;*****************************************************************************
-;* predict-a.asm: h264 encoder library
+;* predict-a.asm: x86 intra prediction
;*****************************************************************************
-;* Copyright (C) 2005-2008 x264 project
+;* Copyright (C) 2005-2011 x264 project
;*
;* Authors: Loren Merritt <lorenm@u.washington.edu>
;* Holger Lubitz <holger@lubitz.org>
;* You should have received a copy of the GNU General Public License
;* along with this program; if not, write to the Free Software
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
+;*
+;* This program is also available under a commercial proprietary license.
+;* For more information, contact us at licensing@x264.com.
;*****************************************************************************
%include "x86inc.asm"
%include "x86util.asm"
+SECTION_RODATA
+
+pw_76543210:
+pw_3210: dw 0, 1, 2, 3, 4, 5, 6, 7
+pw_43210123: dw -3, -2, -1, 0, 1, 2, 3, 4
+pw_m3: times 8 dw -3
+pb_00s_ff: times 8 db 0
+pb_0s_ff: times 7 db 0
+ db 0xff
+
+SECTION .text
+
+cextern pb_0
+cextern pb_1
+cextern pb_3
+cextern pw_1
+cextern pw_2
+cextern pw_4
+cextern pw_8
+cextern pw_16
+cextern pw_ff00
+cextern pb_reverse
+cextern pw_pixel_max
+
%macro STORE8x8 2
- add r0, 4*FDEC_STRIDE
- movq [r0 + -4*FDEC_STRIDE], %1
- movq [r0 + -3*FDEC_STRIDE], %1
- movq [r0 + -2*FDEC_STRIDE], %1
- movq [r0 + -1*FDEC_STRIDE], %1
- movq [r0 + 0*FDEC_STRIDE], %2
- movq [r0 + 1*FDEC_STRIDE], %2
- movq [r0 + 2*FDEC_STRIDE], %2
- movq [r0 + 3*FDEC_STRIDE], %2
+ add r0, 4*FDEC_STRIDEB
+ mova [r0 + -4*FDEC_STRIDEB], %1
+ mova [r0 + -3*FDEC_STRIDEB], %1
+ mova [r0 + -2*FDEC_STRIDEB], %1
+ mova [r0 + -1*FDEC_STRIDEB], %1
+ mova [r0 + 0*FDEC_STRIDEB], %2
+ mova [r0 + 1*FDEC_STRIDEB], %2
+ mova [r0 + 2*FDEC_STRIDEB], %2
+ mova [r0 + 3*FDEC_STRIDEB], %2
%endmacro
-%macro STORE16x16 2
+%macro STORE16x16 2-4
+%ifidn %0, 4
+ mov r1d, 8
+.loop:
+ mova [r0 + 0*FDEC_STRIDEB + 0], %1
+ mova [r0 + 1*FDEC_STRIDEB + 0], %1
+ mova [r0 + 0*FDEC_STRIDEB + 8], %2
+ mova [r0 + 1*FDEC_STRIDEB + 8], %2
+ mova [r0 + 0*FDEC_STRIDEB +16], %3
+ mova [r0 + 1*FDEC_STRIDEB +16], %3
+ mova [r0 + 0*FDEC_STRIDEB +24], %4
+ mova [r0 + 1*FDEC_STRIDEB +24], %4
+ add r0, 2*FDEC_STRIDEB
+ dec r1d
+ jg .loop
+%else
mov r1d, 4
.loop:
- movq [r0 + 0*FDEC_STRIDE], %1
- movq [r0 + 1*FDEC_STRIDE], %1
- movq [r0 + 2*FDEC_STRIDE], %1
- movq [r0 + 3*FDEC_STRIDE], %1
- movq [r0 + 0*FDEC_STRIDE + 8], %2
- movq [r0 + 1*FDEC_STRIDE + 8], %2
- movq [r0 + 2*FDEC_STRIDE + 8], %2
- movq [r0 + 3*FDEC_STRIDE + 8], %2
+ mova [r0 + 0*FDEC_STRIDE], %1
+ mova [r0 + 1*FDEC_STRIDE], %1
+ mova [r0 + 2*FDEC_STRIDE], %1
+ mova [r0 + 3*FDEC_STRIDE], %1
+ mova [r0 + 0*FDEC_STRIDE + 8], %2
+ mova [r0 + 1*FDEC_STRIDE + 8], %2
+ mova [r0 + 2*FDEC_STRIDE + 8], %2
+ mova [r0 + 3*FDEC_STRIDE + 8], %2
add r0, 4*FDEC_STRIDE
dec r1d
jg .loop
+%endif
%endmacro
-%macro STORE16x16_SSE2 1
- add r0, 4*FDEC_STRIDE
- movdqa [r0 + -4*FDEC_STRIDE], %1
- movdqa [r0 + -3*FDEC_STRIDE], %1
- movdqa [r0 + -2*FDEC_STRIDE], %1
- movdqa [r0 + -1*FDEC_STRIDE], %1
- movdqa [r0 + 0*FDEC_STRIDE], %1
- movdqa [r0 + 1*FDEC_STRIDE], %1
- movdqa [r0 + 2*FDEC_STRIDE], %1
- movdqa [r0 + 3*FDEC_STRIDE], %1
- add r0, 8*FDEC_STRIDE
- movdqa [r0 + -4*FDEC_STRIDE], %1
- movdqa [r0 + -3*FDEC_STRIDE], %1
- movdqa [r0 + -2*FDEC_STRIDE], %1
- movdqa [r0 + -1*FDEC_STRIDE], %1
- movdqa [r0 + 0*FDEC_STRIDE], %1
- movdqa [r0 + 1*FDEC_STRIDE], %1
- movdqa [r0 + 2*FDEC_STRIDE], %1
- movdqa [r0 + 3*FDEC_STRIDE], %1
+%macro STORE16x16_SSE2 1-2
+%ifidn %0,2
+ mov r1d, 4
+.loop
+ mova [r0+0*FDEC_STRIDEB+ 0], %1
+ mova [r0+0*FDEC_STRIDEB+16], %2
+ mova [r0+1*FDEC_STRIDEB+ 0], %1
+ mova [r0+1*FDEC_STRIDEB+16], %2
+ mova [r0+2*FDEC_STRIDEB+ 0], %1
+ mova [r0+2*FDEC_STRIDEB+16], %2
+ mova [r0+3*FDEC_STRIDEB+ 0], %1
+ mova [r0+3*FDEC_STRIDEB+16], %2
+ add r0, 4*FDEC_STRIDEB
+ dec r1d
+ jg .loop
+%else
+ add r0, 4*FDEC_STRIDEB
+ mova [r0 + -4*FDEC_STRIDEB], %1
+ mova [r0 + -3*FDEC_STRIDEB], %1
+ mova [r0 + -2*FDEC_STRIDEB], %1
+ mova [r0 + -1*FDEC_STRIDEB], %1
+ mova [r0 + 0*FDEC_STRIDEB], %1
+ mova [r0 + 1*FDEC_STRIDEB], %1
+ mova [r0 + 2*FDEC_STRIDEB], %1
+ mova [r0 + 3*FDEC_STRIDEB], %1
+ add r0, 8*FDEC_STRIDEB
+ mova [r0 + -4*FDEC_STRIDEB], %1
+ mova [r0 + -3*FDEC_STRIDEB], %1
+ mova [r0 + -2*FDEC_STRIDEB], %1
+ mova [r0 + -1*FDEC_STRIDEB], %1
+ mova [r0 + 0*FDEC_STRIDEB], %1
+ mova [r0 + 1*FDEC_STRIDEB], %1
+ mova [r0 + 2*FDEC_STRIDEB], %1
+ mova [r0 + 3*FDEC_STRIDEB], %1
+%endif
%endmacro
-SECTION_RODATA
-
-ALIGN 16
-pb_1: times 16 db 1
-pb_3: times 16 db 3
-pw_2: times 4 dw 2
-pw_4: times 4 dw 4
-pw_8: times 8 dw 8
-pw_76543210:
-pw_3210: dw 0, 1, 2, 3, 4, 5, 6, 7
-pb_00s_ff: times 8 db 0
-pb_0s_ff: times 7 db 0
- db 0xff
-pw_ff00: times 8 dw 0xff00
-pb_reverse: db 7, 6, 5, 4, 3, 2, 1, 0
-
-SECTION .text
-
; dest, left, right, src, tmp
; output: %1 = (t[n-1] + t[n]*2 + t[n+1] + 2) >> 2
-%macro PRED8x8_LOWPASS0 6
- mov%6 %5, %2
- pavgb %2, %3
- pxor %3, %5
- mov%6 %1, %4
- pand %3, [pb_1]
- psubusb %2, %3
- pavgb %1, %2
-%endmacro
-%macro PRED8x8_LOWPASS 5
- PRED8x8_LOWPASS0 %1, %2, %3, %4, %5, q
-%endmacro
-%macro PRED8x8_LOWPASS_XMM 5
- PRED8x8_LOWPASS0 %1, %2, %3, %4, %5, dqa
+%macro PRED8x8_LOWPASS 5-6
+%ifidn %1, w
+ paddw %3, %4
+ psrlw %3, 1
+ pavgw %2, %5, %3
+%else
+ mova %6, %3
+ pavgb %3, %4
+ pxor %4, %6
+ pand %4, [pb_1]
+ psubusb %3, %4
+ pavgb %2, %5, %3
+%endif
%endmacro
%macro LOAD_PLANE_ARGS 0
%endmacro
;-----------------------------------------------------------------------------
-; void predict_4x4_ddl_mmxext( uint8_t *src )
+; void predict_4x4_ddl( pixel *src )
;-----------------------------------------------------------------------------
-cglobal predict_4x4_ddl_mmxext, 1,1
- movq mm1, [r0-FDEC_STRIDE]
- movq mm2, mm1
- movq mm3, mm1
- movq mm4, mm1
- psllq mm1, 8
- pxor mm2, mm1
- psrlq mm2, 8
- pxor mm3, mm2
- PRED8x8_LOWPASS mm0, mm1, mm3, mm4, mm5
+%macro PREDICT_4x4_DDL 4
+cglobal predict_4x4_ddl_%1, 1,1
