generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
ba48a11130
vpx/vp8/common/arm/neon/vp8_subpixelvariance16x16_neon.asm

424 lines
13 KiB
NASM
Raw Normal View History

Initial WebM release
2010-05-18 17:58:33 +02:00
;
Use WebM in copyright notice for consistency Changes 'The VP8 project' to 'The WebM project', for consistency with other webmproject.org repositories. Fixes issue #97. Change-Id: I37c13ed5fbdb9d334ceef71c6350e9febed9bbba
2010-09-09 14:16:39 +02:00
; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
Initial WebM release
2010-05-18 17:58:33 +02:00
;
cosmetics: trim trailing whitespace When the license headers were updated, they accidentally contained trailing whitespace, so unfortunately we have to touch all the files again. Change-Id: I236c05fade06589e417179c0444cb39b09e4200d
2010-06-18 18:39:21 +02:00
; Use of this source code is governed by a BSD-style license
LICENSE: update with latest text Change-Id: Ieebea089095d9073b3a94932791099f614ce120c
2010-06-04 22:19:40 +02:00
; that can be found in the LICENSE file in the root of the source
; tree. An additional intellectual property rights grant can be found
cosmetics: trim trailing whitespace When the license headers were updated, they accidentally contained trailing whitespace, so unfortunately we have to touch all the files again. Change-Id: I236c05fade06589e417179c0444cb39b09e4200d
2010-06-18 18:39:21 +02:00
; in the file PATENTS. All contributing project authors may
LICENSE: update with latest text Change-Id: Ieebea089095d9073b3a94932791099f614ce120c
2010-06-04 22:19:40 +02:00
; be found in the AUTHORS file in the root of the source tree.
Initial WebM release
2010-05-18 17:58:33 +02:00
;
Fix TEXTRELs in the ARM asm. Besides imposing a performance penalty at startup in most configurations, these relocations break the dynamic linker for native Fennec, since it does not support them at all. Change-Id: Id5dc768609354ebb4379966eb61a7313e6fd18de
2012-05-02 19:11:36 +02:00
bilinear_taps_coeff
DCD 128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
;-----------------
Modify calling of NEON code in sub-pixel search In vp8_find_best_sub_pixel_step_iteratively(), many times xoffset and yoffset are specific values - (4,0) (0,4) and (4,4). Modified code to call simplified NEON version at these specific offsets to help with the performance. Change-Id: Iaf896a0f7aae4697bd36a49e182525dd1ef1ab4d
2011-01-18 20:19:52 +01:00
EXPORT |vp8_sub_pixel_variance16x16_neon_func|
Initial WebM release
2010-05-18 17:58:33 +02:00
ARM
REQUIRE8
PRESERVE8
AREA ||.text||, CODE, READONLY, ALIGN=2
; r0 unsigned char *src_ptr,
; r1 int src_pixels_per_line,
; r2 int xoffset,
; r3 int yoffset,
; stack(r4) unsigned char *dst_ptr,
; stack(r5) int dst_pixels_per_line,
; stack(r6) unsigned int *sse
;note: most of the code is copied from bilinear_predict16x16_neon and vp8_variance16x16_neon.
Modify calling of NEON code in sub-pixel search In vp8_find_best_sub_pixel_step_iteratively(), many times xoffset and yoffset are specific values - (4,0) (0,4) and (4,4). Modified code to call simplified NEON version at these specific offsets to help with the performance. Change-Id: Iaf896a0f7aae4697bd36a49e182525dd1ef1ab4d
2011-01-18 20:19:52 +01:00
|vp8_sub_pixel_variance16x16_neon_func| PROC
Initial WebM release
2010-05-18 17:58:33 +02:00
push {r4-r6, lr}
Fix TEXTRELs in the ARM asm. Besides imposing a performance penalty at startup in most configurations, these relocations break the dynamic linker for native Fennec, since it does not support them at all. Change-Id: Id5dc768609354ebb4379966eb61a7313e6fd18de
