The Tor Browser / file comparison
comparison: media/libtheora/lib/x86_vc/mmxloop.h
media/libtheora/lib/x86_vc/mmxloop.h
- branch
- TOR_BUG_9701
- changeset 15
- b8a032363ba2
equal
deleted
inserted
replaced
| -1:000000000000 | 0:2fdb98987476 |
|---|---|
| 1 #if !defined(_x86_vc_mmxloop_H) | |
| 2 # define _x86_vc_mmxloop_H (1) | |
| 3 # include <stddef.h> | |
| 4 # include "x86int.h" | |
| 5 | |
| 6 #if defined(OC_X86_ASM) | |
| 7 | |
| 8 /*On entry, mm0={a0,...,a7}, mm1={b0,...,b7}, mm2={c0,...,c7}, mm3={d0,...d7}. | |
| 9 On exit, mm1={b0+lflim(R_0,L),...,b7+lflim(R_7,L)} and | |
| 10 mm2={c0-lflim(R_0,L),...,c7-lflim(R_7,L)}; mm0 and mm3 are clobbered.*/ | |
| 11 #define OC_LOOP_FILTER8_MMX __asm{ \ | |
| 12 /*mm7=0*/ \ | |
| 13 __asm pxor mm7,mm7 \ | |
| 14 /*mm6:mm0={a0,...,a7}*/ \ | |
| 15 __asm movq mm6,mm0 \ | |
| 16 __asm punpcklbw mm0,mm7 \ | |
| 17 __asm punpckhbw mm6,mm7 \ | |
| 18 /*mm3:mm5={d0,...,d7}*/ \ | |
| 19 __asm movq mm5,mm3 \ | |
| 20 __asm punpcklbw mm3,mm7 \ | |
| 21 __asm punpckhbw mm5,mm7 \ | |
| 22 /*mm6:mm0={a0-d0,...,a7-d7}*/ \ | |
| 23 __asm psubw mm0,mm3 \ | |
| 24 __asm psubw mm6,mm5 \ | |
| 25 /*mm3:mm1={b0,...,b7}*/ \ | |
| 26 __asm movq mm3,mm1 \ | |
| 27 __asm punpcklbw mm1,mm7 \ | |
| 28 __asm movq mm4,mm2 \ | |
| 29 __asm punpckhbw mm3,mm7 \ | |
| 30 /*mm5:mm4={c0,...,c7}*/ \ | |
| 31 __asm movq mm5,mm2 \ | |
| 32 __asm punpcklbw mm4,mm7 \ | |
| 33 __asm punpckhbw mm5,mm7 \ | |
| 34 /*mm7={3}x4 \ | |
| 35 mm5:mm4={c0-b0,...,c7-b7}*/ \ | |
| 36 __asm pcmpeqw mm7,mm7 \ | |
| 37 __asm psubw mm4,mm1 \ | |
| 38 __asm psrlw mm7,14 \ | |
| 39 __asm psubw mm5,mm3 \ | |
| 40 /*Scale by 3.*/ \ | |
| 41 __asm pmullw mm4,mm7 \ | |
| 42 __asm pmullw mm5,mm7 \ | |
| 43 /*mm7={4}x4 \ | |
| 44 mm5:mm4=f={a0-d0+3*(c0-b0),...,a7-d7+3*(c7-b7)}*/ \ | |
| 45 __asm psrlw mm7,1 \ | |
| 46 __asm paddw mm4,mm0 \ | |
| 47 __asm psllw mm7,2 \ | |
| 48 __asm movq mm0,[LL] \ | |
| 49 __asm paddw mm5,mm6 \ | |
| 50 /*R_i has the range [-127,128], so we compute -R_i instead. \ | |
| 51 mm4=-R_i=-(f+4>>3)=0xFF^(f-4>>3)*/ \ | |
| 52 __asm psubw mm4,mm7 \ | |
| 53 __asm psubw mm5,mm7 \ | |
