Skip to content

GitLab

Commit 3c755577 authored by Scott LaVarnway's avatar Scott LaVarnway Committed by Gerrit Code Review
Browse files
Options

Merge "Added predictor stride argument(s) to subtract functions"

parents 120a4640 edd98b73
Hide whitespace changes
Inline Side-by-side
Showing with 374 additions and 662 deletions
vp8/encoder/arm/armv6/vp8_subtract_armv6.asm
View file @ 3c755577
......@@ -72,22 +72,23 @@ loop_block
; r0 short *diff
; r1 unsigned char *usrc
; r2 unsigned char *vsrc
; r3 unsigned char *pred
; stack int stride
; r3 int src_stride
; sp unsigned char *upred
; sp unsigned char *vpred
; sp int pred_stride
|vp8_subtract_mbuv_armv6| PROC
stmfd sp!, {r4-r12, lr}
stmfd sp!, {r4-r11}
add r0, r0, #512 ; set *diff point to Cb
add r3, r3, #256 ; set *pred point to Cb
mov r4, #8 ; loop count
ldr r5, [sp, #40] ; stride
ldr r5, [sp, #32] ; upred
ldr r12, [sp, #40] ; pred_stride
; Subtract U block
loop_u
ldr r6, [r1] ; src (A)
ldr r7, [r3], #4 ; pred (A)
ldr r6, [r1] ; usrc (A)
ldr r7, [r5] ; upred (A)
uxtb16 r8, r6 ; [s2 | s0] (A)
uxtb16 r9, r7 ; [p2 | p0] (A)
......@@ -97,8 +98,8 @@ loop_u
usub16 r6, r8, r9 ; [d2 | d0] (A)
usub16 r7, r10, r11 ; [d3 | d1] (A)
ldr r10, [r1, #4] ; src (B)
ldr r11, [r3], #4 ; pred (B)
ldr r10, [r1, #4] ; usrc (B)
ldr r11, [r5, #4] ; upred (B)
pkhbt r8, r6, r7, lsl #16 ; [d1 | d0] (A)
pkhtb r9, r7, r6, asr #16 ; [d3 | d2] (A)
......@@ -114,7 +115,8 @@ loop_u
usub16 r6, r8, r9 ; [d2 | d0] (B)
usub16 r7, r10, r11 ; [d3 | d1] (B)
add r1, r1, r5 ; update usrc pointer
add r1, r1, r3 ; update usrc pointer
add r5, r5, r12 ; update upred pointer
pkhbt r8, r6, r7, lsl #16 ; [d1 | d0] (B)
pkhtb r9, r7, r6, asr #16 ; [d3 | d2] (B)
......@@ -125,12 +127,13 @@ loop_u
bne loop_u
ldr r5, [sp, #36] ; vpred
mov r4, #8 ; loop count
; Subtract V block
loop_v
ldr r6, [r2] ; src (A)
ldr r7, [r3], #4 ; pred (A)
ldr r6, [r2] ; vsrc (A)
ldr r7, [r5] ; vpred (A)
uxtb16 r8, r6 ; [s2 | s0] (A)
uxtb16 r9, r7 ; [p2 | p0] (A)
......@@ -140,8 +143,8 @@ loop_v
usub16 r6, r8, r9 ; [d2 | d0] (A)
usub16 r7, r10, r11 ; [d3 | d1] (A)
ldr r10, [r2, #4] ; src (B)
ldr r11, [r3], #4 ; pred (B)
ldr r10, [r2, #4] ; vsrc (B)
ldr r11, [r5, #4] ; vpred (B)
pkhbt r8, r6, r7, lsl #16 ; [d1 | d0] (A)
pkhtb r9, r7, r6, asr #16 ; [d3 | d2] (A)
......@@ -157,7 +160,8 @@ loop_v
usub16 r6, r8, r9 ; [d2 | d0] (B)
usub16 r7, r10, r11 ; [d3 | d1] (B)
add r2, r2, r5 ; update vsrc pointer
add r2, r2, r3 ; update vsrc pointer
add r5, r5, r12 ; update vpred pointer
