Merge "Remove struct params from vp8_denoiser_filter"
diff --git a/configure b/configure
index 29683cd..690ac48 100755
--- a/configure
+++ b/configure
@@ -709,6 +709,7 @@
soft_enable webm_io
;;
*-android-*)
+ soft_enable webm_io
# GTestLog must be modified to use Android logging utilities.
;;
*-darwin-*)
diff --git a/test/android/Android.mk b/test/android/Android.mk
index 13af601..4e750b2 100644
--- a/test/android/Android.mk
+++ b/test/android/Android.mk
@@ -14,8 +14,14 @@
BINDINGS_DIR := $(CUR_WD)/../../..
LOCAL_PATH := $(CUR_WD)/../../..
+#libwebm
+include $(CLEAR_VARS)
+include $(BINDINGS_DIR)/libvpx/third_party/libwebm/Android.mk
+LOCAL_PATH := $(CUR_WD)/../../..
+
#libvpx
include $(CLEAR_VARS)
+LOCAL_STATIC_LIBRARIES := libwebm
include $(BINDINGS_DIR)/libvpx/build/make/Android.mk
LOCAL_PATH := $(CUR_WD)/../..
@@ -33,7 +39,7 @@
include $(CLEAR_VARS)
LOCAL_ARM_MODE := arm
LOCAL_MODULE := libvpx_test
-LOCAL_STATIC_LIBRARIES := gtest
+LOCAL_STATIC_LIBRARIES := gtest libwebm
LOCAL_SHARED_LIBRARIES := vpx
include $(LOCAL_PATH)/test/test.mk
LOCAL_C_INCLUDES := $(BINDINGS_DIR)
diff --git a/test/register_state_check.h b/test/register_state_check.h
index 7e3d053..5987fe3 100644
--- a/test/register_state_check.h
+++ b/test/register_state_check.h
@@ -82,8 +82,8 @@
} // namespace libvpx_test
-#elif defined(CONFIG_SHARED) && defined(HAVE_NEON) \
- && !CONFIG_SHARED && HAVE_NEON
+#elif defined(CONFIG_SHARED) && defined(HAVE_NEON) && defined(CONFIG_VP9) \
+ && !CONFIG_SHARED && HAVE_NEON && CONFIG_VP9
#include "vpx/vpx_integer.h"
diff --git a/third_party/libwebm/Android.mk b/third_party/libwebm/Android.mk
new file mode 100644
index 0000000..13868b6
--- /dev/null
+++ b/third_party/libwebm/Android.mk
@@ -0,0 +1,11 @@
+LOCAL_PATH := $(call my-dir)
+include $(CLEAR_VARS)
+
+LOCAL_CPP_EXTENSION := .cpp
+LOCAL_SRC_FILES := mkvmuxer.cpp \
+ mkvmuxerutil.cpp \
+ mkvparser.cpp \
+ mkvreader.cpp \
+ mkvwriter.cpp
+LOCAL_MODULE := libwebm
+include $(BUILD_STATIC_LIBRARY)
diff --git a/vp8/common/arm/neon/iwalsh_neon.asm b/vp8/common/arm/neon/iwalsh_neon.asm
deleted file mode 100644
index e8ea2a6..0000000
--- a/vp8/common/arm/neon/iwalsh_neon.asm
+++ /dev/null
@@ -1,87 +0,0 @@
-;
-; Copyright (c) 2010 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 |vp8_short_inv_walsh4x4_neon|
-
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA |.text|, CODE, READONLY ; name this block of code
-
-;short vp8_short_inv_walsh4x4_neon(short *input, short *mb_dqcoeff)
-|vp8_short_inv_walsh4x4_neon| PROC
-
- ; read in all four lines of values: d0->d3
- vld1.i16 {q0-q1}, [r0@128]
-
- ; first for loop
- vadd.s16 d4, d0, d3 ;a = [0] + [12]
- vadd.s16 d6, d1, d2 ;b = [4] + [8]
- vsub.s16 d5, d0, d3 ;d = [0] - [12]
- vsub.s16 d7, d1, d2 ;c = [4] - [8]
-
- vadd.s16 q0, q2, q3 ; a+b d+c
- vsub.s16 q1, q2, q3 ; a-b d-c
-
- vtrn.32 d0, d2 ;d0: 0 1 8 9
- ;d2: 2 3 10 11
- vtrn.32 d1, d3 ;d1: 4 5 12 13
- ;d3: 6 7 14 15
-
- vtrn.16 d0, d1 ;d0: 0 4 8 12
- ;d1: 1 5 9 13
- vtrn.16 d2, d3 ;d2: 2 6 10 14
- ;d3: 3 7 11 15
-
- ; second for loop
-
- vadd.s16 d4, d0, d3 ;a = [0] + [3]
- vadd.s16 d6, d1, d2 ;b = [1] + [2]
- vsub.s16 d5, d0, d3 ;d = [0] - [3]
- vsub.s16 d7, d1, d2 ;c = [1] - [2]
-
- vmov.i16 q8, #3
-
- vadd.s16 q0, q2, q3 ; a+b d+c
- vsub.s16 q1, q2, q3 ; a-b d-c
-
- vadd.i16 q0, q0, q8 ;e/f += 3
- vadd.i16 q1, q1, q8 ;g/h += 3
-
- vshr.s16 q0, q0, #3 ;e/f >> 3
- vshr.s16 q1, q1, #3 ;g/h >> 3
-
- mov r2, #64
- add r3, r1, #32
-
- vst1.i16 d0[0], [r1],r2
- vst1.i16 d1[0], [r3],r2
- vst1.i16 d2[0], [r1],r2
- vst1.i16 d3[0], [r3],r2
-
- vst1.i16 d0[1], [r1],r2
- vst1.i16 d1[1], [r3],r2
- vst1.i16 d2[1], [r1],r2
- vst1.i16 d3[1], [r3],r2
-
- vst1.i16 d0[2], [r1],r2
- vst1.i16 d1[2], [r3],r2
- vst1.i16 d2[2], [r1],r2
- vst1.i16 d3[2], [r3],r2
-
- vst1.i16 d0[3], [r1],r2
- vst1.i16 d1[3], [r3],r2
- vst1.i16 d2[3], [r1]
- vst1.i16 d3[3], [r3]
-
- bx lr
- ENDP ; |vp8_short_inv_walsh4x4_neon|
-
- END
diff --git a/vp8/common/arm/neon/iwalsh_neon.c b/vp8/common/arm/neon/iwalsh_neon.c
new file mode 100644
index 0000000..6ea9dd7
--- /dev/null
+++ b/vp8/common/arm/neon/iwalsh_neon.c
@@ -0,0 +1,102 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+void vp8_short_inv_walsh4x4_neon(
+ int16_t *input,
+ int16_t *mb_dqcoeff) {
+ int16x8_t q0s16, q1s16, q2s16, q3s16;
+ int16x4_t d4s16, d5s16, d6s16, d7s16;
+ int16x4x2_t v2tmp0, v2tmp1;
+ int32x2x2_t v2tmp2, v2tmp3;
+ int16x8_t qAdd3;
+
+ q0s16 = vld1q_s16(input);
+ q1s16 = vld1q_s16(input + 8);
+
+ // 1st for loop
+ d4s16 = vadd_s16(vget_low_s16(q0s16), vget_high_s16(q1s16));
+ d6s16 = vadd_s16(vget_high_s16(q0s16), vget_low_s16(q1s16));
+ d5s16 = vsub_s16(vget_low_s16(q0s16), vget_high_s16(q1s16));
+ d7s16 = vsub_s16(vget_high_s16(q0s16), vget_low_s16(q1s16));
+
+ q2s16 = vcombine_s16(d4s16, d5s16);
+ q3s16 = vcombine_s16(d6s16, d7s16);
+
+ q0s16 = vaddq_s16(q2s16, q3s16);
+ q1s16 = vsubq_s16(q2s16, q3s16);
+
+ v2tmp2 = vtrn_s32(vreinterpret_s32_s16(vget_low_s16(q0s16)),
+ vreinterpret_s32_s16(vget_low_s16(q1s16)));
+ v2tmp3 = vtrn_s32(vreinterpret_s32_s16(vget_high_s16(q0s16)),
+ vreinterpret_s32_s16(vget_high_s16(q1s16)));
+ v2tmp0 = vtrn_s16(vreinterpret_s16_s32(v2tmp2.val[0]),
+ vreinterpret_s16_s32(v2tmp3.val[0]));
+ v2tmp1 = vtrn_s16(vreinterpret_s16_s32(v2tmp2.val[1]),
+ vreinterpret_s16_s32(v2tmp3.val[1]));
+
+ // 2nd for loop
+ d4s16 = vadd_s16(v2tmp0.val[0], v2tmp1.val[1]);
+ d6s16 = vadd_s16(v2tmp0.val[1], v2tmp1.val[0]);
+ d5s16 = vsub_s16(v2tmp0.val[0], v2tmp1.val[1]);
+ d7s16 = vsub_s16(v2tmp0.val[1], v2tmp1.val[0]);
+ q2s16 = vcombine_s16(d4s16, d5s16);
+ q3s16 = vcombine_s16(d6s16, d7s16);
+
+ qAdd3 = vdupq_n_s16(3);
+
+ q0s16 = vaddq_s16(q2s16, q3s16);
+ q1s16 = vsubq_s16(q2s16, q3s16);
+
+ q0s16 = vaddq_s16(q0s16, qAdd3);
+ q1s16 = vaddq_s16(q1s16, qAdd3);
+
+ q0s16 = vshrq_n_s16(q0s16, 3);
+ q1s16 = vshrq_n_s16(q1s16, 3);
+
+ // store
+ vst1_lane_s16(mb_dqcoeff, vget_low_s16(q0s16), 0);
+ mb_dqcoeff += 16;
+ vst1_lane_s16(mb_dqcoeff, vget_high_s16(q0s16), 0);
+ mb_dqcoeff += 16;
+ vst1_lane_s16(mb_dqcoeff, vget_low_s16(q1s16), 0);
+ mb_dqcoeff += 16;
+ vst1_lane_s16(mb_dqcoeff, vget_high_s16(q1s16), 0);
+ mb_dqcoeff += 16;
+
+ vst1_lane_s16(mb_dqcoeff, vget_low_s16(q0s16), 1);
+ mb_dqcoeff += 16;
+ vst1_lane_s16(mb_dqcoeff, vget_high_s16(q0s16), 1);
+ mb_dqcoeff += 16;
+ vst1_lane_s16(mb_dqcoeff, vget_low_s16(q1s16), 1);
+ mb_dqcoeff += 16;
+ vst1_lane_s16(mb_dqcoeff, vget_high_s16(q1s16), 1);
+ mb_dqcoeff += 16;
+
+ vst1_lane_s16(mb_dqcoeff, vget_low_s16(q0s16), 2);
+ mb_dqcoeff += 16;
+ vst1_lane_s16(mb_dqcoeff, vget_high_s16(q0s16), 2);
+ mb_dqcoeff += 16;
+ vst1_lane_s16(mb_dqcoeff, vget_low_s16(q1s16), 2);
+ mb_dqcoeff += 16;
+ vst1_lane_s16(mb_dqcoeff, vget_high_s16(q1s16), 2);
+ mb_dqcoeff += 16;
+
+ vst1_lane_s16(mb_dqcoeff, vget_low_s16(q0s16), 3);
+ mb_dqcoeff += 16;
+ vst1_lane_s16(mb_dqcoeff, vget_high_s16(q0s16), 3);
+ mb_dqcoeff += 16;
+ vst1_lane_s16(mb_dqcoeff, vget_low_s16(q1s16), 3);
+ mb_dqcoeff += 16;
+ vst1_lane_s16(mb_dqcoeff, vget_high_s16(q1s16), 3);
+ mb_dqcoeff += 16;
+ return;
+}
diff --git a/vp8/common/arm/neon/loopfilter_neon.asm b/vp8/common/arm/neon/loopfilter_neon.asm
deleted file mode 100644
index c4f09c7..0000000
--- a/vp8/common/arm/neon/loopfilter_neon.asm
+++ /dev/null
@@ -1,409 +0,0 @@
-;
-; Copyright (c) 2010 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 |vp8_loop_filter_horizontal_edge_y_neon|
- EXPORT |vp8_loop_filter_horizontal_edge_uv_neon|
- EXPORT |vp8_loop_filter_vertical_edge_y_neon|
- EXPORT |vp8_loop_filter_vertical_edge_uv_neon|
- ARM
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
-; r0 unsigned char *src
-; r1 int pitch
-; r2 unsigned char blimit
-; r3 unsigned char limit
-; sp unsigned char thresh,
-|vp8_loop_filter_horizontal_edge_y_neon| PROC
- push {lr}
- vpush {d8-d15}
-
- vdup.u8 q0, r2 ; duplicate blimit
- vdup.u8 q1, r3 ; duplicate limit
- sub r2, r0, r1, lsl #2 ; move src pointer down by 4 lines
- ldr r3, [sp, #68] ; load thresh
- add r12, r2, r1
- add r1, r1, r1
-
- vdup.u8 q2, r3 ; duplicate thresh
-
- vld1.u8 {q3}, [r2@128], r1 ; p3
- vld1.u8 {q4}, [r12@128], r1 ; p2
- vld1.u8 {q5}, [r2@128], r1 ; p1
- vld1.u8 {q6}, [r12@128], r1 ; p0
- vld1.u8 {q7}, [r2@128], r1 ; q0
- vld1.u8 {q8}, [r12@128], r1 ; q1
- vld1.u8 {q9}, [r2@128] ; q2
- vld1.u8 {q10}, [r12@128] ; q3
-
- sub r2, r2, r1, lsl #1
- sub r12, r12, r1, lsl #1
-
- bl vp8_loop_filter_neon
-
- vst1.u8 {q5}, [r2@128], r1 ; store op1
- vst1.u8 {q6}, [r12@128], r1 ; store op0
- vst1.u8 {q7}, [r2@128], r1 ; store oq0
- vst1.u8 {q8}, [r12@128], r1 ; store oq1
-
- vpop {d8-d15}
- pop {pc}
- ENDP ; |vp8_loop_filter_horizontal_edge_y_neon|
-
-
-; r0 unsigned char *u,
-; r1 int pitch,
-; r2 unsigned char blimit
-; r3 unsigned char limit
-; sp unsigned char thresh,
-; sp+4 unsigned char *v
-|vp8_loop_filter_horizontal_edge_uv_neon| PROC
- push {lr}
- vpush {d8-d15}
-
- vdup.u8 q0, r2 ; duplicate blimit
- vdup.u8 q1, r3 ; duplicate limit
- ldr r12, [sp, #68] ; load thresh
- ldr r2, [sp, #72] ; load v ptr
- vdup.u8 q2, r12 ; duplicate thresh
-
- sub r3, r0, r1, lsl #2 ; move u pointer down by 4 lines
- sub r12, r2, r1, lsl #2 ; move v pointer down by 4 lines
-
- vld1.u8 {d6}, [r3@64], r1 ; p3
- vld1.u8 {d7}, [r12@64], r1 ; p3
- vld1.u8 {d8}, [r3@64], r1 ; p2
- vld1.u8 {d9}, [r12@64], r1 ; p2
- vld1.u8 {d10}, [r3@64], r1 ; p1
- vld1.u8 {d11}, [r12@64], r1 ; p1
- vld1.u8 {d12}, [r3@64], r1 ; p0
- vld1.u8 {d13}, [r12@64], r1 ; p0
- vld1.u8 {d14}, [r3@64], r1 ; q0
- vld1.u8 {d15}, [r12@64], r1 ; q0
- vld1.u8 {d16}, [r3@64], r1 ; q1
- vld1.u8 {d17}, [r12@64], r1 ; q1
- vld1.u8 {d18}, [r3@64], r1 ; q2
- vld1.u8 {d19}, [r12@64], r1 ; q2
- vld1.u8 {d20}, [r3@64] ; q3
- vld1.u8 {d21}, [r12@64] ; q3
-
- bl vp8_loop_filter_neon
-
- sub r0, r0, r1, lsl #1
- sub r2, r2, r1, lsl #1
-
- vst1.u8 {d10}, [r0@64], r1 ; store u op1
- vst1.u8 {d11}, [r2@64], r1 ; store v op1
- vst1.u8 {d12}, [r0@64], r1 ; store u op0
- vst1.u8 {d13}, [r2@64], r1 ; store v op0
- vst1.u8 {d14}, [r0@64], r1 ; store u oq0
- vst1.u8 {d15}, [r2@64], r1 ; store v oq0
- vst1.u8 {d16}, [r0@64] ; store u oq1
- vst1.u8 {d17}, [r2@64] ; store v oq1
-
- vpop {d8-d15}
- pop {pc}
- ENDP ; |vp8_loop_filter_horizontal_edge_uv_neon|
-
-; void vp8_loop_filter_vertical_edge_y_neon(unsigned char *src, int pitch,
-; const signed char *flimit,
-; const signed char *limit,
-; const signed char *thresh,
-; int count)
-; r0 unsigned char *src
-; r1 int pitch
-; r2 unsigned char blimit
-; r3 unsigned char limit
-; sp unsigned char thresh,
-
-|vp8_loop_filter_vertical_edge_y_neon| PROC
- push {lr}
- vpush {d8-d15}
-
- vdup.u8 q0, r2 ; duplicate blimit
- vdup.u8 q1, r3 ; duplicate limit
- sub r2, r0, #4 ; src ptr down by 4 columns
- add r1, r1, r1
- ldr r3, [sp, #68] ; load thresh
- add r12, r2, r1, asr #1
-
- vld1.u8 {d6}, [r2], r1
- vld1.u8 {d8}, [r12], r1
- vld1.u8 {d10}, [r2], r1
- vld1.u8 {d12}, [r12], r1
- vld1.u8 {d14}, [r2], r1
- vld1.u8 {d16}, [r12], r1
- vld1.u8 {d18}, [r2], r1
- vld1.u8 {d20}, [r12], r1
-
- vld1.u8 {d7}, [r2], r1 ; load second 8-line src data
- vld1.u8 {d9}, [r12], r1
- vld1.u8 {d11}, [r2], r1
- vld1.u8 {d13}, [r12], r1
- vld1.u8 {d15}, [r2], r1
- vld1.u8 {d17}, [r12], r1
- vld1.u8 {d19}, [r2]
- vld1.u8 {d21}, [r12]
-
- ;transpose to 8x16 matrix
- vtrn.32 q3, q7
- vtrn.32 q4, q8
- vtrn.32 q5, q9
- vtrn.32 q6, q10
-
- vdup.u8 q2, r3 ; duplicate thresh
-
- vtrn.16 q3, q5
- vtrn.16 q4, q6
- vtrn.16 q7, q9
- vtrn.16 q8, q10
-
- vtrn.8 q3, q4
- vtrn.8 q5, q6
- vtrn.8 q7, q8
- vtrn.8 q9, q10
-
- bl vp8_loop_filter_neon
-
- vswp d12, d11
- vswp d16, d13
-
- sub r0, r0, #2 ; dst ptr
-
- vswp d14, d12
- vswp d16, d15
-
- add r12, r0, r1, asr #1
-
- ;store op1, op0, oq0, oq1
- vst4.8 {d10[0], d11[0], d12[0], d13[0]}, [r0], r1
- vst4.8 {d10[1], d11[1], d12[1], d13[1]}, [r12], r1
- vst4.8 {d10[2], d11[2], d12[2], d13[2]}, [r0], r1
- vst4.8 {d10[3], d11[3], d12[3], d13[3]}, [r12], r1
- vst4.8 {d10[4], d11[4], d12[4], d13[4]}, [r0], r1
- vst4.8 {d10[5], d11[5], d12[5], d13[5]}, [r12], r1
- vst4.8 {d10[6], d11[6], d12[6], d13[6]}, [r0], r1
- vst4.8 {d10[7], d11[7], d12[7], d13[7]}, [r12], r1
-
- vst4.8 {d14[0], d15[0], d16[0], d17[0]}, [r0], r1
- vst4.8 {d14[1], d15[1], d16[1], d17[1]}, [r12], r1
- vst4.8 {d14[2], d15[2], d16[2], d17[2]}, [r0], r1
- vst4.8 {d14[3], d15[3], d16[3], d17[3]}, [r12], r1
- vst4.8 {d14[4], d15[4], d16[4], d17[4]}, [r0], r1
- vst4.8 {d14[5], d15[5], d16[5], d17[5]}, [r12], r1
- vst4.8 {d14[6], d15[6], d16[6], d17[6]}, [r0]
- vst4.8 {d14[7], d15[7], d16[7], d17[7]}, [r12]
-
- vpop {d8-d15}
- pop {pc}
- ENDP ; |vp8_loop_filter_vertical_edge_y_neon|
-
-; void vp8_loop_filter_vertical_edge_uv_neon(unsigned char *u, int pitch
-; const signed char *flimit,
-; const signed char *limit,
-; const signed char *thresh,
-; unsigned char *v)
-; r0 unsigned char *u,
-; r1 int pitch,
-; r2 unsigned char blimit
-; r3 unsigned char limit
-; sp unsigned char thresh,
-; sp+4 unsigned char *v
-|vp8_loop_filter_vertical_edge_uv_neon| PROC
- push {lr}
- vpush {d8-d15}
-
- vdup.u8 q0, r2 ; duplicate blimit
- sub r12, r0, #4 ; move u pointer down by 4 columns
- ldr r2, [sp, #72] ; load v ptr
- vdup.u8 q1, r3 ; duplicate limit
- sub r3, r2, #4 ; move v pointer down by 4 columns
-
- vld1.u8 {d6}, [r12], r1 ;load u data
- vld1.u8 {d7}, [r3], r1 ;load v data
- vld1.u8 {d8}, [r12], r1
- vld1.u8 {d9}, [r3], r1
- vld1.u8 {d10}, [r12], r1
- vld1.u8 {d11}, [r3], r1
- vld1.u8 {d12}, [r12], r1
- vld1.u8 {d13}, [r3], r1
- vld1.u8 {d14}, [r12], r1
- vld1.u8 {d15}, [r3], r1
- vld1.u8 {d16}, [r12], r1
- vld1.u8 {d17}, [r3], r1
- vld1.u8 {d18}, [r12], r1
- vld1.u8 {d19}, [r3], r1
- vld1.u8 {d20}, [r12]
- vld1.u8 {d21}, [r3]
-
- ldr r12, [sp, #68] ; load thresh
-
- ;transpose to 8x16 matrix
- vtrn.32 q3, q7
- vtrn.32 q4, q8
- vtrn.32 q5, q9
- vtrn.32 q6, q10
-
- vdup.u8 q2, r12 ; duplicate thresh
-
- vtrn.16 q3, q5
- vtrn.16 q4, q6
- vtrn.16 q7, q9
- vtrn.16 q8, q10
-
- vtrn.8 q3, q4
- vtrn.8 q5, q6
- vtrn.8 q7, q8
- vtrn.8 q9, q10
-
- bl vp8_loop_filter_neon
-
- vswp d12, d11
- vswp d16, d13
- vswp d14, d12
- vswp d16, d15
-
- sub r0, r0, #2
- sub r2, r2, #2
-
- ;store op1, op0, oq0, oq1
- vst4.8 {d10[0], d11[0], d12[0], d13[0]}, [r0], r1
- vst4.8 {d14[0], d15[0], d16[0], d17[0]}, [r2], r1
- vst4.8 {d10[1], d11[1], d12[1], d13[1]}, [r0], r1
- vst4.8 {d14[1], d15[1], d16[1], d17[1]}, [r2], r1
- vst4.8 {d10[2], d11[2], d12[2], d13[2]}, [r0], r1
- vst4.8 {d14[2], d15[2], d16[2], d17[2]}, [r2], r1
- vst4.8 {d10[3], d11[3], d12[3], d13[3]}, [r0], r1
- vst4.8 {d14[3], d15[3], d16[3], d17[3]}, [r2], r1
- vst4.8 {d10[4], d11[4], d12[4], d13[4]}, [r0], r1
- vst4.8 {d14[4], d15[4], d16[4], d17[4]}, [r2], r1
- vst4.8 {d10[5], d11[5], d12[5], d13[5]}, [r0], r1
- vst4.8 {d14[5], d15[5], d16[5], d17[5]}, [r2], r1
- vst4.8 {d10[6], d11[6], d12[6], d13[6]}, [r0], r1
- vst4.8 {d14[6], d15[6], d16[6], d17[6]}, [r2], r1
- vst4.8 {d10[7], d11[7], d12[7], d13[7]}, [r0]
- vst4.8 {d14[7], d15[7], d16[7], d17[7]}, [r2]
-
- vpop {d8-d15}
- pop {pc}
- ENDP ; |vp8_loop_filter_vertical_edge_uv_neon|
-
-; void vp8_loop_filter_neon();
-; This is a helper function for the loopfilters. The invidual functions do the
-; necessary load, transpose (if necessary) and store.
