Commit 992b5e2d authored by Scott LaVarnway's avatar Scott LaVarnway
Browse files

sse2 intrinsic version of vp8_mbloop_filter_horizontal_edge()

First sse2 version of vp8_mbloop_filter_horizontal_edge().  For now,
intrinsics are being used until the bitstream is finalized.  This function
will be revisited later for further performance improvements.
For the test clip used, a 31+% decoder performance improvement
was seen.  This will vary depending on material.

Change-Id: I03ed3a7182478bdd1f094644ff3e0442625600e7
parent edf099d2
......@@ -7,8 +7,6 @@
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <stdlib.h>
#include "vpx_config.h"
#include "loopfilter.h"
......@@ -94,6 +92,7 @@ static __inline void vp8_filter(signed char mask, uc hev, uc *op1,
*op1 = u ^ 0x80;
}
void vp8_loop_filter_horizontal_edge_c
(
unsigned char *s,
......@@ -218,6 +217,7 @@ static __inline void vp8_mbfilter(signed char mask, uc hev, uc flat,
Filter2 = vp8_signed_char_clamp(vp8_filter + 3);
Filter1 >>= 3;
Filter2 >>= 3;
u = vp8_signed_char_clamp(qs0 - Filter1);
*oq0 = u ^ 0x80;
u = vp8_signed_char_clamp(ps0 + Filter2);
......@@ -272,7 +272,6 @@ void vp8_mbloop_filter_horizontal_edge_c
}
void vp8_mbloop_filter_vertical_edge_c
(
unsigned char *s,
......
......@@ -134,13 +134,13 @@ prototype void vp8_loop_filter_bv8x8 "unsigned char *y, unsigned char *u, unsign
specialize vp8_loop_filter_bv8x8;
prototype void vp8_loop_filter_mbh "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"
specialize vp8_loop_filter_mbh;
specialize vp8_loop_filter_mbh sse2
prototype void vp8_loop_filter_bh "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"
specialize vp8_loop_filter_bh;
prototype void vp8_loop_filter_bh8x8 "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"
specialize vp8_loop_filter_bh8x8;
specialize vp8_loop_filter_bh8x8 sse2
prototype void vp8_loop_filter_simple_mbv "unsigned char *y, int ystride, const unsigned char *blimit"
specialize vp8_loop_filter_simple_mbv mmx sse2 media neon
......
......@@ -594,323 +594,6 @@ sym(vp8_loop_filter_vertical_edge_mmx):
ret
;void vp8_mbloop_filter_horizontal_edge_mmx
;(
; unsigned char *src_ptr,
; int src_pixel_step,
; const char *blimit,
; const char *limit,
; const char *thresh,
; int count
;)
global sym(vp8_mbloop_filter_horizontal_edge_mmx)
sym(vp8_mbloop_filter_horizontal_edge_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 6
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?
