Browse code
VP8: armv6 optimizations.
From 52.503s (~40fps) to 27.973sec (~80fps) decoding of 480p sintel
trailer, i.e. a ~2x speedup overall, on a Nexus S.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Ronald S. Bultje authored on 2011/08/25 05:58:37Showing 5 changed files
- libavcodec/arm/Makefile
- libavcodec/arm/asm.S
- libavcodec/arm/vp8dsp_armv6.S
- libavcodec/arm/vp8dsp_init_arm.c
- libavcodec/arm/vp8dsp_neon.S
| ... | ... |
@@ -11,7 +11,8 @@ ARMV6-OBJS-$(CONFIG_MPEGAUDIODSP) += arm/mpegaudiodsp_fixed_armv6.o |
| 11 | 11 |
OBJS-$(CONFIG_VP5_DECODER) += arm/vp56dsp_init_arm.o |
| 12 | 12 |
OBJS-$(CONFIG_VP6_DECODER) += arm/vp56dsp_init_arm.o |
| 13 | 13 |
OBJS-$(CONFIG_VP8_DECODER) += arm/vp8dsp_init_arm.o |
| 14 |
-ARMV6-OBJS-$(CONFIG_VP8_DECODER) += arm/vp8_armv6.o |
|
| 14 |
+ARMV6-OBJS-$(CONFIG_VP8_DECODER) += arm/vp8_armv6.o \ |
|
| 15 |
+ arm/vp8dsp_armv6.o |
|
| 15 | 16 |
|
| 16 | 17 |
OBJS-$(CONFIG_H264DSP) += arm/h264dsp_init_arm.o |
| 17 | 18 |
OBJS-$(CONFIG_H264PRED) += arm/h264pred_init_arm.o |
| ... | ... |
@@ -97,6 +97,12 @@ T add \rn, \rn, \rm |
| 97 | 97 |
T ldr \rt, [\rn] |
| 98 | 98 |
.endm |
| 99 | 99 |
|
| 100 |
+.macro ldr_dpren rt, rn, rm:vararg |
|
| 101 |
+A ldr \rt, [\rn, -\rm] |
|
| 102 |
+T sub \rt, \rn, \rm |
|
| 103 |
+T ldr \rt, [\rt] |
|
| 104 |
+.endm |
|
| 105 |
+ |
|
| 100 | 106 |
.macro ldr_post rt, rn, rm:vararg |
| 101 | 107 |
A ldr \rt, [\rn], \rm |
| 102 | 108 |
T ldr \rt, [\rn] |
| ... | ... |
@@ -133,6 +139,12 @@ T ldrh \rt, [\rn] |
| 133 | 133 |
T add \rn, \rn, \rm |
| 134 | 134 |
.endm |
| 135 | 135 |
|
| 136 |
+.macro ldrb_post rt, rn, rm |
|
| 137 |
+A ldrb \rt, [\rn], \rm |
|
| 138 |
+T ldrb \rt, [\rn] |
|
| 139 |
+T add \rn, \rn, \rm |
|
| 140 |
+.endm |
|
| 141 |
+ |
|
| 136 | 142 |
.macro str_post rt, rn, rm:vararg |
| 137 | 143 |
A str \rt, [\rn], \rm |
| 138 | 144 |
T str \rt, [\rn] |
| 139 | 145 |
new file mode 100644 |
| ... | ... |
@@ -0,0 +1,2328 @@ |
| 0 |
+/** |
|
| 1 |
+ * VP8 ARMv6 optimisations |
|
| 2 |
+ * |
|
| 3 |
+ * Copyright (c) 2011 The WebM project authors. All Rights Reserved. |
|
| 4 |
+ * Copyright (c) 2010 Rob Clark <rob@ti.com> |
|
| 5 |
+ * Copyright (c) 2011 Mans Rullgard <mans@mansr.com> |
|
| 6 |
+ * |
|
| 7 |
+ * This file is part of Libav. |
|
| 8 |
+ * |
|
| 9 |
+ * Libav is free software; you can redistribute it and/or |
|
| 10 |
+ * modify it under the terms of the GNU Lesser General Public |
|
| 11 |
+ * License as published by the Free Software Foundation; either |
|
| 12 |
+ * version 2.1 of the License, or (at your option) any later version. |
|
| 13 |
+ * |
|
| 14 |
+ * Libav is distributed in the hope that it will be useful, |
|
| 15 |
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of |
|
| 16 |
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
|
| 17 |
+ * Lesser General Public License for more details. |
|
| 18 |
+ * |
|
| 19 |
+ * You should have received a copy of the GNU Lesser General Public |
|
| 20 |
+ * License along with Libav; if not, write to the Free Software |
|
| 21 |
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
|
| 22 |
+ * |
|
| 23 |
+ * This code was partially ported from libvpx, which uses this license: |
|
| 24 |
+ * |
|
| 25 |
+ * Use of this source code is governed by a BSD-style license |
|
| 26 |
+ * that can be found in the LICENSE file in the root of the source |
|
| 27 |
+ * tree. An additional intellectual property rights grant can be found |
|
| 28 |
+ * in the file PATENTS. All contributing project authors may |
|
| 29 |
+ * be found in the AUTHORS file in the root of the source tree. |
|
| 30 |
+ * |
|
| 31 |
+ * (Note that the "LICENSE", "AUTHORS" and "PATENTS" files can be |
|
| 32 |
+ * found in the libvpx source tree.) |
|
| 33 |
+ */ |
|
| 34 |
+ |
|
| 35 |
+#include "asm.S" |
|
| 36 |
+ |
|
| 37 |
+@ idct |
|
| 38 |
+ |
|
| 39 |
+@ void vp8_luma_dc_wht(DCTELEM block[4][4][16], DCTELEM dc[16]) |
|
| 40 |
+function ff_vp8_luma_dc_wht_armv6, export=1 |
|
| 41 |
+ push {r4 - r10, lr}
|
|
| 42 |
+ |
|
| 43 |
+ @ load dc[] and zero memory |
|
| 44 |
+ mov r12, #0 |
|
| 45 |
+ ldr r2, [r1] @ dc0[0,1] |
|
| 46 |
+ ldr r3, [r1, #4] @ dc0[2,3] |
|
| 47 |
+ ldr r4, [r1, #8] @ dc1[0,1] |
|
| 48 |
+ ldr r5, [r1, #12] @ dc1[2,3] |
|
| 49 |
+ ldr r6, [r1, #16] @ dc2[0,1] |
|
| 50 |
+ ldr r7, [r1, #20] @ dc2[2,3] |
|
| 51 |
+ ldr r8, [r1, #24] @ dc3[0,1] |
|
| 52 |
+ ldr r9, [r1, #28] @ dc3[2,3] |
|
| 53 |
+ str r12,[r1] |
|
| 54 |
+ str r12,[r1, #4] |
|
| 55 |
+ str r12,[r1, #8] |
|
| 56 |
+ str r12,[r1, #12] |
|
| 57 |
+ str r12,[r1, #16] |
|
| 58 |
+ str r12,[r1, #20] |
|
| 59 |
+ str r12,[r1, #24] |
|
| 60 |
+ str r12,[r1, #28] |
|
| 61 |
+ |
|
| 62 |
+ @ loop1 |
|
| 63 |
+ uadd16 r12, r2, r8 @ t0[0,1] |
|
| 64 |
+ uadd16 r14, r3, r9 @ t0[2,3] |
|
| 65 |
+ usub16 r2, r2, r8 @ t3[0,1] |
|
| 66 |
+ usub16 r3, r3, r9 @ t3[2,3] |
|
| 67 |
+ uadd16 r8, r4, r6 @ t1[0,1] |
|
| 68 |
+ uadd16 r9, r5, r7 @ t1[2,3] |
|
| 69 |
+ usub16 r4, r4, r6 @ t2[0,1] |
|
| 70 |
+ usub16 r5, r5, r7 @ t2[2,3] |
|
| 71 |
+ |
|
| 72 |
+ uadd16 r6, r12, r8 @ dc0[0,1] |
|
| 73 |
+ uadd16 r7, r14, r9 @ dc0[2,3] |
|
| 74 |
+ usub16 r12, r12, r8 @ dc2[0,1] |
|
| 75 |
+ usub16 r14, r14, r9 @ dc2[2,3] |
|
| 76 |
+ uadd16 r8, r2, r4 @ dc1[0,1] |
|
| 77 |
+ uadd16 r9, r3, r5 @ dc1[2,3] |
|
| 78 |
+ usub16 r2, r2, r4 @ dc3[0,1] |
|
| 79 |
+ usub16 r3, r3, r5 @ dc3[2,3] |
|
| 80 |
+ |
|
| 81 |
+ mov r1, #3 |
|
| 82 |
+ orr r1, r1, #0x30000 @ 3 | 3 (round) |
|
| 83 |
+ |
|
| 84 |
+ @ "transpose" |
|
| 85 |
+ pkhbt r4, r6, r8, lsl #16 @ dc{0,1}[0]
|
|
| 86 |
+ pkhtb r6, r8, r6, asr #16 @ dc{0,1}[1]
|
|
| 87 |
+ pkhbt r5, r12, r2, lsl #16 @ dc{2,3}[0]
|
|
| 88 |
+ pkhtb r12, r2, r12, asr #16 @ dc{2,3}[1]
|
|
| 89 |
+ pkhbt r8, r7, r9, lsl #16 @ dc{0,1}[2]
|
|
| 90 |
+ uadd16 r4, r4, r1 |
|
| 91 |
+ uadd16 r5, r5, r1 |
|
| 92 |
+ pkhtb r7, r9, r7, asr #16 @ dc{0,1}[3]
|
|
| 93 |
+ pkhbt r2, r14, r3, lsl #16 @ dc{2,3}[2]
|
|
| 94 |
+ pkhtb r14, r3, r14, asr #16 @ dc{2,3}[3]
|
|
| 95 |
+ |
|
| 96 |
+ @ loop2 |
|
| 97 |
+ uadd16 r9, r4, r7 @ t0[0,1] |
|
| 98 |
+ uadd16 r3, r5, r14 @ t0[2,3] |
|
| 99 |
+ usub16 r4, r4, r7 @ t3[0,1] |
|
| 100 |
+ usub16 r5, r5, r14 @ t3[2,3] |
|
| 101 |
+ uadd16 r7, r6, r8 @ t1[0,1] |
|
| 102 |
+ uadd16 r14, r12, r2 @ t1[2,3] |
|
| 103 |
+ usub16 r6, r6, r8 @ t2[0,1] |
|
| 104 |
+ usub16 r12, r12, r2 @ t2[2,3] |
|
| 105 |
+ |
|
| 106 |
+ uadd16 r8, r9, r7 @ block[0,1][0] |
|
| 107 |
+ uadd16 r2, r3, r14 @ block[2,3][0] |
|
| 108 |
+ usub16 r9, r9, r7 @ block[0,1][2] |
|
| 109 |
+ usub16 r3, r3, r14 @ block[2,3][2] |
|
| 110 |
+ uadd16 r7, r4, r6 @ block[0,1][1] |
|
| 111 |
+ uadd16 r14, r5, r12 @ block[2,3][1] |
|
| 112 |
+ usub16 r4, r4, r6 @ block[0,1][3] |
|
| 113 |
+ usub16 r5, r5, r12 @ block[2,3][3] |
|
| 114 |
+ |
|
| 115 |
+ @ store |
|
| 116 |
+ mov r6, r8, asr #19 @ block[1][0] |
|
| 117 |
+ mov r12, r7, asr #19 @ block[1][1] |
|
| 118 |
+ mov r1, r9, asr #19 @ block[1][2] |
|
| 119 |
+ mov r10, r4, asr #19 @ block[1][3] |
|
| 120 |
+ sxth r8, r8 |
|
| 121 |
+ sxth r7, r7 |
|
| 122 |
+ sxth r9, r9 |
|
| 123 |
+ sxth r4, r4 |
|
| 124 |
+ asr r8, #3 @ block[0][0] |
|
| 125 |
+ asr r7, #3 @ block[0][1] |
|
| 126 |
+ asr r9, #3 @ block[0][2] |
|
| 127 |
+ asr r4, #3 @ block[0][3] |
|
| 128 |
+ |
|
| 129 |
+ strh r8, [r0], #32 |
|
| 130 |
+ strh r7, [r0], #32 |
|
| 131 |
+ strh r9, [r0], #32 |
|
| 132 |
+ strh r4, [r0], #32 |
|
| 133 |
+ strh r6, [r0], #32 |
|
| 134 |
+ strh r12,[r0], #32 |
|
| 135 |
+ strh r1, [r0], #32 |
|
| 136 |
+ strh r10,[r0], #32 |
|
| 137 |
+ |
|
| 138 |
+ mov r6, r2, asr #19 @ block[3][0] |
|
| 139 |
+ mov r12, r14, asr #19 @ block[3][1] |
|
| 140 |
+ mov r1, r3, asr #19 @ block[3][2] |
|
| 141 |
+ mov r10, r5, asr #19 @ block[3][3] |
|
| 142 |
+ sxth r2, r2 |
|
| 143 |
+ sxth r14, r14 |
|
| 144 |
+ sxth r3, r3 |
|
| 145 |
+ sxth r5, r5 |
|
| 146 |
+ asr r2, #3 @ block[2][0] |
|
| 147 |
+ asr r14, #3 @ block[2][1] |
|
| 148 |
+ asr r3, #3 @ block[2][2] |
|
| 149 |
+ asr r5, #3 @ block[2][3] |
|
| 150 |
+ |
|
| 151 |
+ strh r2, [r0], #32 |
|
| 152 |
+ strh r14,[r0], #32 |
|
| 153 |
+ strh r3, [r0], #32 |
|
| 154 |
+ strh r5, [r0], #32 |
|
| 155 |
+ strh r6, [r0], #32 |
|
| 156 |
+ strh r12,[r0], #32 |
|
| 157 |
+ strh r1, [r0], #32 |
|
| 158 |
+ strh r10,[r0], #32 |
|
| 159 |
+ |
|
| 160 |
+ pop {r4 - r10, pc}
|
|
| 161 |
+endfunc |
|
| 162 |
+ |
|
| 163 |
+@ void vp8_luma_dc_wht_dc(DCTELEM block[4][4][16], DCTELEM dc[16]) |
|
| 164 |
+function ff_vp8_luma_dc_wht_dc_armv6, export=1 |
|
| 165 |
+ ldrsh r2, [r1] |
|
| 166 |
+ mov r3, #0 |
|
| 167 |
+ add r2, r2, #3 |
|
| 168 |
+ strh r3, [r1] |
|
| 169 |
+ asr r2, r2, #3 |
|
| 170 |
+ .rept 16 |
|
| 171 |
+ strh r2, [r0], #32 |
|
| 172 |
+ .endr |
|
| 173 |
+ bx lr |
|
| 174 |
+endfunc |
|
| 175 |
+ |
|
| 176 |
+@ void vp8_idct_add(uint8_t *dst, DCTELEM block[16], int stride) |
|
| 177 |
+function ff_vp8_idct_add_armv6, export=1 |
|
| 178 |
+ push {r4 - r11, lr}
|
|
| 179 |
+ sub sp, sp, #32 |
|
| 180 |
+ |
|
| 181 |
+ mov r3, #0x00004E00 @ cos |
|
| 182 |
+ orr r3, r3, #0x0000007B @ cospi8sqrt2minus1 = 20091 |
|
| 183 |
+ mov r4, #0x00008A00 @ sin |
|
| 184 |
+ orr r4, r4, #0x0000008C @ sinpi8sqrt2 = 35468 |
|
| 185 |
+ mov r5, #0x2 @ i=2 |
|
| 186 |
+1: |
|
| 187 |
+ ldr r6, [r1, #8] @ i5 | i4 = block1[1] | block1[0] |
|
| 188 |
+ ldr r12,[r1, #24] @ i13 | i12 = block3[1] | block3[0] |
|
| 189 |
+ ldr r14,[r1, #16] @ i9 | i8 = block2[1] | block2[0] |
|
| 190 |
+ |
|
| 191 |
+ smulwt r9, r3, r6 @ (ip[5] * cospi8sqrt2minus1) >> 16 |
|
| 192 |
+ smulwb r7, r3, r6 @ (ip[4] * cospi8sqrt2minus1) >> 16 |
|
| 193 |
+ smulwt r10, r4, r6 @ (ip[5] * sinpi8sqrt2) >> 16 |
|
| 194 |
+ smulwb r8, r4, r6 @ (ip[4] * sinpi8sqrt2) >> 16 |
|
| 195 |
+ pkhbt r7, r7, r9, lsl #16 @ 5c | 4c |
|
| 196 |
+ smulwt r11, r3, r12 @ (ip[13] * cospi8sqrt2minus1) >> 16 |
|
| 197 |
+ pkhbt r8, r8, r10, lsl #16 @ 5s | 4s = t2 first half |
|
| 198 |
+ uadd16 r6, r6, r7 @ 5c+5 | 4c+4 = t3 first half |
|
| 199 |
+ smulwt r7, r4, r12 @ (ip[13] * sinpi8sqrt2) >> 16 |
|
| 200 |
+ smulwb r9, r3, r12 @ (ip[12] * cospi8sqrt2minus1) >> 16 |
|
| 201 |
+ smulwb r10, r4, r12 @ (ip[12] * sinpi8sqrt2) >> 16 |
|
| 202 |
+ |
|
| 203 |
+ subs r5, r5, #1 @ i-- |
|
| 204 |
+ pkhbt r9, r9, r11, lsl #16 @ 13c | 12c |
|
| 205 |
+ ldr r11,[r1] @ i1 | i0 |
|
| 206 |
+ pkhbt r10, r10, r7, lsl #16 @ 13s | 12s = t3 second half |
|
| 207 |
+ uadd16 r7, r12, r9 @ 13c+13 | 12c+12 = t2 second half |
|
| 208 |
+ usub16 r7, r8, r7 @ c = t2 |
|
| 209 |
+ uadd16 r6, r6, r10 @ d = t3 |
|
| 210 |
+ uadd16 r10, r11, r14 @ a = t0 |
|
| 211 |
+ usub16 r8, r11, r14 @ b = t1 |
|
| 212 |
+ uadd16 r9, r10, r6 @ a+d = tmp{0,1}[0]
|
|
| 213 |
+ usub16 r10, r10, r6 @ a-d = tmp{0,1}[3]
|
|
| 214 |
+ uadd16 r6, r8, r7 @ b+c = tmp{0,1}[1]
|
|
| 215 |
+ usub16 r7, r8, r7 @ b-c = tmp{0,1}[2]
|
|
| 216 |
+ mov r8, #0 |
|
| 217 |
+ str r6, [sp, #8] @ o5 | o4 |
|
| 218 |
+ str r7, [sp, #16] @ o9 | o8 |
|
| 219 |
+ str r10,[sp, #24] @ o13 | o12 |
|
| 220 |
+ str r9, [sp], #4 @ o1 | o0 |
|
| 221 |
+ str r8, [r1, #24] |
|
| 222 |
+ str r8, [r1, #16] |
|
| 223 |
+ str r8, [r1, #8] |
|
| 224 |
+ str r8, [r1], #4 |
|
| 225 |
+ bne 1b |
|
| 226 |
+ |
|
| 227 |
+ mov r5, #0x2 @ i=2 |
|
| 228 |
+ sub sp, sp, #8 |
|
| 229 |
+2: |
|
| 230 |
+ ldr r6, [sp, #8] @ i5 | i4 = tmp{0,1}[1]
|
|
| 231 |
+ ldr r14,[sp, #4] @ i3 | i2 = tmp{2,3}[0]
|
|
| 232 |
+ ldr r12,[sp, #12] @ i7 | i6 = tmp{2,3}[1]
|
|
| 233 |
+ ldr r1, [sp], #16 @ i1 | i0 = tmp{0,1}[0]
|
|
| 234 |
+ smulwt r9, r3, r6 @ (ip[5] * cospi8sqrt2minus1) >> 16 |
|
| 235 |
+ smulwt r7, r3, r1 @ (ip[1] * cospi8sqrt2minus1) >> 16 |
|
| 236 |
+ smulwt r10, r4, r6 @ (ip[5] * sinpi8sqrt2) >> 16 |
|
| 237 |
+ smulwt r8, r4, r1 @ (ip[1] * sinpi8sqrt2) >> 16 |
|
| 238 |
+ pkhbt r11, r1, r6, lsl #16 @ i4 | i0 = t0/t1 first half |
|
| 239 |
+ pkhbt r7, r7, r9, lsl #16 @ 5c | 1c |
|
| 240 |
+ pkhbt r8, r8, r10, lsl #16 @ 5s | 1s = temp1 = t2 first half |
|
| 241 |
+ pkhtb r1, r6, r1, asr #16 @ i5 | i1 |
|
| 242 |
+ uadd16 r1, r7, r1 @ 5c+5 | 1c+1 = temp2 (d) = t3 first half |
|
| 243 |
+ pkhbt r9, r14, r12, lsl #16 @ i6 | i2 = t0/t1 second half |
|
| 244 |
+ uadd16 r10, r11, r9 @ a = t0 |
|
| 245 |
+ usub16 r9, r11, r9 @ b = t1 |
|
| 246 |
+ pkhtb r6, r12, r14, asr #16 @ i7 | i3 |
|
| 247 |
+ subs r5, r5, #0x1 @ i-- |
|
| 248 |
+ smulwt r7, r3, r6 @ (ip[7] * cospi8sqrt2minus1) >> 16 |
|
| 249 |
+ smulwt r11, r4, r6 @ (ip[7] * sinpi8sqrt2) >> 16 |
|
| 250 |
+ smulwb r12, r3, r6 @ (ip[3] * cospi8sqrt2minus1) >> 16 |
|
| 251 |
+ smulwb r14, r4, r6 @ (ip[3] * sinpi8sqrt2) >> 16 |
|
| 252 |
+ |
|
| 253 |
+ pkhbt r7, r12, r7, lsl #16 @ 7c | 3c |
|
| 254 |
+ pkhbt r11, r14, r11, lsl #16 @ 7s | 3s = temp1 (d) = t3 second half |
|
| 255 |
+ mov r14, #0x4 @ set up 4's |
|
| 256 |
+ orr r14, r14, #0x40000 @ 4|4 |
|
| 257 |
+ uadd16 r6, r7, r6 @ 7c+7 | 3c+3 = temp2 (c) = t2 second half |
|
| 258 |
+ usub16 r12, r8, r6 @ c (o5 | o1) = t2 |
|
| 259 |
+ uadd16 r6, r11, r1 @ d (o7 | o3) = t3 |
|
| 260 |
+ uadd16 r10, r10, r14 @ t0 + 4 |
|
| 261 |
+ uadd16 r9, r9, r14 @ t1 + 4 |
|
| 262 |
+ uadd16 r7, r10, r6 @ a+d = dst{0,1}[0]
|
|
| 263 |
+ usub16 r6, r10, r6 @ a-d = dst{0,1}[3]
|
|
| 264 |
+ uadd16 r10, r9, r12 @ b+c = dst{0,1}[1]
|
|
| 265 |
+ usub16 r1, r9, r12 @ b-c = dst{0,1}[2]
|
|
| 266 |
+ |
|
| 267 |
+ mov r9, r6, asr #3 @ o[1][3] |
|
| 268 |
+ mov r12, r1, asr #3 @ o[1][2] |
|
| 269 |
+ pkhtb r8, r12, r7, asr #19 @ o[1][0,2] |
|
| 270 |
+ pkhtb r11, r9, r10, asr #19 @ o[1][1,3] |
|
| 271 |
+ ldr r12,[r0] |
|
| 272 |
+ ldr r9, [r0, r2] |
|
| 273 |
+ sxth r7, r7 |
|
| 274 |
+ sxth r6, r6 |
|
| 275 |
+ sxth r10, r10 |
|
| 276 |
+ sxth r1, r1 |
|
| 277 |
+ asr r7, #3 @ o[0][0] |
|
| 278 |
+ asr r10, #3 @ o[0][1] |
|
| 279 |
+ pkhbt r7, r7, r1, lsl #13 @ o[0][0,2] |
|
| 280 |
+ pkhbt r10, r10, r6, lsl #13 @ o[0][1,3] |
|
| 281 |
+ |
|
| 282 |
+ uxtab16 r7, r7, r12 |
|
| 283 |
+ uxtab16 r10, r10, r12, ror #8 |
|
| 284 |
+ uxtab16 r8, r8, r9 |
|
| 285 |
+ uxtab16 r11, r11, r9, ror #8 |
|
| 286 |
+ usat16 r7, #8, r7 |
|
| 287 |
+ usat16 r10, #8, r10 |
|
| 288 |
+ usat16 r8, #8, r8 |
|
| 289 |
+ usat16 r11, #8, r11 |
|
| 290 |
+ orr r7, r7, r10, lsl #8 |
|
| 291 |
+ orr r8, r8, r11, lsl #8 |
|
| 292 |
+ str r8, [r0, r2] |
|
| 293 |
+ str_post r7, r0, r2, lsl #1 |
|
| 294 |
+ |
|
| 295 |
+ bne 2b |
|
| 296 |
+ |
|
| 297 |
+ pop {r4 - r11, pc}
|
|
| 298 |
+endfunc |
|
| 299 |
+ |
|
| 300 |
+@ void vp8_idct_dc_add(uint8_t *dst, DCTELEM block[16], int stride) |
|
| 301 |
+function ff_vp8_idct_dc_add_armv6, export=1 |
|
| 302 |
+ push {r4 - r5, lr}
|
|
| 303 |
+ ldrsh r3, [r1] |
|
| 304 |