-
-; r0-r3 PRESERVE
-; q0 flimit
-; q1 limit
-; q2 thresh
-; q3 p3
-; q4 p2
-; q5 p1
-; q6 p0
-; q7 q0
-; q8 q1
-; q9 q2
-; q10 q3
-|vp8_loop_filter_neon| PROC
-
- ; vp8_filter_mask
- vabd.u8 q11, q3, q4 ; abs(p3 - p2)
- vabd.u8 q12, q4, q5 ; abs(p2 - p1)
- vabd.u8 q13, q5, q6 ; abs(p1 - p0)
- vabd.u8 q14, q8, q7 ; abs(q1 - q0)
- vabd.u8 q3, q9, q8 ; abs(q2 - q1)
- vabd.u8 q4, q10, q9 ; abs(q3 - q2)
-
- vmax.u8 q11, q11, q12
- vmax.u8 q12, q13, q14
- vmax.u8 q3, q3, q4
- vmax.u8 q15, q11, q12
-
- vabd.u8 q9, q6, q7 ; abs(p0 - q0)
-
- ; vp8_hevmask
- vcgt.u8 q13, q13, q2 ; (abs(p1 - p0) > thresh)*-1
- vcgt.u8 q14, q14, q2 ; (abs(q1 - q0) > thresh)*-1
- vmax.u8 q15, q15, q3
-
- vmov.u8 q10, #0x80 ; 0x80
-
- vabd.u8 q2, q5, q8 ; a = abs(p1 - q1)
- vqadd.u8 q9, q9, q9 ; b = abs(p0 - q0) * 2
-
- vcge.u8 q15, q1, q15
-
- ; vp8_filter() function
- ; convert to signed
- veor q7, q7, q10 ; qs0
- vshr.u8 q2, q2, #1 ; a = a / 2
- veor q6, q6, q10 ; ps0
-
- veor q5, q5, q10 ; ps1
- vqadd.u8 q9, q9, q2 ; a = b + a
-
- veor q8, q8, q10 ; qs1
-
- vmov.u8 q10, #3 ; #3
-
- vsubl.s8 q2, d14, d12 ; ( qs0 - ps0)
- vsubl.s8 q11, d15, d13
-
- vcge.u8 q9, q0, q9 ; (a > flimit * 2 + limit) * -1
-
- vmovl.u8 q4, d20
-
- vqsub.s8 q1, q5, q8 ; vp8_filter = clamp(ps1-qs1)
- vorr q14, q13, q14 ; vp8_hevmask
-
- vmul.i16 q2, q2, q4 ; 3 * ( qs0 - ps0)
- vmul.i16 q11, q11, q4
-
- vand q1, q1, q14 ; vp8_filter &= hev
- vand q15, q15, q9 ; vp8_filter_mask
-
- vaddw.s8 q2, q2, d2
- vaddw.s8 q11, q11, d3
-
- vmov.u8 q9, #4 ; #4
-
- ; vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0))
- vqmovn.s16 d2, q2
- vqmovn.s16 d3, q11
- vand q1, q1, q15 ; vp8_filter &= mask
-
- vqadd.s8 q2, q1, q10 ; Filter2 = clamp(vp8_filter+3)
- vqadd.s8 q1, q1, q9 ; Filter1 = clamp(vp8_filter+4)
- vshr.s8 q2, q2, #3 ; Filter2 >>= 3
- vshr.s8 q1, q1, #3 ; Filter1 >>= 3
-
-
- vqadd.s8 q11, q6, q2 ; u = clamp(ps0 + Filter2)
- vqsub.s8 q10, q7, q1 ; u = clamp(qs0 - Filter1)
-
- ; outer tap adjustments: ++vp8_filter >> 1
- vrshr.s8 q1, q1, #1
- vbic q1, q1, q14 ; vp8_filter &= ~hev
- vmov.u8 q0, #0x80 ; 0x80
- vqadd.s8 q13, q5, q1 ; u = clamp(ps1 + vp8_filter)
- vqsub.s8 q12, q8, q1 ; u = clamp(qs1 - vp8_filter)
-
- veor q6, q11, q0 ; *op0 = u^0x80
- veor q7, q10, q0 ; *oq0 = u^0x80
- veor q5, q13, q0 ; *op1 = u^0x80
- veor q8, q12, q0 ; *oq1 = u^0x80
-
- bx lr
- ENDP ; |vp8_loop_filter_horizontal_edge_y_neon|
-
-;-----------------
-
- END
diff --git a/vp8/common/arm/neon/loopfilter_neon.c b/vp8/common/arm/neon/loopfilter_neon.c
new file mode 100644
index 0000000..47f522b
--- /dev/null
+++ b/vp8/common/arm/neon/loopfilter_neon.c
@@ -0,0 +1,538 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+static inline void vp8_loop_filter_neon(
+ uint8x16_t qblimit, // flimit
+ uint8x16_t qlimit, // limit
+ uint8x16_t qthresh, // thresh
+ uint8x16_t q3, // p3
+ uint8x16_t q4, // p2
+ uint8x16_t q5, // p1
+ uint8x16_t q6, // p0
+ uint8x16_t q7, // q0
+ uint8x16_t q8, // q1
+ uint8x16_t q9, // q2
+ uint8x16_t q10, // q3
+ uint8x16_t *q5r, // p1
+ uint8x16_t *q6r, // p0
+ uint8x16_t *q7r, // q0
+ uint8x16_t *q8r) { // q1
+ uint8x16_t q0u8, q1u8, q2u8, q11u8, q12u8, q13u8, q14u8, q15u8;
+ int16x8_t q2s16, q11s16;
+ uint16x8_t q4u16;
+ int8x16_t q1s8, q2s8, q10s8, q11s8, q12s8, q13s8;
+ int8x8_t d2s8, d3s8;
+
+ q11u8 = vabdq_u8(q3, q4);
+ q12u8 = vabdq_u8(q4, q5);
+ q13u8 = vabdq_u8(q5, q6);
+ q14u8 = vabdq_u8(q8, q7);
+ q3 = vabdq_u8(q9, q8);
+ q4 = vabdq_u8(q10, q9);
+
+ q11u8 = vmaxq_u8(q11u8, q12u8);
+ q12u8 = vmaxq_u8(q13u8, q14u8);
+ q3 = vmaxq_u8(q3, q4);
+ q15u8 = vmaxq_u8(q11u8, q12u8);
+
+ q9 = vabdq_u8(q6, q7);
+
+ // vp8_hevmask
+ q13u8 = vcgtq_u8(q13u8, qthresh);
+ q14u8 = vcgtq_u8(q14u8, qthresh);
+ q15u8 = vmaxq_u8(q15u8, q3);
+
+ q2u8 = vabdq_u8(q5, q8);
+ q9 = vqaddq_u8(q9, q9);
+
+ q15u8 = vcgeq_u8(qlimit, q15u8);
+
+ // vp8_filter() function
+ // convert to signed
+ q10 = vdupq_n_u8(0x80);
+ q8 = veorq_u8(q8, q10);
+ q7 = veorq_u8(q7, q10);
+ q6 = veorq_u8(q6, q10);
+ q5 = veorq_u8(q5, q10);
+
+ q2u8 = vshrq_n_u8(q2u8, 1);
+ q9 = vqaddq_u8(q9, q2u8);
+
+ q10 = vdupq_n_u8(3);
+
+ q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q7)),
+ vget_low_s8(vreinterpretq_s8_u8(q6)));
+ q11s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q7)),
+ vget_high_s8(vreinterpretq_s8_u8(q6)));
+
+ q9 = vcgeq_u8(qblimit, q9);
+
+ q1s8 = vqsubq_s8(vreinterpretq_s8_u8(q5),
+ vreinterpretq_s8_u8(q8));
+
+ q14u8 = vorrq_u8(q13u8, q14u8);
+
+ q4u16 = vmovl_u8(vget_low_u8(q10));
+ q2s16 = vmulq_s16(q2s16, vreinterpretq_s16_u16(q4u16));
+ q11s16 = vmulq_s16(q11s16, vreinterpretq_s16_u16(q4u16));
+
+ q1u8 = vandq_u8(vreinterpretq_u8_s8(q1s8), q14u8);
+ q15u8 = vandq_u8(q15u8, q9);
+
+ q1s8 = vreinterpretq_s8_u8(q1u8);
+ q2s16 = vaddw_s8(q2s16, vget_low_s8(q1s8));
+ q11s16 = vaddw_s8(q11s16, vget_high_s8(q1s8));
+
+ q9 = vdupq_n_u8(4);
+ // vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0))
+ d2s8 = vqmovn_s16(q2s16);
+ d3s8 = vqmovn_s16(q11s16);
+ q1s8 = vcombine_s8(d2s8, d3s8);
+ q1u8 = vandq_u8(vreinterpretq_u8_s8(q1s8), q15u8);
+ q1s8 = vreinterpretq_s8_u8(q1u8);
+
+ q2s8 = vqaddq_s8(q1s8, vreinterpretq_s8_u8(q10));
+ q1s8 = vqaddq_s8(q1s8, vreinterpretq_s8_u8(q9));
+ q2s8 = vshrq_n_s8(q2s8, 3);
+ q1s8 = vshrq_n_s8(q1s8, 3);
+
+ q11s8 = vqaddq_s8(vreinterpretq_s8_u8(q6), q2s8);
+ q10s8 = vqsubq_s8(vreinterpretq_s8_u8(q7), q1s8);
+
+ q1s8 = vrshrq_n_s8(q1s8, 1);
+ q1s8 = vbicq_s8(q1s8, vreinterpretq_s8_u8(q14u8));
+
+ q13s8 = vqaddq_s8(vreinterpretq_s8_u8(q5), q1s8);
+ q12s8 = vqsubq_s8(vreinterpretq_s8_u8(q8), q1s8);
+
+ q0u8 = vdupq_n_u8(0x80);
+ *q8r = veorq_u8(vreinterpretq_u8_s8(q12s8), q0u8);
+ *q7r = veorq_u8(vreinterpretq_u8_s8(q10s8), q0u8);
+ *q6r = veorq_u8(vreinterpretq_u8_s8(q11s8), q0u8);
+ *q5r = veorq_u8(vreinterpretq_u8_s8(q13s8), q0u8);
+ return;
+}
+
+void vp8_loop_filter_horizontal_edge_y_neon(
+ unsigned char *src,
+ int pitch,
+ unsigned char blimit,
+ unsigned char limit,
+ unsigned char thresh) {
+ uint8x16_t qblimit, qlimit, qthresh, q3, q4;
+ uint8x16_t q5, q6, q7, q8, q9, q10;
+
+ qblimit = vdupq_n_u8(blimit);
+ qlimit = vdupq_n_u8(limit);
+ qthresh = vdupq_n_u8(thresh);
+ src -= (pitch << 2);
+
+ q3 = vld1q_u8(src);
+ src += pitch;
+ q4 = vld1q_u8(src);
+ src += pitch;
+ q5 = vld1q_u8(src);
+ src += pitch;
+ q6 = vld1q_u8(src);
+ src += pitch;
+ q7 = vld1q_u8(src);
+ src += pitch;
+ q8 = vld1q_u8(src);
+ src += pitch;
+ q9 = vld1q_u8(src);
+ src += pitch;
+ q10 = vld1q_u8(src);
+
+ vp8_loop_filter_neon(qblimit, qlimit, qthresh, q3, q4,
+ q5, q6, q7, q8, q9, q10,
+ &q5, &q6, &q7, &q8);
+
+ src -= (pitch * 5);
+ vst1q_u8(src, q5);
+ src += pitch;
+ vst1q_u8(src, q6);
+ src += pitch;
+ vst1q_u8(src, q7);
+ src += pitch;
+ vst1q_u8(src, q8);
+ return;
+}
+
+void vp8_loop_filter_horizontal_edge_uv_neon(
+ unsigned char *u,
+ int pitch,
+ unsigned char blimit,
+ unsigned char limit,
+ unsigned char thresh,
+ unsigned char *v) {
+ uint8x16_t qblimit, qlimit, qthresh, q3, q4;
+ uint8x16_t q5, q6, q7, q8, q9, q10;
+ uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14;
+ uint8x8_t d15, d16, d17, d18, d19, d20, d21;
+
+ qblimit = vdupq_n_u8(blimit);
+ qlimit = vdupq_n_u8(limit);
+ qthresh = vdupq_n_u8(thresh);
+
+ u -= (pitch << 2);
+ v -= (pitch << 2);
+
+ d6 = vld1_u8(u);
+ u += pitch;
+ d7 = vld1_u8(v);
+ v += pitch;
+ d8 = vld1_u8(u);
+ u += pitch;
+ d9 = vld1_u8(v);
+ v += pitch;
+ d10 = vld1_u8(u);
+ u += pitch;
+ d11 = vld1_u8(v);
+ v += pitch;
+ d12 = vld1_u8(u);
+ u += pitch;
+ d13 = vld1_u8(v);
+ v += pitch;
+ d14 = vld1_u8(u);
+ u += pitch;
+ d15 = vld1_u8(v);
+ v += pitch;
+ d16 = vld1_u8(u);
+ u += pitch;
+ d17 = vld1_u8(v);
+ v += pitch;
+ d18 = vld1_u8(u);
+ u += pitch;
+ d19 = vld1_u8(v);
+ v += pitch;
+ d20 = vld1_u8(u);
+ d21 = vld1_u8(v);
+
+ q3 = vcombine_u8(d6, d7);
+ q4 = vcombine_u8(d8, d9);
+ q5 = vcombine_u8(d10, d11);
+ q6 = vcombine_u8(d12, d13);
+ q7 = vcombine_u8(d14, d15);
+ q8 = vcombine_u8(d16, d17);
+ q9 = vcombine_u8(d18, d19);
+ q10 = vcombine_u8(d20, d21);
+
+ vp8_loop_filter_neon(qblimit, qlimit, qthresh, q3, q4,
+ q5, q6, q7, q8, q9, q10,
+ &q5, &q6, &q7, &q8);
+
+ u -= (pitch * 5);
+ vst1_u8(u, vget_low_u8(q5));
+ u += pitch;
+ vst1_u8(u, vget_low_u8(q6));
+ u += pitch;
+ vst1_u8(u, vget_low_u8(q7));
+ u += pitch;
+ vst1_u8(u, vget_low_u8(q8));
+
+ v -= (pitch * 5);
+ vst1_u8(v, vget_high_u8(q5));
+ v += pitch;
+ vst1_u8(v, vget_high_u8(q6));
+ v += pitch;
+ vst1_u8(v, vget_high_u8(q7));
+ v += pitch;
+ vst1_u8(v, vget_high_u8(q8));
+ return;
+}
+
+void vp8_loop_filter_vertical_edge_y_neon(
+ unsigned char *src,
+ int pitch,
+ unsigned char blimit,
+ unsigned char limit,
+ unsigned char thresh) {
+ unsigned char *s, *d;
+ uint8x16_t qblimit, qlimit, qthresh, q3, q4;
+ uint8x16_t q5, q6, q7, q8, q9, q10;
+ uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14;
+ uint8x8_t d15, d16, d17, d18, d19, d20, d21;
+ uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3;
+ uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7;
+ uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11;
+ uint8x8x4_t q4ResultH, q4ResultL;
+
+ qblimit = vdupq_n_u8(blimit);
+ qlimit = vdupq_n_u8(limit);
+ qthresh = vdupq_n_u8(thresh);
+
+ s = src - 4;
+ d6 = vld1_u8(s);
+ s += pitch;
+ d8 = vld1_u8(s);
+ s += pitch;
+ d10 = vld1_u8(s);
+ s += pitch;
+ d12 = vld1_u8(s);
+ s += pitch;
+ d14 = vld1_u8(s);
+ s += pitch;
+ d16 = vld1_u8(s);
+ s += pitch;
+ d18 = vld1_u8(s);
+ s += pitch;
+ d20 = vld1_u8(s);
+ s += pitch;
+ d7 = vld1_u8(s);
+ s += pitch;
+ d9 = vld1_u8(s);
+ s += pitch;
+ d11 = vld1_u8(s);
+ s += pitch;
+ d13 = vld1_u8(s);
+ s += pitch;
+ d15 = vld1_u8(s);
+ s += pitch;
+ d17 = vld1_u8(s);
+ s += pitch;
+ d19 = vld1_u8(s);
+ s += pitch;
+ d21 = vld1_u8(s);
+
+ q3 = vcombine_u8(d6, d7);
+ q4 = vcombine_u8(d8, d9);
+ q5 = vcombine_u8(d10, d11);
+ q6 = vcombine_u8(d12, d13);
+ q7 = vcombine_u8(d14, d15);
+ q8 = vcombine_u8(d16, d17);
+ q9 = vcombine_u8(d18, d19);
+ q10 = vcombine_u8(d20, d21);
+
+ q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7));
+ q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8));
+ q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9));
+ q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10));
+
+ q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]),
+ vreinterpretq_u16_u32(q2tmp2.val[0]));
+ q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]),
+ vreinterpretq_u16_u32(q2tmp3.val[0]));
+ q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]),
+ vreinterpretq_u16_u32(q2tmp2.val[1]));
+ q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]),
+ vreinterpretq_u16_u32(q2tmp3.val[1]));
+
+ q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]),
+ vreinterpretq_u8_u16(q2tmp5.val[0]));
+ q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]),
+ vreinterpretq_u8_u16(q2tmp5.val[1]));
+ q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]),
+ vreinterpretq_u8_u16(q2tmp7.val[0]));
+ q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]),
+ vreinterpretq_u8_u16(q2tmp7.val[1]));
+
+ q3 = q2tmp8.val[0];
+ q4 = q2tmp8.val[1];
+ q5 = q2tmp9.val[0];
+ q6 = q2tmp9.val[1];
+ q7 = q2tmp10.val[0];
+ q8 = q2tmp10.val[1];
+ q9 = q2tmp11.val[0];
+ q10 = q2tmp11.val[1];
+
+ vp8_loop_filter_neon(qblimit, qlimit, qthresh, q3, q4,
+ q5, q6, q7, q8, q9, q10,
+ &q5, &q6, &q7, &q8);
+
+ q4ResultL.val[0] = vget_low_u8(q5); // d10
+ q4ResultL.val[1] = vget_low_u8(q6); // d12
+ q4ResultL.val[2] = vget_low_u8(q7); // d14
+ q4ResultL.val[3] = vget_low_u8(q8); // d16
+ q4ResultH.val[0] = vget_high_u8(q5); // d11
+ q4ResultH.val[1] = vget_high_u8(q6); // d13
+ q4ResultH.val[2] = vget_high_u8(q7); // d15
+ q4ResultH.val[3] = vget_high_u8(q8); // d17
+
+ d = src - 2;
+ vst4_lane_u8(d, q4ResultL, 0);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultL, 1);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultL, 2);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultL, 3);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultL, 4);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultL, 5);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultL, 6);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultL, 7);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultH, 0);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultH, 1);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultH, 2);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultH, 3);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultH, 4);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultH, 5);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultH, 6);