movsxd rcx, dword ptr arg(5) ;count
.next8_mbh:
mov rdx, arg(3) ;limit
movq mm7, [rdx]
mov rdi, rsi ; rdi points to row +1 for indirect addressing
add rdi, rax
; calculate breakout conditions
movq mm2, [rdi+2*rax] ; q3
movq mm1, [rsi+2*rax] ; q2
movq mm6, mm1 ; q2
psubusb mm1, mm2 ; q2-=q3
psubusb mm2, mm6 ; q3-=q2
por mm1, mm2 ; abs(q3-q2)
psubusb mm1, mm7
; mm1 = abs(q3-q2), mm6 =q2, mm7 = limit
movq mm4, [rsi+rax] ; q1
movq mm3, mm4 ; q1
psubusb mm4, mm6 ; q1-=q2
psubusb mm6, mm3 ; q2-=q1
por mm4, mm6 ; abs(q2-q1)
psubusb mm4, mm7
por mm1, mm4
; mm1 = mask, mm3=q1, mm7 = limit
movq mm4, [rsi] ; q0
movq mm0, mm4 ; q0
psubusb mm4, mm3 ; q0-=q1
psubusb mm3, mm0 ; q1-=q0
por mm4, mm3 ; abs(q0-q1)
movq t0, mm4 ; save to t0
psubusb mm4, mm7
por mm1, mm4
; mm1 = mask, mm0=q0, mm7 = limit, t0 = abs(q0-q1)
neg rax ; negate pitch to deal with above border
movq mm2, [rsi+4*rax] ; p3
movq mm4, [rdi+4*rax] ; p2
movq mm5, mm4 ; p2
psubusb mm4, mm2 ; p2-=p3
psubusb mm2, mm5 ; p3-=p2
por mm4, mm2 ; abs(p3 - p2)
psubusb mm4, mm7
por mm1, mm4
; mm1 = mask, mm0=q0, mm7 = limit, t0 = abs(q0-q1)
movq mm4, [rsi+2*rax] ; p1
movq mm3, mm4 ; p1
psubusb mm4, mm5 ; p1-=p2
psubusb mm5, mm3 ; p2-=p1
por mm4, mm5 ; abs(p2 - p1)
psubusb mm4, mm7
por mm1, mm4
movq mm2, mm3 ; p1
; mm1 = mask, mm0=q0, mm7 = limit, t0 = abs(q0-q1)
movq mm4, [rsi+rax] ; p0
movq mm5, mm4 ; p0
psubusb mm4, mm3 ; p0-=p1
psubusb mm3, mm5 ; p1-=p0
por mm4, mm3 ; abs(p1 - p0)
movq t1, mm4 ; save to t1
psubusb mm4, mm7
por mm1, mm4
; mm1 = mask, mm0=q0, mm7 = limit, t0 = abs(q0-q1) t1 = abs(p1-p0)
; mm5 = p0
movq mm3, [rdi] ; q1
movq mm4, mm3 ; q1
psubusb mm3, mm2 ; q1-=p1
psubusb mm2, mm4 ; p1-=q1
por mm2, mm3 ; abs(p1-q1)
pand mm2, [GLOBAL(tfe)] ; set lsb of each byte to zero
psrlw mm2, 1 ; abs(p1-q1)/2
movq mm6, mm5 ; p0
movq mm3, mm0 ; q0
psubusb mm5, mm3 ; p0-=q0
psubusb mm3, mm6 ; q0-=p0
por mm5, mm3 ; abs(p0 - q0)
paddusb mm5, mm5 ; abs(p0-q0)*2
paddusb mm5, mm2 ; abs (p0 - q0) *2 + abs(p1-q1)/2
mov rdx, arg(2) ;blimit ; get blimit
movq mm7, [rdx] ; blimit
psubusb mm5, mm7 ; abs (p0 - q0) *2 + abs(p1-q1)/2 > blimit
por mm1, mm5
pxor mm5, mm5
pcmpeqb mm1, mm5 ; mask mm1
; mm1 = mask, mm0=q0, mm7 = blimit, t0 = abs(q0-q1) t1 = abs(p1-p0)
; mm6 = p0,
; calculate high edge variance
mov rdx, arg(4) ;thresh ; get thresh
movq mm7, [rdx] ;
movq mm4, t0 ; get abs (q1 - q0)
psubusb mm4, mm7
movq mm3, t1 ; get abs (p1 - p0)
psubusb mm3, mm7
paddb mm4, mm3 ; abs(q1 - q0) > thresh || abs(p1 - p0) > thresh
pcmpeqb mm4, mm5
pcmpeqb mm5, mm5
pxor mm4, mm5
; mm1 = mask, mm0=q0, mm7 = thresh, t0 = abs(q0-q1) t1 = abs(p1-p0)
; mm6 = p0, mm4=hev
; start work on filters
movq mm2, [rsi+2*rax] ; p1
movq mm7, [rdi] ; q1
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 ; 1 * (q0 - p0) + (p1 - q1)