+ movu m1, [r0-FDEC_STRIDEB]
+ psll%2 m2, m1, %3
+ mova m3, m1
+ mova m4, m1
+ pxor m1, m2
+ psrl%2 m1, %3
+ pxor m3, m1
+ PRED8x8_LOWPASS %4, m0, m2, m3, m4, m5
%assign Y 0
%rep 4
- psrlq mm0, 8
- movd [r0+Y*FDEC_STRIDE], mm0
+ psrl%2 m0, %3
+ movh [r0+Y*FDEC_STRIDEB], m0
%assign Y (Y+1)
%endrep
RET
+%endmacro
+
+%ifdef HIGH_BIT_DEPTH
+INIT_XMM
+PREDICT_4x4_DDL sse2, dq, 2, w
+INIT_AVX
+PREDICT_4x4_DDL avx , dq, 2, w
+INIT_MMX
+%define PALIGNR PALIGNR_MMX
+cglobal predict_4x4_ddl_mmxext, 1,2
+ mova m1, [r0-2*FDEC_STRIDE+4]
+ mova m2, [r0-2*FDEC_STRIDE+0]
+ mova m3, [r0-2*FDEC_STRIDE+2]
+ PRED8x8_LOWPASS w, m0, m1, m2, m3
+ mova [r0+0*FDEC_STRIDE], m0
+
+ mova m5, [r0-2*FDEC_STRIDE+6]
+ mova m6, [r0-2*FDEC_STRIDE+8]
+ pshufw m7, m6, 0xF9
+ PRED8x8_LOWPASS w, m4, m7, m5, m6
+ mova [r0+6*FDEC_STRIDE], m4
+
+ psllq m0, 16
+ PALIGNR m4, m0, 6, m1
+ mova [r0+4*FDEC_STRIDE], m4
+
+ psllq m0, 16
+ PALIGNR m4, m0, 6, m0
+ mova [r0+2*FDEC_STRIDE], m4
+ RET
+%else
+INIT_MMX
+PREDICT_4x4_DDL mmxext, q , 8, b
+%endif
;-----------------------------------------------------------------------------
-; void predict_4x4_ddr_mmxext( uint8_t *src )
+; void predict_4x4_ddr( pixel *src )
;-----------------------------------------------------------------------------
-%macro PREDICT_4x4 1
+%macro PREDICT_4x4 7
cglobal predict_4x4_ddr_%1, 1,1
- movq mm1, [r0+1*FDEC_STRIDE-8]
- movq mm2, [r0+0*FDEC_STRIDE-8]
- punpckhbw mm2, [r0-1*FDEC_STRIDE-8]
- movd mm3, [r0-1*FDEC_STRIDE]
- punpckhwd mm1, mm2
- PALIGNR mm3, mm1, 5, mm1
- movq mm1, mm3
- PALIGNR mm3, [r0+2*FDEC_STRIDE-8], 7, mm4
- movq mm2, mm3
- PALIGNR mm3, [r0+3*FDEC_STRIDE-8], 7, mm4
- PRED8x8_LOWPASS mm0, mm3, mm1, mm2, mm4
+ movu m1, [r0+1*FDEC_STRIDEB-8*SIZEOF_PIXEL]
+ movq m2, [r0+0*FDEC_STRIDEB-8]
+%ifdef HIGH_BIT_DEPTH
+ movh m4, [r0-1*FDEC_STRIDEB-4*SIZEOF_PIXEL]
+ punpckl%2 m2, m4
+ movh m3, [r0-1*FDEC_STRIDEB]
+ punpckh%3 m1, m2
+ PALIGNR m3, m1, 5*SIZEOF_PIXEL, m1
+ mova m1, m3
+ movhps m4, [r0+2*FDEC_STRIDEB-4*SIZEOF_PIXEL]
+ PALIGNR m3, m4, 7*SIZEOF_PIXEL, m4
+ mova m2, m3
+ movhps m4, [r0+3*FDEC_STRIDEB-4*SIZEOF_PIXEL]
+ PALIGNR m3, m4, 7*SIZEOF_PIXEL, m4
+%else
+ punpckh%2 m2, [r0-1*FDEC_STRIDEB-8*SIZEOF_PIXEL]
+ movh m3, [r0-1*FDEC_STRIDEB]
+ punpckh%3 m1, m2
+ PALIGNR m3, m1, 5*SIZEOF_PIXEL, m1
+ mova m1, m3
+ PALIGNR m3, [r0+2*FDEC_STRIDEB-8*SIZEOF_PIXEL], 7*SIZEOF_PIXEL, m4
+ mova m2, m3
+ PALIGNR m3, [r0+3*FDEC_STRIDEB-8*SIZEOF_PIXEL], 7*SIZEOF_PIXEL, m4
+%endif
+ PRED8x8_LOWPASS %5, m0, m3, m1, m2, m4
%assign Y 3
- movd [r0+Y*FDEC_STRIDE], mm0
+ movh [r0+Y*FDEC_STRIDEB], m0
%rep 3
%assign Y (Y-1)
- psrlq mm0, 8
- movd [r0+Y*FDEC_STRIDE], mm0
+ psrl%4 m0, %7
+ movh [r0+Y*FDEC_STRIDEB], m0
%endrep
RET
-cglobal predict_4x4_vr_%1, 1,1
- movd mm0, [r0-1*FDEC_STRIDE] ; ........t3t2t1t0
- movq mm7, mm0
- PALIGNR mm0, [r0-1*FDEC_STRIDE-8], 7, mm1 ; ......t3t2t1t0lt
- pavgb mm7, mm0
- PALIGNR mm0, [r0+0*FDEC_STRIDE-8], 7, mm1 ; ....t3t2t1t0ltl0
- movq mm1, mm0
- PALIGNR mm0, [r0+1*FDEC_STRIDE-8], 7, mm2 ; ..t3t2t1t0ltl0l1
- movq mm2, mm0
- PALIGNR mm0, [r0+2*FDEC_STRIDE-8], 7, mm3 ; t3t2t1t0ltl0l1l2
- PRED8x8_LOWPASS mm3, mm1, mm0, mm2, mm4
- movq mm1, mm3
- psrlq mm3, 16
- psllq mm1, 48
- movd [r0+0*FDEC_STRIDE], mm7
- movd [r0+1*FDEC_STRIDE], mm3
- PALIGNR mm7, mm1, 7, mm2
- psllq mm1, 8
- movd [r0+2*FDEC_STRIDE], mm7
- PALIGNR mm3, mm1, 7, mm1
- movd [r0+3*FDEC_STRIDE], mm3
+cglobal predict_4x4_vr_%1, 1,1,6*(mmsize/16)
+ movh m0, [r0-1*FDEC_STRIDEB] ; ........t3t2t1t0
+ mova m5, m0
+%ifdef HIGH_BIT_DEPTH
+ movhps m1, [r0-1*FDEC_STRIDEB-4*SIZEOF_PIXEL]
+ PALIGNR m0, m1, 7*SIZEOF_PIXEL, m1 ; ......t3t2t1t0lt
+ pavg%5 m5, m0
+ movhps m1, [r0+0*FDEC_STRIDEB-4*SIZEOF_PIXEL]
+ PALIGNR m0, m1, 7*SIZEOF_PIXEL, m1 ; ....t3t2t1t0ltl0
+ mova m1, m0
+ movhps m2, [r0+1*FDEC_STRIDEB-4*SIZEOF_PIXEL]
+ PALIGNR m0, m2, 7*SIZEOF_PIXEL, m2 ; ..t3t2t1t0ltl0l1
+ mova m2, m0
+ movhps m3, [r0+2*FDEC_STRIDEB-4*SIZEOF_PIXEL]
+ PALIGNR m0, m3, 7*SIZEOF_PIXEL, m3 ; t3t2t1t0ltl0l1l2
+%else
+ PALIGNR m0, [r0-1*FDEC_STRIDEB-8*SIZEOF_PIXEL], 7*SIZEOF_PIXEL, m1 ; ......t3t2t1t0lt
+ pavg%5 m5, m0
+ PALIGNR m0, [r0+0*FDEC_STRIDEB-8*SIZEOF_PIXEL], 7*SIZEOF_PIXEL, m1 ; ....t3t2t1t0ltl0
+ mova m1, m0
+ PALIGNR m0, [r0+1*FDEC_STRIDEB-8*SIZEOF_PIXEL], 7*SIZEOF_PIXEL, m2 ; ..t3t2t1t0ltl0l1
+ mova m2, m0
+ PALIGNR m0, [r0+2*FDEC_STRIDEB-8*SIZEOF_PIXEL], 7*SIZEOF_PIXEL, m3 ; t3t2t1t0ltl0l1l2
+%endif
+ PRED8x8_LOWPASS %5, m3, m1, m0, m2, m4
+ psll%4 m1, m3, %7*6
+ psrl%4 m3, %7*2
+ movh [r0+0*FDEC_STRIDEB], m5
+ movh [r0+1*FDEC_STRIDEB], m3
+ PALIGNR m5, m1, 7*SIZEOF_PIXEL, m2
+ psll%4 m1, %7
+ movh [r0+2*FDEC_STRIDEB], m5
+ PALIGNR m3, m1, 7*SIZEOF_PIXEL, m1
+ movh [r0+3*FDEC_STRIDEB], m3
RET
-cglobal predict_4x4_hd_%1, 1,1
- movd mm0, [r0-1*FDEC_STRIDE-4] ; lt ..
- punpckldq mm0, [r0-1*FDEC_STRIDE] ; t3 t2 t1 t0 lt .. .. ..
- psllq mm0, 8 ; t2 t1 t0 lt .. .. .. ..
- movq mm1, [r0+3*FDEC_STRIDE-8] ; l3
- punpckhbw mm1, [r0+2*FDEC_STRIDE-8] ; l2 l3
- movq mm2, [r0+1*FDEC_STRIDE-8] ; l1
- punpckhbw mm2, [r0+0*FDEC_STRIDE-8] ; l0 l1
- punpckhwd mm1, mm2 ; l0 l1 l2 l3
- punpckhdq mm1, mm0 ; t2 t1 t0 lt l0 l1 l2 l3
- movq mm0, mm1
- movq mm2, mm1
- movq mm7, mm1
- psrlq mm0, 16 ; .. .. t2 t1 t0 lt l0 l1
- psrlq mm2, 8 ; .. t2 t1 t0 lt l0 l1 l2
- pavgb mm7, mm2
- PRED8x8_LOWPASS mm3, mm1, mm0, mm2, mm4
- punpcklbw mm7, mm3
- psrlq mm3, 32
- PALIGNR mm3, mm7, 6, mm6
+cglobal predict_4x4_hd_%1, 1,1,6*(mmsize/16)
+ movh m0, [r0-1*FDEC_STRIDEB-4*SIZEOF_PIXEL] ; lt ..
+%ifdef HIGH_BIT_DEPTH
+ movh m1, [r0-1*FDEC_STRIDEB]
+ punpckl%6 m0, m1 ; t3 t2 t1 t0 lt .. .. ..
+ psll%4 m0, %7 ; t2 t1 t0 lt .. .. .. ..
+ movh m1, [r0+3*FDEC_STRIDEB-4*SIZEOF_PIXEL] ; l3
+ movh m2, [r0+2*FDEC_STRIDEB-4*SIZEOF_PIXEL]
+ punpckl%2 m1, m2 ; l2 l3
+ movh m2, [r0+1*FDEC_STRIDEB-4*SIZEOF_PIXEL] ; l1
+ movh m3, [r0+0*FDEC_STRIDEB-4*SIZEOF_PIXEL]
+ punpckl%2 m2, m3 ; l0 l1
+%else
+ punpckl%6 m0, [r0-1*FDEC_STRIDEB] ; t3 t2 t1 t0 lt .. .. ..
+ psll%4 m0, %7 ; t2 t1 t0 lt .. .. .. ..