2012-05-02 19:11:36 +02:00
adr r12, bilinear_taps_coeff
Initial WebM release
2010-05-18 17:58:33 +02:00
ldr r4, [sp, #16] ;load *dst_ptr from stack
ldr r5, [sp, #20] ;load dst_pixels_per_line from stack
ldr r6, [sp, #24] ;load *sse from stack
cmp r2, #0 ;skip first_pass filter if xoffset=0
beq secondpass_bfilter16x16_only
add r2, r12, r2, lsl #3 ;calculate filter location
cmp r3, #0 ;skip second_pass filter if yoffset=0
vld1.s32 {d31}, [r2] ;load first_pass filter
beq firstpass_bfilter16x16_only
sub sp, sp, #272 ;reserve space on stack for temporary storage
vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data
mov lr, sp
vld1.u8 {d5, d6, d7}, [r0], r1
mov r2, #3 ;loop counter
vld1.u8 {d8, d9, d10}, [r0], r1
vdup.8 d0, d31[0] ;first_pass filter (d0 d1)
vld1.u8 {d11, d12, d13}, [r0], r1
vdup.8 d1, d31[4]
;First Pass: output_height lines x output_width columns (17x16)
vp8e_filt_blk2d_fp16x16_loop_neon
pld [r0]
pld [r0, r1]
pld [r0, r1, lsl #1]
vmull.u8 q7, d2, d0 ;(src_ptr[0] * Filter[0])
vmull.u8 q8, d3, d0
vmull.u8 q9, d5, d0
vmull.u8 q10, d6, d0
vmull.u8 q11, d8, d0
vmull.u8 q12, d9, d0
vmull.u8 q13, d11, d0
vmull.u8 q14, d12, d0
vext.8 d2, d2, d3, #1 ;construct src_ptr[1]
vext.8 d5, d5, d6, #1
vext.8 d8, d8, d9, #1
vext.8 d11, d11, d12, #1
vmlal.u8 q7, d2, d1 ;(src_ptr[0] * Filter[1])
vmlal.u8 q9, d5, d1
vmlal.u8 q11, d8, d1
vmlal.u8 q13, d11, d1
vext.8 d3, d3, d4, #1
vext.8 d6, d6, d7, #1
vext.8 d9, d9, d10, #1
vext.8 d12, d12, d13, #1
vmlal.u8 q8, d3, d1 ;(src_ptr[0] * Filter[1])
vmlal.u8 q10, d6, d1
vmlal.u8 q12, d9, d1
vmlal.u8 q14, d12, d1
subs r2, r2, #1
vqrshrn.u16 d14, q7, #7 ;shift/round/saturate to u8
vqrshrn.u16 d15, q8, #7
vqrshrn.u16 d16, q9, #7
vqrshrn.u16 d17, q10, #7
vqrshrn.u16 d18, q11, #7
vqrshrn.u16 d19, q12, #7
vqrshrn.u16 d20, q13, #7
vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data
vqrshrn.u16 d21, q14, #7
vld1.u8 {d5, d6, d7}, [r0], r1
vst1.u8 {d14, d15, d16, d17}, [lr]! ;store result
vld1.u8 {d8, d9, d10}, [r0], r1
vst1.u8 {d18, d19, d20, d21}, [lr]!
vld1.u8 {d11, d12, d13}, [r0], r1
bne vp8e_filt_blk2d_fp16x16_loop_neon
;First-pass filtering for rest 5 lines
vld1.u8 {d14, d15, d16}, [r0], r1
vmull.u8 q9, d2, d0 ;(src_ptr[0] * Filter[0])
vmull.u8 q10, d3, d0
vmull.u8 q11, d5, d0
vmull.u8 q12, d6, d0
vmull.u8 q13, d8, d0
vmull.u8 q14, d9, d0
vext.8 d2, d2, d3, #1 ;construct src_ptr[1]
vext.8 d5, d5, d6, #1
vext.8 d8, d8, d9, #1
vmlal.u8 q9, d2, d1 ;(src_ptr[0] * Filter[1])
vmlal.u8 q11, d5, d1
vmlal.u8 q13, d8, d1
vext.8 d3, d3, d4, #1
vext.8 d6, d6, d7, #1
vext.8 d9, d9, d10, #1
vmlal.u8 q10, d3, d1 ;(src_ptr[0] * Filter[1])
vmlal.u8 q12, d6, d1
vmlal.u8 q14, d9, d1
vmull.u8 q1, d11, d0
vmull.u8 q2, d12, d0
vmull.u8 q3, d14, d0
vmull.u8 q4, d15, d0
vext.8 d11, d11, d12, #1 ;construct src_ptr[1]
vext.8 d14, d14, d15, #1
vmlal.u8 q1, d11, d1 ;(src_ptr[0] * Filter[1])
vmlal.u8 q3, d14, d1
vext.8 d12, d12, d13, #1
vext.8 d15, d15, d16, #1
vmlal.u8 q2, d12, d1 ;(src_ptr[0] * Filter[1])
vmlal.u8 q4, d15, d1
vqrshrn.u16 d10, q9, #7 ;shift/round/saturate to u8
vqrshrn.u16 d11, q10, #7
vqrshrn.u16 d12, q11, #7
vqrshrn.u16 d13, q12, #7
vqrshrn.u16 d14, q13, #7
vqrshrn.u16 d15, q14, #7
vqrshrn.u16 d16, q1, #7
vqrshrn.u16 d17, q2, #7
vqrshrn.u16 d18, q3, #7
vqrshrn.u16 d19, q4, #7
vst1.u8 {d10, d11, d12, d13}, [lr]! ;store result
vst1.u8 {d14, d15, d16, d17}, [lr]!
vst1.u8 {d18, d19}, [lr]!