+ d += pitch;
+ vst4_lane_u8(d, q4ResultH, 7);
+ return;
+}
+
+void vp8_loop_filter_vertical_edge_uv_neon(
+ unsigned char *u,
+ int pitch,
+ unsigned char blimit,
+ unsigned char limit,
+ unsigned char thresh,
+ unsigned char *v) {
+ unsigned char *us, *ud;
+ unsigned char *vs, *vd;
+ uint8x16_t qblimit, qlimit, qthresh, q3, q4;
+ uint8x16_t q5, q6, q7, q8, q9, q10;
+ uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14;
+ uint8x8_t d15, d16, d17, d18, d19, d20, d21;
+ uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3;
+ uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7;
+ uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11;
+ uint8x8x4_t q4ResultH, q4ResultL;
+
+ qblimit = vdupq_n_u8(blimit);
+ qlimit = vdupq_n_u8(limit);
+ qthresh = vdupq_n_u8(thresh);
+
+ us = u - 4;
+ d6 = vld1_u8(us);
+ us += pitch;
+ d8 = vld1_u8(us);
+ us += pitch;
+ d10 = vld1_u8(us);
+ us += pitch;
+ d12 = vld1_u8(us);
+ us += pitch;
+ d14 = vld1_u8(us);
+ us += pitch;
+ d16 = vld1_u8(us);
+ us += pitch;
+ d18 = vld1_u8(us);
+ us += pitch;
+ d20 = vld1_u8(us);
+
+ vs = v - 4;
+ d7 = vld1_u8(vs);
+ vs += pitch;
+ d9 = vld1_u8(vs);
+ vs += pitch;
+ d11 = vld1_u8(vs);
+ vs += pitch;
+ d13 = vld1_u8(vs);
+ vs += pitch;
+ d15 = vld1_u8(vs);
+ vs += pitch;
+ d17 = vld1_u8(vs);
+ vs += pitch;
+ d19 = vld1_u8(vs);
+ vs += pitch;
+ d21 = vld1_u8(vs);
+
+ q3 = vcombine_u8(d6, d7);
+ q4 = vcombine_u8(d8, d9);
+ q5 = vcombine_u8(d10, d11);
+ q6 = vcombine_u8(d12, d13);
+ q7 = vcombine_u8(d14, d15);
+ q8 = vcombine_u8(d16, d17);
+ q9 = vcombine_u8(d18, d19);
+ q10 = vcombine_u8(d20, d21);
+
+ q2tmp0 = vtrnq_u32(vreinterpretq_u32_u8(q3), vreinterpretq_u32_u8(q7));
+ q2tmp1 = vtrnq_u32(vreinterpretq_u32_u8(q4), vreinterpretq_u32_u8(q8));
+ q2tmp2 = vtrnq_u32(vreinterpretq_u32_u8(q5), vreinterpretq_u32_u8(q9));
+ q2tmp3 = vtrnq_u32(vreinterpretq_u32_u8(q6), vreinterpretq_u32_u8(q10));
+
+ q2tmp4 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[0]),
+ vreinterpretq_u16_u32(q2tmp2.val[0]));
+ q2tmp5 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[0]),
+ vreinterpretq_u16_u32(q2tmp3.val[0]));
+ q2tmp6 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp0.val[1]),
+ vreinterpretq_u16_u32(q2tmp2.val[1]));
+ q2tmp7 = vtrnq_u16(vreinterpretq_u16_u32(q2tmp1.val[1]),
+ vreinterpretq_u16_u32(q2tmp3.val[1]));
+
+ q2tmp8 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[0]),
+ vreinterpretq_u8_u16(q2tmp5.val[0]));
+ q2tmp9 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp4.val[1]),
+ vreinterpretq_u8_u16(q2tmp5.val[1]));
+ q2tmp10 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[0]),
+ vreinterpretq_u8_u16(q2tmp7.val[0]));
+ q2tmp11 = vtrnq_u8(vreinterpretq_u8_u16(q2tmp6.val[1]),
+ vreinterpretq_u8_u16(q2tmp7.val[1]));
+
+ q3 = q2tmp8.val[0];
+ q4 = q2tmp8.val[1];
+ q5 = q2tmp9.val[0];
+ q6 = q2tmp9.val[1];
+ q7 = q2tmp10.val[0];
+ q8 = q2tmp10.val[1];
+ q9 = q2tmp11.val[0];
+ q10 = q2tmp11.val[1];
+
+ vp8_loop_filter_neon(qblimit, qlimit, qthresh, q3, q4,
+ q5, q6, q7, q8, q9, q10,
+ &q5, &q6, &q7, &q8);
+
+ q4ResultL.val[0] = vget_low_u8(q5); // d10
+ q4ResultL.val[1] = vget_low_u8(q6); // d12
+ q4ResultL.val[2] = vget_low_u8(q7); // d14
+ q4ResultL.val[3] = vget_low_u8(q8); // d16
+ ud = u - 2;
+ vst4_lane_u8(ud, q4ResultL, 0);
+ ud += pitch;
+ vst4_lane_u8(ud, q4ResultL, 1);
+ ud += pitch;
+ vst4_lane_u8(ud, q4ResultL, 2);
+ ud += pitch;
+ vst4_lane_u8(ud, q4ResultL, 3);
+ ud += pitch;
+ vst4_lane_u8(ud, q4ResultL, 4);
+ ud += pitch;
+ vst4_lane_u8(ud, q4ResultL, 5);
+ ud += pitch;
+ vst4_lane_u8(ud, q4ResultL, 6);
+ ud += pitch;
+ vst4_lane_u8(ud, q4ResultL, 7);
+
+ q4ResultH.val[0] = vget_high_u8(q5); // d11
+ q4ResultH.val[1] = vget_high_u8(q6); // d13
+ q4ResultH.val[2] = vget_high_u8(q7); // d15
+ q4ResultH.val[3] = vget_high_u8(q8); // d17
+ vd = v - 2;
+ vst4_lane_u8(vd, q4ResultH, 0);
+ vd += pitch;
+ vst4_lane_u8(vd, q4ResultH, 1);
+ vd += pitch;
+ vst4_lane_u8(vd, q4ResultH, 2);
+ vd += pitch;
+ vst4_lane_u8(vd, q4ResultH, 3);
+ vd += pitch;
+ vst4_lane_u8(vd, q4ResultH, 4);
+ vd += pitch;
+ vst4_lane_u8(vd, q4ResultH, 5);
+ vd += pitch;
+ vst4_lane_u8(vd, q4ResultH, 6);
+ vd += pitch;
+ vst4_lane_u8(vd, q4ResultH, 7);
+ return;
+}
diff --git a/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.asm b/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.asm
deleted file mode 100644
index 6eb0651..0000000
--- a/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.asm
+++ /dev/null
@@ -1,117 +0,0 @@
-;
-; Copyright (c) 2010 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 |vp8_loop_filter_bhs_neon|
- EXPORT |vp8_loop_filter_mbhs_neon|
- ARM
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
-; r0 unsigned char *s, PRESERVE
-; r1 int p, PRESERVE
-; q1 limit, PRESERVE
-
-|vp8_loop_filter_simple_horizontal_edge_neon| PROC
- vpush {d8-d15}
- sub r3, r0, r1, lsl #1 ; move src pointer down by 2 lines
-
- vld1.u8 {q7}, [r0@128], r1 ; q0
- vld1.u8 {q5}, [r3@128], r1 ; p0
- vld1.u8 {q8}, [r0@128] ; q1
- vld1.u8 {q6}, [r3@128] ; p1
-
- vabd.u8 q15, q6, q7 ; abs(p0 - q0)
- vabd.u8 q14, q5, q8 ; abs(p1 - q1)
-
- vqadd.u8 q15, q15, q15 ; abs(p0 - q0) * 2
- vshr.u8 q14, q14, #1 ; abs(p1 - q1) / 2
- vmov.u8 q0, #0x80 ; 0x80
- vmov.s16 q13, #3
- vqadd.u8 q15, q15, q14 ; abs(p0 - q0) * 2 + abs(p1 - q1) / 2
-
- veor q7, q7, q0 ; qs0: q0 offset to convert to a signed value
- veor q6, q6, q0 ; ps0: p0 offset to convert to a signed value
- veor q5, q5, q0 ; ps1: p1 offset to convert to a signed value
- veor q8, q8, q0 ; qs1: q1 offset to convert to a signed value
-
- vcge.u8 q15, q1, q15 ; (abs(p0 - q0)*2 + abs(p1-q1)/2 > limit)*-1
-
- vsubl.s8 q2, d14, d12 ; ( qs0 - ps0)
- vsubl.s8 q3, d15, d13
-
- vqsub.s8 q4, q5, q8 ; q4: vp8_filter = vp8_signed_char_clamp(ps1-qs1)
-
- vmul.s16 q2, q2, q13 ; 3 * ( qs0 - ps0)
- vmul.s16 q3, q3, q13
-
- vmov.u8 q10, #0x03 ; 0x03
- vmov.u8 q9, #0x04 ; 0x04
-
- vaddw.s8 q2, q2, d8 ; vp8_filter + 3 * ( qs0 - ps0)
- vaddw.s8 q3, q3, d9
-
- vqmovn.s16 d8, q2 ; vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * ( qs0 - ps0))
- vqmovn.s16 d9, q3
-
- vand q14, q4, q15 ; vp8_filter &= mask
-
- vqadd.s8 q2, q14, q10 ; Filter2 = vp8_signed_char_clamp(vp8_filter+3)
- vqadd.s8 q3, q14, q9 ; Filter1 = vp8_signed_char_clamp(vp8_filter+4)
- vshr.s8 q2, q2, #3 ; Filter2 >>= 3
- vshr.s8 q4, q3, #3 ; Filter1 >>= 3
-
- sub r0, r0, r1
-
- ;calculate output
- vqadd.s8 q11, q6, q2 ; u = vp8_signed_char_clamp(ps0 + Filter2)
- vqsub.s8 q10, q7, q4 ; u = vp8_signed_char_clamp(qs0 - Filter1)
-
- veor q6, q11, q0 ; *op0 = u^0x80
- veor q7, q10, q0 ; *oq0 = u^0x80
-
- vst1.u8 {q6}, [r3@128] ; store op0
- vst1.u8 {q7}, [r0@128] ; store oq0
-
- vpop {d8-d15}
- bx lr
- ENDP ; |vp8_loop_filter_simple_horizontal_edge_neon|
-
-; r0 unsigned char *y
-; r1 int ystride
-; r2 const unsigned char *blimit
-
-|vp8_loop_filter_bhs_neon| PROC
- push {r4, lr}
- ldrb r3, [r2] ; load blim from mem
- vdup.s8 q1, r3 ; duplicate blim
-
- add r0, r0, r1, lsl #2 ; src = y_ptr + 4 * y_stride
- bl vp8_loop_filter_simple_horizontal_edge_neon
- ; vp8_loop_filter_simple_horizontal_edge_neon preserves r0, r1 and q1
- add r0, r0, r1, lsl #2 ; src = y_ptr + 8* y_stride
- bl vp8_loop_filter_simple_horizontal_edge_neon
- add r0, r0, r1, lsl #2 ; src = y_ptr + 12 * y_stride
- pop {r4, lr}
- b vp8_loop_filter_simple_horizontal_edge_neon
- ENDP ;|vp8_loop_filter_bhs_neon|
-
-; r0 unsigned char *y
-; r1 int ystride
-; r2 const unsigned char *blimit
-
-|vp8_loop_filter_mbhs_neon| PROC
- ldrb r3, [r2] ; load blim from mem
- vdup.s8 q1, r3 ; duplicate mblim
- b vp8_loop_filter_simple_horizontal_edge_neon
- ENDP ;|vp8_loop_filter_bhs_neon|
-
- END
diff --git a/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.c b/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.c
new file mode 100644
index 0000000..3d024e0
--- /dev/null
+++ b/vp8/common/arm/neon/loopfiltersimplehorizontaledge_neon.c
@@ -0,0 +1,110 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+static inline void vp8_loop_filter_simple_horizontal_edge_neon(
+ unsigned char *s,
+ int p,
+ const unsigned char *blimit) {
+ uint8_t *sp;
+ uint8x16_t qblimit, q0u8;
+ uint8x16_t q5u8, q6u8, q7u8, q8u8, q9u8, q10u8, q14u8, q15u8;
+ int16x8_t q2s16, q3s16, q13s16;
+ int8x8_t d8s8, d9s8;
+ int8x16_t q2s8, q3s8, q4s8, q10s8, q11s8, q14s8;
+
+ qblimit = vdupq_n_u8(*blimit);
+
+ sp = s - (p << 1);
+ q5u8 = vld1q_u8(sp);
+ sp += p;
+ q6u8 = vld1q_u8(sp);
+ sp += p;
+ q7u8 = vld1q_u8(sp);
+ sp += p;
+ q8u8 = vld1q_u8(sp);
+
+ q15u8 = vabdq_u8(q6u8, q7u8);
+ q14u8 = vabdq_u8(q5u8, q8u8);
+
+ q15u8 = vqaddq_u8(q15u8, q15u8);
+ q14u8 = vshrq_n_u8(q14u8, 1);
+ q0u8 = vdupq_n_u8(0x80);
+ q13s16 = vdupq_n_s16(3);
+ q15u8 = vqaddq_u8(q15u8, q14u8);
+
+ q5u8 = veorq_u8(q5u8, q0u8);
+ q6u8 = veorq_u8(q6u8, q0u8);
+ q7u8 = veorq_u8(q7u8, q0u8);
+ q8u8 = veorq_u8(q8u8, q0u8);
+
+ q15u8 = vcgeq_u8(qblimit, q15u8);
+
+ q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q7u8)),
+ vget_low_s8(vreinterpretq_s8_u8(q6u8)));
+ q3s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q7u8)),
+ vget_high_s8(vreinterpretq_s8_u8(q6u8)));
+
+ q4s8 = vqsubq_s8(vreinterpretq_s8_u8(q5u8),
+ vreinterpretq_s8_u8(q8u8));
+
+ q2s16 = vmulq_s16(q2s16, q13s16);
+ q3s16 = vmulq_s16(q3s16, q13s16);
+
+ q10u8 = vdupq_n_u8(3);
+ q9u8 = vdupq_n_u8(4);
+
+ q2s16 = vaddw_s8(q2s16, vget_low_s8(q4s8));
+ q3s16 = vaddw_s8(q3s16, vget_high_s8(q4s8));
+
+ d8s8 = vqmovn_s16(q2s16);
+ d9s8 = vqmovn_s16(q3s16);
+ q4s8 = vcombine_s8(d8s8, d9s8);
+
+ q14s8 = vandq_s8(q4s8, vreinterpretq_s8_u8(q15u8));
+
+ q2s8 = vqaddq_s8(q14s8, vreinterpretq_s8_u8(q10u8));
+ q3s8 = vqaddq_s8(q14s8, vreinterpretq_s8_u8(q9u8));
+ q2s8 = vshrq_n_s8(q2s8, 3);
+ q3s8 = vshrq_n_s8(q3s8, 3);
+
+ q11s8 = vqaddq_s8(vreinterpretq_s8_u8(q6u8), q2s8);
+ q10s8 = vqsubq_s8(vreinterpretq_s8_u8(q7u8), q3s8);
+
+ q6u8 = veorq_u8(vreinterpretq_u8_s8(q11s8), q0u8);
+ q7u8 = veorq_u8(vreinterpretq_u8_s8(q10s8), q0u8);
+
+ vst1q_u8(s, q7u8);
+ s -= p;
+ vst1q_u8(s, q6u8);
+ return;
+}
+
+void vp8_loop_filter_bhs_neon(
+ unsigned char *y_ptr,
+ int y_stride,
+ const unsigned char *blimit) {
+ y_ptr += y_stride * 4;
+ vp8_loop_filter_simple_horizontal_edge_neon(y_ptr, y_stride, blimit);
+ y_ptr += y_stride * 4;
+ vp8_loop_filter_simple_horizontal_edge_neon(y_ptr, y_stride, blimit);
+ y_ptr += y_stride * 4;
+ vp8_loop_filter_simple_horizontal_edge_neon(y_ptr, y_stride, blimit);
+ return;
+}
+
+void vp8_loop_filter_mbhs_neon(
+ unsigned char *y_ptr,
+ int y_stride,
+ const unsigned char *blimit) {
+ vp8_loop_filter_simple_horizontal_edge_neon(y_ptr, y_stride, blimit);
+ return;
+}
diff --git a/vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.asm b/vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.asm
deleted file mode 100644
index 78d13c8..0000000
--- a/vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.asm
+++ /dev/null
@@ -1,156 +0,0 @@
-;
-; Copyright (c) 2010 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 |vp8_loop_filter_bvs_neon|
- EXPORT |vp8_loop_filter_mbvs_neon|
- ARM
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
-; r0 unsigned char *s, PRESERVE
-; r1 int p, PRESERVE
-; q1 limit, PRESERVE
-
-|vp8_loop_filter_simple_vertical_edge_neon| PROC
- vpush {d8-d15}
-
- sub r0, r0, #2 ; move src pointer down by 2 columns
- add r12, r1, r1
- add r3, r0, r1
-
- vld4.8 {d6[0], d7[0], d8[0], d9[0]}, [r0], r12
- vld4.8 {d6[1], d7[1], d8[1], d9[1]}, [r3], r12
- vld4.8 {d6[2], d7[2], d8[2], d9[2]}, [r0], r12
- vld4.8 {d6[3], d7[3], d8[3], d9[3]}, [r3], r12
- vld4.8 {d6[4], d7[4], d8[4], d9[4]}, [r0], r12
- vld4.8 {d6[5], d7[5], d8[5], d9[5]}, [r3], r12
- vld4.8 {d6[6], d7[6], d8[6], d9[6]}, [r0], r12
- vld4.8 {d6[7], d7[7], d8[7], d9[7]}, [r3], r12
-
- vld4.8 {d10[0], d11[0], d12[0], d13[0]}, [r0], r12
- vld4.8 {d10[1], d11[1], d12[1], d13[1]}, [r3], r12
- vld4.8 {d10[2], d11[2], d12[2], d13[2]}, [r0], r12
