summaryrefslogtreecommitdiff
path: root/vp9/common/arm
diff options
context:
space:
mode:
authorMans Rullgard <mans@mansr.com>2013-08-14 16:18:08 +0100
committerMans Rullgard <mans@mansr.com>2013-08-14 16:27:55 +0100
commit0f1deccf86973ebd9c470edb92748bc32467733c (patch)
treed94fc3ea92391861d1295eec146d77201fd34b1d /vp9/common/arm
parent635ba269be143743f3513a3dc16db39d71b1945a (diff)
downloadlibvpx-0f1deccf86973ebd9c470edb92748bc32467733c.tar
libvpx-0f1deccf86973ebd9c470edb92748bc32467733c.tar.gz
libvpx-0f1deccf86973ebd9c470edb92748bc32467733c.tar.bz2
libvpx-0f1deccf86973ebd9c470edb92748bc32467733c.zip
vp9: neon: add vp9_convolve_avg_neon
Change-Id: I33cff9ac4f2234558f6f87729f9b2e88a33fbf58
Diffstat (limited to 'vp9/common/arm')
-rw-r--r--vp9/common/arm/neon/vp9_avg_neon.asm116
1 files changed, 116 insertions, 0 deletions
diff --git a/vp9/common/arm/neon/vp9_avg_neon.asm b/vp9/common/arm/neon/vp9_avg_neon.asm
new file mode 100644
index 000000000..7d2453021
--- /dev/null
+++ b/vp9/common/arm/neon/vp9_avg_neon.asm
@@ -0,0 +1,116 @@
+;
+; Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+;
+; Use of this source code is governed by a BSD-style license
+; that can be found in the LICENSE file in the root of the source
+; tree. An additional intellectual property rights grant can be found
+; in the file PATENTS. All contributing project authors may
+; be found in the AUTHORS file in the root of the source tree.
+;
+
+ EXPORT |vp9_convolve_avg_neon|
+ ARM
+ REQUIRE8
+ PRESERVE8
+
+ AREA ||.text||, CODE, READONLY, ALIGN=2
+
+|vp9_convolve_avg_neon| PROC
+ push {r4-r6, lr}
+ ldrd r4, r5, [sp, #32]
+ mov r6, r2
+
+ cmp r4, #32
+ bgt avg64
+ beq avg32
+ cmp r4, #8
+ bgt avg16
+ beq avg8
+ b avg4
+
+avg64
+ sub lr, r1, #32
+ sub r4, r3, #32
+avg64_h
+ pld [r0, r1, lsl #1]
+ vld1.8 {q0-q1}, [r0]!
+ vld1.8 {q2-q3}, [r0], lr
+ pld [r2, r3]
+ vld1.8 {q8-q9}, [r6@128]!
+ vld1.8 {q10-q11}, [r6@128], r4
+ vrhadd.u8 q0, q0, q8
+ vrhadd.u8 q1, q1, q9
+ vrhadd.u8 q2, q2, q10
+ vrhadd.u8 q3, q3, q11
+ vst1.8 {q0-q1}, [r2@128]!
+ vst1.8 {q2-q3}, [r2@128], r4
+ subs r5, r5, #1
+ bgt avg64_h
+ pop {r4-r6, pc}
+
+avg32
+ vld1.8 {q0-q1}, [r0], r1
+ vld1.8 {q2-q3}, [r0], r1
+ vld1.8 {q8-q9}, [r6@128], r3
+ vld1.8 {q10-q11}, [r6@128], r3
+ pld [r0]
+ vrhadd.u8 q0, q0, q8
+ pld [r0, r1]
+ vrhadd.u8 q1, q1, q9
+ pld [r6]
+ vrhadd.u8 q2, q2, q10
+ pld [r6, r3]
+ vrhadd.u8 q3, q3, q11
+ vst1.8 {q0-q1}, [r2@128], r3
+ vst1.8 {q2-q3}, [r2@128], r3
+ subs r5, r5, #2
+ bgt avg32
+ pop {r4-r6, pc}
+
+avg16
+ vld1.8 {q0}, [r0], r1
+ vld1.8 {q1}, [r0], r1
+ vld1.8 {q2}, [r6@128], r3
+ vld1.8 {q3}, [r6@128], r3
+ pld [r0]
+ pld [r0, r1]
+ vrhadd.u8 q0, q0, q2
+ pld [r6]
+ pld [r6, r3]
+ vrhadd.u8 q1, q1, q3
+ vst1.8 {q0}, [r2@128], r3
+ vst1.8 {q1}, [r2@128], r3
+ subs r5, r5, #2
+ bgt avg16
+ pop {r4-r6, pc}
+
+avg8
+ vld1.8 {d0}, [r0], r1
+ vld1.8 {d1}, [r0], r1
+ vld1.8 {d2}, [r6@64], r3
+ vld1.8 {d3}, [r6@64], r3
+ pld [r0]
+ pld [r0, r1]
+ vrhadd.u8 q0, q0, q1
+ pld [r6]
+ pld [r6, r3]
+ vst1.8 {d0}, [r2@64], r3
+ vst1.8 {d1}, [r2@64], r3
+ subs r5, r5, #2
+ bgt avg8
+ pop {r4-r6, pc}
+
+avg4
+ vld1.32 {d0[0]}, [r0], r1
+ vld1.32 {d0[1]}, [r0], r1
+ vld1.32 {d2[0]}, [r6@32], r3
+ vld1.32 {d2[1]}, [r6@32], r3
+ vrhadd.u8 d0, d0, d2
+ vst1.32 {d0[0]}, [r2@32], r3
+ vst1.32 {d0[1]}, [r2@32], r3
+ subs r5, r5, #2
+ bgt avg4
+ pop {r4-r6, pc}
+ ENDP
+
+ END