*op0 = signed_char_clamp(p0 + filter2) ^ 0x80;
}
-void vp9_loop_filter_simple_horizontal_edge_c(uint8_t *s, int p,
- const uint8_t *blimit) {
- int i = 0;
-
- do {
- const int8_t mask = simple_filter_mask(blimit[0], s[-2 * p], s[-1 * p],
- s[0 * p], s[1 * p]);
- simple_filter(mask, s - 2 * p, s - 1 * p, s, s + 1 * p);
- ++s;
- } while (++i < 16);
-}
-
-void vp9_loop_filter_simple_vertical_edge_c(uint8_t *s, int p,
- const uint8_t *blimit) {
- int i = 0;
-
- do {
- const int8_t mask = simple_filter_mask(blimit[0], s[-2], s[-1], s[0], s[1]);
- simple_filter(mask, s - 2, s - 1, s, s + 1);
- s += p;
- } while (++i < 16);
-}
-
/* Vertical MB Filtering */
void vp9_loop_filter_mbv_c(uint8_t *y_ptr, uint8_t *u_ptr,
uint8_t *v_ptr, int y_stride, int uv_stride,
lfi->blim, lfi->lim, lfi->hev_thr, 1);
}
-void vp9_loop_filter_bhs_c(uint8_t *y, int y_stride, const uint8_t *blimit) {
- vp9_loop_filter_simple_horizontal_edge_c(y + 4 * y_stride, y_stride, blimit);
- vp9_loop_filter_simple_horizontal_edge_c(y + 8 * y_stride, y_stride, blimit);
- vp9_loop_filter_simple_horizontal_edge_c(y + 12 * y_stride, y_stride, blimit);
-}
void vp9_loop_filter_bv8x8_c(uint8_t *y, uint8_t *u, uint8_t *v,
int y_stride, int uv_stride,
lfi->blim, lfi->lim, lfi->hev_thr, 1);
}
-void vp9_loop_filter_bvs_c(uint8_t *y, int y_stride, const uint8_t *blimit) {
- vp9_loop_filter_simple_vertical_edge_c(y + 4, y_stride, blimit);
- vp9_loop_filter_simple_vertical_edge_c(y + 8, y_stride, blimit);
- vp9_loop_filter_simple_vertical_edge_c(y + 12, y_stride, blimit);
-}
-
static INLINE void wide_mbfilter(int8_t mask, uint8_t hev,
uint8_t flat, uint8_t flat2,
uint8_t *op7, uint8_t *op6, uint8_t *op5,
prototype void vp9_loop_filter_bh8x8 "uint8_t *y, uint8_t *u, uint8_t *v, int ystride, int uv_stride, struct loop_filter_info *lfi"
specialize vp9_loop_filter_bh8x8 sse2
-prototype void vp9_loop_filter_simple_mbv "uint8_t *y, int ystride, const uint8_t *blimit"
-specialize vp9_loop_filter_simple_mbv mmx sse2
-vp9_loop_filter_simple_mbv_c=vp9_loop_filter_simple_vertical_edge_c
-vp9_loop_filter_simple_mbv_mmx=vp9_loop_filter_simple_vertical_edge_mmx
-vp9_loop_filter_simple_mbv_sse2=vp9_loop_filter_simple_vertical_edge_sse2
-
-prototype void vp9_loop_filter_simple_mbh "uint8_t *y, int ystride, const uint8_t *blimit"
-specialize vp9_loop_filter_simple_mbh mmx sse2
-vp9_loop_filter_simple_mbh_c=vp9_loop_filter_simple_horizontal_edge_c
-vp9_loop_filter_simple_mbh_mmx=vp9_loop_filter_simple_horizontal_edge_mmx
-vp9_loop_filter_simple_mbh_sse2=vp9_loop_filter_simple_horizontal_edge_sse2
-
-prototype void vp9_loop_filter_simple_bv "uint8_t *y, int ystride, const uint8_t *blimit"
-specialize vp9_loop_filter_simple_bv mmx sse2
-vp9_loop_filter_simple_bv_c=vp9_loop_filter_bvs_c
-vp9_loop_filter_simple_bv_mmx=vp9_loop_filter_bvs_mmx
-vp9_loop_filter_simple_bv_sse2=vp9_loop_filter_bvs_sse2
-
-prototype void vp9_loop_filter_simple_bh "uint8_t *y, int ystride, const uint8_t *blimit"
-specialize vp9_loop_filter_simple_bh mmx sse2
-vp9_loop_filter_simple_bh_c=vp9_loop_filter_bhs_c
-vp9_loop_filter_simple_bh_mmx=vp9_loop_filter_bhs_mmx
-vp9_loop_filter_simple_bh_sse2=vp9_loop_filter_bhs_sse2
-
prototype void vp9_lpf_mbh_w "unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, int y_stride, int uv_stride, struct loop_filter_info *lfi"
specialize vp9_lpf_mbh_w sse2
}
-void vp9_loop_filter_bhs_mmx(unsigned char *y_ptr, int y_stride,
- const unsigned char *blimit) {
- vp9_loop_filter_simple_horizontal_edge_mmx(y_ptr + 4 * y_stride,
- y_stride, blimit);
