yading@10: ;****************************************************************************** yading@10: ;* MMX/SSE2-optimized functions for the VP3 decoder yading@10: ;* Copyright (c) 2007 Aurelien Jacobs yading@10: ;* yading@10: ;* This file is part of FFmpeg. yading@10: ;* yading@10: ;* FFmpeg is free software; you can redistribute it and/or yading@10: ;* modify it under the terms of the GNU Lesser General Public yading@10: ;* License as published by the Free Software Foundation; either yading@10: ;* version 2.1 of the License, or (at your option) any later version. yading@10: ;* yading@10: ;* FFmpeg is distributed in the hope that it will be useful, yading@10: ;* but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: ;* Lesser General Public License for more details. yading@10: ;* yading@10: ;* You should have received a copy of the GNU Lesser General Public yading@10: ;* License along with FFmpeg; if not, write to the Free Software yading@10: ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: ;****************************************************************************** yading@10: yading@10: %include "libavutil/x86/x86util.asm" yading@10: yading@10: ; MMX-optimized functions cribbed from the original VP3 source code. yading@10: yading@10: SECTION_RODATA yading@10: yading@10: vp3_idct_data: times 8 dw 64277 yading@10: times 8 dw 60547 yading@10: times 8 dw 54491 yading@10: times 8 dw 46341 yading@10: times 8 dw 36410 yading@10: times 8 dw 25080 yading@10: times 8 dw 12785 yading@10: yading@10: pb_7: times 8 db 0x07 yading@10: pb_1F: times 8 db 0x1f yading@10: pb_81: times 8 db 0x81 yading@10: yading@10: cextern pb_1 yading@10: cextern pb_3 yading@10: cextern pb_80 yading@10: yading@10: cextern pw_8 yading@10: yading@10: SECTION .text yading@10: yading@10: ; this is off by one or two for some cases when filter_limit is greater than 63 yading@10: ; in: p0 in mm6, p1 in mm4, p2 in mm2, p3 in mm1 yading@10: ; out: p1 in mm4, p2 in mm3 yading@10: %macro VP3_LOOP_FILTER 0 yading@10: movq m7, m6 yading@10: pand m6, [pb_7] ; p0&7 yading@10: psrlw m7, 3 yading@10: pand m7, [pb_1F] ; p0>>3 yading@10: movq m3, m2 ; p2 yading@10: pxor m2, m4 yading@10: pand m2, [pb_1] ; (p2^p1)&1 yading@10: movq m5, m2 yading@10: paddb m2, m2 yading@10: paddb m2, m5 ; 3*(p2^p1)&1 yading@10: paddb m2, m6 ; extra bits lost in shifts yading@10: pcmpeqb m0, m0 yading@10: pxor m1, m0 ; 255 - p3 yading@10: pavgb m1, m2 ; (256 - p3 + extrabits) >> 1 yading@10: pxor m0, m4 ; 255 - p1 yading@10: pavgb m0, m3 ; (256 + p2-p1) >> 1 yading@10: paddb m1, [pb_3] yading@10: pavgb m1, m0 ; 128+2+( p2-p1 - p3) >> 2 yading@10: pavgb m1, m0 ; 128+1+(3*(p2-p1) - p3) >> 3 yading@10: paddusb m7, m1 ; d+128+1 yading@10: movq m6, [pb_81] yading@10: psubusb m6, m7 yading@10: psubusb m7, [pb_81] yading@10: yading@10: movq m5, [r2+516] ; flim yading@10: pminub m6, m5 yading@10: pminub m7, m5 yading@10: movq m0, m6 yading@10: movq m1, m7 yading@10: paddb m6, m6 yading@10: paddb m7, m7 yading@10: pminub