| 54 __asm psraw mm4,3 \ | |
| 55 __asm psraw mm5,3 \ | |
| 56 __asm pcmpeqb mm7,mm7 \ | |
| 57 __asm packsswb mm4,mm5 \ | |
| 58 __asm pxor mm6,mm6 \ | |
| 59 __asm pxor mm4,mm7 \ | |
| 60 __asm packuswb mm1,mm3 \ | |
| 61 /*Now compute lflim of -mm4 cf. Section 7.10 of the sepc.*/ \ | |
| 62 /*There's no unsigned byte+signed byte with unsigned saturation op code, so \ | |
| 63 we have to split things by sign (the other option is to work in 16 bits, \ | |
| 64 but working in 8 bits gives much better parallelism). \ | |
| 65 We compute abs(R_i), but save a mask of which terms were negative in mm6. \ | |
| 66 Then we compute mm4=abs(lflim(R_i,L))=min(abs(R_i),max(2*L-abs(R_i),0)). \ | |
| 67 Finally, we split mm4 into positive and negative pieces using the mask in \ | |
| 68 mm6, and add and subtract them as appropriate.*/ \ | |
| 69 /*mm4=abs(-R_i)*/ \ | |
| 70 /*mm7=255-2*L*/ \ | |
| 71 __asm pcmpgtb mm6,mm4 \ | |
| 72 __asm psubb mm7,mm0 \ | |
| 73 __asm pxor mm4,mm6 \ | |
| 74 __asm psubb mm7,mm0 \ | |
| 75 __asm psubb mm4,mm6 \ | |
| 76 /*mm7=255-max(2*L-abs(R_i),0)*/ \ | |
| 77 __asm paddusb mm7,mm4 \ | |
| 78 /*mm4=min(abs(R_i),max(2*L-abs(R_i),0))*/ \ | |
| 79 __asm paddusb mm4,mm7 \ | |
| 80 __asm psubusb mm4,mm7 \ | |
| 81 /*Now split mm4 by the original sign of -R_i.*/ \ | |
| 82 __asm movq mm5,mm4 \ | |
| 83 __asm pand mm4,mm6 \ | |
| 84 __asm pandn mm6,mm5 \ | |
| 85 /*mm1={b0+lflim(R_0,L),...,b7+lflim(R_7,L)}*/ \ | |
| 86 /*mm2={c0-lflim(R_0,L),...,c7-lflim(R_7,L)}*/ \ | |
| 87 __asm paddusb mm1,mm4 \ | |
| 88 __asm psubusb mm2,mm4 \ | |
| 89 __asm psubusb mm1,mm6 \ | |
| 90 __asm paddusb mm2,mm6 \ | |
| 91 } | |
| 92 | |
| 93 #define OC_LOOP_FILTER_V_MMX(_pix,_ystride,_ll) \ | |
| 94 do{ \ | |
| 95 /*Used local variable pix__ in order to fix compilation errors like: \ | |
| 96 "error C2425: 'SHL' : non-constant expression in 'second operand'".*/ \ | |
| 97 unsigned char *pix__; \ | |
| 98 unsigned char *ll__; \ | |
| 99 ll__=(_ll); \ | |
| 100 pix__=(_pix); \ | |
| 101 __asm mov YSTRIDE,_ystride \ | |
| 102 __asm mov LL,ll__ \ | |
| 103 __asm mov PIX,pix__ \ | |
