;*
;* You should have received a copy of the GNU Lesser General Public
;* License along with Libav; if not, write to the Free Software
-;* 51, Inc., Foundation Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************
%include "x86inc.asm"
; out: %4 = |%1-%2|>%3
; clobbers: %5
%macro DIFF_GT2 5
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
psubusb %5, %2, %1
psubusb %4, %1, %2
%else
; out: m1=p0' m2=q0'
; clobbers: m0,3-6
%macro DEBLOCK_P0_Q0 0
- pxor m5, m1, m2 ; p0^q0
- pand m5, [pb_1] ; (p0^q0)&1
pcmpeqb m4, m4
+ pxor m5, m1, m2 ; p0^q0
pxor m3, m4
+ pand m5, [pb_1] ; (p0^q0)&1
pavgb m3, m0 ; (p1 - q1 + 256)>>1
- pavgb m3, [pb_3] ; (((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2
pxor m4, m1
+ pavgb m3, [pb_3] ; (((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2
pavgb m4, m2 ; (q0 - p0 + 256)>>1
pavgb m3, m5
- paddusb m3, m4 ; d+128+33
mova m6, [pb_A1]
+ paddusb m3, m4 ; d+128+33
psubusb m6, m3
psubusb m3, [pb_A1]
pminub m6, m7
mova %4, %2
%endmacro
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
;-----------------------------------------------------------------------------
; void deblock_v_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
;-----------------------------------------------------------------------------
; void deblock_h_luma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
;-----------------------------------------------------------------------------
INIT_MMX
-cglobal deblock_h_luma_8_%1, 5,7
- movsxd r10, r1d
- lea r11, [r10+r10*2]
+cglobal deblock_h_luma_8_%1, 5,9
+ movsxd r7, r1d
+ lea r8, [r7+r7*2]
lea r6, [r0-4]
- lea r5, [r0-4+r11]
-%ifdef WIN64
+ lea r5, [r0-4+r8]
+%if WIN64
sub rsp, 0x98
%define pix_tmp rsp+0x30
%else
%endif
; transpose 6x16 -> tmp space
- TRANSPOSE6x8_MEM PASS8ROWS(r6, r5, r10, r11), pix_tmp
- lea r6, [r6+r10*8]
- lea r5, [r5+r10*8]
- TRANSPOSE6x8_MEM PASS8ROWS(r6, r5, r10, r11), pix_tmp+8
+ TRANSPOSE6x8_MEM PASS8ROWS(r6, r5, r7, r8), pix_tmp
+ lea r6, [r6+r7*8]
+ lea r5, [r5+r7*8]
+ TRANSPOSE6x8_MEM PASS8ROWS(r6, r5, r7, r8), pix_tmp+8
; vertical filter
; alpha, beta, tc0 are still in r2d, r3d, r4
- ; don't backup r6, r5, r10, r11 because deblock_v_luma_sse2 doesn't use them
+ ; don't backup r6, r5, r7, r8 because deblock_v_luma_sse2 doesn't use them
lea r0, [pix_tmp+0x30]
mov r1d, 0x10
-%ifdef WIN64
+%if WIN64
mov [rsp+0x20], r4
%endif
call deblock_v_luma_8_%1
movq m1, [pix_tmp+0x28]
movq m2, [pix_tmp+0x38]
movq m3, [pix_tmp+0x48]
- TRANSPOSE8x4B_STORE PASS8ROWS(r6, r5, r10, r11)
+ TRANSPOSE8x4B_STORE PASS8ROWS(r6, r5, r7, r8)
- shl r10, 3
- sub r6, r10
- sub r5, r10
- shr r10, 3
+ shl r7, 3
+ sub r6, r7
+ sub r5, r7
+ shr r7, 3
movq m0, [pix_tmp+0x10]
movq m1, [pix_tmp+0x20]
movq m2, [pix_tmp+0x30]
movq m3, [pix_tmp+0x40]
- TRANSPOSE8x4B_STORE PASS8ROWS(r6, r5, r10, r11)
+ TRANSPOSE8x4B_STORE PASS8ROWS(r6, r5, r7, r8)
-%ifdef WIN64
+%if WIN64
add rsp, 0x98
%else
add rsp, 0x68
LOAD_MASK r2, r3
mov r3, r4mp
+ pcmpeqb m3, m3
movd m4, [r3] ; tc0
punpcklbw m4, m4
punpcklbw m4, m4 ; tc = 4x tc0[3], 4x tc0[2], 4x tc0[1], 4x tc0[0]
mova [esp+%3], m4 ; tc
- pcmpeqb m3, m3
pcmpgtb m4, m3
+ mova m3, [r4] ; p2
pand m4, m7
mova [esp], m4 ; mask
- mova m3, [r4] ; p2
DIFF_GT2 m1, m3, m5, m6, m7 ; |p2-p0| > beta-1
pand m6, m4
pand m4, [esp+%3] ; tc
mova m4, [r0+2*r1] ; q2
DIFF_GT2 m2, m4, m5, m6, m3 ; |q2-q0| > beta-1
- mova m5, [esp] ; mask
- pand m6, m5
+ pand m6, [esp] ; mask
mova m5, [esp+%3] ; tc
- pand m5, m6
psubb m7, m6
+ pand m5, m6
mova m3, [r0+r1]
LUMA_Q1 m3, m4, [r0+2*r1], [r0+r1], m5, m6
; transpose 16x4 -> original space (only the middle 4 rows were changed by the filter)
mov r0, r0mp
sub r0, 2
- lea r1, [r0+r4]
movq m0, [pix_tmp+0x10]
movq m1, [pix_tmp+0x20]
+ lea r1, [r0+r4]
