ref: a8eee97b43b532c9ee35280bfdf910ab468004a4
parent: a7fa1314da6a4945523e126319fd3cbe9d2082a9
author: Linfeng Zhang <linfengz@google.com>
date: Thu Oct 6 13:01:30 EDT 2016
Check in vpx_lpf_vertical_4_dual_neon() assembly This replaces its C version. Change-Id: Ie39e9324305fdc0fff610ced608a037e44a85a1a
--- a/vpx_dsp/arm/loopfilter_4_neon.asm
+++ b/vpx_dsp/arm/loopfilter_4_neon.asm
@@ -11,6 +11,7 @@
EXPORT |vpx_lpf_horizontal_4_neon|
EXPORT |vpx_lpf_vertical_4_neon|
EXPORT |vpx_lpf_horizontal_4_dual_neon|
+ EXPORT |vpx_lpf_vertical_4_dual_neon|
ARM
AREA ||.text||, CODE, READONLY, ALIGN=2
@@ -54,7 +55,7 @@
sub r2, r2, r1, lsl #1
sub r3, r3, r1, lsl #1
- bl vpx_loop_filter_neon
+ bl filter4_8
vst1.u8 {d4}, [r2@64], r1 ; store op1 vst1.u8 {d5}, [r3@64], r1 ; store op0@@ -114,7 +115,7 @@
vtrn.8 d7, d16
vtrn.8 d17, d18
- bl vpx_loop_filter_neon
+ bl filter4_8
sub r0, r0, #2
@@ -131,7 +132,7 @@
pop {pc}ENDP ; |vpx_lpf_vertical_4_neon|
-; void vpx_loop_filter_neon();
+; void filter4_8();
; This is a helper function for the loopfilters. The invidual functions do the
; necessary load, transpose (if necessary) and store. The function does not use
; registers d8-d15.
@@ -155,7 +156,7 @@
; d5 op0
; d6 oq0
; d7 oq1
-|vpx_loop_filter_neon| PROC
+|filter4_8| PROC
; filter_mask
vabd.u8 d19, d3, d4 ; m1 = abs(p3 - p2)
vabd.u8 d20, d4, d5 ; m2 = abs(p2 - p1)
@@ -245,7 +246,7 @@
veor d7, d20, d18 ; *oq1 = u^0x80
bx lr
- ENDP ; |vpx_loop_filter_neon|
+ ENDP ; |filter4_8|
;void vpx_lpf_horizontal_4_dual_neon(uint8_t *s, int p,
; const uint8_t *blimit0,
@@ -300,7 +301,7 @@
sub r2, r2, r1, lsl #1
sub r3, r3, r1, lsl #1
- bl vpx_loop_filter_neon_16
+ bl filter4_16
vst1.u8 {q5}, [r2@64], r1 ; store op1 vst1.u8 {q6}, [r3@64], r1 ; store op0@@ -312,7 +313,122 @@
pop {pc}ENDP ; |vpx_lpf_horizontal_4_dual_neon|
-; void vpx_loop_filter_neon_16();
+;void vpx_lpf_vertical_4_dual_neon(uint8_t *s, int p,
+; const uint8_t *blimit0,
+; const uint8_t *limit0,
+; const uint8_t *thresh0,
+; const uint8_t *blimit1,
+; const uint8_t *limit1,
+; const uint8_t *thresh1)
+; r0 uint8_t *s,
+; r1 int p,
+; r2 const uint8_t *blimit0,
+; r3 const uint8_t *limit0,
+; sp const uint8_t *thresh0,
+; sp+4 const uint8_t *blimit1,
+; sp+8 const uint8_t *limit1,
+; sp+12 const uint8_t *thresh1,
+
+|vpx_lpf_vertical_4_dual_neon| PROC
+ push {lr}+
+ ldr r12, [sp, #4] ; load thresh0
+ vld1.8 {d0}, [r2] ; load blimit0 to first half q+ vld1.8 {d2}, [r3] ; load limit0 to first half q+
+ ldr r2, [sp, #8] ; load blimit1
+
+ vld1.8 {d4}, [r12] ; load thresh0 to first half q+
+ ldr r3, [sp, #12] ; load limit1
+ ldr r12, [sp, #16] ; load thresh1
