DRC | cdc8ac3 | 2009-06-25 20:38:31 +0000 | [diff] [blame] | 1 | ; |
| 2 | ; jcqnts2f.asm - sample data conversion and quantization (64-bit SSE & SSE2) |
| 3 | ; |
| 4 | ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB |
| 5 | ; Copyright 2009 D. R. Commander |
| 6 | ; |
| 7 | ; Based on |
| 8 | ; x86 SIMD extension for IJG JPEG library |
| 9 | ; Copyright (C) 1999-2006, MIYASAKA Masaru. |
| 10 | ; For conditions of distribution and use, see copyright notice in jsimdext.inc |
| 11 | ; |
| 12 | ; This file should be assembled with NASM (Netwide Assembler), |
| 13 | ; can *not* be assembled with Microsoft's MASM or any compatible |
| 14 | ; assembler (including Borland's Turbo Assembler). |
| 15 | ; NASM is available from http://nasm.sourceforge.net/ or |
| 16 | ; http://sourceforge.net/project/showfiles.php?group_id=6208 |
| 17 | ; |
| 18 | ; [TAB8] |
| 19 | |
| 20 | %include "jsimdext.inc" |
| 21 | %include "jdct.inc" |
| 22 | |
| 23 | ; -------------------------------------------------------------------------- |
| 24 | SECTION SEG_TEXT |
| 25 | BITS 64 |
| 26 | ; |
| 27 | ; Load data into workspace, applying unsigned->signed conversion |
| 28 | ; |
| 29 | ; GLOBAL(void) |
| 30 | ; jsimd_convsamp_float_sse2 (JSAMPARRAY sample_data, JDIMENSION start_col, |
| 31 | ; FAST_FLOAT * workspace); |
| 32 | ; |
| 33 | |
| 34 | ; r10 = JSAMPARRAY sample_data |
| 35 | ; r11 = JDIMENSION start_col |
| 36 | ; r12 = FAST_FLOAT * workspace |
| 37 | |
| 38 | align 16 |
| 39 | global EXTN(jsimd_convsamp_float_sse2) |
| 40 | |
| 41 | EXTN(jsimd_convsamp_float_sse2): |
| 42 | push rbp |
| 43 | mov rbp,rsp |
| 44 | push rbx |
| 45 | collect_args |
| 46 | |
| 47 | pcmpeqw xmm7,xmm7 |
| 48 | psllw xmm7,7 |
| 49 | packsswb xmm7,xmm7 ; xmm7 = PB_CENTERJSAMPLE (0x808080..) |
| 50 | |
| 51 | mov rsi, r10 |
| 52 | mov rax, r11 |
| 53 | mov rdi, r12 |
| 54 | mov rcx, DCTSIZE/2 |
| 55 | .convloop: |
| 56 | mov rbx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) |
| 57 | mov rdx, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) |
| 58 | |
| 59 | movq xmm0, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE] |
| 60 | movq xmm1, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE] |
| 61 | |
| 62 | psubb xmm0,xmm7 ; xmm0=(01234567) |
| 63 | psubb xmm1,xmm7 ; xmm1=(89ABCDEF) |
| 64 | |
| 65 | punpcklbw xmm0,xmm0 ; xmm0=(*0*1*2*3*4*5*6*7) |
| 66 | punpcklbw xmm1,xmm1 ; xmm1=(*8*9*A*B*C*D*E*F) |
| 67 | |
| 68 | punpcklwd xmm2,xmm0 ; xmm2=(***0***1***2***3) |
| 69 | punpckhwd xmm0,xmm0 ; xmm0=(***4***5***6***7) |
| 70 | punpcklwd xmm3,xmm1 ; xmm3=(***8***9***A***B) |
| 71 | punpckhwd xmm1,xmm1 ; xmm1=(***C***D***E***F) |
| 72 | |
| 73 | psrad xmm2,(DWORD_BIT-BYTE_BIT) ; xmm2=(0123) |
| 74 | psrad xmm0,(DWORD_BIT-BYTE_BIT) ; xmm0=(4567) |
| 75 | cvtdq2ps xmm2,xmm2 ; xmm2=(0123) |
| 76 | cvtdq2ps xmm0,xmm0 ; xmm0=(4567) |