paddsb mm2, mm0 ; 2 * (q0 - p0)
paddsb mm2, mm0 ; 3 * (q0 - p0) + (p1 - q1)
pand mm1, mm2 ; mask filter values we don't care about
; mm1 = vp8_filter, mm4=hev, mm6=ps0, mm3=qs0
movq mm2, mm1 ; vp8_filter
pand mm2, mm4; ; Filter2 = vp8_filter & hev
movq mm5, mm2 ;
paddsb mm5, [GLOBAL(t3)];
pxor mm0, mm0 ; 0
pxor mm7, mm7 ; 0
punpcklbw mm0, mm5 ; e0f0g0h0
psraw mm0, 11 ; sign extended shift right by 3
punpckhbw mm7, mm5 ; a0b0c0d0
psraw mm7, 11 ; sign extended shift right by 3
packsswb mm0, mm7 ; Filter2 >>=3;
movq mm5, mm0 ; Filter2
paddsb mm2, [GLOBAL(t4)] ; vp8_signed_char_clamp(Filter2 + 4)
pxor mm0, mm0 ; 0
pxor mm7, mm7 ; 0
punpcklbw mm0, mm2 ; e0f0g0h0
psraw mm0, 11 ; sign extended shift right by 3
punpckhbw mm7, mm2 ; a0b0c0d0
psraw mm7, 11 ; sign extended shift right by 3
packsswb mm0, mm7 ; Filter2 >>=3;
; mm0= filter2 mm1 = vp8_filter, mm3 =qs0 mm5=s mm4 =hev mm6=ps0
psubsb mm3, mm0 ; qs0 =qs0 - filter1
paddsb mm6, mm5 ; ps0 =ps0 + Fitler2
; mm1=vp8_filter, mm3=qs0, mm4 =hev mm6=ps0
; vp8_filter &= ~hev;
; Filter2 = vp8_filter;
pandn mm4, mm1 ; vp8_filter&=~hev
; mm3=qs0, mm4=filter2, mm6=ps0
; u = vp8_signed_char_clamp((63 + Filter2 * 27)>>7);
; s = vp8_signed_char_clamp(qs0 - u);
; *oq0 = s^0x80;
; s = vp8_signed_char_clamp(ps0 + u);
; *op0 = s^0x80;
pxor mm0, mm0
pxor mm1, mm1
pxor mm2, mm2
punpcklbw mm1, mm4
punpckhbw mm2, mm4
pmulhw mm1, [GLOBAL(s27)]
pmulhw mm2, [GLOBAL(s27)]
paddw mm1, [GLOBAL(s63)]
paddw mm2, [GLOBAL(s63)]
psraw mm1, 7
psraw mm2, 7
packsswb mm1, mm2
psubsb mm3, mm1
paddsb mm6, mm1
pxor mm3, [GLOBAL(t80)]
pxor mm6, [GLOBAL(t80)]
movq [rsi+rax], mm6
movq [rsi], mm3
; roughly 2/7th difference across boundary
; u = vp8_signed_char_clamp((63 + Filter2 * 18)>>7);
; s = vp8_signed_char_clamp(qs1 - u);
; *oq1 = s^0x80;
; s = vp8_signed_char_clamp(ps1 + u);
; *op1 = s^0x80;
pxor mm1, mm1
pxor mm2, mm2
punpcklbw mm1, mm4
punpckhbw mm2, mm4
pmulhw mm1, [GLOBAL(s18)]
pmulhw mm2, [GLOBAL(s18)]
paddw mm1, [GLOBAL(s63)]
paddw mm2, [GLOBAL(s63)]
psraw mm1, 7
psraw mm2, 7
packsswb mm1, mm2
movq mm3, [rdi]
movq mm6, [rsi+rax*2] ; p1
pxor mm3, [GLOBAL(t80)]
pxor mm6, [GLOBAL(t80)]
paddsb mm6, mm1
psubsb mm3, mm1
pxor mm6, [GLOBAL(t80)]
pxor mm3, [GLOBAL(t80)]
movq [rdi], mm3
movq [rsi+rax*2], mm6
; roughly 1/7th difference across boundary
; u = vp8_signed_char_clamp((63 + Filter2 * 9)>>7);
; s = vp8_signed_char_clamp(qs2 - u);
; *oq2 = s^0x80;
; s = vp8_signed_char_clamp(ps2 + u);
; *op2 = s^0x80;
pxor mm1, mm1
pxor mm2, mm2
punpcklbw mm1, mm4
punpckhbw mm2, mm4
pmulhw mm1, [GLOBAL(s9)]
pmulhw mm2, [GLOBAL(s9)]
paddw mm1, [GLOBAL(s63)]
paddw mm2, [GLOBAL(s63)]
psraw mm1, 7
psraw mm2, 7
packsswb mm1, mm2
movq mm6, [rdi+rax*4]
neg rax
movq mm3, [rdi+rax ]
pxor mm6, [GLOBAL(t80)]