- vp9_loop_filter_simple_horizontal_edge_mmx(y_ptr + 8 * y_stride,
- y_stride, blimit);
- vp9_loop_filter_simple_horizontal_edge_mmx(y_ptr + 12 * y_stride,
- y_stride, blimit);
-}
-
/* Vertical B Filtering */
void vp9_loop_filter_bv_mmx(unsigned char *y_ptr,
unsigned char *u_ptr, unsigned char *v_ptr,
lfi->blim, lfi->lim, lfi->hev_thr, 1);
}
-void vp9_loop_filter_bvs_mmx(unsigned char *y_ptr, int y_stride,
- const unsigned char *blimit) {
- vp9_loop_filter_simple_vertical_edge_mmx(y_ptr + 4, y_stride, blimit);
- vp9_loop_filter_simple_vertical_edge_mmx(y_ptr + 8, y_stride, blimit);
- vp9_loop_filter_simple_vertical_edge_mmx(y_ptr + 12, y_stride, blimit);
-}
v_ptr + 4 * uv_stride);
}
-void vp9_loop_filter_bhs_sse2(unsigned char *y_ptr, int y_stride,
- const unsigned char *blimit) {
- vp9_loop_filter_simple_horizontal_edge_sse2(y_ptr + 4 * y_stride,
- y_stride, blimit);
- vp9_loop_filter_simple_horizontal_edge_sse2(y_ptr + 8 * y_stride,
- y_stride, blimit);
- vp9_loop_filter_simple_horizontal_edge_sse2(y_ptr + 12 * y_stride,
- y_stride, blimit);
-}
-
/* Vertical B Filtering */
void vp9_loop_filter_bv_sse2(unsigned char *y_ptr,
unsigned char *u_ptr, unsigned char *v_ptr,
v_ptr + 4);
}
-void vp9_loop_filter_bvs_sse2(unsigned char *y_ptr, int y_stride,
- const unsigned char *blimit) {
- vp9_loop_filter_simple_vertical_edge_sse2(y_ptr + 4, y_stride, blimit);
- vp9_loop_filter_simple_vertical_edge_sse2(y_ptr + 8, y_stride, blimit);
- vp9_loop_filter_simple_vertical_edge_sse2(y_ptr + 12, y_stride, blimit);
-}
pop rbp
ret
-
-;void vp9_loop_filter_simple_horizontal_edge_mmx
-;(
-; unsigned char *src_ptr,
-; int src_pixel_step,
-; const char *blimit
-;)
-global sym(vp9_loop_filter_simple_horizontal_edge_mmx) PRIVATE
-sym(vp9_loop_filter_simple_horizontal_edge_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 3
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- mov rsi, arg(0) ;src_ptr
- movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch?
-
- mov rcx, 2 ; count
-.nexts8_h:
- mov rdx, arg(2) ;blimit ; get blimit
- movq mm3, [rdx] ;
-
- mov rdi, rsi ; rdi points to row +1 for indirect addressing
- add rdi, rax
- neg rax
-
- ; calculate mask
- movq mm1, [rsi+2*rax] ; p1
- movq mm0, [rdi] ; q1
- movq mm2, mm1
- movq mm7, mm0
- movq mm4, mm0
- psubusb mm0, mm1 ; q1-=p1
- psubusb mm1, mm4 ; p1-=q1
- por mm1, mm0 ; abs(p1-q1)
- pand mm1, [GLOBAL(tfe)] ; set lsb of each byte to zero
- psrlw mm1, 1 ; abs(p1-q1)/2
-
- movq mm5, [rsi+rax] ; p0
- movq mm4, [rsi] ; q0
- movq mm0, mm4 ; q0
- movq mm6, mm5 ; p0
- psubusb mm5, mm4 ; p0-=q0
- psubusb mm4, mm6 ; q0-=p0
- por mm5, mm4 ; abs(p0 - q0)
- paddusb mm5, mm5 ; abs(p0-q0)*2
- paddusb mm5, mm1 ; abs (p0 - q0) *2 + abs(p1-q1)/2
-
- psubusb mm5, mm3 ; abs(p0 - q0) *2 + abs(p1-q1)/2 > blimit
- pxor mm3, mm3
- pcmpeqb mm5, mm3
-
- ; start work on filters
- pxor mm2, [GLOBAL(t80)] ; p1 offset to convert to signed values
- pxor mm7, [GLOBAL(t80)] ; q1 offset to convert to signed values
- psubsb mm2, mm7 ; p1 - q1
-
- pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values
- pxor mm0, [GLOBAL(t80)] ; offset to convert to signed values
- movq mm3, mm0 ; q0
- psubsb mm0, mm6 ; q0 - p0
- paddsb mm2, mm0 ; p1 - q1 + 1 * (q0 - p0)
- paddsb mm2, mm0 ; p1 - q1 + 2 * (q0 - p0)
- paddsb mm2, mm0 ; p1 - q1 + 3 * (q0 - p0)
- pand mm5, mm2 ; mask filter values we don't care about
-
- ; do + 4 side
- paddsb mm5, [GLOBAL(t4)] ; 3* (q0 - p0) + (p1 - q1) + 4
-
- movq mm0, mm5 ; get a copy of filters
- psllw mm0, 8 ; shift left 8
- psraw mm0, 3 ; arithmetic shift right 11
- psrlw mm0, 8
