VP9 common for ARMv8 by using NEON intrinsics 11

Add vp9_idct16x16_1_add_neon.c
- vp9_idct16x16_1_add_neon

Change-Id: I7c6524024ad4cb4e66aa38f1c887e733503c39df
Signed-off-by: James Yu <james.yu@linaro.org>

diff --git a/vp9/common/arm/neon/vp9_idct16x16_1_add_neon.c b/vp9/common/arm/neon/vp9_idct16x16_1_add_neon.c
new file mode 100644 (file)
index 0000000..3c8c6a9
--- /dev/null
+++ b/vp9/common/arm/neon/vp9_idct16x16_1_add_neon.c
@@ -0,0 +1,59 @@
+/*
+ *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <arm_neon.h>
+#include "vp9/common/vp9_idct.h"
+
+void vp9_idct16x16_1_add_neon(
+        int16_t *input,
+        uint8_t *dest,
+        int dest_stride) {
+    uint8x8_t d2u8, d3u8, d30u8, d31u8;
+    uint64x1_t d2u64, d3u64, d4u64, d5u64;
+    uint16x8_t q0u16, q9u16, q10u16, q11u16, q12u16;
+    int16x8_t q0s16;
+    uint8_t *d1, *d2;
+    int16_t i, j, a1, cospi_16_64 = 11585;
+    int16_t out = dct_const_round_shift(input[0] * cospi_16_64);
+    out = dct_const_round_shift(out * cospi_16_64);
+    a1 = ROUND_POWER_OF_TWO(out, 6);
+
+    q0s16 = vdupq_n_s16(a1);
+    q0u16 = vreinterpretq_u16_s16(q0s16);
+
+    for (d1 = d2 = dest, i = 0; i < 4; i++) {
+        for (j = 0; j < 2; j++) {
+            d2u64 = vld1_u64((const uint64_t *)d1);
+            d3u64 = vld1_u64((const uint64_t *)(d1 + 8));
+            d1 += dest_stride;
+            d4u64 = vld1_u64((const uint64_t *)d1);
+            d5u64 = vld1_u64((const uint64_t *)(d1 + 8));
+            d1 += dest_stride;
+
+            q9u16 = vaddw_u8(q0u16, vreinterpret_u8_u64(d2u64));
+            q10u16 = vaddw_u8(q0u16, vreinterpret_u8_u64(d3u64));
+            q11u16 = vaddw_u8(q0u16, vreinterpret_u8_u64(d4u64));
+            q12u16 = vaddw_u8(q0u16, vreinterpret_u8_u64(d5u64));
+
+            d2u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16));
+            d3u8 = vqmovun_s16(vreinterpretq_s16_u16(q10u16));
+            d30u8 = vqmovun_s16(vreinterpretq_s16_u16(q11u16));
+            d31u8 = vqmovun_s16(vreinterpretq_s16_u16(q12u16));
+
+            vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d2u8));
+            vst1_u64((uint64_t *)(d2 + 8), vreinterpret_u64_u8(d3u8));
+            d2 += dest_stride;
+            vst1_u64((uint64_t *)d2, vreinterpret_u64_u8(d30u8));
+            vst1_u64((uint64_t *)(d2 + 8), vreinterpret_u64_u8(d31u8));
+            d2 += dest_stride;
+        }
+    }
+    return;
+}

diff --git a/vp9/common/arm/neon/vp9_idct16x16_1_add_neon.asm b/vp9/common/arm/neon/vp9_idct16x16_1_add_neon_asm.asm
similarity index 100%
rename from vp9/common/arm/neon/vp9_idct16x16_1_add_neon.asm
rename to vp9/common/arm/neon/vp9_idct16x16_1_add_neon_asm.asm

diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index 05f7ddf99818fdd4a5fa599b63a48e9d39934d9e..90df14857e4f80b984d607802495453eba0af493 100644 (file)
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -440,8 +440,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
     $vp9_idct8x8_12_add_neon_asm=vp9_idct8x8_12_add_neon;
 
     add_proto qw/void vp9_idct16x16_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
-    specialize qw/vp9_idct16x16_1_add sse2 neon_asm dspr2/;
-    $vp9_idct16x16_1_add_neon_asm=vp9_idct16x16_1_add_neon;
+    specialize qw/vp9_idct16x16_1_add sse2 neon dspr2/;
 
     add_proto qw/void vp9_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
     specialize qw/vp9_idct16x16_256_add sse2 ssse3 neon_asm dspr2/;

diff --git a/vp9/vp9_common.mk b/vp9/vp9_common.mk
index 370be3351d96a53bcb46ad98b1e13047be372e44..35f4d99a77045769e352f58b4e23f02f98455ec7 100644 (file)
--- a/vp9/vp9_common.mk
+++ b/vp9/vp9_common.mk
@@ -136,7 +136,6 @@ VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_loopfilter_16_neon_asm$(
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_dc_only_idct_add_neon$(ASM)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct4x4_add_neon$(ASM)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct8x8_add_neon$(ASM)
-VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct16x16_1_add_neon$(ASM)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct16x16_add_neon$(ASM)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_idct32x32_add_neon$(ASM)
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_iht4x4_add_neon$(ASM)
@@ -153,6 +152,7 @@ VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_avg_neon_asm$(ASM)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_neon_asm$(ASM)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_copy_neon_asm$(ASM)
+VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_1_add_neon_asm$(ASM)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct32x32_1_add_neon_asm$(ASM)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct4x4_1_add_neon_asm$(ASM)
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_1_add_neon_asm$(ASM)
@@ -165,6 +165,7 @@ VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_avg_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve8_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_convolve_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_copy_neon.c
+VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_1_add_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct32x32_1_add_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct4x4_1_add_neon.c
 VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_1_add_neon.c