pxor mm3, [GLOBAL(t80)]
paddsb mm6, mm1
psubsb mm3, mm1
pxor mm6, [GLOBAL(t80)]
pxor mm3, [GLOBAL(t80)]
movq [rdi+rax ], mm3
neg rax
movq [rdi+rax*4], mm6
;EARLY_BREAK_OUT:
neg rax
add rsi,8
dec rcx
jnz .next8_mbh
add rsp, 32
pop rsp
; begin epilog
pop rdi
pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
ret
;void vp8_mbloop_filter_vertical_edge_mmx
;(
; unsigned char *src_ptr,
......
......@@ -567,115 +567,6 @@ sym(vp8_loop_filter_horizontal_edge_uv_sse2):
%endmacro
;void vp8_mbloop_filter_horizontal_edge_sse2
;(
; unsigned char *src_ptr,
; int src_pixel_step,
; const char *blimit,
; const char *limit,
; const char *thresh,
; int count
;)
global sym(vp8_mbloop_filter_horizontal_edge_sse2)
sym(vp8_mbloop_filter_horizontal_edge_sse2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 6
SAVE_XMM 7
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[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
mov rdx, arg(3) ;limit
movdqa xmm7, XMMWORD PTR [rdx]
lea rdi, [rsi+rax] ; rdi points to row +1 for indirect addressing
; calculate breakout conditions and high edge variance
LFH_FILTER_AND_HEV_MASK 1
; filter and write back the results
MB_FILTER_AND_WRITEBACK 1
add rsp, 32
pop rsp
; begin epilog
pop rdi
pop rsi
RESTORE_GOT
RESTORE_XMM
UNSHADOW_ARGS
pop rbp
ret
;void vp8_mbloop_filter_horizontal_edge_uv_sse2
;(
; unsigned char *u,
; int src_pixel_step,
; const char *blimit,
; const char *limit,
; const char *thresh,
; unsigned char *v
;)
global sym(vp8_mbloop_filter_horizontal_edge_uv_sse2)
sym(vp8_mbloop_filter_horizontal_edge_uv_sse2):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 6
SAVE_XMM 7
GET_GOT rbx
push rsi
push rdi
; end prolog
ALIGN_STACK 16, rax
sub rsp, 96 ; reserve 96 bytes
%define q2 [rsp + 0] ;__declspec(align(16)) char q2[16];
%define q1 [rsp + 16] ;__declspec(align(16)) char q1[16];
%define p2 [rsp + 32] ;__declspec(align(16)) char p2[16];
%define p1 [rsp + 48] ;__declspec(align(16)) char p1[16];
%define t0 [rsp + 64] ;__declspec(align(16)) char t0[16];
%define t1 [rsp + 80] ;__declspec(align(16)) char t1[16];
mov rsi, arg(0) ; u
mov rdi, arg(5) ; v
movsxd rax, dword ptr arg(1) ; src_pixel_step
mov rcx, rax
neg rax ; negate pitch to deal with above border
mov rdx, arg(3) ;limit
movdqa xmm7, XMMWORD PTR [rdx]
lea rsi, [rsi + rcx]
lea rdi, [rdi + rcx]
; calculate breakout conditions and high edge variance
LFH_FILTER_AND_HEV_MASK 0
; filter and write back the results
MB_FILTER_AND_WRITEBACK 0
add rsp, 96
pop rsp
; begin epilog
pop rdi
pop rsi
RESTORE_GOT
RESTORE_XMM
UNSHADOW_ARGS
pop rbp
ret
%macro TRANSPOSE_16X8 2
movq xmm4, QWORD PTR [rsi] ; xx xx xx xx xx xx xx xx 07 06 05 04 03 02 01 00
movq xmm1, QWORD PTR [rdi] ; xx xx xx xx xx xx xx xx 17 16 15 14 13 12 11 10
......