+ mov r4, #0 |
|
| 305 |
+ add r3, r3, #4 |
|
| 306 |
+ asr r3, #3 |
|
| 307 |
+ strh r4, [r1], #32 |
|
| 308 |
+ ldr r4, [r0, r2] |
|
| 309 |
+ ldr_post r5, r0, r2, lsl #1 |
|
| 310 |
+ pkhbt r3, r3, r3, lsl #16 |
|
| 311 |
+ |
|
| 312 |
+ uxtab16 lr, r3, r5 @ a1+2 | a1+0 |
|
| 313 |
+ uxtab16 r5, r3, r5, ror #8 @ a1+3 | a1+1 |
|
| 314 |
+ uxtab16 r12, r3, r4 |
|
| 315 |
+ uxtab16 r4, r3, r4, ror #8 |
|
| 316 |
+ usat16 lr, #8, lr |
|
| 317 |
+ usat16 r5, #8, r5 |
|
| 318 |
+ usat16 r12, #8, r12 |
|
| 319 |
+ usat16 r4, #8, r4 |
|
| 320 |
+ orr lr, lr, r5, lsl #8 |
|
| 321 |
+ orr r12, r12, r4, lsl #8 |
|
| 322 |
+ ldr r5, [r0] |
|
| 323 |
+ ldr r4, [r0, r2] |
|
| 324 |
+ sub r0, r0, r2, lsl #1 |
|
| 325 |
+ str r12,[r0, r2] |
|
| 326 |
+ str_post lr, r0, r2, lsl #1 |
|
| 327 |
+ |
|
| 328 |
+ uxtab16 lr, r3, r5 |
|
| 329 |
+ uxtab16 r5, r3, r5, ror #8 |
|
| 330 |
+ uxtab16 r12, r3, r4 |
|
| 331 |
+ uxtab16 r4, r3, r4, ror #8 |
|
| 332 |
+ usat16 lr, #8, lr |
|
| 333 |
+ usat16 r5, #8, r5 |
|
| 334 |
+ usat16 r12, #8, r12 |
|
| 335 |
+ usat16 r4, #8, r4 |
|
| 336 |
+ orr lr, lr, r5, lsl #8 |
|
| 337 |
+ orr r12, r12, r4, lsl #8 |
|
| 338 |
+ |
|
| 339 |
+ str r12,[r0, r2] |
|
| 340 |
+ str_post lr, r0, r2, lsl #1 |
|
| 341 |
+ |
|
| 342 |
+ pop {r4 - r5, pc}
|
|
| 343 |
+endfunc |
|
| 344 |
+ |
|
| 345 |
+@ void vp8_idct_dc_add4uv(uint8_t *dst, DCTELEM block[4][16], int stride) |
|
| 346 |
+function ff_vp8_idct_dc_add4uv_armv6, export=1 |
|
| 347 |
+ push {lr}
|
|
| 348 |
+ |
|
| 349 |
+ bl ff_vp8_idct_dc_add_armv6 |
|
| 350 |
+ sub r0, r0, r2, lsl #2 |
|
| 351 |
+ add r0, r0, #4 |
|
| 352 |
+ bl ff_vp8_idct_dc_add_armv6 |
|
| 353 |
+ sub r0, r0, #4 |
|
| 354 |
+ bl ff_vp8_idct_dc_add_armv6 |
|
| 355 |
+ sub r0, r0, r2, lsl #2 |
|
| 356 |
+ add r0, r0, #4 |
|
| 357 |
+ bl ff_vp8_idct_dc_add_armv6 |
|
| 358 |
+ |
|
| 359 |
+ pop {pc}
|
|
| 360 |
+endfunc |
|
| 361 |
+ |
|
| 362 |
+@ void vp8_idct_dc_add4y(uint8_t *dst, DCTELEM block[4][16], int stride) |
|
| 363 |
+function ff_vp8_idct_dc_add4y_armv6, export=1 |
|
| 364 |
+ push {lr}
|
|
| 365 |
+ |
|
| 366 |
+ bl ff_vp8_idct_dc_add_armv6 |
|
| 367 |
+ sub r0, r0, r2, lsl #2 |
|
| 368 |
+ add r0, r0, #4 |
|
| 369 |
+ bl ff_vp8_idct_dc_add_armv6 |
|
| 370 |
+ sub r0, r0, r2, lsl #2 |
|
| 371 |
+ add r0, r0, #4 |
|
| 372 |
+ bl ff_vp8_idct_dc_add_armv6 |
|
| 373 |
+ sub r0, r0, r2, lsl #2 |
|
| 374 |
+ add r0, r0, #4 |
|
| 375 |
+ bl ff_vp8_idct_dc_add_armv6 |
|
| 376 |
+ |
|
| 377 |
+ pop {pc}
|
|
| 378 |
+endfunc |
|
| 379 |
+ |
|
| 380 |
+@ loopfilter |
|
| 381 |
+ |
|
| 382 |
+@ void vp8_v_loop_filter16_simple(uint8_t *dst, int stride, int flim) |
|
| 383 |
+function ff_vp8_v_loop_filter16_simple_armv6, export=1 |
|
| 384 |
+ push {r4 - r11, lr}
|
|
| 385 |
+ |
|
| 386 |
+ ldr_dpren r3, r0, r1, lsl #1 @ p1 |
|
| 387 |
+ ldr_dpren r4, r0, r1 @ p0 |
|
| 388 |
+ ldr r5, [r0] @ q0 |
|
| 389 |
+ ldr r6, [r0, r1] @ q1 |
|
| 390 |
+ orr r2, r2, r2, lsl #16 |
|
| 391 |
+ mov r9, #4 @ count |
|
| 392 |
+ mov lr, #0 @ need 0 in a couple places |
|
| 393 |
+ orr r12, r2, r2, lsl #8 @ splat int -> byte |
|
| 394 |
+ ldr r2, c0x80808080 |
|
| 395 |
+ |
|
| 396 |
+1: |
|
| 397 |
+ @ vp8_simple_filter_mask() |
|
| 398 |
+ uqsub8 r7, r3, r6 @ p1 - q1 |
|
| 399 |
+ uqsub8 r8, r6, r3 @ q1 - p1 |
|
| 400 |
+ uqsub8 r10, r4, r5 @ p0 - q0 |
|
| 401 |
+ uqsub8 r11, r5, r4 @ q0 - p0 |
|
| 402 |
+ orr r8, r8, r7 @ abs(p1 - q1) |
|
| 403 |
+ orr r10, r10, r11 @ abs(p0 - q0) |
|
| 404 |
+ uqadd8 r10, r10, r10 @ abs(p0 - q0) * 2 |
|
| 405 |
+ uhadd8 r8, r8, lr @ abs(p1 - q2) >> 1 |
|
| 406 |
+ uqadd8 r10, r10, r8 @ abs(p0 - q0)*2 + abs(p1 - q1)/2 |
|
| 407 |
+ mvn r8, #0 |
|
| 408 |
+ usub8 r10, r12, r10 @ compare to flimit. usub8 sets GE flags |
|
| 409 |
+ sel r10, r8, lr @ filter mask: F or 0 |
|
| 410 |
+ cmp r10, #0 |
|
| 411 |
+ beq 2f @ skip filtering if all masks are 0x00 |
|
| 412 |
+ |
|
| 413 |
+ @ vp8_simple_filter() |
|
| 414 |
+ eor r3, r3, r2 @ p1 offset to convert to a signed value |
|
| 415 |
+ eor r6, r6, r2 @ q1 offset to convert to a signed value |
|
| 416 |
+ eor r4, r4, r2 @ p0 offset to convert to a signed value |
|
| 417 |
+ eor r5, r5, r2 @ q0 offset to convert to a signed value |
|
| 418 |
+ |
|
| 419 |
+ qsub8 r3, r3, r6 @ vp8_filter = p1 - q1 |
|
| 420 |
+ qsub8 r6, r5, r4 @ q0 - p0 |
|
| 421 |
+ qadd8 r3, r3, r6 @ += q0 - p0 |
|
| 422 |
+ ldr r7, c0x04040404 |
|
| 423 |
+ qadd8 r3, r3, r6 @ += q0 - p0 |
|
| 424 |
+ ldr r8, c0x03030303 |
|
| 425 |
+ qadd8 r3, r3, r6 @ vp8_filter = p1-q1 + 3*(q0-p0)) |
|
| 426 |
+ @STALL |
|
| 427 |
+ and r3, r3, r10 @ vp8_filter &= mask |
|
| 428 |
+ |
|
| 429 |
+ qadd8 r7, r3, r7 @ Filter1 = vp8_filter + 4 |
|
| 430 |
+ qadd8 r8, r3, r8 @ Filter2 = vp8_filter + 3 |
|
| 431 |
+ |
|
| 432 |
+ shadd8 r7, r7, lr |
|
| 433 |
+ shadd8 r8, r8, lr |
|
| 434 |
+ shadd8 r7, r7, lr |
|
| 435 |
+ shadd8 r8, r8, lr |
|
| 436 |
+ shadd8 r7, r7, lr @ Filter1 >>= 3 |
|
| 437 |
+ shadd8 r8, r8, lr @ Filter2 >>= 3 |
|
| 438 |
+ |
|
| 439 |
+ qsub8 r5, r5, r7 @ u = q0 - Filter1 |
|
| 440 |
+ qadd8 r4, r4, r8 @ u = p0 + Filter2 |
|
| 441 |
+ eor r5, r5, r2 @ *oq0 = u^0x80 |
|
| 442 |
+ eor r4, r4, r2 @ *op0 = u^0x80 |
|
| 443 |
+T sub r7, r0, r1 |
|
| 444 |
+ str r5, [r0] @ store oq0 result |
|
| 445 |
+A str r4, [r0, -r1] @ store op0 result |
|
| 446 |
+T str r4, [r7] |
|
| 447 |
+ |
|
| 448 |
+2: |
|
| 449 |
+ subs r9, r9, #1 @ counter-- |
|
| 450 |
+ add r0, r0, #4 @ next row |
|
| 451 |
+T itttt ne |
|
| 452 |
+A ldrne r3, [r0, -r1, lsl #1] @ p1 |
|
| 453 |
+T subne r3, r0, r1, lsl #1 |
|
| 454 |
+T ldrne r3, [r3] @ p1 |
|
| 455 |
+A ldrne r4, [r0, -r1] @ p0 |
|
| 456 |
+T subne r4, r0, r1 |
|
| 457 |
+T ldrne r4, [r4] @ p0 |
|
| 458 |
+T itt ne |
|
| 459 |
+ ldrne r5, [r0] @ q0 |
|
| 460 |
+ ldrne r6, [r0, r1] @ q1 |
|
| 461 |
+ |
|
| 462 |
+ bne 1b |
|
| 463 |
+ |
|
| 464 |
+ pop {r4 - r11, pc}
|
|
| 465 |
+endfunc |
|
| 466 |
+ |
|
| 467 |
+c0x01010101: .long 0x01010101 |
|
| 468 |
+c0x03030303: .long 0x03030303 |
|
| 469 |
+c0x04040404: .long 0x04040404 |
|
| 470 |
+c0x7F7F7F7F: .long 0x7F7F7F7F |
|
| 471 |
+c0x80808080: .long 0x80808080 |
|
| 472 |
+ |
|
| 473 |
+@ void vp8_v_loop_filter16_inner(uint8_t *dst, int stride, |
|
| 474 |
+@ int fE, int fI, int hev_thresh) |
|
| 475 |
+@ and |
|
| 476 |
+@ void vp8_v_loop_filter8uv_inner(uint8_t *dstU, uint8_t *dstV, int stride, |
|
| 477 |
+@ int fE, int fI, int hev_thresh) |
|
| 478 |
+@ call: |
|
| 479 |
+@ void vp8_v_loop_filter_inner(uint8_t *dst, int stride, |
|
| 480 |
+@ int fE, int fI, int hev_thresh, int count) |
|
| 481 |
+function ff_vp8_v_loop_filter_inner_armv6, export=1 |
|
| 482 |
+ push {r4 - r11, lr}
|
|
| 483 |
+ |
|
| 484 |
+ sub r0, r0, r1, lsl #2 @ move r0 pointer down by 4 lines |
|
| 485 |
+ ldr r5, [sp, #40] @ counter |
|
| 486 |
+ ldr r6, [sp, #36] @ load thresh address |
|
| 487 |
+ sub sp, sp, #16 @ create temp buffer |
|
| 488 |
+ |
|
| 489 |
+ ldr r10,[r0, r1] @ p2 |
|
| 490 |
+ ldr_post r9, r0, r1, lsl #1 @ p3 |
|
| 491 |
+ ldr r12,[r0, r1] @ p0 |
|
| 492 |
+ ldr_post r11, r0, r1, lsl #1 @ p1 |
|
| 493 |
+ |
|
| 494 |
+ orr r2, r2, r2, lsl #16 |
|
| 495 |
+ orr r3, r3, r3, lsl #16 |
|
| 496 |
+ orr r6, r6, r6, lsl #16 |
|
| 497 |
+ orr r4, r2, r2, lsl #8 @ flimE splat int -> byte |
|
| 498 |
+ orr r2, r3, r3, lsl #8 @ flimI splat int -> byte |
|
| 499 |
+ orr r3, r6, r6, lsl #8 @ thresh splat int -> byte |
|
| 500 |
+ |
|
| 501 |
+1: |
|
| 502 |
+ @ vp8_filter_mask() function |
|
| 503 |
+ @ calculate breakout conditions |
|
| 504 |
+ uqsub8 r6, r9, r10 @ p3 - p2 |
|
| 505 |
+ uqsub8 r7, r10, r9 @ p2 - p3 |
|
| 506 |
+ uqsub8 r8, r10, r11 @ p2 - p1 |
|
| 507 |
+ uqsub8 r10, r11, r10 @ p1 - p2 |
|
| 508 |
+ |
|
| 509 |
+ orr r6, r6, r7 @ abs (p3-p2) |
|
| 510 |
+ orr r8, r8, r10 @ abs (p2-p1) |
|
| 511 |
+ uqsub8 lr, r6, r2 @ compare to limit. lr: vp8_filter_mask |
|
| 512 |
+ uqsub8 r8, r8, r2 @ compare to limit |
|
| 513 |
+ uqsub8 r6, r11, r12 @ p1 - p0 |
|
| 514 |
+ orr lr, lr, r8 |
|
| 515 |
+ uqsub8 r7, r12, r11 @ p0 - p1 |
|
| 516 |
+ ldr r10,[r0, r1] @ q1 |
|
| 517 |
+ ldr_post r9, r0, r1, lsl #1 @ q0 |
|
| 518 |
+ orr r6, r6, r7 @ abs (p1-p0) |
|
| 519 |
+ uqsub8 r7, r6, r2 @ compare to limit |
|
| 520 |
+ uqsub8 r8, r6, r3 @ compare to thresh -- save r8 for later |
|
| 521 |
+ orr lr, lr, r7 |
|
| 522 |
+ |
|
| 523 |
+ uqsub8 r6, r11, r10 @ p1 - q1 |
|
| 524 |
+ uqsub8 r7, r10, r11 @ q1 - p1 |
|
| 525 |
+ uqsub8 r11, r12, r9 @ p0 - q0 |
|
| 526 |
+ uqsub8 r12, r9, r12 @ q0 - p0 |
|
| 527 |
+ orr r6, r6, r7 @ abs (p1-q1) |
|
| 528 |
+ ldr r7, c0x7F7F7F7F |
|
| 529 |
+ orr r12, r11, r12 @ abs (p0-q0) |
|
| 530 |
+ ldr_post r11, r0, r1 @ q2 |
|
| 531 |
+ uqadd8 r12, r12, r12 @ abs (p0-q0) * 2 |
|
| 532 |
+ and r6, r7, r6, lsr #1 @ abs (p1-q1) / 2 |
|
| 533 |
+ uqsub8 r7, r9, r10 @ q0 - q1 |
|
| 534 |
+ uqadd8 r12, r12, r6 @ abs (p0-q0)*2 + abs (p1-q1)/2 |
|
| 535 |
+ uqsub8 r6, r10, r9 @ q1 - q0 |
|
| 536 |
+ uqsub8 r12, r12, r4 @ compare to flimit |
|
| 537 |
+ uqsub8 r9, r11, r10 @ q2 - q1 |
|
| 538 |
+ |
|
| 539 |
+ orr lr, lr, r12 |
|
| 540 |
+ |
|
| 541 |
+ ldr_post r12, r0, r1 @ q3 |
|
| 542 |
+ uqsub8 r10, r10, r11 @ q1 - q2 |
|
| 543 |
+ orr r6, r7, r6 @ abs (q1-q0) |
|
| 544 |
+ orr r10, r9, r10 @ abs (q2-q1) |
|
| 545 |
+ uqsub8 r7, r6, r2 @ compare to limit |
|
| 546 |
+ uqsub8 r10, r10, r2 @ compare to limit |
|
| 547 |
+ uqsub8 r6, r6, r3 @ compare to thresh -- save r6 for later |
|
| 548 |
+ orr lr, lr, r7 |
|
| 549 |
+ orr lr, lr, r10 |
|
| 550 |
+ |
|
| 551 |
+ uqsub8 r10, r12, r11 @ q3 - q2 |
|
| 552 |
+ uqsub8 r9, r11, r12 @ q2 - q3 |
|
| 553 |
+ |
|
| 554 |
+ mvn r11, #0 @ r11 == -1 |
|
| 555 |
+ |
|
| 556 |
+ orr r10, r10, r9 @ abs (q3-q2) |
|
| 557 |
+ uqsub8 r10, r10, r2 @ compare to limit |
|
| 558 |
+ |
|
| 559 |
+ mov r12, #0 |
|
| 560 |
+ orr lr, lr, r10 |
|
| 561 |
+ sub r0, r0, r1, lsl #2 |
|
| 562 |
+ |
|
| 563 |
+ usub8 lr, r12, lr @ use usub8 instead of ssub8 |
|
| 564 |
+ sel lr, r11, r12 @ filter mask: lr |
|
| 565 |
+ |
|
| 566 |
+ cmp lr, #0 |
|
| 567 |
+ beq 2f @ skip filtering |
|
| 568 |
+ |
|
| 569 |
+ sub r0, r0, r1, lsl #1 @ move r0 pointer down by 6 lines |
|
| 570 |
+ |
|
| 571 |
+ @vp8_hevmask() function |
|
| 572 |
+ @calculate high edge variance |
|
| 573 |
+ orr r10, r6, r8 @ calculate vp8_hevmask |
|
| 574 |
+ |
|
| 575 |
+ usub8 r10, r12, r10 @ use usub8 instead of ssub8 |
|
| 576 |
+ sel r6, r12, r11 @ obtain vp8_hevmask: r6 |
|
| 577 |
+ |
|
| 578 |
+ @vp8_filter() function |
|
| 579 |
+ ldr r8, [r0, r1] @ p0 |
|
| 580 |
+ ldr_post r7, r0, r1, lsl #1 @ p1 |
|
| 581 |
+ ldr r12, c0x80808080 |
|
| 582 |
+ ldr r10,[r0, r1] @ q1 |
|
| 583 |
+ ldr_post r9, r0, r1, lsl #1 @ q0 |
|
| 584 |
+ |
|
| 585 |
+ eor r7, r7, r12 @ p1 offset to convert to a signed value |
|
| 586 |
+ eor r8, r8, r12 @ p0 offset to convert to a signed value |
|
| 587 |
+ eor r9, r9, r12 @ q0 offset to convert to a signed value |
|
| 588 |
+ eor r10, r10, r12 @ q1 offset to convert to a signed value |
|
| 589 |
+ |
|
| 590 |
+ str r9, [sp] @ store qs0 temporarily |
|
| 591 |
+ str r8, [sp, #4] @ store ps0 temporarily |
|
| 592 |
+ str r10,[sp, #8] @ store qs1 temporarily |
|
| 593 |
+ str r7, [sp, #12] @ store ps1 temporarily |
|
| 594 |
+ |
|
| 595 |
+ qsub8 r7, r7, r10 @ vp8_signed_char_clamp(ps1-qs1) |
|
| 596 |
+ qsub8 r8, r9, r8 @ vp8_signed_char_clamp(vp8_filter + 3 * ( qs0 - ps0)) |
|
| 597 |
+ |
|
| 598 |
+ and r7, r7, r6 @ vp8_filter (r7) &= hev |
|
| 599 |
+ |
|
| 600 |
+ qadd8 r7, r7, r8 |
|
| 601 |
+ ldr r9, c0x03030303 @ r9 = 3 --modified for vp8 |
|
| 602 |
+ |
|
| 603 |
+ qadd8 r7, r7, r8 |
|
| 604 |
+ ldr r10, c0x04040404 |
|
| 605 |
+ |
|
| 606 |
+ qadd8 r7, r7, r8 |
|
| 607 |
+ and r7, r7, lr @ vp8_filter &= mask@ |
|
| 608 |
+ |
|
| 609 |
+ qadd8 r8, r7, r9 @ Filter2 (r8) = vp8_signed_char_clamp(vp8_filter+3) |
|
| 610 |
+ qadd8 r7, r7, r10 @ vp8_filter = vp8_signed_char_clamp(vp8_filter+4) |
|
| 611 |
+ |
|
| 612 |
+ mov r9, #0 |
|
| 613 |
+ shadd8 r8, r8, r9 @ Filter2 >>= 3 |
|
| 614 |
+ shadd8 r7, r7, r9 @ vp8_filter >>= 3 |
|
| 615 |
+ shadd8 r8, r8, r9 |
|
| 616 |
+ shadd8 r7, r7, r9 |
|
| 617 |
+ shadd8 lr, r8, r9 @ lr: Filter2 |
|
| 618 |
+ shadd8 r7, r7, r9 @ r7: filter |
|
| 619 |
+ |
|
| 620 |
+ @calculate output |
|
| 621 |
+ |
|
| 622 |
+ ldr r8, [sp] @ load qs0 |
|
| 623 |
+ ldr r9, [sp, #4] @ load ps0 |
|
| 624 |
+ |
|
| 625 |
+ ldr r10, c0x01010101 |
|
| 626 |
+ |
|
| 627 |
+ qsub8 r8, r8, r7 @ u = vp8_signed_char_clamp(qs0 - vp8_filter) |
|
| 628 |
+ qadd8 r9, r9, lr @ u = vp8_signed_char_clamp(ps0 + Filter2) |
|
| 629 |
+ |
|
| 630 |
+ mov lr, #0 |
|
| 631 |
+ sadd8 r7, r7, r10 @ vp8_filter += 1 |
|
| 632 |
+ shadd8 r7, r7, lr @ vp8_filter >>= 1 |
|
| 633 |
+ |
|
| 634 |
+ ldr r11,[sp, #12] @ load ps1 |
|
| 635 |
+ ldr r10,[sp, #8] @ load qs1 |
|
| 636 |
+ |
|
| 637 |
+ bic r7, r7, r6 @ vp8_filter &= ~hev |
|
| 638 |
+ sub r0, r0, r1, lsl #2 |
|
| 639 |
+ |
|
| 640 |
+ qadd8 r11, r11, r7 @ u = vp8_signed_char_clamp(ps1 + vp8_filter) |
|
| 641 |
+ qsub8 r10, r10, r7 @ u = vp8_signed_char_clamp(qs1 - vp8_filter) |
|
| 642 |
+ |
|
| 643 |
+ eor r11, r11, r12 @ *op1 = u^0x80 |
|
| 644 |
+ eor r9, r9, r12 @ *op0 = u^0x80 |
|
| 645 |
+ eor r8, r8, r12 @ *oq0 = u^0x80 |
|
| 646 |
+ eor r10, r10, r12 @ *oq1 = u^0x80 |
|
| 647 |
+ str r9, [r0, r1] @ store op0 result |
|
| 648 |
+ str_post r11, r0, r1, lsl #1 @ store op1 |
|
| 649 |
+ str r10,[r0, r1] @ store oq1 |
|
| 650 |
+ str_post r8, r0, r1, lsl #1 @ store oq0 result |
|
| 651 |
+ |
|
| 652 |
+ sub r0, r0, r1, lsl #1 |
|
| 653 |
+ |
|
| 654 |
+2: |
|
| 655 |
+ add r0, r0, #4 |
|
| 656 |
+ sub r0, r0, r1, lsl #2 |
|
| 657 |
+ |
|
| 658 |
+ subs r5, r5, #1 |
|
| 659 |
+T ittt ne |
|
| 660 |
+ ldrne r10,[r0, r1] @ p2 |
|
| 661 |
+A ldrne r9, [r0], r1, lsl #1 @ p3 |
|
| 662 |
+T ldrne r9, [r0] @ p3 |
|
| 663 |
+T addne r0, r0, r1, lsl #1 |
|
| 664 |
+T ittt ne |
|
| 665 |
+ ldrne r12,[r0, r1] @ p0 |
|
| 666 |
+A ldrne r11,[r0], r1, lsl #1 @ p1 |
|
| 667 |
+T ldrne r11,[r0] @ p3 |
|
| 668 |
+T addne r0, r0, r1, lsl #1 |
|
| 669 |
+ |
|
| 670 |
+ bne 1b |
|
| 671 |
+ |
|
| 672 |
+ add sp, sp, #16 |
|
| 673 |
+ pop {r4 - r11, pc}
|
|
| 674 |
+endfunc |
|
| 675 |
+ |
|
| 676 |
+@ void vp8_v_loop_filter16(uint8_t *dst, int stride, |
|
| 677 |
+@ int fE, int fI, int hev_thresh) |
|
| 678 |
+@ and |
|
| 679 |
+@ void vp8_v_loop_filter8uv(uint8_t *dstU, uint8_t *dstV, int stride, |
|
| 680 |
+@ int fE, int fI, int hev_thresh) |
|
| 681 |
+@ call: |
|
| 682 |
+@ void vp8_v_loop_filter(uint8_t *dst, int stride, |
|
| 683 |
+@ int fE, int fI, int hev_thresh, int count) |
|
| 684 |