| 77 | psrad xmm3,(DWORD_BIT-BYTE_BIT) ; xmm3=(89AB) |
| 78 | psrad xmm1,(DWORD_BIT-BYTE_BIT) ; xmm1=(CDEF) |
| 79 | cvtdq2ps xmm3,xmm3 ; xmm3=(89AB) |
| 80 | cvtdq2ps xmm1,xmm1 ; xmm1=(CDEF) |
| 81 | |
| 82 | movaps XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_FAST_FLOAT)], xmm2 |
| 83 | movaps XMMWORD [XMMBLOCK(0,1,rdi,SIZEOF_FAST_FLOAT)], xmm0 |
| 84 | movaps XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_FAST_FLOAT)], xmm3 |
| 85 | movaps XMMWORD [XMMBLOCK(1,1,rdi,SIZEOF_FAST_FLOAT)], xmm1 |
| 86 | |
| 87 | add rsi, byte 2*SIZEOF_JSAMPROW |
| 88 | add rdi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT |
| 89 | dec rcx |
| 90 | jnz short .convloop |
| 91 | |
| 92 | uncollect_args |
| 93 | pop rbx |
| 94 | pop rbp |
| 95 | ret |
| 96 | |
| 97 | |
| 98 | ; -------------------------------------------------------------------------- |
| 99 | ; |
| 100 | ; Quantize/descale the coefficients, and store into coef_block |
| 101 | ; |
| 102 | ; GLOBAL(void) |
| 103 | ; jsimd_quantize_float_sse2 (JCOEFPTR coef_block, FAST_FLOAT * divisors, |
| 104 | ; FAST_FLOAT * workspace); |
| 105 | ; |
| 106 | |
| 107 | ; r10 = JCOEFPTR coef_block |
| 108 | ; r11 = FAST_FLOAT * divisors |
| 109 | ; r12 = FAST_FLOAT * workspace |
| 110 | |
| 111 | align 16 |
| 112 | global EXTN(jsimd_quantize_float_sse2) |
| 113 | |
| 114 | EXTN(jsimd_quantize_float_sse2): |
| 115 | push rbp |
| 116 | mov rbp,rsp |
| 117 | collect_args |
| 118 | |
| 119 | mov rsi, r12 |
| 120 | mov rdx, r11 |
| 121 | mov rdi, r10 |
| 122 | mov rax, DCTSIZE2/16 |
| 123 | .quantloop: |
| 124 | movaps xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_FAST_FLOAT)] |
| 125 | movaps xmm1, XMMWORD [XMMBLOCK(0,1,rsi,SIZEOF_FAST_FLOAT)] |
| 126 | mulps xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)] |
| 127 | mulps xmm1, XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)] |
| 128 | movaps xmm2, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_FAST_FLOAT)] |
| 129 | movaps xmm3, XMMWORD [XMMBLOCK(1,1,rsi,SIZEOF_FAST_FLOAT)] |
| 130 | mulps xmm2, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)] |
| 131 | mulps xmm3, XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)] |
| 132 | |
| 133 | cvtps2dq xmm0,xmm0 |
| 134 | cvtps2dq xmm1,xmm1 |
| 135 | cvtps2dq xmm2,xmm2 |
| 136 | cvtps2dq xmm3,xmm3 |
| 137 | |
| 138 | packssdw xmm0,xmm1 |
| 139 | packssdw xmm2,xmm3 |
| 140 | |
| 141 | movdqa XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_JCOEF)], xmm0 |
| 142 | movdqa XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_JCOEF)], xmm2 |
| 143 | |
| 144 | add rsi, byte 16*SIZEOF_FAST_FLOAT |
| 145 | add rdx, byte 16*SIZEOF_FAST_FLOAT |
| 146 | add rdi, byte 16*SIZEOF_JCOEF |
| 147 | dec rax |
| 148 | jnz short .quantloop |
| 149 | |
| 150 | uncollect_args |
| 151 | pop rbp |
| 152 | ret |