- vld4.8 {d10[3], d11[3], d12[3], d13[3]}, [r3], r12
- vld4.8 {d10[4], d11[4], d12[4], d13[4]}, [r0], r12
- vld4.8 {d10[5], d11[5], d12[5], d13[5]}, [r3], r12
- vld4.8 {d10[6], d11[6], d12[6], d13[6]}, [r0], r12
- vld4.8 {d10[7], d11[7], d12[7], d13[7]}, [r3]
-
- vswp d7, d10
- vswp d12, d9
-
- ;vp8_filter_mask() function
- ;vp8_hevmask() function
- sub r0, r0, r1, lsl #4
- vabd.u8 q15, q5, q4 ; abs(p0 - q0)
- vabd.u8 q14, q3, q6 ; abs(p1 - q1)
-
- vqadd.u8 q15, q15, q15 ; abs(p0 - q0) * 2
- vshr.u8 q14, q14, #1 ; abs(p1 - q1) / 2
- vmov.u8 q0, #0x80 ; 0x80
- vmov.s16 q11, #3
- vqadd.u8 q15, q15, q14 ; abs(p0 - q0) * 2 + abs(p1 - q1) / 2
-
- veor q4, q4, q0 ; qs0: q0 offset to convert to a signed value
- veor q5, q5, q0 ; ps0: p0 offset to convert to a signed value
- veor q3, q3, q0 ; ps1: p1 offset to convert to a signed value
- veor q6, q6, q0 ; qs1: q1 offset to convert to a signed value
-
- vcge.u8 q15, q1, q15 ; abs(p0 - q0)*2 + abs(p1-q1)/2 > flimit*2 + limit)*-1
-
- vsubl.s8 q2, d8, d10 ; ( qs0 - ps0)
- vsubl.s8 q13, d9, d11
-
- vqsub.s8 q14, q3, q6 ; vp8_filter = vp8_signed_char_clamp(ps1-qs1)
-
- vmul.s16 q2, q2, q11 ; 3 * ( qs0 - ps0)
- vmul.s16 q13, q13, q11
-
- vmov.u8 q11, #0x03 ; 0x03
- vmov.u8 q12, #0x04 ; 0x04
-
- vaddw.s8 q2, q2, d28 ; vp8_filter + 3 * ( qs0 - ps0)
- vaddw.s8 q13, q13, d29
-
- vqmovn.s16 d28, q2 ; vp8_filter = vp8_signed_char_clamp(vp8_filter + 3 * ( qs0 - ps0))
- vqmovn.s16 d29, q13
-
- add r0, r0, #1
- add r3, r0, r1
-
- vand q14, q14, q15 ; vp8_filter &= mask
-
- vqadd.s8 q2, q14, q11 ; Filter2 = vp8_signed_char_clamp(vp8_filter+3)
- vqadd.s8 q3, q14, q12 ; Filter1 = vp8_signed_char_clamp(vp8_filter+4)
- vshr.s8 q2, q2, #3 ; Filter2 >>= 3
- vshr.s8 q14, q3, #3 ; Filter1 >>= 3
-
- ;calculate output
- vqadd.s8 q11, q5, q2 ; u = vp8_signed_char_clamp(ps0 + Filter2)
- vqsub.s8 q10, q4, q14 ; u = vp8_signed_char_clamp(qs0 - Filter1)
-
- veor q6, q11, q0 ; *op0 = u^0x80
- veor q7, q10, q0 ; *oq0 = u^0x80
- add r12, r1, r1
- vswp d13, d14
-
- ;store op1, op0, oq0, oq1
- vst2.8 {d12[0], d13[0]}, [r0], r12
- vst2.8 {d12[1], d13[1]}, [r3], r12
- vst2.8 {d12[2], d13[2]}, [r0], r12
- vst2.8 {d12[3], d13[3]}, [r3], r12
- vst2.8 {d12[4], d13[4]}, [r0], r12
- vst2.8 {d12[5], d13[5]}, [r3], r12
- vst2.8 {d12[6], d13[6]}, [r0], r12
- vst2.8 {d12[7], d13[7]}, [r3], r12
- vst2.8 {d14[0], d15[0]}, [r0], r12
- vst2.8 {d14[1], d15[1]}, [r3], r12
- vst2.8 {d14[2], d15[2]}, [r0], r12
- vst2.8 {d14[3], d15[3]}, [r3], r12
- vst2.8 {d14[4], d15[4]}, [r0], r12
- vst2.8 {d14[5], d15[5]}, [r3], r12
- vst2.8 {d14[6], d15[6]}, [r0], r12
- vst2.8 {d14[7], d15[7]}, [r3]
-
- vpop {d8-d15}
- bx lr
- ENDP ; |vp8_loop_filter_simple_vertical_edge_neon|
-
-; r0 unsigned char *y
-; r1 int ystride
-; r2 const unsigned char *blimit
-
-|vp8_loop_filter_bvs_neon| PROC
- push {r4, lr}
- ldrb r3, [r2] ; load blim from mem
- mov r4, r0
- add r0, r0, #4
- vdup.s8 q1, r3 ; duplicate blim
- bl vp8_loop_filter_simple_vertical_edge_neon
- ; vp8_loop_filter_simple_vertical_edge_neon preserves r1 and q1
- add r0, r4, #8
- bl vp8_loop_filter_simple_vertical_edge_neon
- add r0, r4, #12
- pop {r4, lr}
- b vp8_loop_filter_simple_vertical_edge_neon
- ENDP ;|vp8_loop_filter_bvs_neon|
-
-; r0 unsigned char *y
-; r1 int ystride
-; r2 const unsigned char *blimit
-
-|vp8_loop_filter_mbvs_neon| PROC
- ldrb r3, [r2] ; load mblim from mem
- vdup.s8 q1, r3 ; duplicate mblim
- b vp8_loop_filter_simple_vertical_edge_neon
- ENDP ;|vp8_loop_filter_bvs_neon|
- END
diff --git a/vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.c b/vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.c
new file mode 100644
index 0000000..68bcde4
--- /dev/null
+++ b/vp8/common/arm/neon/loopfiltersimpleverticaledge_neon.c
@@ -0,0 +1,183 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+static inline void vp8_loop_filter_simple_vertical_edge_neon(
+ unsigned char *s,
+ int p,
+ const unsigned char *blimit) {
+ int pitch;
+ unsigned char *src1, *src2;
+ uint8x16_t qblimit, q0u8;
+ uint8x16_t q3u8, q4u8, q5u8, q6u8, q7u8, q11u8, q12u8, q14u8, q15u8;
+ int16x8_t q2s16, q13s16, q11s16;
+ int8x8_t d28s8, d29s8;
+ int8x16_t q2s8, q3s8, q10s8, q11s8, q14s8;
+ uint8x8x4_t d0u8x4; // d6, d7, d8, d9
+ uint8x8x4_t d1u8x4; // d10, d11, d12, d13
+ uint8x8x2_t d2u8x2; // d12, d13
+ uint8x8x2_t d3u8x2; // d14, d15
+
+ pitch = p << 1;
+ qblimit = vdupq_n_u8(*blimit);
+
+ src1 = s - 2;
+ d0u8x4 = vld4_lane_u8(src1, d0u8x4, 0);
+ src1 += pitch;
+ d0u8x4 = vld4_lane_u8(src1, d0u8x4, 2);
+ src1 += pitch;
+ d0u8x4 = vld4_lane_u8(src1, d0u8x4, 4);
+ src1 += pitch;
+ d0u8x4 = vld4_lane_u8(src1, d0u8x4, 6);
+ src1 += pitch;
+ d1u8x4 = vld4_lane_u8(src1, d1u8x4, 0);
+ src1 += pitch;
+ d1u8x4 = vld4_lane_u8(src1, d1u8x4, 2);
+ src1 += pitch;
+ d1u8x4 = vld4_lane_u8(src1, d1u8x4, 4);
+ src1 += pitch;
+ d1u8x4 = vld4_lane_u8(src1, d1u8x4, 6);
+
+ src2 = s - 2 + p;
+ d0u8x4 = vld4_lane_u8(src2, d0u8x4, 1);
+ src2 += pitch;
+ d0u8x4 = vld4_lane_u8(src2, d0u8x4, 3);
+ src2 += pitch;
+ d0u8x4 = vld4_lane_u8(src2, d0u8x4, 5);
+ src2 += pitch;
+ d0u8x4 = vld4_lane_u8(src2, d0u8x4, 7);
+ src2 += pitch;
+ d1u8x4 = vld4_lane_u8(src2, d1u8x4, 1);
+ src2 += pitch;
+ d1u8x4 = vld4_lane_u8(src2, d1u8x4, 3);
+ src2 += pitch;
+ d1u8x4 = vld4_lane_u8(src2, d1u8x4, 5);
+ src2 += pitch;
+ d1u8x4 = vld4_lane_u8(src2, d1u8x4, 7);
+
+ // vswp d7, d10
+ // vswp d12, d9
+ q3u8 = vcombine_u8(d0u8x4.val[0], d1u8x4.val[0]); // d6 d10
+ q4u8 = vcombine_u8(d0u8x4.val[2], d1u8x4.val[2]); // d8 d12
+ q5u8 = vcombine_u8(d0u8x4.val[1], d1u8x4.val[1]); // d7 d11
+ q6u8 = vcombine_u8(d0u8x4.val[3], d1u8x4.val[3]); // d9 d13
+
+ q15u8 = vabdq_u8(q5u8, q4u8);
+ q14u8 = vabdq_u8(q3u8, q6u8);
+
+ q15u8 = vqaddq_u8(q15u8, q15u8);
+ q14u8 = vshrq_n_u8(q14u8, 1);
+ q0u8 = vdupq_n_u8(0x80);
+ q11s16 = vdupq_n_s16(3);
+ q15u8 = vqaddq_u8(q15u8, q14u8);
+
+ q3u8 = veorq_u8(q3u8, q0u8);
+ q4u8 = veorq_u8(q4u8, q0u8);
+ q5u8 = veorq_u8(q5u8, q0u8);
+ q6u8 = veorq_u8(q6u8, q0u8);
+
+ q15u8 = vcgeq_u8(qblimit, q15u8);
+
+ q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q4u8)),
+ vget_low_s8(vreinterpretq_s8_u8(q5u8)));
+ q13s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q4u8)),
+ vget_high_s8(vreinterpretq_s8_u8(q5u8)));
+
+ q14s8 = vqsubq_s8(vreinterpretq_s8_u8(q3u8),
+ vreinterpretq_s8_u8(q6u8));
+
+ q2s16 = vmulq_s16(q2s16, q11s16);
+ q13s16 = vmulq_s16(q13s16, q11s16);
+
+ q11u8 = vdupq_n_u8(3);
+ q12u8 = vdupq_n_u8(4);
+
+ q2s16 = vaddw_s8(q2s16, vget_low_s8(q14s8));
+ q13s16 = vaddw_s8(q13s16, vget_high_s8(q14s8));
+
+ d28s8 = vqmovn_s16(q2s16);
+ d29s8 = vqmovn_s16(q13s16);
+ q14s8 = vcombine_s8(d28s8, d29s8);
+
+ q14s8 = vandq_s8(q14s8, vreinterpretq_s8_u8(q15u8));
+
+ q2s8 = vqaddq_s8(q14s8, vreinterpretq_s8_u8(q11u8));
+ q3s8 = vqaddq_s8(q14s8, vreinterpretq_s8_u8(q12u8));
+ q2s8 = vshrq_n_s8(q2s8, 3);
+ q14s8 = vshrq_n_s8(q3s8, 3);
+
+ q11s8 = vqaddq_s8(vreinterpretq_s8_u8(q5u8), q2s8);
+ q10s8 = vqsubq_s8(vreinterpretq_s8_u8(q4u8), q14s8);
+
+ q6u8 = veorq_u8(vreinterpretq_u8_s8(q11s8), q0u8);
+ q7u8 = veorq_u8(vreinterpretq_u8_s8(q10s8), q0u8);
+
+ d2u8x2.val[0] = vget_low_u8(q6u8); // d12
+ d2u8x2.val[1] = vget_low_u8(q7u8); // d14
+ d3u8x2.val[0] = vget_high_u8(q6u8); // d13
+ d3u8x2.val[1] = vget_high_u8(q7u8); // d15
+
+ src1 = s - 1;
+ vst2_lane_u8(src1, d2u8x2, 0);
+ src1 += pitch;
+ vst2_lane_u8(src1, d2u8x2, 2);
+ src1 += pitch;
+ vst2_lane_u8(src1, d2u8x2, 4);
+ src1 += pitch;
+ vst2_lane_u8(src1, d2u8x2, 6);
+ src1 += pitch;
+ vst2_lane_u8(src1, d3u8x2, 0);
+ src1 += pitch;
+ vst2_lane_u8(src1, d3u8x2, 2);
+ src1 += pitch;
+ vst2_lane_u8(src1, d3u8x2, 4);
+ src1 += pitch;
+ vst2_lane_u8(src1, d3u8x2, 6);
+
+ src2 = s - 1 + p;
+ vst2_lane_u8(src2, d2u8x2, 1);
+ src2 += pitch;
+ vst2_lane_u8(src2, d2u8x2, 3);
+ src2 += pitch;
+ vst2_lane_u8(src2, d2u8x2, 5);
+ src2 += pitch;
+ vst2_lane_u8(src2, d2u8x2, 7);
+ src2 += pitch;
+ vst2_lane_u8(src2, d3u8x2, 1);
+ src2 += pitch;
+ vst2_lane_u8(src2, d3u8x2, 3);
+ src2 += pitch;
+ vst2_lane_u8(src2, d3u8x2, 5);
+ src2 += pitch;
+ vst2_lane_u8(src2, d3u8x2, 7);
+ return;
+}
+
+void vp8_loop_filter_bvs_neon(
+ unsigned char *y_ptr,
+ int y_stride,
+ const unsigned char *blimit) {
+ y_ptr += 4;
+ vp8_loop_filter_simple_vertical_edge_neon(y_ptr, y_stride, blimit);
+ y_ptr += 4;
+ vp8_loop_filter_simple_vertical_edge_neon(y_ptr, y_stride, blimit);
+ y_ptr += 4;
+ vp8_loop_filter_simple_vertical_edge_neon(y_ptr, y_stride, blimit);
+ return;
+}
+
+void vp8_loop_filter_mbvs_neon(
+ unsigned char *y_ptr,
+ int y_stride,
+ const unsigned char *blimit) {
+ vp8_loop_filter_simple_vertical_edge_neon(y_ptr, y_stride, blimit);
+ return;
+}
diff --git a/vp8/common/arm/neon/mbloopfilter_neon.asm b/vp8/common/arm/neon/mbloopfilter_neon.asm
deleted file mode 100644
index d200c30..0000000
--- a/vp8/common/arm/neon/mbloopfilter_neon.asm
+++ /dev/null
@@ -1,481 +0,0 @@
-;
-; Copyright (c) 2010 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 |vp8_mbloop_filter_horizontal_edge_y_neon|
- EXPORT |vp8_mbloop_filter_horizontal_edge_uv_neon|
- EXPORT |vp8_mbloop_filter_vertical_edge_y_neon|
- EXPORT |vp8_mbloop_filter_vertical_edge_uv_neon|
- ARM
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
-; void vp8_mbloop_filter_horizontal_edge_y_neon(unsigned char *src, int pitch,
-; const unsigned char *blimit,
-; const unsigned char *limit,
-; const unsigned char *thresh)
-; r0 unsigned char *src,
-; r1 int pitch,
-; r2 unsigned char blimit
-; r3 unsigned char limit
-; sp unsigned char thresh,
-|vp8_mbloop_filter_horizontal_edge_y_neon| PROC
- push {lr}
- vpush {d8-d15}
-
- add r1, r1, r1 ; double stride
- ldr r12, [sp, #68] ; load thresh
- sub r0, r0, r1, lsl #1 ; move src pointer down by 4 lines
- vdup.u8 q2, r12 ; thresh
- add r12, r0, r1, lsr #1 ; move src pointer up by 1 line
-
- vld1.u8 {q3}, [r0@128], r1 ; p3
- vld1.u8 {q4}, [r12@128], r1 ; p2
- vld1.u8 {q5}, [r0@128], r1 ; p1
- vld1.u8 {q6}, [r12@128], r1 ; p0
- vld1.u8 {q7}, [r0@128], r1 ; q0
- vld1.u8 {q8}, [r12@128], r1 ; q1
- vld1.u8 {q9}, [r0@128], r1 ; q2
- vld1.u8 {q10}, [r12@128], r1 ; q3
-
- bl vp8_mbloop_filter_neon
-
- sub r12, r12, r1, lsl #2
- add r0, r12, r1, lsr #1
-
- vst1.u8 {q4}, [r12@128],r1 ; store op2
- vst1.u8 {q5}, [r0@128],r1 ; store op1
- vst1.u8 {q6}, [r12@128], r1 ; store op0
- vst1.u8 {q7}, [r0@128],r1 ; store oq0
- vst1.u8 {q8}, [r12@128] ; store oq1
- vst1.u8 {q9}, [r0@128] ; store oq2
-
- vpop {d8-d15}
- pop {pc}
- ENDP ; |vp8_mbloop_filter_horizontal_edge_y_neon|
-
-; void vp8_mbloop_filter_horizontal_edge_uv_neon(unsigned char *u, int pitch,
-; const unsigned char *blimit,
-; const unsigned char *limit,
-; const unsigned char *thresh,
-; unsigned char *v)
-; r0 unsigned char *u,
-; r1 int pitch,
-; r2 unsigned char blimit
-; r3 unsigned char limit
-; sp unsigned char thresh,
-; sp+4 unsigned char *v
-
-|vp8_mbloop_filter_horizontal_edge_uv_neon| PROC
- push {lr}
- vpush {d8-d15}
-
- ldr r12, [sp, #68] ; load thresh
- sub r0, r0, r1, lsl #2 ; move u pointer down by 4 lines
- vdup.u8 q2, r12 ; thresh
- ldr r12, [sp, #72] ; load v ptr
- sub r12, r12, r1, lsl #2 ; move v pointer down by 4 lines
-
- vld1.u8 {d6}, [r0@64], r1 ; p3
- vld1.u8 {d7}, [r12@64], r1 ; p3
- vld1.u8 {d8}, [r0@64], r1 ; p2
- vld1.u8 {d9}, [r12@64], r1 ; p2
- vld1.u8 {d10}, [r0@64], r1 ; p1
- vld1.u8 {d11}, [r12@64], r1 ; p1
- vld1.u8 {d12}, [r0@64], r1 ; p0
- vld1.u8 {d13}, [r12@64], r1 ; p0
- vld1.u8 {d14}, [r0@64], r1 ; q0
- vld1.u8 {d15}, [r12@64], r1 ; q0
- vld1.u8 {d16}, [r0@64], r1 ; q1
- vld1.u8 {d17}, [r12@64], r1 ; q1
- vld1.u8 {d18}, [r0@64], r1 ; q2
- vld1.u8 {d19}, [r12@64], r1 ; q2
- vld1.u8 {d20}, [r0@64], r1 ; q3
- vld1.u8 {d21}, [r12@64], r1 ; q3
-
- bl vp8_mbloop_filter_neon
-
- sub r0, r0, r1, lsl #3
- sub r12, r12, r1, lsl #3
-
- add r0, r0, r1
- add r12, r12, r1
-
- vst1.u8 {d8}, [r0@64], r1 ; store u op2
- vst1.u8 {d9}, [r12@64], r1 ; store v op2
- vst1.u8 {d10}, [r0@64], r1 ; store u op1
- vst1.u8 {d11}, [r12@64], r1 ; store v op1
- vst1.u8 {d12}, [r0@64], r1 ; store u op0
- vst1.u8 {d13}, [r12@64], r1 ; store v op0
- vst1.u8 {d14}, [r0@64], r1 ; store u oq0
- vst1.u8 {d15}, [r12@64], r1 ; store v oq0
- vst1.u8 {d16}, [r0@64], r1 ; store u oq1
- vst1.u8 {d17}, [r12@64], r1 ; store v oq1
- vst1.u8 {d18}, [r0@64], r1 ; store u oq2
