michael@0 | 137 | cvtps2dq xmm2,xmm2 |
| michael@0 | 138 | cvtps2dq xmm3,xmm3 |
| michael@0 | 139 | |
| michael@0 | 140 | packssdw xmm0,xmm1 |
| michael@0 | 141 | packssdw xmm2,xmm3 |
| michael@0 | 142 | |
| michael@0 | 143 | movdqa XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_JCOEF)], xmm0 |
| michael@0 | 144 | movdqa XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_JCOEF)], xmm2 |
| michael@0 | 145 | |
| michael@0 | 146 | add rsi, byte 16*SIZEOF_FAST_FLOAT |
| michael@0 | 147 | add rdx, byte 16*SIZEOF_FAST_FLOAT |
| michael@0 | 148 | add rdi, byte 16*SIZEOF_JCOEF |
| michael@0 | 149 | dec rax |
| michael@0 | 150 | jnz short .quantloop |
| michael@0 | 151 | |
| michael@0 | 152 | uncollect_args |
| michael@0 | 153 | pop rbp |
| michael@0 | 154 | ret |
| michael@0 | 155 | |
| michael@0 | 156 | ; For some reason, the OS X linker does not honor the request to align the |
| michael@0 | 157 | ; segment unless we do this. |
| michael@0 | 158 | align 16 |