......@@ -9,35 +9,26 @@
*/
#include <emmintrin.h> // SSE2
#include "vpx_config.h"
#include "vp8/common/loopfilter.h"
prototype_loopfilter(vp8_mbloop_filter_vertical_edge_mmx);
prototype_loopfilter(vp8_mbloop_filter_horizontal_edge_mmx);
prototype_loopfilter(vp8_loop_filter_vertical_edge_mmx);
prototype_loopfilter(vp8_loop_filter_horizontal_edge_mmx);
prototype_loopfilter(vp8_loop_filter_vertical_edge_sse2);
prototype_loopfilter(vp8_loop_filter_horizontal_edge_sse2);
prototype_loopfilter(vp8_mbloop_filter_vertical_edge_sse2);
prototype_loopfilter(vp8_mbloop_filter_horizontal_edge_sse2);
extern loop_filter_uvfunction vp8_loop_filter_horizontal_edge_uv_sse2;
extern loop_filter_uvfunction vp8_loop_filter_vertical_edge_uv_sse2;
extern loop_filter_uvfunction vp8_mbloop_filter_horizontal_edge_uv_sse2;
extern loop_filter_uvfunction vp8_mbloop_filter_vertical_edge_uv_sse2;
#if HAVE_MMX
/* Horizontal MB filtering */
void vp8_loop_filter_mbh_mmx(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
int y_stride, int uv_stride, struct loop_filter_info *lfi) {
vp8_mbloop_filter_horizontal_edge_mmx(y_ptr, y_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 2);
if (u_ptr)
vp8_mbloop_filter_horizontal_edge_mmx(u_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
if (v_ptr)
vp8_mbloop_filter_horizontal_edge_mmx(v_ptr, uv_stride, lfi->mblim, lfi->lim, lfi->hev_thr, 1);
}
......@@ -57,15 +48,7 @@ void vp8_loop_filter_mbv_mmx(unsigned char *y_ptr, unsigned char *u_ptr, unsigne
/* Horizontal B Filtering */
void vp8_loop_filter_bh_mmx(unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr,
int y_stride, int uv_stride, struct loop_filter_info *lfi) {
vp8_loop_filter_horizontal_edge_mmx(y_ptr + 4 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
vp8_loop_filter_horizontal_edge_mmx(y_ptr + 8 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
vp8_loop_filter_horizontal_edge_mmx(y_ptr + 12 * y_stride, y_stride, lfi->blim, lfi->lim, lfi->hev_thr, 2);
if (u_ptr)
vp8_loop_filter_horizontal_edge_mmx(u_ptr + 4 * uv_stride, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1);
if (v_ptr)
vp8_loop_filter_horizontal_edge_mmx(v_ptr + 4 * uv_stride, uv_stride, lfi->blim, lfi->lim, lfi->hev_thr, 1);
}
......@@ -99,16 +82,288 @@ void vp8_loop_filter_bvs_mmx(unsigned char *y_ptr, int y_stride, const unsigned
#endif
/* Horizontal MB filtering */
#if HAVE_SSE2
