media/libjpeg/simd/jcqnt3dn.asm@b8a032363ba2 (annotated)
media/libjpeg/simd/jcqnt3dn.asm
Thu, 22 Jan 2015 13:21:57 +0100
- author
- Michael Schloh von Bennewitz <michael@schloh.com>
- date
- Thu, 22 Jan 2015 13:21:57 +0100
- branch
- TOR_BUG_9701
- changeset 15
- b8a032363ba2
- permissions
- -rw-r--r--
Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6
| michael@0 | 1 | ; |
| michael@0 | 2 | ; jcqnt3dn.asm - sample data conversion and quantization (3DNow! & MMX) |
| michael@0 | 3 | ; |
| michael@0 | 4 | ; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB |
| michael@0 | 5 | ; |
| michael@0 | 6 | ; Based on |
| michael@0 | 7 | ; x86 SIMD extension for IJG JPEG library |
| michael@0 | 8 | ; Copyright (C) 1999-2006, MIYASAKA Masaru. |
| michael@0 | 9 | ; For conditions of distribution and use, see copyright notice in jsimdext.inc |
| michael@0 | 10 | ; |
| michael@0 | 11 | ; This file should be assembled with NASM (Netwide Assembler), |
| michael@0 | 12 | ; can *not* be assembled with Microsoft's MASM or any compatible |
| michael@0 | 13 | ; assembler (including Borland's Turbo Assembler). |
| michael@0 | 14 | ; NASM is available from http://nasm.sourceforge.net/ or |
| michael@0 | 15 | ; http://sourceforge.net/project/showfiles.php?group_id=6208 |
| michael@0 | 16 | ; |
| michael@0 | 17 | ; [TAB8] |
| michael@0 | 18 | |
| michael@0 | 19 | %include "jsimdext.inc" |
| michael@0 | 20 | %include "jdct.inc" |
| michael@0 | 21 | |
| michael@0 | 22 | ; -------------------------------------------------------------------------- |
| michael@0 | 23 | SECTION SEG_TEXT |
| michael@0 | 24 | BITS 32 |
| michael@0 | 25 | ; |
| michael@0 | 26 | ; Load data into workspace, applying unsigned->signed conversion |
| michael@0 | 27 | ; |
| michael@0 | 28 | ; GLOBAL(void) |
| michael@0 | 29 | ; jsimd_convsamp_float_3dnow (JSAMPARRAY sample_data, JDIMENSION start_col, |
| michael@0 | 30 | ; FAST_FLOAT * workspace); |
| michael@0 | 31 | ; |
| michael@0 | 32 | |
| michael@0 | 33 | %define sample_data ebp+8 ; JSAMPARRAY sample_data |
| michael@0 | 34 | %define start_col ebp+12 ; JDIMENSION start_col |
| michael@0 | 35 | %define workspace ebp+16 ; FAST_FLOAT * workspace |
| michael@0 | 36 | |
| michael@0 | 37 | align 16 |
| michael@0 | 38 | global EXTN(jsimd_convsamp_float_3dnow) |
| michael@0 | 39 | |
| michael@0 | 40 | EXTN(jsimd_convsamp_float_3dnow): |
| michael@0 | 41 | push ebp |
| michael@0 | 42 | mov ebp,esp |
| michael@0 | 43 | push ebx |
| michael@0 | 44 | ; push ecx ; need not be preserved |
| michael@0 | 45 | ; push edx ; need not be preserved |
| michael@0 | 46 | push esi |
| michael@0 | 47 | push edi |
| michael@0 | 48 | |
| michael@0 | 49 | pcmpeqw mm7,mm7 |
| michael@0 | 50 | psllw mm7,7 |
| michael@0 | 51 | packsswb mm7,mm7 ; mm7 = PB_CENTERJSAMPLE (0x808080..) |
| michael@0 | 52 | |
| michael@0 | 53 | mov esi, JSAMPARRAY [sample_data] ; (JSAMPROW *) |
| michael@0 | 54 | mov eax, JDIMENSION [start_col] |
| michael@0 | 55 | mov edi, POINTER [workspace] ; (DCTELEM *) |
