Merge "Fork VP9 and VP10 codebase"

configure

@@ -37,6 +37,7 @@ Advanced options:
   ${toggle_vp9_highbitdepth}      use VP9 high bit depth (10/12) profiles
   ${toggle_vp8}                   VP8 codec support
   ${toggle_vp9}                   VP9 codec support
+  ${toggle_vp10}                  VP10 codec support
   ${toggle_internal_stats}        output of encoder internal stats for debug, if supported (encoders)
   ${toggle_postproc}              postprocessing
   ${toggle_vp9_postproc}          vp9 specific postprocessing
@@ -191,6 +192,7 @@ fi
 # disable codecs when their source directory does not exist
 [ -d "${source_path}/vp8" ] || disable_feature vp8
 [ -d "${source_path}/vp9" ] || disable_feature vp9
+[ -d "${source_path}/vp10" ] || disable_feature vp10
 
 # install everything except the sources, by default. sources will have
 # to be enabled when doing dist builds, since that's no longer a common
@@ -212,10 +214,13 @@ CODECS="
     vp8_decoder
     vp9_encoder
     vp9_decoder
+    vp10_encoder
+    vp10_decoder
 "
 CODEC_FAMILIES="
     vp8
     vp9
+    vp10
 "
 
 ARCH_LIST="

libs.mk

@@ -109,6 +109,40 @@ endif
 VP9_PREFIX=vp9/
 $(BUILD_PFX)$(VP9_PREFIX)%.c.o: CFLAGS += -Wextra
 
+#  VP10 make file
+ifneq ($(CONFIG_VP10_ENCODER)$(CONFIG_VP10_DECODER),)
+  VP10_PREFIX=vp10/
+  include $(SRC_PATH_BARE)/$(VP10_PREFIX)vp10_common.mk
+endif
+
+ifeq ($(CONFIG_VP10_ENCODER),yes)
+  VP10_PREFIX=vp10/
+  include $(SRC_PATH_BARE)/$(VP10_PREFIX)vp10cx.mk
+  CODEC_SRCS-yes += $(addprefix $(VP10_PREFIX),$(call enabled,VP10_CX_SRCS))
+  CODEC_EXPORTS-yes += $(addprefix $(VP10_PREFIX),$(VP10_CX_EXPORTS))
+  CODEC_SRCS-yes += $(VP10_PREFIX)vp10cx.mk vpx/vp8.h vpx/vp8cx.h
+  INSTALL-LIBS-yes += include/vpx/vp8.h include/vpx/vp8cx.h
+  INSTALL-LIBS-$(CONFIG_SPATIAL_SVC) += include/vpx/svc_context.h
+  INSTALL_MAPS += include/vpx/% $(SRC_PATH_BARE)/$(VP10_PREFIX)/%
+  CODEC_DOC_SRCS += vpx/vp8.h vpx/vp8cx.h
+  CODEC_DOC_SECTIONS += vp9 vp9_encoder
+endif
+
+ifeq ($(CONFIG_VP10_DECODER),yes)
+  VP10_PREFIX=vp10/
+  include $(SRC_PATH_BARE)/$(VP10_PREFIX)vp10dx.mk
+  CODEC_SRCS-yes += $(addprefix $(VP10_PREFIX),$(call enabled,VP10_DX_SRCS))
+  CODEC_EXPORTS-yes += $(addprefix $(VP10_PREFIX),$(VP10_DX_EXPORTS))
+  CODEC_SRCS-yes += $(VP10_PREFIX)vp10dx.mk vpx/vp8.h vpx/vp8dx.h
+  INSTALL-LIBS-yes += include/vpx/vp8.h include/vpx/vp8dx.h
+  INSTALL_MAPS += include/vpx/% $(SRC_PATH_BARE)/$(VP10_PREFIX)/%
+  CODEC_DOC_SRCS += vpx/vp8.h vpx/vp8dx.h
+  CODEC_DOC_SECTIONS += vp9 vp9_decoder
+endif
+
+VP10_PREFIX=vp10/
+$(BUILD_PFX)$(VP10_PREFIX)%.c.o: CFLAGS += -Wextra
+
 ifeq ($(CONFIG_ENCODERS),yes)
   CODEC_DOC_SECTIONS += encoder
 endif

test/codec_factory.h

@@ -13,10 +13,10 @@
 #include "./vpx_config.h"
 #include "vpx/vpx_decoder.h"
 #include "vpx/vpx_encoder.h"
-#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER
+#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER || CONFIG_VP10_ENCODER
 #include "vpx/vp8cx.h"
 #endif
-#if CONFIG_VP8_DECODER || CONFIG_VP9_DECODER
+#if CONFIG_VP8_DECODER || CONFIG_VP9_DECODER || CONFIG_VP10_DECODER
 #include "vpx/vp8dx.h"
 #endif
 