m6, m5 yading@10: pminub m7, m5 yading@10: psubb m6, m0 yading@10: psubb m7, m1 yading@10: paddusb m4, m7 yading@10: psubusb m4, m6 yading@10: psubusb m3, m7 yading@10: paddusb m3, m6 yading@10: %endmacro yading@10: yading@10: %macro STORE_4_WORDS 1 yading@10: movd r2d, %1 yading@10: mov [r0 -1], r2w yading@10: psrlq %1, 32 yading@10: shr r2, 16 yading@10: mov [r0+r1 -1], r2w yading@10: movd r2d, %1 yading@10: mov [r0+r1*2-1], r2w yading@10: shr r2, 16 yading@10: mov [r0+r3 -1], r2w yading@10: %endmacro yading@10: yading@10: INIT_MMX mmxext yading@10: cglobal vp3_v_loop_filter, 3, 4 yading@10: %if ARCH_X86_64 yading@10: movsxd r1, r1d yading@10: %endif yading@10: mov r3, r1 yading@10: neg r1 yading@10: movq m6, [r0+r1*2] yading@10: movq m4, [r0+r1 ] yading@10: movq m2, [r0 ] yading@10: movq m1, [r0+r3 ] yading@10: yading@10: VP3_LOOP_FILTER yading@10: yading@10: movq [r0+r1], m4 yading@10: movq [r0 ], m3 yading@10: RET yading@10: yading@10: cglobal vp3_h_loop_filter, 3, 4 yading@10: %if ARCH_X86_64 yading@10: movsxd r1, r1d yading@10: %endif yading@10: lea r3, [r1*3] yading@10: yading@10: movd m6, [r0 -2] yading@10: movd m4, [r0+r1 -2] yading@10: movd m2, [r0+r1*2-2] yading@10: movd m1, [r0+r3 -2] yading@10: lea r0, [r0+r1*4 ] yading@10: punpcklbw m6, [r0 -2] yading@10: punpcklbw m4, [r0+r1 -2] yading@10: punpcklbw m2, [r0+r1*2-2] yading@10: punpcklbw m1, [r0+r3 -2] yading@10: sub r0, r3 yading@10: sub r0, r1 yading@10: yading@10: TRANSPOSE4x4B 6, 4, 2, 1, 0 yading@10: VP3_LOOP_FILTER yading@10: SBUTTERFLY bw, 4, 3, 5 yading@10: yading@10: STORE_4_WORDS m4 yading@10: lea r0, [r0+r1*4 ] yading@10: STORE_4_WORDS m3 yading@10: RET yading@10: yading@10: ; from original comments: The Macro does IDct on 4 1-D Dcts yading@10: %macro BeginIDCT 0 yading@10: movq m2, I(3) yading@10: movq m6, C(3) yading@10: movq m4, m2 yading@10: movq m7, J(5) yading@10: pmulhw m4, m6 ; r4 = c3*i3 - i3 yading@10: movq m1, C(5) yading@10: pmulhw m6, m7 ; r6 = c3*i5 - i5 yading@10: movq m5, m1 yading@10: pmulhw m1, m2 ; r1 = c5*i3 - i3 yading@10: movq m3, I(1) yading@10: pmulhw m5, m7 ; r5 = c5*i5 - i5 yading@10: movq m0, C(1) yading@10: paddw m4, m2 ; r4 = c3*i3 yading@10: paddw m6, m7 ; r6 = c3*i5 yading@10: paddw m2, m1 ; r2 = c5*i3 yading@10: movq m1, J(7) yading@10: paddw m7, m5 ; r7 = c5*i5 yading@10: movq m5, m0 ; r5 = c1 yading@10: pmulhw m0, m3 ; r0 = c1*i1 - i1 yading@10: paddsw m4, m7 ; r4 = C = c3*i3 + c5*i5 yading@10: pmulhw m5, m1 ; r5 = c1*i7 - i7 yading@10: movq m7, C(7) yading@10: psubsw m6, m2 ; r6 = D = c3*i5 - c5*i3 yading@10: paddw m0, m3 ; r0 = c1*i1 yading@10: pmulhw m3, m7 ; r3 = c7*i1 yading@10: movq m2, I(2) yading@10: pmulhw m7, m1 ; r7 = c7*i7 yading@10: paddw m5, m1 ; r5 = c1*i7 yading@10: movq m1, m2 ; r1 = i2 yading@10: pmulhw m2, C(2) ; r2 = c2*i2 - i2 yading@10: psubsw m3, m5 ; r3 = B = c7*i1 - c1*i7 yading@10: movq m5, J(6) yading@10: paddsw m0, m7 ; r0 = A = c1*i1 + c7*i7 yading@10: movq m7, m5 ; r7 = i6 yading@10: psubsw m0, m4 ; r0 = A - C