| michael@0 | 56 | mov ecx, DCTSIZE/2 |
| michael@0 | 57 | alignx 16,7 |
| michael@0 | 58 | .convloop: |
| michael@0 | 59 | mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) |
| michael@0 | 60 | mov edx, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) |
| michael@0 | 61 | |
| michael@0 | 62 | movq mm0, MMWORD [ebx+eax*SIZEOF_JSAMPLE] |
| michael@0 | 63 | movq mm1, MMWORD [edx+eax*SIZEOF_JSAMPLE] |
| michael@0 | 64 | |
| michael@0 | 65 | psubb mm0,mm7 ; mm0=(01234567) |
| michael@0 | 66 | psubb mm1,mm7 ; mm1=(89ABCDEF) |
| michael@0 | 67 | |
| michael@0 | 68 | punpcklbw mm2,mm0 ; mm2=(*0*1*2*3) |
| michael@0 | 69 | punpckhbw mm0,mm0 ; mm0=(*4*5*6*7) |
| michael@0 | 70 | punpcklbw mm3,mm1 ; mm3=(*8*9*A*B) |
| michael@0 | 71 | punpckhbw mm1,mm1 ; mm1=(*C*D*E*F) |
| michael@0 | 72 | |
| michael@0 | 73 | punpcklwd mm4,mm2 ; mm4=(***0***1) |
| michael@0 | 74 | punpckhwd mm2,mm2 ; mm2=(***2***3) |
| michael@0 | 75 | punpcklwd mm5,mm0 ; mm5=(***4***5) |
| michael@0 | 76 | punpckhwd mm0,mm0 ; mm0=(***6***7) |
| michael@0 | 77 | |
| michael@0 | 78 | psrad mm4,(DWORD_BIT-BYTE_BIT) ; mm4=(01) |
| michael@0 | 79 | psrad mm2,(DWORD_BIT-BYTE_BIT) ; mm2=(23) |
| michael@0 | 80 | pi2fd mm4,mm4 |
| michael@0 | 81 | pi2fd mm2,mm2 |
| michael@0 | 82 | psrad mm5,(DWORD_BIT-BYTE_BIT) ; mm5=(45) |
| michael@0 | 83 | psrad mm0,(DWORD_BIT-BYTE_BIT) ; mm0=(67) |
| michael@0 | 84 | pi2fd mm5,mm5 |
| michael@0 | 85 | pi2fd mm0,mm0 |
| michael@0 | 86 | |
| michael@0 | 87 | movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], mm4 |
| michael@0 | 88 | movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], mm2 |
| michael@0 | 89 | movq MMWORD [MMBLOCK(0,2,edi,SIZEOF_FAST_FLOAT)], mm5 |
| michael@0 | 90 | movq MMWORD [MMBLOCK(0,3,edi,SIZEOF_FAST_FLOAT)], mm0 |
| michael@0 | 91 | |
| michael@0 | 92 | punpcklwd mm6,mm3 ; mm6=(***8***9) |
| michael@0 | 93 | punpckhwd mm3,mm3 ; mm3=(***A***B) |
| michael@0 | 94 | punpcklwd mm4,mm1 ; mm4=(***C***D) |
| michael@0 | 95 | punpckhwd mm1,mm1 ; mm1=(***E***F) |
| michael@0 | 96 | |
| michael@0 | 97 | psrad mm6,(DWORD_BIT-BYTE_BIT) ; mm6=(89) |
| michael@0 | 98 | psrad mm3,(DWORD_BIT-BYTE_BIT) ; mm3=(AB) |
| michael@0 | 99 | pi2fd mm6,mm6 |
| michael@0 | 100 | pi2fd mm3,mm3 |
| michael@0 | 101 | psrad mm4,(DWORD_BIT-BYTE_BIT) ; mm4=(CD) |
| michael@0 | 102 | psrad mm1,(DWORD_BIT-BYTE_BIT) ; mm1=(EF) |
| michael@0 | 103 | pi2fd mm4,mm4 |
| michael@0 | 104 | pi2fd mm1,mm1 |
| michael@0 | 105 | |
| michael@0 | 106 | movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], mm6 |
| michael@0 | 107 | movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], mm3 |
| michael@0 | 108 | movq MMWORD [MMBLOCK(1,2,edi,SIZEOF_FAST_FLOAT)], mm4 |
| michael@0 | 109 | movq MMWORD [MMBLOCK(1,3,edi,SIZEOF_FAST_FLOAT)], mm1 |
| michael@0 | 110 | |
| michael@0 | 111 | add esi, byte 2*SIZEOF_JSAMPROW |
| michael@0 | 112 | add edi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT |
| michael@0 | 113 | dec ecx |
