vpx/vp9/common/arm/neon/vp9_convolve8_neon.asm
Mans Rullgard 4fa93bcef4 vp9: neon: use aligned stores in convolve functions
The destination is block-aligned so it is safe to use aligned
stores.

Change-Id: I38261e4fa40bc60e6472edffece59e372908da7e
2013-08-16 14:25:08 +01:00

281 lines
7.4 KiB
NASM

;
; Copyright (c) 2013 The WebM project authors. All Rights Reserved.
;
; Use of this source code is governed by a BSD-style license
; that can be found in the LICENSE file in the root of the source
; tree. An additional intellectual property rights grant can be found
; in the file PATENTS. All contributing project authors may
; be found in the AUTHORS file in the root of the source tree.
;
; These functions are only valid when:
; x_step_q4 == 16
; w%4 == 0
; h%4 == 0
; taps == 8
; VP9_FILTER_WEIGHT == 128
; VP9_FILTER_SHIFT == 7
EXPORT |vp9_convolve8_horiz_neon|
EXPORT |vp9_convolve8_vert_neon|
IMPORT |vp9_convolve8_horiz_c|
IMPORT |vp9_convolve8_vert_c|
ARM
REQUIRE8
PRESERVE8
AREA ||.text||, CODE, READONLY, ALIGN=2
; Multiply and accumulate by q0
MACRO
MULTIPLY_BY_Q0 $dst, $src0, $src1, $src2, $src3, $src4, $src5, $src6, $src7
vmull.s16 $dst, $src0, d0[0]
vmlal.s16 $dst, $src1, d0[1]
vmlal.s16 $dst, $src2, d0[2]
vmlal.s16 $dst, $src3, d0[3]
vmlal.s16 $dst, $src4, d1[0]
vmlal.s16 $dst, $src5, d1[1]
vmlal.s16 $dst, $src6, d1[2]
vmlal.s16 $dst, $src7, d1[3]
MEND
; r0 const uint8_t *src
; r1 int src_stride
; r2 uint8_t *dst
; r3 int dst_stride
; sp[]const int16_t *filter_x
; sp[]int x_step_q4
; sp[]const int16_t *filter_y ; unused
; sp[]int y_step_q4 ; unused
; sp[]int w
; sp[]int h
|vp9_convolve8_horiz_neon| PROC
ldr r12, [sp, #4] ; x_step_q4
cmp r12, #16
bne vp9_convolve8_horiz_c
push {r4-r10, lr}
sub r0, r0, #3 ; adjust for taps
ldr r5, [sp, #32] ; filter_x
ldr r6, [sp, #48] ; w
ldr r7, [sp, #52] ; h
vld1.s16 {q0}, [r5] ; filter_x
sub r8, r1, r1, lsl #2 ; -src_stride * 3
add r8, r8, #4 ; -src_stride * 3 + 4
sub r4, r3, r3, lsl #2 ; -dst_stride * 3
add r4, r4, #4 ; -dst_stride * 3 + 4
rsb r9, r6, r1, lsl #2 ; reset src for outer loop
sub r9, r9, #7
rsb r12, r6, r3, lsl #2 ; reset dst for outer loop
mov r10, r6 ; w loop counter
loop_horiz_v
vld1.8 {d24}, [r0], r1
vld1.8 {d25}, [r0], r1
vld1.8 {d26}, [r0], r1
vld1.8 {d27}, [r0], r8
vtrn.16 q12, q13
vtrn.8 d24, d25
vtrn.8 d26, d27
pld [r0, r1, lsl #2]
vmovl.u8 q8, d24
vmovl.u8 q9, d25
vmovl.u8 q10, d26
vmovl.u8 q11, d27
; save a few instructions in the inner loop
vswp d17, d18
vmov d23, d21
add r0, r0, #3
loop_horiz
add r5, r0, #64
vld1.32 {d28[]}, [r0], r1
vld1.32 {d29[]}, [r0], r1
vld1.32 {d31[]}, [r0], r1
vld1.32 {d30[]}, [r0], r8
pld [r5]
vtrn.16 d28, d31
vtrn.16 d29, d30
vtrn.8 d28, d29
vtrn.8 d31, d30
pld [r5, r1]