- movq mm1, mm5 ; get a copy of filters
- psraw mm1, 11 ; arithmetic shift right 11
- psllw mm1, 8 ; shift left 8 to put it back
-
- por mm0, mm1 ; put the two together to get result
-
- psubsb mm3, mm0 ; q0-= q0 add
- pxor mm3, [GLOBAL(t80)] ; unoffset
- movq [rsi], mm3 ; write back
-
-
- ; now do +3 side
- psubsb mm5, [GLOBAL(t1s)] ; +3 instead of +4
-
- movq mm0, mm5 ; get a copy of filters
- psllw mm0, 8 ; shift left 8
- psraw mm0, 3 ; arithmetic shift right 11
- psrlw mm0, 8
- psraw mm5, 11 ; arithmetic shift right 11
- psllw mm5, 8 ; shift left 8 to put it back
- por mm0, mm5 ; put the two together to get result
-
-
- paddsb mm6, mm0 ; p0+= p0 add
- pxor mm6, [GLOBAL(t80)] ; unoffset
- movq [rsi+rax], mm6 ; write back
-
- add rsi,8
- neg rax
- dec rcx
- jnz .nexts8_h
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
-;void vp9_loop_filter_simple_vertical_edge_mmx
-;(
-; unsigned char *src_ptr,
-; int src_pixel_step,
-; const char *blimit
-;)
-global sym(vp9_loop_filter_simple_vertical_edge_mmx) PRIVATE
-sym(vp9_loop_filter_simple_vertical_edge_mmx):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 3
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 32 ; reserve 32 bytes
- %define t0 [rsp + 0] ;__declspec(align(16)) char t0[8];
- %define t1 [rsp + 16] ;__declspec(align(16)) char t1[8];
-
- mov rsi, arg(0) ;src_ptr
- movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch?
-
- lea rsi, [rsi + rax*4- 2]; ;
- mov rcx, 2 ; count
-.nexts8_v:
-
- lea rdi, [rsi + rax];
- movd mm0, [rdi + rax * 2] ; xx xx xx xx 73 72 71 70
-
- movd mm6, [rsi + rax * 2] ; xx xx xx xx 63 62 61 60
- punpcklbw mm6, mm0 ; 73 63 72 62 71 61 70 60
-
- movd mm0, [rsi + rax] ; xx xx xx xx 53 52 51 50
- movd mm4, [rsi] ; xx xx xx xx 43 42 41 40
-
- punpcklbw mm4, mm0 ; 53 43 52 42 51 41 50 40
- movq mm5, mm4 ; 53 43 52 42 51 41 50 40
-
- punpcklwd mm4, mm6 ; 71 61 51 41 70 60 50 40
- punpckhwd mm5, mm6 ; 73 63 53 43 72 62 52 42
-
- neg rax
-
- movd mm7, [rsi + rax] ; xx xx xx xx 33 32 31 30
- movd mm6, [rsi + rax * 2] ; xx xx xx xx 23 22 21 20
-
- punpcklbw mm6, mm7 ; 33 23 32 22 31 21 30 20
- movd mm1, [rdi + rax * 4] ; xx xx xx xx 13 12 11 10
-
- movd mm0, [rsi + rax * 4] ; xx xx xx xx 03 02 01 00
- punpcklbw mm0, mm1 ; 13 03 12 02 11 01 10 00
-
- movq mm2, mm0 ; 13 03 12 02 11 01 10 00
- punpcklwd mm0, mm6 ; 31 21 11 01 30 20 10 00
-
- punpckhwd mm2, mm6 ; 33 23 13 03 32 22 12 02
- movq mm1, mm0 ; 13 03 12 02 11 01 10 00
-
- punpckldq mm0, mm4 ; 70 60 50 40 30 20 10 00 = p1
- movq mm3, mm2 ; 33 23 13 03 32 22 12 02
-
- punpckhdq mm1, mm4 ; 71 61 51 41 31 21 11 01 = p0
- punpckldq mm2, mm5 ; 72 62 52 42 32 22 12 02 = q0
-
- punpckhdq mm3, mm5 ; 73 63 53 43 33 23 13 03 = q1
-
-
- ; calculate mask
- movq mm6, mm0 ; p1
- movq mm7, mm3 ; q1
- psubusb mm7, mm6 ; q1-=p1
- psubusb mm6, mm3 ; p1-=q1
- por mm6, mm7 ; abs(p1-q1)
- pand mm6, [GLOBAL(tfe)] ; set lsb of each byte to zero
- psrlw mm6, 1 ; abs(p1-q1)/2
-
- movq mm5, mm1 ; p0
- movq mm4, mm2 ; q0
-
- psubusb mm5, mm2 ; p0-=q0
- psubusb mm4, mm1 ; q0-=p0
-
- por mm5, mm4 ; abs(p0 - q0)
- paddusb mm5, mm5 ; abs(p0-q0)*2
- paddusb mm5, mm6 ; abs (p0 - q0) *2 + abs(p1-q1)/2
-
- mov rdx, arg(2) ;blimit ; get blimit
- movq mm7, [rdx]
-
- psubusb mm5, mm7 ; abs(p0 - q0) *2 + abs(p1-q1)/2 > blimit
- pxor mm7, mm7
- pcmpeqb mm5, mm7 ; mm5 = mask
-
- ; start work on filters
- movq t0, mm0
- movq t1, mm3
-
- pxor mm0, [GLOBAL(t80)] ; p1 offset to convert to signed values
- pxor mm3, [GLOBAL(t80)] ; q1 offset to convert to signed values
-
- psubsb mm0, mm3 ; p1 - q1