- vst1.u8 {d19}, [r12@64], r1 ; store v oq2
-
- vpop {d8-d15}
- pop {pc}
- ENDP ; |vp8_mbloop_filter_horizontal_edge_uv_neon|
-
-; void vp8_mbloop_filter_vertical_edge_y_neon(unsigned char *src, int pitch,
-; const unsigned char *blimit,
-; const unsigned char *limit,
-; const unsigned char *thresh)
-; r0 unsigned char *src,
-; r1 int pitch,
-; r2 unsigned char blimit
-; r3 unsigned char limit
-; sp unsigned char thresh,
-|vp8_mbloop_filter_vertical_edge_y_neon| PROC
- push {lr}
- vpush {d8-d15}
-
- ldr r12, [sp, #68] ; load thresh
- sub r0, r0, #4 ; move src pointer down by 4 columns
- vdup.s8 q2, r12 ; thresh
- add r12, r0, r1, lsl #3 ; move src pointer down by 8 lines
-
- vld1.u8 {d6}, [r0], r1 ; load first 8-line src data
- vld1.u8 {d7}, [r12], r1 ; load second 8-line src data
- vld1.u8 {d8}, [r0], r1
- vld1.u8 {d9}, [r12], r1
- vld1.u8 {d10}, [r0], r1
- vld1.u8 {d11}, [r12], r1
- vld1.u8 {d12}, [r0], r1
- vld1.u8 {d13}, [r12], r1
- vld1.u8 {d14}, [r0], r1
- vld1.u8 {d15}, [r12], r1
- vld1.u8 {d16}, [r0], r1
- vld1.u8 {d17}, [r12], r1
- vld1.u8 {d18}, [r0], r1
- vld1.u8 {d19}, [r12], r1
- vld1.u8 {d20}, [r0], r1
- vld1.u8 {d21}, [r12], r1
-
- ;transpose to 8x16 matrix
- vtrn.32 q3, q7
- vtrn.32 q4, q8
- vtrn.32 q5, q9
- vtrn.32 q6, q10
-
- vtrn.16 q3, q5
- vtrn.16 q4, q6
- vtrn.16 q7, q9
- vtrn.16 q8, q10
-
- vtrn.8 q3, q4
- vtrn.8 q5, q6
- vtrn.8 q7, q8
- vtrn.8 q9, q10
-
- sub r0, r0, r1, lsl #3
-
- bl vp8_mbloop_filter_neon
-
- sub r12, r12, r1, lsl #3
-
- ;transpose to 16x8 matrix
- vtrn.32 q3, q7
- vtrn.32 q4, q8
- vtrn.32 q5, q9
- vtrn.32 q6, q10
-
- vtrn.16 q3, q5
- vtrn.16 q4, q6
- vtrn.16 q7, q9
- vtrn.16 q8, q10
-
- vtrn.8 q3, q4
- vtrn.8 q5, q6
- vtrn.8 q7, q8
- vtrn.8 q9, q10
-
- ;store op2, op1, op0, oq0, oq1, oq2
- vst1.8 {d6}, [r0], r1
- vst1.8 {d7}, [r12], r1
- vst1.8 {d8}, [r0], r1
- vst1.8 {d9}, [r12], r1
- vst1.8 {d10}, [r0], r1
- vst1.8 {d11}, [r12], r1
- vst1.8 {d12}, [r0], r1
- vst1.8 {d13}, [r12], r1
- vst1.8 {d14}, [r0], r1
- vst1.8 {d15}, [r12], r1
- vst1.8 {d16}, [r0], r1
- vst1.8 {d17}, [r12], r1
- vst1.8 {d18}, [r0], r1
- vst1.8 {d19}, [r12], r1
- vst1.8 {d20}, [r0]
- vst1.8 {d21}, [r12]
-
- vpop {d8-d15}
- pop {pc}
- ENDP ; |vp8_mbloop_filter_vertical_edge_y_neon|
-
-; void vp8_mbloop_filter_vertical_edge_uv_neon(unsigned char *u, int pitch,
-; const unsigned char *blimit,
-; const unsigned char *limit,
-; const unsigned char *thresh,
-; unsigned char *v)
-; r0 unsigned char *u,
-; r1 int pitch,
-; r2 const signed char *flimit,
-; r3 const signed char *limit,
-; sp const signed char *thresh,
-; sp+4 unsigned char *v
-|vp8_mbloop_filter_vertical_edge_uv_neon| PROC
- push {lr}
- vpush {d8-d15}
-
- ldr r12, [sp, #68] ; load thresh
- sub r0, r0, #4 ; move u pointer down by 4 columns
- vdup.u8 q2, r12 ; thresh
- ldr r12, [sp, #72] ; load v ptr
- sub r12, r12, #4 ; move v pointer down by 4 columns
-
- vld1.u8 {d6}, [r0], r1 ;load u data
- vld1.u8 {d7}, [r12], r1 ;load v data
- vld1.u8 {d8}, [r0], r1
- vld1.u8 {d9}, [r12], r1
- vld1.u8 {d10}, [r0], r1
- vld1.u8 {d11}, [r12], r1
- vld1.u8 {d12}, [r0], r1
- vld1.u8 {d13}, [r12], r1
- vld1.u8 {d14}, [r0], r1
- vld1.u8 {d15}, [r12], r1
- vld1.u8 {d16}, [r0], r1
- vld1.u8 {d17}, [r12], r1
- vld1.u8 {d18}, [r0], r1
- vld1.u8 {d19}, [r12], r1
- vld1.u8 {d20}, [r0], r1
- vld1.u8 {d21}, [r12], r1
-
- ;transpose to 8x16 matrix
- vtrn.32 q3, q7
- vtrn.32 q4, q8
- vtrn.32 q5, q9
- vtrn.32 q6, q10
-
- vtrn.16 q3, q5
- vtrn.16 q4, q6
- vtrn.16 q7, q9
- vtrn.16 q8, q10
-
- vtrn.8 q3, q4
- vtrn.8 q5, q6
- vtrn.8 q7, q8
- vtrn.8 q9, q10
-
- sub r0, r0, r1, lsl #3
-
- bl vp8_mbloop_filter_neon
-
- sub r12, r12, r1, lsl #3
-
- ;transpose to 16x8 matrix
- vtrn.32 q3, q7
- vtrn.32 q4, q8
- vtrn.32 q5, q9
- vtrn.32 q6, q10
-
- vtrn.16 q3, q5
- vtrn.16 q4, q6
- vtrn.16 q7, q9
- vtrn.16 q8, q10
-
- vtrn.8 q3, q4
- vtrn.8 q5, q6
- vtrn.8 q7, q8
- vtrn.8 q9, q10
-
- ;store op2, op1, op0, oq0, oq1, oq2
- vst1.8 {d6}, [r0], r1
- vst1.8 {d7}, [r12], r1
- vst1.8 {d8}, [r0], r1
- vst1.8 {d9}, [r12], r1
- vst1.8 {d10}, [r0], r1
- vst1.8 {d11}, [r12], r1
- vst1.8 {d12}, [r0], r1
- vst1.8 {d13}, [r12], r1
- vst1.8 {d14}, [r0], r1
- vst1.8 {d15}, [r12], r1
- vst1.8 {d16}, [r0], r1
- vst1.8 {d17}, [r12], r1
- vst1.8 {d18}, [r0], r1
- vst1.8 {d19}, [r12], r1
- vst1.8 {d20}, [r0]
- vst1.8 {d21}, [r12]
-
- vpop {d8-d15}
- pop {pc}
- ENDP ; |vp8_mbloop_filter_vertical_edge_uv_neon|
-
-; void vp8_mbloop_filter_neon()
-; This is a helper function for the macroblock loopfilters. The individual
-; functions do the necessary load, transpose (if necessary), preserve (if
-; necessary) and store.
-
-; r0,r1 PRESERVE
-; r2 mblimit
-; r3 limit
-
-; q2 thresh
-; q3 p3 PRESERVE
-; q4 p2
-; q5 p1
-; q6 p0
-; q7 q0
-; q8 q1
-; q9 q2
-; q10 q3 PRESERVE
-
-|vp8_mbloop_filter_neon| PROC
-
- ; vp8_filter_mask
- vabd.u8 q11, q3, q4 ; abs(p3 - p2)
- vabd.u8 q12, q4, q5 ; abs(p2 - p1)
- vabd.u8 q13, q5, q6 ; abs(p1 - p0)
- vabd.u8 q14, q8, q7 ; abs(q1 - q0)
- vabd.u8 q1, q9, q8 ; abs(q2 - q1)
- vabd.u8 q0, q10, q9 ; abs(q3 - q2)
-
- vmax.u8 q11, q11, q12
- vmax.u8 q12, q13, q14
- vmax.u8 q1, q1, q0
- vmax.u8 q15, q11, q12
-
- vabd.u8 q12, q6, q7 ; abs(p0 - q0)
-
- ; vp8_hevmask
- vcgt.u8 q13, q13, q2 ; (abs(p1 - p0) > thresh) * -1
- vcgt.u8 q14, q14, q2 ; (abs(q1 - q0) > thresh) * -1
- vmax.u8 q15, q15, q1
-
- vdup.u8 q1, r3 ; limit
- vdup.u8 q2, r2 ; mblimit
-
- vmov.u8 q0, #0x80 ; 0x80
-
- vcge.u8 q15, q1, q15
-
- vabd.u8 q1, q5, q8 ; a = abs(p1 - q1)
- vqadd.u8 q12, q12, q12 ; b = abs(p0 - q0) * 2
- vmov.u16 q11, #3 ; #3
-
- ; vp8_filter
- ; convert to signed
- veor q7, q7, q0 ; qs0
- vshr.u8 q1, q1, #1 ; a = a / 2
- veor q6, q6, q0 ; ps0
- veor q5, q5, q0 ; ps1
-
- vqadd.u8 q12, q12, q1 ; a = b + a
-
- veor q8, q8, q0 ; qs1
- veor q4, q4, q0 ; ps2
- veor q9, q9, q0 ; qs2
-
- vorr q14, q13, q14 ; vp8_hevmask
-
- vcge.u8 q12, q2, q12 ; (a > flimit * 2 + limit) * -1
-
- vsubl.s8 q2, d14, d12 ; qs0 - ps0
- vsubl.s8 q13, d15, d13
-
- vqsub.s8 q1, q5, q8 ; vp8_filter = clamp(ps1-qs1)
-
- vmul.i16 q2, q2, q11 ; 3 * ( qs0 - ps0)
-
- vand q15, q15, q12 ; vp8_filter_mask
-
- vmul.i16 q13, q13, q11
-
- vmov.u8 q12, #3 ; #3
-
- vaddw.s8 q2, q2, d2 ; vp8_filter + 3 * ( qs0 - ps0)
- vaddw.s8 q13, q13, d3
-
- vmov.u8 q11, #4 ; #4
-
- ; vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0))
- vqmovn.s16 d2, q2
- vqmovn.s16 d3, q13
-
- vand q1, q1, q15 ; vp8_filter &= mask
-
- vmov.u16 q15, #63 ; #63
-
- vand q13, q1, q14 ; Filter2 &= hev
-
- vqadd.s8 q2, q13, q11 ; Filter1 = clamp(Filter2+4)
- vqadd.s8 q13, q13, q12 ; Filter2 = clamp(Filter2+3)
-
- vmov q0, q15
-
- vshr.s8 q2, q2, #3 ; Filter1 >>= 3
- vshr.s8 q13, q13, #3 ; Filter2 >>= 3
-
- vmov q11, q15
- vmov q12, q15
-
- vqsub.s8 q7, q7, q2 ; qs0 = clamp(qs0 - Filter1)
-
- vqadd.s8 q6, q6, q13 ; ps0 = clamp(ps0 + Filter2)
-
- vbic q1, q1, q14 ; vp8_filter &= ~hev
-
- ; roughly 1/7th difference across boundary
- ; roughly 2/7th difference across boundary
- ; roughly 3/7th difference across boundary
-
- vmov.u8 d5, #9 ; #9
- vmov.u8 d4, #18 ; #18
-
- vmov q13, q15
- vmov q14, q15
-
- vmlal.s8 q0, d2, d5 ; 63 + Filter2 * 9
- vmlal.s8 q11, d3, d5
- vmov.u8 d5, #27 ; #27
- vmlal.s8 q12, d2, d4 ; 63 + Filter2 * 18
- vmlal.s8 q13, d3, d4
- vmlal.s8 q14, d2, d5 ; 63 + Filter2 * 27
- vmlal.s8 q15, d3, d5
-
- vqshrn.s16 d0, q0, #7 ; u = clamp((63 + Filter2 * 9)>>7)
- vqshrn.s16 d1, q11, #7
- vqshrn.s16 d24, q12, #7 ; u = clamp((63 + Filter2 * 18)>>7)
- vqshrn.s16 d25, q13, #7
- vqshrn.s16 d28, q14, #7 ; u = clamp((63 + Filter2 * 27)>>7)
- vqshrn.s16 d29, q15, #7
-
- vmov.u8 q1, #0x80 ; 0x80
-
- vqsub.s8 q11, q9, q0 ; s = clamp(qs2 - u)
- vqadd.s8 q0, q4, q0 ; s = clamp(ps2 + u)
- vqsub.s8 q13, q8, q12 ; s = clamp(qs1 - u)
- vqadd.s8 q12, q5, q12 ; s = clamp(ps1 + u)
- vqsub.s8 q15, q7, q14 ; s = clamp(qs0 - u)
- vqadd.s8 q14, q6, q14 ; s = clamp(ps0 + u)
-
- veor q9, q11, q1 ; *oq2 = s^0x80
- veor q4, q0, q1 ; *op2 = s^0x80
- veor q8, q13, q1 ; *oq1 = s^0x80
- veor q5, q12, q1 ; *op2 = s^0x80
- veor q7, q15, q1 ; *oq0 = s^0x80
- veor q6, q14, q1 ; *op0 = s^0x80
-
- bx lr
- ENDP ; |vp8_mbloop_filter_neon|
-
-;-----------------
-
- END
diff --git a/vp8/common/arm/neon/mbloopfilter_neon.c b/vp8/common/arm/neon/mbloopfilter_neon.c
new file mode 100644
index 0000000..385ce36
--- /dev/null
+++ b/vp8/common/arm/neon/mbloopfilter_neon.c
@@ -0,0 +1,592 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+static inline void vp8_mbloop_filter_neon(
+ uint8x16_t qblimit, // mblimit
+ uint8x16_t qlimit, // limit
+ uint8x16_t qthresh, // thresh
+ uint8x16_t q3, // p2
+ uint8x16_t q4, // p2
+ uint8x16_t q5, // p1
+ uint8x16_t q6, // p0
+ uint8x16_t q7, // q0
+ uint8x16_t q8, // q1
+ uint8x16_t q9, // q2
+ uint8x16_t q10, // q3
+ uint8x16_t *q4r, // p1
+ uint8x16_t *q5r, // p1
+ uint8x16_t *q6r, // p0
+ uint8x16_t *q7r, // q0
+ uint8x16_t *q8r, // q1
+ uint8x16_t *q9r) { // q1
+ uint8x16_t q0u8, q1u8, q11u8, q12u8, q13u8, q14u8, q15u8;
+ int16x8_t q0s16, q2s16, q11s16, q12s16, q13s16, q14s16, q15s16;
+ int8x16_t q1s8, q6s8, q7s8, q2s8, q11s8, q13s8;
+ uint16x8_t q0u16, q11u16, q12u16, q13u16, q14u16, q15u16;
+ int8x16_t q0s8, q12s8, q14s8, q15s8;
+ int8x8_t d0, d1, d2, d3, d4, d5, d24, d25, d28, d29;
+
+ q11u8 = vabdq_u8(q3, q4);
+ q12u8 = vabdq_u8(q4, q5);
+ q13u8 = vabdq_u8(q5, q6);
+ q14u8 = vabdq_u8(q8, q7);
+ q1u8 = vabdq_u8(q9, q8);
+ q0u8 = vabdq_u8(q10, q9);
+
+ q11u8 = vmaxq_u8(q11u8, q12u8);
+ q12u8 = vmaxq_u8(q13u8, q14u8);
+ q1u8 = vmaxq_u8(q1u8, q0u8);
+ q15u8 = vmaxq_u8(q11u8, q12u8);
+
+ q12u8 = vabdq_u8(q6, q7);
+
+ // vp8_hevmask
+ q13u8 = vcgtq_u8(q13u8, qthresh);
+ q14u8 = vcgtq_u8(q14u8, qthresh);
+ q15u8 = vmaxq_u8(q15u8, q1u8);
+
+ q15u8 = vcgeq_u8(qlimit, q15u8);
+
+ q1u8 = vabdq_u8(q5, q8);
+ q12u8 = vqaddq_u8(q12u8, q12u8);
+
+ // vp8_filter() function
+ // convert to signed
+ q0u8 = vdupq_n_u8(0x80);
+ q9 = veorq_u8(q9, q0u8);
+ q8 = veorq_u8(q8, q0u8);
+ q7 = veorq_u8(q7, q0u8);
+ q6 = veorq_u8(q6, q0u8);
+ q5 = veorq_u8(q5, q0u8);
+ q4 = veorq_u8(q4, q0u8);
+
+ q1u8 = vshrq_n_u8(q1u8, 1);
+ q12u8 = vqaddq_u8(q12u8, q1u8);
+
+ q14u8 = vorrq_u8(q13u8, q14u8);
+ q12u8 = vcgeq_u8(qblimit, q12u8);
+
+ q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q7)),
+ vget_low_s8(vreinterpretq_s8_u8(q6)));
+ q13s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q7)),
+ vget_high_s8(vreinterpretq_s8_u8(q6)));
+
+ q1s8 = vqsubq_s8(vreinterpretq_s8_u8(q5),
+ vreinterpretq_s8_u8(q8));
+
+ q11s16 = vdupq_n_s16(3);
+ q2s16 = vmulq_s16(q2s16, q11s16);
+ q13s16 = vmulq_s16(q13s16, q11s16);
+
+ q15u8 = vandq_u8(q15u8, q12u8);
+
+ q2s16 = vaddw_s8(q2s16, vget_low_s8(q1s8));
+ q13s16 = vaddw_s8(q13s16, vget_high_s8(q1s8));
+
+ q12u8 = vdupq_n_u8(3);
+ q11u8 = vdupq_n_u8(4);
+ // vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0))
+ d2 = vqmovn_s16(q2s16);
+ d3 = vqmovn_s16(q13s16);
+ q1s8 = vcombine_s8(d2, d3);
+ q1s8 = vandq_s8(q1s8, vreinterpretq_s8_u8(q15u8));
+ q13s8 = vandq_s8(q1s8, vreinterpretq_s8_u8(q14u8));
+
+ q2s8 = vqaddq_s8(q13s8, vreinterpretq_s8_u8(q11u8));
+ q13s8 = vqaddq_s8(q13s8, vreinterpretq_s8_u8(q12u8));
+ q2s8 = vshrq_n_s8(q2s8, 3);
+ q13s8 = vshrq_n_s8(q13s8, 3);
+
+ q7s8 = vqsubq_s8(vreinterpretq_s8_u8(q7), q2s8);
+ q6s8 = vqaddq_s8(vreinterpretq_s8_u8(q6), q13s8);
+
+ q1s8 = vbicq_s8(q1s8, vreinterpretq_s8_u8(q14u8));
+
+ q0u16 = q11u16 = q12u16 = q13u16 = q14u16 = q15u16 = vdupq_n_u16(63);
+ d5 = vdup_n_s8(9);
+ d4 = vdup_n_s8(18);
+
+ q0s16 = vmlal_s8(vreinterpretq_s16_u16(q0u16), vget_low_s8(q1s8), d5);
+ q11s16 = vmlal_s8(vreinterpretq_s16_u16(q11u16), vget_high_s8(q1s8), d5);
+ d5 = vdup_n_s8(27);
+ q12s16 = vmlal_s8(vreinterpretq_s16_u16(q12u16), vget_low_s8(q1s8), d4);
+ q13s16 = vmlal_s8(vreinterpretq_s16_u16(q13u16), vget_high_s8(q1s8), d4);
+ q14s16 = vmlal_s8(vreinterpretq_s16_u16(q14u16), vget_low_s8(q1s8), d5);
+ q15s16 = vmlal_s8(vreinterpretq_s16_u16(q15u16), vget_high_s8(q1s8), d5);
+
+ d0 = vqshrn_n_s16(q0s16 , 7);
+ d1 = vqshrn_n_s16(q11s16, 7);
+ d24 = vqshrn_n_s16(q12s16, 7);
+ d25 = vqshrn_n_s16(q13s16, 7);
+ d28 = vqshrn_n_s16(q14s16, 7);
+ d29 = vqshrn_n_s16(q15s16, 7);
+
+ q0s8 = vcombine_s8(d0, d1);
+ q12s8 = vcombine_s8(d24, d25);
+ q14s8 = vcombine_s8(d28, d29);
+
+ q11s8 = vqsubq_s8(vreinterpretq_s8_u8(q9), q0s8);