void vp8_mbloop_filter_horizontal_edge_c_sse2
(
unsigned char *s,
int p,
const unsigned char *_blimit,
const unsigned char *_limit,
const unsigned char *_thresh,
int count
) {
DECLARE_ALIGNED(16, unsigned char, flat_op2[16]);
DECLARE_ALIGNED(16, unsigned char, flat_op1[16]);
DECLARE_ALIGNED(16, unsigned char, flat_op0[16]);
DECLARE_ALIGNED(16, unsigned char, flat_oq2[16]);
DECLARE_ALIGNED(16, unsigned char, flat_oq1[16]);
DECLARE_ALIGNED(16, unsigned char, flat_oq0[16]);
__m128i mask, hev, flat;
__m128i thresh, limit, blimit;
const __m128i zero = _mm_set1_epi16(0);
__m128i p4, p3, p2, p1, p0, q0, q1, q2, q3, q4;
thresh = _mm_shuffle_epi32(_mm_cvtsi32_si128(_thresh[0] * 0x01010101), 0);
limit = _mm_shuffle_epi32(_mm_cvtsi32_si128(_limit[0] * 0x01010101), 0);
blimit = _mm_shuffle_epi32(_mm_cvtsi32_si128(_blimit[0] * 0x01010101), 0);
p4 = _mm_loadu_si128((__m128i *)(s - 5 * p));
p3 = _mm_loadu_si128((__m128i *)(s - 4 * p));
p2 = _mm_loadu_si128((__m128i *)(s - 3 * p));
p1 = _mm_loadu_si128((__m128i *)(s - 2 * p));
p0 = _mm_loadu_si128((__m128i *)(s - 1 * p));
q0 = _mm_loadu_si128((__m128i *)(s - 0 * p));
q1 = _mm_loadu_si128((__m128i *)(s + 1 * p));
q2 = _mm_loadu_si128((__m128i *)(s + 2 * p));
q3 = _mm_loadu_si128((__m128i *)(s + 3 * p));
q4 = _mm_loadu_si128((__m128i *)(s + 4 * p));
{
const __m128i abs_p1p0 = _mm_or_si128(_mm_subs_epu8(p1, p0),
_mm_subs_epu8(p0, p1));
const __m128i abs_q1q0 = _mm_or_si128(_mm_subs_epu8(q1, q0),
_mm_subs_epu8(q0, q1));
const __m128i one = _mm_set1_epi8(1);
const __m128i fe = _mm_set1_epi8(0xfe);
const __m128i ff = _mm_cmpeq_epi8(abs_p1p0, abs_p1p0);
__m128i abs_p0q0 = _mm_or_si128(_mm_subs_epu8(p0, q0),
_mm_subs_epu8(q0, p0));
__m128i abs_p1q1 = _mm_or_si128(_mm_subs_epu8(p1, q1),
_mm_subs_epu8(q1, p1));
__m128i work;
flat = _mm_max_epu8(abs_p1p0, abs_q1q0);
hev = _mm_subs_epu8(flat, thresh);
hev = _mm_xor_si128(_mm_cmpeq_epi8(hev, zero), ff);
abs_p0q0 =_mm_adds_epu8(abs_p0q0, abs_p0q0);
abs_p1q1 = _mm_srli_epi16(_mm_and_si128(abs_p1q1, fe), 1);
mask = _mm_subs_epu8(_mm_adds_epu8(abs_p0q0, abs_p1q1), blimit);
mask = _mm_xor_si128(_mm_cmpeq_epi8(mask, zero), ff);
// mask |= (abs(p0 - q0) * 2 + abs(p1 - q1) / 2 > blimit) * -1;
mask = _mm_max_epu8(flat, mask);
// mask |= (abs(p1 - p0) > limit) * -1;
// mask |= (abs(q1 - q0) > limit) * -1;
work = _mm_max_epu8(_mm_or_si128(_mm_subs_epu8(p2, p1),
_mm_subs_epu8(p1, p2)),