+function ff_vp8_v_loop_filter_armv6, export=1 |
|
| 685 |
+ push {r4 - r11, lr}
|
|
| 686 |
+ |
|
| 687 |
+ sub r0, r0, r1, lsl #2 @ move r0 pointer down by 4 lines |
|
| 688 |
+ ldr r5, [sp, #40] @ counter |
|
| 689 |
+ ldr r6, [sp, #36] @ load thresh address |
|
| 690 |
+ sub sp, sp, #16 @ create temp buffer |
|
| 691 |
+ |
|
| 692 |
+ ldr r10,[r0, r1] @ p2 |
|
| 693 |
+ ldr_post r9, r0, r1, lsl #1 @ p3 |
|
| 694 |
+ ldr r12,[r0, r1] @ p0 |
|
| 695 |
+ ldr_post r11, r0, r1, lsl #1 @ p1 |
|
| 696 |
+ |
|
| 697 |
+ orr r2, r2, r2, lsl #16 |
|
| 698 |
+ orr r3, r3, r3, lsl #16 |
|
| 699 |
+ orr r6, r6, r6, lsl #16 |
|
| 700 |
+ orr r4, r2, r2, lsl #8 @ flimE splat int -> byte |
|
| 701 |
+ orr r2, r3, r3, lsl #8 @ flimI splat int -> byte |
|
| 702 |
+ orr r3, r6, r6, lsl #8 @ thresh splat int -> byte |
|
| 703 |
+ |
|
| 704 |
+1: |
|
| 705 |
+ @ vp8_filter_mask() function |
|
| 706 |
+ @ calculate breakout conditions |
|
| 707 |
+ uqsub8 r6, r9, r10 @ p3 - p2 |
|
| 708 |
+ uqsub8 r7, r10, r9 @ p2 - p3 |
|
| 709 |
+ uqsub8 r8, r10, r11 @ p2 - p1 |
|
| 710 |
+ uqsub8 r10, r11, r10 @ p1 - p2 |
|
| 711 |
+ |
|
| 712 |
+ orr r6, r6, r7 @ abs (p3-p2) |
|
| 713 |
+ orr r8, r8, r10 @ abs (p2-p1) |
|
| 714 |
+ uqsub8 lr, r6, r2 @ compare to limit. lr: vp8_filter_mask |
|
| 715 |
+ uqsub8 r8, r8, r2 @ compare to limit |
|
| 716 |
+ |
|
| 717 |
+ uqsub8 r6, r11, r12 @ p1 - p0 |
|
| 718 |
+ orr lr, lr, r8 |
|
| 719 |
+ uqsub8 r7, r12, r11 @ p0 - p1 |
|
| 720 |
+ ldr r10,[r0, r1] @ q1 |
|
| 721 |
+ ldr_post r9, r0, r1, lsl #1 @ q0 |
|
| 722 |
+ orr r6, r6, r7 @ abs (p1-p0) |
|
| 723 |
+ uqsub8 r7, r6, r2 @ compare to limit |
|
| 724 |
+ uqsub8 r8, r6, r3 @ compare to thresh -- save r8 for later |
|
| 725 |
+ orr lr, lr, r7 |
|
| 726 |
+ |
|
| 727 |
+ uqsub8 r6, r11, r10 @ p1 - q1 |
|
| 728 |
+ uqsub8 r7, r10, r11 @ q1 - p1 |
|
| 729 |
+ uqsub8 r11, r12, r9 @ p0 - q0 |
|
| 730 |
+ uqsub8 r12, r9, r12 @ q0 - p0 |
|
| 731 |
+ orr r6, r6, r7 @ abs (p1-q1) |
|
| 732 |
+ ldr r7, c0x7F7F7F7F |
|
| 733 |
+ orr r12, r11, r12 @ abs (p0-q0) |
|
| 734 |
+ ldr_post r11, r0, r1 @ q2 |
|
| 735 |
+ uqadd8 r12, r12, r12 @ abs (p0-q0) * 2 |
|
| 736 |
+ and r6, r7, r6, lsr #1 @ abs (p1-q1) / 2 |
|
| 737 |
+ uqsub8 r7, r9, r10 @ q0 - q1 |
|
| 738 |
+ uqadd8 r12, r12, r6 @ abs (p0-q0)*2 + abs (p1-q1)/2 |
|
| 739 |
+ uqsub8 r6, r10, r9 @ q1 - q0 |
|
| 740 |
+ uqsub8 r12, r12, r4 @ compare to flimit |
|
| 741 |
+ uqsub8 r9, r11, r10 @ q2 - q1 |
|
| 742 |
+ |
|
| 743 |
+ orr lr, lr, r12 |
|
| 744 |
+ |
|
| 745 |
+ ldr_post r12, r0, r1 @ q3 |
|
| 746 |
+ |
|
| 747 |
+ uqsub8 r10, r10, r11 @ q1 - q2 |
|
| 748 |
+ orr r6, r7, r6 @ abs (q1-q0) |
|
| 749 |
+ orr r10, r9, r10 @ abs (q2-q1) |
|
| 750 |
+ uqsub8 r7, r6, r2 @ compare to limit |
|
| 751 |
+ uqsub8 r10, r10, r2 @ compare to limit |
|
| 752 |
+ uqsub8 r6, r6, r3 @ compare to thresh -- save r6 for later |
|
| 753 |
+ orr lr, lr, r7 |
|
| 754 |
+ orr lr, lr, r10 |
|
| 755 |
+ |
|
| 756 |
+ uqsub8 r10, r12, r11 @ q3 - q2 |
|
| 757 |
+ uqsub8 r9, r11, r12 @ q2 - q3 |
|
| 758 |
+ |
|
| 759 |
+ mvn r11, #0 @ r11 == -1 |
|
| 760 |
+ |
|
| 761 |
+ orr r10, r10, r9 @ abs (q3-q2) |
|
| 762 |
+ uqsub8 r10, r10, r2 @ compare to limit |
|
| 763 |
+ |
|
| 764 |
+ mov r12, #0 |
|
| 765 |
+ |
|
| 766 |
+ orr lr, lr, r10 |
|
| 767 |
+ |
|
| 768 |
+ usub8 lr, r12, lr @ use usub8 instead of ssub8 |
|
| 769 |
+ sel lr, r11, r12 @ filter mask: lr |
|
| 770 |
+ |
|
| 771 |
+ cmp lr, #0 |
|
| 772 |
+ beq 2f @ skip filtering |
|
| 773 |
+ |
|
| 774 |
+ @vp8_hevmask() function |
|
| 775 |
+ @calculate high edge variance |
|
| 776 |
+ sub r0, r0, r1, lsl #2 @ move r0 pointer down by 6 lines |
|
| 777 |
+ sub r0, r0, r1, lsl #1 |
|
| 778 |
+ |
|
| 779 |
+ orr r10, r6, r8 |
|
| 780 |
+ |
|
| 781 |
+ usub8 r10, r12, r10 |
|
| 782 |
+ sel r6, r12, r11 @ hev mask: r6 |
|
| 783 |
+ |
|
| 784 |
+ @vp8_mbfilter() function |
|
| 785 |
+ @p2, q2 are only needed at the end. Do not need to load them in now. |
|
| 786 |
+ ldr r8, [r0, r1] @ p0 |
|
| 787 |
+ ldr_post r7, r0, r1, lsl #1 @ p1 |
|
| 788 |
+ ldr r12, c0x80808080 |
|
| 789 |
+ ldr_post r9, r0, r1 @ q0 |
|
| 790 |
+ ldr r10,[r0] @ q1 |
|
| 791 |
+ |
|
| 792 |
+ eor r7, r7, r12 @ ps1 |
|
| 793 |
+ eor r8, r8, r12 @ ps0 |
|
| 794 |
+ eor r9, r9, r12 @ qs0 |
|
| 795 |
+ eor r10, r10, r12 @ qs1 |
|
| 796 |
+ |
|
| 797 |
+ qsub8 r12, r9, r8 @ vp8_signed_char_clamp(vp8_filter + 3 * ( qs0 - ps0)) |
|
| 798 |
+ str r7, [sp, #12] @ store ps1 temporarily |
|
| 799 |
+ qsub8 r7, r7, r10 @ vp8_signed_char_clamp(ps1-qs1) |
|
| 800 |
+ str r10,[sp, #8] @ store qs1 temporarily |
|
| 801 |
+ qadd8 r7, r7, r12 |
|
| 802 |
+ str r9, [sp] @ store qs0 temporarily |
|
| 803 |
+ qadd8 r7, r7, r12 |
|
| 804 |
+ str r8, [sp, #4] @ store ps0 temporarily |
|
| 805 |
+ qadd8 r7, r7, r12 @ vp8_filter: r7 |
|
| 806 |
+ |
|
| 807 |
+ ldr r10, c0x03030303 @ r10 = 3 --modified for vp8 |
|
| 808 |
+ ldr r9, c0x04040404 |
|
| 809 |
+ |
|
| 810 |
+ and r7, r7, lr @ vp8_filter &= mask (lr is free) |
|
| 811 |
+ |
|
| 812 |
+ mov r12, r7 @ Filter2: r12 |
|
| 813 |
+ and r12, r12, r6 @ Filter2 &= hev |
|
| 814 |
+ |
|
| 815 |
+ @save bottom 3 bits so that we round one side +4 and the other +3 |
|
| 816 |
+ qadd8 r8, r12, r9 @ Filter1 (r8) = vp8_signed_char_clamp(Filter2+4) |
|
| 817 |
+ qadd8 r12, r12, r10 @ Filter2 (r12) = vp8_signed_char_clamp(Filter2+3) |
|
| 818 |
+ |
|
| 819 |
+ mov r10, #0 |
|
| 820 |
+ shadd8 r8, r8, r10 @ Filter1 >>= 3 |
|
| 821 |
+ shadd8 r12, r12, r10 @ Filter2 >>= 3 |
|
| 822 |
+ shadd8 r8, r8, r10 |
|
| 823 |
+ shadd8 r12, r12, r10 |
|
| 824 |
+ shadd8 r8, r8, r10 @ r8: Filter1 |
|
| 825 |
+ shadd8 r12, r12, r10 @ r12: Filter2 |
|
| 826 |
+ |
|
| 827 |
+ ldr r9, [sp] @ load qs0 |
|
| 828 |
+ ldr r11,[sp, #4] @ load ps0 |
|
| 829 |
+ |
|
| 830 |
+ qsub8 r9, r9, r8 @ qs0 = vp8_signed_char_clamp(qs0 - Filter1) |
|
| 831 |
+ qadd8 r11, r11, r12 @ ps0 = vp8_signed_char_clamp(ps0 + Filter2) |
|
| 832 |
+ |
|
| 833 |
+ bic r12, r7, r6 @ vp8_filter &= ~hev ( r6 is free) |
|
| 834 |
+ |
|
| 835 |
+ @roughly 3/7th difference across boundary |
|
| 836 |
+ mov lr, #0x1b @ 27 |
|
| 837 |
+ mov r7, #0x3f @ 63 |
|
| 838 |
+ |
|
| 839 |
+ sxtb16 r6, r12 |
|
| 840 |
+ sxtb16 r10, r12, ror #8 |
|
| 841 |
+ smlabb r8, r6, lr, r7 |
|
| 842 |
+ smlatb r6, r6, lr, r7 |
|
| 843 |
+ smlabb r7, r10, lr, r7 |
|
| 844 |
+ smultb r10, r10, lr |
|
| 845 |
+ ssat r8, #8, r8, asr #7 |
|
| 846 |
+ ssat r6, #8, r6, asr #7 |
|
| 847 |
+ add r10, r10, #63 |
|
| 848 |
+ ssat r7, #8, r7, asr #7 |
|
| 849 |
+ ssat r10, #8, r10, asr #7 |
|
| 850 |
+ |
|
| 851 |
+ ldr lr, c0x80808080 |
|
| 852 |
+ |
|
| 853 |
+ pkhbt r6, r8, r6, lsl #16 |
|
| 854 |
+ pkhbt r10, r7, r10, lsl #16 |
|
| 855 |
+ uxtb16 r6, r6 |
|
| 856 |
+ uxtb16 r10, r10 |
|
| 857 |
+ |
|
| 858 |
+ sub r0, r0, r1 |
|
| 859 |
+ |
|
| 860 |
+ orr r10, r6, r10, lsl #8 @ u = vp8_signed_char_clamp((63 + Filter2 * 27)>>7) |
|
| 861 |
+ |
|
| 862 |
+ qsub8 r8, r9, r10 @ s = vp8_signed_char_clamp(qs0 - u) |
|
| 863 |
+ qadd8 r10, r11, r10 @ s = vp8_signed_char_clamp(ps0 + u) |
|
| 864 |
+ eor r8, r8, lr @ *oq0 = s^0x80 |
|
| 865 |
+ str r8, [r0] @ store *oq0 |
|
| 866 |
+ sub r0, r0, r1 |
|
| 867 |
+ eor r10, r10, lr @ *op0 = s^0x80 |
|
| 868 |
+ str r10,[r0] @ store *op0 |
|
| 869 |
+ |
|
| 870 |
+ @roughly 2/7th difference across boundary |
|
| 871 |
+ mov lr, #0x12 @ 18 |
|
| 872 |
+ mov r7, #0x3f @ 63 |
|
| 873 |
+ |
|
| 874 |
+ sxtb16 r6, r12 |
|
| 875 |
+ sxtb16 r10, r12, ror #8 |
|
| 876 |
+ smlabb r8, r6, lr, r7 |
|
| 877 |
+ smlatb r6, r6, lr, r7 |
|
| 878 |
+ smlabb r9, r10, lr, r7 |
|
| 879 |
+ smlatb r10, r10, lr, r7 |
|
| 880 |
+ ssat r8, #8, r8, asr #7 |
|
| 881 |
+ ssat r6, #8, r6, asr #7 |
|
| 882 |
+ ssat r9, #8, r9, asr #7 |
|
| 883 |
+ ssat r10, #8, r10, asr #7 |
|
| 884 |
+ |
|
| 885 |
+ ldr lr, c0x80808080 |
|
| 886 |
+ |
|
| 887 |
+ pkhbt r6, r8, r6, lsl #16 |
|
| 888 |
+ pkhbt r10, r9, r10, lsl #16 |
|
| 889 |
+ |
|
| 890 |
+ ldr r9, [sp, #8] @ load qs1 |
|
| 891 |
+ ldr r11, [sp, #12] @ load ps1 |
|
| 892 |
+ |
|
| 893 |
+ uxtb16 r6, r6 |
|
| 894 |
+ uxtb16 r10, r10 |
|
| 895 |
+ |
|
| 896 |
+ sub r0, r0, r1 |
|
| 897 |
+ |
|
| 898 |
+ orr r10, r6, r10, lsl #8 @ u = vp8_signed_char_clamp((63 + Filter2 * 18)>>7) |
|
| 899 |
+ |
|
| 900 |
+ qadd8 r11, r11, r10 @ s = vp8_signed_char_clamp(ps1 + u) |
|
| 901 |
+ qsub8 r8, r9, r10 @ s = vp8_signed_char_clamp(qs1 - u) |
|
| 902 |
+ eor r11, r11, lr @ *op1 = s^0x80 |
|
| 903 |
+ str_post r11, r0, r1 @ store *op1 |
|
| 904 |
+ eor r8, r8, lr @ *oq1 = s^0x80 |
|
| 905 |
+ add r0, r0, r1, lsl #1 |
|
| 906 |
+ |
|
| 907 |
+ mov r7, #0x3f @ 63 |
|
| 908 |
+ |
|
| 909 |
+ str_post r8, r0, r1 @ store *oq1 |
|
| 910 |
+ |
|
| 911 |
+ @roughly 1/7th difference across boundary |
|
| 912 |
+ mov lr, #0x9 @ 9 |
|
| 913 |
+ ldr r9, [r0] @ load q2 |
|
| 914 |
+ |
|
| 915 |
+ sxtb16 r6, r12 |
|
| 916 |
+ sxtb16 r10, r12, ror #8 |
|
| 917 |
+ smlabb r8, r6, lr, r7 |
|
| 918 |
+ smlatb r6, r6, lr, r7 |
|
| 919 |
+ smlabb r12, r10, lr, r7 |
|
| 920 |
+ smlatb r10, r10, lr, r7 |
|
| 921 |
+ ssat r8, #8, r8, asr #7 |
|
| 922 |
+ ssat r6, #8, r6, asr #7 |
|
| 923 |
+ ssat r12, #8, r12, asr #7 |
|
| 924 |
+ ssat r10, #8, r10, asr #7 |
|
| 925 |
+ |
|
| 926 |
+ sub r0, r0, r1, lsl #2 |
|
| 927 |
+ |
|
| 928 |
+ pkhbt r6, r8, r6, lsl #16 |
|
| 929 |
+ pkhbt r10, r12, r10, lsl #16 |
|
| 930 |
+ |
|
| 931 |
+ sub r0, r0, r1 |
|
| 932 |
+ ldr lr, c0x80808080 |
|
| 933 |
+ |
|
| 934 |
+ ldr r11, [r0] @ load p2 |
|
| 935 |
+ |
|
| 936 |
+ uxtb16 r6, r6 |
|
| 937 |
+ uxtb16 r10, r10 |
|
| 938 |
+ |
|
| 939 |
+ eor r9, r9, lr |
|
| 940 |
+ eor r11, r11, lr |
|
| 941 |
+ |
|
| 942 |
+ orr r10, r6, r10, lsl #8 @ u = vp8_signed_char_clamp((63 + Filter2 * 9)>>7) |
|
| 943 |
+ |
|
| 944 |
+ qadd8 r8, r11, r10 @ s = vp8_signed_char_clamp(ps2 + u) |
|
| 945 |
+ qsub8 r10, r9, r10 @ s = vp8_signed_char_clamp(qs2 - u) |
|
| 946 |
+ eor r8, r8, lr @ *op2 = s^0x80 |
|
| 947 |
+ str_post r8, r0, r1, lsl #2 @ store *op2 |
|
| 948 |
+ add r0, r0, r1 |
|
| 949 |
+ eor r10, r10, lr @ *oq2 = s^0x80 |
|
| 950 |
+ str_post r10, r0, r1, lsl #1 @ store *oq2 |
|
| 951 |
+ |
|
| 952 |
+2: |
|
| 953 |
+ add r0, r0, #4 |
|
| 954 |
+ sub r0, r0, r1, lsl #3 |
|
| 955 |
+ subs r5, r5, #1 |
|
| 956 |
+ |
|
| 957 |
+T ittt ne |
|
| 958 |
+ ldrne r10,[r0, r1] @ p2 |
|
| 959 |
+A ldrne r9, [r0], r1, lsl #1 @ p3 |
|
| 960 |
+T ldrne r9, [r0] @ p3 |
|
| 961 |
+T addne r0, r0, r1, lsl #1 |
|
| 962 |
+T ittt ne |
|
| 963 |
+ ldrne r12,[r0, r1] @ p0 |
|
| 964 |
+A ldrne r11,[r0], r1, lsl #1 @ p1 |
|
| 965 |
+T ldrne r11,[r0] @ p3 |
|
| 966 |
+T addne r0, r0, r1, lsl #1 |
|
| 967 |
+ |
|
| 968 |
+ bne 1b |
|
| 969 |
+ |
|
| 970 |
+ add sp, sp, #16 |
|
| 971 |
+ pop {r4 - r11, pc}
|
|
| 972 |
+endfunc |
|
| 973 |
+ |
|
| 974 |
+.macro TRANSPOSE_MATRIX i0, i1, i2, i3, o3, o2, o1, o0 |
|
| 975 |
+ @ input: $0, $1, $2, $3 |
|
| 976 |
+ @ output: $4, $5, $6, $7 |
|
| 977 |
+ @ i0: 03 02 01 00 |
|
| 978 |
+ @ i1: 13 12 11 10 |
|
| 979 |
+ @ i2: 23 22 21 20 |
|
| 980 |
+ @ i3: 33 32 31 30 |
|
| 981 |
+ @ o3 o2 o1 o0 |
|
| 982 |
+ |
|
| 983 |
+ uxtb16 \o1, \i1 @ xx 12 xx 10 |
|
| 984 |
+ uxtb16 \o0, \i0 @ xx 02 xx 00 |
|
| 985 |
+ uxtb16 \o3, \i3 @ xx 32 xx 30 |
|
| 986 |
+ uxtb16 \o2, \i2 @ xx 22 xx 20 |
|
| 987 |
+ orr \o1, \o0, \o1, lsl #8 @ 12 02 10 00 |
|
| 988 |
+ orr \o3, \o2, \o3, lsl #8 @ 32 22 30 20 |
|
| 989 |
+ |
|
| 990 |
+ uxtb16 \i1, \i1, ror #8 @ xx 13 xx 11 |
|
| 991 |
+ uxtb16 \i3, \i3, ror #8 @ xx 33 xx 31 |
|
| 992 |
+ uxtb16 \i0, \i0, ror #8 @ xx 03 xx 01 |
|
| 993 |
+ uxtb16 \i2, \i2, ror #8 @ xx 23 xx 21 |
|
| 994 |
+ orr \i0, \i0, \i1, lsl #8 @ 13 03 11 01 |
|
| 995 |
+ orr \i2, \i2, \i3, lsl #8 @ 33 23 31 21 |
|
| 996 |
+ |
|
| 997 |
+ pkhtb \o2, \o3, \o1, asr #16 @ 32 22 12 02 -- p1 |
|
| 998 |
+ pkhbt \o0, \o1, \o3, lsl #16 @ 30 20 10 00 -- p3 |
|
| 999 |
+ |
|
| 1000 |
+ pkhtb \o3, \i2, \i0, asr #16 @ 33 23 13 03 -- p0 |
|
| 1001 |
+ pkhbt \o1, \i0, \i2, lsl #16 @ 31 21 11 01 -- p2 |
|
| 1002 |
+.endm |
|
| 1003 |
+ |
|
| 1004 |
+@ void vp8_h_loop_filter16_simple(uint8_t *dst, int stride, int flim) |
|
| 1005 |
+function ff_vp8_h_loop_filter16_simple_armv6, export=1 |
|
| 1006 |
+ push {r4 - r11, lr}
|
|
| 1007 |
+ orr r12, r2, r2, lsl #16 |
|
| 1008 |
+ ldr r2, c0x80808080 |
|
| 1009 |
+ orr r12, r12, r12, lsl #8 |
|
| 1010 |
+ |
|
| 1011 |
+ @ load soure data to r7, r8, r9, r10 |
|
| 1012 |
+ sub r0, r0, #2 |
|
| 1013 |
+ ldr r8, [r0, r1] |
|
| 1014 |
+ ldr_post r7, r0, r1, lsl #1 |
|
| 1015 |
+ ldr r10,[r0, r1] |
|
| 1016 |
+ ldr_post r9, r0, r1, lsl #1 |
|
| 1017 |
+ add r0, r0, #2 |
|
| 1018 |
+ |
|
| 1019 |
+ mov r11, #4 @ count (r11) for 4-in-parallel |
|
| 1020 |
+1: |
|
| 1021 |
+ @transpose r7, r8, r9, r10 to r3, r4, r5, r6 |
|
| 1022 |
+ TRANSPOSE_MATRIX r7, r8, r9, r10, r6, r5, r4, r3 |
|
| 1023 |
+ |
|
| 1024 |
+ @ vp8_simple_filter_mask() function |
|
| 1025 |
+ uqsub8 r7, r3, r6 @ p1 - q1 |
|
| 1026 |
+ uqsub8 r8, r6, r3 @ q1 - p1 |
|
| 1027 |
+ uqsub8 r9, r4, r5 @ p0 - q0 |
|
| 1028 |
+ uqsub8 r10, r5, r4 @ q0 - p0 |
|
| 1029 |
+ orr r7, r7, r8 @ abs(p1 - q1) |
|
| 1030 |
+ orr r9, r9, r10 @ abs(p0 - q0) |
|
| 1031 |
+ mov r8, #0 |
|
| 1032 |
+ uqadd8 r9, r9, r9 @ abs(p0 - q0) * 2 |
|
| 1033 |
+ uhadd8 r7, r7, r8 @ abs(p1 - q1) / 2 |
|
| 1034 |
+ uqadd8 r7, r7, r9 @ abs(p0 - q0)*2 + abs(p1 - q1)/2 |
|
| 1035 |
+ mvn r10, #0 @ r10 == -1 |
|
| 1036 |
+ |
|
| 1037 |
+ usub8 r7, r12, r7 @ compare to flimit |
|
| 1038 |
+ sel lr, r10, r8 @ filter mask |
|
| 1039 |
+ |
|
| 1040 |
+ cmp lr, #0 |
|
| 1041 |
+ beq 2f @ skip filtering |
|
| 1042 |
+ |
|
| 1043 |
+ @vp8_simple_filter() function |
|
| 1044 |
+ eor r3, r3, r2 @ p1 offset to convert to a signed value |
|
| 1045 |
+ eor r6, r6, r2 @ q1 offset to convert to a signed value |
|
| 1046 |
+ eor r4, r4, r2 @ p0 offset to convert to a signed value |
|
| 1047 |
+ eor r5, r5, r2 @ q0 offset to convert to a signed value |
|
| 1048 |
+ |
|
| 1049 |
+ qsub8 r3, r3, r6 @ vp8_filter = p1 - q1 |
|
| 1050 |
+ qsub8 r6, r5, r4 @ q0 - p0 |
|
| 1051 |
+ |
|
| 1052 |
+ qadd8 r3, r3, r6 @ vp8_filter += q0 - p0 |
|
| 1053 |
+ ldr r9, c0x03030303 @ r9 = 3 |
|
| 1054 |
+ |
|
| 1055 |
+ qadd8 r3, r3, r6 @ vp8_filter += q0 - p0 |
|
| 1056 |
+ ldr r7, c0x04040404 |
|
| 1057 |
+ |
|
| 1058 |
+ qadd8 r3, r3, r6 @ vp8_filter = p1-q1 + 3*(q0-p0)) |
|
| 1059 |
+ @STALL |
|
| 1060 |
+ and r3, r3, lr @ vp8_filter &= mask |
|
| 1061 |
+ |
|
| 1062 |
+ qadd8 r9, r3, r9 @ Filter2 = vp8_filter + 3 |
|
| 1063 |
+ qadd8 r3, r3, r7 @ Filter1 = vp8_filter + 4 |
|
| 1064 |
+ |
|
| 1065 |
+ shadd8 r9, r9, r8 |
|
| 1066 |
+ shadd8 r3, r3, r8 |