yading@10: pmulhw m5, C(2) ; r5 = c2*i6 - i6 yading@10: paddw m2, m1 ; r2 = c2*i2 yading@10: pmulhw m1, C(6) ; r1 = c6*i2 yading@10: paddsw m4, m4 ; r4 = C + C yading@10: paddsw m4, m0 ; r4 = C. = A + C yading@10: psubsw m3, m6 ; r3 = B - D yading@10: paddw m5, m7 ; r5 = c2*i6 yading@10: paddsw m6, m6 ; r6 = D + D yading@10: pmulhw m7, C(6) ; r7 = c6*i6 yading@10: paddsw m6, m3 ; r6 = D. = B + D yading@10: movq I(1), m4 ; save C. at I(1) yading@10: psubsw m1, m5 ; r1 = H = c6*i2 - c2*i6 yading@10: movq m4, C(4) yading@10: movq m5, m3 ; r5 = B - D yading@10: pmulhw m3, m4 ; r3 = (c4 - 1) * (B - D) yading@10: paddsw m7, m2 ; r3 = (c4 - 1) * (B - D) yading@10: movq I(2), m6 ; save D. at I(2) yading@10: movq m2, m0 ; r2 = A - C yading@10: movq m6, I(0) yading@10: pmulhw m0, m4 ; r0 = (c4 - 1) * (A - C) yading@10: paddw m5, m3 ; r5 = B. = c4 * (B - D) yading@10: movq m3, J(4) yading@10: psubsw m5, m1 ; r5 = B.. = B. - H yading@10: paddw m2, m0 ; r0 = A. = c4 * (A - C) yading@10: psubsw m6, m3 ; r6 = i0 - i4 yading@10: movq m0, m6 yading@10: pmulhw m6, m4 ; r6 = (c4 - 1) * (i0 - i4) yading@10: paddsw m3, m3 ; r3 = i4 + i4 yading@10: paddsw m1, m1 ; r1 = H + H yading@10: paddsw m3, m0 ; r3 = i0 + i4 yading@10: paddsw m1, m5 ; r1 = H. = B + H yading@10: pmulhw m4, m3 ; r4 = (c4 - 1) * (i0 + i4) yading@10: paddsw m6, m0 ; r6 = F = c4 * (i0 - i4) yading@10: psubsw m6, m2 ; r6 = F. = F - A. yading@10: paddsw m2, m2 ; r2 = A. + A. yading@10: movq m0, I(1) ; r0 = C. yading@10: paddsw m2, m6 ; r2 = A.. = F + A. yading@10: paddw m4, m3 ; r4 = E = c4 * (i0 + i4) yading@10: psubsw m2, m1 ; r2 = R2 = A.. - H. yading@10: %endmacro yading@10: yading@10: ; RowIDCT gets ready to transpose yading@10: %macro RowIDCT 0 yading@10: BeginIDCT yading@10: movq m3, I(2) ; r3 = D. yading@10: psubsw m4, m7 ; r4 = E. = E - G yading@10: paddsw m1, m1 ; r1 = H. + H. yading@10: paddsw m7, m7 ; r7 = G + G yading@10: paddsw m1, m2 ; r1 = R1 = A.. + H. yading@10: paddsw m7, m4 ; r1 = R1 = A.. + H. yading@10: psubsw m4, m3 ; r4 = R4 = E. - D. yading@10: paddsw m3, m3 yading@10: psubsw m6, m5 ; r6 = R6 = F. - B.. yading@10: paddsw m5, m5 yading@10: paddsw m3, m4 ; r3 = R3 = E. + D. yading@10: paddsw m5, m6 ; r5 = R5 = F. + B.. yading@10: psubsw m7, m0 ; r7 = R7 = G. - C. yading@10: paddsw m0, m0 yading@10: movq I(1), m1 ; save R1 yading@10: paddsw m0, m7 ; r0 = R0 = G. + C. yading@10: %endmacro yading@10: yading@10: ; Column IDCT normalizes and stores final results yading@10: %macro ColumnIDCT 0 yading@10: BeginIDCT yading@10: paddsw m2, OC_8 ; adjust R2 (and R1) for shift yading@10: paddsw m1, m1 ; r1 = H. + H. yading@10: paddsw m1, m2 ; r1 = R1 = A.. + H. yading@10: psraw m2, 4 ; r2 = NR2 yading@10: psubsw m4, m7 ; r4 = E. = E - G yading@10: psraw m1, 4 ; r1 = NR2 yading@10: movq m3, I(2) ; r3 = D. yading@10: paddsw m7, m7 ; r7 = G + G yading@10: movq I(2), m2 ; store NR2 at I2 yading@10: paddsw m7, m4 ; r7 = G. = E + G yading@10: movq I(1), m1 ; store NR1 at I1 yading@10: psubsw m4, m3 ; r4 = R4 = E. - D. yading@10: paddsw m4, OC_8 ; adjust R4 (and R3) for shift yading@10: paddsw m3, m3 ; r3 = D. + D. yading@10: paddsw m3, m4 ; r3 = R3 = E. + D. yading@10: psraw m4, 4 ; r4 = NR4 yading@10: psubsw m6, m5 ; r6 = R6 = F. - B.. yading@10: psraw m3, 4 ; r3 = NR3 yading@10: paddsw m6, OC_8 ; adjust R6 (and R5) for shift yading@10: paddsw m5, m5 ; r5 = B.. + B.. yading@10: paddsw m5, m6 ; r5 = R5 = F. + B.. yading@10: psraw m6, 4 ; r6 = NR6 yading@10: movq J(4), m4 ; store NR4 at J4 yading@10: psraw m5, 4 ; r5 = NR5 yading@10: movq I(3), m3 ; store NR3 at I3 yading@10: psubsw m7, m0 ; r7 = R7 = G. - C. yading@10: paddsw m7, OC_8 ; adjust R7 (and R0) for shift yading@10: paddsw m0, m0 ; r0 = C. + C. yading@10: paddsw m0, m7 ; r0 = R0 = G. + C. yading@10: psraw m7, 4 ; r7 = NR7 yading@10: movq J(6), m6 ; store NR6 at J6 yading@10: psraw m0, 4 ; r0 = NR0 yading@10: movq J(5), m5 ; store NR5 at J5 yading@10: movq J(7), m7 ; store NR7 at J7 yading@10: movq I(0), m0 ; store NR0 at I0 yading@10: %endmacro yading@10: yading@10: ; Following macro does two 4x4 transposes in place. yading@10: ; yading@10: ; At entry (we assume): yading@10: ; yading@10: ; r0 = a3 a2 a1 a0 yading@10: ; I(1) = b3 b2 b1 b0 yading@10: ; r2 = c3 c2 c1 c0 yading@10: ; r3 = d3 d2 d1 d0 yading@10: ; yading@10: ; r4 = e3 e2 e1 e0 yading@10: ; r5 = f3 f2 f1 f0 yading@10: ; r6 = g3 g2 g1 g0 yading@10: ; r7 = h3 h2 h1 h0 yading@10: ; yading@10: ; At exit, we have: yading@10: ; yading@10: ; I(0) = d0 c0 b0 a0 yading@10: ; I(1) = d1 c1 b1 a1 yading@10: ; I(2) = d2 c2 b2 a2 yading@10: ; I(3) = d3 c3 b3 a3 yading@10: ; yading@10: ; J(4) = h0 g0 f0 e0 yading@10: ; J(5) = h1 g1 f1 e1 yading@10: ; J(6) = h2 g2 f2 e2 yading@10: ; J(7) = h3 g3 f3 e3 yading@10: ; yading@10: ; I(0) I(1) I(2) I(3) is the transpose of r0 I(1) r2 r3. yading@10: ; J(4) J(5) J(6) J(7) is the transpose of r4 r5 r6 r7. yading@10: ; yading@10: ; Since r1 is free at entry, we calculate the Js first. yading@10: %macro Transpose 0 yading@10: movq m1, m4 ; r1 = e3 e2 e1 e0 yading@10: punpcklwd m4, m5 ; r4 = f1 e1 f0 e0 yading@10: movq I(0), m0 ; save a3 a2 a1 a0 yading@10: punpckhwd m1, m5 ; r1 = f3 e3 f2 e2 yading@10: movq m0, m6 ; r0 = g3 g2 g1 g0 yading@10: punpcklwd m6, m7 ; r6 = h1 g1 h0 g0 yading@10: movq m5, m4 ; r5 = f1 e1 f0 e0 yading@10: punpckldq m4, m6 ; r4 = h0 g0 f0 e0 = R4 yading@10: punpckhdq m5, m6 ; r5 = h1 g1 f1 e1 = R5 yading@10: movq m6, m1 ; r6 = f3 e3 f2 e2 yading@10: movq J(4), m4 yading@10: punpckhwd m0, m7 ; r0 = h3 g3 h2 g2 yading@10: movq J(5), m5 yading@10: punpckhdq m6, m0 ; r6 = h3 g3 f3 e3 = R7 yading@10: movq m4, I(0) ; r4 = a3 a2 a1 a0 yading@10: punpckldq m1, m0 ; r1 = h2 g2 f2 e2 = R6 yading@10: movq m5, I(1) ; r5 = b3 b2 b1 b0 yading@10: movq m0, m4 ; r0 = a3 a2 a1 a0 yading@10: movq J(7), m6 yading@10: punpcklwd m0, m5 ; r0 = b1 a1 b0 a0 yading@10: movq J(6), m1 yading@10: punpckhwd m4, m5 ; r4 = b3 a3 b2 a2 yading@10: movq m5, m2 ; r5 = c3 c2 c1 c0 yading@10: punpcklwd m2, m3 ; r2 = d1 