+ movu m1, [r0+3*FDEC_STRIDEB-8*SIZEOF_PIXEL] ; l3
+ punpckh%2 m1, [r0+2*FDEC_STRIDEB-8*SIZEOF_PIXEL] ; l2 l3
+ movu m2, [r0+1*FDEC_STRIDEB-8*SIZEOF_PIXEL] ; l1
+ punpckh%2 m2, [r0+0*FDEC_STRIDEB-8*SIZEOF_PIXEL] ; l0 l1
+%endif
+ punpckh%3 m1, m2 ; l0 l1 l2 l3
+ punpckh%6 m1, m0 ; t2 t1 t0 lt l0 l1 l2 l3
+ psrl%4 m2, m1, %7 ; .. t2 t1 t0 lt l0 l1 l2
+ psrl%4 m0, m1, %7*2 ; .. .. t2 t1 t0 lt l0 l1
+ pavg%5 m5, m1, m2
+ PRED8x8_LOWPASS %5, m3, m1, m0, m2, m4
+ punpckl%2 m5, m3
+ psrl%4 m3, %7*4
+ PALIGNR m3, m5, 6*SIZEOF_PIXEL, m4
%assign Y 3
- movd [r0+Y*FDEC_STRIDE], mm7
+ movh [r0+Y*FDEC_STRIDEB], m5
%rep 2
%assign Y (Y-1)
- psrlq mm7, 16
- movd [r0+Y*FDEC_STRIDE], mm7
+ psrl%4 m5, %7*2
+ movh [r0+Y*FDEC_STRIDEB], m5
%endrep
- movd [r0+0*FDEC_STRIDE], mm3
+ movh [r0+0*FDEC_STRIDEB], m3
RET
%endmacro
+%ifdef HIGH_BIT_DEPTH
+INIT_MMX
%define PALIGNR PALIGNR_MMX
-PREDICT_4x4 mmxext
+cglobal predict_4x4_ddr_mmxext, 1,1
+ movq m3, [r0+3*FDEC_STRIDEB-8]
+ psrlq m3, 48
+ PALIGNR m3, [r0+2*FDEC_STRIDEB-8], 6, m6
+ PALIGNR m3, [r0+1*FDEC_STRIDEB-8], 6, m7
+ movq m6, [r0+0*FDEC_STRIDEB-8]
+ PALIGNR m3, m6, 6, m5
+
+ movq m4, [r0-1*FDEC_STRIDEB-8]
+ movq m2, m3
+ movq m1, m3
+ PALIGNR m2, m4, 6, m5
+ movq m1, m2
+ psllq m1, 16
+ PRED8x8_LOWPASS w, m0, m3, m1, m2
+ pshufw m0, m0, 0x1B
+ movq [r0+3*FDEC_STRIDEB], m0
+
+ movq m2, [r0-1*FDEC_STRIDEB-0]
+ movq m5, m2
+ PALIGNR m5, m4, 6, m4
+ movq m3, m5
+ PALIGNR m5, m6, 6, m6
+ PRED8x8_LOWPASS w, m1, m5, m2, m3
+ movq [r0+0*FDEC_STRIDEB], m1
+
+ psllq m0, 16
+ PALIGNR m1, m0, 6, m2
+ movq [r0+1*FDEC_STRIDEB], m1
+ psllq m0, 16
+ PALIGNR m1, m0, 6, m0
+ movq [r0+2*FDEC_STRIDEB], m1
+ psrlq m1, 16
+ movd [r0+3*FDEC_STRIDEB+4], m1
+ RET
+
+cglobal predict_4x4_hd_mmxext, 1,1
+ mova m0, [r0+1*FDEC_STRIDEB-8]
+ punpckhwd m0, [r0+0*FDEC_STRIDEB-8]
+ mova m1, [r0+3*FDEC_STRIDEB-8]
+ punpckhwd m1, [r0+2*FDEC_STRIDEB-8]
+ punpckhdq m1, m0
+ mova m0, m1
+ mova m4, m1
+
+ movu m3, [r0-1*FDEC_STRIDEB-2]
+ mova m7, m3
+ punpckhdq m4, [r0-1*FDEC_STRIDEB-6]
+ PALIGNR m3, m1, 2, m2
+ PRED8x8_LOWPASS w, m2, m4, m1, m3, m6
+
+ pavgw m0, m3
+ mova m5, m0
+ punpcklwd m5, m2
+ mova m4, m0
+ punpckhwd m4, m2
+ mova [r0+3*FDEC_STRIDEB], m5
+ mova [r0+1*FDEC_STRIDEB], m4
+
+ mova m4, m7
+ mova m6, [r0-1*FDEC_STRIDEB+0]
+ PALIGNR m7, [r0+0*FDEC_STRIDEB-8], 6, m5
+ PRED8x8_LOWPASS w, m3, m7, m6, m4, m1
+
+ PALIGNR m3, m0, 6, m5
+ mova [r0+0*FDEC_STRIDEB], m3
+ psrlq m0, 16
+ psrlq m2, 16
+ punpcklwd m0, m2
+ mova [r0+2*FDEC_STRIDEB], m0
+ RET
+
+INIT_XMM
+%define PALIGNR PALIGNR_MMX
+PREDICT_4x4 sse2 , wd, dq, dq, w, qdq, 2
%define PALIGNR PALIGNR_SSSE3
-PREDICT_4x4 ssse3
+PREDICT_4x4 ssse3 , wd, dq, dq, w, qdq, 2
+INIT_AVX
+PREDICT_4x4 avx , wd, dq, dq, w, qdq, 2
+%else
+INIT_MMX
+%define PALIGNR PALIGNR_MMX
+PREDICT_4x4 mmxext, bw, wd, q , b, dq , 8
+%define PALIGNR PALIGNR_SSSE3
+PREDICT_4x4 ssse3 , bw, wd, q , b, dq , 8
+%endif
;-----------------------------------------------------------------------------
-; void predict_4x4_hu_mmxext( uint8_t *src )
+; void predict_4x4_hu( pixel *src )
;-----------------------------------------------------------------------------
+%ifdef HIGH_BIT_DEPTH
+INIT_MMX
+cglobal predict_4x4_hu_mmxext, 1,1
+ movq m0, [r0+0*FDEC_STRIDEB-4*2]
+ punpckhwd m0, [r0+1*FDEC_STRIDEB-4*2]
+ movq m1, [r0+2*FDEC_STRIDEB-4*2]
+ punpckhwd m1, [r0+3*FDEC_STRIDEB-4*2]
+ punpckhdq m0, m1
+ pshufw m1, m1, 0xFF
+ movq [r0+3*FDEC_STRIDEB], m1
+ movd [r0+2*FDEC_STRIDEB+4], m1
+ mova m2, m0
+ psrlq m2, 16
+ pavgw m2, m0
+
+ pshufw m1, m0, 11111001b
+ pshufw m5, m0, 11111110b
+ PRED8x8_LOWPASS w, m3, m0, m5, m1, m7
+ movq m6, m2
+ punpcklwd m6, m3
+ mova [r0+0*FDEC_STRIDEB], m6
+ psrlq m2, 16
+ psrlq m3, 16
+ punpcklwd m2, m3
+ mova [r0+1*FDEC_STRIDEB], m2
+ psrlq m2, 32
+ movd [r0+2*FDEC_STRIDEB+0], m2
+ RET
+
+%else ; !HIGH_BIT_DEPTH
+INIT_MMX
cglobal predict_4x4_hu_mmxext, 1,1
movq mm0, [r0+0*FDEC_STRIDE-8]
punpckhbw mm0, [r0+1*FDEC_STRIDE-8]
psrlq mm2, 16
psrlq mm3, 8
pavgb mm7, mm3
- PRED8x8_LOWPASS mm4, mm0, mm2, mm3, mm5
+ PRED8x8_LOWPASS b, mm4, mm0, mm2, mm3, mm5
punpcklbw mm7, mm4
%assign Y 0
movd [r0+Y*FDEC_STRIDE], mm7
%endrep
movd [r0+3*FDEC_STRIDE], mm1
RET
+%endif ; HIGH_BIT_DEPTH
;-----------------------------------------------------------------------------
-; void predict_4x4_vl_mmxext( uint8_t *src )
+; void predict_4x4_vl( pixel *src )
;-----------------------------------------------------------------------------
-cglobal predict_4x4_vl_mmxext, 1,1
- movq mm1, [r0-FDEC_STRIDE]
- movq mm3, mm1
- movq mm2, mm1
- psrlq mm3, 8
- psrlq mm2, 16
- movq mm4, mm3
- pavgb mm4, mm1
-
- PRED8x8_LOWPASS mm0, mm1, mm2, mm3, mm5
-
- movd [r0+0*FDEC_STRIDE], mm4
- movd [r0+1*FDEC_STRIDE], mm0
- psrlq mm4, 8
- psrlq mm0, 8
- movd [r0+2*FDEC_STRIDE], mm4
- movd [r0+3*FDEC_STRIDE], mm0
+%macro PREDICT_4x4_V1 4
+cglobal predict_4x4_vl_%1, 1,1,6*(mmsize/16)
+ movu m1, [r0-FDEC_STRIDEB]
+ psrl%2 m3, m1, %3
+ psrl%2 m2, m1, %3*2
+ pavg%4 m4, m3, m1
+ PRED8x8_LOWPASS %4, m0, m1, m2, m3, m5
+ movh [r0+0*FDEC_STRIDEB], m4
+ movh [r0+1*FDEC_STRIDEB], m0
+ psrl%2 m4, %3
+ psrl%2 m0, %3
+ movh [r0+2*FDEC_STRIDEB], m4
+ movh [r0+3*FDEC_STRIDEB], m0
RET
+%endmacro
+
+%ifdef HIGH_BIT_DEPTH
+INIT_XMM
+PREDICT_4x4_V1 sse2, dq, 2, w
+%ifdef ARCH_X86_64
+INIT_AVX
+PREDICT_4x4_V1 avx , dq, 2, w
+%endif
+
+INIT_MMX
+%define PALIGNR PALIGNR_MMX
+cglobal predict_4x4_vl_mmxext, 1,4
+ mova m1, [r0-FDEC_STRIDEB+0]
+ mova m2, [r0-FDEC_STRIDEB+8]
+ mova m3, m2
+ PALIGNR m2, m1, 4, m6
+ PALIGNR m3, m1, 2, m5
+ mova m4, m3
+ pavgw m4, m1
+ mova [r0+0*FDEC_STRIDEB], m4
+ psrlq m4, 16
+ mova [r0+2*FDEC_STRIDEB], m4
+ PRED8x8_LOWPASS w, m0, m1, m2, m3, m6
+ mova [r0+1*FDEC_STRIDEB], m0
+ psrlq m0, 16
+ mova [r0+3*FDEC_STRIDEB], m0
+
+ movzx r1d, word [r0-FDEC_STRIDEB+ 8]
+ movzx r2d, word [r0-FDEC_STRIDEB+10]
+ movzx r3d, word [r0-FDEC_STRIDEB+12]
+ lea r1d, [r1+r2+1]
+ add r3d, r2d
+ lea r3d, [r3+r1+1]
+ shr r1d, 1
+ shr r3d, 2
+ mov [r0+2*FDEC_STRIDEB+6], r1w
+ mov [r0+3*FDEC_STRIDEB+6], r3w
+ RET
+%else
+INIT_MMX
+PREDICT_4x4_V1 mmxext, q , 8, b
+%endif
;-----------------------------------------------------------------------------
-; void predict_4x4_dc( uint8_t *src )
+; void predict_4x4_dc( pixel *src )
;-----------------------------------------------------------------------------
+%ifdef HIGH_BIT_DEPTH
+INIT_MMX
+cglobal predict_4x4_dc_mmxext, 1,1
+ mova m2, [r0+0*FDEC_STRIDEB-4*SIZEOF_PIXEL]
+ paddw m2, [r0+1*FDEC_STRIDEB-4*SIZEOF_PIXEL]
+ paddw m2, [r0+2*FDEC_STRIDEB-4*SIZEOF_PIXEL]
+ paddw m2, [r0+3*FDEC_STRIDEB-4*SIZEOF_PIXEL]
+ psrlq m2, 48
+ mova m0, [r0-FDEC_STRIDEB]
+ HADDW m0, m1
+ paddw m0, [pw_4]
+ paddw m0, m2
+ psrlw m0, 3
+ SPLATW m0, m0
+ mova [r0+0*FDEC_STRIDEB], m0
+ mova [r0+1*FDEC_STRIDEB], m0
+ mova [r0+2*FDEC_STRIDEB], m0
+ mova [r0+3*FDEC_STRIDEB], m0
+ RET
+%else
+INIT_MMX
cglobal predict_4x4_dc_mmxext, 1,4
pxor mm7, mm7
movd mm0, [r0-FDEC_STRIDE]
mov [r0+FDEC_STRIDE*2], r1d
mov [r0+FDEC_STRIDE*3], r1d
RET
+%endif ; HIGH_BIT_DEPTH
-%macro PREDICT_FILTER 1
+%macro PREDICT_FILTER 6
;-----------------------------------------------------------------------------
-;void predict_8x8_filter( uint8_t *src, uint8_t edge[33], int i_neighbor, int i_filters )
+;void predict_8x8_filter( pixel *src, pixel edge[33], int i_neighbor, int i_filters )
;-----------------------------------------------------------------------------
-
-cglobal predict_8x8_filter_%1, 4,5
- add r0, 0x58
-%define src r0-0x58
+cglobal predict_8x8_filter_%1, 4,5,7*(mmsize/16)
+ add r0, 0x58*SIZEOF_PIXEL
+%define src r0-0x58*SIZEOF_PIXEL
%ifndef ARCH_X86_64
mov r4, r1
%define t1 r4
%define t1 r1
%define t4 r4
%endif
- test r3b, 0x01
+ test r3b, 0x01
je .check_top
- movq mm0, [src+0*FDEC_STRIDE-8]
- punpckhbw mm0, [src-1*FDEC_STRIDE-8]
- movq mm1, [src+2*FDEC_STRIDE-8]