|
| 1067 |
+ shadd8 r9, r9, r8 |
|
| 1068 |
+ shadd8 r3, r3, r8 |
|
| 1069 |
+ shadd8 r9, r9, r8 @ Filter2 >>= 3 |
|
| 1070 |
+ shadd8 r3, r3, r8 @ Filter1 >>= 3 |
|
| 1071 |
+ |
|
| 1072 |
+ @calculate output |
|
| 1073 |
+ sub r0, r0, r1, lsl #2 |
|
| 1074 |
+ |
|
| 1075 |
+ qadd8 r4, r4, r9 @ u = p0 + Filter2 |
|
| 1076 |
+ qsub8 r5, r5, r3 @ u = q0 - Filter1 |
|
| 1077 |
+ eor r4, r4, r2 @ *op0 = u^0x80 |
|
| 1078 |
+ eor r5, r5, r2 @ *oq0 = u^0x80 |
|
| 1079 |
+ |
|
| 1080 |
+ strb r4, [r0, #-1] @ store the result |
|
| 1081 |
+ mov r4, r4, lsr #8 |
|
| 1082 |
+ strb_post r5, r0, r1 |
|
| 1083 |
+ mov r5, r5, lsr #8 |
|
| 1084 |
+ |
|
| 1085 |
+ strb r4, [r0, #-1] |
|
| 1086 |
+ mov r4, r4, lsr #8 |
|
| 1087 |
+ strb_post r5, r0, r1 |
|
| 1088 |
+ mov r5, r5, lsr #8 |
|
| 1089 |
+ |
|
| 1090 |
+ strb r4, [r0, #-1] |
|
| 1091 |
+ mov r4, r4, lsr #8 |
|
| 1092 |
+ strb_post r5, r0, r1 |
|
| 1093 |
+ mov r5, r5, lsr #8 |
|
| 1094 |
+ |
|
| 1095 |
+ strb r4, [r0, #-1] |
|
| 1096 |
+ strb_post r5, r0, r1 |
|
| 1097 |
+ |
|
| 1098 |
+2: |
|
| 1099 |
+ subs r11, r11, #1 |
|
| 1100 |
+ |
|
| 1101 |
+ @ load soure data to r7, r8, r9, r10 |
|
| 1102 |
+ sub r0, r0, #2 |
|
| 1103 |
+T ittt ne |
|
| 1104 |
+ ldrne r8, [r0, r1] |
|
| 1105 |
+A ldrne r7, [r0], r1, lsl #1 |
|
| 1106 |
+T ldrne r7, [r0] |
|
| 1107 |
+T addne r0, r0, r1, lsl #1 |
|
| 1108 |
+T ittt ne |
|
| 1109 |
+ ldrne r10,[r0, r1] |
|
| 1110 |
+A ldrne r9, [r0], r1, lsl #1 |
|
| 1111 |
+T ldrne r9, [r0] |
|
| 1112 |
+T addne r0, r0, r1, lsl #1 |
|
| 1113 |
+ add r0, r0, #2 |
|
| 1114 |
+ |
|
| 1115 |
+ bne 1b |
|
| 1116 |
+ |
|
| 1117 |
+ pop {r4 - r11, pc}
|
|
| 1118 |
+endfunc |
|
| 1119 |
+ |
|
| 1120 |
+@ void vp8_h_loop_filter16_inner(uint8_t *dst, int stride, |
|
| 1121 |
+@ int fE, int fI, int hev_thresh) |
|
| 1122 |
+@ and |
|
| 1123 |
+@ void vp8_h_loop_filter8uv_inner(uint8_t *dstU, uint8_t *dstV, int stride, |
|
| 1124 |
+@ int fE, int fI, int hev_thresh) |
|
| 1125 |
+@ call: |
|
| 1126 |
+@ void vp8_h_loop_filter_inner(uint8_t *dst, int stride, |
|
| 1127 |
+@ int fE, int fI, int hev_thresh, int count) |
|
| 1128 |
+function ff_vp8_h_loop_filter_inner_armv6, export=1 |
|
| 1129 |
+ push {r4 - r11, lr}
|
|
| 1130 |
+ |
|
| 1131 |
+ sub r0, r0, #4 @ move r0 pointer down by 4 |
|
| 1132 |
+ ldr r5, [sp, #40] @ counter |
|
| 1133 |
+ ldr r9, [sp, #36] @ load thresh address |
|
| 1134 |
+ sub sp, sp, #16 @ create temp buffer |
|
| 1135 |
+ |
|
| 1136 |
+ ldr r7, [r0, r1] @ transpose will make it into p3-p0 |
|
| 1137 |
+ ldr_post r6, r0, r1, lsl #1 @ load source data |
|
| 1138 |
+ ldr lr, [r0, r1] |
|
| 1139 |
+ ldr_post r8, r0, r1, lsl #1 |
|
| 1140 |
+ |
|
| 1141 |
+ orr r2, r2, r2, lsl #16 |
|
| 1142 |
+ orr r3, r3, r3, lsl #16 |
|
| 1143 |
+ orr r9, r9, r9, lsl #16 |
|
| 1144 |
+ orr r4, r2, r2, lsl #8 @ flimE splat int -> byte |
|
| 1145 |
+ orr r2, r3, r3, lsl #8 @ flimI splat int -> byte |
|
| 1146 |
+ orr r3, r9, r9, lsl #8 @ thresh splat int -> byte |
|
| 1147 |
+ |
|
| 1148 |
+1: |
|
| 1149 |
+ @ vp8_filter_mask() function |
|
| 1150 |
+ @ calculate breakout conditions |
|
| 1151 |
+ @ transpose the source data for 4-in-parallel operation |
|
| 1152 |
+ TRANSPOSE_MATRIX r6, r7, r8, lr, r12, r11, r10, r9 |
|
| 1153 |
+ |
|
| 1154 |
+ uqsub8 r7, r9, r10 @ p3 - p2 |
|
| 1155 |
+ uqsub8 r8, r10, r9 @ p2 - p3 |
|
| 1156 |
+ uqsub8 r9, r10, r11 @ p2 - p1 |
|
| 1157 |
+ uqsub8 r10, r11, r10 @ p1 - p2 |
|
| 1158 |
+ orr r7, r7, r8 @ abs (p3-p2) |
|
| 1159 |
+ orr r10, r9, r10 @ abs (p2-p1) |
|
| 1160 |
+ uqsub8 lr, r7, r2 @ compare to limit. lr: vp8_filter_mask |
|
| 1161 |
+ uqsub8 r10, r10, r2 @ compare to limit |
|
| 1162 |
+ |
|
| 1163 |
+ sub r0, r0, r1, lsl #2 @ move r0 pointer down by 4 lines |
|
| 1164 |
+ |
|
| 1165 |
+ orr lr, lr, r10 |
|
| 1166 |
+ |
|
| 1167 |
+ uqsub8 r6, r11, r12 @ p1 - p0 |
|
| 1168 |
+ uqsub8 r7, r12, r11 @ p0 - p1 |
|
| 1169 |
+ add r0, r0, #4 @ move r0 pointer up by 4 |
|
| 1170 |
+ orr r6, r6, r7 @ abs (p1-p0) |
|
| 1171 |
+ str r11,[sp, #12] @ save p1 |
|
| 1172 |
+ uqsub8 r10, r6, r2 @ compare to limit |
|
| 1173 |
+ uqsub8 r11, r6, r3 @ compare to thresh |
|
| 1174 |
+ orr lr, lr, r10 |
|
| 1175 |
+ |
|
| 1176 |
+ @ transpose uses 8 regs(r6 - r12 and lr). Need to save reg value now |
|
| 1177 |
+ @ transpose the source data for 4-in-parallel operation |
|
| 1178 |
+ str r11,[sp] @ push r11 to stack |
|
| 1179 |
+ ldr r7, [r0, r1] |
|
| 1180 |
+ ldr_post r6, r0, r1, lsl #1 @ load source data |
|
| 1181 |
+ str r12,[sp, #4] @ save current reg before load q0 - q3 data |
|
| 1182 |
+ str lr, [sp, #8] |
|
| 1183 |
+ ldr lr, [r0, r1] |
|
| 1184 |
+ ldr_post r8, r0, r1, lsl #1 |
|
| 1185 |
+ |
|
| 1186 |
+ TRANSPOSE_MATRIX r6, r7, r8, lr, r12, r11, r10, r9 |
|
| 1187 |
+ |
|
| 1188 |
+ ldr lr, [sp, #8] @ load back (f)limit accumulator |
|
| 1189 |
+ |
|
| 1190 |
+ uqsub8 r6, r12, r11 @ q3 - q2 |
|
| 1191 |
+ uqsub8 r7, r11, r12 @ q2 - q3 |
|
| 1192 |
+ uqsub8 r12, r11, r10 @ q2 - q1 |
|
| 1193 |
+ uqsub8 r11, r10, r11 @ q1 - q2 |
|
| 1194 |
+ orr r6, r6, r7 @ abs (q3-q2) |
|
| 1195 |
+ orr r7, r12, r11 @ abs (q2-q1) |
|
| 1196 |
+ uqsub8 r6, r6, r2 @ compare to limit |
|
| 1197 |
+ uqsub8 r7, r7, r2 @ compare to limit |
|
| 1198 |
+ ldr r11,[sp, #4] @ load back p0 |
|
| 1199 |
+ ldr r12,[sp, #12] @ load back p1 |
|
| 1200 |
+ orr lr, lr, r6 |
|
| 1201 |
+ orr lr, lr, r7 |
|
| 1202 |
+ |
|
| 1203 |
+ uqsub8 r6, r11, r9 @ p0 - q0 |
|
| 1204 |
+ uqsub8 r7, r9, r11 @ q0 - p0 |
|
| 1205 |
+ uqsub8 r8, r12, r10 @ p1 - q1 |
|
| 1206 |
+ uqsub8 r11, r10, r12 @ q1 - p1 |
|
| 1207 |
+ orr r6, r6, r7 @ abs (p0-q0) |
|
| 1208 |
+ ldr r7, c0x7F7F7F7F |
|
| 1209 |
+ orr r8, r8, r11 @ abs (p1-q1) |
|
| 1210 |
+ uqadd8 r6, r6, r6 @ abs (p0-q0) * 2 |
|
| 1211 |
+ and r8, r7, r8, lsr #1 @ abs (p1-q1) / 2 |
|
| 1212 |
+ uqsub8 r11, r10, r9 @ q1 - q0 |
|
| 1213 |
+ uqadd8 r6, r8, r6 @ abs (p0-q0)*2 + abs (p1-q1)/2 |
|
| 1214 |
+ uqsub8 r12, r9, r10 @ q0 - q1 |
|
| 1215 |
+ uqsub8 r6, r6, r4 @ compare to flimit |
|
| 1216 |
+ |
|
| 1217 |
+ orr r9, r11, r12 @ abs (q1-q0) |
|
| 1218 |
+ uqsub8 r8, r9, r2 @ compare to limit |
|
| 1219 |
+ uqsub8 r10, r9, r3 @ compare to thresh |
|
| 1220 |
+ orr lr, lr, r6 |
|
| 1221 |
+ orr lr, lr, r8 |
|
| 1222 |
+ |
|
| 1223 |
+ mvn r11, #0 @ r11 == -1 |
|
| 1224 |
+ mov r12, #0 |
|
| 1225 |
+ |
|
| 1226 |
+ usub8 lr, r12, lr |
|
| 1227 |
+ ldr r9, [sp] @ load the compared result |
|
| 1228 |
+ sel lr, r11, r12 @ filter mask: lr |
|
| 1229 |
+ |
|
| 1230 |
+ cmp lr, #0 |
|
| 1231 |
+ beq 2f @ skip filtering |
|
| 1232 |
+ |
|
| 1233 |
+ @vp8_hevmask() function |
|
| 1234 |
+ @calculate high edge variance |
|
| 1235 |
+ sub r0, r0, r1, lsl #2 @ move r0 pointer down by 4 lines |
|
| 1236 |
+ |
|
| 1237 |
+ orr r9, r9, r10 |
|
| 1238 |
+ |
|
| 1239 |
+ ldrh r7, [r0, #-2] |
|
| 1240 |
+ ldrh_post r8, r0, r1 |
|
| 1241 |
+ |
|
| 1242 |
+ usub8 r9, r12, r9 |
|
| 1243 |
+ sel r6, r12, r11 @ hev mask: r6 |
|
| 1244 |
+ |
|
| 1245 |
+ @vp8_filter() function |
|
| 1246 |
+ @ load soure data to r6, r11, r12, lr |
|
| 1247 |
+ ldrh r9, [r0, #-2] |
|
| 1248 |
+ ldrh_post r10, r0, r1 |
|
| 1249 |
+ |
|
| 1250 |
+ pkhbt r12, r7, r8, lsl #16 |
|
| 1251 |
+ |
|
| 1252 |
+ ldrh r7, [r0, #-2] |
|
| 1253 |
+ ldrh_post r8, r0, r1 |
|
| 1254 |
+ |
|
| 1255 |
+ pkhbt r11, r9, r10, lsl #16 |
|
| 1256 |
+ |
|
| 1257 |
+ ldrh r9, [r0, #-2] |
|
| 1258 |
+ ldrh_post r10, r0, r1 |
|
| 1259 |
+ |
|
| 1260 |
+ @ Transpose needs 8 regs(r6 - r12, and lr). Save r6 and lr first |
|
| 1261 |
+ str r6, [sp] |
|
| 1262 |
+ str lr, [sp, #4] |
|
| 1263 |
+ |
|
| 1264 |
+ pkhbt r6, r7, r8, lsl #16 |
|
| 1265 |
+ pkhbt lr, r9, r10, lsl #16 |
|
| 1266 |
+ |
|
| 1267 |
+ @transpose r12, r11, r6, lr to r7, r8, r9, r10 |
|
| 1268 |
+ TRANSPOSE_MATRIX r12, r11, r6, lr, r10, r9, r8, r7 |
|
| 1269 |
+ |
|
| 1270 |
+ @load back hev_mask r6 and filter_mask lr |
|
| 1271 |
+ ldr r12, c0x80808080 |
|
| 1272 |
+ ldr r6, [sp] |
|
| 1273 |
+ ldr lr, [sp, #4] |
|
| 1274 |
+ |
|
| 1275 |
+ eor r7, r7, r12 @ p1 offset to convert to a signed value |
|
| 1276 |
+ eor r8, r8, r12 @ p0 offset to convert to a signed value |
|
| 1277 |
+ eor r9, r9, r12 @ q0 offset to convert to a signed value |
|
| 1278 |
+ eor r10, r10, r12 @ q1 offset to convert to a signed value |
|
| 1279 |
+ |
|
| 1280 |
+ str r9, [sp] @ store qs0 temporarily |
|
| 1281 |
+ str r8, [sp, #4] @ store ps0 temporarily |
|
| 1282 |
+ str r10,[sp, #8] @ store qs1 temporarily |
|
| 1283 |
+ str r7, [sp, #12] @ store ps1 temporarily |
|
| 1284 |
+ |
|
| 1285 |
+ qsub8 r7, r7, r10 @ vp8_signed_char_clamp(ps1-qs1) |
|
| 1286 |
+ qsub8 r8, r9, r8 @ vp8_signed_char_clamp(vp8_filter + 3 * ( qs0 - ps0)) |
|
| 1287 |
+ |
|
| 1288 |
+ and r7, r7, r6 @ vp8_filter (r7) &= hev (r7 : filter) |
|
| 1289 |
+ |
|
| 1290 |
+ qadd8 r7, r7, r8 |
|
| 1291 |
+ ldr r9, c0x03030303 @ r9 = 3 --modified for vp8 |
|
| 1292 |
+ |
|
| 1293 |
+ qadd8 r7, r7, r8 |
|
| 1294 |
+ ldr r10, c0x04040404 |
|
| 1295 |
+ |
|
| 1296 |
+ qadd8 r7, r7, r8 |
|
| 1297 |
+ |
|
| 1298 |
+ and r7, r7, lr @ vp8_filter &= mask |
|
| 1299 |
+ |
|
| 1300 |
+ qadd8 r8, r7, r9 @ Filter2 (r8) = vp8_signed_char_clamp(vp8_filter+3) |
|
| 1301 |
+ qadd8 r7, r7, r10 @ vp8_filter = vp8_signed_char_clamp(vp8_filter+4) |
|
| 1302 |
+ |
|
| 1303 |
+ mov r9, #0 |
|
| 1304 |
+ shadd8 r8, r8, r9 @ Filter2 >>= 3 |
|
| 1305 |
+ shadd8 r7, r7, r9 @ vp8_filter >>= 3 |
|
| 1306 |
+ shadd8 r8, r8, r9 |
|
| 1307 |
+ shadd8 r7, r7, r9 |
|
| 1308 |
+ shadd8 lr, r8, r9 @ lr: filter2 |
|
| 1309 |
+ shadd8 r7, r7, r9 @ r7: filter |
|
| 1310 |
+ |
|
| 1311 |
+ @calculate output |
|
| 1312 |
+ ldr r8, [sp] @ load qs0 |
|
| 1313 |
+ ldr r9, [sp, #4] @ load ps0 |
|
| 1314 |
+ |
|
| 1315 |
+ ldr r10, c0x01010101 |
|
| 1316 |
+ |
|
| 1317 |
+ qsub8 r8, r8, r7 @ u = vp8_signed_char_clamp(qs0 - vp8_filter) |
|
| 1318 |
+ qadd8 r9, r9, lr @ u = vp8_signed_char_clamp(ps0 + Filter2) |
|
| 1319 |
+ |
|
| 1320 |
+ eor r8, r8, r12 |
|
| 1321 |
+ eor r9, r9, r12 |
|
| 1322 |
+ |
|
| 1323 |
+ mov lr, #0 |
|
| 1324 |
+ |
|
| 1325 |
+ sadd8 r7, r7, r10 |
|
| 1326 |
+ shadd8 r7, r7, lr |
|
| 1327 |
+ |
|
| 1328 |
+ ldr r10,[sp, #8] @ load qs1 |
|
| 1329 |
+ ldr r11,[sp, #12] @ load ps1 |
|
| 1330 |
+ |
|
| 1331 |
+ bic r7, r7, r6 @ r7: vp8_filter |
|
| 1332 |
+ |
|
| 1333 |
+ qsub8 r10, r10, r7 @ u = vp8_signed_char_clamp(qs1 - vp8_filter) |
|
| 1334 |
+ qadd8 r11, r11, r7 @ u = vp8_signed_char_clamp(ps1 + vp8_filter) |
|
| 1335 |
+ eor r10, r10, r12 |
|
| 1336 |
+ eor r11, r11, r12 |
|
| 1337 |
+ |
|
| 1338 |
+ sub r0, r0, r1, lsl #2 |
|
| 1339 |
+ |
|
| 1340 |
+ @we can use TRANSPOSE_MATRIX macro to transpose output - input: q1, q0, p0, p1 |
|
| 1341 |
+ TRANSPOSE_MATRIX r11, r9, r8, r10, lr, r12, r7, r6 |
|
| 1342 |
+ |
|
| 1343 |
+ strh r6, [r0, #-2] @ store the result |
|
| 1344 |
+ mov r6, r6, lsr #16 |
|
| 1345 |
+ strh_post r6, r0, r1 |
|
| 1346 |
+ |
|
| 1347 |
+ strh r7, [r0, #-2] |
|
| 1348 |
+ mov r7, r7, lsr #16 |
|
| 1349 |
+ strh_post r7, r0, r1 |
|
| 1350 |
+ |
|
| 1351 |
+ strh r12, [r0, #-2] |
|
| 1352 |
+ mov r12, r12, lsr #16 |
|
| 1353 |
+ strh_post r12, r0, r1 |
|
| 1354 |
+ |
|
| 1355 |
+ strh lr, [r0, #-2] |
|
| 1356 |
+ mov lr, lr, lsr #16 |
|
| 1357 |
+ strh_post lr, r0, r1 |
|
| 1358 |
+ |
|
| 1359 |
+2: |
|
| 1360 |
+ sub r0, r0, #4 |
|
| 1361 |
+ subs r5, r5, #1 |
|
| 1362 |
+ |
|
| 1363 |
+T ittt ne |
|
| 1364 |
+ ldrne r7, [r0, r1] |
|
| 1365 |
+A ldrne r6, [r0], r1, lsl #1 @ load source data |
|
| 1366 |
+T ldrne r6, [r0] @ load source data |
|
| 1367 |
+T addne r0, r0, r1, lsl #1 |
|
| 1368 |
+T ittt ne |
|
| 1369 |
+ ldrne lr, [r0, r1] |
|
| 1370 |
+A ldrne r8, [r0], r1, lsl #1 |
|
| 1371 |
+T ldrne r8, [r0] |
|
| 1372 |
+T addne r0, r0, r1, lsl #1 |
|
| 1373 |
+ |
|
| 1374 |
+ bne 1b |
|
| 1375 |
+ |
|
| 1376 |
+ add sp, sp, #16 |
|
| 1377 |
+ pop {r4 - r11, pc}
|
|
| 1378 |
+endfunc |
|
| 1379 |
+ |
|
| 1380 |
+@ void vp8_h_loop_filter16(uint8_t *dst, int stride, |
|
| 1381 |
+@ int fE, int fI, int hev_thresh) |
|
| 1382 |
+@ and |
|
| 1383 |
+@ void vp8_h_loop_filter8uv(uint8_t *dstU, uint8_t *dstV, int stride, |
|
| 1384 |
+@ int fE, int fI, int hev_thresh) |
|
| 1385 |
+@ call: |
|
| 1386 |
+@ void vp8_h_loop_filter(uint8_t *dst, int stride, |
|
| 1387 |
+@ int fE, int fI, int hev_thresh, int count) |
|
| 1388 |
+function ff_vp8_h_loop_filter_armv6, export=1 |
|
| 1389 |
+ push {r4 - r11, lr}
|
|
| 1390 |
+ |
|
| 1391 |
+ sub r0, r0, #4 @ move r0 pointer down by 4 |
|
| 1392 |
+ ldr r5, [sp, #40] @ counter |
|
| 1393 |
+ ldr r9, [sp, #36] @ load thresh address |
|
| 1394 |
+ sub sp, sp, #16 @ create temp buffer |
|
| 1395 |
+ |
|
| 1396 |
+ ldr r7, [r0, r1] @ transpose will make it into p3-p0 |
|
| 1397 |
+ ldr_post r6, r0, r1, lsl #1 @ load source data |
|
| 1398 |
+ ldr lr, [r0, r1] |
|
| 1399 |
+ ldr_post r8, r0, r1, lsl #1 |
|
| 1400 |
+ |
|
| 1401 |
+ orr r2, r2, r2, lsl #16 |
|
| 1402 |
+ orr r3, r3, r3, lsl #16 |
|
| 1403 |
+ orr r9, r9, r9, lsl #16 |
|
| 1404 |
+ orr r4, r2, r2, lsl #8 @ flimE splat int -> byte |
|
| 1405 |
+ orr r2, r3, r3, lsl #8 @ flimI splat int -> byte |
|
| 1406 |
+ orr r3, r9, r9, lsl #8 @ thresh splat int -> byte |
|
| 1407 |
+ |
|
| 1408 |
+1: |
|
| 1409 |
+ @ vp8_filter_mask() function |
|
| 1410 |
+ @ calculate breakout conditions |
|
| 1411 |
+ @ transpose the source data for 4-in-parallel operation |
|
| 1412 |
+ TRANSPOSE_MATRIX r6, r7, r8, lr, r12, r11, r10, r9 |
|
| 1413 |
+ |
|
| 1414 |
+ uqsub8 r7, r9, r10 @ p3 - p2 |
|
| 1415 |
+ uqsub8 r8, r10, r9 @ p2 - p3 |
|
| 1416 |
+ uqsub8 r9, r10, r11 @ p2 - p1 |
|
| 1417 |
+ uqsub8 r10, r11, r10 @ p1 - p2 |
|
| 1418 |
+ orr r7, r7, r8 @ abs (p3-p2) |
|
| 1419 |
+ orr r10, r9, r10 @ abs (p2-p1) |
|
| 1420 |
+ uqsub8 lr, r7, r2 @ compare to limit. lr: vp8_filter_mask |
|
| 1421 |
+ uqsub8 r10, r10, r2 @ compare to limit |
|
| 1422 |
+ |
|
| 1423 |
+ sub r0, r0, r1, lsl #2 @ move r0 pointer down by 4 lines |
|
| 1424 |
+ |
|
| 1425 |
+ orr lr, lr, r10 |
|
| 1426 |
+ |
|
| 1427 |
+ uqsub8 r6, r11, r12 @ p1 - p0 |
|
| 1428 |
+ uqsub8 r7, r12, r11 @ p0 - p1 |
|
| 1429 |
+ add r0, r0, #4 @ move r0 pointer up by 4 |
|
| 1430 |
+ orr r6, r6, r7 @ abs (p1-p0) |
|
| 1431 |
+ str r11,[sp, #12] @ save p1 |
|
| 1432 |
+ uqsub8 r10, r6, r2 @ compare to limit |
|
| 1433 |
+ uqsub8 r11, r6, r3 @ compare to thresh |
|
| 1434 |
+ orr lr, lr, r10 |
|
| 1435 |
+ |
|
| 1436 |
+ @ transpose uses 8 regs(r6 - r12 and lr). Need to save reg value now |
|
| 1437 |
+ @ transpose the source data for 4-in-parallel operation |
|
| 1438 |
+ str r11,[sp] @ push r11 to stack |
|
| 1439 |
+ ldr r7, [r0, r1] |