+ q0s8 = vqaddq_s8(vreinterpretq_s8_u8(q4), q0s8);
+ q13s8 = vqsubq_s8(vreinterpretq_s8_u8(q8), q12s8);
+ q12s8 = vqaddq_s8(vreinterpretq_s8_u8(q5), q12s8);
+ q15s8 = vqsubq_s8((q7s8), q14s8);
+ q14s8 = vqaddq_s8((q6s8), q14s8);
+
+ q1u8 = vdupq_n_u8(0x80);
+ *q9r = veorq_u8(vreinterpretq_u8_s8(q11s8), q1u8);
+ *q8r = veorq_u8(vreinterpretq_u8_s8(q13s8), q1u8);
+ *q7r = veorq_u8(vreinterpretq_u8_s8(q15s8), q1u8);
+ *q6r = veorq_u8(vreinterpretq_u8_s8(q14s8), q1u8);
+ *q5r = veorq_u8(vreinterpretq_u8_s8(q12s8), q1u8);
+ *q4r = veorq_u8(vreinterpretq_u8_s8(q0s8), q1u8);
+ return;
+}
+
+void vp8_mbloop_filter_horizontal_edge_y_neon(
+ unsigned char *src,
+ int pitch,
+ unsigned char blimit,
+ unsigned char limit,
+ unsigned char thresh) {
+ uint8x16_t qblimit, qlimit, qthresh, q3, q4;
+ uint8x16_t q5, q6, q7, q8, q9, q10;
+
+ qblimit = vdupq_n_u8(blimit);
+ qlimit = vdupq_n_u8(limit);
+ qthresh = vdupq_n_u8(thresh);
+
+ src -= (pitch << 2);
+
+ q3 = vld1q_u8(src);
+ src += pitch;
+ q4 = vld1q_u8(src);
+ src += pitch;
+ q5 = vld1q_u8(src);
+ src += pitch;
+ q6 = vld1q_u8(src);
+ src += pitch;
+ q7 = vld1q_u8(src);
+ src += pitch;
+ q8 = vld1q_u8(src);
+ src += pitch;
+ q9 = vld1q_u8(src);
+ src += pitch;
+ q10 = vld1q_u8(src);
+
+ vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4,
+ q5, q6, q7, q8, q9, q10,
+ &q4, &q5, &q6, &q7, &q8, &q9);
+
+ src -= (pitch * 6);
+ vst1q_u8(src, q4);
+ src += pitch;
+ vst1q_u8(src, q5);
+ src += pitch;
+ vst1q_u8(src, q6);
+ src += pitch;
+ vst1q_u8(src, q7);
+ src += pitch;
+ vst1q_u8(src, q8);
+ src += pitch;
+ vst1q_u8(src, q9);
+ return;
+}
+
+void vp8_mbloop_filter_horizontal_edge_uv_neon(
+ unsigned char *u,
+ int pitch,
+ unsigned char blimit,
+ unsigned char limit,
+ unsigned char thresh,
+ unsigned char *v) {
+ uint8x16_t qblimit, qlimit, qthresh, q3, q4;
+ uint8x16_t q5, q6, q7, q8, q9, q10;
+ uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14;
+ uint8x8_t d15, d16, d17, d18, d19, d20, d21;
+
+ qblimit = vdupq_n_u8(blimit);
+ qlimit = vdupq_n_u8(limit);
+ qthresh = vdupq_n_u8(thresh);
+
+ u -= (pitch << 2);
+ v -= (pitch << 2);
+
+ d6 = vld1_u8(u);
+ u += pitch;
+ d7 = vld1_u8(v);
+ v += pitch;
+ d8 = vld1_u8(u);
+ u += pitch;
+ d9 = vld1_u8(v);
+ v += pitch;
+ d10 = vld1_u8(u);
+ u += pitch;
+ d11 = vld1_u8(v);
+ v += pitch;
+ d12 = vld1_u8(u);
+ u += pitch;
+ d13 = vld1_u8(v);
+ v += pitch;
+ d14 = vld1_u8(u);
+ u += pitch;
+ d15 = vld1_u8(v);
+ v += pitch;
+ d16 = vld1_u8(u);
+ u += pitch;
+ d17 = vld1_u8(v);
+ v += pitch;
+ d18 = vld1_u8(u);
+ u += pitch;
+ d19 = vld1_u8(v);
+ v += pitch;
+ d20 = vld1_u8(u);
+ d21 = vld1_u8(v);
+
+ q3 = vcombine_u8(d6, d7);
+ q4 = vcombine_u8(d8, d9);
+ q5 = vcombine_u8(d10, d11);
+ q6 = vcombine_u8(d12, d13);
+ q7 = vcombine_u8(d14, d15);
+ q8 = vcombine_u8(d16, d17);
+ q9 = vcombine_u8(d18, d19);
+ q10 = vcombine_u8(d20, d21);
+
+ vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4,
+ q5, q6, q7, q8, q9, q10,
+ &q4, &q5, &q6, &q7, &q8, &q9);
+
+ u -= (pitch * 6);
+ v -= (pitch * 6);
+ vst1_u8(u, vget_low_u8(q4));
+ u += pitch;
+ vst1_u8(v, vget_high_u8(q4));
+ v += pitch;
+ vst1_u8(u, vget_low_u8(q5));
+ u += pitch;
+ vst1_u8(v, vget_high_u8(q5));
+ v += pitch;
+ vst1_u8(u, vget_low_u8(q6));
+ u += pitch;
+ vst1_u8(v, vget_high_u8(q6));
+ v += pitch;
+ vst1_u8(u, vget_low_u8(q7));
+ u += pitch;
+ vst1_u8(v, vget_high_u8(q7));
+ v += pitch;
+ vst1_u8(u, vget_low_u8(q8));
+ u += pitch;
+ vst1_u8(v, vget_high_u8(q8));
+ v += pitch;
+ vst1_u8(u, vget_low_u8(q9));
+ vst1_u8(v, vget_high_u8(q9));
+ return;
+}
+
+void vp8_mbloop_filter_vertical_edge_y_neon(
+ unsigned char *src,
+ int pitch,
+ unsigned char blimit,
+ unsigned char limit,
+ unsigned char thresh) {
+ unsigned char *s1, *s2;
+ uint8x16_t qblimit, qlimit, qthresh, q3, q4;
+ uint8x16_t q5, q6, q7, q8, q9, q10;
+ uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14;
+ uint8x8_t d15, d16, d17, d18, d19, d20, d21;
+ uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3;
+ uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7;
+ uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11;
+
+ qblimit = vdupq_n_u8(blimit);
+ qlimit = vdupq_n_u8(limit);
+ qthresh = vdupq_n_u8(thresh);
+
+ s1 = src - 4;
+ s2 = s1 + 8 * pitch;
+ d6 = vld1_u8(s1);
+ s1 += pitch;
+ d7 = vld1_u8(s2);
+ s2 += pitch;
+ d8 = vld1_u8(s1);
+ s1 += pitch;
+ d9 = vld1_u8(s2);
+ s2 += pitch;
+ d10 = vld1_u8(s1);
+ s1 += pitch;
+ d11 = vld1_u8(s2);
+ s2 += pitch;
+ d12 = vld1_u8(s1);
+ s1 += pitch;
+ d13 = vld1_u8(s2);
+ s2 += pitch;
+ d14 = vld1_u8(s1);
+ s1 += pitch;
+ d15 = vld1_u8(s2);
+ s2 += pitch;
+ d16 = vld1_u8(s1);
+ s1 += pitch;
+ d17 = vld1_u8(s2);
+ s2 += pitch;
+ d18 = vld1_u8(s1);
+ s1 += pitch;
+ d19 = vld1_u8(s2);
+ s2 += pitch;
+ d20 = vld1_u8(s1);
+ d21 = vld1_u8(s2);
+
+ q3 = vcombine_u8(d6, d7);
+ q4 = vcombine_u8(d8, d9);
+ q5 = vcombine_u8(d10, d11);
+ q6 = vcombine_u8(d12, d13);
+ q7 = vcombine_u8(d14, d15);
+ q8 = vcombine_u8(d16, d17);
+ q9 = vcombine_u8(d18, d19);
+ q10 = vcombine_u8(d20, d21);
+
+ q2tmp0 = vtrnq_u32((uint32x4_t)q3, (uint32x4_t)q7);
+ q2tmp1 = vtrnq_u32((uint32x4_t)q4, (uint32x4_t)q8);
+ q2tmp2 = vtrnq_u32((uint32x4_t)q5, (uint32x4_t)q9);
+ q2tmp3 = vtrnq_u32((uint32x4_t)q6, (uint32x4_t)q10);
+
+ q2tmp4 = vtrnq_u16((uint16x8_t)q2tmp0.val[0], (uint16x8_t)q2tmp2.val[0]);
+ q2tmp5 = vtrnq_u16((uint16x8_t)q2tmp1.val[0], (uint16x8_t)q2tmp3.val[0]);
+ q2tmp6 = vtrnq_u16((uint16x8_t)q2tmp0.val[1], (uint16x8_t)q2tmp2.val[1]);
+ q2tmp7 = vtrnq_u16((uint16x8_t)q2tmp1.val[1], (uint16x8_t)q2tmp3.val[1]);
+
+ q2tmp8 = vtrnq_u8((uint8x16_t)q2tmp4.val[0], (uint8x16_t)q2tmp5.val[0]);
+ q2tmp9 = vtrnq_u8((uint8x16_t)q2tmp4.val[1], (uint8x16_t)q2tmp5.val[1]);
+ q2tmp10 = vtrnq_u8((uint8x16_t)q2tmp6.val[0], (uint8x16_t)q2tmp7.val[0]);
+ q2tmp11 = vtrnq_u8((uint8x16_t)q2tmp6.val[1], (uint8x16_t)q2tmp7.val[1]);
+
+ q3 = q2tmp8.val[0];
+ q4 = q2tmp8.val[1];
+ q5 = q2tmp9.val[0];
+ q6 = q2tmp9.val[1];
+ q7 = q2tmp10.val[0];
+ q8 = q2tmp10.val[1];
+ q9 = q2tmp11.val[0];
+ q10 = q2tmp11.val[1];
+
+ vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4,
+ q5, q6, q7, q8, q9, q10,
+ &q4, &q5, &q6, &q7, &q8, &q9);
+
+ q2tmp0 = vtrnq_u32((uint32x4_t)q3, (uint32x4_t)q7);
+ q2tmp1 = vtrnq_u32((uint32x4_t)q4, (uint32x4_t)q8);
+ q2tmp2 = vtrnq_u32((uint32x4_t)q5, (uint32x4_t)q9);
+ q2tmp3 = vtrnq_u32((uint32x4_t)q6, (uint32x4_t)q10);
+
+ q2tmp4 = vtrnq_u16((uint16x8_t)q2tmp0.val[0], (uint16x8_t)q2tmp2.val[0]);
+ q2tmp5 = vtrnq_u16((uint16x8_t)q2tmp1.val[0], (uint16x8_t)q2tmp3.val[0]);
+ q2tmp6 = vtrnq_u16((uint16x8_t)q2tmp0.val[1], (uint16x8_t)q2tmp2.val[1]);
+ q2tmp7 = vtrnq_u16((uint16x8_t)q2tmp1.val[1], (uint16x8_t)q2tmp3.val[1]);
+
+ q2tmp8 = vtrnq_u8((uint8x16_t)q2tmp4.val[0], (uint8x16_t)q2tmp5.val[0]);
+ q2tmp9 = vtrnq_u8((uint8x16_t)q2tmp4.val[1], (uint8x16_t)q2tmp5.val[1]);
+ q2tmp10 = vtrnq_u8((uint8x16_t)q2tmp6.val[0], (uint8x16_t)q2tmp7.val[0]);
+ q2tmp11 = vtrnq_u8((uint8x16_t)q2tmp6.val[1], (uint8x16_t)q2tmp7.val[1]);
+
+ q3 = q2tmp8.val[0];
+ q4 = q2tmp8.val[1];
+ q5 = q2tmp9.val[0];
+ q6 = q2tmp9.val[1];
+ q7 = q2tmp10.val[0];
+ q8 = q2tmp10.val[1];
+ q9 = q2tmp11.val[0];
+ q10 = q2tmp11.val[1];
+
+ s1 -= 7 * pitch;
+ s2 -= 7 * pitch;
+
+ vst1_u8(s1, vget_low_u8(q3));
+ s1 += pitch;
+ vst1_u8(s2, vget_high_u8(q3));
+ s2 += pitch;
+ vst1_u8(s1, vget_low_u8(q4));
+ s1 += pitch;
+ vst1_u8(s2, vget_high_u8(q4));
+ s2 += pitch;
+ vst1_u8(s1, vget_low_u8(q5));
+ s1 += pitch;
+ vst1_u8(s2, vget_high_u8(q5));
+ s2 += pitch;
+ vst1_u8(s1, vget_low_u8(q6));
+ s1 += pitch;
+ vst1_u8(s2, vget_high_u8(q6));
+ s2 += pitch;
+ vst1_u8(s1, vget_low_u8(q7));
+ s1 += pitch;
+ vst1_u8(s2, vget_high_u8(q7));
+ s2 += pitch;
+ vst1_u8(s1, vget_low_u8(q8));
+ s1 += pitch;
+ vst1_u8(s2, vget_high_u8(q8));
+ s2 += pitch;
+ vst1_u8(s1, vget_low_u8(q9));
+ s1 += pitch;
+ vst1_u8(s2, vget_high_u8(q9));
+ s2 += pitch;
+ vst1_u8(s1, vget_low_u8(q10));
+ vst1_u8(s2, vget_high_u8(q10));
+ return;
+}
+
+void vp8_mbloop_filter_vertical_edge_uv_neon(
+ unsigned char *u,
+ int pitch,
+ unsigned char blimit,
+ unsigned char limit,
+ unsigned char thresh,
+ unsigned char *v) {
+ unsigned char *us, *ud;
+ unsigned char *vs, *vd;
+ uint8x16_t qblimit, qlimit, qthresh, q3, q4;
+ uint8x16_t q5, q6, q7, q8, q9, q10;
+ uint8x8_t d6, d7, d8, d9, d10, d11, d12, d13, d14;
+ uint8x8_t d15, d16, d17, d18, d19, d20, d21;
+ uint32x4x2_t q2tmp0, q2tmp1, q2tmp2, q2tmp3;
+ uint16x8x2_t q2tmp4, q2tmp5, q2tmp6, q2tmp7;
+ uint8x16x2_t q2tmp8, q2tmp9, q2tmp10, q2tmp11;
+
+ qblimit = vdupq_n_u8(blimit);
+ qlimit = vdupq_n_u8(limit);
+ qthresh = vdupq_n_u8(thresh);
+
+ us = u - 4;
+ vs = v - 4;
+ d6 = vld1_u8(us);
+ us += pitch;
+ d7 = vld1_u8(vs);
+ vs += pitch;
+ d8 = vld1_u8(us);
+ us += pitch;
+ d9 = vld1_u8(vs);
+ vs += pitch;
+ d10 = vld1_u8(us);
+ us += pitch;
+ d11 = vld1_u8(vs);
+ vs += pitch;
+ d12 = vld1_u8(us);
+ us += pitch;
+ d13 = vld1_u8(vs);
+ vs += pitch;
+ d14 = vld1_u8(us);
+ us += pitch;
+ d15 = vld1_u8(vs);
+ vs += pitch;
+ d16 = vld1_u8(us);
+ us += pitch;
+ d17 = vld1_u8(vs);
+ vs += pitch;
+ d18 = vld1_u8(us);
+ us += pitch;
+ d19 = vld1_u8(vs);
+ vs += pitch;
+ d20 = vld1_u8(us);
+ d21 = vld1_u8(vs);
+
+ q3 = vcombine_u8(d6, d7);
+ q4 = vcombine_u8(d8, d9);
+ q5 = vcombine_u8(d10, d11);
+ q6 = vcombine_u8(d12, d13);
+ q7 = vcombine_u8(d14, d15);
+ q8 = vcombine_u8(d16, d17);
+ q9 = vcombine_u8(d18, d19);
+ q10 = vcombine_u8(d20, d21);
+
+ q2tmp0 = vtrnq_u32((uint32x4_t)q3, (uint32x4_t)q7);
+ q2tmp1 = vtrnq_u32((uint32x4_t)q4, (uint32x4_t)q8);
+ q2tmp2 = vtrnq_u32((uint32x4_t)q5, (uint32x4_t)q9);
+ q2tmp3 = vtrnq_u32((uint32x4_t)q6, (uint32x4_t)q10);
+
+ q2tmp4 = vtrnq_u16((uint16x8_t)q2tmp0.val[0], (uint16x8_t)q2tmp2.val[0]);
+ q2tmp5 = vtrnq_u16((uint16x8_t)q2tmp1.val[0], (uint16x8_t)q2tmp3.val[0]);
+ q2tmp6 = vtrnq_u16((uint16x8_t)q2tmp0.val[1], (uint16x8_t)q2tmp2.val[1]);
+ q2tmp7 = vtrnq_u16((uint16x8_t)q2tmp1.val[1], (uint16x8_t)q2tmp3.val[1]);
+
+ q2tmp8 = vtrnq_u8((uint8x16_t)q2tmp4.val[0], (uint8x16_t)q2tmp5.val[0]);
+ q2tmp9 = vtrnq_u8((uint8x16_t)q2tmp4.val[1], (uint8x16_t)q2tmp5.val[1]);
+ q2tmp10 = vtrnq_u8((uint8x16_t)q2tmp6.val[0], (uint8x16_t)q2tmp7.val[0]);
+ q2tmp11 = vtrnq_u8((uint8x16_t)q2tmp6.val[1], (uint8x16_t)q2tmp7.val[1]);
+
+ q3 = q2tmp8.val[0];
+ q4 = q2tmp8.val[1];
+ q5 = q2tmp9.val[0];
+ q6 = q2tmp9.val[1];
+ q7 = q2tmp10.val[0];
+ q8 = q2tmp10.val[1];
+ q9 = q2tmp11.val[0];
+ q10 = q2tmp11.val[1];
+
+ vp8_mbloop_filter_neon(qblimit, qlimit, qthresh, q3, q4,
+ q5, q6, q7, q8, q9, q10,
+ &q4, &q5, &q6, &q7, &q8, &q9);
+
+ q2tmp0 = vtrnq_u32((uint32x4_t)q3, (uint32x4_t)q7);
+ q2tmp1 = vtrnq_u32((uint32x4_t)q4, (uint32x4_t)q8);
+ q2tmp2 = vtrnq_u32((uint32x4_t)q5, (uint32x4_t)q9);
+ q2tmp3 = vtrnq_u32((uint32x4_t)q6, (uint32x4_t)q10);
+
+ q2tmp4 = vtrnq_u16((uint16x8_t)q2tmp0.val[0], (uint16x8_t)q2tmp2.val[0]);
+ q2tmp5 = vtrnq_u16((uint16x8_t)q2tmp1.val[0], (uint16x8_t)q2tmp3.val[0]);
+ q2tmp6 = vtrnq_u16((uint16x8_t)q2tmp0.val[1], (uint16x8_t)q2tmp2.val[1]);
+ q2tmp7 = vtrnq_u16((uint16x8_t)q2tmp1.val[1], (uint16x8_t)q2tmp3.val[1]);
+
+ q2tmp8 = vtrnq_u8((uint8x16_t)q2tmp4.val[0], (uint8x16_t)q2tmp5.val[0]);
+ q2tmp9 = vtrnq_u8((uint8x16_t)q2tmp4.val[1], (uint8x16_t)q2tmp5.val[1]);
+ q2tmp10 = vtrnq_u8((uint8x16_t)q2tmp6.val[0], (uint8x16_t)q2tmp7.val[0]);
+ q2tmp11 = vtrnq_u8((uint8x16_t)q2tmp6.val[1], (uint8x16_t)q2tmp7.val[1]);
+
+ q3 = q2tmp8.val[0];
+ q4 = q2tmp8.val[1];
+ q5 = q2tmp9.val[0];
+ q6 = q2tmp9.val[1];
+ q7 = q2tmp10.val[0];
+ q8 = q2tmp10.val[1];
+ q9 = q2tmp11.val[0];
+ q10 = q2tmp11.val[1];
+
+ ud = u - 4;
+ vst1_u8(ud, vget_low_u8(q3));
+ ud += pitch;
+ vst1_u8(ud, vget_low_u8(q4));
+ ud += pitch;
+ vst1_u8(ud, vget_low_u8(q5));
+ ud += pitch;
+ vst1_u8(ud, vget_low_u8(q6));
+ ud += pitch;
+ vst1_u8(ud, vget_low_u8(q7));
+ ud += pitch;
+ vst1_u8(ud, vget_low_u8(q8));
+ ud += pitch;
+ vst1_u8(ud, vget_low_u8(q9));
+ ud += pitch;
+ vst1_u8(ud, vget_low_u8(q10));
+
+ vd = v - 4;
+ vst1_u8(vd, vget_high_u8(q3));
+ vd += pitch;
+ vst1_u8(vd, vget_high_u8(q4));
+ vd += pitch;
+ vst1_u8(vd, vget_high_u8(q5));
+ vd += pitch;
+ vst1_u8(vd, vget_high_u8(q6));
+ vd += pitch;
+ vst1_u8(vd, vget_high_u8(q7));
+ vd += pitch;
+ vst1_u8(vd, vget_high_u8(q8));
+ vd += pitch;