+ vld1.8 {d1}, [r2] ; load blimit1 to 2nd half q+
+ sub r2, r0, #4 ; s[-4]
+
+ vld1.8 {d3}, [r3] ; load limit1 to 2nd half q+ vld1.8 {d5}, [r12] ; load thresh1 to 2nd half q+
+ vpush {d8-d15} ; save neon registers+
+ vld1.u8 {d6}, [r2], r1 ; 00 01 02 03 04 05 06 07+ vld1.u8 {d8}, [r2], r1 ; 10 11 12 13 14 15 16 17+ vld1.u8 {d10}, [r2], r1 ; 20 21 22 23 24 25 26 27+ vld1.u8 {d12}, [r2], r1 ; 30 31 32 33 34 35 36 37+ vld1.u8 {d14}, [r2], r1 ; 40 41 42 43 44 45 46 47+ vld1.u8 {d16}, [r2], r1 ; 50 51 52 53 54 55 56 57+ vld1.u8 {d18}, [r2], r1 ; 60 61 62 63 64 65 66 67+ vld1.u8 {d20}, [r2], r1 ; 70 71 72 73 74 75 76 77+ vld1.u8 {d7}, [r2], r1 ; 80 81 82 83 84 85 86 87+ vld1.u8 {d9}, [r2], r1 ; 90 91 92 93 94 95 96 97+ vld1.u8 {d11}, [r2], r1 ; A0 A1 A2 A3 A4 A5 A6 A7+ vld1.u8 {d13}, [r2], r1 ; B0 B1 B2 B3 B4 B5 B6 B7+ vld1.u8 {d15}, [r2], r1 ; C0 C1 C2 C3 C4 C5 C6 C7+ vld1.u8 {d17}, [r2], r1 ; D0 D1 D2 D3 D4 D5 D6 D7+ vld1.u8 {d19}, [r2], r1 ; E0 E1 E2 E3 E4 E5 E6 E7+ vld1.u8 {d21}, [r2] ; F0 F1 F2 F3 F4 F5 F6 F7+
+ vtrn.8 q3, q4 ; q3 : 00 10 02 12 04 14 06 16 80 90 82 92 84 94 86 96
+ ; q4 : 01 11 03 13 05 15 07 17 81 91 83 93 85 95 87 97
+ vtrn.8 q5, q6 ; q5 : 20 30 22 32 24 34 26 36 A0 B0 A2 B2 A4 B4 A6 B6
+ ; q6 : 21 31 23 33 25 35 27 37 A1 B1 A3 B3 A5 B5 A7 B7
+ vtrn.8 q7, q8 ; q7 : 40 50 42 52 44 54 46 56 C0 D0 C2 D2 C4 D4 C6 D6
+ ; q8 : 41 51 43 53 45 55 47 57 C1 D1 C3 D3 C5 D5 C7 D7
+ vtrn.8 q9, q10 ; q9 : 60 70 62 72 64 74 66 76 E0 F0 E2 F2 E4 F4 E6 F6
+ ; q10: 61 71 63 73 65 75 67 77 E1 F1 E3 F3 E5 F5 E7 F7
+
+ vtrn.16 q3, q5 ; q3 : 00 10 20 30 04 14 24 34 80 90 A0 B0 84 94 A4 B4
+ ; q5 : 02 12 22 32 06 16 26 36 82 92 A2 B2 86 96 A6 B6
+ vtrn.16 q4, q6 ; q4 : 01 11 21 31 05 15 25 35 81 91 A1 B1 85 95 A5 B5
+ ; q6 : 03 13 23 33 07 17 27 37 83 93 A3 B3 87 97 A7 B7
+ vtrn.16 q7, q9 ; q7 : 40 50 60 70 44 54 64 74 C0 D0 E0 F0 C4 D4 E4 F4
+ ; q9 : 42 52 62 72 46 56 66 76 C2 D2 E2 F2 C6 D6 E6 F6
+ vtrn.16 q8, q10 ; q8 : 41 51 61 71 45 55 65 75 C1 D1 E1 F1 C5 D5 E5 F5
+ ; q10: 43 53 63 73 47 57 67 77 C3 D3 E3 F3 C7 D7 E7 F7
+
+ vtrn.32 q3, q7 ; q3 : 00 10 20 30 40 50 60 70 80 90 A0 B0 C0 D0 E0 F0
+ ; q7 : 04 14 24 34 44 54 64 74 84 94 A4 B4 C4 D4 E4 F4
+ vtrn.32 q5, q9 ; q5 : 02 12 22 32 42 52 62 72 82 92 A2 B2 C2 D2 E2 F2
+ ; q9 : 06 16 26 36 46 56 66 76 86 96 A6 B6 C6 D6 E6 F6
+ vtrn.32 q4, q8 ; q4 : 01 11 21 31 41 51 61 71 81 91 A1 B1 C1 D1 E1 F1
+ ; q8 : 05 15 25 35 45 55 65 75 85 95 A5 B5 C5 D5 E5 F5
+ vtrn.32 q6, q10 ; q6 : 03 13 23 33 43 53 63 73 83 93 A3 B3 C3 D3 E3 F3
+ ; q10: 07 17 27 37 47 57 67 77 87 97 A7 B7 C7 D7 E7 F7
+
+ bl filter4_16
+
+ sub r0, #2
+
+ vmov d0, d11
+ vmov d1, d13
