VP8 for ARMv8 by using NEON intrinsics 12

Add sad_neon.c
- vp8_sad16x16_neon
- vp8_sad16x8_neon
- vp8_sad8x8_neon
- vp8_sad8x16_neon
- vp8_sad4x4_neon

Change-Id: I08eaae49ec03fb91b394354660a5df0367cea311
Signed-off-by: James Yu <james.yu@linaro.org>

4 files changed, 185 insertions, 429 deletions

diff --git a/vp8/common/arm/neon/sad16_neon.asm b/vp8/common/arm/neon/sad16_neon.asm
deleted file mode 100644
index 7197e5655..000000000
--- 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 6b849d933..000000000
--- 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 000000000..6595ac051
--- /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/vp8_common.mk b/vp8/vp8_common.mk
index 1b658352b..964e378e0 100644
--- a/vp8/vp8_common.mk
+++ b/vp8/vp8_common.mk
@@ -160,8 +160,6 @@ VP8_COMMON_SRCS-$(HAVE_MEDIA)  += common/arm/armv6/vp8_variance_halfpixvar16x16_
 
 # common (neon)
 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 +184,7 @@ 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
 
 
 $(eval $(call rtcd_h_template,vp8_rtcd,vp8/common/rtcd_defs.pl))