- movq mm6, mm1 ; p0
-
- movq mm7, mm2 ; q0
- pxor mm6, [GLOBAL(t80)] ; offset to convert to signed values
-
- pxor mm7, [GLOBAL(t80)] ; offset to convert to signed values
- movq mm3, mm7 ; offseted ; q0
-
- psubsb mm7, mm6 ; q0 - p0
- paddsb mm0, mm7 ; p1 - q1 + 1 * (q0 - p0)
-
- paddsb mm0, mm7 ; p1 - q1 + 2 * (q0 - p0)
- paddsb mm0, mm7 ; p1 - q1 + 3 * (q0 - p0)
-
- pand mm5, mm0 ; mask filter values we don't care about
-
- paddsb mm5, [GLOBAL(t4)] ; 3* (q0 - p0) + (p1 - q1) + 4
-
- movq mm0, mm5 ; get a copy of filters
- psllw mm0, 8 ; shift left 8
- psraw mm0, 3 ; arithmetic shift right 11
- psrlw mm0, 8
-
- movq mm7, mm5 ; get a copy of filters
- psraw mm7, 11 ; arithmetic shift right 11
- psllw mm7, 8 ; shift left 8 to put it back
-
- por mm0, mm7 ; put the two together to get result
-
- psubsb mm3, mm0 ; q0-= q0sz add
- pxor mm3, [GLOBAL(t80)] ; unoffset
-
- ; now do +3 side
- psubsb mm5, [GLOBAL(t1s)] ; +3 instead of +4
-
- movq mm0, mm5 ; get a copy of filters
- psllw mm0, 8 ; shift left 8
- psraw mm0, 3 ; arithmetic shift right 11
- psrlw mm0, 8
-
- psraw mm5, 11 ; arithmetic shift right 11
- psllw mm5, 8 ; shift left 8 to put it back
- por mm0, mm5 ; put the two together to get result
-
- paddsb mm6, mm0 ; p0+= p0 add
- pxor mm6, [GLOBAL(t80)] ; unoffset
-
-
- movq mm0, t0
- movq mm4, t1
-
- ; mm0 = 70 60 50 40 30 20 10 00
- ; mm6 = 71 61 51 41 31 21 11 01
- ; mm3 = 72 62 52 42 32 22 12 02
- ; mm4 = 73 63 53 43 33 23 13 03
- ; transpose back to write out
-
- movq mm1, mm0 ;
- punpcklbw mm0, mm6 ; 31 30 21 20 11 10 01 00
-
- punpckhbw mm1, mm6 ; 71 70 61 60 51 50 41 40
- movq mm2, mm3 ;
-
- punpcklbw mm2, mm4 ; 33 32 23 22 13 12 03 02
- movq mm5, mm1 ; 71 70 61 60 51 50 41 40
-
- punpckhbw mm3, mm4 ; 73 72 63 62 53 52 43 42
- movq mm6, mm0 ; 31 30 21 20 11 10 01 00
-
- punpcklwd mm0, mm2 ; 13 12 11 10 03 02 01 00
- punpckhwd mm6, mm2 ; 33 32 31 30 23 22 21 20
-
- movd [rsi+rax*4], mm0 ; write 03 02 01 00
- punpcklwd mm1, mm3 ; 53 52 51 50 43 42 41 40
-
- psrlq mm0, 32 ; xx xx xx xx 13 12 11 10
- punpckhwd mm5, mm3 ; 73 72 71 70 63 62 61 60
-
- movd [rdi+rax*4], mm0 ; write 13 12 11 10
- movd [rsi+rax*2], mm6 ; write 23 22 21 20
-
- psrlq mm6, 32 ; 33 32 31 30
- movd [rsi], mm1 ; write 43 42 41 40
-
- movd [rsi + rax], mm6 ; write 33 32 31 30
- neg rax
-
- movd [rsi + rax*2], mm5 ; write 63 62 61 60
- psrlq mm1, 32 ; 53 52 51 50
-
- movd [rdi], mm1 ; write out 53 52 51 50
- psrlq mm5, 32 ; 73 72 71 70
-
- movd [rdi + rax*2], mm5 ; write 73 72 71 70
-
- lea rsi, [rsi+rax*8] ; next 8
-
- dec rcx
- jnz .nexts8_v
-
- add rsp, 32
- pop rsp
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
-
-;void fast_loop_filter_vertical_edges_mmx(unsigned char *y_ptr,
-; int y_stride,
-; loop_filter_info *lfi)
-;{
-;
-;
-; vp9_loop_filter_simple_vertical_edge_mmx(y_ptr+4, y_stride, lfi->flim,lfi->lim,lfi->thr,2);
-; vp9_loop_filter_simple_vertical_edge_mmx(y_ptr+8, y_stride, lfi->flim,lfi->lim,lfi->thr,2);
-; vp9_loop_filter_simple_vertical_edge_mmx(y_ptr+12, y_stride, lfi->flim,lfi->lim,lfi->thr,2);
-;}
-
SECTION_RODATA
align 16
tfe:
pop rbp
ret
-;void vp9_loop_filter_simple_horizontal_edge_sse2
-;(
-; unsigned char *src_ptr,
-; int src_pixel_step,
-; const char *blimit,
-;)
-global sym(vp9_loop_filter_simple_horizontal_edge_sse2) PRIVATE
-sym(vp9_loop_filter_simple_horizontal_edge_sse2):
- push rbp
- mov rbp, rsp
- SHADOW_ARGS_TO_STACK 3
- SAVE_XMM 7
- GET_GOT rbx
- push rsi
- push rdi
- ; end prolog
-
- mov rsi, arg(0) ;src_ptr
- movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch?