+ vst1_u8(vd, vget_high_u8(q9));
+ vd += pitch;
+ vst1_u8(vd, vget_high_u8(q10));
+ return;
+}
diff --git a/vp8/common/arm/neon/sad16_neon.asm b/vp8/common/arm/neon/sad16_neon.asm
deleted file mode 100644
index 7197e56..0000000
--- a/vp8/common/arm/neon/sad16_neon.asm
+++ /dev/null
@@ -1,212 +0,0 @@
-;
-; Copyright (c) 2010 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 |vp8_sad16x16_neon|
- EXPORT |vp8_sad16x8_neon|
-
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
-; r0 unsigned char *src_ptr
-; r1 int src_stride
-; r2 unsigned char *ref_ptr
-; r3 int ref_stride
-|vp8_sad16x16_neon| PROC
-;;
- vpush {d8-d15}
- vld1.8 {q0}, [r0], r1
- vld1.8 {q4}, [r2], r3
-
- vld1.8 {q1}, [r0], r1
- vld1.8 {q5}, [r2], r3
-
- vabdl.u8 q12, d0, d8
- vabdl.u8 q13, d1, d9
-
- vld1.8 {q2}, [r0], r1
- vld1.8 {q6}, [r2], r3
-
- vabal.u8 q12, d2, d10
- vabal.u8 q13, d3, d11
-
- vld1.8 {q3}, [r0], r1
- vld1.8 {q7}, [r2], r3
-
- vabal.u8 q12, d4, d12
- vabal.u8 q13, d5, d13
-
-;;
- vld1.8 {q0}, [r0], r1
- vld1.8 {q4}, [r2], r3
-
- vabal.u8 q12, d6, d14
- vabal.u8 q13, d7, d15
-
- vld1.8 {q1}, [r0], r1
- vld1.8 {q5}, [r2], r3
-
- vabal.u8 q12, d0, d8
- vabal.u8 q13, d1, d9
-
- vld1.8 {q2}, [r0], r1
- vld1.8 {q6}, [r2], r3
-
- vabal.u8 q12, d2, d10
- vabal.u8 q13, d3, d11
-
- vld1.8 {q3}, [r0], r1
- vld1.8 {q7}, [r2], r3
-
- vabal.u8 q12, d4, d12
- vabal.u8 q13, d5, d13
-
-;;
- vld1.8 {q0}, [r0], r1
- vld1.8 {q4}, [r2], r3
-
- vabal.u8 q12, d6, d14
- vabal.u8 q13, d7, d15
-
- vld1.8 {q1}, [r0], r1
- vld1.8 {q5}, [r2], r3
-
- vabal.u8 q12, d0, d8
- vabal.u8 q13, d1, d9
-
- vld1.8 {q2}, [r0], r1
- vld1.8 {q6}, [r2], r3
-
- vabal.u8 q12, d2, d10
- vabal.u8 q13, d3, d11
-
- vld1.8 {q3}, [r0], r1
- vld1.8 {q7}, [r2], r3
-
- vabal.u8 q12, d4, d12
- vabal.u8 q13, d5, d13
-
-;;
- vld1.8 {q0}, [r0], r1
- vld1.8 {q4}, [r2], r3
-
- vabal.u8 q12, d6, d14
- vabal.u8 q13, d7, d15
-
- vld1.8 {q1}, [r0], r1
- vld1.8 {q5}, [r2], r3
-
- vabal.u8 q12, d0, d8
- vabal.u8 q13, d1, d9
-
- vld1.8 {q2}, [r0], r1
- vld1.8 {q6}, [r2], r3
-
- vabal.u8 q12, d2, d10
- vabal.u8 q13, d3, d11
-
- vld1.8 {q3}, [r0]
- vld1.8 {q7}, [r2]
-
- vabal.u8 q12, d4, d12
- vabal.u8 q13, d5, d13
-
- vabal.u8 q12, d6, d14
- vabal.u8 q13, d7, d15
-
- vadd.u16 q0, q12, q13
-
- vpaddl.u16 q1, q0
- vpaddl.u32 q0, q1
-
- vadd.u32 d0, d0, d1
-
- vmov.32 r0, d0[0]
-
- vpop {d8-d15}
- bx lr
-
- ENDP
-
-;==============================
-;unsigned int vp8_sad16x8_c(
-; unsigned char *src_ptr,
-; int src_stride,
-; unsigned char *ref_ptr,
-; int ref_stride)
-|vp8_sad16x8_neon| PROC
- vpush {d8-d15}
-
- vld1.8 {q0}, [r0], r1
- vld1.8 {q4}, [r2], r3
-
- vld1.8 {q1}, [r0], r1
- vld1.8 {q5}, [r2], r3
-
- vabdl.u8 q12, d0, d8
- vabdl.u8 q13, d1, d9
-
- vld1.8 {q2}, [r0], r1
- vld1.8 {q6}, [r2], r3
-
- vabal.u8 q12, d2, d10
- vabal.u8 q13, d3, d11
-
- vld1.8 {q3}, [r0], r1
- vld1.8 {q7}, [r2], r3
-
- vabal.u8 q12, d4, d12
- vabal.u8 q13, d5, d13
-
- vld1.8 {q0}, [r0], r1
- vld1.8 {q4}, [r2], r3
-
- vabal.u8 q12, d6, d14
- vabal.u8 q13, d7, d15
-
- vld1.8 {q1}, [r0], r1
- vld1.8 {q5}, [r2], r3
-
- vabal.u8 q12, d0, d8
- vabal.u8 q13, d1, d9
-
- vld1.8 {q2}, [r0], r1
- vld1.8 {q6}, [r2], r3
-
- vabal.u8 q12, d2, d10
- vabal.u8 q13, d3, d11
-
- vld1.8 {q3}, [r0], r1
- vld1.8 {q7}, [r2], r3
-
- vabal.u8 q12, d4, d12
- vabal.u8 q13, d5, d13
-
- vabal.u8 q12, d6, d14
- vabal.u8 q13, d7, d15
-
- vadd.u16 q0, q12, q13
-
- vpaddl.u16 q1, q0
- vpaddl.u32 q0, q1
-
- vadd.u32 d0, d0, d1
-
- vmov.32 r0, d0[0]
-
- vpop {d8-d15}
- bx lr
-
- ENDP
-
- END
diff --git a/vp8/common/arm/neon/sad8_neon.asm b/vp8/common/arm/neon/sad8_neon.asm
deleted file mode 100644
index 6b849d9..0000000
--- a/vp8/common/arm/neon/sad8_neon.asm
+++ /dev/null
@@ -1,215 +0,0 @@
-;
-; Copyright (c) 2010 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 |vp8_sad8x8_neon|
- EXPORT |vp8_sad8x16_neon|
- EXPORT |vp8_sad4x4_neon|
-
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-; unsigned int vp8_sad8x8_c(
-; unsigned char *src_ptr,
-; int src_stride,
-; unsigned char *ref_ptr,
-; int ref_stride)
-
-|vp8_sad8x8_neon| PROC
- vpush {d8-d15}
- vld1.8 {d0}, [r0], r1
- vld1.8 {d8}, [r2], r3
-
- vld1.8 {d2}, [r0], r1
- vld1.8 {d10}, [r2], r3
-
- vabdl.u8 q12, d0, d8
-
- vld1.8 {d4}, [r0], r1
- vld1.8 {d12}, [r2], r3
-
- vabal.u8 q12, d2, d10
-
- vld1.8 {d6}, [r0], r1
- vld1.8 {d14}, [r2], r3
-
- vabal.u8 q12, d4, d12
-
- vld1.8 {d0}, [r0], r1
- vld1.8 {d8}, [r2], r3
-
- vabal.u8 q12, d6, d14
-
- vld1.8 {d2}, [r0], r1
- vld1.8 {d10}, [r2], r3
-
- vabal.u8 q12, d0, d8
-
- vld1.8 {d4}, [r0], r1
- vld1.8 {d12}, [r2], r3
-
- vabal.u8 q12, d2, d10
-
- vld1.8 {d6}, [r0], r1
- vld1.8 {d14}, [r2], r3
-
- vabal.u8 q12, d4, d12
- vabal.u8 q12, d6, d14
-
- vpaddl.u16 q1, q12
- vpaddl.u32 q0, q1
- vadd.u32 d0, d0, d1
-
- vmov.32 r0, d0[0]
-
- vpop {d8-d15}
- bx lr
-
- ENDP
-
-;============================
-;unsigned int vp8_sad8x16_c(
-; unsigned char *src_ptr,
-; int src_stride,
-; unsigned char *ref_ptr,
-; int ref_stride)
-
-|vp8_sad8x16_neon| PROC
- vpush {d8-d15}
- vld1.8 {d0}, [r0], r1
- vld1.8 {d8}, [r2], r3
-
- vld1.8 {d2}, [r0], r1
- vld1.8 {d10}, [r2], r3
-
- vabdl.u8 q12, d0, d8
-
- vld1.8 {d4}, [r0], r1
- vld1.8 {d12}, [r2], r3
-
- vabal.u8 q12, d2, d10
-
- vld1.8 {d6}, [r0], r1
- vld1.8 {d14}, [r2], r3
-
- vabal.u8 q12, d4, d12
-
- vld1.8 {d0}, [r0], r1
- vld1.8 {d8}, [r2], r3
-
- vabal.u8 q12, d6, d14
-
- vld1.8 {d2}, [r0], r1
- vld1.8 {d10}, [r2], r3
-
- vabal.u8 q12, d0, d8
-
- vld1.8 {d4}, [r0], r1
- vld1.8 {d12}, [r2], r3
-
- vabal.u8 q12, d2, d10
-
- vld1.8 {d6}, [r0], r1
- vld1.8 {d14}, [r2], r3
-
- vabal.u8 q12, d4, d12
-
- vld1.8 {d0}, [r0], r1
- vld1.8 {d8}, [r2], r3
-
- vabal.u8 q12, d6, d14
-
- vld1.8 {d2}, [r0], r1
- vld1.8 {d10}, [r2], r3
-
- vabal.u8 q12, d0, d8
-
- vld1.8 {d4}, [r0], r1
- vld1.8 {d12}, [r2], r3
-
- vabal.u8 q12, d2, d10
-
- vld1.8 {d6}, [r0], r1
- vld1.8 {d14}, [r2], r3
-
- vabal.u8 q12, d4, d12
-
- vld1.8 {d0}, [r0], r1
- vld1.8 {d8}, [r2], r3
-
- vabal.u8 q12, d6, d14
-
- vld1.8 {d2}, [r0], r1
- vld1.8 {d10}, [r2], r3
-
- vabal.u8 q12, d0, d8
-
- vld1.8 {d4}, [r0], r1
- vld1.8 {d12}, [r2], r3
-
- vabal.u8 q12, d2, d10
-
- vld1.8 {d6}, [r0], r1
- vld1.8 {d14}, [r2], r3
-
- vabal.u8 q12, d4, d12
- vabal.u8 q12, d6, d14
-
- vpaddl.u16 q1, q12
- vpaddl.u32 q0, q1
- vadd.u32 d0, d0, d1
-
- vmov.32 r0, d0[0]
-
- vpop {d8-d15}
- bx lr
-
- ENDP
-
-;===========================
-;unsigned int vp8_sad4x4_c(
-; unsigned char *src_ptr,
-; int src_stride,
-; unsigned char *ref_ptr,
-; int ref_stride)
-
-|vp8_sad4x4_neon| PROC
- vpush {d8-d15}
- vld1.8 {d0}, [r0], r1
- vld1.8 {d8}, [r2], r3
-
- vld1.8 {d2}, [r0], r1
- vld1.8 {d10}, [r2], r3
-
- vabdl.u8 q12, d0, d8
-
- vld1.8 {d4}, [r0], r1
- vld1.8 {d12}, [r2], r3
-
- vabal.u8 q12, d2, d10
-
- vld1.8 {d6}, [r0], r1
- vld1.8 {d14}, [r2], r3
-
- vabal.u8 q12, d4, d12
- vabal.u8 q12, d6, d14
-
- vpaddl.u16 d1, d24
- vpaddl.u32 d0, d1
- vmov.32 r0, d0[0]
-
- vpop {d8-d15}
- bx lr
-
- ENDP
-
- END
diff --git a/vp8/common/arm/neon/sad_neon.c b/vp8/common/arm/neon/sad_neon.c
new file mode 100644
index 0000000..6595ac0
--- /dev/null
+++ b/vp8/common/arm/neon/sad_neon.c
@@ -0,0 +1,184 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+unsigned int vp8_sad8x8_neon(
+ unsigned char *src_ptr,
+ int src_stride,
+ unsigned char *ref_ptr,
+ int ref_stride) {
+ uint8x8_t d0, d8;
+ uint16x8_t q12;
+ uint32x4_t q1;
+ uint64x2_t q3;
+ uint32x2_t d5;
+ int i;
+
+ d0 = vld1_u8(src_ptr);
+ src_ptr += src_stride;
+ d8 = vld1_u8(ref_ptr);
+ ref_ptr += ref_stride;
+ q12 = vabdl_u8(d0, d8);
+
+ for (i = 0; i < 7; i++) {
+ d0 = vld1_u8(src_ptr);
+ src_ptr += src_stride;
+ d8 = vld1_u8(ref_ptr);
+ ref_ptr += ref_stride;
+ q12 = vabal_u8(q12, d0, d8);
+ }
+
+ q1 = vpaddlq_u16(q12);
+ q3 = vpaddlq_u32(q1);
+ d5 = vadd_u32(vreinterpret_u32_u64(vget_low_u64(q3)),
+ vreinterpret_u32_u64(vget_high_u64(q3)));
+
+ return vget_lane_u32(d5, 0);
+}
+
+unsigned int vp8_sad8x16_neon(
+ unsigned char *src_ptr,
+ int src_stride,
+ unsigned char *ref_ptr,
+ int ref_stride) {
+ uint8x8_t d0, d8;
+ uint16x8_t q12;
+ uint32x4_t q1;
+ uint64x2_t q3;
+ uint32x2_t d5;
+ int i;
+
+ d0 = vld1_u8(src_ptr);
+ src_ptr += src_stride;
+ d8 = vld1_u8(ref_ptr);
+ ref_ptr += ref_stride;
+ q12 = vabdl_u8(d0, d8);
+
+ for (i = 0; i < 15; i++) {
+ d0 = vld1_u8(src_ptr);
+ src_ptr += src_stride;
+ d8 = vld1_u8(ref_ptr);
+ ref_ptr += ref_stride;
+ q12 = vabal_u8(q12, d0, d8);
+ }
+
+ q1 = vpaddlq_u16(q12);
+ q3 = vpaddlq_u32(q1);
+ d5 = vadd_u32(vreinterpret_u32_u64(vget_low_u64(q3)),
+ vreinterpret_u32_u64(vget_high_u64(q3)));
+
+ return vget_lane_u32(d5, 0);
+}
+
+unsigned int vp8_sad4x4_neon(
+ unsigned char *src_ptr,
+ int src_stride,
+ unsigned char *ref_ptr,
+ int ref_stride) {
+ uint8x8_t d0, d8;
+ uint16x8_t q12;
+ uint32x2_t d1;
+ uint64x1_t d3;
+ int i;
+
+ d0 = vld1_u8(src_ptr);
+ src_ptr += src_stride;
+ d8 = vld1_u8(ref_ptr);
+ ref_ptr += ref_stride;
+ q12 = vabdl_u8(d0, d8);
+
+ for (i = 0; i < 3; i++) {
+ d0 = vld1_u8(src_ptr);
+ src_ptr += src_stride;
+ d8 = vld1_u8(ref_ptr);
+ ref_ptr += ref_stride;
+ q12 = vabal_u8(q12, d0, d8);
+ }
+
+ d1 = vpaddl_u16(vget_low_u16(q12));
+ d3 = vpaddl_u32(d1);
+
+ return vget_lane_u32(vreinterpret_u32_u64(d3), 0);
+}
+
+unsigned int vp8_sad16x16_neon(
+ unsigned char *src_ptr,
+ int src_stride,
+ unsigned char *ref_ptr,
+ int ref_stride) {
+ uint8x16_t q0, q4;
+ uint16x8_t q12, q13;
+ uint32x4_t q1;
+ uint64x2_t q3;
+ uint32x2_t d5;
+ int i;
+
+ q0 = vld1q_u8(src_ptr);
+ src_ptr += src_stride;
+ q4 = vld1q_u8(ref_ptr);
+ ref_ptr += ref_stride;
+ q12 = vabdl_u8(vget_low_u8(q0), vget_low_u8(q4));
+ q13 = vabdl_u8(vget_high_u8(q0), vget_high_u8(q4));
+
+ for (i = 0; i < 15; i++) {
+ q0 = vld1q_u8(src_ptr);
+ src_ptr += src_stride;
+ q4 = vld1q_u8(ref_ptr);
+ ref_ptr += ref_stride;
+ q12 = vabal_u8(q12, vget_low_u8(q0), vget_low_u8(q4));
+ q13 = vabal_u8(q13, vget_high_u8(q0), vget_high_u8(q4));
+ }
+
+ q12 = vaddq_u16(q12, q13);
+ q1 = vpaddlq_u16(q12);
+ q3 = vpaddlq_u32(q1);
+ d5 = vadd_u32(vreinterpret_u32_u64(vget_low_u64(q3)),
+ vreinterpret_u32_u64(vget_high_u64(q3)));
+
+ return vget_lane_u32(d5, 0);
+}
+
+unsigned int vp8_sad16x8_neon(
+ unsigned char *src_ptr,
+ int src_stride,
+ unsigned char *ref_ptr,
+ int ref_stride) {
+ uint8x16_t q0, q4;
+ uint16x8_t q12, q13;
+ uint32x4_t q1;
+ uint64x2_t q3;
+ uint32x2_t d5;
+ int i;
+
+ q0 = vld1q_u8(src_ptr);
+ src_ptr += src_stride;
+ q4 = vld1q_u8(ref_ptr);
+ ref_ptr += ref_stride;
+ q12 = vabdl_u8(vget_low_u8(q0), vget_low_u8(q4));
+ q13 = vabdl_u8(vget_high_u8(q0), vget_high_u8(q4));
+
+ for (i = 0; i < 7; i++) {
+ q0 = vld1q_u8(src_ptr);
+ src_ptr += src_stride;
+ q4 = vld1q_u8(ref_ptr);
+ ref_ptr += ref_stride;
+ q12 = vabal_u8(q12, vget_low_u8(q0), vget_low_u8(q4));
+ q13 = vabal_u8(q13, vget_high_u8(q0), vget_high_u8(q4));
+ }
+
+ q12 = vaddq_u16(q12, q13);
+ q1 = vpaddlq_u16(q12);
+ q3 = vpaddlq_u32(q1);
+ d5 = vadd_u32(vreinterpret_u32_u64(vget_low_u64(q3)),
+ vreinterpret_u32_u64(vget_high_u64(q3)));
+
+ return vget_lane_u32(d5, 0);
+}
diff --git a/vp8/common/arm/neon/shortidct4x4llm_neon.asm b/vp8/common/arm/neon/shortidct4x4llm_neon.asm
deleted file mode 100644
index 87ca887..0000000
--- a/vp8/common/arm/neon/shortidct4x4llm_neon.asm
+++ /dev/null
@@ -1,142 +0,0 @@
-;
-; Copyright (c) 2010 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 |vp8_short_idct4x4llm_neon|
- ARM
- REQUIRE8
- PRESERVE8
-
- AREA ||.text||, CODE, READONLY, ALIGN=2
-
-;*************************************************************
-;void vp8_short_idct4x4llm_c(short *input, unsigned char *pred, int pitch,
-; unsigned char *dst, int stride)
-;r0 short * input
-;r1 short * pred
-;r2 int pitch
-;r3 unsigned char dst
-;sp int stride
-;*************************************************************
-
-; static const int cospi8sqrt2minus1=20091;
-; static const int sinpi8sqrt2 =35468;
-; static const int rounding = 0;
-
-; Optimization note: The resulted data from dequantization are signed
-; 13-bit data that is in the range of [-4096, 4095]. This allows to
-; use "vqdmulh"(neon) instruction since it won't go out of range
-; (13+16+1=30bits<32bits). This instruction gives the high half
-; result of the multiplication that is needed in IDCT.