pkhbt r8, r6, r7, lsl #16 ; [d1 | d0] (B)
pkhtb r9, r7, r6, asr #16 ; [d3 | d2] (B)
......@@ -168,23 +172,25 @@ loop_v
bne loop_v
ldmfd sp!, {r4-r12, pc}
ldmfd sp!, {r4-r11}
bx lr
ENDP
; r0 short *diff
; r1 unsigned char *src
; r2 unsigned char *pred
; r3 int stride
; r2 int src_stride
; r3 unsigned char *pred
; sp int pred_stride
|vp8_subtract_mby_armv6| PROC
stmfd sp!, {r4-r11}
ldr r12, [sp, #32] ; pred_stride
mov r4, #16
loop
ldr r6, [r1] ; src (A)
ldr r7, [r2], #4 ; pred (A)
ldr r7, [r3] ; pred (A)
uxtb16 r8, r6 ; [s2 | s0] (A)
uxtb16 r9, r7 ; [p2 | p0] (A)
......@@ -195,7 +201,7 @@ loop
usub16 r7, r10, r11 ; [d3 | d1] (A)
ldr r10, [r1, #4] ; src (B)
ldr r11, [r2], #4 ; pred (B)
ldr r11, [r3, #4] ; pred (B)
pkhbt r8, r6, r7, lsl #16 ; [d1 | d0] (A)
pkhtb r9, r7, r6, asr #16 ; [d3 | d2] (A)
......@@ -212,7 +218,7 @@ loop
usub16 r7, r10, r11 ; [d3 | d1] (B)
ldr r10, [r1, #8] ; src (C)
ldr r11, [r2], #4 ; pred (C)
ldr r11, [r3, #8] ; pred (C)
pkhbt r8, r6, r7, lsl #16 ; [d1 | d0] (B)
pkhtb r9, r7, r6, asr #16 ; [d3 | d2] (B)
......@@ -229,10 +235,10 @@ loop
usub16 r7, r10, r11 ; [d3 | d1] (C)
ldr r10, [r1, #12] ; src (D)
ldr r11, [r2], #4 ; pred (D)
ldr r11, [r3, #12] ; pred (D)
pkhbt r8, r6, r7, lsl #16 ; [d1 | d0] (C)
pkhtb r9, r7, r6, asr #16 ; [d3 | d2] (C)
pkhbt r8, r6, r7, lsl #16 ; [d1 | d0] (C)
pkhtb r9, r7, r6, asr #16 ; [d3 | d2] (C)
str r8, [r0], #4 ; diff (C)
uxtb16 r8, r10 ; [s2 | s0] (D)
......@@ -245,7 +251,8 @@ loop
usub16 r6, r8, r9 ; [d2 | d0] (D)
usub16 r7, r10, r11 ; [d3 | d1] (D)
add r1, r1, r3 ; update src pointer
add r1, r1, r2 ; update src pointer
add r3, r3, r12 ; update pred pointer
pkhbt r8, r6, r7, lsl #16 ; [d1 | d0] (D)
pkhtb r9, r7, r6, asr #16 ; [d3 | d2] (D)
......@@ -257,7 +264,7 @@ loop
bne loop
ldmfd sp!, {r4-r11}
mov pc, lr
bx lr
ENDP
......
vp8/encoder/arm/neon/subtract_neon.asm
View file @ 3c755577
......@@ -61,19 +61,24 @@
;==========================================
;void vp8_subtract_mby_neon(short *diff, unsigned char *src, unsigned char *pred, int stride)
;void vp8_subtract_mby_neon(short *diff, unsigned char *src, int src_stride
; unsigned char *pred, int pred_stride)
|vp8_subtract_mby_neon| PROC
push {r4-r7}
mov r12, #4
ldr r4, [sp, #16] ; pred_stride
mov r6, #32 ; "diff" stride x2
add r5, r0, #16 ; second diff pointer
subtract_mby_loop
vld1.8 {q0}, [r1], r3 ;load src
vld1.8 {q1}, [r2]! ;load pred
vld1.8 {q2}, [r1], r3
vld1.8 {q3}, [r2]!
vld1.8 {q4}, [r1], r3
vld1.8 {q5}, [r2]!
vld1.8 {q6}, [r1], r3
vld1.8 {q7}, [r2]!