- mov rdx, arg(2) ;blimit
- movdqa xmm3, XMMWORD PTR [rdx]
-
- mov rdi, rsi ; rdi points to row +1 for indirect addressing
- add rdi, rax
- neg rax
-
- ; calculate mask
- movdqa xmm1, [rsi+2*rax] ; p1
- movdqa xmm0, [rdi] ; q1
- movdqa xmm2, xmm1
- movdqa xmm7, xmm0
- movdqa xmm4, xmm0
- psubusb xmm0, xmm1 ; q1-=p1
- psubusb xmm1, xmm4 ; p1-=q1
- por xmm1, xmm0 ; abs(p1-q1)
- pand xmm1, [GLOBAL(tfe)] ; set lsb of each byte to zero
- psrlw xmm1, 1 ; abs(p1-q1)/2
-
- movdqa xmm5, [rsi+rax] ; p0
- movdqa xmm4, [rsi] ; q0
- movdqa xmm0, xmm4 ; q0
- movdqa xmm6, xmm5 ; p0
- psubusb xmm5, xmm4 ; p0-=q0
- psubusb xmm4, xmm6 ; q0-=p0
- por xmm5, xmm4 ; abs(p0 - q0)
- paddusb xmm5, xmm5 ; abs(p0-q0)*2
- paddusb xmm5, xmm1 ; abs (p0 - q0) *2 + abs(p1-q1)/2
-
- psubusb xmm5, xmm3 ; abs(p0 - q0) *2 + abs(p1-q1)/2 > blimit
- pxor xmm3, xmm3
- pcmpeqb xmm5, xmm3
-
- ; start work on filters
- pxor xmm2, [GLOBAL(t80)] ; p1 offset to convert to signed values
- pxor xmm7, [GLOBAL(t80)] ; q1 offset to convert to signed values
- psubsb xmm2, xmm7 ; p1 - q1
-
- pxor xmm6, [GLOBAL(t80)] ; offset to convert to signed values
- pxor xmm0, [GLOBAL(t80)] ; offset to convert to signed values
- movdqa xmm3, xmm0 ; q0
- psubsb xmm0, xmm6 ; q0 - p0
- paddsb xmm2, xmm0 ; p1 - q1 + 1 * (q0 - p0)
- paddsb xmm2, xmm0 ; p1 - q1 + 2 * (q0 - p0)
- paddsb xmm2, xmm0 ; p1 - q1 + 3 * (q0 - p0)
- pand xmm5, xmm2 ; mask filter values we don't care about
-
- ; do + 4 side
- paddsb xmm5, [GLOBAL(t4)] ; 3* (q0 - p0) + (p1 - q1) + 4
-
- movdqa xmm0, xmm5 ; get a copy of filters
- psllw xmm0, 8 ; shift left 8
- psraw xmm0, 3 ; arithmetic shift right 11
- psrlw xmm0, 8
- movdqa xmm1, xmm5 ; get a copy of filters
- psraw xmm1, 11 ; arithmetic shift right 11
- psllw xmm1, 8 ; shift left 8 to put it back
-
- por xmm0, xmm1 ; put the two together to get result
-
- psubsb xmm3, xmm0 ; q0-= q0 add
- pxor xmm3, [GLOBAL(t80)] ; unoffset
- movdqa [rsi], xmm3 ; write back
-
- ; now do +3 side
- psubsb xmm5, [GLOBAL(t1s)] ; +3 instead of +4
-
- movdqa xmm0, xmm5 ; get a copy of filters
- psllw xmm0, 8 ; shift left 8
- psraw xmm0, 3 ; arithmetic shift right 11
- psrlw xmm0, 8
- psraw xmm5, 11 ; arithmetic shift right 11
- psllw xmm5, 8 ; shift left 8 to put it back
- por xmm0, xmm5 ; put the two together to get result
-
-
- paddsb xmm6, xmm0 ; p0+= p0 add
- pxor xmm6, [GLOBAL(t80)] ; unoffset
- movdqa [rsi+rax], xmm6 ; write back
-
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
-
-;void vp9_loop_filter_simple_vertical_edge_sse2
-;(
-; unsigned char *src_ptr,
-; int src_pixel_step,
-; const char *blimit,
-;)
-global sym(vp9_loop_filter_simple_vertical_edge_sse2) PRIVATE
-sym(vp9_loop_filter_simple_vertical_edge_sse2):
- push rbp ; save old base pointer value.