c1 d0 c0 yading@10: movq m1, m0 ; r1 = b1 a1 b0 a0 yading@10: punpckldq m0, m2 ; r0 = d0 c0 b0 a0 = R0 yading@10: punpckhdq m1, m2 ; r1 = d1 c1 b1 a1 = R1 yading@10: movq m2, m4 ; r2 = b3 a3 b2 a2 yading@10: movq I(0), m0 yading@10: punpckhwd m5, m3 ; r5 = d3 c3 d2 c2 yading@10: movq I(1), m1 yading@10: punpckhdq m4, m5 ; r4 = d3 c3 b3 a3 = R3 yading@10: punpckldq m2, m5 ; r2 = d2 c2 b2 a2 = R2 yading@10: movq I(3), m4 yading@10: movq I(2), m2 yading@10: %endmacro yading@10: yading@10: %macro VP3_1D_IDCT_SSE2 0 yading@10: movdqa m2, I(3) ; xmm2 = i3 yading@10: movdqa m6, C(3) ; xmm6 = c3 yading@10: movdqa m4, m2 ; xmm4 = i3 yading@10: movdqa m7, I(5) ; xmm7 = i5 yading@10: pmulhw m4, m6 ; xmm4 = c3 * i3 - i3 yading@10: movdqa m1, C(5) ; xmm1 = c5 yading@10: pmulhw m6, m7 ; xmm6 = c3 * i5 - i5 yading@10: movdqa m5, m1 ; xmm5 = c5 yading@10: pmulhw m1, m2 ; xmm1 = c5 * i3 - i3 yading@10: movdqa m3, I(1) ; xmm3 = i1 yading@10: pmulhw m5, m7 ; xmm5 = c5 * i5 - i5 yading@10: movdqa m0, C(1) ; xmm0 = c1 yading@10: paddw m4, m2 ; xmm4 = c3 * i3 yading@10: paddw m6, m7 ; xmm6 = c3 * i5 yading@10: paddw m2, m1 ; xmm2 = c5 * i3 yading@10: movdqa m1, I(7) ; xmm1 = i7 yading@10: paddw m7, m5 ; xmm7 = c5 * i5 yading@10: movdqa m5, m0 ; xmm5 = c1 yading@10: pmulhw m0, m3 ; xmm0 = c1 * i1 - i1 yading@10: paddsw m4, m7 ; xmm4 = c3 * i3 + c5 * i5 = C yading@10: pmulhw m5, m1 ; xmm5 = c1 * i7 - i7 yading@10: movdqa m7, C(7) ; xmm7 = c7 yading@10: psubsw m6, m2 ; xmm6 = c3 * i5 - c5 * i3 = D yading@10: paddw m0, m3 ; xmm0 = c1 * i1 yading@10: pmulhw m3, m7 ; xmm3 = c7 * i1 yading@10: movdqa m2, I(2) ; xmm2 = i2 yading@10: pmulhw m7, m1 ; xmm7 = c7 * i7 yading@10: paddw m5, m1 ; xmm5 = c1 * i7 yading@10: movdqa m1, m2 ; xmm1 = i2 yading@10: pmulhw m2, C(2) ; xmm2 = i2 * c2 -i2 yading@10: psubsw m3, m5 ; xmm3 = c7 * i1 - c1 * i7 = B yading@10: movdqa m5, I(6) ; xmm5 = i6 yading@10: paddsw m0, m7 ; xmm0 = c1 * i1 + c7 * i7 = A yading@10: movdqa m7, m5 ; xmm7 = i6 yading@10: psubsw m0, m4 ; xmm0 = A - C yading@10: pmulhw m5, C(2) ; xmm5 = c2 * i6 - i6 yading@10: paddw m2, m1 ; xmm2 = i2 * c2 yading@10: pmulhw m1, C(6) ; xmm1 = c6 * i2 yading@10: paddsw m4, m4 ; xmm4 = C + C yading@10: paddsw m4, m0 ; xmm4 = A + C = C. yading@10: psubsw m3, m6 ; xmm3 = B - D yading@10: paddw m5, m7 ; xmm5 = c2 * i6 yading@10: paddsw m6, m6 ; xmm6 = D + D yading@10: pmulhw m7, C(6) ; xmm7 = c6 * i6 yading@10: paddsw m6, m3 ; xmm6 = B + D = D. yading@10: movdqa I(1), m4 ; Save C. at I(1) yading@10: psubsw m1, m5 ; xmm1 = c6 * i2 - c2 * i6 = H yading@10: movdqa m4, C(4) ; xmm4 = C4 yading@10: movdqa m5, m3 ; xmm5 = B - D yading@10: pmulhw m3, m4 ; xmm3 = ( c4 -1 ) * ( B - D ) yading@10: paddsw m7, m2 ; xmm7 = c2 * i2 + c6 * i6 = G yading@10: movdqa I(2), m6 ; save D. at I(2) yading@10: movdqa m2, m0 ; xmm2 = A - C yading@10: movdqa m6, I(0) ; xmm6 = i0 yading@10: pmulhw m0, m4 ; xmm0 = ( c4 - 1 ) * ( A - C ) = A. yading@10: paddw m5, m3 ; xmm5 = c4 * ( B - D ) = B. yading@10: movdqa m3, I(4) ; xmm3 = i4 yading@10: psubsw m5, m1 ; xmm5 = B. - H = B.. yading@10: paddw m2, m0 ; xmm2 = c4 * ( A - C) = A. yading@10: psubsw m6, m3 ; xmm6 = i0 - i4 yading@10: movdqa m0, m6 ; xmm0 = i0 - i4 yading@10: pmulhw m6, m4 ; xmm6 = (c4 - 1) * (i0 - i4) = F yading@10: paddsw m3, m3 ; xmm3 = i4 + i4 yading@10: paddsw m1, m1 ; xmm1 = H + H yading@10: paddsw m3, m0 ; xmm3 = i0 + i4 yading@10: paddsw m1, m5 ; xmm1 = B. + H = H. yading@10: pmulhw m4, m3 ; xmm4 = ( c4 - 1 ) * ( i0 + i4 ) yading@10: paddw m6, m0 ; xmm6 = c4 * ( i0 - i4 ) yading@10: psubsw m6, m2 ; xmm6 = F - A. = F. yading@10: paddsw m2, m2 ; xmm2 = A. + A. yading@10: movdqa m0, I(1) ; Load C. from I(1) yading@10: paddsw m2, m6 ; xmm2 = F + A. = A.. yading@10: paddw m4, m3 ; xmm4 = c4 * ( i0 + i4 ) = 3 yading@10: psubsw m2, m1 ; xmm2 = A.. - H. = R2 yading@10: ADD(m2) ; Adjust R2 and R1 before shifting yading@10: paddsw m1, m1 ; xmm1 = H. + H. yading@10: paddsw m1, m2 ; xmm1 = A.. + H. = R1 yading@10: SHIFT(m2) ; xmm2 = op2 yading@10: psubsw m4, m7 ; xmm4 = E - G = E. yading@10: SHIFT(m1) ; xmm1 = op1 yading@10: movdqa m3, I(2) ; Load D. from I(2) yading@10: paddsw m7, m7 ; xmm7 = G + G yading@10: paddsw m7, m4 ; xmm7 = E + G = G. yading@10: psubsw m4, m3 ; xmm4 = E. - D. = R4 yading@10: ADD(m4) ; Adjust R4 and R3 before shifting yading@10: paddsw m3, m3 ; xmm3 = D. + D. yading@10: paddsw m3, m4 ; xmm3 = E. + D. = R3 yading@10: SHIFT(m4) ; xmm4 = op4 yading@10: psubsw m6, m5 ; xmm6 = F. - B..= R6 yading@10: SHIFT(m3) ; xmm3 = op3 yading@10: ADD(m6) ; Adjust R6 and R5 before shifting yading@10: paddsw m5, m5 ; xmm5 = B.. + B.. yading@10: paddsw m5, m6 ; xmm5 = F. + B.. = R5 yading@10: SHIFT(m6) ; xmm6 = op6 yading@10: SHIFT(m5) ; xmm5 = op5 yading@10: psubsw m7, m0 ; xmm7 = G. - C. = R7 yading@10: ADD(m7) ; Adjust R7 and R0 before shifting yading@10: paddsw m0, m0 ; xmm0 = C. + C. yading@10: paddsw m0, m7 ; xmm0 = G. + C. yading@10: SHIFT(m7) ; xmm7 = op7 yading@10: SHIFT(m0) ; xmm0 = op0 yading@10: %endmacro yading@10: yading@10: %macro PUT_BLOCK 8 yading@10: movdqa O(0), m%1 yading@10: movdqa O(1), m%2 yading@10: movdqa O(2), m%3 yading@10: movdqa O(3), m%4 yading@10: movdqa O(4), m%5 yading@10: movdqa O(5), m%6 yading@10: movdqa O(6), m%7 yading@10: movdqa O(7), m%8 yading@10: %endmacro yading@10: yading@10: %macro VP3_IDCT 1 yading@10: %if mmsize == 16 yading@10: %define I(x) [%1+16*x] yading@10: %define O(x) [%1+16*x] yading@10: %define C(x) [vp3_idct_data+16*(x-1)] yading@10: %define SHIFT(x) yading@10: %define ADD(x) yading@10: VP3_1D_IDCT_SSE2 yading@10: %if ARCH_X86_64 yading@10: TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, 8 yading@10: %else yading@10: TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, [%1], [%1+16] yading@10: %endif yading@10: PUT_BLOCK 0, 1, 2, 3, 4, 5, 6, 7 yading@10: yading@10: %define SHIFT(x) psraw x, 4 yading@10: %define ADD(x) paddsw x, [pw_8] yading@10: VP3_1D_IDCT_SSE2 yading@10: PUT_BLOCK 0, 1, 2, 3, 4, 5, 6, 7 yading@10: %else ; mmsize == 8 yading@10: ; eax = quantized input yading@10: ; ebx = dequantizer matrix yading@10: ; ecx = IDCT constants yading@10: ; M(I) = ecx + MaskOffset(0) + I * 8 yading@10: ; C(I) = ecx + CosineOffset(32) + (I-1) * 8 yading@10: ; edx = output yading@10: ; r0..r7 = mm0..mm7 yading@10: %define OC_8 [pw_8] yading@10: %define C(x) [vp3_idct_data+16*(x-1)] yading@10: yading@10: ; at this point, function has completed dequantization + dezigzag + yading@10: ; partial transposition; now do the idct itself yading@10: %define I(x) [%1+16*x] yading@10: %define J(x) [%1+16*x] yading@10: RowIDCT yading@10: Transpose yading@10: yading@10: %define I(x) [%1+16*x+8] yading@10: %define J(x) [%1+16*x+8] yading@10: RowIDCT yading@10: Transpose yading@10: yading@10: %define I(x) [%1+16* x] yading@10: %define J(x) [%1+16*(x-4)+8] yading@10: ColumnIDCT yading@10: yading@10: %define I(x) [%1+16* x +64] yading@10: %define J(x) [%1+16*(x-4)+72] yading@10: ColumnIDCT yading@10: %endif ; mmsize == 16/8 yading@10: %endmacro yading@10: yading@10: %macro vp3_idct_funcs 0 yading@10: cglobal vp3_idct_put, 3, 4, 9 yading@10: VP3_IDCT r2 yading@10: yading@10: movsxdifnidn r1, r1d yading@10: mova m4, [pb_80] yading@10: lea r3, [r1*3] yading@10: %assign %%i 0 yading@10: %rep 16/mmsize yading@10: mova m0, [r2+mmsize*0+%%i] yading@10: mova m1, [r2+mmsize*2+%%i] yading@10: mova m2, [r2+mmsize*4+%%i] yading@10: mova m3, [r2+mmsize*6+%%i] yading@10: %if mmsize == 8 yading@10: packsswb m0, [r2+mmsize*8+%%i] yading@10: packsswb m1, [r2+mmsize*10+%%i] yading@10: packsswb m2, [r2+mmsize*12+%%i] yading@10: packsswb m3, [r2+mmsize*14+%%i] yading@10: %else yading@10: packsswb m0, [r2+mmsize*1+%%i] yading@10: packsswb m1, [r2+mmsize*3+%%i] yading@10: packsswb m2, [r2+mmsize*5+%%i] yading@10: packsswb m3, [r2+mmsize*7+%%i] yading@10: %endif yading@10: paddb m0, m4 yading@10: paddb m1, m4 yading@10: paddb m2, m4 yading@10: paddb m3, m4 yading@10: movq [r0 ], m0 yading@10: %if mmsize == 8 yading@10: movq [r0+r1 ], m1 yading@10: movq [r0+r1*2], m2 yading@10: movq [r0+r3 ], m3 yading@10: %else yading@10: movhps [r0+r1 ], m0 yading@10: movq [r0+r1*2], m1 yading@10: movhps [r0+r3 ], m1 yading@10: %endif yading@10: %if %%i == 0 yading@10: lea r0, [r0+r1*4] yading@10: %endif yading@10: %if mmsize == 16 yading@10: movq [r0 ], m2 yading@10: movhps [r0+r1 ], m2 yading@10: movq [r0+r1*2], m3 yading@10: movhps [r0+r3 ], m3 yading@10: %endif yading@10: %assign %%i %%i+8 yading@10: %endrep yading@10: yading@10: pxor m0, m0 yading@10: %assign %%offset 0 yading@10: %rep 128/mmsize yading@10: mova [r2+%%offset], m0 yading@10: %assign %%offset %%offset+mmsize yading@10: %endrep yading@10: RET yading@10: yading@10: cglobal vp3_idct_add, 3, 4, 9 yading@10: VP3_IDCT r2 yading@10: yading@10: movsxdifnidn r1, r1d yading@10: lea r3, [r1*3] yading@10: pxor m4, m4 yading@10: %if mmsize == 16 yading@10: %assign %%i 0 yading@10: %rep 2 yading@10: movq m0, [r0] yading@10: movq m1, [r0+r1] yading@10: movq m2, [r0+r1*2] yading@10: movq