vld1.8 {q0}, [r1], r2 ;load src
vld1.8 {q1}, [r3], r4 ;load pred
vld1.8 {q2}, [r1], r2
vld1.8 {q3}, [r3], r4
vld1.8 {q4}, [r1], r2
vld1.8 {q5}, [r3], r4
vld1.8 {q6}, [r1], r2
vld1.8 {q7}, [r3], r4
vsubl.u8 q8, d0, d2
vsubl.u8 q9, d1, d3
......@@ -84,46 +89,53 @@ subtract_mby_loop
vsubl.u8 q14, d12, d14
vsubl.u8 q15, d13, d15
vst1.16 {q8}, [r0]! ;store diff
vst1.16 {q9}, [r0]!
vst1.16 {q10}, [r0]!
vst1.16 {q11}, [r0]!
vst1.16 {q12}, [r0]!
vst1.16 {q13}, [r0]!
vst1.16 {q14}, [r0]!
vst1.16 {q15}, [r0]!
vst1.16 {q8}, [r0], r6 ;store diff
vst1.16 {q9}, [r5], r6
vst1.16 {q10}, [r0], r6
vst1.16 {q11}, [r5], r6
vst1.16 {q12}, [r0], r6
vst1.16 {q13}, [r5], r6
vst1.16 {q14}, [r0], r6
vst1.16 {q15}, [r5], r6
subs r12, r12, #1
bne subtract_mby_loop
pop {r4-r7}
bx lr
ENDP
;=================================
;void vp8_subtract_mbuv_neon(short *diff, unsigned char *usrc, unsigned char *vsrc, unsigned char *pred, int stride)
;void vp8_subtract_mbuv_c(short *diff, unsigned char *usrc, unsigned char *vsrc,
; int src_stride, unsigned char *upred,
; unsigned char *vpred, int pred_stride)
|vp8_subtract_mbuv_neon| PROC
ldr r12, [sp]
push {r4-r7}
ldr r4, [sp, #16] ; upred
ldr r5, [sp, #20] ; vpred
ldr r6, [sp, #24] ; pred_stride
add r0, r0, #512 ; short *udiff = diff + 256;
mov r12, #32 ; "diff" stride x2
add r7, r0, #16 ; second diff pointer
;u
add r0, r0, #512 ; short *udiff = diff + 256;
add r3, r3, #256 ; unsigned char *upred = pred + 256;
vld1.8 {d0}, [r1], r12 ;load src
vld1.8 {d1}, [r3]! ;load pred
vld1.8 {d2}, [r1], r12
vld1.8 {d3}, [r3]!
vld1.8 {d4}, [r1], r12
vld1.8 {d5}, [r3]!
vld1.8 {d6}, [r1], r12
vld1.8 {d7}, [r3]!
vld1.8 {d8}, [r1], r12
vld1.8 {d9}, [r3]!
vld1.8 {d10}, [r1], r12
vld1.8 {d11}, [r3]!
vld1.8 {d12}, [r1], r12
vld1.8 {d13}, [r3]!
vld1.8 {d14}, [r1], r12
vld1.8 {d15}, [r3]!
vld1.8 {d0}, [r1], r3 ;load usrc
vld1.8 {d1}, [r4], r6 ;load upred
vld1.8 {d2}, [r1], r3
vld1.8 {d3}, [r4], r6
vld1.8 {d4}, [r1], r3
vld1.8 {d5}, [r4], r6
vld1.8 {d6}, [r1], r3
vld1.8 {d7}, [r4], r6
vld1.8 {d8}, [r1], r3
vld1.8 {d9}, [r4], r6
vld1.8 {d10}, [r1], r3
vld1.8 {d11}, [r4], r6
vld1.8 {d12}, [r1], r3
vld1.8 {d13}, [r4], r6
vld1.8 {d14}, [r1], r3
vld1.8 {d15}, [r4], r6
vsubl.u8 q8, d0, d1
vsubl.u8 q9, d2, d3
......@@ -134,32 +146,32 @@ subtract_mby_loop
vsubl.u8 q14, d12, d13
vsubl.u8 q15, d14, d15
vst1.16 {q8}, [r0]! ;store diff
vst1.16 {q9}, [r0]!
vst1.16 {q10}, [r0]!
vst1.16 {q11}, [r0]!
vst1.16 {q12}, [r0]!
vst1.16 {q13}, [r0]!
vst1.16 {q14}, [r0]!
vst1.16 {q15}, [r0]!