movq m2, [pix_tmp+0x30]
movq m3, [pix_tmp+0x40]
TRANSPOSE8x4B_STORE PASS8ROWS(r0, r1, r3, r4)
%macro LUMA_INTRA_P012 4 ; p0..p3 in memory
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
pavgb t0, p2, p1
pavgb t1, p0, q0
%else
%endif
pavgb t0, t1 ; ((p2+p1+1)/2 + (p0+q0+1)/2 + 1)/2
mova t5, t1
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
paddb t2, p2, p1
paddb t3, p0, q0
%else
pand t2, mpb_1
psubb t0, t2 ; p1' = (p2+p1+p0+q0+2)/4;
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
pavgb t1, p2, q1
psubb t2, p2, q1
%else
%define t1 m5
%define t2 m6
%define t3 m7
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
%define p2 m8
%define q2 m9
%define t4 m10
; void deblock_v_luma_intra( uint8_t *pix, int stride, int alpha, int beta )
;-----------------------------------------------------------------------------
cglobal deblock_%2_luma_intra_8_%1, 4,6,16
-%ifndef ARCH_X86_64
+%if ARCH_X86_64 == 0
sub esp, 0x60
%endif
lea r4, [r1*4]
mova p0, [r4+r5]
mova q0, [r0]
mova q1, [r0+r1]
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
pxor mpb_0, mpb_0
mova mpb_1, [pb_1]
LOAD_MASK r2d, r3d, t5 ; m5=beta-1, t5=alpha-1, m7=mask0
LUMA_INTRA_SWAP_PQ
LUMA_INTRA_P012 [r0], [r0+r1], [r0+2*r1], [r0+r5]
.end:
-%ifndef ARCH_X86_64
+%if ARCH_X86_64 == 0
add esp, 0x60
%endif
RET
INIT_MMX
-%ifdef ARCH_X86_64
+%if ARCH_X86_64
;-----------------------------------------------------------------------------
; void deblock_h_luma_intra( uint8_t *pix, int stride, int alpha, int beta )
;-----------------------------------------------------------------------------
-cglobal deblock_h_luma_intra_8_%1, 4,7
- movsxd r10, r1d
- lea r11, [r10*3]
+cglobal deblock_h_luma_intra_8_%1, 4,9
+ movsxd r7, r1d
+ lea r8, [r7*3]
lea r6, [r0-4]
- lea r5, [r0-4+r11]
+ lea r5, [r0-4+r8]
sub rsp, 0x88
%define pix_tmp rsp
; transpose 8x16 -> tmp space
- TRANSPOSE8x8_MEM PASS8ROWS(r6, r5, r10, r11), PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30)
- lea r6, [r6+r10*8]
- lea r5, [r5+r10*8]
- TRANSPOSE8x8_MEM PASS8ROWS(r6, r5, r10, r11), PASS8ROWS(pix_tmp+8, pix_tmp+0x38, 0x10, 0x30)
+ TRANSPOSE8x8_MEM PASS8ROWS(r6, r5, r7, r8), PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30)
+ lea r6, [r6+r7*8]
+ lea r5, [r5+r7*8]
+ TRANSPOSE8x8_MEM PASS8ROWS(r6, r5, r7, r8), PASS8ROWS(pix_tmp+8, pix_tmp+0x38, 0x10, 0x30)
lea r0, [pix_tmp+0x40]
mov r1, 0x10
call deblock_v_luma_intra_8_%1
; transpose 16x6 -> original space (but we can't write only 6 pixels, so really 16x8)
- lea r5, [r6+r11]
- TRANSPOSE8x8_MEM PASS8ROWS(pix_tmp+8, pix_tmp+0x38, 0x10, 0x30), PASS8ROWS(r6, r5, r10, r11)
- shl r10, 3
- sub r6, r10
- sub r5, r10
- shr r10, 3
- TRANSPOSE8x8_MEM PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30), PASS8ROWS(r6, r5, r10, r11)
+ lea r5, [r6+r8]
+ TRANSPOSE8x8_MEM PASS8ROWS(pix_tmp+8, pix_tmp+0x38, 0x10, 0x30), PASS8ROWS(r6, r5, r7, r8)
+ shl r7, 3
+ sub r6, r7
+ sub r5, r7
+ shr r7, 3
+ TRANSPOSE8x8_MEM PASS8ROWS(pix_tmp, pix_tmp+0x30, 0x10, 0x30), PASS8ROWS(r6, r5, r7, r8)
add rsp, 0x88
RET
%else
DEBLOCK_LUMA_INTRA sse2, v
INIT_AVX
DEBLOCK_LUMA_INTRA avx , v
-%ifndef ARCH_X86_64
+%if ARCH_X86_64 == 0
INIT_MMX
DEBLOCK_LUMA_INTRA mmxext, v8
%endif
; void ff_deblock_h_chroma( uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0 )
;-----------------------------------------------------------------------------
cglobal deblock_h_chroma_8_mmxext, 5,7
-%ifdef ARCH_X86_64
+%if UNIX64
%define buf0 [rsp-24]
%define buf1 [rsp-16]
+%elif WIN64
+ sub rsp, 16
+ %define buf0 [rsp]
+ %define buf1 [rsp+8]
%else
%define buf0 r0m
%define buf1 r2m
movq m0, buf0
movq m3, buf1
TRANSPOSE8x4B_STORE PASS8ROWS(t5, r0, r1, t6)
+%if WIN64
+ add rsp, 16
+%endif
RET
ALIGN 16
projects / ffmpeg / blobdiff