- punpckhbw mm1, [src+1*FDEC_STRIDE-8]
- punpckhwd mm1, mm0
- movq mm2, [src+4*FDEC_STRIDE-8]
- punpckhbw mm2, [src+3*FDEC_STRIDE-8]
- movq mm3, [src+6*FDEC_STRIDE-8]
- punpckhbw mm3, [src+5*FDEC_STRIDE-8]
- punpckhwd mm3, mm2
- punpckhdq mm3, mm1
- movq mm0, [src+7*FDEC_STRIDE-8]
- movq mm1, [src-1*FDEC_STRIDE]
- movq mm4, mm3
- movq mm2, mm3
- PALIGNR mm4, mm0, 7, mm0
- PALIGNR mm1, mm2, 1, mm2
- test r2b, 0x08
+ mova m0, [src+0*FDEC_STRIDEB-8*SIZEOF_PIXEL]
+ punpckh%2%3 m0, [src-1*FDEC_STRIDEB-8*SIZEOF_PIXEL]
+ mova m1, [src+2*FDEC_STRIDEB-8*SIZEOF_PIXEL]
+ punpckh%2%3 m1, [src+1*FDEC_STRIDEB-8*SIZEOF_PIXEL]
+ punpckh%3%4 m1, m0
+ mova m2, [src+4*FDEC_STRIDEB-8*SIZEOF_PIXEL]
+ punpckh%2%3 m2, [src+3*FDEC_STRIDEB-8*SIZEOF_PIXEL]
+ mova m3, [src+6*FDEC_STRIDEB-8*SIZEOF_PIXEL]
+ punpckh%2%3 m3, [src+5*FDEC_STRIDEB-8*SIZEOF_PIXEL]
+ punpckh%3%4 m3, m2
+ punpckh%4%5 m3, m1
+ mova m0, [src+7*FDEC_STRIDEB-8*SIZEOF_PIXEL]
+ mova m1, [src-1*FDEC_STRIDEB]
+ mova m4, m3
+ mova m2, m3
+ PALIGNR m4, m0, 7*SIZEOF_PIXEL, m0
+ PALIGNR m1, m2, 1*SIZEOF_PIXEL, m2
+ test r2b, 0x08
je .fix_lt_1
.do_left:
- movq mm0, mm4
- PRED8x8_LOWPASS mm2, mm1, mm4, mm3, mm5
- movq [t1+8], mm2
- movq mm4, mm0
- PRED8x8_LOWPASS mm1, mm3, mm0, mm4, mm5
- movd t4, mm1
- mov [t1+7], t4b
+ mova m0, m4
+ PRED8x8_LOWPASS %2, m2, m1, m4, m3, m5
+ mova [t1+8*SIZEOF_PIXEL], m2
+ mova m4, m0
+ PRED8x8_LOWPASS %2, m1, m3, m0, m4, m5
+ movd t4, m1
+ mov [t1+7*SIZEOF_PIXEL], t4%2
.check_top:
test r3b, 0x02
je .done
- movq mm0, [src-1*FDEC_STRIDE-8]
- movq mm3, [src-1*FDEC_STRIDE]
- movq mm1, [src-1*FDEC_STRIDE+8]
- movq mm2, mm3
- movq mm4, mm3
- PALIGNR mm2, mm0, 7, mm0
- PALIGNR mm1, mm4, 1, mm4
+ mova m0, [src-1*FDEC_STRIDEB-8*SIZEOF_PIXEL]
+ mova m3, [src-1*FDEC_STRIDEB]
+ mova m1, [src-1*FDEC_STRIDEB+8*SIZEOF_PIXEL]
+ mova m2, m3
+ mova m4, m3
+ PALIGNR m2, m0, 7*SIZEOF_PIXEL, m0
+ PALIGNR m1, m4, 1*SIZEOF_PIXEL, m4
test r2b, 0x08
je .fix_lt_2
test r2b, 0x04
je .fix_tr_1
.do_top:
- PRED8x8_LOWPASS mm4, mm2, mm1, mm3, mm5
- movq [t1+16], mm4
+ PRED8x8_LOWPASS %2, m4, m2, m1, m3, m5
+ mova [t1+16*SIZEOF_PIXEL], m4
test r3b, 0x04
je .done
test r2b, 0x04
je .fix_tr_2
- movq mm0, [src-1*FDEC_STRIDE+8]
- movq mm5, mm0
- movq mm2, mm0
- movq mm4, mm0
- psrlq mm5, 56
- PALIGNR mm2, mm3, 7, mm3
- PALIGNR mm5, mm4, 1, mm4
- PRED8x8_LOWPASS mm1, mm2, mm5, mm0, mm4
+ mova m0, [src-1*FDEC_STRIDEB+8*SIZEOF_PIXEL]
+ mova m2, m0
+ mova m4, m0
+ psrl%5 m5, m0, 7*%6
+ PALIGNR m2, m3, 7*SIZEOF_PIXEL, m3
+ PALIGNR m5, m4, 1*SIZEOF_PIXEL, m4
+ PRED8x8_LOWPASS %2, m1, m2, m5, m0, m4
jmp .do_topright
.fix_tr_2:
- punpckhbw mm3, mm3
- pshufw mm1, mm3, 0xFF
+ punpckh%2%3 m3, m3
+ pshuf%3 m1, m3, 0xFF
.do_topright:
- movq [t1+24], mm1
- psrlq mm1, 56
- movd t4, mm1
- mov [t1+32], t4b
+ mova [t1+24*SIZEOF_PIXEL], m1
+ psrl%5 m1, 7*%6
+ movd t4, m1
+ mov [t1+32*SIZEOF_PIXEL], t4%2
.done:
REP_RET
.fix_lt_1:
- movq mm5, mm3
- pxor mm5, mm4
- psrlq mm5, 56
- psllq mm5, 48
- pxor mm1, mm5
+ pxor m5, m3, m4
+ psrl%5 m5, 7*%6
+ psll%5 m5, 6*%6
+ pxor m1, m5
jmp .do_left
.fix_lt_2:
- movq mm5, mm3
- pxor mm5, mm2
- psllq mm5, 56
- psrlq mm5, 56
- pxor mm2, mm5
- test r2b, 0x04
+ pxor m5, m3, m2
+ psll%5 m5, 7*%6
+ psrl%5 m5, 7*%6
+ pxor m2, m5
+ test r2b, 0x04
jne .do_top
.fix_tr_1:
- movq mm5, mm3
- pxor mm5, mm1
- psrlq mm5, 56
- psllq mm5, 56
- pxor mm1, mm5
+ pxor m5, m3, m1
+ psrl%5 m5, 7*%6
+ psll%5 m5, 7*%6
+ pxor m1, m5
jmp .do_top
%endmacro
+%ifdef HIGH_BIT_DEPTH
+INIT_XMM
+%define PALIGNR PALIGNR_MMX
+PREDICT_FILTER sse2 , w, d, q, dq, 2
+%define PALIGNR PALIGNR_SSSE3
+PREDICT_FILTER ssse3 , w, d, q, dq, 2
+INIT_AVX
+PREDICT_FILTER avx , w, d, q, dq, 2
+%else
+INIT_MMX
%define PALIGNR PALIGNR_MMX
-PREDICT_FILTER mmxext
+PREDICT_FILTER mmxext, b, w, d, q , 8
%define PALIGNR PALIGNR_SSSE3
-PREDICT_FILTER ssse3
+PREDICT_FILTER ssse3 , b, w, d, q , 8
+%endif
;-----------------------------------------------------------------------------
-; void predict_8x8_v_mmxext( uint8_t *src, uint8_t *edge )
+; void predict_8x8_v( pixel *src, pixel *edge )
;-----------------------------------------------------------------------------
-cglobal predict_8x8_v_mmxext, 2,2
- movq mm0, [r1+16]
- STORE8x8 mm0, mm0
+%macro PREDICT_8x8_V 1
+cglobal predict_8x8_v_%1, 2,2
+ mova m0, [r1+16*SIZEOF_PIXEL]
+ STORE8x8 m0, m0
RET
+%endmacro
+
+%ifdef HIGH_BIT_DEPTH
+INIT_XMM
+PREDICT_8x8_V sse2
+%else
+INIT_MMX
+PREDICT_8x8_V mmxext
+%endif
;-----------------------------------------------------------------------------
-; void predict_8x8_h_mmxext( uint8_t *src, uint8_t edge[33] )
+; void predict_8x8_h( pixel *src, pixel edge[33] )
;-----------------------------------------------------------------------------
-
-INIT_MMX
-cglobal predict_8x8_h_mmxext, 2,2
- movu m3, [r1+7]
- mova m7, m3
- punpckhbw m3, m3
- punpcklbw m7, m7
- pshufw m0, m3, 0xff
- pshufw m1, m3, 0xaa
- pshufw m2, m3, 0x55
- pshufw m3, m3, 0x00
- pshufw m4, m7, 0xff
- pshufw m5, m7, 0xaa
- pshufw m6, m7, 0x55
- pshufw m7, m7, 0x00
+%macro PREDICT_8x8_H 3
+cglobal predict_8x8_h_%1, 2,2
+ movu m1, [r1+7*SIZEOF_PIXEL]
+ add r0, 4*FDEC_STRIDEB
+ punpckl%2 m2, m1, m1
+ punpckh%2 m1, m1
%assign n 0
%rep 8
- mova [r0+n*FDEC_STRIDE], m %+ n
+%assign i 1+n/4
+ SPLAT%3 m0, m %+ i, (3-n)&3
+ mova [r0+(n-4)*FDEC_STRIDEB], m0
%assign n n+1
%endrep
RET
+%endmacro
+
+%ifdef HIGH_BIT_DEPTH
+INIT_XMM
+PREDICT_8x8_H sse2 , wd, D
+%else
+INIT_MMX
+PREDICT_8x8_H mmxext, bw, W
+%endif
;-----------------------------------------------------------------------------
-; void predict_8x8_dc_mmxext( uint8_t *src, uint8_t *edge );
+; void predict_8x8_dc( pixel *src, pixel *edge );
;-----------------------------------------------------------------------------
+%ifdef HIGH_BIT_DEPTH
+INIT_XMM
+cglobal predict_8x8_dc_sse2, 2,2
+ movu m0, [r1+14]
+ paddw m0, [r1+32]
+ HADDW m0, m1
+ paddw m0, [pw_8]
+ psrlw m0, 4
+ SPLATW m0, m0
+ STORE8x8 m0, m0
+ REP_RET
+
+%else
+INIT_MMX
cglobal predict_8x8_dc_mmxext, 2,2
pxor mm0, mm0
pxor mm1, mm1
packuswb mm0, mm0
STORE8x8 mm0, mm0
RET
+%endif ; HIGH_BIT_DEPTH
;-----------------------------------------------------------------------------
-; void predict_8x8_dc_top_mmxext( uint8_t *src, uint8_t *edge );
+; void predict_8x8_dc_top ( pixel *src, pixel *edge );
+; void predict_8x8_dc_left( pixel *src, pixel *edge );
;-----------------------------------------------------------------------------
-%macro PRED8x8_DC 2
+%ifdef HIGH_BIT_DEPTH
+%macro PREDICT_8x8_DC 3
+cglobal %1, 2,2
+ %3 m0, [r1+%2]
+ HADDW m0, m1
+ paddw m0, [pw_4]
+ psrlw m0, 3
+ SPLATW m0, m0
+ STORE8x8 m0, m0
+ RET
+%endmacro
+INIT_XMM
+PREDICT_8x8_DC predict_8x8_dc_top_sse2 , 32, mova
+PREDICT_8x8_DC predict_8x8_dc_left_sse2, 14, movu
+
+%else
+%macro PREDICT_8x8_DC 2
cglobal %1, 2,2
pxor mm0, mm0
psadbw mm0, [r1+%2]
STORE8x8 mm0, mm0
RET
%endmacro
-
-PRED8x8_DC predict_8x8_dc_top_mmxext, 16
-PRED8x8_DC predict_8x8_dc_left_mmxext, 7
-
-%ifndef ARCH_X86_64
-; sse2 is faster even on amd, so there's no sense in spending exe size on these
-; functions if we know sse2 is available.
-
-;-----------------------------------------------------------------------------
-; void predict_8x8_ddl_mmxext( uint8_t *src, uint8_t *edge )
-;-----------------------------------------------------------------------------
-cglobal predict_8x8_ddl_mmxext, 2,2
- movq mm5, [r1+16]
- movq mm2, [r1+17]
- movq mm3, [r1+23]
- movq mm4, [r1+25]
- movq mm1, mm5
- psllq mm1, 8
- add r0, FDEC_STRIDE*4
- PRED8x8_LOWPASS mm0, mm1, mm2, mm5, mm7
- PRED8x8_LOWPASS mm1, mm3, mm4, [r1+24], mm6
+INIT_MMX
+PREDICT_8x8_DC predict_8x8_dc_top_mmxext, 16
+PREDICT_8x8_DC predict_8x8_dc_left_mmxext, 7
+%endif ; HIGH_BIT_DEPTH
+
+; sse2 is faster even on amd for 8-bit, so there's no sense in spending exe
+; size on the 8-bit mmx functions below if we know sse2 is available.