- mov rbp, rsp ; set new base pointer value.
- SHADOW_ARGS_TO_STACK 3
- SAVE_XMM 7
- GET_GOT rbx ; save callee-saved reg
- push rsi
- push rdi
- ; end prolog
-
- ALIGN_STACK 16, rax
- sub rsp, 32 ; reserve 32 bytes
- %define t0 [rsp + 0] ;__declspec(align(16)) char t0[16];
- %define t1 [rsp + 16] ;__declspec(align(16)) char t1[16];
-
- mov rsi, arg(0) ;src_ptr
- movsxd rax, dword ptr arg(1) ;src_pixel_step ; destination pitch?
-
- lea rsi, [rsi - 2 ]
- lea rdi, [rsi + rax]
- lea rdx, [rsi + rax*4]
- lea rcx, [rdx + rax]
-
- movd xmm0, [rsi] ; (high 96 bits unused) 03 02 01 00
- movd xmm1, [rdx] ; (high 96 bits unused) 43 42 41 40
- movd xmm2, [rdi] ; 13 12 11 10
- movd xmm3, [rcx] ; 53 52 51 50
- punpckldq xmm0, xmm1 ; (high 64 bits unused) 43 42 41 40 03 02 01 00
- punpckldq xmm2, xmm3 ; 53 52 51 50 13 12 11 10
-
- movd xmm4, [rsi + rax*2] ; 23 22 21 20
- movd xmm5, [rdx + rax*2] ; 63 62 61 60
- movd xmm6, [rdi + rax*2] ; 33 32 31 30
- movd xmm7, [rcx + rax*2] ; 73 72 71 70
- punpckldq xmm4, xmm5 ; 63 62 61 60 23 22 21 20
- punpckldq xmm6, xmm7 ; 73 72 71 70 33 32 31 30
-
- punpcklbw xmm0, xmm2 ; 53 43 52 42 51 41 50 40 13 03 12 02 11 01 10 00
- punpcklbw xmm4, xmm6 ; 73 63 72 62 71 61 70 60 33 23 32 22 31 21 30 20
-
- movdqa xmm1, xmm0
- punpcklwd xmm0, xmm4 ; 33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00
- punpckhwd xmm1, xmm4 ; 73 63 53 43 72 62 52 42 71 61 51 41 70 60 50 40
-
- movdqa xmm2, xmm0
- punpckldq xmm0, xmm1 ; 71 61 51 41 31 21 11 01 70 60 50 40 30 20 10 00
- punpckhdq xmm2, xmm1 ; 73 63 53 43 33 23 13 03 72 62 52 42 32 22 12 02
-
- movdqa t0, xmm0 ; save to t0
- movdqa t1, xmm2 ; save to t1
-
- lea rsi, [rsi + rax*8]
- lea rdi, [rsi + rax]
- lea rdx, [rsi + rax*4]
- lea rcx, [rdx + rax]
-
- movd xmm4, [rsi] ; 83 82 81 80
- movd xmm1, [rdx] ; c3 c2 c1 c0
- movd xmm6, [rdi] ; 93 92 91 90
- movd xmm3, [rcx] ; d3 d2 d1 d0
- punpckldq xmm4, xmm1 ; c3 c2 c1 c0 83 82 81 80
- punpckldq xmm6, xmm3 ; d3 d2 d1 d0 93 92 91 90
-
- movd xmm0, [rsi + rax*2] ; a3 a2 a1 a0
- movd xmm5, [rdx + rax*2] ; e3 e2 e1 e0
- movd xmm2, [rdi + rax*2] ; b3 b2 b1 b0
- movd xmm7, [rcx + rax*2] ; f3 f2 f1 f0
- punpckldq xmm0, xmm5 ; e3 e2 e1 e0 a3 a2 a1 a0
- punpckldq xmm2, xmm7 ; f3 f2 f1 f0 b3 b2 b1 b0
-
- punpcklbw xmm4, xmm6 ; d3 c3 d2 c2 d1 c1 d0 c0 93 83 92 82 91 81 90 80
- punpcklbw xmm0, xmm2 ; f3 e3 f2 e2 f1 e1 f0 e0 b3 a3 b2 a2 b1 a1 b0 a0
-
- movdqa xmm1, xmm4
- punpcklwd xmm4, xmm0 ; b3 a3 93 83 b2 a2 92 82 b1 a1 91 81 b0 a0 90 80
- punpckhwd xmm1, xmm0 ; f3 e3 d3 c3 f2 e2 d2 c2 f1 e1 d1 c1 f0 e0 d0 c0
-
- movdqa xmm6, xmm4
- punpckldq xmm4, xmm1 ; f1 e1 d1 c1 b1 a1 91 81 f0 e0 d0 c0 b0 a0 90 80
- punpckhdq xmm6, xmm1 ; f3 e3 d3 c3 b3 a3 93 83 f2 e2 d2 c2 b2 a2 92 82
-
- movdqa xmm0, t0 ; 71 61 51 41 31 21 11 01 70 60 50 40 30 20 10 00