vst1.16 {q8}, [r0], r12 ;store diff
vst1.16 {q9}, [r7], r12
vst1.16 {q10}, [r0], r12
vst1.16 {q11}, [r7], r12
vst1.16 {q12}, [r0], r12
vst1.16 {q13}, [r7], r12
vst1.16 {q14}, [r0], r12
vst1.16 {q15}, [r7], r12
;v
vld1.8 {d0}, [r2], r12 ;load src
vld1.8 {d1}, [r3]! ;load pred
vld1.8 {d2}, [r2], r12
vld1.8 {d3}, [r3]!
vld1.8 {d4}, [r2], r12
vld1.8 {d5}, [r3]!
vld1.8 {d6}, [r2], r12
vld1.8 {d7}, [r3]!
vld1.8 {d8}, [r2], r12
vld1.8 {d9}, [r3]!
vld1.8 {d10}, [r2], r12
vld1.8 {d11}, [r3]!
vld1.8 {d12}, [r2], r12
vld1.8 {d13}, [r3]!
vld1.8 {d14}, [r2], r12
vld1.8 {d15}, [r3]!
vld1.8 {d0}, [r2], r3 ;load vsrc
vld1.8 {d1}, [r5], r6 ;load vpred
vld1.8 {d2}, [r2], r3
vld1.8 {d3}, [r5], r6
vld1.8 {d4}, [r2], r3
vld1.8 {d5}, [r5], r6
vld1.8 {d6}, [r2], r3
vld1.8 {d7}, [r5], r6
vld1.8 {d8}, [r2], r3
vld1.8 {d9}, [r5], r6
vld1.8 {d10}, [r2], r3
vld1.8 {d11}, [r5], r6
vld1.8 {d12}, [r2], r3
vld1.8 {d13}, [r5], r6
vld1.8 {d14}, [r2], r3
vld1.8 {d15}, [r5], r6
vsubl.u8 q8, d0, d1
vsubl.u8 q9, d2, d3
......@@ -170,16 +182,18 @@ subtract_mby_loop
vsubl.u8 q14, d12, d13
vsubl.u8 q15, d14, d15
vst1.16 {q8}, [r0]! ;store diff
vst1.16 {q9}, [r0]!
vst1.16 {q10}, [r0]!
vst1.16 {q11}, [r0]!
vst1.16 {q12}, [r0]!
vst1.16 {q13}, [r0]!
vst1.16 {q14}, [r0]!
vst1.16 {q15}, [r0]!
vst1.16 {q8}, [r0], r12 ;store diff
vst1.16 {q9}, [r7], r12
vst1.16 {q10}, [r0], r12
vst1.16 {q11}, [r7], r12
vst1.16 {q12}, [r0], r12
vst1.16 {q13}, [r7], r12
vst1.16 {q14}, [r0], r12
vst1.16 {q15}, [r7], r12
pop {r4-r7}
bx lr
ENDP
END
vp8/encoder/encodeintra.c
View file @ 3c755577
......@@ -100,7 +100,7 @@ void vp8_encode_intra16x16mby(const VP8_ENCODER_RTCD *rtcd, MACROBLOCK *x)
RECON_INVOKE(&rtcd->common->recon, build_intra_predictors_mby)(&x->e_mbd);
ENCODEMB_INVOKE(&rtcd->encodemb, submby)(x->src_diff, *(b->base_src),
x->e_mbd.predictor, b->src_stride);
b->src_stride, x->e_mbd.predictor, 16);
vp8_transform_intra_mby(x);
......@@ -115,7 +115,9 @@ void vp8_encode_intra16x16mbuv(const VP8_ENCODER_RTCD *rtcd, MACROBLOCK *x)
{
RECON_INVOKE(&rtcd->common->recon, build_intra_predictors_mbuv)(&x->e_mbd);
ENCODEMB_INVOKE(&rtcd->encodemb, submbuv)(x->src_diff, x->src.u_buffer, x->src.v_buffer, x->e_mbd.predictor, x->src.uv_stride);
ENCODEMB_INVOKE(&rtcd->encodemb, submbuv)(x->src_diff, x->src.u_buffer,
x->src.v_buffer, x->src.uv_stride, &x->e_mbd.predictor[256],
&x->e_mbd.predictor[320], 8);
vp8_transform_mbuv(x);
......