| michael@0 | 114 | jnz near .convloop |
| michael@0 | 115 | |
| michael@0 | 116 | femms ; empty MMX/3DNow! state |
| michael@0 | 117 | |
| michael@0 | 118 | pop edi |
| michael@0 | 119 | pop esi |
| michael@0 | 120 | ; pop edx ; need not be preserved |
| michael@0 | 121 | ; pop ecx ; need not be preserved |
| michael@0 | 122 | pop ebx |
| michael@0 | 123 | pop ebp |
| michael@0 | 124 | ret |
| michael@0 | 125 | |
| michael@0 | 126 | |
| michael@0 | 127 | ; -------------------------------------------------------------------------- |
| michael@0 | 128 | ; |
| michael@0 | 129 | ; Quantize/descale the coefficients, and store into coef_block |
| michael@0 | 130 | ; |
| michael@0 | 131 | ; GLOBAL(void) |
| michael@0 | 132 | ; jsimd_quantize_float_3dnow (JCOEFPTR coef_block, FAST_FLOAT * divisors, |
| michael@0 | 133 | ; FAST_FLOAT * workspace); |
| michael@0 | 134 | ; |
| michael@0 | 135 | |
| michael@0 | 136 | %define coef_block ebp+8 ; JCOEFPTR coef_block |
| michael@0 | 137 | %define divisors ebp+12 ; FAST_FLOAT * divisors |
| michael@0 | 138 | %define workspace ebp+16 ; FAST_FLOAT * workspace |
| michael@0 | 139 | |
| michael@0 | 140 | align 16 |
| michael@0 | 141 | global EXTN(jsimd_quantize_float_3dnow) |
| michael@0 | 142 | |
| michael@0 | 143 | EXTN(jsimd_quantize_float_3dnow): |
| michael@0 | 144 | push ebp |
| michael@0 | 145 | mov ebp,esp |
| michael@0 | 146 | ; push ebx ; unused |
| michael@0 | 147 | ; push ecx ; unused |
| michael@0 | 148 | ; push edx ; need not be preserved |
| michael@0 | 149 | push esi |
| michael@0 | 150 | push edi |
| michael@0 | 151 | |
| michael@0 | 152 | mov eax, 0x4B400000 ; (float)0x00C00000 (rndint_magic) |
| michael@0 | 153 | movd mm7,eax |
| michael@0 | 154 | punpckldq mm7,mm7 ; mm7={12582912.0F 12582912.0F} |
| michael@0 | 155 | |
| michael@0 | 156 | mov esi, POINTER [workspace] |
| michael@0 | 157 | mov edx, POINTER [divisors] |
| michael@0 | 158 | mov edi, JCOEFPTR [coef_block] |
| michael@0 | 159 | mov eax, DCTSIZE2/16 |
| michael@0 | 160 | alignx 16,7 |
| michael@0 | 161 | .quantloop: |
| michael@0 | 162 | movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_FAST_FLOAT)] |
| michael@0 | 163 | movq mm1, MMWORD [MMBLOCK(0,1,esi,SIZEOF_FAST_FLOAT)] |
| michael@0 | 164 | pfmul mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)] |
| michael@0 | 165 | pfmul mm1, MMWORD [MMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)] |
| michael@0 | 166 | movq mm2, MMWORD [MMBLOCK(0,2,esi,SIZEOF_FAST_FLOAT)] |
| michael@0 | 167 | movq mm3, MMWORD [MMBLOCK(0,3,esi,SIZEOF_FAST_FLOAT)] |
| michael@0 | 168 | pfmul mm2, MMWORD [MMBLOCK(0,2,edx,SIZEOF_FAST_FLOAT)] |
| michael@0 | 169 | pfmul mm3, MMWORD [MMBLOCK(0,3,edx,SIZEOF_FAST_FLOAT)] |
| michael@0 | 170 | |
| michael@0 | 171 | pfadd mm0,mm7 ; mm0=(00 ** 01 **) |
| michael@0 | 172 | pfadd mm1,mm7 ; mm1=(02 ** 03 **) |
| michael@0 | 173 | pfadd mm2,mm7 ; mm0=(04 ** 05 **) |
| michael@0 | 174 | pfadd mm3,mm7 ; mm1=(06 ** 07 **) |
| michael@0 | 175 | |
| michael@0 | 176 | movq mm4,mm0 |