+%macro PREDICT_8x8 4
+;-----------------------------------------------------------------------------
+; void predict_8x8_ddl( pixel *src, pixel *edge )
+;-----------------------------------------------------------------------------
+cglobal predict_8x8_ddl_%1, 2,2,8*(mmsize/16)
+ mova m5, [r1+16*SIZEOF_PIXEL]
+ movu m2, [r1+17*SIZEOF_PIXEL]
+ movu m3, [r1+23*SIZEOF_PIXEL]
+ movu m4, [r1+25*SIZEOF_PIXEL]
+ psll%3 m1, m5, %4
+ add r0, FDEC_STRIDEB*4
+ PRED8x8_LOWPASS %2, m0, m1, m2, m5, m7
+%if avx_enabled == 1
+ INIT_XMM
+ PRED8x8_LOWPASS %2, m1, m3, m4, [r1+24*SIZEOF_PIXEL], m6
+ INIT_AVX
+%else
+ PRED8x8_LOWPASS %2, m1, m3, m4, [r1+24*SIZEOF_PIXEL], m6
+%endif
%assign Y 3
%rep 6
- movq [r0+Y*FDEC_STRIDE], mm1
- movq mm2, mm0
- psllq mm1, 8
- psrlq mm2, 56
- psllq mm0, 8
- por mm1, mm2
+ mova [r0+Y*FDEC_STRIDEB], m1
+ psll%3 m1, %4
+ psrl%3 m2, m0, 7*%4
+ psll%3 m0, %4
+ por m1, m2
%assign Y (Y-1)
%endrep
- movq [r0+Y*FDEC_STRIDE], mm1
- psllq mm1, 8
- psrlq mm0, 56
- por mm1, mm0
+ mova [r0+Y*FDEC_STRIDEB], m1
+ psll%3 m1, %4
+ psrl%3 m0, 7*%4
+ por m1, m0
%assign Y (Y-1)
- movq [r0+Y*FDEC_STRIDE], mm1
+ mova [r0+Y*FDEC_STRIDEB], m1
RET
;-----------------------------------------------------------------------------
-; void predict_8x8_ddr_mmxext( uint8_t *src, uint8_t *edge )
+; void predict_8x8_ddr( pixel *src, pixel *edge )
;-----------------------------------------------------------------------------
-cglobal predict_8x8_ddr_mmxext, 2,2
- movq mm1, [r1+7]
- movq mm2, [r1+9]
- movq mm3, [r1+15]
- movq mm4, [r1+17]
- add r0, FDEC_STRIDE*4
- PRED8x8_LOWPASS mm0, mm1, mm2, [r1+8], mm7
- PRED8x8_LOWPASS mm1, mm3, mm4, [r1+16], mm6
+%if avx_enabled == 0
+cglobal predict_8x8_ddr_%1, 2,2,7*(mmsize/16)
+ movu m1, [r1+ 7*SIZEOF_PIXEL]
+ movu m2, [r1+ 9*SIZEOF_PIXEL]
+ movu m3, [r1+15*SIZEOF_PIXEL]
+ movu m4, [r1+17*SIZEOF_PIXEL]
+ add r0, FDEC_STRIDEB*4
+ PRED8x8_LOWPASS %2, m0, m1, m2, [r1+ 8*SIZEOF_PIXEL], m5
+ PRED8x8_LOWPASS %2, m1, m3, m4, [r1+16*SIZEOF_PIXEL], m6
%assign Y 3
%rep 6
- movq [r0+Y*FDEC_STRIDE], mm0
- movq mm2, mm1
- psrlq mm0, 8
- psllq mm2, 56
- psrlq mm1, 8
- por mm0, mm2
+ mova [r0+Y*FDEC_STRIDEB], m0
+ psrl%3 m0, %4
+ psll%3 m2, m1, 7*%4
+ psrl%3 m1, %4
+ por m0, m2
%assign Y (Y-1)
%endrep
- movq [r0+Y*FDEC_STRIDE], mm0
- psrlq mm0, 8
- psllq mm1, 56
- por mm0, mm1
+ mova [r0+Y*FDEC_STRIDEB], m0
+ psrl%3 m0, %4
+ psll%3 m1, 7*%4
+ por m0, m1
%assign Y (Y-1)
- movq [r0+Y*FDEC_STRIDE], mm0
+ mova [r0+Y*FDEC_STRIDEB], m0
RET
+%endif
+%endmacro ; PREDICT_8x8
-;-----------------------------------------------------------------------------
-; void predict_8x8_hu_mmxext( uint8_t *src, uint8_t *edge )
-;-----------------------------------------------------------------------------
-%define PALIGNR PALIGNR_MMX
-cglobal predict_8x8_hu_mmxext, 2,2
- movq mm1, [r1+7] ; l0 l1 l2 l3 l4 l5 l6 l7
- add r0, 4*FDEC_STRIDE
- pshufw mm0, mm1, 00011011b ; l6 l7 l4 l5 l2 l3 l0 l1
- psllq mm1, 56 ; l7 .. .. .. .. .. .. ..
- movq mm2, mm0
- psllw mm0, 8
- psrlw mm2, 8
- por mm2, mm0 ; l7 l6 l5 l4 l3 l2 l1 l0
- movq mm3, mm2
- movq mm4, mm2
- movq mm5, mm2
- psrlq mm2, 8
- psrlq mm3, 16
- por mm2, mm1 ; l7 l7 l6 l5 l4 l3 l2 l1
- punpckhbw mm1, mm1
- por mm3, mm1 ; l7 l7 l7 l6 l5 l4 l3 l2
- pavgb mm4, mm2
- PRED8x8_LOWPASS mm1, mm3, mm5, mm2, mm6
- movq mm5, mm4
- punpcklbw mm4, mm1 ; p4 p3 p2 p1
- punpckhbw mm5, mm1 ; p8 p7 p6 p5
- movq mm6, mm5
- movq mm7, mm5
- movq mm0, mm5
- PALIGNR mm5, mm4, 2, mm1
- pshufw mm1, mm6, 11111001b
- PALIGNR mm6, mm4, 4, mm2
- pshufw mm2, mm7, 11111110b
- PALIGNR mm7, mm4, 6, mm3
- pshufw mm3, mm0, 11111111b
- movq [r0-4*FDEC_STRIDE], mm4
- movq [r0-3*FDEC_STRIDE], mm5
- movq [r0-2*FDEC_STRIDE], mm6
- movq [r0-1*FDEC_STRIDE], mm7
- movq [r0+0*FDEC_STRIDE], mm0
- movq [r0+1*FDEC_STRIDE], mm1
- movq [r0+2*FDEC_STRIDE], mm2
- movq [r0+3*FDEC_STRIDE], mm3
- RET
+%ifdef HIGH_BIT_DEPTH
+INIT_XMM
+PREDICT_8x8 sse2 , w, dq, 2
+INIT_AVX
+PREDICT_8x8 avx , w, dq, 2
+%elifndef ARCH_X86_64
+INIT_MMX
+PREDICT_8x8 mmxext, b, q , 8
+%endif
;-----------------------------------------------------------------------------
-; void predict_8x8_vr_core_mmxext( uint8_t *src, uint8_t *edge )
-;-----------------------------------------------------------------------------
+; void predict_8x8_hu( pixel *src, pixel *edge )
+;-----------------------------------------------------------------------------
+%macro PREDICT_8x8_HU 6
+cglobal predict_8x8_hu_%1, 2,2,8*(mmsize/16)
+ movu m1, [r1+7*SIZEOF_PIXEL] ; l0 l1 l2 l3 l4 l5 l6 l7
+ add r0, 4*FDEC_STRIDEB
+ pshuf%4 m0, m1, 00011011b ; l6 l7 l4 l5 l2 l3 l0 l1
+ psll%3 m1, 7*%6 ; l7 .. .. .. .. .. .. ..
+ mova m2, m0
+ psll%4 m0, 8*SIZEOF_PIXEL
+ psrl%4 m2, 8*SIZEOF_PIXEL
+ por m2, m0 ; l7 l6 l5 l4 l3 l2 l1 l0
+ mova m4, m2
+ mova m5, m2
+ psrl%3 m3, m2, 2*%6
+ psrl%3 m2, %6
+ por m2, m1 ; l7 l7 l6 l5 l4 l3 l2 l1
+ punpckh%5 m1, m1
+ por m3, m1 ; l7 l7 l7 l6 l5 l4 l3 l2
+ pavg%2 m4, m2
+ PRED8x8_LOWPASS %2, m1, m3, m5, m2, m6
+ punpckh%5 m5, m4, m1 ; p8 p7 p6 p5
+ punpckl%5 m4, m1 ; p4 p3 p2 p1
+ mova m6, m5
+ mova m7, m5
+ mova m0, m5
+ PALIGNR m5, m4, 2*SIZEOF_PIXEL, m1
+ pshuf%4 m1, m6, 11111001b
+ PALIGNR m6, m4, 4*SIZEOF_PIXEL, m2
+ pshuf%4 m2, m7, 11111110b
+ PALIGNR m7, m4, 6*SIZEOF_PIXEL, m3
+ pshuf%4 m3, m0, 11111111b
+ mova [r0-4*FDEC_STRIDEB], m4
+ mova [r0-3*FDEC_STRIDEB], m5
+ mova [r0-2*FDEC_STRIDEB], m6
+ mova [r0-1*FDEC_STRIDEB], m7
+ mova [r0+0*FDEC_STRIDEB], m0
+ mova [r0+1*FDEC_STRIDEB], m1
+ mova [r0+2*FDEC_STRIDEB], m2
+ mova [r0+3*FDEC_STRIDEB], m3
+ RET
+%endmacro
-; fills only some pixels:
-; f01234567
-; 0........
-; 1,,,,,,,,
-; 2 .......
-; 3 ,,,,,,,
-; 4 ......
-; 5 ,,,,,,
-; 6 .....
-; 7 ,,,,,
-
-cglobal predict_8x8_vr_core_mmxext, 2,2
- movq mm2, [r1+16]
- movq mm3, [r1+15]
- movq mm1, [r1+14]
- movq mm4, mm3
- pavgb mm3, mm2
- add r0, FDEC_STRIDE*4
- PRED8x8_LOWPASS mm0, mm1, mm2, mm4, mm7
+%ifdef HIGH_BIT_DEPTH
+INIT_XMM
+%define PALIGNR PALIGNR_MMX
+PREDICT_8x8_HU sse2 , w, dq, d, wd, 2
+%define PALIGNR PALIGNR_SSSE3
+PREDICT_8x8_HU ssse3 , w, dq, d, wd, 2
+INIT_AVX
+PREDICT_8x8_HU avx , w, dq, d, wd, 2
+%elifndef ARCH_X86_64
+INIT_MMX
+%define PALIGNR PALIGNR_MMX
+PREDICT_8x8_HU mmxext, b, q , w, bw, 8
+%endif
-%assign Y -4
-%rep 3
- movq [r0+ Y *FDEC_STRIDE], mm3
- movq [r0+(Y+1)*FDEC_STRIDE], mm0
- psllq mm3, 8
- psllq mm0, 8
-%assign Y (Y+2)
+;-----------------------------------------------------------------------------
+; void predict_8x8_vr( pixel *src, pixel *edge )
+;-----------------------------------------------------------------------------
+%macro PREDICT_8x8_VR 4
+cglobal predict_8x8_vr_%1, 2,3,7*(mmsize/16)
+ mova m2, [r1+16*SIZEOF_PIXEL]
+ movu m3, [r1+15*SIZEOF_PIXEL]
+ movu m1, [r1+14*SIZEOF_PIXEL]
+ pavg%2 m4, m3, m2
+ add r0, FDEC_STRIDEB*4
+ PRED8x8_LOWPASS %2, m0, m1, m2, m3, m5
+ mova [r0-4*FDEC_STRIDEB], m4
+ mova [r0-3*FDEC_STRIDEB], m0
+ mova m5, m0
+ mova m6, m4
+ mova m1, [r1+8*SIZEOF_PIXEL]
+ mova m2, m1
+ psll%3 m2, %4
+ mova m3, m1
+ psll%3 m3, 2*%4
+ PRED8x8_LOWPASS %2, m0, m1, m3, m2, m4
+
+%assign Y -2
+%rep 5
+ %assign i (5 + ((Y+3)&1))
+ PALIGNR m %+ i, m0, 7*SIZEOF_PIXEL, m2
+ mova [r0+Y*FDEC_STRIDEB], m %+ i
+ psll%3 m0, %4
+%assign Y (Y+1)
%endrep
- movq [r0+ Y *FDEC_STRIDE], mm3
- movq [r0+(Y+1)*FDEC_STRIDE], mm0
-
+ PALIGNR m5, m0, 7*SIZEOF_PIXEL, m0
+ mova [r0+Y*FDEC_STRIDEB], m5
RET
+%endmacro
+
+%ifdef HIGH_BIT_DEPTH
+INIT_XMM
+%define PALIGNR PALIGNR_MMX
+PREDICT_8x8_VR sse2 , w, dq, 2
+%define PALIGNR PALIGNR_SSSE3
+PREDICT_8x8_VR ssse3 , w, dq, 2
+INIT_AVX
+PREDICT_8x8_VR avx , w, dq, 2
+%else
+INIT_MMX
+%define PALIGNR PALIGNR_MMX
+PREDICT_8x8_VR mmxext, b, q , 8
+%endif
;-----------------------------------------------------------------------------
-; void predict_8x8c_p_core_mmxext( uint8_t *src, int i00, int b, int c )
+; void predict_8x8c_p_core( uint8_t *src, int i00, int b, int c )