|
| 1440 |
+ ldr_post r6, r0, r1, lsl #1 @ load source data |
|
| 1441 |
+ str r12,[sp, #4] @ save current reg before load q0 - q3 data |
|
| 1442 |
+ str lr, [sp, #8] |
|
| 1443 |
+ ldr lr, [r0, r1] |
|
| 1444 |
+ ldr_post r8, r0, r1, lsl #1 |
|
| 1445 |
+ |
|
| 1446 |
+ TRANSPOSE_MATRIX r6, r7, r8, lr, r12, r11, r10, r9 |
|
| 1447 |
+ |
|
| 1448 |
+ ldr lr, [sp, #8] @ load back (f)limit accumulator |
|
| 1449 |
+ |
|
| 1450 |
+ uqsub8 r6, r12, r11 @ q3 - q2 |
|
| 1451 |
+ uqsub8 r7, r11, r12 @ q2 - q3 |
|
| 1452 |
+ uqsub8 r12, r11, r10 @ q2 - q1 |
|
| 1453 |
+ uqsub8 r11, r10, r11 @ q1 - q2 |
|
| 1454 |
+ orr r6, r6, r7 @ abs (q3-q2) |
|
| 1455 |
+ orr r7, r12, r11 @ abs (q2-q1) |
|
| 1456 |
+ uqsub8 r6, r6, r2 @ compare to limit |
|
| 1457 |
+ uqsub8 r7, r7, r2 @ compare to limit |
|
| 1458 |
+ ldr r11,[sp, #4] @ load back p0 |
|
| 1459 |
+ ldr r12,[sp, #12] @ load back p1 |
|
| 1460 |
+ orr lr, lr, r6 |
|
| 1461 |
+ orr lr, lr, r7 |
|
| 1462 |
+ |
|
| 1463 |
+ uqsub8 r6, r11, r9 @ p0 - q0 |
|
| 1464 |
+ uqsub8 r7, r9, r11 @ q0 - p0 |
|
| 1465 |
+ uqsub8 r8, r12, r10 @ p1 - q1 |
|
| 1466 |
+ uqsub8 r11, r10, r12 @ q1 - p1 |
|
| 1467 |
+ orr r6, r6, r7 @ abs (p0-q0) |
|
| 1468 |
+ ldr r7, c0x7F7F7F7F |
|
| 1469 |
+ orr r8, r8, r11 @ abs (p1-q1) |
|
| 1470 |
+ uqadd8 r6, r6, r6 @ abs (p0-q0) * 2 |
|
| 1471 |
+ and r8, r7, r8, lsr #1 @ abs (p1-q1) / 2 |
|
| 1472 |
+ uqsub8 r11, r10, r9 @ q1 - q0 |
|
| 1473 |
+ uqadd8 r6, r8, r6 @ abs (p0-q0)*2 + abs (p1-q1)/2 |
|
| 1474 |
+ uqsub8 r12, r9, r10 @ q0 - q1 |
|
| 1475 |
+ uqsub8 r6, r6, r4 @ compare to flimit |
|
| 1476 |
+ |
|
| 1477 |
+ orr r9, r11, r12 @ abs (q1-q0) |
|
| 1478 |
+ uqsub8 r8, r9, r2 @ compare to limit |
|
| 1479 |
+ uqsub8 r10, r9, r3 @ compare to thresh |
|
| 1480 |
+ orr lr, lr, r6 |
|
| 1481 |
+ orr lr, lr, r8 |
|
| 1482 |
+ |
|
| 1483 |
+ mvn r11, #0 @ r11 == -1 |
|
| 1484 |
+ mov r12, #0 |
|
| 1485 |
+ |
|
| 1486 |
+ usub8 lr, r12, lr |
|
| 1487 |
+ ldr r9, [sp] @ load the compared result |
|
| 1488 |
+ sel lr, r11, r12 @ filter mask: lr |
|
| 1489 |
+ |
|
| 1490 |
+ cmp lr, #0 |
|
| 1491 |
+ beq 2f @ skip filtering |
|
| 1492 |
+ |
|
| 1493 |
+ |
|
| 1494 |
+ @vp8_hevmask() function |
|
| 1495 |
+ @calculate high edge variance |
|
| 1496 |
+ sub r0, r0, r1, lsl #2 @ move r0 pointer down by 4 lines |
|
| 1497 |
+ |
|
| 1498 |
+ orr r9, r9, r10 |
|
| 1499 |
+ |
|
| 1500 |
+ ldrh r7, [r0, #-2] |
|
| 1501 |
+ ldrh_post r8, r0, r1 |
|
| 1502 |
+ |
|
| 1503 |
+ usub8 r9, r12, r9 |
|
| 1504 |
+ sel r6, r12, r11 @ hev mask: r6 |
|
| 1505 |
+ |
|
| 1506 |
+ |
|
| 1507 |
+ @ vp8_mbfilter() function |
|
| 1508 |
+ @ p2, q2 are only needed at the end. do not need to load them in now. |
|
| 1509 |
+ @ Transpose needs 8 regs(r6 - r12, and lr). Save r6 and lr first |
|
| 1510 |
+ @ load soure data to r6, r11, r12, lr |
|
| 1511 |
+ ldrh r9, [r0, #-2] |
|
| 1512 |
+ ldrh_post r10, r0, r1 |
|
| 1513 |
+ |
|
| 1514 |
+ pkhbt r12, r7, r8, lsl #16 |
|
| 1515 |
+ |
|
| 1516 |
+ ldrh r7, [r0, #-2] |
|
| 1517 |
+ ldrh_post r8, r0, r1 |
|
| 1518 |
+ |
|
| 1519 |
+ pkhbt r11, r9, r10, lsl #16 |
|
| 1520 |
+ |
|
| 1521 |
+ ldrh r9, [r0, #-2] |
|
| 1522 |
+ ldrh_post r10, r0, r1 |
|
| 1523 |
+ |
|
| 1524 |
+ str r6, [sp] @ save r6 |
|
| 1525 |
+ str lr, [sp, #4] @ save lr |
|
| 1526 |
+ |
|
| 1527 |
+ pkhbt r6, r7, r8, lsl #16 |
|
| 1528 |
+ pkhbt lr, r9, r10, lsl #16 |
|
| 1529 |
+ |
|
| 1530 |
+ @transpose r12, r11, r6, lr to p1, p0, q0, q1 |
|
| 1531 |
+ TRANSPOSE_MATRIX r12, r11, r6, lr, r10, r9, r8, r7 |
|
| 1532 |
+ |
|
| 1533 |
+ @load back hev_mask r6 and filter_mask lr |
|
| 1534 |
+ ldr r12, c0x80808080 |
|
| 1535 |
+ ldr r6, [sp] |
|
| 1536 |
+ ldr lr, [sp, #4] |
|
| 1537 |
+ |
|
| 1538 |
+ eor r7, r7, r12 @ ps1 |
|
| 1539 |
+ eor r8, r8, r12 @ ps0 |
|
| 1540 |
+ eor r9, r9, r12 @ qs0 |
|
| 1541 |
+ eor r10, r10, r12 @ qs1 |
|
| 1542 |
+ |
|
| 1543 |
+ qsub8 r12, r9, r8 @ vp8_signed_char_clamp(vp8_filter + 3 * ( qs0 - ps0)) |
|
| 1544 |
+ str r7, [sp, #12] @ store ps1 temporarily |
|
| 1545 |
+ qsub8 r7, r7, r10 @ vp8_signed_char_clamp(ps1-qs1) |
|
| 1546 |
+ str r10,[sp, #8] @ store qs1 temporarily |
|
| 1547 |
+ qadd8 r7, r7, r12 |
|
| 1548 |
+ str r9, [sp] @ store qs0 temporarily |
|
| 1549 |
+ qadd8 r7, r7, r12 |
|
| 1550 |
+ str r8, [sp, #4] @ store ps0 temporarily |
|
| 1551 |
+ qadd8 r7, r7, r12 @ vp8_filter: r7 |
|
| 1552 |
+ |
|
| 1553 |
+ ldr r10, c0x03030303 @ r10 = 3 --modified for vp8 |
|
| 1554 |
+ ldr r9, c0x04040404 |
|
| 1555 |
+ |
|
| 1556 |
+ and r7, r7, lr @ vp8_filter &= mask (lr is free) |
|
| 1557 |
+ |
|
| 1558 |
+ mov r12, r7 @ Filter2: r12 |
|
| 1559 |
+ and r12, r12, r6 @ Filter2 &= hev |
|
| 1560 |
+ |
|
| 1561 |
+ @save bottom 3 bits so that we round one side +4 and the other +3 |
|
| 1562 |
+ qadd8 r8, r12, r9 @ Filter1 (r8) = vp8_signed_char_clamp(Filter2+4) |
|
| 1563 |
+ qadd8 r12, r12, r10 @ Filter2 (r12) = vp8_signed_char_clamp(Filter2+3) |
|
| 1564 |
+ |
|
| 1565 |
+ mov r10, #0 |
|
| 1566 |
+ shadd8 r8, r8, r10 @ Filter1 >>= 3 |
|
| 1567 |
+ shadd8 r12, r12, r10 @ Filter2 >>= 3 |
|
| 1568 |
+ shadd8 r8, r8, r10 |
|
| 1569 |
+ shadd8 r12, r12, r10 |
|
| 1570 |
+ shadd8 r8, r8, r10 @ r8: Filter1 |
|
| 1571 |
+ shadd8 r12, r12, r10 @ r12: Filter2 |
|
| 1572 |
+ |
|
| 1573 |
+ ldr r9, [sp] @ load qs0 |
|
| 1574 |
+ ldr r11,[sp, #4] @ load ps0 |
|
| 1575 |
+ |
|
| 1576 |
+ qsub8 r9, r9, r8 @ qs0 = vp8_signed_char_clamp(qs0 - Filter1) |
|
| 1577 |
+ qadd8 r11, r11, r12 @ ps0 = vp8_signed_char_clamp(ps0 + Filter2) |
|
| 1578 |
+ |
|
| 1579 |
+ bic r12, r7, r6 @vp8_filter &= ~hev ( r6 is free) |
|
| 1580 |
+ |
|
| 1581 |
+ @roughly 3/7th difference across boundary |
|
| 1582 |
+ mov lr, #0x1b @ 27 |
|
| 1583 |
+ mov r7, #0x3f @ 63 |
|
| 1584 |
+ |
|
| 1585 |
+ sxtb16 r6, r12 |
|
| 1586 |
+ sxtb16 r10, r12, ror #8 |
|
| 1587 |
+ smlabb r8, r6, lr, r7 |
|
| 1588 |
+ smlatb r6, r6, lr, r7 |
|
| 1589 |
+ smlabb r7, r10, lr, r7 |
|
| 1590 |
+ smultb r10, r10, lr |
|
| 1591 |
+ ssat r8, #8, r8, asr #7 |
|
| 1592 |
+ ssat r6, #8, r6, asr #7 |
|
| 1593 |
+ add r10, r10, #63 |
|
| 1594 |
+ ssat r7, #8, r7, asr #7 |
|
| 1595 |
+ ssat r10, #8, r10, asr #7 |
|
| 1596 |
+ |
|
| 1597 |
+ ldr lr, c0x80808080 |
|
| 1598 |
+ |
|
| 1599 |
+ pkhbt r6, r8, r6, lsl #16 |
|
| 1600 |
+ pkhbt r10, r7, r10, lsl #16 |
|
| 1601 |
+ uxtb16 r6, r6 |
|
| 1602 |
+ uxtb16 r10, r10 |
|
| 1603 |
+ |
|
| 1604 |
+ sub r0, r0, r1, lsl #2 @ move r0 pointer down by 4 lines |
|
| 1605 |
+ |
|
| 1606 |
+ orr r10, r6, r10, lsl #8 @ u = vp8_signed_char_clamp((63 + Filter2 * 27)>>7) |
|
| 1607 |
+ |
|
| 1608 |
+ qsub8 r8, r9, r10 @ s = vp8_signed_char_clamp(qs0 - u) |
|
| 1609 |
+ qadd8 r10, r11, r10 @ s = vp8_signed_char_clamp(ps0 + u) |
|
| 1610 |
+ eor r8, r8, lr @ *oq0 = s^0x80 |
|
| 1611 |
+ eor r10, r10, lr @ *op0 = s^0x80 |
|
| 1612 |
+ |
|
| 1613 |
+ strb r10,[r0, #-1] @ store op0 result |
|
| 1614 |
+ strb_post r8, r0, r1 @ store oq0 result |
|
| 1615 |
+ mov r10, r10, lsr #8 |
|
| 1616 |
+ mov r8, r8, lsr #8 |
|
| 1617 |
+ strb r10,[r0, #-1] |
|
| 1618 |
+ strb_post r8, r0, r1 |
|
| 1619 |
+ mov r10, r10, lsr #8 |
|
| 1620 |
+ mov r8, r8, lsr #8 |
|
| 1621 |
+ strb r10,[r0, #-1] |
|
| 1622 |
+ strb_post r8, r0, r1 |
|
| 1623 |
+ mov r10, r10, lsr #8 |
|
| 1624 |
+ mov r8, r8, lsr #8 |
|
| 1625 |
+ strb r10,[r0, #-1] |
|
| 1626 |
+ strb_post r8, r0, r1 |
|
| 1627 |
+ |
|
| 1628 |
+ @roughly 2/7th difference across boundary |
|
| 1629 |
+ mov lr, #0x12 @ 18 |
|
| 1630 |
+ mov r7, #0x3f @ 63 |
|
| 1631 |
+ |
|
| 1632 |
+ sxtb16 r6, r12 |
|
| 1633 |
+ sxtb16 r10, r12, ror #8 |
|
| 1634 |
+ smlabb r8, r6, lr, r7 |
|
| 1635 |
+ smlatb r6, r6, lr, r7 |
|
| 1636 |
+ smlabb r9, r10, lr, r7 |
|
| 1637 |
+ smlatb r10, r10, lr, r7 |
|
| 1638 |
+ ssat r8, #8, r8, asr #7 |
|
| 1639 |
+ ssat r6, #8, r6, asr #7 |
|
| 1640 |
+ ssat r9, #8, r9, asr #7 |
|
| 1641 |
+ ssat r10, #8, r10, asr #7 |
|
| 1642 |
+ |
|
| 1643 |
+ sub r0, r0, r1, lsl #2 @ move r0 pointer down by 4 lines |
|
| 1644 |
+ |
|
| 1645 |
+ pkhbt r6, r8, r6, lsl #16 |
|
| 1646 |
+ pkhbt r10, r9, r10, lsl #16 |
|
| 1647 |
+ |
|
| 1648 |
+ ldr r9, [sp, #8] @ load qs1 |
|
| 1649 |
+ ldr r11,[sp, #12] @ load ps1 |
|
| 1650 |
+ ldr lr, c0x80808080 |
|
| 1651 |
+ |
|
| 1652 |
+ uxtb16 r6, r6 |
|
| 1653 |
+ uxtb16 r10, r10 |
|
| 1654 |
+ |
|
| 1655 |
+ add r0, r0, #2 |
|
| 1656 |
+ |
|
| 1657 |
+ orr r10, r6, r10, lsl #8 @ u = vp8_signed_char_clamp((63 + Filter2 * 18)>>7) |
|
| 1658 |
+ |
|
| 1659 |
+ qsub8 r8, r9, r10 @ s = vp8_signed_char_clamp(qs1 - u) |
|
| 1660 |
+ qadd8 r10, r11, r10 @ s = vp8_signed_char_clamp(ps1 + u) |
|
| 1661 |
+ eor r8, r8, lr @ *oq1 = s^0x80 |
|
| 1662 |
+ eor r10, r10, lr @ *op1 = s^0x80 |
|
| 1663 |
+ |
|
| 1664 |
+ ldrb r11,[r0, #-5] @ load p2 for 1/7th difference across boundary |
|
| 1665 |
+ strb r10,[r0, #-4] @ store op1 |
|
| 1666 |
+ strb r8, [r0, #-1] @ store oq1 |
|
| 1667 |
+ ldrb_post r9, r0, r1 @ load q2 for 1/7th difference across boundary |
|
| 1668 |
+ |
|
| 1669 |
+ mov r10, r10, lsr #8 |
|
| 1670 |
+ mov r8, r8, lsr #8 |
|
| 1671 |
+ |
|
| 1672 |
+ ldrb r6, [r0, #-5] |
|
| 1673 |
+ strb r10,[r0, #-4] |
|
| 1674 |
+ strb r8, [r0, #-1] |
|
| 1675 |
+ ldrb_post r7, r0, r1 |
|
| 1676 |
+ |
|
| 1677 |
+ mov r10, r10, lsr #8 |
|
| 1678 |
+ mov r8, r8, lsr #8 |
|
| 1679 |
+ orr r11, r11, r6, lsl #8 |
|
| 1680 |
+ orr r9, r9, r7, lsl #8 |
|
| 1681 |
+ |
|
| 1682 |
+ ldrb r6, [r0, #-5] |
|
| 1683 |
+ strb r10,[r0, #-4] |
|
| 1684 |
+ strb r8, [r0, #-1] |
|
| 1685 |
+ ldrb_post r7, r0, r1 |
|
| 1686 |
+ |
|
| 1687 |
+ mov r10, r10, lsr #8 |
|
| 1688 |
+ mov r8, r8, lsr #8 |
|
| 1689 |
+ orr r11, r11, r6, lsl #16 |
|
| 1690 |
+ orr r9, r9, r7, lsl #16 |
|
| 1691 |
+ |
|
| 1692 |
+ ldrb r6, [r0, #-5] |
|
| 1693 |
+ strb r10,[r0, #-4] |
|
| 1694 |
+ strb r8, [r0, #-1] |
|
| 1695 |
+ ldrb_post r7, r0, r1 |
|
| 1696 |
+ orr r11, r11, r6, lsl #24 |
|
| 1697 |
+ orr r9, r9, r7, lsl #24 |
|
| 1698 |
+ |
|
| 1699 |
+ @roughly 1/7th difference across boundary |
|
| 1700 |
+ eor r9, r9, lr |
|
| 1701 |
+ eor r11, r11, lr |
|
| 1702 |
+ |
|
| 1703 |
+ mov lr, #0x9 @ 9 |
|
| 1704 |
+ mov r7, #0x3f @ 63 |
|
| 1705 |
+ |
|
| 1706 |
+ sxtb16 r6, r12 |
|
| 1707 |
+ sxtb16 r10, r12, ror #8 |
|
| 1708 |
+ smlabb r8, r6, lr, r7 |
|
| 1709 |
+ smlatb r6, r6, lr, r7 |
|
| 1710 |
+ smlabb r12, r10, lr, r7 |
|
| 1711 |
+ smlatb r10, r10, lr, r7 |
|
| 1712 |
+ ssat r8, #8, r8, asr #7 |
|
| 1713 |
+ ssat r6, #8, r6, asr #7 |
|
| 1714 |
+ ssat r12, #8, r12, asr #7 |
|
| 1715 |
+ ssat r10, #8, r10, asr #7 |
|
| 1716 |
+ |
|
| 1717 |
+ sub r0, r0, r1, lsl #2 |
|
| 1718 |
+ |
|
| 1719 |
+ pkhbt r6, r8, r6, lsl #16 |
|
| 1720 |
+ pkhbt r10, r12, r10, lsl #16 |
|
| 1721 |
+ |
|
| 1722 |
+ uxtb16 r6, r6 |
|
| 1723 |
+ uxtb16 r10, r10 |
|
| 1724 |
+ |
|
| 1725 |
+ ldr lr, c0x80808080 |
|
| 1726 |
+ |
|
| 1727 |
+ orr r10, r6, r10, lsl #8 @ u = vp8_signed_char_clamp((63 + Filter2 * 9)>>7) |
|
| 1728 |
+ |
|
| 1729 |
+ qadd8 r8, r11, r10 @ s = vp8_signed_char_clamp(ps2 + u) |
|
| 1730 |
+ qsub8 r10, r9, r10 @ s = vp8_signed_char_clamp(qs2 - u) |
|
| 1731 |
+ eor r8, r8, lr @ *op2 = s^0x80 |
|
| 1732 |
+ eor r10, r10, lr @ *oq2 = s^0x80 |
|
| 1733 |
+ |
|
| 1734 |
+ strb r8, [r0, #-5] @ store *op2 |
|
| 1735 |
+ strb_post r10, r0, r1 @ store *oq2 |
|
| 1736 |
+ mov r8, r8, lsr #8 |
|
| 1737 |
+ mov r10, r10, lsr #8 |
|
| 1738 |
+ strb r8, [r0, #-5] |
|
| 1739 |
+ strb_post r10, r0, r1 |
|
| 1740 |
+ mov r8, r8, lsr #8 |
|
| 1741 |
+ mov r10, r10, lsr #8 |
|
| 1742 |
+ strb r8, [r0, #-5] |
|
| 1743 |
+ strb_post r10, r0, r1 |
|
| 1744 |
+ mov r8, r8, lsr #8 |
|
| 1745 |
+ mov r10, r10, lsr #8 |
|
| 1746 |
+ strb r8, [r0, #-5] |
|
| 1747 |
+ strb_post r10, r0, r1 |
|
| 1748 |
+ |
|
| 1749 |
+ @adjust r0 pointer for next loop |
|
| 1750 |
+ sub r0, r0, #2 |
|
| 1751 |
+ |
|
| 1752 |
+2: |
|
| 1753 |
+ sub r0, r0, #4 |
|
| 1754 |
+ subs r5, r5, #1 |
|
| 1755 |
+ |
|
| 1756 |
+T ittt ne |
|
| 1757 |
+ ldrne r7, [r0, r1] |
|
| 1758 |
+A ldrne r6, [r0], r1, lsl #1 @ load source data |
|
| 1759 |
+T ldrne r6, [r0] |
|
| 1760 |
+T addne r0, r0, r1, lsl #1 |
|
| 1761 |
+T ittt ne |
|
| 1762 |
+ ldrne lr, [r0, r1] |
|
| 1763 |
+A ldrne r8, [r0], r1, lsl #1 |
|
| 1764 |
+T ldrne r8, [r0] |
|
| 1765 |
+T addne r0, r0, r1, lsl #1 |
|
| 1766 |
+ |
|
| 1767 |
+ bne 1b |
|
| 1768 |
+ |
|
| 1769 |
+ add sp, sp, #16 |
|
| 1770 |
+ pop {r4 - r11, pc}
|
|
| 1771 |
+endfunc |
|
| 1772 |
+ |
|
| 1773 |
+@ MC |
|
| 1774 |
+ |
|
| 1775 |
+@ void put_vp8_pixels16(uint8_t *dst, int dststride, uint8_t *src, |
|
| 1776 |
+@ int srcstride, int h, int mx, int my) |
|
| 1777 |
+function ff_put_vp8_pixels16_armv6, export=1 |
|
| 1778 |
+ push {r4 - r11}
|
|
| 1779 |
+ ldr r12,[sp, #32] @ h |
|
| 1780 |
+1: |
|
| 1781 |
+ subs r12, r12, #2 |
|
| 1782 |
+ ldr r5, [r2, #4] |
|
| 1783 |
+ ldr r6, [r2, #8] |
|
| 1784 |
+ ldr r7, [r2, #12] |
|
| 1785 |
+ ldr_post r4, r2, r3 |
|
| 1786 |
+ ldr r9, [r2, #4] |
|
| 1787 |
+ ldr r10,[r2, #8] |
|
| 1788 |
+ ldr r11,[r2, #12] |
|
| 1789 |
+ ldr_post r8, r2, r3 |
|
| 1790 |
+ strd r6, r7, [r0, #8] |
|
| 1791 |
+ strd_post r4, r5, r0, r1 |
|
| 1792 |
+ strd r10, r11,[r0, #8] |
|
| 1793 |
+ strd_post r8, r9, r0, r1 |
|
| 1794 |
+ bgt 1b |
|
| 1795 |
+ pop {r4 - r11}
|
|
| 1796 |
+ bx lr |
|
| 1797 |
+endfunc |
|
| 1798 |
+ |
|
| 1799 |
+@ void put_vp8_pixels8(uint8_t *dst, int dststride, uint8_t *src, |
|
| 1800 |
+@ int srcstride, int h, int mx, int my) |
|
| 1801 |
+function ff_put_vp8_pixels8_armv6, export=1 |
|
| 1802 |
+ push {r4 - r11}
|
|
| 1803 |
+ ldr r12,[sp, #32] @ h |
|
| 1804 |
+1: |
|
| 1805 |
+ subs r12, r12, #4 |
|
| 1806 |
+ ldr r5, [r2, #4] |
|
| 1807 |
+ ldr_post r4, r2, r3 |
|
| 1808 |
+ ldr r7, [r2, #4] |
|
| 1809 |
+ ldr_post r6, r2, r3 |
|
| 1810 |
+ ldr r9, [r2, #4] |
|
| 1811 |
+ ldr_post r8, r2, r3 |
|
| 1812 |
+ ldr r11,[r2, #4] |
|
| 1813 |
+ ldr_post r10, r2, r3 |
|
| 1814 |
+ strd_post r4, r5, r0, r1 |
|
| 1815 |
+ strd_post r6, r7, r0, r1 |
|
| 1816 |
+ strd_post r8, r9, r0, r1 |
|
| 1817 |
+ strd_post r10, r11, r0, r1 |
|
| 1818 |
+ bgt 1b |
|
| 1819 |
+ pop {r4 - r11}
|
|
| 1820 |
+ bx lr |
|
| 1821 |
+endfunc |
|
| 1822 |
+ |
|
| 1823 |
+@ void put_vp8_pixels4(uint8_t *dst, int dststride, uint8_t *src, |
|
| 1824 |
+@ int srcstride, int h, int mx, int my) |
|
| 1825 |
+function ff_put_vp8_pixels4_armv6, export=1 |
|
| 1826 |
+ ldr r12, [sp, #0] @ h |
|
| 1827 |
+ push {r4 - r6, lr}
|
|
| 1828 |
+1: |
|
| 1829 |
+ subs r12, r12, #4 |
|
| 1830 |
+ ldr r5, [r2, r3] |
|
| 1831 |
+ ldr_post r4, r2, r3, lsl #1 |
|
| 1832 |
+ ldr lr, [r2, r3] |
|
| 1833 |
+ ldr_post r6, r2, r3, lsl #1 |
|