| 104 __asm sub PIX,YSTRIDE \ | |
| 105 __asm sub PIX,YSTRIDE \ | |
| 106 /*mm0={a0,...,a7}*/ \ | |
| 107 __asm movq mm0,[PIX] \ | |
| 108 /*ystride3=_ystride*3*/ \ | |
| 109 __asm lea YSTRIDE3,[YSTRIDE+YSTRIDE*2] \ | |
| 110 /*mm3={d0,...,d7}*/ \ | |
| 111 __asm movq mm3,[PIX+YSTRIDE3] \ | |
| 112 /*mm1={b0,...,b7}*/ \ | |
| 113 __asm movq mm1,[PIX+YSTRIDE] \ | |
| 114 /*mm2={c0,...,c7}*/ \ | |
| 115 __asm movq mm2,[PIX+YSTRIDE*2] \ | |
| 116 OC_LOOP_FILTER8_MMX \ | |
| 117 /*Write it back out.*/ \ | |
| 118 __asm movq [PIX+YSTRIDE],mm1 \ | |
| 119 __asm movq [PIX+YSTRIDE*2],mm2 \ | |
| 120 } \ | |
| 121 while(0) | |
| 122 | |
| 123 #define OC_LOOP_FILTER_H_MMX(_pix,_ystride,_ll) \ | |
| 124 do{ \ | |
| 125 /*Used local variable ll__ in order to fix compilation errors like: \ | |
| 126 "error C2443: operand size conflict".*/ \ | |
| 127 unsigned char *ll__; \ | |
| 128 unsigned char *pix__; \ | |
| 129 ll__=(_ll); \ | |
| 130 pix__=(_pix)-2; \ | |
| 131 __asm mov PIX,pix__ \ | |
| 132 __asm mov YSTRIDE,_ystride \ | |
| 133 __asm mov LL,ll__ \ | |
| 134 /*x x x x d0 c0 b0 a0*/ \ | |
| 135 __asm movd mm0,[PIX] \ | |
| 136 /*x x x x d1 c1 b1 a1*/ \ | |
| 137 __asm movd mm1,[PIX+YSTRIDE] \ | |
| 138 /*ystride3=_ystride*3*/ \ | |
| 139 __asm lea YSTRIDE3,[YSTRIDE+YSTRIDE*2] \ | |
| 140 /*x x x x d2 c2 b2 a2*/ \ | |
| 141 __asm movd mm2,[PIX+YSTRIDE*2] \ | |
| 142 /*x x x x d3 c3 b3 a3*/ \ | |
| 143 __asm lea D,[PIX+YSTRIDE*4] \ | |
| 144 __asm movd mm3,[PIX+YSTRIDE3] \ | |
| 145 /*x x x x d4 c4 b4 a4*/ \ | |
| 146 __asm movd mm4,[D] \ | |
| 147 /*x x x x d5 c5 b5 a5*/ \ | |
| 148 __asm movd mm5,[D+YSTRIDE] \ | |
| 149 /*x x x x d6 c6 b6 a6*/ \ | |
| 150 __asm movd mm6,[D+YSTRIDE*2] \ | |
| 151 /*x x x x d7 c7 b7 a7*/ \ | |
| 152 __asm movd mm7,[D+YSTRIDE3] \ | |
| 153 /*mm0=d1 d0 c1 c0 b1 b0 a1 a0*/ \ | |
| 154 __asm punpcklbw mm0,mm1 \ | |
| 155 /*mm2=d3 d2 c3 c2 b3 b2 a3 a2*/ \ | |
| 156 __asm punpcklbw mm2,mm3 \ | |
| 157 /*mm3=d1 d0 c1 c0 b1 b0 a1 a0*/ \ | |
| 158 __asm movq mm3,mm0 \ | |
| 159 /*mm0=b3 b2 b1 b0 a3 a2 a1 a0*/ \ | |
| 160 __asm punpcklwd mm0,mm2 \ | |