; extract to s16
vtrn.32 q14, q15
vmovl.u8 q12, d28
vmovl.u8 q13, d29
pld [r5, r1, lsl #1]
; src[] * filter_x
MULTIPLY_BY_Q0 q1, d16, d17, d20, d22, d18, d19, d23, d24
MULTIPLY_BY_Q0 q2, d17, d20, d22, d18, d19, d23, d24, d26
MULTIPLY_BY_Q0 q14, d20, d22, d18, d19, d23, d24, d26, d27
MULTIPLY_BY_Q0 q15, d22, d18, d19, d23, d24, d26, d27, d25
pld [r5, -r8]
; += 64 >> 7
vqrshrun.s32 d2, q1, #7
vqrshrun.s32 d3, q2, #7
vqrshrun.s32 d4, q14, #7
vqrshrun.s32 d5, q15, #7
; saturate
vqmovn.u16 d2, q1
vqmovn.u16 d3, q2
; transpose
vtrn.16 d2, d3
vtrn.32 d2, d3
vtrn.8 d2, d3
vst1.u32 {d2[0]}, [r2@32], r3
vst1.u32 {d3[0]}, [r2@32], r3
vst1.u32 {d2[1]}, [r2@32], r3
vst1.u32 {d3[1]}, [r2@32], r4
vmov q8, q9
vmov d20, d23
vmov q11, q12
vmov q9, q13
subs r6, r6, #4 ; w -= 4
bgt loop_horiz
; outer loop
mov r6, r10 ; restore w counter
add r0, r0, r9 ; src += src_stride * 4 - w
add r2, r2, r12 ; dst += dst_stride * 4 - w
subs r7, r7, #4 ; h -= 4
bgt loop_horiz_v
pop {r4-r10, pc}
ENDP
|vp9_convolve8_vert_neon| PROC
ldr r12, [sp, #12]
cmp r12, #16
bne vp9_convolve8_vert_c
push {r4-r8, lr}
; adjust for taps
sub r0, r0, r1
sub r0, r0, r1, lsl #1
ldr r4, [sp, #32] ; filter_y
ldr r6, [sp, #40] ; w
ldr lr, [sp, #44] ; h
vld1.s16 {q0}, [r4] ; filter_y
lsl r1, r1, #1
lsl r3, r3, #1
loop_vert_h
mov r4, r0
add r7, r0, r1, asr #1
mov r5, r2
add r8, r2, r3, asr #1
mov r12, lr ; h loop counter
vld1.u32 {d16[0]}, [r4], r1
vld1.u32 {d16[1]}, [r7], r1
vld1.u32 {d18[0]}, [r4], r1
vld1.u32 {d18[1]}, [r7], r1
vld1.u32 {d20[0]}, [r4], r1
vld1.u32 {d20[1]}, [r7], r1
vld1.u32 {d22[0]}, [r4], r1
vmovl.u8 q8, d16
vmovl.u8 q9, d18
vmovl.u8 q10, d20
vmovl.u8 q11, d22
loop_vert
; always process a 4x4 block at a time
vld1.u32 {d24[0]}, [r7], r1
vld1.u32 {d26[0]}, [r4], r1
vld1.u32 {d26[1]}, [r7], r1
vld1.u32 {d24[1]}, [r4], r1
; extract to s16
vmovl.u8 q12, d24
vmovl.u8 q13, d26
pld [r5]
pld [r8]
; src[] * filter_y
MULTIPLY_BY_Q0 q1, d16, d17, d18, d19, d20, d21, d22, d24
pld [r5, r3]
pld [r8, r3]
MULTIPLY_BY_Q0 q2, d17, d18, d19, d20, d21, d22, d24, d26
pld [r7]
pld [r4]
MULTIPLY_BY_Q0 q14, d18, d19, d20, d21, d22, d24, d26, d27
pld [r7, r1]
pld [r4, r1]
MULTIPLY_BY_Q0 q15, d19, d20, d21, d22, d24, d26, d27, d25
; += 64 >> 7
vqrshrun.s32 d2, q1, #7
vqrshrun.s32 d3, q2, #7
vqrshrun.s32 d4, q14, #7
vqrshrun.s32 d5, q15, #7
; saturate
vqmovn.u16 d2, q1
vqmovn.u16 d3, q2
vst1.u32 {d2[0]}, [r5@32], r3
vst1.u32 {d2[1]}, [r8@32], r3
vst1.u32 {d3[0]}, [r5@32], r3
vst1.u32 {d3[1]}, [r8@32], r3
vmov q8, q10
vmov d18, d22
vmov d19, d24
vmov q10, q13
vmov d22, d25
subs r12, r12, #4 ; h -= 4
bgt loop_vert
; outer loop
add r0, r0, #4
add r2, r2, #4
subs r6, r6, #4 ; w -= 4
bgt loop_vert_h
pop {r4-r8, pc}
ENDP
END