-
-|vp8_short_idct4x4llm_neon| PROC
- vpush {d8-d15}
-
- adr r12, idct_coeff
- vld1.16 {q1, q2}, [r0]
- vld1.16 {d0}, [r12]
-
- vswp d3, d4 ;q2(vp[4] vp[12])
- ldr r0, [sp, #64] ; stride
-
- vqdmulh.s16 q3, q2, d0[2]
- vqdmulh.s16 q4, q2, d0[0]
-
- vqadd.s16 d12, d2, d3 ;a1
- vqsub.s16 d13, d2, d3 ;b1
-
- vshr.s16 q3, q3, #1
- vshr.s16 q4, q4, #1
-
- vqadd.s16 q3, q3, q2 ;modify since sinpi8sqrt2 > 65536/2 (negtive number)
- vqadd.s16 q4, q4, q2
-
- ;d6 - c1:temp1
- ;d7 - d1:temp2
- ;d8 - d1:temp1
- ;d9 - c1:temp2
-
- vqsub.s16 d10, d6, d9 ;c1
- vqadd.s16 d11, d7, d8 ;d1
-
- vqadd.s16 d2, d12, d11
- vqadd.s16 d3, d13, d10
- vqsub.s16 d4, d13, d10
- vqsub.s16 d5, d12, d11
-
- vtrn.32 d2, d4
- vtrn.32 d3, d5
- vtrn.16 d2, d3
- vtrn.16 d4, d5
-
- vswp d3, d4
-
- vqdmulh.s16 q3, q2, d0[2]
- vqdmulh.s16 q4, q2, d0[0]
-
- vqadd.s16 d12, d2, d3 ;a1
- vqsub.s16 d13, d2, d3 ;b1
-
- vshr.s16 q3, q3, #1
- vshr.s16 q4, q4, #1
-
- vqadd.s16 q3, q3, q2 ;modify since sinpi8sqrt2 > 65536/2 (negtive number)
- vqadd.s16 q4, q4, q2
-
- vqsub.s16 d10, d6, d9 ;c1
- vqadd.s16 d11, d7, d8 ;d1
-
- vqadd.s16 d2, d12, d11
- vqadd.s16 d3, d13, d10
- vqsub.s16 d4, d13, d10
- vqsub.s16 d5, d12, d11
-
- vrshr.s16 d2, d2, #3
- vrshr.s16 d3, d3, #3
- vrshr.s16 d4, d4, #3
- vrshr.s16 d5, d5, #3
-
- vtrn.32 d2, d4
- vtrn.32 d3, d5
- vtrn.16 d2, d3
- vtrn.16 d4, d5
-
- ; load prediction data
- vld1.32 d6[0], [r1], r2
- vld1.32 d6[1], [r1], r2
- vld1.32 d7[0], [r1], r2
- vld1.32 d7[1], [r1], r2
-
- ; add prediction and residual
- vaddw.u8 q1, q1, d6
- vaddw.u8 q2, q2, d7
-
- vqmovun.s16 d1, q1
- vqmovun.s16 d2, q2
-
- ; store to destination
- vst1.32 d1[0], [r3], r0
- vst1.32 d1[1], [r3], r0
- vst1.32 d2[0], [r3], r0
- vst1.32 d2[1], [r3], r0
-
- vpop {d8-d15}
- bx lr
-
- ENDP
-
-;-----------------
-
-idct_coeff
- DCD 0x4e7b4e7b, 0x8a8c8a8c
-
-;20091, 20091, 35468, 35468
-
- END
diff --git a/vp8/common/arm/neon/shortidct4x4llm_neon.c b/vp8/common/arm/neon/shortidct4x4llm_neon.c
new file mode 100644
index 0000000..373afa6
--- /dev/null
+++ b/vp8/common/arm/neon/shortidct4x4llm_neon.c
@@ -0,0 +1,123 @@
+/*
+ * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+
+static const int16_t cospi8sqrt2minus1 = 20091;
+static const int16_t sinpi8sqrt2 = 35468;
+
+void vp8_short_idct4x4llm_neon(
+ int16_t *input,
+ unsigned char *pred_ptr,
+ int pred_stride,
+ unsigned char *dst_ptr,
+ int dst_stride) {
+ int i;
+ uint32x2_t d6u32 = vdup_n_u32(0);
+ uint8x8_t d1u8;
+ int16x4_t d2, d3, d4, d5, d10, d11, d12, d13;
+ uint16x8_t q1u16;
+ int16x8_t q1s16, q2s16, q3s16, q4s16;
+ int32x2x2_t v2tmp0, v2tmp1;
+ int16x4x2_t v2tmp2, v2tmp3;
+
+ d2 = vld1_s16(input);
+ d3 = vld1_s16(input + 4);
+ d4 = vld1_s16(input + 8);
+ d5 = vld1_s16(input + 12);
+
+ // 1st for loop
+ q1s16 = vcombine_s16(d2, d4); // Swap d3 d4 here
+ q2s16 = vcombine_s16(d3, d5);
+
+ q3s16 = vqdmulhq_n_s16(q2s16, sinpi8sqrt2);
+ q4s16 = vqdmulhq_n_s16(q2s16, cospi8sqrt2minus1);
+
+ d12 = vqadd_s16(vget_low_s16(q1s16), vget_high_s16(q1s16)); // a1
+ d13 = vqsub_s16(vget_low_s16(q1s16), vget_high_s16(q1s16)); // b1
+
+ q3s16 = vshrq_n_s16(q3s16, 1);
+ q4s16 = vshrq_n_s16(q4s16, 1);
+
+ q3s16 = vqaddq_s16(q3s16, q2s16);
+ q4s16 = vqaddq_s16(q4s16, q2s16);
+
+ d10 = vqsub_s16(vget_low_s16(q3s16), vget_high_s16(q4s16)); // c1
+ d11 = vqadd_s16(vget_high_s16(q3s16), vget_low_s16(q4s16)); // d1
+
+ d2 = vqadd_s16(d12, d11);
+ d3 = vqadd_s16(d13, d10);
+ d4 = vqsub_s16(d13, d10);
+ d5 = vqsub_s16(d12, d11);
+
+ v2tmp0 = vtrn_s32(vreinterpret_s32_s16(d2), vreinterpret_s32_s16(d4));
+ v2tmp1 = vtrn_s32(vreinterpret_s32_s16(d3), vreinterpret_s32_s16(d5));
+ v2tmp2 = vtrn_s16(vreinterpret_s16_s32(v2tmp0.val[0]),
+ vreinterpret_s16_s32(v2tmp1.val[0]));
+ v2tmp3 = vtrn_s16(vreinterpret_s16_s32(v2tmp0.val[1]),
+ vreinterpret_s16_s32(v2tmp1.val[1]));
+
+ // 2nd for loop
+ q1s16 = vcombine_s16(v2tmp2.val[0], v2tmp3.val[0]);
+ q2s16 = vcombine_s16(v2tmp2.val[1], v2tmp3.val[1]);
+
+ q3s16 = vqdmulhq_n_s16(q2s16, sinpi8sqrt2);
+ q4s16 = vqdmulhq_n_s16(q2s16, cospi8sqrt2minus1);
+
+ d12 = vqadd_s16(vget_low_s16(q1s16), vget_high_s16(q1s16)); // a1
+ d13 = vqsub_s16(vget_low_s16(q1s16), vget_high_s16(q1s16)); // b1
+
+ q3s16 = vshrq_n_s16(q3s16, 1);
+ q4s16 = vshrq_n_s16(q4s16, 1);
+
+ q3s16 = vqaddq_s16(q3s16, q2s16);
+ q4s16 = vqaddq_s16(q4s16, q2s16);
+
+ d10 = vqsub_s16(vget_low_s16(q3s16), vget_high_s16(q4s16)); // c1
+ d11 = vqadd_s16(vget_high_s16(q3s16), vget_low_s16(q4s16)); // d1
+
+ d2 = vqadd_s16(d12, d11);
+ d3 = vqadd_s16(d13, d10);
+ d4 = vqsub_s16(d13, d10);
+ d5 = vqsub_s16(d12, d11);
+
+ d2 = vrshr_n_s16(d2, 3);
+ d3 = vrshr_n_s16(d3, 3);
+ d4 = vrshr_n_s16(d4, 3);
+ d5 = vrshr_n_s16(d5, 3);
+
+ v2tmp0 = vtrn_s32(vreinterpret_s32_s16(d2), vreinterpret_s32_s16(d4));
+ v2tmp1 = vtrn_s32(vreinterpret_s32_s16(d3), vreinterpret_s32_s16(d5));
+ v2tmp2 = vtrn_s16(vreinterpret_s16_s32(v2tmp0.val[0]),
+ vreinterpret_s16_s32(v2tmp1.val[0]));
+ v2tmp3 = vtrn_s16(vreinterpret_s16_s32(v2tmp0.val[1]),
+ vreinterpret_s16_s32(v2tmp1.val[1]));
+
+ q1s16 = vcombine_s16(v2tmp2.val[0], v2tmp2.val[1]);
+ q2s16 = vcombine_s16(v2tmp3.val[0], v2tmp3.val[1]);
+
+ // dc_only_idct_add
+ for (i = 0; i < 2; i++, q1s16 = q2s16) {
+ d6u32 = vld1_lane_u32((const uint32_t *)pred_ptr, d6u32, 0);
+ pred_ptr += pred_stride;
+ d6u32 = vld1_lane_u32((const uint32_t *)pred_ptr, d6u32, 1);
+ pred_ptr += pred_stride;
+
+ q1u16 = vaddw_u8(vreinterpretq_u16_s16(q1s16),
+ vreinterpret_u8_u32(d6u32));
+ d1u8 = vqmovun_s16(vreinterpretq_s16_u16(q1u16));
+
+ vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d1u8), 0);
+ dst_ptr += dst_stride;
+ vst1_lane_u32((uint32_t *)dst_ptr, vreinterpret_u32_u8(d1u8), 1);
+ dst_ptr += dst_stride;
+ }
+ return;
+}
diff --git a/vp8/encoder/x86/quantize_ssse3.asm b/vp8/encoder/x86/quantize_ssse3.asm
deleted file mode 100644
index 7b1dc11..0000000
--- a/vp8/encoder/x86/quantize_ssse3.asm
+++ /dev/null
@@ -1,138 +0,0 @@
-;
-; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
-;
-; Use of this source code is governed by a BSD-style license and patent
-; grant that can be found in the LICENSE file in the root of the source
-; tree. All contributing project authors may be found in the AUTHORS
-; file in the root of the source tree.
-;
-
-
-%include "vpx_ports/x86_abi_support.asm"
-%include "vp8_asm_enc_offsets.asm"
-
-
-; void vp8_fast_quantize_b_ssse3 | arg
-; (BLOCK *b, | 0
-; BLOCKD *d) | 1
-;
-
-global sym(vp8_fast_quantize_b_ssse3) PRIVATE
-sym(vp8_fast_quantize_b_ssse3):
- push rbp
- mov rbp, rsp
- GET_GOT rbx
-
-%if ABI_IS_32BIT
- push rdi
- push rsi
-%else
- %if LIBVPX_YASM_WIN64
- push rdi
- push rsi
- %endif
-%endif
- ; end prolog
-
-%if ABI_IS_32BIT
- mov rdi, arg(0) ; BLOCK *b
- mov rsi, arg(1) ; BLOCKD *d
-%else
- %if LIBVPX_YASM_WIN64
- mov rdi, rcx ; BLOCK *b
- mov rsi, rdx ; BLOCKD *d
- %else
- ;mov rdi, rdi ; BLOCK *b
- ;mov rsi, rsi ; BLOCKD *d
- %endif
-%endif
-
- mov rax, [rdi + vp8_block_coeff]
- mov rcx, [rdi + vp8_block_round]
- mov rdx, [rdi + vp8_block_quant_fast]
-
- ; coeff
- movdqa xmm0, [rax]
- movdqa xmm4, [rax + 16]
-
- ; round
- movdqa xmm2, [rcx]
- movdqa xmm3, [rcx + 16]
-
- movdqa xmm1, xmm0
- movdqa xmm5, xmm4
-
- ; sz = z >> 15
- psraw xmm0, 15
- psraw xmm4, 15
-
- pabsw xmm1, xmm1
- pabsw xmm5, xmm5
-
- paddw xmm1, xmm2
- paddw xmm5, xmm3
-
- ; quant_fast
- pmulhw xmm1, [rdx]
- pmulhw xmm5, [rdx + 16]
-
- mov rax, [rsi + vp8_blockd_qcoeff]
- mov rdi, [rsi + vp8_blockd_dequant]
- mov rcx, [rsi + vp8_blockd_dqcoeff]
-
- movdqa xmm2, xmm1 ;store y for getting eob
- movdqa xmm3, xmm5
-
- pxor xmm1, xmm0
- pxor xmm5, xmm4
- psubw xmm1, xmm0
- psubw xmm5, xmm4
-
- movdqa [rax], xmm1
- movdqa [rax + 16], xmm5
-
- movdqa xmm0, [rdi]
- movdqa xmm4, [rdi + 16]
-
- pmullw xmm0, xmm1
- pmullw xmm4, xmm5
- pxor xmm1, xmm1
-
- pcmpgtw xmm2, xmm1 ;calculate eob
- pcmpgtw xmm3, xmm1
- packsswb xmm2, xmm3
- pshufb xmm2, [GLOBAL(zz_shuf)]
-
- pmovmskb edx, xmm2
-
- movdqa [rcx], xmm0 ;store dqcoeff
- movdqa [rcx + 16], xmm4 ;store dqcoeff
- mov rcx, [rsi + vp8_blockd_eob]
-
- bsr eax, edx ;count 0
- add eax, 1
-
- cmp edx, 0 ;if all 0, eob=0
- cmove eax, edx
-
- mov BYTE PTR [rcx], al ;store eob
-
- ; begin epilog
-%if ABI_IS_32BIT
- pop rsi
- pop rdi
-%else
- %if LIBVPX_YASM_WIN64
- pop rsi
- pop rdi
- %endif
-%endif
-
- RESTORE_GOT
- pop rbp
- ret
-
-SECTION_RODATA
-align 16
-zz_shuf:
- db 0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15
diff --git a/vp8/encoder/x86/quantize_ssse3.c b/vp8/encoder/x86/quantize_ssse3.c
new file mode 100644
index 0000000..9b4471d
--- /dev/null
+++ b/vp8/encoder/x86/quantize_ssse3.c
@@ -0,0 +1,110 @@
+/*
+ * Copyright (c) 2012 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <tmmintrin.h> /* SSSE3 */
+
+#include "vp8/encoder/block.h"
+
+/* bitscan reverse (bsr) */
+#if defined(_MSC_VER)
+#include <intrin.h>
+#pragma intrinsic(_BitScanReverse)
+static int bsr(int mask) {
+ int eob;
+ _BitScanReverse(&eob, mask);
+ eob++;
+ if (mask == 0)
+ eob = 0;
+ return eob;
+}
+#else
+static int bsr(int mask) {
+ int eob;
+ asm volatile("bsr %1, %0" : "=r" (eob) : "r" (mask) : "flags");
+ eob++;
+ if (mask == 0)
+ eob = 0;
+ return eob;
+}
+#endif
+
+void vp8_fast_quantize_b_ssse3(BLOCK *b, BLOCKD *d) {
+ int eob, mask;
+
+ __m128i z0 = _mm_load_si128((__m128i *)(b->coeff));
+ __m128i z1 = _mm_load_si128((__m128i *)(b->coeff + 8));
+ __m128i round0 = _mm_load_si128((__m128i *)(b->round));
+ __m128i round1 = _mm_load_si128((__m128i *)(b->round + 8));
+ __m128i quant_fast0 = _mm_load_si128((__m128i *)(b->quant_fast));
+ __m128i quant_fast1 = _mm_load_si128((__m128i *)(b->quant_fast + 8));
+ __m128i dequant0 = _mm_load_si128((__m128i *)(d->dequant));
+ __m128i dequant1 = _mm_load_si128((__m128i *)(d->dequant + 8));
+
+ __m128i sz0, sz1, x, x0, x1, y0, y1, zeros, abs0, abs1;
+
+ DECLARE_ALIGNED(16, const uint8_t, pshufb_zig_zag_mask[16]) =
+ { 0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15 };
+ __m128i zig_zag = _mm_load_si128((const __m128i *)pshufb_zig_zag_mask);
+
+ /* sign of z: z >> 15 */
+ sz0 = _mm_srai_epi16(z0, 15);
+ sz1 = _mm_srai_epi16(z1, 15);
+
+ /* x = abs(z) */
+ x0 = _mm_abs_epi16(z0);
+ x1 = _mm_abs_epi16(z1);
+
+ /* x += round */
+ x0 = _mm_add_epi16(x0, round0);
+ x1 = _mm_add_epi16(x1, round1);
+
+ /* y = (x * quant) >> 16 */
+ y0 = _mm_mulhi_epi16(x0, quant_fast0);
+ y1 = _mm_mulhi_epi16(x1, quant_fast1);
+
+ /* ASM saves Y for EOB */
+ /* I think we can ignore that because adding the sign doesn't change anything
+ * and multiplying 0 by dequant is OK as well */
+ abs0 = y0;
+ abs1 = y1;
+
+ /* Restore the sign bit. */
+ y0 = _mm_xor_si128(y0, sz0);
+ y1 = _mm_xor_si128(y1, sz1);
+ x0 = _mm_sub_epi16(y0, sz0);
+ x1 = _mm_sub_epi16(y1, sz1);
+
+ /* qcoeff = x */
+ _mm_store_si128((__m128i *)(d->qcoeff), x0);
+ _mm_store_si128((__m128i *)(d->qcoeff + 8), x1);
+
+ /* x * dequant */
+ x0 = _mm_mullo_epi16(x0, dequant0);
+ x1 = _mm_mullo_epi16(x1, dequant1);
+
+ /* dqcoeff = x * dequant */
+ _mm_store_si128((__m128i *)(d->dqcoeff), x0);
+ _mm_store_si128((__m128i *)(d->dqcoeff + 8), x1);
+
+ zeros = _mm_setzero_si128();
+
+ x0 = _mm_cmpgt_epi16(abs0, zeros);
+ x1 = _mm_cmpgt_epi16(abs1, zeros);
+
+ x = _mm_packs_epi16(x0, x1);
+
+ x = _mm_shuffle_epi8(x, zig_zag);
+
+ mask = _mm_movemask_epi8(x);
+
+ eob = bsr(mask);
+
+ *d->eob = 0xFF & eob;
+}
diff --git a/vp8/vp8_common.mk b/vp8/vp8_common.mk
index cdb2716..2c611ed 100644
--- a/vp8/vp8_common.mk
+++ b/vp8/vp8_common.mk
@@ -159,14 +159,6 @@
VP8_COMMON_SRCS-$(HAVE_MEDIA) += common/arm/armv6/vp8_variance_halfpixvar16x16_hv_armv6$(ASM)
# common (neon)
-VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/iwalsh_neon$(ASM)
-VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/loopfilter_neon$(ASM)
-VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/loopfiltersimplehorizontaledge_neon$(ASM)
-VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/loopfiltersimpleverticaledge_neon$(ASM)
-VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/mbloopfilter_neon$(ASM)
-VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/shortidct4x4llm_neon$(ASM)
-VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sad8_neon$(ASM)
-VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sad16_neon$(ASM)
VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sixtappredict4x4_neon$(ASM)
VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sixtappredict8x4_neon$(ASM)
VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sixtappredict8x8_neon$(ASM)
@@ -186,6 +178,13 @@
VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/dequant_idct_neon.c
VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/dequantizeb_neon.c
VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/idct_dequant_full_2x_neon.c
+VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/iwalsh_neon.c
+VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/loopfilter_neon.c
+VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/loopfiltersimplehorizontaledge_neon.c
+VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/loopfiltersimpleverticaledge_neon.c
+VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/mbloopfilter_neon.c
+VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/sad_neon.c
+VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/shortidct4x4llm_neon.c
$(eval $(call rtcd_h_template,vp8_rtcd,vp8/common/rtcd_defs.pl))
diff --git a/vp8/vp8cx.mk b/vp8/vp8cx.mk
index d7c6dd1..607382b 100644
--- a/vp8/vp8cx.mk
+++ b/vp8/vp8cx.mk
@@ -88,6 +88,7 @@
VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/dct_sse2.asm
VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/fwalsh_sse2.asm
VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/quantize_sse2.c
+VP8_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/quantize_ssse3.c
ifeq ($(CONFIG_TEMPORAL_DENOISING),yes)
VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/denoising_sse2.c
@@ -96,7 +97,6 @@
VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/subtract_sse2.asm
VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/temporal_filter_apply_sse2.asm
VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp8_enc_stubs_sse2.c
-VP8_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/quantize_ssse3.asm
VP8_CX_SRCS-$(HAVE_SSE4_1) += encoder/x86/quantize_sse4.asm
VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/quantize_mmx.asm
VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/encodeopt.asm