| 161 /*mm3=d3 d2 d1 d0 c3 c2 c1 c0*/ \ | |
| 162 __asm punpckhwd mm3,mm2 \ | |
| 163 /*mm1=b3 b2 b1 b0 a3 a2 a1 a0*/ \ | |
| 164 __asm movq mm1,mm0 \ | |
| 165 /*mm4=d5 d4 c5 c4 b5 b4 a5 a4*/ \ | |
| 166 __asm punpcklbw mm4,mm5 \ | |
| 167 /*mm6=d7 d6 c7 c6 b7 b6 a7 a6*/ \ | |
| 168 __asm punpcklbw mm6,mm7 \ | |
| 169 /*mm5=d5 d4 c5 c4 b5 b4 a5 a4*/ \ | |
| 170 __asm movq mm5,mm4 \ | |
| 171 /*mm4=b7 b6 b5 b4 a7 a6 a5 a4*/ \ | |
| 172 __asm punpcklwd mm4,mm6 \ | |
| 173 /*mm5=d7 d6 d5 d4 c7 c6 c5 c4*/ \ | |
| 174 __asm punpckhwd mm5,mm6 \ | |
| 175 /*mm2=d3 d2 d1 d0 c3 c2 c1 c0*/ \ | |
| 176 __asm movq mm2,mm3 \ | |
| 177 /*mm0=a7 a6 a5 a4 a3 a2 a1 a0*/ \ | |
| 178 __asm punpckldq mm0,mm4 \ | |
| 179 /*mm1=b7 b6 b5 b4 b3 b2 b1 b0*/ \ | |
| 180 __asm punpckhdq mm1,mm4 \ | |
| 181 /*mm2=c7 c6 c5 c4 c3 c2 c1 c0*/ \ | |
| 182 __asm punpckldq mm2,mm5 \ | |
| 183 /*mm3=d7 d6 d5 d4 d3 d2 d1 d0*/ \ | |
| 184 __asm punpckhdq mm3,mm5 \ | |
| 185 OC_LOOP_FILTER8_MMX \ | |
| 186 /*mm2={b0+R_0'',...,b7+R_7''}*/ \ | |
| 187 __asm movq mm0,mm1 \ | |
| 188 /*mm1={b0+R_0'',c0-R_0'',...,b3+R_3'',c3-R_3''}*/ \ | |
| 189 __asm punpcklbw mm1,mm2 \ | |
| 190 /*mm2={b4+R_4'',c4-R_4'',...,b7+R_7'',c7-R_7''}*/ \ | |
| 191 __asm punpckhbw mm0,mm2 \ | |
| 192 /*[d]=c1 b1 c0 b0*/ \ | |
| 193 __asm movd D,mm1 \ | |
| 194 __asm mov [PIX+1],D_WORD \ | |
| 195 __asm psrlq mm1,32 \ | |
| 196 __asm shr D,16 \ | |
| 197 __asm mov [PIX+YSTRIDE+1],D_WORD \ | |
| 198 /*[d]=c3 b3 c2 b2*/ \ | |
| 199 __asm movd D,mm1 \ | |
| 200 __asm mov [PIX+YSTRIDE*2+1],D_WORD \ | |
| 201 __asm shr D,16 \ | |
| 202 __asm mov [PIX+YSTRIDE3+1],D_WORD \ | |
| 203 __asm lea PIX,[PIX+YSTRIDE*4] \ | |
| 204 /*[d]=c5 b5 c4 b4*/ \ | |
| 205 __asm movd D,mm0 \ | |
| 206 __asm mov [PIX+1],D_WORD \ | |
| 207 __asm psrlq mm0,32 \ | |
| 208 __asm shr D,16 \ | |
| 209 __asm mov [PIX+YSTRIDE+1],D_WORD \ | |
| 210 /*[d]=c7 b7 c6 b6*/ \ | |
| 211 __asm movd D,mm0 \ | |
| 212 __asm mov [PIX+YSTRIDE*2+1],D_WORD \ | |
| 213 __asm shr D,16 \ | |
| 214 __asm mov [PIX+YSTRIDE3+1],D_WORD \ | |
| 215 } \ | |
| 216 while(0) | |
| 217 | |
| 218 # endif | |
| 219 #endif |