;-----------------------------------------------------------------------------
+%ifndef ARCH_X86_64
+INIT_MMX
cglobal predict_8x8c_p_core_mmxext, 1,2
LOAD_PLANE_ARGS
movq mm1, mm2
dec r1d
jg .loop
REP_RET
+%endif ; !ARCH_X86_64
+
+INIT_XMM
+%ifdef HIGH_BIT_DEPTH
+cglobal predict_8x8c_p_core_sse2, 1,1,7
+ movd m0, r1m
+ movd m2, r2m
+ movd m4, r3m
+ mova m3, [pw_pixel_max]
+ pxor m1, m1
+ SPLATW m0, m0, 0
+ SPLATW m2, m2, 0
+ SPLATW m4, m4, 0
+ pmullw m2, [pw_43210123] ; b
+ pmullw m5, m4, [pw_m3] ; c
+ paddw m5, [pw_16]
+ mov r1d, 8
+.loop:
+ paddsw m6, m2, m5
+ paddsw m6, m0
+ psraw m6, 5
+ CLIPW m6, m1, m3
+ mova [r0], m6
+ paddw m5, m4
+ add r0, FDEC_STRIDEB
+ dec r1d
+ jg .loop
+ REP_RET
+%else ; !HIGH_BIT_DEPTH
+cglobal predict_8x8c_p_core_sse2, 1,1
+ movd m0, r1m
+ movd m2, r2m
+ movd m4, r3m
+ SPLATW m0, m0, 0
+ SPLATW m2, m2, 0
+ SPLATW m4, m4, 0
+ pmullw m2, [pw_76543210]
+ paddsw m0, m2 ; m0 = {i+0*b, i+1*b, i+2*b, i+3*b, i+4*b, i+5*b, i+6*b, i+7*b}
+ paddsw m3, m0, m4
+ paddsw m4, m4
+call .loop
+ add r0, FDEC_STRIDE*4
+.loop:
+ paddsw m1, m3, m4
+ paddsw m5, m0, m4
+ psraw m3, 5
+ psraw m0, 5
+ packuswb m0, m3
+ movq [r0+FDEC_STRIDE*0], m0
+ movhps [r0+FDEC_STRIDE*1], m0
+ paddsw m0, m5, m4
+ paddsw m3, m1, m4
+ psraw m5, 5
+ psraw m1, 5
+ packuswb m5, m1
+ movq [r0+FDEC_STRIDE*2], m5
+ movhps [r0+FDEC_STRIDE*3], m5
+ RET
+%endif ; HIGH_BIT_DEPTH
;-----------------------------------------------------------------------------
-; void predict_16x16_p_core_mmxext( uint8_t *src, int i00, int b, int c )
+; void predict_16x16_p_core( uint8_t *src, int i00, int b, int c )
;-----------------------------------------------------------------------------
+%ifndef ARCH_X86_64
cglobal predict_16x16_p_core_mmxext, 1,2
LOAD_PLANE_ARGS
movq mm5, mm2
dec r1d
jg .loop
REP_RET
-
%endif ; !ARCH_X86_64
+%macro PREDICT_16x16_P 1
+cglobal predict_16x16_p_core_%1, 1,2,8
+ movd m0, r1m
+ movd m1, r2m
+ movd m2, r3m
+ SPLATW m0, m0, 0
+ SPLATW m1, m1, 0
+ SPLATW m2, m2, 0
+ pmullw m3, m1, [pw_76543210]
+ psllw m1, 3
+%ifdef HIGH_BIT_DEPTH
+ pxor m6, m6
+ mov r1d, 16
+.loop:
+ mova m4, m0
+ mova m5, m0
+ mova m7, m3
+ paddsw m7, m6
+ paddsw m4, m7
+ paddsw m7, m1
+ paddsw m5, m7
+ psraw m4, 5
+ psraw m5, 5
+ CLIPW m4, [pb_0], [pw_pixel_max]
+ CLIPW m5, [pb_0], [pw_pixel_max]
+ mova [r0], m4
+ mova [r0+16], m5
+ add r0, FDEC_STRIDEB
+ paddw m6, m2
+ dec r1d
+ jg .loop
+%else ; !HIGH_BIT_DEPTH
+ paddsw m0, m3 ; m0 = {i+ 0*b, i+ 1*b, i+ 2*b, i+ 3*b, i+ 4*b, i+ 5*b, i+ 6*b, i+ 7*b}
+ paddsw m1, m0 ; m1 = {i+ 8*b, i+ 9*b, i+10*b, i+11*b, i+12*b, i+13*b, i+14*b, i+15*b}
+ paddsw m7, m2, m2
+ mov r1d, 8
+ALIGN 4
+.loop:
+ psraw m3, m0, 5
+ psraw m4, m1, 5
+ paddsw m5, m0, m2
+ paddsw m6, m1, m2
+ psraw m5, 5
+ psraw m6, 5
+ packuswb m3, m4
+ packuswb m5, m6
+ mova [r0+FDEC_STRIDE*0], m3
+ mova [r0+FDEC_STRIDE*1], m5
+ paddsw m0, m7
+ paddsw m1, m7
+ add r0, FDEC_STRIDE*2
+ dec r1d
+ jg .loop
+%endif ; !HIGH_BIT_DEPTH
+ REP_RET
+%endmacro ; PREDICT_16x16_P
+
+INIT_XMM
+PREDICT_16x16_P sse2
+%ifndef HIGH_BIT_DEPTH
+INIT_AVX
+PREDICT_16x16_P avx
+%endif
+
+%ifndef HIGH_BIT_DEPTH
+%macro PREDICT_8x8 1
;-----------------------------------------------------------------------------
-; void predict_8x8_ddl_sse2( uint8_t *src, uint8_t *edge )
+; void predict_8x8_ddl( uint8_t *src, uint8_t *edge )
;-----------------------------------------------------------------------------
-cglobal predict_8x8_ddl_sse2, 2,2
+cglobal predict_8x8_ddl_%1, 2,2
movdqa xmm3, [r1+16]
movdqu xmm2, [r1+17]
- movdqa xmm1, xmm3
- pslldq xmm1, 1
+ pslldq xmm1, xmm3, 1
add r0, FDEC_STRIDE*4
- PRED8x8_LOWPASS_XMM xmm0, xmm1, xmm2, xmm3, xmm4
+ PRED8x8_LOWPASS b, xmm0, xmm1, xmm2, xmm3, xmm4
%assign Y -4
%rep 8
RET
;-----------------------------------------------------------------------------
-; void predict_8x8_ddr_sse2( uint8_t *src, uint8_t *edge )
+; void predict_8x8_ddr( uint8_t *src, uint8_t *edge )
;-----------------------------------------------------------------------------
-cglobal predict_8x8_ddr_sse2, 2,2
+cglobal predict_8x8_ddr_%1, 2,2
movdqu xmm3, [r1+8]
movdqu xmm1, [r1+7]
- movdqa xmm2, xmm3
- psrldq xmm2, 1
+ psrldq xmm2, xmm3, 1
add r0, FDEC_STRIDE*4
- PRED8x8_LOWPASS_XMM xmm0, xmm1, xmm2, xmm3, xmm4
+ PRED8x8_LOWPASS b, xmm0, xmm1, xmm2, xmm3, xmm4
- movdqa xmm1, xmm0
- psrldq xmm1, 1
+ psrldq xmm1, xmm0, 1
%assign Y 3
%rep 3
movq [r0+Y*FDEC_STRIDE], xmm0
%endrep
movq [r0-3*FDEC_STRIDE], xmm0
movq [r0-4*FDEC_STRIDE], xmm1
-
RET
;-----------------------------------------------------------------------------
-; void predict_8x8_vl_sse2( uint8_t *src, uint8_t *edge )
+; void predict_8x8_vl( uint8_t *src, uint8_t *edge )
;-----------------------------------------------------------------------------
-cglobal predict_8x8_vl_sse2, 2,2
+cglobal predict_8x8_vl_%1, 2,2
movdqa xmm4, [r1+16]
- movdqa xmm2, xmm4
- movdqa xmm1, xmm4
- movdqa xmm3, xmm4
- psrldq xmm2, 1
- pslldq xmm1, 1
- pavgb xmm3, xmm2
+ pslldq xmm1, xmm4, 1
+ psrldq xmm2, xmm4, 1
+ pavgb xmm3, xmm4, xmm2
add r0, FDEC_STRIDE*4
- PRED8x8_LOWPASS_XMM xmm0, xmm1, xmm2, xmm4, xmm5
+ PRED8x8_LOWPASS b, xmm0, xmm1, xmm2, xmm4, xmm5
; xmm0: (t0 + 2*t1 + t2 + 2) >> 2
; xmm3: (t0 + t1 + 1) >> 1
RET
;-----------------------------------------------------------------------------
-; void predict_8x8_vr_sse2( uint8_t *src, uint8_t *edge )
+; void predict_8x8_vr( uint8_t *src, uint8_t *edge )
;-----------------------------------------------------------------------------
-cglobal predict_8x8_vr_sse2, 2,2,7
+cglobal predict_8x8_vr_%1, 2,2,7
movdqu xmm0, [r1+8]
movdqa xmm6, [pw_ff00]
add r0, 4*FDEC_STRIDE
- movdqa xmm1, xmm0
movdqa xmm2, xmm0
movdqa xmm3, xmm0
+ pslldq xmm1, xmm0, 2
pslldq xmm0, 1
- pslldq xmm1, 2
pavgb xmm2, xmm0
- PRED8x8_LOWPASS_XMM xmm4, xmm3, xmm1, xmm0, xmm5
+ PRED8x8_LOWPASS b, xmm4, xmm3, xmm1, xmm0, xmm5
pandn xmm6, xmm4
movdqa xmm5, xmm4
psrlw xmm4, 8
%assign Y (Y-2)
%endrep
RET
+%endmacro ; PREDICT_8x8
-;-----------------------------------------------------------------------------
-; void predict_8x8_hd_mmxext( uint8_t *src, uint8_t *edge )
-;-----------------------------------------------------------------------------
-%define PALIGNR PALIGNR_MMX
-cglobal predict_8x8_hd_mmxext, 2,2
- add r0, 4*FDEC_STRIDE
- movq mm0, [r1] ; l7 .. .. .. .. .. .. ..
- movq mm1, [r1+8] ; lt l0 l1 l2 l3 l4 l5 l6
- movq mm2, [r1+16] ; t7 t6 t5 t4 t3 t2 t1 t0
- movq mm3, mm1 ; lt l0 l1 l2 l3 l4 l5 l6
- movq mm4, mm2 ; t7 t6 t5 t4 t3 t2 t1 t0
- PALIGNR mm2, mm1, 7, mm5 ; t6 t5 t4 t3 t2 t1 t0 lt
- PALIGNR mm1, mm0, 7, mm6 ; l0 l1 l2 l3 l4 l5 l6 l7
- PALIGNR mm4, mm3, 1, mm7 ; t0 lt l0 l1 l2 l3 l4 l5
- movq mm5, mm3
- pavgb mm3, mm1
- PRED8x8_LOWPASS mm0, mm4, mm1, mm5, mm7
- movq mm4, mm2
- movq mm1, mm2 ; t6 t5 t4 t3 t2 t1 t0 lt
- psrlq mm4, 16 ; .. .. t6 t5 t4 t3 t2 t1
- psrlq mm1, 8 ; .. t6 t5 t4 t3 t2 t1 t0
- PRED8x8_LOWPASS mm6, mm4, mm2, mm1, mm5
- ; .. p11 p10 p9
- movq mm7, mm3
- punpcklbw mm3, mm0 ; p4 p3 p2 p1
- punpckhbw mm7, mm0 ; p8 p7 p6 p5
- movq mm1, mm7
- movq mm0, mm7
- movq mm4, mm7
- movq [r0+3*FDEC_STRIDE], mm3
- PALIGNR mm7, mm3, 2, mm5
- movq [r0+2*FDEC_STRIDE], mm7
- PALIGNR mm1, mm3, 4, mm5
- movq [r0+1*FDEC_STRIDE], mm1
- PALIGNR mm0, mm3, 6, mm3
- movq [r0+0*FDEC_STRIDE], mm0
- movq mm2, mm6
- movq mm3, mm6
- movq [r0-1*FDEC_STRIDE], mm4
- PALIGNR mm6, mm4, 2, mm5
- movq [r0-2*FDEC_STRIDE], mm6
- PALIGNR mm2, mm4, 4, mm5
- movq [r0-3*FDEC_STRIDE], mm2
- PALIGNR mm3, mm4, 6, mm4
- movq [r0-4*FDEC_STRIDE], mm3
+INIT_XMM
+PREDICT_8x8 sse2
+INIT_AVX
+PREDICT_8x8 avx
+
+%endif ; !HIGH_BIT_DEPTH
+
+;-----------------------------------------------------------------------------
+; void predict_8x8_hd( pixel *src, pixel *edge )
+;-----------------------------------------------------------------------------
+%macro PREDICT_8x8_HD 5
+cglobal predict_8x8_hd_%1, 2,2,8*(mmsize/16)
+ add r0, 4*FDEC_STRIDEB
+ mova m0, [r1] ; l7 .. .. .. .. .. .. ..