| 1834 |
+ str r5, [r0, r1] |
|
| 1835 |
+ str_post r4, r0, r1, lsl #1 |
|
| 1836 |
+ str lr, [r0, r1] |
|
| 1837 |
+ str_post r6, r0, r1, lsl #1 |
|
| 1838 |
+ bgt 1b |
|
| 1839 |
+ pop {r4 - r6, pc}
|
|
| 1840 |
+endfunc |
|
| 1841 |
+ |
|
| 1842 |
+@ note: worst case sum of all 6-tap filter values * 255 is 0x7f80 so 16 bit |
|
| 1843 |
+@ arithmatic can be used to apply filters |
|
| 1844 |
+const sixtap_filters_13245600, align=4 |
|
| 1845 |
+ .short 2, 108, -11, 36, -8, 1, 0, 0 |
|
| 1846 |
+ .short 3, 77, -16, 77, -16, 3, 0, 0 |
|
| 1847 |
+ .short 1, 36, -8, 108, -11, 2, 0, 0 |
|
| 1848 |
+endconst |
|
| 1849 |
+const fourtap_filters_1324, align=4 |
|
| 1850 |
+ .short -6, 12, 123, -1 |
|
| 1851 |
+ .short -9, 50, 93, -6 |
|
| 1852 |
+ .short -6, 93, 50, -9 |
|
| 1853 |
+ .short -1, 123, 12, -6 |
|
| 1854 |
+endconst |
|
| 1855 |
+ |
|
| 1856 |
+@ void put_vp8_epel_h6(uint8_t *dst, int dststride, uint8_t *src, |
|
| 1857 |
+@ int srcstride, int w, int h, int mx) |
|
| 1858 |
+function ff_put_vp8_epel_h6_armv6, export=1 |
|
| 1859 |
+ push {r4 - r11, lr}
|
|
| 1860 |
+ |
|
| 1861 |
+ sub r2, r2, #2 |
|
| 1862 |
+ movrel lr, sixtap_filters_13245600 - 16 |
|
| 1863 |
+ ldr r12,[sp, #44] @ vp8_filter index |
|
| 1864 |
+ ldr r4, [sp, #36] @ width |
|
| 1865 |
+ add lr, lr, r12, lsl #3 |
|
| 1866 |
+ sub r3, r3, r4 @ src_stride - block_width |
|
| 1867 |
+ sub r1, r1, r4 @ dst_stride - block_width |
|
| 1868 |
+ lsr r4, #2 |
|
| 1869 |
+ |
|
| 1870 |
+ str r4, [sp, #36] @ "4-in-parallel" loop counter @40 |
|
| 1871 |
+ str r3, [sp, #44] @ src_stride - block_width @48 |
|
| 1872 |
+ push {r1} @ dst_stride - block_width @0
|
|
| 1873 |
+ @ height @44 |
|
| 1874 |
+ |
|
| 1875 |
+ ldr r1, [lr], #4 @ coefficients |
|
| 1876 |
+ ldr r3, [lr], #4 |
|
| 1877 |
+ ldr lr, [lr] |
|
| 1878 |
+1: |
|
| 1879 |
+ @ 3 loads, 10 shuffles and then mul/acc/add/shr |
|
| 1880 |
+ @ o0: i0/i1/i2/i3/i4/i5 -> i0/i2 (ld1) | i1/i3 (ld1) | i4/i5 (ld2) |
|
| 1881 |
+ @ o1: i1/i2/i3/i4/i5/i6 -> i1/i3 (ld1) | i2/i4 (ld2) | i5/i6 (ld2/3) |
|
| 1882 |
+ @ o2: i2/i3/i4/i5/i6/i7 -> i2/i4 (ld2) | i3/i5 (ld2) | i6/i7 (ld3) |
|
| 1883 |
+ @ o3: i3/i4/i5/i6/i7/i8 -> i3/i5 (ld2) | i4/i6 (ld2/3) | i7/i8 (ld3) |
|
| 1884 |
+ ldr r7, [r2, #5] @ ld3 -> src[5-8] |
|
| 1885 |
+ ldr r6, [r2, #2] @ ld2 -> src[2-5] |
|
| 1886 |
+ ldr r5, [r2], #4 @ ld1 -> src[0-3] |
|
| 1887 |
+ |
|
| 1888 |
+ pkhtb r7, r7, r7, asr #8 @ src[8,7,7,6] |
|
| 1889 |
+ uxtb16 r9, r6, ror #8 @ src[5] | src[3] |
|
| 1890 |
+ uxtb16 r6, r6 @ src[4] | src[2] |
|
| 1891 |
+ uxtb16 r8, r5, ror #8 @ src[3] | src[1] |
|
| 1892 |
+ uxtb16 r11, r7, ror #8 @ src[8] | src[7] |
|
| 1893 |
+ uxtb16 r7, r7 @ src[7] | src[6] |
|
| 1894 |
+ pkhtb r10, r9, r6, asr #16 @ src[5] | src[4] |
|
| 1895 |
+ uxtb16 r5, r5 @ src[2] | src[0] |
|
| 1896 |
+ |
|
| 1897 |
+ smuad r11, r11, lr @ filter[3][2] -> r11 |
|
| 1898 |
+ subs r4, r4, #1 |
|
| 1899 |
+ pkhbt r12, r10, r7, lsl #16 @ src[6] | src[4] |
|
| 1900 |
+ smuad r7, r7, lr @ filter[2][2] -> r7 |
|
| 1901 |
+ smuad r5, r5, r1 @ filter[0][0] -> r5 |
|
| 1902 |
+ smlad r11, r9, r1, r11 @ filter[3][0] -> r11 |
|
| 1903 |
+ smlad r7, r9, r3, r7 @ filter[2][1] -> r7 |
|
| 1904 |
+ smuad r9, r8, r1 @ filter[1][0] -> r9 |
|
| 1905 |
+ smlad r5, r8, r3, r5 @ filter[0][1] -> r5 |
|
| 1906 |
+ pkhtb r8, r12, r10, asr #16 @ src[6] | src[5] |
|
| 1907 |
+ smlad r11, r12, r3, r11 @ filter[3][1] -> r11 |
|
| 1908 |
+ smlad r9, r6, r3, r9 @ filter[1][1] -> r9 |
|
| 1909 |
+ smlad r5, r10, lr, r5 @ filter[0][2] -> r5 |
|
| 1910 |
+ smlad r7, r6, r1, r7 @ filter[2][0] -> r7 |
|
| 1911 |
+ smlad r9, r8, lr, r9 @ filter[1][2] -> r9 |
|
| 1912 |
+ |
|
| 1913 |
+ add r5, r5, #0x40 @ round_shift_and_clamp[0] |
|
| 1914 |
+ add r9, r9, #0x40 @ round_shift_and_clamp[1] |
|
| 1915 |
+ add r7, r7, #0x40 @ round_shift_and_clamp[2] |
|
| 1916 |
+ add r11, r11, #0x40 @ round_shift_and_clamp[3] |
|
| 1917 |
+ |
|
| 1918 |
+ usat r5, #8, r5, asr #7 |
|
| 1919 |
+ usat r9, #8, r9, asr #7 |
|
| 1920 |
+ usat r7, #8, r7, asr #7 |
|
| 1921 |
+ usat r11, #8, r11, asr #7 |
|
| 1922 |
+ |
|
| 1923 |
+ strb r5, [r0], #1 @ store res[0] |
|
| 1924 |
+ strb r9, [r0], #1 @ store res[1] |
|
| 1925 |
+ strb r7, [r0], #1 @ store res[2] |
|
| 1926 |
+ strb r11,[r0], #1 @ store res[3] |
|
| 1927 |
+ |
|
| 1928 |
+ bne 1b |
|
| 1929 |
+ |
|
| 1930 |
+ ldr r12,[sp, #44] @ height = outer-loop counter |
|
| 1931 |
+ subs r12, r12, #1 |
|
| 1932 |
+T itttt ne |
|
| 1933 |
+ ldrne r4, [sp, #40] @ 4-in-parallel loop counter |
|
| 1934 |
+ ldrne r5, [sp, #48] |
|
| 1935 |
+ ldrne r6, [sp] |
|
| 1936 |
+ strne r12,[sp, #44] |
|
| 1937 |
+ add r2, r2, r5 @ move to next input/output lines |
|
| 1938 |
+ add r0, r0, r6 |
|
| 1939 |
+ |
|
| 1940 |
+ bne 1b |
|
| 1941 |
+ |
|
| 1942 |
+ add sp, sp, #4 @ restore stack after push{r1} above
|
|
| 1943 |
+ pop {r4 - r11, pc}
|
|
| 1944 |
+endfunc |
|
| 1945 |
+ |
|
| 1946 |
+@ void put_vp8_epel_v6(uint8_t *dst, int dststride, uint8_t *src, |
|
| 1947 |
+@ int srcstride, int w, int h, int my) |
|
| 1948 |
+function ff_put_vp8_epel_v6_armv6, export=1 |
|
| 1949 |
+ push {r4 - r11, lr}
|
|
| 1950 |
+ |
|
| 1951 |
+ movrel lr, sixtap_filters_13245600 - 16 |
|
| 1952 |
+ ldr r12,[sp, #44] @ vp8_filter index |
|
| 1953 |
+ ldr r4, [sp, #36] @ width |
|
| 1954 |
+ add lr, lr, r12, lsl #3 |
|
| 1955 |
+ sub r1, r1, r4 @ dst_stride - block_width |
|
| 1956 |
+ lsr r4, #2 |
|
| 1957 |
+ |
|
| 1958 |
+ str r4, [sp, #36] @ "4-in-parallel" loop counter @40 |
|
| 1959 |
+ str r3, [sp, #44] @ src_stride - block_width @48 |
|
| 1960 |
+ push {r1} @ dst_stride - block_width @0
|
|
| 1961 |
+ @ height @44 |
|
| 1962 |
+1: |
|
| 1963 |
+ add r1, r3, r3, lsl #1 @ stride * 3 |
|
| 1964 |
+ ldr_dpren r5, r2, r3 @ src[0,1,2,3 + stride * 1] |
|
| 1965 |
+ ldr r6, [r2, r3] @ src[0,1,2,3 + stride * 3] |
|
| 1966 |
+ ldr r7, [r2, r3, lsl #1] @ src[0,1,2,3 + stride * 4] |
|
| 1967 |
+ ldr r8, [r2, r1] @ src[0,1,2,3 + stride * 5] |
|
| 1968 |
+ |
|
| 1969 |
+ @ byte -> word and "transpose" |
|
| 1970 |
+ uxtb16 r9, r5, ror #8 @ src[3 + stride*1] | src[1 + stride*1] |
|
| 1971 |
+ uxtb16 r10, r6, ror #8 @ src[3 + stride*3] | src[1 + stride*3] |
|
| 1972 |
+ uxtb16 r11, r7, ror #8 @ src[3 + stride*4] | src[1 + stride*4] |
|
| 1973 |
+ uxtb16 r12, r8, ror #8 @ src[3 + stride*5] | src[1 + stride*5] |
|
| 1974 |
+ uxtb16 r5, r5 @ src[2 + stride*1] | src[0 + stride*1] |
|
| 1975 |
+ uxtb16 r6, r6 @ src[2 + stride*3] | src[0 + stride*3] |
|
| 1976 |
+ uxtb16 r7, r7 @ src[2 + stride*4] | src[0 + stride*4] |
|
| 1977 |
+ uxtb16 r8, r8 @ src[2 + stride*5] | src[0 + stride*5] |
|
| 1978 |
+ pkhbt r1, r9, r10, lsl #16 @ src[1 + stride*3] | src[1 + stride*1] |
|
| 1979 |
+ pkhtb r9, r10, r9, asr #16 @ src[3 + stride*3] | src[3 + stride*1] |
|
| 1980 |
+ pkhbt r10, r11, r12, lsl #16 @ src[1 + stride*5] | src[1 + stride*4] |
|
| 1981 |
+ pkhtb r11, r12, r11, asr #16 @ src[3 + stride*5] | src[3 + stride*4] |
|
| 1982 |
+ pkhbt r12, r5, r6, lsl #16 @ src[0 + stride*3] | src[0 + stride*1] |
|
| 1983 |
+ pkhtb r5, r6, r5, asr #16 @ src[2 + stride*3] | src[2 + stride*1] |
|
| 1984 |
+ pkhbt r6, r7, r8, lsl #16 @ src[0 + stride*5] | src[0 + stride*4] |
|
| 1985 |
+ pkhtb r7, r8, r7, asr #16 @ src[2 + stride*5] | src[2 + stride*4] |
|
| 1986 |
+ |
|
| 1987 |
+ ldr r8, [lr, #4] @ stall - if only I had more registers... |
|
| 1988 |
+ smuad r12, r12, r8 @ filter[0][1] |
|
| 1989 |
+ smuad r1, r1, r8 @ filter[1][1] |
|
| 1990 |
+ smuad r5, r5, r8 @ filter[2][1] |
|
| 1991 |
+ smuad r9, r9, r8 @ filter[3][1] |
|
| 1992 |
+ ldr r8, [lr, #8] @ stall - if only I had more registers... |
|
| 1993 |
+ smlad r12, r6, r8, r12 @ filter[0][2] |
|
| 1994 |
+ smlad r1, r10, r8, r1 @ filter[1][2] |
|
| 1995 |
+ ldr_dpren r6, r2, r3, lsl #1 @ src[0,1,2,3 + stride * 0] |
|
| 1996 |
+ ldr r10,[r2], #4 @ src[0,1,2,3 + stride * 2] |
|
| 1997 |
+ smlad r5, r7, r8, r5 @ filter[2][2] |
|
| 1998 |
+ smlad r9, r11, r8, r9 @ filter[3][2] |
|
| 1999 |
+ |
|
| 2000 |
+ uxtb16 r7, r6, ror #8 @ src[3 + stride*0] | src[1 + stride*0] |
|
| 2001 |
+ uxtb16 r11, r10, ror #8 @ src[3 + stride*2] | src[1 + stride*2] |
|
| 2002 |
+ uxtb16 r6, r6 @ src[2 + stride*0] | src[0 + stride*0] |
|
| 2003 |
+ uxtb16 r10, r10 @ src[2 + stride*2] | src[0 + stride*2] |
|
| 2004 |
+ |
|
| 2005 |
+ pkhbt r8, r7, r11, lsl #16 @ src[1 + stride*2] | src[1 + stride*0] |
|
| 2006 |
+ pkhtb r7, r11, r7, asr #16 @ src[3 + stride*2] | src[3 + stride*0] |
|
| 2007 |
+ pkhbt r11, r6, r10, lsl #16 @ src[0 + stride*2] | src[0 + stride*0] |
|
| 2008 |
+ pkhtb r6, r10, r6, asr #16 @ src[2 + stride*2] | src[2 + stride*0] |
|
| 2009 |
+ |
|
| 2010 |
+ ldr r10,[lr] @ stall - if only I had more registers... |
|
| 2011 |
+ subs r4, r4, #1 @ counter-- |
|
| 2012 |
+ smlad r12, r11, r10, r12 @ filter[0][0] |
|
| 2013 |
+ smlad r1, r8, r10, r1 @ filter[1][0] |
|
| 2014 |
+ smlad r5, r6, r10, r5 @ filter[2][0] |
|
| 2015 |
+ smlad r9, r7, r10, r9 @ filter[3][0] |
|
| 2016 |
+ |
|
| 2017 |
+ add r12, r12, #0x40 @ round_shift_and_clamp[0] |
|
| 2018 |
+ add r1, r1, #0x40 @ round_shift_and_clamp[1] |
|
| 2019 |
+ add r5, r5, #0x40 @ round_shift_and_clamp[2] |
|
| 2020 |
+ add r9, r9, #0x40 @ round_shift_and_clamp[3] |
|
| 2021 |
+ |
|
| 2022 |
+ usat r12, #8, r12, asr #7 |
|
| 2023 |
+ usat r1, #8, r1, asr #7 |
|
| 2024 |
+ usat r5, #8, r5, asr #7 |
|
| 2025 |
+ usat r9, #8, r9, asr #7 |
|
| 2026 |
+ |
|
| 2027 |
+ strb r12,[r0], #1 @ store res[0] |
|
| 2028 |
+ strb r1, [r0], #1 @ store res[1] |
|
| 2029 |
+ strb r5, [r0], #1 @ store res[2] |
|
| 2030 |
+ strb r9, [r0], #1 @ store res[3] |
|
| 2031 |
+ |
|
| 2032 |
+ bne 1b |
|
| 2033 |
+ |
|
| 2034 |
+ ldr r12,[sp, #44] @ height = outer-loop counter |
|
| 2035 |
+ subs r12, r12, #1 |
|
| 2036 |
+T itttt ne |
|
| 2037 |
+ ldrne r4, [sp, #40] @ 4-in-parallel loop counter |
|
| 2038 |
+ ldrne r6, [sp, #0] |
|
| 2039 |
+ subne r2, r2, r4, lsl #2 |
|
| 2040 |
+ strne r12,[sp, #44] |
|
| 2041 |
+ add r0, r0, r6 |
|
| 2042 |
+ add r2, r2, r3 @ move to next input/output lines |
|
| 2043 |
+ |
|
| 2044 |
+ bne 1b |
|
| 2045 |
+ |
|
| 2046 |
+ add sp, sp, #4 @ restore stack after push{r1} above
|
|
| 2047 |
+ pop {r4 - r11, pc}
|
|
| 2048 |
+endfunc |
|
| 2049 |
+ |
|
| 2050 |
+@ void put_vp8_epel_h4(uint8_t *dst, int dststride, uint8_t *src, |
|
| 2051 |
+@ int srcstride, int w, int h, int mx) |
|
| 2052 |
+function ff_put_vp8_epel_h4_armv6, export=1 |
|
| 2053 |
+ push {r4 - r11, lr}
|
|
| 2054 |
+ |
|
| 2055 |
+ subs r2, r2, #1 |
|
| 2056 |
+ movrel lr, fourtap_filters_1324 - 4 |
|
| 2057 |
+ ldr r4, [sp, #36] @ width |
|
| 2058 |
+ ldr r12,[sp, #44] @ vp8_filter index |
|
| 2059 |
+ add lr, lr, r12, lsl #2 |
|
| 2060 |
+ sub r3, r3, r4 @ src_stride - block_width |
|
| 2061 |
+ sub r1, r1, r4 @ dst_stride - block_width |
|
| 2062 |
+ ldr r5, [lr] |
|
| 2063 |
+ ldr r6, [lr, #4] |
|
| 2064 |
+ asr r4, #2 |
|
| 2065 |
+ |
|
| 2066 |
+ ldr lr, [sp, #40] @ height = outer-loop counter |
|
| 2067 |
+ str r4, [sp, #36] @ "4-in-parallel" inner loop counter |
|
| 2068 |
+1: |
|
| 2069 |
+ @ 3 loads, 5 uxtb16s and then mul/acc/add/shr |
|
| 2070 |
+ @ o0: i0/i1/i2/i3 -> i0/i2(ld1) + i1/i3(ld1) |
|
| 2071 |
+ @ o1: i1/i2/i3/i4 -> i1/i3(ld1) + i2/i4(ld2) |
|
| 2072 |
+ @ o2: i2/i3/i4/i5 -> i2/i4(ld2) + i3/i5(ld2) |
|
| 2073 |
+ @ o3: i3/i4/i5/i6 -> i3/i5(ld2) + i4/i6(ld3) |
|
| 2074 |
+ ldr r9, [r2, #3] @ load source data |
|
| 2075 |
+ ldr r8, [r2, #2] |
|
| 2076 |
+ ldr r7, [r2], #4 |
|
| 2077 |
+ |
|
| 2078 |
+ uxtb16 r9, r9, ror #8 @ src[6] | src[4] |
|
| 2079 |
+ uxtb16 r10, r8, ror #8 @ src[5] | src[3] |
|
| 2080 |
+ uxtb16 r8, r8 @ src[4] | src[2] |
|
| 2081 |
+ uxtb16 r11, r7, ror #8 @ src[3] | src[1] |
|
| 2082 |
+ uxtb16 r7, r7 @ src[2] | src[0] |
|
| 2083 |
+ |
|
| 2084 |
+ smuad r9, r9, r6 @ filter[3][1] -> r9 |
|
| 2085 |
+ smuad r12, r10, r6 @ filter[2][1] -> r12 |
|
| 2086 |
+ smuad r7, r7, r5 @ filter[0][0] -> r7 |
|
| 2087 |
+ smlad r9, r10, r5, r9 @ filter[3][0] -> r9 |
|
| 2088 |
+ smuad r10, r11, r5 @ filter[1][0] -> r10 |
|
| 2089 |
+ smlad r12, r8, r5, r12 @ filter[2][0] -> r12 |
|
| 2090 |
+ smlad r7, r11, r6, r7 @ filter[0][1] -> r7 |
|
| 2091 |
+ smlad r10, r8, r6, r10 @ filter[1][1] -> r10 |
|
| 2092 |
+ |
|
| 2093 |
+ subs r4, r4, #1 @ counter-- |
|
| 2094 |
+ |
|
| 2095 |
+ add r7, r7, #0x40 @ round_shift_and_clamp[0] |
|
| 2096 |
+ add r10, r10, #0x40 @ round_shift_and_clamp[1] |
|
| 2097 |
+ add r12, r12, #0x40 @ round_shift_and_clamp[2] |
|
| 2098 |
+ add r9, r9, #0x40 @ round_shift_and_clamp[3] |
|
| 2099 |
+ |
|
| 2100 |
+ usat r7, #8, r7, asr #7 |
|
| 2101 |
+ usat r10, #8, r10, asr #7 |
|
| 2102 |
+ usat r12, #8, r12, asr #7 |
|
| 2103 |
+ usat r9, #8, r9, asr #7 |
|
| 2104 |
+ |
|
| 2105 |
+ strb r7, [r0], #1 @ store res[0] |
|
| 2106 |
+ strb r10,[r0], #1 @ store res[1] |
|
| 2107 |
+ strb r12,[r0], #1 @ store res[2] |
|
| 2108 |
+ strb r9, [r0], #1 @ store res[3] |
|
| 2109 |
+ |
|
| 2110 |
+ bne 1b |
|
| 2111 |
+ |
|
| 2112 |
+ subs lr, lr, #1 |
|
| 2113 |
+T it ne |
|
| 2114 |
+ ldrne r4, [sp, #36] @ 4-in-parallel loop counter |
|
| 2115 |
+ add r2, r2, r3 @ move to next input/output lines |
|
| 2116 |
+ add r0, r0, r1 |
|
| 2117 |
+ |
|
| 2118 |
+ bne 1b |
|
| 2119 |
+ |
|
| 2120 |
+ pop {r4 - r11, pc}
|
|
| 2121 |
+endfunc |
|
| 2122 |
+ |
|
| 2123 |
+@ void put_vp8_epel_v4(uint8_t *dst, int dststride, uint8_t *src, |
|
| 2124 |
+@ int srcstride, int w, int h, int my) |
|
| 2125 |
+function ff_put_vp8_epel_v4_armv6, export=1 |
|
| 2126 |
+ push {r4 - r11, lr}
|
|
| 2127 |
+ |
|
| 2128 |
+ movrel lr, fourtap_filters_1324 - 4 |
|
| 2129 |
+ ldr r12,[sp, #44] @ vp8_filter index |
|
| 2130 |
+ ldr r4, [sp, #36] @ width |
|
| 2131 |
+ add lr, lr, r12, lsl #2 |
|
| 2132 |
+ sub r1, r1, r4 @ dst_stride - block_width |
|
| 2133 |
+ asr r4, #2 |
|
| 2134 |
+ ldr r5, [lr] |
|
| 2135 |
+ ldr r6, [lr, #4] |
|
| 2136 |
+ |
|
| 2137 |
+ str r4, [sp, #36] @ "4-in-parallel" loop counter @40 |
|
| 2138 |
+ str r3, [sp, #44] @ src_stride @48 |
|
| 2139 |
+ push {r1} @ dst_stride - block_width @36
|
|
| 2140 |
+ @ height @44 |
|
| 2141 |
+1: |
|
| 2142 |
+ ldr lr, [r2, r3, lsl #1] @ load source pixels |
|
| 2143 |
+ ldr r12,[r2, r3] |
|
| 2144 |
+ ldr_dpren r7, r2, r3 |
|
| 2145 |
+ ldr r11,[r2], #4 |
|
| 2146 |
+ |
|
| 2147 |
+ @ byte -> word and "transpose" |
|
| 2148 |
+ uxtb16 r8, lr, ror #8 @ src[3 + stride*3] | src[1 + stride*3] |
|
| 2149 |
+ uxtb16 r9, r12, ror #8 @ src[3 + stride*2] | src[1 + stride*2] |
|
| 2150 |
+ uxtb16 r3, r7, ror #8 @ src[3 + stride*0] | src[1 + stride*0] |
|
| 2151 |
+ uxtb16 r1, r11, ror #8 @ src[3 + stride*1] | src[1 + stride*1] |