vp8/encoder/encodemb.c
View file @ 3c755577
......@@ -48,12 +48,12 @@ void vp8_subtract_b_c(BLOCK *be, BLOCKD *bd, int pitch)
}
}
void vp8_subtract_mbuv_c(short *diff, unsigned char *usrc, unsigned char *vsrc, unsigned char *pred, int stride)
void vp8_subtract_mbuv_c(short *diff, unsigned char *usrc, unsigned char *vsrc,
int src_stride, unsigned char *upred,
unsigned char *vpred, int pred_stride)
{
short *udiff = diff + 256;
short *vdiff = diff + 320;
unsigned char *upred = pred + 256;
unsigned char *vpred = pred + 320;
int r, c;
......@@ -65,8 +65,8 @@ void vp8_subtract_mbuv_c(short *diff, unsigned char *usrc, unsigned char *vsrc,
}
udiff += 8;
upred += 8;
usrc += stride;
upred += pred_stride;
usrc += src_stride;
}
for (r = 0; r < 8; r++)
......@@ -77,12 +77,13 @@ void vp8_subtract_mbuv_c(short *diff, unsigned char *usrc, unsigned char *vsrc,
}
vdiff += 8;
vpred += 8;
vsrc += stride;
vpred += pred_stride;
vsrc += src_stride;
}
}
void vp8_subtract_mby_c(short *diff, unsigned char *src, unsigned char *pred, int stride)
void vp8_subtract_mby_c(short *diff, unsigned char *src, int src_stride,
unsigned char *pred, int pred_stride)
{
int r, c;
......@@ -94,8 +95,8 @@ void vp8_subtract_mby_c(short *diff, unsigned char *src, unsigned char *pred, in
}
diff += 16;
pred += 16;
src += stride;
pred += pred_stride;
src += src_stride;
}
}
......@@ -103,8 +104,11 @@ static void vp8_subtract_mb(const VP8_ENCODER_RTCD *rtcd, MACROBLOCK *x)
{
BLOCK *b = &x->block[0];
ENCODEMB_INVOKE(&rtcd->encodemb, submby)(x->src_diff, *(b->base_src), x->e_mbd.predictor, b->src_stride);
ENCODEMB_INVOKE(&rtcd->encodemb, submbuv)(x->src_diff, x->src.u_buffer, x->src.v_buffer, x->e_mbd.predictor, x->src.uv_stride);
ENCODEMB_INVOKE(&rtcd->encodemb, submby)(x->src_diff, *(b->base_src),
b->src_stride, x->e_mbd.predictor, 16);
ENCODEMB_INVOKE(&rtcd->encodemb, submbuv)(x->src_diff, x->src.u_buffer,
x->src.v_buffer, x->src.uv_stride, &x->e_mbd.predictor[256],
&x->e_mbd.predictor[320], 8);
}
static void build_dcblock(MACROBLOCK *x)
......@@ -641,7 +645,8 @@ void vp8_encode_inter16x16y(const VP8_ENCODER_RTCD *rtcd, MACROBLOCK *x)
vp8_build_inter16x16_predictors_mby(&x->e_mbd);
ENCODEMB_INVOKE(&rtcd->encodemb, submby)(x->src_diff, *(b->base_src), x->e_mbd.predictor, b->src_stride);
ENCODEMB_INVOKE(&rtcd->encodemb, submby)(x->src_diff, *(b->base_src),
b->src_stride, x->e_mbd.predictor, 16);
transform_mby(x);
......
vp8/encoder/encodemb.h
View file @ 3c755577
......@@ -28,11 +28,13 @@
void (sym)(BLOCK *be,BLOCKD *bd, int pitch)
#define prototype_submby(sym) \
void (sym)(short *diff, unsigned char *src, unsigned char *pred, int stride)
void (sym)(short *diff, unsigned char *src, int src_stride, \
unsigned char *pred, int pred_stride)
#define prototype_submbuv(sym) \
void (sym)(short *diff, unsigned char *usrc, unsigned char *vsrc,\
unsigned char *pred, int stride)
int src_stride, unsigned char *upred, unsigned char *vpred,\
int pred_stride)
#if ARCH_X86 || ARCH_X86_64
#include "x86/encodemb_x86.h"
......