+ mova m1, [r1+ 8*SIZEOF_PIXEL] ; lt l0 l1 l2 l3 l4 l5 l6
+ mova m2, [r1+16*SIZEOF_PIXEL] ; t7 t6 t5 t4 t3 t2 t1 t0
+ mova m3, m1 ; lt l0 l1 l2 l3 l4 l5 l6
+ mova m4, m2 ; t7 t6 t5 t4 t3 t2 t1 t0
+ PALIGNR m2, m1, 7*SIZEOF_PIXEL, m5 ; t6 t5 t4 t3 t2 t1 t0 lt
+ PALIGNR m1, m0, 7*SIZEOF_PIXEL, m6 ; l0 l1 l2 l3 l4 l5 l6 l7
+ PALIGNR m4, m3, 1*SIZEOF_PIXEL, m7 ; t0 lt l0 l1 l2 l3 l4 l5
+ mova m5, m3
+ pavg%2 m3, m1
+ PRED8x8_LOWPASS %2, m0, m4, m1, m5, m7
+ psrl%3 m4, m2, 2*%5 ; .. .. t6 t5 t4 t3 t2 t1
+ psrl%3 m1, m2, %5 ; .. t6 t5 t4 t3 t2 t1 t0
+ PRED8x8_LOWPASS %2, m6, m4, m2, m1, m5
+ ; .. p11 p10 p9
+ punpckh%4 m7, m3, m0 ; p8 p7 p6 p5
+ punpckl%4 m3, m0 ; p4 p3 p2 p1
+ mova m1, m7
+ mova m0, m7
+ mova m4, m7
+ mova [r0+3*FDEC_STRIDEB], m3
+ PALIGNR m7, m3, 2*SIZEOF_PIXEL, m5
+ mova [r0+2*FDEC_STRIDEB], m7
+ PALIGNR m1, m3, 4*SIZEOF_PIXEL, m5
+ mova [r0+1*FDEC_STRIDEB], m1
+ PALIGNR m0, m3, 6*SIZEOF_PIXEL, m3
+ mova [r0+0*FDEC_STRIDEB], m0
+ mova m2, m6
+ mova m3, m6
+ mova [r0-1*FDEC_STRIDEB], m4
+ PALIGNR m6, m4, 2*SIZEOF_PIXEL, m5
+ mova [r0-2*FDEC_STRIDEB], m6
+ PALIGNR m2, m4, 4*SIZEOF_PIXEL, m5
+ mova [r0-3*FDEC_STRIDEB], m2
+ PALIGNR m3, m4, 6*SIZEOF_PIXEL, m4
+ mova [r0-4*FDEC_STRIDEB], m3
RET
+%endmacro
+
+%ifdef HIGH_BIT_DEPTH
+INIT_XMM
+%define PALIGNR PALIGNR_MMX
+PREDICT_8x8_HD sse2 , w, dq, wd, 2
+%define PALIGNR PALIGNR_SSSE3
+PREDICT_8x8_HD ssse3 , w, dq, wd, 2
+INIT_AVX
+PREDICT_8x8_HD avx , w, dq, wd, 2
+%else
+INIT_MMX
+%define PALIGNR PALIGNR_MMX
+PREDICT_8x8_HD mmxext, b, q , bw, 8
;-----------------------------------------------------------------------------
-; void predict_8x8_hd_ssse3( uint8_t *src, uint8_t *edge )
+; void predict_8x8_hd( uint8_t *src, uint8_t *edge )
;-----------------------------------------------------------------------------
%macro PREDICT_8x8_HD 1
cglobal predict_8x8_hd_%1, 2,2
PALIGNR xmm1, xmm0, 7, xmm4
PALIGNR xmm2, xmm0, 9, xmm5
PALIGNR xmm3, xmm0, 8, xmm0
- movdqa xmm4, xmm1
- pavgb xmm4, xmm3
- PRED8x8_LOWPASS_XMM xmm0, xmm1, xmm2, xmm3, xmm5
+ pavgb xmm4, xmm1, xmm3
+ PRED8x8_LOWPASS b, xmm0, xmm1, xmm2, xmm3, xmm5
punpcklbw xmm4, xmm0
movhlps xmm0, xmm4
PREDICT_8x8_HD sse2
%define PALIGNR PALIGNR_SSSE3
PREDICT_8x8_HD ssse3
+INIT_AVX
+PREDICT_8x8_HD avx
INIT_MMX
%define PALIGNR PALIGNR_MMX
+%endif ; HIGH_BIT_DEPTH
+INIT_MMX
;-----------------------------------------------------------------------------
-; void predict_8x8_hu_sse2( uint8_t *src, uint8_t *edge )
+; void predict_8x8_hu( uint8_t *src, uint8_t *edge )
;-----------------------------------------------------------------------------
%macro PREDICT_8x8_HU 1
cglobal predict_8x8_hu_%1, 2,2
por mm3, mm1 ; l7 l7 l7 l6 l5 l4 l3 l2
%endif
pavgb mm4, mm2
- PRED8x8_LOWPASS mm1, mm3, mm5, mm2, mm6
+ PRED8x8_LOWPASS b, mm1, mm3, mm5, mm2, mm6
movq2dq xmm0, mm4
movq2dq xmm1, mm1
RET
%endmacro
+%ifndef HIGH_BIT_DEPTH
PREDICT_8x8_HU sse2
PREDICT_8x8_HU ssse3
+%endif
;-----------------------------------------------------------------------------
-; void predict_8x8c_v_mmx( uint8_t *src )
+; void predict_8x8c_v( uint8_t *src )
;-----------------------------------------------------------------------------
+
+%macro PREDICT_8x8C_V 1
+cglobal predict_8x8c_v_%1, 1,1
+ mova m0, [r0 - FDEC_STRIDEB]
+ STORE8x8 m0, m0
+ RET
+%endmacro
+
+%ifdef HIGH_BIT_DEPTH
+INIT_XMM
+PREDICT_8x8C_V sse2
+%else
+INIT_MMX
+PREDICT_8x8C_V mmx
+%endif
+
+%ifdef HIGH_BIT_DEPTH
+
+INIT_MMX
cglobal predict_8x8c_v_mmx, 1,1
- movq mm0, [r0 - FDEC_STRIDE]
- STORE8x8 mm0, mm0
+ mova m0, [r0 - FDEC_STRIDEB]
+ mova m1, [r0 - FDEC_STRIDEB + 8]
+%assign n 0
+%rep 8
+ mova [r0 + (n&1)*FDEC_STRIDEB], m0
+ mova [r0 + (n&1)*FDEC_STRIDEB + 8], m1
+%if (n&1) && (n!=7)
+ add r0, FDEC_STRIDEB*2
+%endif
+%assign n n+1
+%endrep
RET
+%endif
+
;-----------------------------------------------------------------------------
-; void predict_8x8c_h_mmxext( uint8_t *src )
+; void predict_8x8c_h( uint8_t *src )
;-----------------------------------------------------------------------------
+%ifdef HIGH_BIT_DEPTH
+%macro PREDICT_8x8C_H 1
-%macro PRED_8x8C_H 1
+cglobal predict_8x8c_h_%1, 1,1
+ add r0, FDEC_STRIDEB*4
+%assign n -4
+%rep 8
+ movd m0, [r0+FDEC_STRIDEB*n-SIZEOF_PIXEL*2]
+ SPLATW m0, m0, 1
+ mova [r0+FDEC_STRIDEB*n], m0
+%assign n n+1
+%endrep
+ RET
+
+%endmacro
+
+INIT_XMM
+PREDICT_8x8C_H sse2
+
+%else
+
+%macro PREDICT_8x8C_H 1
cglobal predict_8x8c_h_%1, 1,1
%ifidn %1, ssse3
mova m1, [pb_3]
%endif
-%assign n 0
+ add r0, FDEC_STRIDE*4
+%assign n -4
%rep 8
SPLATB m0, r0+FDEC_STRIDE*n-1, m1
mova [r0+FDEC_STRIDE*n], m0
INIT_MMX
%define SPLATB SPLATB_MMX
-PRED_8x8C_H mmxext
+PREDICT_8x8C_H mmxext
%define SPLATB SPLATB_SSSE3
-PRED_8x8C_H ssse3
+PREDICT_8x8C_H ssse3
+%endif
;-----------------------------------------------------------------------------
-; void predict_8x8c_dc_core_mmxext( uint8_t *src, int s2, int s3 )
+; void predict_8x8c_dc( pixel *src )
;-----------------------------------------------------------------------------
-cglobal predict_8x8c_dc_core_mmxext, 1,1
- movq mm0, [r0 - FDEC_STRIDE]
- pxor mm1, mm1
- pxor mm2, mm2
- punpckhbw mm1, mm0
- punpcklbw mm0, mm2
- psadbw mm1, mm2 ; s1
- psadbw mm0, mm2 ; s0
-%ifdef ARCH_X86_64
- movd mm4, r1d
- movd mm5, r2d
- paddw mm0, mm4
- pshufw mm2, mm5, 0
+%macro PREDICT_8x8C_DC 1
+cglobal predict_8x8c_dc_%1, 1,3
+ pxor m7, m7
+%ifdef HIGH_BIT_DEPTH
+ movq m0, [r0-FDEC_STRIDEB+0]
+ movq m1, [r0-FDEC_STRIDEB+8]
+ HADDW m0, m2
+ HADDW m1, m2
%else
- paddw mm0, r1m
- pshufw mm2, r2m, 0
+ movd m0, [r0-FDEC_STRIDEB+0]
+ movd m1, [r0-FDEC_STRIDEB+4]
+ psadbw m0, m7 ; s0
+ psadbw m1, m7 ; s1
%endif
- psrlw mm0, 3
- paddw mm1, [pw_2]
- movq mm3, mm2
- pshufw mm1, mm1, 0
- pshufw mm0, mm0, 0 ; dc0 (w)
- paddw mm3, mm1
- psrlw mm3, 3 ; dc3 (w)
- psrlw mm2, 2 ; dc2 (w)
- psrlw mm1, 2 ; dc1 (w)
+ add r0, FDEC_STRIDEB*4
+
+ movzx r1d, pixel [r0-FDEC_STRIDEB*4-SIZEOF_PIXEL]
+ movzx r2d, pixel [r0-FDEC_STRIDEB*3-SIZEOF_PIXEL]
+ add r1d, r2d
+ movzx r2d, pixel [r0-FDEC_STRIDEB*2-SIZEOF_PIXEL]
+ add r1d, r2d
+ movzx r2d, pixel [r0-FDEC_STRIDEB*1-SIZEOF_PIXEL]
+ add r1d, r2d
+ movd m2, r1d ; s2
+
+ movzx r1d, pixel [r0+FDEC_STRIDEB*0-SIZEOF_PIXEL]
+ movzx r2d, pixel [r0+FDEC_STRIDEB*1-SIZEOF_PIXEL]
+ add r1d, r2d
+ movzx r2d, pixel [r0+FDEC_STRIDEB*2-SIZEOF_PIXEL]
+ add r1d, r2d
+ movzx r2d, pixel [r0+FDEC_STRIDEB*3-SIZEOF_PIXEL]
+ add r1d, r2d
+ movd m3, r1d ; s3
+
+ punpcklwd m0, m1
+ punpcklwd m2, m3
+ punpckldq m0, m2 ; s0, s1, s2, s3
+ pshufw m3, m0, 11110110b ; s2, s1, s3, s3
+ pshufw m0, m0, 01110100b ; s0, s1, s3, s1
+ paddw m0, m3
+ psrlw m0, 2
+ pavgw m0, m7 ; s0+s2, s1, s3, s1+s3
+%ifdef HIGH_BIT_DEPTH
+%ifidn %1, sse2
+ movq2dq xmm0, m0
+ punpcklwd xmm0, xmm0
+ pshufd xmm1, xmm0, 11111010b
+ punpckldq xmm0, xmm0
+%assign n 0
+%rep 8
+%assign i (0 + (n/4))
+ movdqa [r0+FDEC_STRIDEB*(n-4)+0], xmm %+ i
+%assign n n+1
+%endrep
+%else
+ pshufw m1, m0, 0x00
+ pshufw m2, m0, 0x55
+ pshufw m3, m0, 0xaa
+ pshufw m4, m0, 0xff
+%assign n 0
+%rep 8
+%assign i (1 + (n/4)*2)
+%assign j (2 + (n/4)*2)
+ movq [r0+FDEC_STRIDEB*(n-4)+0], m %+ i
+ movq [r0+FDEC_STRIDEB*(n-4)+8], m %+ j
+%assign n n+1
+%endrep
+%endif
+%else
+ packuswb m0, m0
+ punpcklbw m0, m0
+ movq m1, m0
+ punpcklbw m0, m0
+ punpckhbw m1, m1
+%assign n 0
+%rep 8
+%assign i (0 + (n/4))
+ movq [r0+FDEC_STRIDEB*(n-4)], m %+ i
+%assign n n+1
+%endrep
+%endif
+ RET
+%endmacro
- packuswb mm0, mm1 ; dc0,dc1 (b)
- packuswb mm2, mm3 ; dc2,dc3 (b)
+INIT_MMX
+PREDICT_8x8C_DC mmxext
+%ifdef HIGH_BIT_DEPTH
+PREDICT_8x8C_DC sse2
+%endif
+
+%ifdef HIGH_BIT_DEPTH
- STORE8x8 mm0, mm2
+INIT_XMM
+cglobal predict_8x8c_dc_top_sse2, 1,1
+ pxor m2, m2
+ mova m0, [r0 - FDEC_STRIDEB]
+ pmaddwd m0, [pw_1]
+ pshufd m1, m0, 0x31
+ paddd m0, m1
+ psrlw m0, 1
+ pavgw m0, m2
+ pshuflw m0, m0, 0
+ pshufhw m0, m0, 0
+ STORE8x8 m0, m0
RET
+%else
+
cglobal predict_8x8c_dc_top_mmxext, 1,1
movq mm0, [r0 - FDEC_STRIDE]
pxor mm1, mm1
STORE8x8 mm0, mm0
RET
-;-----------------------------------------------------------------------------