| michael@0 | 177 | punpcklwd mm0,mm1 ; mm0=(00 02 ** **) |
| michael@0 | 178 | punpckhwd mm4,mm1 ; mm4=(01 03 ** **) |
| michael@0 | 179 | movq mm5,mm2 |
| michael@0 | 180 | punpcklwd mm2,mm3 ; mm2=(04 06 ** **) |
| michael@0 | 181 | punpckhwd mm5,mm3 ; mm5=(05 07 ** **) |
| michael@0 | 182 | |
| michael@0 | 183 | punpcklwd mm0,mm4 ; mm0=(00 01 02 03) |
| michael@0 | 184 | punpcklwd mm2,mm5 ; mm2=(04 05 06 07) |
| michael@0 | 185 | |
| michael@0 | 186 | movq mm6, MMWORD [MMBLOCK(1,0,esi,SIZEOF_FAST_FLOAT)] |
| michael@0 | 187 | movq mm1, MMWORD [MMBLOCK(1,1,esi,SIZEOF_FAST_FLOAT)] |
| michael@0 | 188 | pfmul mm6, MMWORD [MMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)] |
| michael@0 | 189 | pfmul mm1, MMWORD [MMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)] |
| michael@0 | 190 | movq mm3, MMWORD [MMBLOCK(1,2,esi,SIZEOF_FAST_FLOAT)] |
| michael@0 | 191 | movq mm4, MMWORD [MMBLOCK(1,3,esi,SIZEOF_FAST_FLOAT)] |
| michael@0 | 192 | pfmul mm3, MMWORD [MMBLOCK(1,2,edx,SIZEOF_FAST_FLOAT)] |
| michael@0 | 193 | pfmul mm4, MMWORD [MMBLOCK(1,3,edx,SIZEOF_FAST_FLOAT)] |
| michael@0 | 194 | |
| michael@0 | 195 | pfadd mm6,mm7 ; mm0=(10 ** 11 **) |
| michael@0 | 196 | pfadd mm1,mm7 ; mm4=(12 ** 13 **) |
| michael@0 | 197 | pfadd mm3,mm7 ; mm0=(14 ** 15 **) |
| michael@0 | 198 | pfadd mm4,mm7 ; mm4=(16 ** 17 **) |
| michael@0 | 199 | |
| michael@0 | 200 | movq mm5,mm6 |
| michael@0 | 201 | punpcklwd mm6,mm1 ; mm6=(10 12 ** **) |
| michael@0 | 202 | punpckhwd mm5,mm1 ; mm5=(11 13 ** **) |
| michael@0 | 203 | movq mm1,mm3 |
| michael@0 | 204 | punpcklwd mm3,mm4 ; mm3=(14 16 ** **) |
| michael@0 | 205 | punpckhwd mm1,mm4 ; mm1=(15 17 ** **) |
| michael@0 | 206 | |
| michael@0 | 207 | punpcklwd mm6,mm5 ; mm6=(10 11 12 13) |
| michael@0 | 208 | punpcklwd mm3,mm1 ; mm3=(14 15 16 17) |
| michael@0 | 209 | |
| michael@0 | 210 | movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm0 |
| michael@0 | 211 | movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_JCOEF)], mm2 |
| michael@0 | 212 | movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm6 |
| michael@0 | 213 | movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_JCOEF)], mm3 |
| michael@0 | 214 | |
| michael@0 | 215 | add esi, byte 16*SIZEOF_FAST_FLOAT |
| michael@0 | 216 | add edx, byte 16*SIZEOF_FAST_FLOAT |
| michael@0 | 217 | add edi, byte 16*SIZEOF_JCOEF |
| michael@0 | 218 | dec eax |
| michael@0 | 219 | jnz near .quantloop |
| michael@0 | 220 | |
| michael@0 | 221 | femms ; empty MMX/3DNow! state |
| michael@0 | 222 | |
| michael@0 | 223 | pop edi |
| michael@0 | 224 | pop esi |
| michael@0 | 225 | ; pop edx ; need not be preserved |
| michael@0 | 226 | ; pop ecx ; unused |
| michael@0 | 227 | ; pop ebx ; unused |
| michael@0 | 228 | pop ebp |
| michael@0 | 229 | ret |
| michael@0 | 230 | |
| michael@0 | 231 | ; For some reason, the OS X linker does not honor the request to align the |
| michael@0 | 232 | ; segment unless we do this. |
| michael@0 | 233 | align 16 |