|
| 2152 |
+ uxtb16 lr, lr @ src[2 + stride*3] | src[0 + stride*3] |
|
| 2153 |
+ uxtb16 r12, r12 @ src[2 + stride*2] | src[0 + stride*2] |
|
| 2154 |
+ uxtb16 r7, r7 @ src[2 + stride*0] | src[0 + stride*0] |
|
| 2155 |
+ uxtb16 r11, r11 @ src[2 + stride*1] | src[0 + stride*1] |
|
| 2156 |
+ pkhbt r10, r1, r8, lsl #16 @ src[1 + stride*3] | src[1 + stride*1] |
|
| 2157 |
+ pkhtb r1, r8, r1, asr #16 @ src[3 + stride*3] | src[3 + stride*1] |
|
| 2158 |
+ pkhbt r8, r3, r9, lsl #16 @ src[1 + stride*2] | src[1 + stride*0] |
|
| 2159 |
+ pkhtb r3, r9, r3, asr #16 @ src[3 + stride*2] | src[3 + stride*0] |
|
| 2160 |
+ pkhbt r9, r11, lr, lsl #16 @ src[0 + stride*3] | src[0 + stride*1] |
|
| 2161 |
+ pkhtb r11, lr, r11, asr #16 @ src[2 + stride*3] | src[2 + stride*1] |
|
| 2162 |
+ pkhbt lr, r7, r12, lsl #16 @ src[0 + stride*2] | src[0 + stride*0] |
|
| 2163 |
+ pkhtb r7, r12, r7, asr #16 @ src[2 + stride*2] | src[2 + stride*0] |
|
| 2164 |
+ |
|
| 2165 |
+ smuad r9, r9, r6 @ filter[0][1] |
|
| 2166 |
+ smuad r10, r10, r6 @ filter[1][1] |
|
| 2167 |
+ smuad r11, r11, r6 @ filter[2][1] |
|
| 2168 |
+ smuad r1, r1, r6 @ filter[3][1] |
|
| 2169 |
+ smlad r9, lr, r5, r9 @ filter[0][0] |
|
| 2170 |
+ smlad r10, r8, r5, r10 @ filter[1][0] |
|
| 2171 |
+ smlad r11, r7, r5, r11 @ filter[2][0] |
|
| 2172 |
+ smlad r1, r3, r5, r1 @ filter[3][0] |
|
| 2173 |
+ |
|
| 2174 |
+ subs r4, r4, #1 @ counter-- |
|
| 2175 |
+ ldr r3, [sp, #48] @ FIXME prevent clobber of r3 above? |
|
| 2176 |
+ |
|
| 2177 |
+ add r9, r9, #0x40 @ round_shift_and_clamp[0] |
|
| 2178 |
+ add r10, r10, #0x40 @ round_shift_and_clamp[1] |
|
| 2179 |
+ add r11, r11, #0x40 @ round_shift_and_clamp[2] |
|
| 2180 |
+ add r1, r1, #0x40 @ round_shift_and_clamp[3] |
|
| 2181 |
+ |
|
| 2182 |
+ usat r9, #8, r9, asr #7 |
|
| 2183 |
+ usat r10, #8, r10, asr #7 |
|
| 2184 |
+ usat r11, #8, r11, asr #7 |
|
| 2185 |
+ usat r1, #8, r1, asr #7 |
|
| 2186 |
+ |
|
| 2187 |
+ strb r9, [r0], #1 @ store result |
|
| 2188 |
+ strb r10,[r0], #1 |
|
| 2189 |
+ strb r11,[r0], #1 |
|
| 2190 |
+ strb r1, [r0], #1 |
|
| 2191 |
+ |
|
| 2192 |
+ bne 1b |
|
| 2193 |
+ |
|
| 2194 |
+ ldr r12,[sp, #44] @ height = outer-loop counter |
|
| 2195 |
+ subs r12, r12, #1 |
|
| 2196 |
+T ittt ne |
|
| 2197 |
+ ldrne r4, [sp, #40] @ 4-in-parallel loop counter |
|
| 2198 |
+ ldrne r9, [sp, #0] |
|
| 2199 |
+ strne r12,[sp, #44] |
|
| 2200 |
+ sub r2, r2, r4, lsl #2 |
|
| 2201 |
+ add r0, r0, r9 |
|
| 2202 |
+ add r2, r2, r3 @ move to next input/output lines |
|
| 2203 |
+ |
|
| 2204 |
+ bne 1b |
|
| 2205 |
+ |
|
| 2206 |
+ add sp, sp, #4 @ restore stack after push{r1} above
|
|
| 2207 |
+ pop {r4 - r11, pc}
|
|
| 2208 |
+endfunc |
|
| 2209 |
+ |
|
| 2210 |
+@ void put_vp8_bilin_h(uint8_t *dst, int dststride, uint8_t *src, |
|
| 2211 |
+@ int srcstride, int w, int h, int mx) |
|
| 2212 |
+function ff_put_vp8_bilin_h_armv6, export=1 |
|
| 2213 |
+ push {r4 - r9, lr}
|
|
| 2214 |
+ |
|
| 2215 |
+ ldr r8, [sp, #36] @ vp8_filter index |
|
| 2216 |
+ ldr r12,[sp, #32] @ height = outer-loop counter |
|
| 2217 |
+ ldr r4, [sp, #28] @ width |
|
| 2218 |
+ lsl r5, r8, #16 @ mx << 16 |
|
| 2219 |
+ sub r3, r3, r4 @ src_stride - block_width |
|
| 2220 |
+ sub r1, r1, r4 @ dst_stride - block_width |
|
| 2221 |
+ asr r4, #2 |
|
| 2222 |
+ sub r5, r5, r8 @ (mx << 16) | (-mx) |
|
| 2223 |
+ str r4, [sp, #28] @ "4-in-parallel" loop counter |
|
| 2224 |
+ add r5, r5, #8 @ (8 - mx) | (mx << 16) = filter coefficients |
|
| 2225 |
+1: |
|
| 2226 |
+ ldrb r6, [r2], #1 @ load source data |
|
| 2227 |
+ ldrb r7, [r2], #1 |
|
| 2228 |
+ ldrb r8, [r2], #1 |
|
| 2229 |
+ ldrb r9, [r2], #1 |
|
| 2230 |
+ ldrb lr, [r2] |
|
| 2231 |
+ |
|
| 2232 |
+ pkhbt r6, r6, r7, lsl #16 @ src[1] | src[0] |
|
| 2233 |
+ pkhbt r7, r7, r8, lsl #16 @ src[2] | src[1] |
|
| 2234 |
+ pkhbt r8, r8, r9, lsl #16 @ src[3] | src[2] |
|
| 2235 |
+ pkhbt r9, r9, lr, lsl #16 @ src[4] | src[3] |
|
| 2236 |
+ |
|
| 2237 |
+ smuad r6, r6, r5 @ apply the filter |
|
| 2238 |
+ smuad r7, r7, r5 |
|
| 2239 |
+ smuad r8, r8, r5 |
|
| 2240 |
+ smuad r9, r9, r5 |
|
| 2241 |
+ |
|
| 2242 |
+ subs r4, r4, #1 @ counter-- |
|
| 2243 |
+ |
|
| 2244 |
+ add r6, r6, #0x4 @ round_shift_and_clamp |
|
| 2245 |
+ add r7, r7, #0x4 |
|
| 2246 |
+ add r8, r8, #0x4 |
|
| 2247 |
+ add r9, r9, #0x4 |
|
| 2248 |
+ |
|
| 2249 |
+ asr r6, #3 |
|
| 2250 |
+ asr r7, #3 |
|
| 2251 |
+ pkhbt r6, r6, r8, lsl #13 |
|
| 2252 |
+ pkhbt r7, r7, r9, lsl #13 |
|
| 2253 |
+ orr r6, r6, r7, lsl #8 |
|
| 2254 |
+ str r6, [r0], #4 @ store result |
|
| 2255 |
+ |
|
| 2256 |
+ bne 1b |
|
| 2257 |
+ |
|
| 2258 |
+ ldr r4, [sp, #28] @ 4-in-parallel loop counter |
|
| 2259 |
+ subs r12, r12, #1 |
|
| 2260 |
+ |
|
| 2261 |
+ add r2, r2, r3 @ move to next input/output lines |
|
| 2262 |
+ add r0, r0, r1 |
|
| 2263 |
+ |
|
| 2264 |
+ bne 1b |
|
| 2265 |
+ |
|
| 2266 |
+ pop {r4 - r9, pc}
|
|
| 2267 |
+endfunc |
|
| 2268 |
+ |
|
| 2269 |
+@ void put_vp8_bilin_v(uint8_t *dst, int dststride, uint8_t *src, |
|
| 2270 |
+@ int srcstride, int w, int h, int my) |
|
| 2271 |
+function ff_put_vp8_bilin_v_armv6, export=1 |
|
| 2272 |
+ push {r4 - r11, lr}
|
|
| 2273 |
+ |
|
| 2274 |
+ ldr r11,[sp, #44] @ vp8_filter index |
|
| 2275 |
+ ldr r4, [sp, #36] @ width |
|
| 2276 |
+ mov r5, r11, lsl #16 @ mx << 16 |
|
| 2277 |
+ ldr r12,[sp, #40] @ height = outer-loop counter |
|
| 2278 |
+ sub r1, r1, r4 |
|
| 2279 |
+ sub r5, r5, r11 @ (mx << 16) | (-mx) |
|
| 2280 |
+ asr r4, #2 |
|
| 2281 |
+ add r5, r5, #8 @ (8 - mx) | (mx << 16) = filter coefficients |
|
| 2282 |
+ str r4, [sp, #36] @ "4-in-parallel" loop counter |
|
| 2283 |
+1: |
|
| 2284 |
+ ldrb r10,[r2, r3] @ load the data |
|
| 2285 |
+ ldrb r6, [r2], #1 |
|
| 2286 |
+ ldrb r11,[r2, r3] |
|
| 2287 |
+ ldrb r7, [r2], #1 |
|
| 2288 |
+ ldrb lr, [r2, r3] |
|
| 2289 |
+ ldrb r8, [r2], #1 |
|
| 2290 |
+ ldrb r9, [r2, r3] |
|
| 2291 |
+ pkhbt r6, r6, r10, lsl #16 |
|
| 2292 |
+ ldrb r10,[r2], #1 |
|
| 2293 |
+ pkhbt r7, r7, r11, lsl #16 |
|
| 2294 |
+ pkhbt r8, r8, lr, lsl #16 |
|
| 2295 |
+ pkhbt r9, r10, r9, lsl #16 |
|
| 2296 |
+ |
|
| 2297 |
+ smuad r6, r6, r5 @ apply the filter |
|
| 2298 |
+ smuad r7, r7, r5 |
|
| 2299 |
+ smuad r8, r8, r5 |
|
| 2300 |
+ smuad r9, r9, r5 |
|
| 2301 |
+ |
|
| 2302 |
+ subs r4, r4, #1 @ counter-- |
|
| 2303 |
+ |
|
| 2304 |
+ add r6, r6, #0x4 @ round_shift_and_clamp |
|
| 2305 |
+ add r7, r7, #0x4 |
|
| 2306 |
+ add r8, r8, #0x4 |
|
| 2307 |
+ add r9, r9, #0x4 |
|
| 2308 |
+ |
|
| 2309 |
+ asr r6, #3 |
|
| 2310 |
+ asr r7, #3 |
|
| 2311 |
+ pkhbt r6, r6, r8, lsl #13 |
|
| 2312 |
+ pkhbt r7, r7, r9, lsl #13 |
|
| 2313 |
+ orr r6, r6, r7, lsl #8 |
|
| 2314 |
+ str r6, [r0], #4 @ store result |
|
| 2315 |
+ |
|
| 2316 |
+ bne 1b |
|
| 2317 |
+ |
|
| 2318 |
+ ldr r4, [sp, #36] @ 4-in-parallel loop counter |
|
| 2319 |
+ subs r12, r12, #1 |
|
| 2320 |
+ |
|
| 2321 |
+ add r2, r2, r3 @ move to next input/output lines |
|
| 2322 |
+ add r0, r0, r1 |
|
| 2323 |
+ sub r2, r2, r4, lsl #2 |
|
| 2324 |
+ |
|
| 2325 |
+ bne 1b |
|
| 2326 |
+ pop {r4 - r11, pc}
|
|
| 2327 |
+endfunc |
| ... | ... |
@@ -19,13 +19,17 @@ |
| 19 | 19 |
#include <stdint.h> |
| 20 | 20 |
#include "libavcodec/vp8dsp.h" |
| 21 | 21 |
|
| 22 |
-void ff_vp8_luma_dc_wht_neon(DCTELEM block[4][4][16], DCTELEM dc[16]); |
|
| 23 |
-void ff_vp8_luma_dc_wht_dc_neon(DCTELEM block[4][4][16], DCTELEM dc[16]); |
|
| 22 |
+void ff_vp8_luma_dc_wht_dc_armv6(DCTELEM block[4][4][16], DCTELEM dc[16]); |
|
| 24 | 23 |
|
| 25 |
-void ff_vp8_idct_add_neon(uint8_t *dst, DCTELEM block[16], int stride); |
|
| 26 |
-void ff_vp8_idct_dc_add_neon(uint8_t *dst, DCTELEM block[16], int stride); |
|
| 27 |
-void ff_vp8_idct_dc_add4y_neon(uint8_t *dst, DCTELEM block[4][16], int stride); |
|
| 28 |
-void ff_vp8_idct_dc_add4uv_neon(uint8_t *dst, DCTELEM block[4][16], int stride); |
|
| 24 |
+#define idct_funcs(opt) \ |
|
| 25 |
+void ff_vp8_luma_dc_wht_ ## opt(DCTELEM block[4][4][16], DCTELEM dc[16]); \ |
|
| 26 |
+void ff_vp8_idct_add_ ## opt(uint8_t *dst, DCTELEM block[16], int stride); \ |
|
| 27 |
+void ff_vp8_idct_dc_add_ ## opt(uint8_t *dst, DCTELEM block[16], int stride); \ |
|
| 28 |
+void ff_vp8_idct_dc_add4y_ ## opt(uint8_t *dst, DCTELEM block[4][16], int stride); \ |
|
| 29 |
+void ff_vp8_idct_dc_add4uv_ ## opt(uint8_t *dst, DCTELEM block[4][16], int stride) |
|
| 30 |
+ |
|
| 31 |
+idct_funcs(neon); |
|
| 32 |
+idct_funcs(armv6); |
|
| 29 | 33 |
|
| 30 | 34 |
void ff_vp8_v_loop_filter16_neon(uint8_t *dst, int stride, |
| 31 | 35 |
int flim_E, int flim_I, int hev_thresh); |
| ... | ... |
@@ -47,29 +51,106 @@ void ff_vp8_h_loop_filter8uv_inner_neon(uint8_t *dstU, uint8_t *dstV, |
| 47 | 47 |
int stride, int flim_E, int flim_I, |
| 48 | 48 |
int hev_thresh); |
| 49 | 49 |
|
| 50 |
-void ff_vp8_v_loop_filter16_simple_neon(uint8_t *dst, int stride, int flim); |
|
| 51 |
-void ff_vp8_h_loop_filter16_simple_neon(uint8_t *dst, int stride, int flim); |
|
| 50 |
+void ff_vp8_v_loop_filter_inner_armv6(uint8_t *dst, int stride, |
|
| 51 |
+ int flim_E, int flim_I, |
|
| 52 |
+ int hev_thresh, int count); |
|
| 53 |
+void ff_vp8_h_loop_filter_inner_armv6(uint8_t *dst, int stride, |
|
| 54 |
+ int flim_E, int flim_I, |
|
| 55 |
+ int hev_thresh, int count); |
|
| 56 |
+void ff_vp8_v_loop_filter_armv6(uint8_t *dst, int stride, |
|
| 57 |
+ int flim_E, int flim_I, |
|
| 58 |
+ int hev_thresh, int count); |
|
| 59 |
+void ff_vp8_h_loop_filter_armv6(uint8_t *dst, int stride, |
|
| 60 |
+ int flim_E, int flim_I, |
|
| 61 |
+ int hev_thresh, int count); |
|
| 52 | 62 |
|
| 63 |
+static void ff_vp8_v_loop_filter16_armv6(uint8_t *dst, int stride, |
|
| 64 |
+ int flim_E, int flim_I, int hev_thresh) |
|
| 65 |
+{
|
|
| 66 |
+ ff_vp8_v_loop_filter_armv6(dst, stride, flim_E, flim_I, hev_thresh, 4); |
|
| 67 |
+} |
|
| 68 |
+ |
|
| 69 |
+static void ff_vp8_h_loop_filter16_armv6(uint8_t *dst, int stride, |
|
| 70 |
+ int flim_E, int flim_I, int hev_thresh) |
|
| 71 |
+{
|
|
| 72 |
+ ff_vp8_h_loop_filter_armv6(dst, stride, flim_E, flim_I, hev_thresh, 4); |
|
| 73 |
+} |
|
| 53 | 74 |
|
| 54 |
-#define VP8_MC(n) \ |
|
| 55 |
- void ff_put_vp8_##n##_neon(uint8_t *dst, int dststride, \ |
|
| 56 |
- uint8_t *src, int srcstride, \ |
|
| 57 |
- int h, int x, int y) |
|
| 75 |
+static void ff_vp8_v_loop_filter8uv_armv6(uint8_t *dstU, uint8_t *dstV, int stride, |
|
| 76 |
+ int flim_E, int flim_I, int hev_thresh) |
|
| 77 |
+{
|
|
| 78 |
+ ff_vp8_v_loop_filter_armv6(dstU, stride, flim_E, flim_I, hev_thresh, 2); |
|
| 79 |
+ ff_vp8_v_loop_filter_armv6(dstV, stride, flim_E, flim_I, hev_thresh, 2); |
|
| 80 |
+} |
|
| 81 |
+ |
|
| 82 |
+static void ff_vp8_h_loop_filter8uv_armv6(uint8_t *dstU, uint8_t *dstV, int stride, |
|
| 83 |
+ int flim_E, int flim_I, int hev_thresh) |
|
| 84 |
+{
|
|
| 85 |
+ ff_vp8_h_loop_filter_armv6(dstU, stride, flim_E, flim_I, hev_thresh, 2); |
|
| 86 |
+ ff_vp8_h_loop_filter_armv6(dstV, stride, flim_E, flim_I, hev_thresh, 2); |
|
| 87 |
+} |
|
| 88 |
+ |
|
| 89 |
+static void ff_vp8_v_loop_filter16_inner_armv6(uint8_t *dst, int stride, |
|
| 90 |
+ int flim_E, int flim_I, int hev_thresh) |
|
| 91 |
+{
|
|
| 92 |
+ ff_vp8_v_loop_filter_inner_armv6(dst, stride, flim_E, flim_I, hev_thresh, 4); |
|
| 93 |
+} |
|
| 94 |
+ |
|
| 95 |
+static void ff_vp8_h_loop_filter16_inner_armv6(uint8_t *dst, int stride, |
|
| 96 |
+ int flim_E, int flim_I, int hev_thresh) |
|
| 97 |
+{
|
|
| 98 |
+ ff_vp8_h_loop_filter_inner_armv6(dst, stride, flim_E, flim_I, hev_thresh, 4); |
|
| 99 |
+} |
|
| 100 |
+ |
|
| 101 |
+static void ff_vp8_v_loop_filter8uv_inner_armv6(uint8_t *dstU, uint8_t *dstV, |
|
| 102 |
+ int stride, int flim_E, int flim_I, |
|
| 103 |
+ int hev_thresh) |
|
| 104 |
+{
|
|
| 105 |
+ ff_vp8_v_loop_filter_inner_armv6(dstU, stride, flim_E, flim_I, hev_thresh, 2); |
|
| 106 |
+ ff_vp8_v_loop_filter_inner_armv6(dstV, stride, flim_E, flim_I, hev_thresh, 2); |
|
| 107 |
+} |
|
| 108 |
+ |
|
| 109 |
+static void ff_vp8_h_loop_filter8uv_inner_armv6(uint8_t *dstU, uint8_t *dstV, |
|
| 110 |
+ int stride, int flim_E, int flim_I, |
|
| 111 |
+ int hev_thresh) |
|
| 112 |
+{
|
|
| 113 |
+ ff_vp8_h_loop_filter_inner_armv6(dstU, stride, flim_E, flim_I, hev_thresh, 2); |
|
| 114 |
+ ff_vp8_h_loop_filter_inner_armv6(dstV, stride, flim_E, flim_I, hev_thresh, 2); |
|
| 115 |
+} |
|
| 116 |
+ |
|
| 117 |
+#define simple_lf_funcs(opt) \ |
|
| 118 |
+void ff_vp8_v_loop_filter16_simple_ ## opt(uint8_t *dst, int stride, int flim); \ |
|
| 119 |
+void ff_vp8_h_loop_filter16_simple_ ## opt(uint8_t *dst, int stride, int flim) |
|
| 120 |
+ |
|
| 121 |
+simple_lf_funcs(neon); |
|
| 122 |
+simple_lf_funcs(armv6); |
|
| 123 |
+ |
|
| 124 |
+#define VP8_MC_OPT(n, opt) \ |
|
| 125 |
+ void ff_put_vp8_##n##_##opt(uint8_t *dst, int dststride, \ |
|
| 126 |
+ uint8_t *src, int srcstride, \ |
|
| 127 |
+ int h, int x, int y) |
|
| 128 |
+ |
|
| 129 |
+#define VP8_MC(n) \ |
|
| 130 |
+ VP8_MC_OPT(n, neon) |
|
| 58 | 131 |
|
| 59 | 132 |
#define VP8_EPEL(w) \ |
| 60 |
- VP8_MC(pixels ## w); \ |
|
| 61 | 133 |
VP8_MC(epel ## w ## _h4); \ |
| 62 | 134 |
VP8_MC(epel ## w ## _h6); \ |
| 63 |
- VP8_MC(epel ## w ## _v4); \ |
|
| 64 | 135 |
VP8_MC(epel ## w ## _h4v4); \ |
| 65 | 136 |
VP8_MC(epel ## w ## _h6v4); \ |
| 137 |
+ VP8_MC(epel ## w ## _v4); \ |
|
| 66 | 138 |
VP8_MC(epel ## w ## _v6); \ |
| 67 | 139 |
VP8_MC(epel ## w ## _h4v6); \ |
| 68 | 140 |
VP8_MC(epel ## w ## _h6v6) |
| 69 | 141 |
|
| 70 | 142 |
VP8_EPEL(16); |
| 143 |
+VP8_MC(pixels16); |
|
| 144 |
+VP8_MC_OPT(pixels16, armv6); |
|
| 71 | 145 |
VP8_EPEL(8); |
| 146 |
+VP8_MC(pixels8); |
|
| 147 |
+VP8_MC_OPT(pixels8, armv6); |
|
| 72 | 148 |
VP8_EPEL(4); |
| 149 |
+VP8_MC_OPT(pixels4, armv6); |
|
| 73 | 150 |
|
| 74 | 151 |
VP8_MC(bilin16_h); |
| 75 | 152 |
VP8_MC(bilin16_v); |
| ... | ... |
@@ -81,83 +162,148 @@ VP8_MC(bilin4_h); |
| 81 | 81 |
VP8_MC(bilin4_v); |
| 82 | 82 |
VP8_MC(bilin4_hv); |
| 83 | 83 |
|
| 84 |
+#define VP8_V6_MC(n) \ |
|
| 85 |
+void ff_put_vp8_##n##_armv6(uint8_t *dst, int dststride, uint8_t *src, \ |
|
| 86 |
+ int srcstride, int w, int h, int mxy) |
|
| 87 |
+ |
|
| 88 |
+VP8_V6_MC(epel_v6); |
|
| 89 |
+VP8_V6_MC(epel_h6); |
|
| 90 |
+VP8_V6_MC(epel_v4); |
|
| 91 |
+VP8_V6_MC(epel_h4); |
|
| 92 |
+VP8_V6_MC(bilin_v); |
|
| 93 |
+VP8_V6_MC(bilin_h); |
|
| 94 |
+ |
|
| 95 |
+#define VP8_EPEL_HV(SIZE, TAPNUMX, TAPNUMY, NAME, HNAME, VNAME, MAXHEIGHT) \ |
|
| 96 |
+static void ff_put_vp8_##NAME##SIZE##_##HNAME##VNAME##_armv6( \ |
|
| 97 |
+ uint8_t *dst, int dststride, uint8_t *src, \ |
|
| 98 |
+ int srcstride, int h, int mx, int my) \ |
|
| 99 |
+{ \
|
|