-; void predict_8x8c_p_core_sse2( uint8_t *src, int i00, int b, int c )
-;-----------------------------------------------------------------------------
-
-cglobal predict_8x8c_p_core_sse2, 1,1
- movd xmm0, r1m
- movd xmm2, r2m
- movd xmm4, r3m
- pshuflw xmm0, xmm0, 0
- pshuflw xmm2, xmm2, 0
- pshuflw xmm4, xmm4, 0
- punpcklqdq xmm0, xmm0
- punpcklqdq xmm2, xmm2
- punpcklqdq xmm4, xmm4
- pmullw xmm2, [pw_76543210]
- paddsw xmm0, xmm2 ; xmm0 = {i+0*b, i+1*b, i+2*b, i+3*b, i+4*b, i+5*b, i+6*b, i+7*b}
- movdqa xmm3, xmm0
- paddsw xmm3, xmm4
- paddsw xmm4, xmm4
-call .loop
- add r0, FDEC_STRIDE*4
-.loop:
- movdqa xmm5, xmm0
- movdqa xmm1, xmm3
- psraw xmm0, 5
- psraw xmm3, 5
- packuswb xmm0, xmm3
- movq [r0+FDEC_STRIDE*0], xmm0
- movhps [r0+FDEC_STRIDE*1], xmm0
- paddsw xmm5, xmm4
- paddsw xmm1, xmm4
- movdqa xmm0, xmm5
- movdqa xmm3, xmm1
- psraw xmm5, 5
- psraw xmm1, 5
- packuswb xmm5, xmm1
- movq [r0+FDEC_STRIDE*2], xmm5
- movhps [r0+FDEC_STRIDE*3], xmm5
- paddsw xmm0, xmm4
- paddsw xmm3, xmm4
- RET
+%endif
;-----------------------------------------------------------------------------
-; void predict_16x16_p_core_sse2( uint8_t *src, int i00, int b, int c )
+; void predict_16x16_v( pixel *src )
;-----------------------------------------------------------------------------
-cglobal predict_16x16_p_core_sse2, 1,2,8
- movd xmm0, r1m
- movd xmm1, r2m
- movd xmm2, r3m
- pshuflw xmm0, xmm0, 0
- pshuflw xmm1, xmm1, 0
- pshuflw xmm2, xmm2, 0
- punpcklqdq xmm0, xmm0
- punpcklqdq xmm1, xmm1
- punpcklqdq xmm2, xmm2
- movdqa xmm3, xmm1
- pmullw xmm3, [pw_76543210]
- psllw xmm1, 3
- paddsw xmm0, xmm3 ; xmm0 = {i+ 0*b, i+ 1*b, i+ 2*b, i+ 3*b, i+ 4*b, i+ 5*b, i+ 6*b, i+ 7*b}
- paddsw xmm1, xmm0 ; xmm1 = {i+ 8*b, i+ 9*b, i+10*b, i+11*b, i+12*b, i+13*b, i+14*b, i+15*b}
- movdqa xmm7, xmm2
- paddsw xmm7, xmm7
- mov r1d, 8
-ALIGN 4
-.loop:
- movdqa xmm3, xmm0
- movdqa xmm4, xmm1
- movdqa xmm5, xmm0
- movdqa xmm6, xmm1
- psraw xmm3, 5
- psraw xmm4, 5
- paddsw xmm5, xmm2
- paddsw xmm6, xmm2
- psraw xmm5, 5
- psraw xmm6, 5
- packuswb xmm3, xmm4
- packuswb xmm5, xmm6
- movdqa [r0+FDEC_STRIDE*0], xmm3
- movdqa [r0+FDEC_STRIDE*1], xmm5
- paddsw xmm0, xmm7
- paddsw xmm1, xmm7
- add r0, FDEC_STRIDE*2
- dec r1d
- jg .loop
+%ifdef HIGH_BIT_DEPTH
+INIT_MMX
+cglobal predict_16x16_v_mmx, 1,2
+ mova m0, [r0 - FDEC_STRIDEB+ 0]
+ mova m1, [r0 - FDEC_STRIDEB+ 8]
+ mova m2, [r0 - FDEC_STRIDEB+16]
+ mova m3, [r0 - FDEC_STRIDEB+24]
+ STORE16x16 m0, m1, m2, m3
REP_RET
-
-;-----------------------------------------------------------------------------
-; void predict_16x16_v_mmx( uint8_t *src )
-;-----------------------------------------------------------------------------
+INIT_XMM
+cglobal predict_16x16_v_sse2, 2,2
+ mova m0, [r0 - FDEC_STRIDEB+ 0]
+ mova m1, [r0 - FDEC_STRIDEB+16]
+ STORE16x16_SSE2 m0, m1
+ REP_RET
+%else
+INIT_MMX
cglobal predict_16x16_v_mmx, 1,2
- movq mm0, [r0 - FDEC_STRIDE]
- movq mm1, [r0 - FDEC_STRIDE + 8]
- STORE16x16 mm0, mm1
+ movq m0, [r0 - FDEC_STRIDE + 0]
+ movq m1, [r0 - FDEC_STRIDE + 8]
+ STORE16x16 m0, m1
REP_RET
-
-;-----------------------------------------------------------------------------
-; void predict_16x16_v_sse2( uint8_t *src )
-;-----------------------------------------------------------------------------
+INIT_XMM
cglobal predict_16x16_v_sse2, 1,1
movdqa xmm0, [r0 - FDEC_STRIDE]
STORE16x16_SSE2 xmm0
RET
+%endif
;-----------------------------------------------------------------------------
-; void predict_16x16_h_mmxext( uint8_t *src )
+; void predict_16x16_h( pixel *src )
;-----------------------------------------------------------------------------
-
-%macro PRED_16x16_H 1
+%macro PREDICT_16x16_H 1
cglobal predict_16x16_h_%1, 1,2
- mov r1, FDEC_STRIDE*12
+ mov r1, 12*FDEC_STRIDEB
+%ifdef HIGH_BIT_DEPTH
+.vloop:
+%assign n 0
+%rep 4
+ movd m0, [r0+r1+n*FDEC_STRIDEB-2*SIZEOF_PIXEL]
+ SPLATW m0, m0, 1
+ mova [r0+r1+n*FDEC_STRIDEB+ 0], m0
+ mova [r0+r1+n*FDEC_STRIDEB+16], m0
+%if mmsize==8
+ mova [r0+r1+n*FDEC_STRIDEB+ 8], m0
+ mova [r0+r1+n*FDEC_STRIDEB+24], m0
+%endif
+%assign n n+1
+%endrep
+
+%else
%ifidn %1, ssse3
mova m1, [pb_3]
%endif
%endif
%assign n n+1
%endrep
- add r1, -FDEC_STRIDE*4
+%endif ; HIGH_BIT_DEPTH
+ sub r1, 4*FDEC_STRIDEB
jge .vloop
REP_RET
%endmacro
-;no SSE2, its slower than MMX on all systems that don't support SSSE3
INIT_MMX
%define SPLATB SPLATB_MMX
-PRED_16x16_H mmxext
+PREDICT_16x16_H mmxext
INIT_XMM
+%ifdef HIGH_BIT_DEPTH
+PREDICT_16x16_H sse2
+%else
+;no SSE2 for 8-bit, it's slower than MMX on all systems that don't support SSSE3
%define SPLATB SPLATB_SSSE3
-PRED_16x16_H ssse3
+PREDICT_16x16_H ssse3
+%endif
;-----------------------------------------------------------------------------
-; void predict_16x16_dc_core_mmxext( uint8_t *src, int i_dc_left )
+; void predict_16x16_dc_core( pixel *src, int i_dc_left )
;-----------------------------------------------------------------------------
%macro PRED16x16_DC 2
- pxor mm0, mm0
- pxor mm1, mm1
- psadbw mm0, [r0 - FDEC_STRIDE]
- psadbw mm1, [r0 - FDEC_STRIDE + 8]
- paddusw mm0, mm1
- paddusw mm0, %1
- psrlw mm0, %2 ; dc
- pshufw mm0, mm0, 0
- packuswb mm0, mm0 ; dc in bytes
- STORE16x16 mm0, mm0
+%ifdef HIGH_BIT_DEPTH
+ mova m0, [r0 - FDEC_STRIDEB+ 0]
+ paddw m0, [r0 - FDEC_STRIDEB+ 8]
+ paddw m0, [r0 - FDEC_STRIDEB+16]
+ paddw m0, [r0 - FDEC_STRIDEB+24]
+ HADDW m0, m1
+ paddw m0, %1
+ psrlw m0, %2
+ SPLATW m0, m0
+ STORE16x16 m0, m0, m0, m0
+%else
+ pxor m0, m0
+ pxor m1, m1
+ psadbw m0, [r0 - FDEC_STRIDE]
+ psadbw m1, [r0 - FDEC_STRIDE + 8]
+ paddusw m0, m1
+ paddusw m0, %1
+ psrlw m0, %2 ; dc
+ pshufw m0, m0, 0
+ packuswb m0, m0 ; dc in bytes
+ STORE16x16 m0, m0
+%endif
%endmacro
+INIT_MMX
cglobal predict_16x16_dc_core_mmxext, 1,2
%ifdef ARCH_X86_64
- movd mm2, r1d
- PRED16x16_DC mm2, 5
+ movd m6, r1d
+ PRED16x16_DC m6, 5
%else
PRED16x16_DC r1m, 5
%endif
REP_RET
+INIT_MMX
cglobal predict_16x16_dc_top_mmxext, 1,2
PRED16x16_DC [pw_8], 4
REP_RET
+INIT_MMX
+%ifdef HIGH_BIT_DEPTH
+cglobal predict_16x16_dc_left_core_mmxext, 1,2
+ movd m0, r1m
+ SPLATW m0, m0
+ STORE16x16 m0, m0, m0, m0
+ REP_RET
+%else
cglobal predict_16x16_dc_left_core_mmxext, 1,1
- movd mm0, r1m
- pshufw mm0, mm0, 0
- packuswb mm0, mm0
- STORE16x16 mm0, mm0
+ movd m0, r1m
+ pshufw m0, m0, 0
+ packuswb m0, m0
+ STORE16x16 m0, m0
REP_RET
+%endif
;-----------------------------------------------------------------------------
-; void predict_16x16_dc_core_sse2( uint8_t *src, int i_dc_left )
+; void predict_16x16_dc_core( pixel *src, int i_dc_left )
;-----------------------------------------------------------------------------
%macro PRED16x16_DC_SSE2 2
- pxor xmm0, xmm0
- psadbw xmm0, [r0 - FDEC_STRIDE]
- movhlps xmm1, xmm0
- paddw xmm0, xmm1
- paddusw xmm0, %1
- psrlw xmm0, %2 ; dc
- pshuflw xmm0, xmm0, 0
- punpcklqdq xmm0, xmm0
- packuswb xmm0, xmm0 ; dc in bytes
- STORE16x16_SSE2 xmm0
+%ifdef HIGH_BIT_DEPTH
+ mova m0, [r0 - FDEC_STRIDEB+ 0]
+ paddw m0, [r0 - FDEC_STRIDEB+16]
+ HADDW m0, m2
+ paddw m0, %1
+ psrlw m0, %2
+ SPLATW m0, m0
+ STORE16x16_SSE2 m0, m0
+%else
+ pxor m0, m0
+ psadbw m0, [r0 - FDEC_STRIDE]
+ movhlps m1, m0
+ paddw m0, m1
+ paddusw m0, %1
+ psrlw m0, %2 ; dc
+ SPLATW m0, m0
+ packuswb m0, m0 ; dc in bytes
+ STORE16x16_SSE2 m0
+%endif
%endmacro
-cglobal predict_16x16_dc_core_sse2, 1,1
- movd xmm2, r1m
- PRED16x16_DC_SSE2 xmm2, 5
- RET
+INIT_XMM
+cglobal predict_16x16_dc_core_sse2, 2,2,4
+ movd m3, r1m
+ PRED16x16_DC_SSE2 m3, 5
+ REP_RET
-cglobal predict_16x16_dc_top_sse2, 1,1
+cglobal predict_16x16_dc_top_sse2, 1,2
PRED16x16_DC_SSE2 [pw_8], 4
- RET
+ REP_RET
+INIT_XMM
+%ifdef HIGH_BIT_DEPTH
+cglobal predict_16x16_dc_left_core_sse2, 1,2
+ movd m0, r1m
+ SPLATW m0, m0
+ STORE16x16_SSE2 m0, m0
+ REP_RET
+%else
cglobal predict_16x16_dc_left_core_sse2, 1,1
- movd xmm0, r1m
- pshuflw xmm0, xmm0, 0
- punpcklqdq xmm0, xmm0
- packuswb xmm0, xmm0
- STORE16x16_SSE2 xmm0
+ movd m0, r1m
+ SPLATW m0, m0
+ packuswb m0, m0
+ STORE16x16_SSE2 m0
RET
+%endif
projects / x264 / blobdiff