;Second pass: 16x16
;secondpass_filter
add r3, r12, r3, lsl #3
sub lr, lr, #272
vld1.u32 {d31}, [r3] ;load second_pass filter
sub sp, sp, #256
mov r3, sp
vld1.u8 {d22, d23}, [lr]! ;load src data
vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1)
vdup.8 d1, d31[4]
mov r12, #4 ;loop counter
vp8e_filt_blk2d_sp16x16_loop_neon
vld1.u8 {d24, d25}, [lr]!
vmull.u8 q1, d22, d0 ;(src_ptr[0] * Filter[0])
vld1.u8 {d26, d27}, [lr]!
vmull.u8 q2, d23, d0
vld1.u8 {d28, d29}, [lr]!
vmull.u8 q3, d24, d0
vld1.u8 {d30, d31}, [lr]!
vmull.u8 q4, d25, d0
vmull.u8 q5, d26, d0
vmull.u8 q6, d27, d0
vmull.u8 q7, d28, d0
vmull.u8 q8, d29, d0
vmlal.u8 q1, d24, d1 ;(src_ptr[pixel_step] * Filter[1])
vmlal.u8 q2, d25, d1
vmlal.u8 q3, d26, d1
vmlal.u8 q4, d27, d1
vmlal.u8 q5, d28, d1
vmlal.u8 q6, d29, d1
vmlal.u8 q7, d30, d1
vmlal.u8 q8, d31, d1
subs r12, r12, #1
vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8
vqrshrn.u16 d3, q2, #7
vqrshrn.u16 d4, q3, #7
vqrshrn.u16 d5, q4, #7
vqrshrn.u16 d6, q5, #7
vqrshrn.u16 d7, q6, #7
vqrshrn.u16 d8, q7, #7
vqrshrn.u16 d9, q8, #7
vst1.u8 {d2, d3}, [r3]! ;store result
vst1.u8 {d4, d5}, [r3]!
vst1.u8 {d6, d7}, [r3]!
vmov q11, q15
vst1.u8 {d8, d9}, [r3]!
bne vp8e_filt_blk2d_sp16x16_loop_neon
b sub_pixel_variance16x16_neon
;--------------------
firstpass_bfilter16x16_only
mov r2, #4 ;loop counter
sub sp, sp, #528 ;reserve space on stack for temporary storage
vdup.8 d0, d31[0] ;first_pass filter (d0 d1)
vdup.8 d1, d31[4]
mov r3, sp
;First Pass: output_height lines x output_width columns (16x16)
vp8e_filt_blk2d_fpo16x16_loop_neon
vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data
vld1.u8 {d5, d6, d7}, [r0], r1
vld1.u8 {d8, d9, d10}, [r0], r1
vld1.u8 {d11, d12, d13}, [r0], r1
pld [r0]
pld [r0, r1]
pld [r0, r1, lsl #1]
vmull.u8 q7, d2, d0 ;(src_ptr[0] * Filter[0])
vmull.u8 q8, d3, d0
vmull.u8 q9, d5, d0
vmull.u8 q10, d6, d0
vmull.u8 q11, d8, d0
vmull.u8 q12, d9, d0
vmull.u8 q13, d11, d0
vmull.u8 q14, d12, d0
vext.8 d2, d2, d3, #1 ;construct src_ptr[1]
vext.8 d5, d5, d6, #1
vext.8 d8, d8, d9, #1
vext.8 d11, d11, d12, #1
vmlal.u8 q7, d2, d1 ;(src_ptr[0] * Filter[1])
vmlal.u8 q9, d5, d1
vmlal.u8 q11, d8, d1
vmlal.u8 q13, d11, d1
vext.8 d3, d3, d4, #1
vext.8 d6, d6, d7, #1
vext.8 d9, d9, d10, #1
vext.8 d12, d12, d13, #1
vmlal.u8 q8, d3, d1 ;(src_ptr[0] * Filter[1])
vmlal.u8 q10, d6, d1
vmlal.u8 q12, d9, d1
vmlal.u8 q14, d12, d1
subs r2, r2, #1
vqrshrn.u16 d14, q7, #7 ;shift/round/saturate to u8
vqrshrn.u16 d15, q8, #7
vqrshrn.u16 d16, q9, #7
vqrshrn.u16 d17, q10, #7
vqrshrn.u16 d18, q11, #7
vqrshrn.u16 d19, q12, #7
vqrshrn.u16 d20, q13, #7
vst1.u8 {d14, d15}, [r3]! ;store result
vqrshrn.u16 d21, q14, #7
vst1.u8 {d16, d17}, [r3]!
vst1.u8 {d18, d19}, [r3]!
vst1.u8 {d20, d21}, [r3]!