+ vmov d2, d15
+ vmov d3, d17
+ vmov d11, d12
+ vmov d12, d14
+ vmov d13, d16
+ vst4.8 {d10[0], d11[0], d12[0], d13[0]}, [r0], r1+ vst4.8 {d10[1], d11[1], d12[1], d13[1]}, [r0], r1+ vst4.8 {d10[2], d11[2], d12[2], d13[2]}, [r0], r1+ vst4.8 {d10[3], d11[3], d12[3], d13[3]}, [r0], r1+ vst4.8 {d10[4], d11[4], d12[4], d13[4]}, [r0], r1+ vst4.8 {d10[5], d11[5], d12[5], d13[5]}, [r0], r1+ vst4.8 {d10[6], d11[6], d12[6], d13[6]}, [r0], r1+ vst4.8 {d10[7], d11[7], d12[7], d13[7]}, [r0], r1+ vst4.8 {d0[0], d1[0], d2[0], d3[0]}, [r0], r1+ vst4.8 {d0[1], d1[1], d2[1], d3[1]}, [r0], r1+ vst4.8 {d0[2], d1[2], d2[2], d3[2]}, [r0], r1+ vst4.8 {d0[3], d1[3], d2[3], d3[3]}, [r0], r1+ vst4.8 {d0[4], d1[4], d2[4], d3[4]}, [r0], r1+ vst4.8 {d0[5], d1[5], d2[5], d3[5]}, [r0], r1+ vst4.8 {d0[6], d1[6], d2[6], d3[6]}, [r0], r1+ vst4.8 {d0[7], d1[7], d2[7], d3[7]}, [r0]+
+ vpop {d8-d15} ; restore neon registers+
+ pop {pc}+ ENDP ; |vpx_lpf_vertical_4_dual_neon|
+
+; void filter4_16();
; This is a helper function for the loopfilters. The invidual functions do the
; necessary load, transpose (if necessary) and store. This function uses
; registers d8-d15, so the calling function must save those registers.
@@ -335,7 +451,7 @@
; q6 op0
; q7 oq0
; q8 oq1
-|vpx_loop_filter_neon_16| PROC
+|filter4_16| PROC
; filter_mask
vabd.u8 q11, q3, q4 ; m1 = abs(p3 - p2)
@@ -428,6 +544,6 @@
veor q8, q12, q10 ; *oq1 = u^0x80
bx lr
- ENDP ; |vpx_loop_filter_neon_16|
+ ENDP ; |filter4_16|
END
--- a/vpx_dsp/arm/loopfilter_vertical_4_dual_neon.c
+++ /dev/null
@@ -1,23 +1,0 @@
-/*
- * Copyright (c) 2014 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.
- */
-
-#include <arm_neon.h>
-
-#include "./vpx_dsp_rtcd.h"
-#include "./vpx_config.h"
-#include "vpx/vpx_integer.h"
-
-void vpx_lpf_vertical_4_dual_neon(uint8_t *s, int p, const uint8_t *blimit0,
- const uint8_t *limit0, const uint8_t *thresh0,
- const uint8_t *blimit1, const uint8_t *limit1,
- const uint8_t *thresh1) {- vpx_lpf_vertical_4_neon(s, p, blimit0, limit0, thresh0);
- vpx_lpf_vertical_4_neon(s + 8 * p, p, blimit1, limit1, thresh1);
-}
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -141,14 +141,11 @@
DSP_SRCS-$(HAVE_AVX2) += x86/loopfilter_avx2.c
ifeq ($(HAVE_NEON_ASM),yes)
-DSP_SRCS-yes += arm/loopfilter_vertical_4_dual_neon.c
DSP_SRCS-yes += arm/loopfilter_16_neon$(ASM)
DSP_SRCS-yes += arm/loopfilter_8_neon$(ASM)
DSP_SRCS-yes += arm/loopfilter_4_neon$(ASM)
else
-ifeq ($(HAVE_NEON),yes)
-DSP_SRCS-yes += arm/loopfilter_neon.c
-endif # HAVE_NEON
+DSP_SRCS-$(HAVE_NEON) += arm/loopfilter_neon.c
endif # HAVE_NEON_ASM
DSP_SRCS-$(HAVE_MSA) += mips/loopfilter_msa.h