- movdqa xmm2, t1 ; 73 63 53 43 33 23 13 03 72 62 52 42 32 22 12 02
- movdqa xmm1, xmm0
- movdqa xmm3, xmm2
-
- punpcklqdq xmm0, xmm4 ; p1 f0 e0 d0 c0 b0 a0 90 80 70 60 50 40 30 20 10 00
- punpckhqdq xmm1, xmm4 ; p0 f1 e1 d1 c1 b1 a1 91 81 71 61 51 41 31 21 11 01
- punpcklqdq xmm2, xmm6 ; q0 f2 e2 d2 c2 b2 a2 92 82 72 62 52 42 32 22 12 02
- punpckhqdq xmm3, xmm6 ; q1 f3 e3 d3 c3 b3 a3 93 83 73 63 53 43 33 23 13 03
-
- ; calculate mask
- movdqa xmm6, xmm0 ; p1
- movdqa xmm7, xmm3 ; q1
- psubusb xmm7, xmm0 ; q1-=p1
- psubusb xmm6, xmm3 ; p1-=q1
- por xmm6, xmm7 ; abs(p1-q1)
- pand xmm6, [GLOBAL(tfe)] ; set lsb of each byte to zero
- psrlw xmm6, 1 ; abs(p1-q1)/2
-
- movdqa xmm5, xmm1 ; p0
- movdqa xmm4, xmm2 ; q0
- psubusb xmm5, xmm2 ; p0-=q0
- psubusb xmm4, xmm1 ; q0-=p0
- por xmm5, xmm4 ; abs(p0 - q0)
- paddusb xmm5, xmm5 ; abs(p0-q0)*2
- paddusb xmm5, xmm6 ; abs (p0 - q0) *2 + abs(p1-q1)/2
-
- mov rdx, arg(2) ;blimit
- movdqa xmm7, XMMWORD PTR [rdx]
-
- psubusb xmm5, xmm7 ; abs(p0 - q0) *2 + abs(p1-q1)/2 > blimit
- pxor xmm7, xmm7
- pcmpeqb xmm5, xmm7 ; mm5 = mask
-
- ; start work on filters
- movdqa t0, xmm0
- movdqa t1, xmm3
-
- pxor xmm0, [GLOBAL(t80)] ; p1 offset to convert to signed values
- pxor xmm3, [GLOBAL(t80)] ; q1 offset to convert to signed values
-
- psubsb xmm0, xmm3 ; p1 - q1
- movdqa xmm6, xmm1 ; p0
-
- movdqa xmm7, xmm2 ; q0
- pxor xmm6, [GLOBAL(t80)] ; offset to convert to signed values
-
- pxor xmm7, [GLOBAL(t80)] ; offset to convert to signed values
- movdqa xmm3, xmm7 ; offseted ; q0
-
- psubsb xmm7, xmm6 ; q0 - p0
- paddsb xmm0, xmm7 ; p1 - q1 + 1 * (q0 - p0)
-
- paddsb xmm0, xmm7 ; p1 - q1 + 2 * (q0 - p0)
- paddsb xmm0, xmm7 ; p1 - q1 + 3 * (q0 - p0)
-
- pand xmm5, xmm0 ; mask filter values we don't care about
-
-
- paddsb xmm5, [GLOBAL(t4)] ; 3* (q0 - p0) + (p1 - q1) + 4
-
- movdqa xmm0, xmm5 ; get a copy of filters
- psllw xmm0, 8 ; shift left 8
-
- psraw xmm0, 3 ; arithmetic shift right 11
- psrlw xmm0, 8
-
- movdqa xmm7, xmm5 ; get a copy of filters
- psraw xmm7, 11 ; arithmetic shift right 11
-
- psllw xmm7, 8 ; shift left 8 to put it back
- por xmm0, xmm7 ; put the two together to get result
-
- psubsb xmm3, xmm0 ; q0-= q0sz add
- pxor xmm3, [GLOBAL(t80)] ; unoffset q0
-
- ; now do +3 side
- psubsb xmm5, [GLOBAL(t1s)] ; +3 instead of +4
- movdqa xmm0, xmm5 ; get a copy of filters
-
- psllw xmm0, 8 ; shift left 8
- psraw xmm0, 3 ; arithmetic shift right 11
-
- psrlw xmm0, 8
- psraw xmm5, 11 ; arithmetic shift right 11
-
- psllw xmm5, 8 ; shift left 8 to put it back
- por xmm0, xmm5 ; put the two together to get result
-
- paddsb xmm6, xmm0 ; p0+= p0 add
- pxor xmm6, [GLOBAL(t80)] ; unoffset p0
-
- movdqa xmm0, t0 ; p1
- movdqa xmm4, t1 ; q1
-
- ; transpose back to write out
- ; p1 f0 e0 d0 c0 b0 a0 90 80 70 60 50 40 30 20 10 00
- ; p0 f1 e1 d1 c1 b1 a1 91 81 71 61 51 41 31 21 11 01