| 100 |
+ DECLARE_ALIGNED(4, uint8_t, tmp)[SIZE * (MAXHEIGHT + TAPNUMY - 1)]; \ |
|
| 101 |
+ uint8_t *tmpptr = tmp + SIZE * (TAPNUMY / 2 - 1); \ |
|
| 102 |
+ src -= srcstride * (TAPNUMY / 2 - 1); \ |
|
| 103 |
+ ff_put_vp8_ ## NAME ## _ ## HNAME ## _armv6(tmp, SIZE, src, srcstride, \ |
|
| 104 |
+ SIZE, h + TAPNUMY - 1, mx); \ |
|
| 105 |
+ ff_put_vp8_ ## NAME ## _ ## VNAME ## _armv6(dst, dststride, tmpptr, SIZE, \ |
|
| 106 |
+ SIZE, h, my); \ |
|
| 107 |
+} |
|
| 108 |
+ |
|
| 109 |
+VP8_EPEL_HV(16, 6, 6, epel, h6, v6, 16); |
|
| 110 |
+VP8_EPEL_HV(16, 2, 2, bilin, h, v, 16); |
|
| 111 |
+VP8_EPEL_HV(8, 6, 6, epel, h6, v6, 16); |
|
| 112 |
+VP8_EPEL_HV(8, 4, 6, epel, h4, v6, 16); |
|
| 113 |
+VP8_EPEL_HV(8, 6, 4, epel, h6, v4, 16); |
|
| 114 |
+VP8_EPEL_HV(8, 4, 4, epel, h4, v4, 16); |
|
| 115 |
+VP8_EPEL_HV(8, 2, 2, bilin, h, v, 16); |
|
| 116 |
+VP8_EPEL_HV(4, 6, 6, epel, h6, v6, 8); |
|
| 117 |
+VP8_EPEL_HV(4, 4, 6, epel, h4, v6, 8); |
|
| 118 |
+VP8_EPEL_HV(4, 6, 4, epel, h6, v4, 8); |
|
| 119 |
+VP8_EPEL_HV(4, 4, 4, epel, h4, v4, 8); |
|
| 120 |
+VP8_EPEL_HV(4, 2, 2, bilin, h, v, 8); |
|
| 121 |
+ |
|
| 122 |
+extern void put_vp8_epel4_v6_c(uint8_t *dst, int d, uint8_t *src, int s, int h, int mx, int my); |
|
| 123 |
+#undef printf |
|
| 124 |
+#define VP8_EPEL_H_OR_V(SIZE, NAME, HV) \ |
|
| 125 |
+static void ff_put_vp8_##NAME##SIZE##_##HV##_armv6( \ |
|
| 126 |
+ uint8_t *dst, int dststride, uint8_t *src, \ |
|
| 127 |
+ int srcstride, int h, int mx, int my) \ |
|
| 128 |
+{ \
|
|
| 129 |
+ ff_put_vp8_## NAME ## _ ## HV ## _armv6(dst, dststride, src, srcstride, \ |
|
| 130 |
+ SIZE, h, mx | my); \ |
|
| 131 |
+} |
|
| 132 |
+ |
|
| 133 |
+VP8_EPEL_H_OR_V(4, epel, h6); |
|
| 134 |
+VP8_EPEL_H_OR_V(4, epel, h4); |
|
| 135 |
+VP8_EPEL_H_OR_V(4, epel, v6); |
|
| 136 |
+VP8_EPEL_H_OR_V(4, epel, v4); |
|
| 137 |
+VP8_EPEL_H_OR_V(4, bilin, v); |
|
| 138 |
+VP8_EPEL_H_OR_V(4, bilin, h); |
|
| 139 |
+VP8_EPEL_H_OR_V(8, epel, h6); |
|
| 140 |
+VP8_EPEL_H_OR_V(8, epel, h4); |
|
| 141 |
+VP8_EPEL_H_OR_V(8, epel, v6); |
|
| 142 |
+VP8_EPEL_H_OR_V(8, epel, v4); |
|
| 143 |
+VP8_EPEL_H_OR_V(8, bilin, v); |
|
| 144 |
+VP8_EPEL_H_OR_V(8, bilin, h); |
|
| 145 |
+VP8_EPEL_H_OR_V(16, epel, h6); |
|
| 146 |
+VP8_EPEL_H_OR_V(16, epel, v6); |
|
| 147 |
+VP8_EPEL_H_OR_V(16, bilin, v); |
|
| 148 |
+VP8_EPEL_H_OR_V(16, bilin, h); |
|
| 149 |
+ |
|
| 84 | 150 |
av_cold void ff_vp8dsp_init_arm(VP8DSPContext *dsp) |
| 85 | 151 |
{
|
| 152 |
+#define set_func_ptrs(opt) \ |
|
| 153 |
+ dsp->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_##opt; \ |
|
| 154 |
+ dsp->vp8_luma_dc_wht_dc = ff_vp8_luma_dc_wht_dc_armv6; \ |
|
| 155 |
+ \ |
|
| 156 |
+ dsp->vp8_idct_add = ff_vp8_idct_add_##opt; \ |
|
| 157 |
+ dsp->vp8_idct_dc_add = ff_vp8_idct_dc_add_##opt; \ |
|
| 158 |
+ dsp->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_##opt; \ |
|
| 159 |
+ dsp->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_##opt; \ |
|
| 160 |
+ \ |
|
| 161 |
+ dsp->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_##opt; \ |
|
| 162 |
+ dsp->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_##opt; \ |
|
| 163 |
+ dsp->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_##opt; \ |
|
| 164 |
+ dsp->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_##opt; \ |
|
| 165 |
+ \ |
|
| 166 |
+ dsp->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_##opt; \ |
|
| 167 |
+ dsp->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_##opt; \ |
|
| 168 |
+ dsp->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_##opt; \ |
|
| 169 |
+ dsp->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_##opt; \ |
|
| 170 |
+ \ |
|
| 171 |
+ dsp->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter16_simple_##opt; \ |
|
| 172 |
+ dsp->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter16_simple_##opt; \ |
|
| 173 |
+ \ |
|
| 174 |
+ dsp->put_vp8_epel_pixels_tab[0][0][0] = ff_put_vp8_pixels16_##opt; \ |
|
| 175 |
+ dsp->put_vp8_epel_pixels_tab[0][0][2] = ff_put_vp8_epel16_h6_##opt; \ |
|
| 176 |
+ dsp->put_vp8_epel_pixels_tab[0][2][0] = ff_put_vp8_epel16_v6_##opt; \ |
|
| 177 |
+ dsp->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_##opt; \ |
|
| 178 |
+ \ |
|
| 179 |
+ dsp->put_vp8_epel_pixels_tab[1][0][0] = ff_put_vp8_pixels8_##opt; \ |
|
| 180 |
+ dsp->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_##opt; \ |
|
| 181 |
+ dsp->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_##opt; \ |
|
| 182 |
+ dsp->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_##opt; \ |
|
| 183 |
+ dsp->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_##opt; \ |
|
| 184 |
+ dsp->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_##opt; \ |
|
| 185 |
+ dsp->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_##opt; \ |
|
| 186 |
+ dsp->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_##opt; \ |
|
| 187 |
+ dsp->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_##opt; \ |
|
| 188 |
+ \ |
|
| 189 |
+ dsp->put_vp8_epel_pixels_tab[2][0][0] = ff_put_vp8_pixels4_armv6; \ |
|
| 190 |
+ dsp->put_vp8_epel_pixels_tab[2][0][1] = ff_put_vp8_epel4_h4_##opt; \ |
|
| 191 |
+ dsp->put_vp8_epel_pixels_tab[2][0][2] = ff_put_vp8_epel4_h6_##opt; \ |
|
| 192 |
+ dsp->put_vp8_epel_pixels_tab[2][1][0] = ff_put_vp8_epel4_v4_##opt; \ |
|
| 193 |
+ dsp->put_vp8_epel_pixels_tab[2][1][1] = ff_put_vp8_epel4_h4v4_##opt; \ |
|
| 194 |
+ dsp->put_vp8_epel_pixels_tab[2][1][2] = ff_put_vp8_epel4_h6v4_##opt; \ |
|
| 195 |
+ dsp->put_vp8_epel_pixels_tab[2][2][0] = ff_put_vp8_epel4_v6_##opt; \ |
|
| 196 |
+ dsp->put_vp8_epel_pixels_tab[2][2][1] = ff_put_vp8_epel4_h4v6_##opt; \ |
|
| 197 |
+ dsp->put_vp8_epel_pixels_tab[2][2][2] = ff_put_vp8_epel4_h6v6_##opt; \ |
|
| 198 |
+ \ |
|
| 199 |
+ dsp->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_##opt; \ |
|
| 200 |
+ dsp->put_vp8_bilinear_pixels_tab[0][0][2] = ff_put_vp8_bilin16_h_##opt; \ |
|
| 201 |
+ dsp->put_vp8_bilinear_pixels_tab[0][2][0] = ff_put_vp8_bilin16_v_##opt; \ |
|
| 202 |
+ dsp->put_vp8_bilinear_pixels_tab[0][2][2] = ff_put_vp8_bilin16_hv_##opt; \ |
|
| 203 |
+ \ |
|
| 204 |
+ dsp->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_##opt; \ |
|
| 205 |
+ dsp->put_vp8_bilinear_pixels_tab[1][0][1] = ff_put_vp8_bilin8_h_##opt; \ |
|
| 206 |
+ dsp->put_vp8_bilinear_pixels_tab[1][0][2] = ff_put_vp8_bilin8_h_##opt; \ |
|
| 207 |
+ dsp->put_vp8_bilinear_pixels_tab[1][1][0] = ff_put_vp8_bilin8_v_##opt; \ |
|
| 208 |
+ dsp->put_vp8_bilinear_pixels_tab[1][1][1] = ff_put_vp8_bilin8_hv_##opt; \ |
|
| 209 |
+ dsp->put_vp8_bilinear_pixels_tab[1][1][2] = ff_put_vp8_bilin8_hv_##opt; \ |
|
| 210 |
+ dsp->put_vp8_bilinear_pixels_tab[1][2][0] = ff_put_vp8_bilin8_v_##opt; \ |
|
| 211 |
+ dsp->put_vp8_bilinear_pixels_tab[1][2][1] = ff_put_vp8_bilin8_hv_##opt; \ |
|
| 212 |
+ dsp->put_vp8_bilinear_pixels_tab[1][2][2] = ff_put_vp8_bilin8_hv_##opt; \ |
|
| 213 |
+ \ |
|
| 214 |
+ dsp->put_vp8_bilinear_pixels_tab[2][0][0] = ff_put_vp8_pixels4_armv6; \ |
|
| 215 |
+ dsp->put_vp8_bilinear_pixels_tab[2][0][1] = ff_put_vp8_bilin4_h_##opt; \ |
|
| 216 |
+ dsp->put_vp8_bilinear_pixels_tab[2][0][2] = ff_put_vp8_bilin4_h_##opt; \ |
|
| 217 |
+ dsp->put_vp8_bilinear_pixels_tab[2][1][0] = ff_put_vp8_bilin4_v_##opt; \ |
|
| 218 |
+ dsp->put_vp8_bilinear_pixels_tab[2][1][1] = ff_put_vp8_bilin4_hv_##opt; \ |
|
| 219 |
+ dsp->put_vp8_bilinear_pixels_tab[2][1][2] = ff_put_vp8_bilin4_hv_##opt; \ |
|
| 220 |
+ dsp->put_vp8_bilinear_pixels_tab[2][2][0] = ff_put_vp8_bilin4_v_##opt; \ |
|
| 221 |
+ dsp->put_vp8_bilinear_pixels_tab[2][2][1] = ff_put_vp8_bilin4_hv_##opt; \ |
|
| 222 |
+ dsp->put_vp8_bilinear_pixels_tab[2][2][2] = ff_put_vp8_bilin4_hv_##opt |
|
| 86 | 223 |
if (HAVE_NEON) {
|
| 87 |
- dsp->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_neon; |
|
| 88 |
- dsp->vp8_luma_dc_wht_dc = ff_vp8_luma_dc_wht_dc_neon; |
|
| 89 |
- |
|
| 90 |
- dsp->vp8_idct_add = ff_vp8_idct_add_neon; |
|
| 91 |
- dsp->vp8_idct_dc_add = ff_vp8_idct_dc_add_neon; |
|
| 92 |
- dsp->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_neon; |
|
| 93 |
- dsp->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_neon; |
|
| 94 |
- |
|
| 95 |
- dsp->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_neon; |
|
| 96 |
- dsp->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_neon; |
|
| 97 |
- dsp->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_neon; |
|
| 98 |
- dsp->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_neon; |
|
| 99 |
- |
|
| 100 |
- dsp->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_neon; |
|
| 101 |
- dsp->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_neon; |
|
| 102 |
- dsp->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_neon; |
|
| 103 |
- dsp->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_neon; |
|
| 104 |
- |
|
| 105 |
- dsp->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter16_simple_neon; |
|
| 106 |
- dsp->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter16_simple_neon; |
|
| 107 |
- |
|
| 108 |
- dsp->put_vp8_epel_pixels_tab[0][0][0] = ff_put_vp8_pixels16_neon; |
|
| 109 |
- dsp->put_vp8_epel_pixels_tab[0][0][2] = ff_put_vp8_epel16_h6_neon; |
|
| 110 |
- dsp->put_vp8_epel_pixels_tab[0][2][0] = ff_put_vp8_epel16_v6_neon; |
|
| 111 |
- dsp->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_neon; |
|
| 112 |
- |
|
| 113 |
- dsp->put_vp8_epel_pixels_tab[1][0][0] = ff_put_vp8_pixels8_neon; |
|
| 114 |
- dsp->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_neon; |
|
| 115 |
- dsp->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_neon; |
|
| 116 |
- dsp->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_neon; |
|
| 117 |
- dsp->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_neon; |
|
| 118 |
- dsp->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_neon; |
|
| 119 |
- dsp->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_neon; |
|
| 120 |
- dsp->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_neon; |
|
| 121 |
- dsp->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_neon; |
|
| 122 |
- |
|
| 123 |
- dsp->put_vp8_epel_pixels_tab[2][0][0] = ff_put_vp8_pixels4_neon; |
|
| 124 |
- dsp->put_vp8_epel_pixels_tab[2][0][1] = ff_put_vp8_epel4_h4_neon; |
|
| 125 |
- dsp->put_vp8_epel_pixels_tab[2][0][2] = ff_put_vp8_epel4_h6_neon; |
|
| 126 |
- dsp->put_vp8_epel_pixels_tab[2][1][0] = ff_put_vp8_epel4_v4_neon; |
|
| 127 |
- dsp->put_vp8_epel_pixels_tab[2][1][1] = ff_put_vp8_epel4_h4v4_neon; |
|
| 128 |
- dsp->put_vp8_epel_pixels_tab[2][1][2] = ff_put_vp8_epel4_h6v4_neon; |
|
| 129 |
- dsp->put_vp8_epel_pixels_tab[2][2][0] = ff_put_vp8_epel4_v6_neon; |
|
| 130 |
- dsp->put_vp8_epel_pixels_tab[2][2][1] = ff_put_vp8_epel4_h4v6_neon; |
|
| 131 |
- dsp->put_vp8_epel_pixels_tab[2][2][2] = ff_put_vp8_epel4_h6v6_neon; |
|
| 132 |
- |
|
| 133 |
- dsp->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_neon; |
|
| 134 |
- dsp->put_vp8_bilinear_pixels_tab[0][0][1] = ff_put_vp8_bilin16_h_neon; |
|
| 135 |
- dsp->put_vp8_bilinear_pixels_tab[0][0][2] = ff_put_vp8_bilin16_h_neon; |
|
| 136 |
- dsp->put_vp8_bilinear_pixels_tab[0][1][0] = ff_put_vp8_bilin16_v_neon; |
|
| 137 |
- dsp->put_vp8_bilinear_pixels_tab[0][1][1] = ff_put_vp8_bilin16_hv_neon; |
|
| 138 |
- dsp->put_vp8_bilinear_pixels_tab[0][1][2] = ff_put_vp8_bilin16_hv_neon; |
|
| 139 |
- dsp->put_vp8_bilinear_pixels_tab[0][2][0] = ff_put_vp8_bilin16_v_neon; |
|
| 140 |
- dsp->put_vp8_bilinear_pixels_tab[0][2][1] = ff_put_vp8_bilin16_hv_neon; |
|
| 141 |
- dsp->put_vp8_bilinear_pixels_tab[0][2][2] = ff_put_vp8_bilin16_hv_neon; |
|
| 142 |
- |
|
| 143 |
- dsp->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_neon; |
|
| 144 |
- dsp->put_vp8_bilinear_pixels_tab[1][0][1] = ff_put_vp8_bilin8_h_neon; |
|
| 145 |
- dsp->put_vp8_bilinear_pixels_tab[1][0][2] = ff_put_vp8_bilin8_h_neon; |
|
| 146 |
- dsp->put_vp8_bilinear_pixels_tab[1][1][0] = ff_put_vp8_bilin8_v_neon; |
|
| 147 |
- dsp->put_vp8_bilinear_pixels_tab[1][1][1] = ff_put_vp8_bilin8_hv_neon; |
|
| 148 |
- dsp->put_vp8_bilinear_pixels_tab[1][1][2] = ff_put_vp8_bilin8_hv_neon; |
|
| 149 |
- dsp->put_vp8_bilinear_pixels_tab[1][2][0] = ff_put_vp8_bilin8_v_neon; |
|
| 150 |
- dsp->put_vp8_bilinear_pixels_tab[1][2][1] = ff_put_vp8_bilin8_hv_neon; |
|
| 151 |
- dsp->put_vp8_bilinear_pixels_tab[1][2][2] = ff_put_vp8_bilin8_hv_neon; |
|
| 152 |
- |
|
| 153 |
- dsp->put_vp8_bilinear_pixels_tab[2][0][0] = ff_put_vp8_pixels4_neon; |
|
| 154 |
- dsp->put_vp8_bilinear_pixels_tab[2][0][1] = ff_put_vp8_bilin4_h_neon; |
|
| 155 |
- dsp->put_vp8_bilinear_pixels_tab[2][0][2] = ff_put_vp8_bilin4_h_neon; |
|
| 156 |
- dsp->put_vp8_bilinear_pixels_tab[2][1][0] = ff_put_vp8_bilin4_v_neon; |
|
| 157 |
- dsp->put_vp8_bilinear_pixels_tab[2][1][1] = ff_put_vp8_bilin4_hv_neon; |
|
| 158 |
- dsp->put_vp8_bilinear_pixels_tab[2][1][2] = ff_put_vp8_bilin4_hv_neon; |
|
| 159 |
- dsp->put_vp8_bilinear_pixels_tab[2][2][0] = ff_put_vp8_bilin4_v_neon; |
|
| 160 |
- dsp->put_vp8_bilinear_pixels_tab[2][2][1] = ff_put_vp8_bilin4_hv_neon; |
|
| 161 |
- dsp->put_vp8_bilinear_pixels_tab[2][2][2] = ff_put_vp8_bilin4_hv_neon; |
|
| 224 |
+ set_func_ptrs(neon); |
|
| 225 |
+ } else if (HAVE_ARMV6) {
|
|
| 226 |
+ set_func_ptrs(armv6); |
|
| 162 | 227 |
} |
| 163 | 228 |
} |
| ... | ... |
@@ -76,18 +76,6 @@ function ff_vp8_luma_dc_wht_neon, export=1 |
| 76 | 76 |
bx lr |
| 77 | 77 |
endfunc |
| 78 | 78 |
|
| 79 |
-function ff_vp8_luma_dc_wht_dc_neon, export=1 |
|
| 80 |
- ldrsh r2, [r1] |
|
| 81 |
- mov r3, #0 |
|
| 82 |
- add r2, r2, #3 |
|
| 83 |
- strh r3, [r1] |
|
| 84 |
- asr r2, r2, #3 |
|
| 85 |
- .rept 16 |
|
| 86 |
- strh r2, [r0], #32 |
|
| 87 |
- .endr |
|
| 88 |
- bx lr |
|
| 89 |
-endfunc |
|
| 90 |
- |
|
| 91 | 79 |
function ff_vp8_idct_add_neon, export=1 |
| 92 | 80 |
vld1.16 {q0-q1}, [r1,:128]
|
| 93 | 81 |
movw r3, #20091 |
| ... | ... |
@@ -741,23 +729,6 @@ function ff_put_vp8_pixels8_neon, export=1 |
| 741 | 741 |
bx lr |
| 742 | 742 |
endfunc |
| 743 | 743 |
|
| 744 |
-function ff_put_vp8_pixels4_neon, export=1 |
|
| 745 |
- ldr r12, [sp, #0] @ h |
|
| 746 |
- push {r4-r6,lr}
|
|
| 747 |
-1: |
|
| 748 |
- subs r12, r12, #4 |
|
| 749 |
- ldr_post r4, r2, r3 |
|
| 750 |
- ldr_post r5, r2, r3 |
|
| 751 |
- ldr_post r6, r2, r3 |
|
| 752 |
- ldr_post lr, r2, r3 |
|
| 753 |
- str_post r4, r0, r1 |
|
| 754 |
- str_post r5, r0, r1 |
|
| 755 |
- str_post r6, r0, r1 |
|
| 756 |
- str_post lr, r0, r1 |
|
| 757 |
- bgt 1b |
|
| 758 |
- pop {r4-r6,pc}
|
|
| 759 |
-endfunc |
|
| 760 |
- |
|
| 761 | 744 |
/* 4/6-tap 8th-pel MC */ |
| 762 | 745 |
|
| 763 | 746 |
.macro vp8_epel8_h6 d, a, b |