bne vp8e_filt_blk2d_fpo16x16_loop_neon
b sub_pixel_variance16x16_neon
;---------------------
secondpass_bfilter16x16_only
;Second pass: 16x16
;secondpass_filter
sub sp, sp, #528 ;reserve space on stack for temporary storage
add r3, r12, r3, lsl #3
mov r12, #4 ;loop counter
vld1.u32 {d31}, [r3] ;load second_pass filter
vld1.u8 {d22, d23}, [r0], r1 ;load src data
mov r3, sp
vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1)
vdup.8 d1, d31[4]
vp8e_filt_blk2d_spo16x16_loop_neon
vld1.u8 {d24, d25}, [r0], r1
vmull.u8 q1, d22, d0 ;(src_ptr[0] * Filter[0])
vld1.u8 {d26, d27}, [r0], r1
vmull.u8 q2, d23, d0
vld1.u8 {d28, d29}, [r0], r1
vmull.u8 q3, d24, d0
vld1.u8 {d30, d31}, [r0], r1
vmull.u8 q4, d25, d0
vmull.u8 q5, d26, d0
vmull.u8 q6, d27, d0
vmull.u8 q7, d28, d0
vmull.u8 q8, d29, d0
vmlal.u8 q1, d24, d1 ;(src_ptr[pixel_step] * Filter[1])
vmlal.u8 q2, d25, d1
vmlal.u8 q3, d26, d1
vmlal.u8 q4, d27, d1
vmlal.u8 q5, d28, d1
vmlal.u8 q6, d29, d1
vmlal.u8 q7, d30, d1
vmlal.u8 q8, d31, d1
vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8
vqrshrn.u16 d3, q2, #7
vqrshrn.u16 d4, q3, #7
vqrshrn.u16 d5, q4, #7
vqrshrn.u16 d6, q5, #7
vqrshrn.u16 d7, q6, #7
vqrshrn.u16 d8, q7, #7
vqrshrn.u16 d9, q8, #7
vst1.u8 {d2, d3}, [r3]! ;store result
subs r12, r12, #1
vst1.u8 {d4, d5}, [r3]!
vmov q11, q15
vst1.u8 {d6, d7}, [r3]!
vst1.u8 {d8, d9}, [r3]!
bne vp8e_filt_blk2d_spo16x16_loop_neon
b sub_pixel_variance16x16_neon
;----------------------------
;variance16x16
sub_pixel_variance16x16_neon
vmov.i8 q8, #0 ;q8 - sum
vmov.i8 q9, #0 ;q9, q10 - sse
vmov.i8 q10, #0
sub r3, r3, #256
mov r12, #8
sub_pixel_variance16x16_neon_loop
vld1.8 {q0}, [r3]! ;Load up source and reference
vld1.8 {q2}, [r4], r5
vld1.8 {q1}, [r3]!
vld1.8 {q3}, [r4], r5
vsubl.u8 q11, d0, d4 ;diff
vsubl.u8 q12, d1, d5
vsubl.u8 q13, d2, d6
vsubl.u8 q14, d3, d7
vpadal.s16 q8, q11 ;sum
vmlal.s16 q9, d22, d22 ;sse
vmlal.s16 q10, d23, d23
subs r12, r12, #1
vpadal.s16 q8, q12
vmlal.s16 q9, d24, d24
vmlal.s16 q10, d25, d25
vpadal.s16 q8, q13
vmlal.s16 q9, d26, d26
vmlal.s16 q10, d27, d27
vpadal.s16 q8, q14
vmlal.s16 q9, d28, d28
vmlal.s16 q10, d29, d29
bne sub_pixel_variance16x16_neon_loop
vadd.u32 q10, q9, q10 ;accumulate sse
vpaddl.s32 q0, q8 ;accumulate sum
vpaddl.u32 q1, q10
vadd.s64 d0, d0, d1
vadd.u64 d1, d2, d3
vmull.s32 q5, d0, d0
vst1.32 {d1[0]}, [r6] ;store sse
Fix variance overflow In the variance calculations the difference is summed and later squared. When the sum exceeds sqrt(2^31) the value is treated as a negative when it is shifted which gives incorrect results. To fix this we cast the result of the multiplication as unsigned. The alternative fix is to shift sum down by 4 before multiplying. However that will reduce precision. For 16x16 blocks the maximum sum is 65280 and sqrt(2^31) is 46340 (and change). PPC change is untested. Change-Id: I1bad27ea0720067def6d71a6da5f789508cec265
2012-02-09 21:38:31 +01:00
vshr.u32 d10, d10, #8
vsub.u32 d0, d1, d10
Initial WebM release
2010-05-18 17:58:33 +02:00
add sp, sp, #528
vmov.32 r0, d0[0] ;return
pop {r4-r6,pc}
ENDP
END
Reference in New Issue Copy Permalink