- ; q0 f2 e2 d2 c2 b2 a2 92 82 72 62 52 42 32 22 12 02
- ; q1 f3 e3 d3 c3 b3 a3 93 83 73 63 53 43 33 23 13 03
- movdqa xmm1, xmm0
- punpcklbw xmm0, xmm6 ; 71 70 61 60 51 50 41 40 31 30 21 20 11 10 01 00
- punpckhbw xmm1, xmm6 ; f1 f0 e1 e0 d1 d0 c1 c0 b1 b0 a1 a0 91 90 81 80
-
- movdqa xmm5, xmm3
- punpcklbw xmm3, xmm4 ; 73 72 63 62 53 52 43 42 33 32 23 22 13 12 03 02
- punpckhbw xmm5, xmm4 ; f3 f2 e3 e2 d3 d2 c3 c2 b3 b2 a3 a2 93 92 83 82
-
- movdqa xmm2, xmm0
- punpcklwd xmm0, xmm3 ; 33 32 31 30 23 22 21 20 13 12 11 10 03 02 01 00
- punpckhwd xmm2, xmm3 ; 73 72 71 70 63 62 61 60 53 52 51 50 43 42 41 40
-
- movdqa xmm3, xmm1
- punpcklwd xmm1, xmm5 ; b3 b2 b1 b0 a3 a2 a1 a0 93 92 91 90 83 82 81 80
- punpckhwd xmm3, xmm5 ; f3 f2 f1 f0 e3 e2 e1 e0 d3 d2 d1 d0 c3 c2 c1 c0
-
- ; write out order: xmm0 xmm2 xmm1 xmm3
- lea rdx, [rsi + rax*4]
-
- movd [rsi], xmm1 ; write the second 8-line result
- psrldq xmm1, 4
- movd [rdi], xmm1
- psrldq xmm1, 4
- movd [rsi + rax*2], xmm1
- psrldq xmm1, 4
- movd [rdi + rax*2], xmm1
-
- movd [rdx], xmm3
- psrldq xmm3, 4
- movd [rcx], xmm3
- psrldq xmm3, 4
- movd [rdx + rax*2], xmm3
- psrldq xmm3, 4
- movd [rcx + rax*2], xmm3
-
- neg rax
- lea rsi, [rsi + rax*8]
- neg rax
- lea rdi, [rsi + rax]
- lea rdx, [rsi + rax*4]
- lea rcx, [rdx + rax]
-
- movd [rsi], xmm0 ; write the first 8-line result
- psrldq xmm0, 4
- movd [rdi], xmm0
- psrldq xmm0, 4
- movd [rsi + rax*2], xmm0
- psrldq xmm0, 4
- movd [rdi + rax*2], xmm0
-
- movd [rdx], xmm2
- psrldq xmm2, 4
- movd [rcx], xmm2
- psrldq xmm2, 4
- movd [rdx + rax*2], xmm2
- psrldq xmm2, 4
- movd [rcx + rax*2], xmm2
-
- add rsp, 32
- pop rsp
- ; begin epilog
- pop rdi
- pop rsi
- RESTORE_GOT
- RESTORE_XMM
- UNSHADOW_ARGS
- pop rbp
- ret
-
SECTION_RODATA
align 16
tfe:
extern prototype_loopfilter_block(vp9_loop_filter_bv_mmx);
extern prototype_loopfilter_block(vp9_loop_filter_mbh_mmx);
extern prototype_loopfilter_block(vp9_loop_filter_bh_mmx);
-extern prototype_simple_loopfilter(vp9_loop_filter_simple_vertical_edge_mmx);
-extern prototype_simple_loopfilter(vp9_loop_filter_bvs_mmx);
-extern prototype_simple_loopfilter(vp9_loop_filter_simple_horizontal_edge_mmx);
-extern prototype_simple_loopfilter(vp9_loop_filter_bhs_mmx);
#endif
#if HAVE_SSE2
extern prototype_loopfilter_block(vp9_loop_filter_bv_sse2);
extern prototype_loopfilter_block(vp9_loop_filter_mbh_sse2);
extern prototype_loopfilter_block(vp9_loop_filter_bh_sse2);
-extern prototype_simple_loopfilter(vp9_loop_filter_simple_vertical_edge_sse2);
-extern prototype_simple_loopfilter(vp9_loop_filter_bvs_sse2);
-extern prototype_simple_loopfilter(vp9_loop_filter_simple_horizontal_edge_sse2);
-extern prototype_simple_loopfilter(vp9_loop_filter_bhs_sse2);
#endif
#endif // LOOPFILTER_X86_H
projects / platform / upstream / libvpx.git / commitdiff