vp8/encoder/rdopt.c
View file @ 3c755577
......@@ -552,7 +552,7 @@ static void macro_block_yrd( MACROBLOCK *mb,
int d;
ENCODEMB_INVOKE(rtcd, submby)( mb->src_diff, *(mb->block[0].base_src),
mb->e_mbd.predictor, mb->block[0].src_stride );
mb->block[0].src_stride, mb->e_mbd.predictor, 16);
// Fdct and building the 2nd order block
for (beptr = mb->block; beptr < mb->block + 16; beptr += 2)
......@@ -800,7 +800,8 @@ static int rd_inter16x16_uv(VP8_COMP *cpi, MACROBLOCK *x, int *rate,
{
vp8_build_inter16x16_predictors_mbuv(&x->e_mbd);
ENCODEMB_INVOKE(IF_RTCD(&cpi->rtcd.encodemb), submbuv)(x->src_diff,
x->src.u_buffer, x->src.v_buffer, x->e_mbd.predictor, x->src.uv_stride);
x->src.u_buffer, x->src.v_buffer, x->src.uv_stride,
&x->e_mbd.predictor[256], &x->e_mbd.predictor[320], 8);
vp8_transform_mbuv(x);
vp8_quantize_mbuv(x);
......@@ -816,7 +817,8 @@ static int rd_inter4x4_uv(VP8_COMP *cpi, MACROBLOCK *x, int *rate,
{
vp8_build_inter4x4_predictors_mbuv(&x->e_mbd);
ENCODEMB_INVOKE(IF_RTCD(&cpi->rtcd.encodemb), submbuv)(x->src_diff,
x->src.u_buffer, x->src.v_buffer, x->e_mbd.predictor, x->src.uv_stride);
x->src.u_buffer, x->src.v_buffer, x->src.uv_stride,
&x->e_mbd.predictor[256], &x->e_mbd.predictor[320], 8);
vp8_transform_mbuv(x);
vp8_quantize_mbuv(x);
......@@ -845,8 +847,8 @@ static void rd_pick_intra_mbuv_mode(VP8_COMP *cpi, MACROBLOCK *x, int *rate, int
RECON_INVOKE(&cpi->rtcd.common->recon, build_intra_predictors_mbuv)
(&x->e_mbd);
ENCODEMB_INVOKE(IF_RTCD(&cpi->rtcd.encodemb), submbuv)(x->src_diff,
x->src.u_buffer, x->src.v_buffer, x->e_mbd.predictor,
x->src.uv_stride);
x->src.u_buffer, x->src.v_buffer, x->src.uv_stride,
&x->e_mbd.predictor[256], &x->e_mbd.predictor[320], 8);
vp8_transform_mbuv(x);
vp8_quantize_mbuv(x);
......
vp8/encoder/x86/subtract_mmx.asm
View file @ 3c755577
......@@ -73,74 +73,71 @@ sym(vp8_subtract_b_mmx_impl):
pop rbp
ret
;void vp8_subtract_mby_mmx(short *diff, unsigned char *src, unsigned char *pred, int stride)
;void vp8_subtract_mby_mmx(short *diff, unsigned char *src, int src_stride,
;unsigned char *pred, int pred_stride)
global sym(vp8_subtract_mby_mmx)
sym(vp8_subtract_mby_mmx):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 4
SHADOW_ARGS_TO_STACK 5
push rsi
push rdi
; end prolog
mov rdi, arg(0) ;diff
mov rsi, arg(1) ;src
movsxd rdx, dword ptr arg(2);src_stride
mov rax, arg(3) ;pred
push rbx
movsxd rbx, dword ptr arg(4);pred_stride
mov rsi, arg(1) ;src
mov rdi, arg(0) ;diff
mov rax, arg(2) ;pred
movsxd rdx, dword ptr arg(3) ;stride
pxor mm0, mm0
mov rcx, 16
mov rcx, 16
pxor mm0, mm0
.submby_loop:
movq mm1, [rsi]
movq mm3, [rax]
movq mm1, [rsi]
movq mm3, [rax]
movq mm2, mm1
movq mm4, mm3
punpcklbw mm1, mm0
punpcklbw mm3, mm0
punpckhbw mm2, mm0
punpckhbw mm4, mm0
psubw mm1, mm3
psubw mm2, mm4
movq mm2, mm1
movq mm4, mm3
movq [rdi], mm1
movq [rdi+8], mm2
punpcklbw mm1, mm0
punpcklbw mm3, mm0
punpckhbw mm2, mm0
punpckhbw mm4, mm0
movq mm1, [rsi+8]
movq mm3, [rax+8]
psubw mm1, mm3
psubw mm2, mm4
movq mm2, mm1