m3, [r0+r3] yading@10: punpcklbw m0, m4 yading@10: punpcklbw m1, m4 yading@10: punpcklbw m2, m4 yading@10: punpcklbw m3, m4 yading@10: paddsw m0, [r2+ 0+%%i] yading@10: paddsw m1, [r2+16+%%i] yading@10: paddsw m2, [r2+32+%%i] yading@10: paddsw m3, [r2+48+%%i] yading@10: packuswb m0, m1 yading@10: packuswb m2, m3 yading@10: movq [r0 ], m0 yading@10: movhps [r0+r1 ], m0 yading@10: movq [r0+r1*2], m2 yading@10: movhps [r0+r3 ], m2 yading@10: %if %%i == 0 yading@10: lea r0, [r0+r1*4] yading@10: %endif yading@10: %assign %%i %%i+64 yading@10: %endrep yading@10: %else yading@10: %assign %%i 0 yading@10: %rep 2 yading@10: movq m0, [r0] yading@10: movq m1, [r0+r1] yading@10: movq m2, [r0+r1*2] yading@10: movq m3, [r0+r3] yading@10: movq m5, m0 yading@10: movq m6, m1 yading@10: movq m7, m2 yading@10: punpcklbw m0, m4 yading@10: punpcklbw m1, m4 yading@10: punpcklbw m2, m4 yading@10: punpckhbw m5, m4 yading@10: punpckhbw m6, m4 yading@10: punpckhbw m7, m4 yading@10: paddsw m0, [r2+ 0+%%i] yading@10: paddsw m1, [r2+16+%%i] yading@10: paddsw m2, [r2+32+%%i] yading@10: paddsw m5, [r2+64+%%i] yading@10: paddsw m6, [r2+80+%%i] yading@10: paddsw m7, [r2+96+%%i] yading@10: packuswb m0, m5 yading@10: movq m5, m3 yading@10: punpcklbw m3, m4 yading@10: punpckhbw m5, m4 yading@10: packuswb m1, m6 yading@10: paddsw m3, [r2+48+%%i] yading@10: paddsw m5, [r2+112+%%i] yading@10: packuswb m2, m7 yading@10: packuswb m3, m5 yading@10: movq [r0 ], m0 yading@10: movq [r0+r1 ], m1 yading@10: movq [r0+r1*2], m2 yading@10: movq [r0+r3 ], m3 yading@10: %if %%i == 0 yading@10: lea r0, [r0+r1*4] yading@10: %endif yading@10: %assign %%i %%i+8 yading@10: %endrep yading@10: %endif yading@10: %assign %%i 0 yading@10: %rep 128/mmsize yading@10: mova [r2+%%i], m4 yading@10: %assign %%i %%i+mmsize yading@10: %endrep yading@10: RET yading@10: %endmacro yading@10: yading@10: %if ARCH_X86_32 yading@10: INIT_MMX mmx yading@10: vp3_idct_funcs yading@10: %endif yading@10: yading@10: INIT_XMM sse2 yading@10: vp3_idct_funcs yading@10: yading@10: %macro DC_ADD 0 yading@10: movq m2, [r0 ] yading@10: movq m3, [r0+r1 ] yading@10: paddusb m2, m0 yading@10: movq m4, [r0+r1*2] yading@10: paddusb m3, m0 yading@10: movq m5, [r0+r2 ] yading@10: paddusb m4, m0 yading@10: paddusb m5, m0 yading@10: psubusb m2, m1 yading@10: psubusb m3, m1 yading@10: movq [r0 ], m2 yading@10: psubusb m4, m1 yading@10: movq [r0+r1 ], m3 yading@10: psubusb m5, m1 yading@10: movq [r0+r1*2], m4 yading@10: movq [r0+r2 ], m5 yading@10: %endmacro yading@10: yading@10: INIT_MMX mmxext yading@10: cglobal vp3_idct_dc_add, 3, 4 yading@10: %if ARCH_X86_64 yading@10: movsxd r1, r1d yading@10: %endif yading@10: movsx r3, word [r2] yading@10: mov word [r2], 0 yading@10: lea r2, [r1*3] yading@10: add r3, 15 yading@10: sar r3, 5 yading@10: movd m0, r3d yading@10: pshufw m0, m0, 0x0 yading@10: pxor m1, m1 yading@10: psubw m1, m0 yading@10: packuswb m0, m0 yading@10: packuswb m1, m1 yading@10: DC_ADD yading@10: lea r0, [r0+r1*4] yading@10: DC_ADD yading@10: RET