@@ -233,6 +233,8 @@ class VP9CodecFactory : public CodecFactory {
                                                int usage) const {
 #if CONFIG_VP9_ENCODER
     return vpx_codec_enc_config_default(&vpx_codec_vp9_cx_algo, cfg, usage);
+#elif CONFIG_VP10_ENCODER
+    return vpx_codec_enc_config_default(&vpx_codec_vp10_cx_algo, cfg, usage);
 #else
     return VPX_CODEC_INCAPABLE;
 #endif
@@ -251,7 +253,96 @@ const libvpx_test::VP9CodecFactory kVP9;
 #define VP9_INSTANTIATE_TEST_CASE(test, ...)
 #endif  // CONFIG_VP9
 
+/*
+ * VP10 Codec Definitions
+ */
+#if CONFIG_VP10
+class VP10Decoder : public Decoder {
+ public:
+  VP10Decoder(vpx_codec_dec_cfg_t cfg, unsigned long deadline)
+      : Decoder(cfg, deadline) {}
 
-}  // namespace libvpx_test
+  VP10Decoder(vpx_codec_dec_cfg_t cfg, const vpx_codec_flags_t flag,
+              unsigned long deadline)  // NOLINT
+      : Decoder(cfg, flag, deadline) {}
+
+ protected:
+  virtual vpx_codec_iface_t* CodecInterface() const {
+#if CONFIG_VP10_DECODER
+    return &vpx_codec_vp10_dx_algo;
+#else
+    return NULL;
+#endif
+  }
+};
+
+class VP10Encoder : public Encoder {
+ public:
+  VP10Encoder(vpx_codec_enc_cfg_t cfg, unsigned long deadline,
+              const unsigned long init_flags, TwopassStatsStore *stats)
+      : Encoder(cfg, deadline, init_flags, stats) {}
+
+ protected:
+  virtual vpx_codec_iface_t* CodecInterface() const {
+#if CONFIG_VP10_ENCODER
+    return &vpx_codec_vp10_cx_algo;
+#else
+    return NULL;
+#endif
+  }
+};
+
+class VP10CodecFactory : public CodecFactory {
+ public:
+  VP10CodecFactory() : CodecFactory() {}
+
+  virtual Decoder* CreateDecoder(vpx_codec_dec_cfg_t cfg,
+                                 unsigned long deadline) const {
+    return CreateDecoder(cfg, 0, deadline);
+  }
 
+  virtual Decoder* CreateDecoder(vpx_codec_dec_cfg_t cfg,
+                                 const vpx_codec_flags_t flags,
+                                 unsigned long deadline) const {  // NOLINT
+#if CONFIG_VP10_DECODER
+    return new VP10Decoder(cfg, flags, deadline);
+#else
+    return NULL;
+#endif
+  }
+
+  virtual Encoder* CreateEncoder(vpx_codec_enc_cfg_t cfg,
+                                 unsigned long deadline,
+                                 const unsigned long init_flags,
+                                 TwopassStatsStore *stats) const {
+#if CONFIG_VP10_ENCODER
+    return new VP10Encoder(cfg, deadline, init_flags, stats);
+#else
+    return NULL;
+#endif
+  }
+
+  virtual vpx_codec_err_t DefaultEncoderConfig(vpx_codec_enc_cfg_t *cfg,
+                                               int usage) const {
+#if CONFIG_VP10_ENCODER
+    return vpx_codec_enc_config_default(&vpx_codec_vp10_cx_algo, cfg, usage);
+#else
+    return VPX_CODEC_INCAPABLE;
+#endif
+  }
+};
+
+const libvpx_test::VP10CodecFactory kVP10;
+
+#define VP10_INSTANTIATE_TEST_CASE(test, ...)\
+  INSTANTIATE_TEST_CASE_P(VP10, test, \
+      ::testing::Combine( \
+          ::testing::Values(static_cast<const libvpx_test::CodecFactory*>( \
+               &libvpx_test::kVP10)), \
+          __VA_ARGS__))
+#else
+#define VP10_INSTANTIATE_TEST_CASE(test, ...)
+#endif  // CONFIG_VP10
+
+}  // namespace libvpx_test
 #endif  // TEST_CODEC_FACTORY_H_

test/decode_api_test.cc

@@ -26,6 +26,9 @@ TEST(DecodeAPI, InvalidParams) {
 #endif
 #if CONFIG_VP9_DECODER
     &vpx_codec_vp9_dx_algo,
+#endif
+#if CONFIG_VP10_DECODER
+    &vpx_codec_vp10_dx_algo,
 #endif
   };
   uint8_t buf[1] = {0};

test/encode_test_driver.h

@@ -16,7 +16,7 @@
 #include "third_party/googletest/src/include/gtest/gtest.h"
 
 #include "./vpx_config.h"
-#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER
+#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER || CONFIG_VP10_ENCODER
 #include "vpx/vp8cx.h"
 #endif
 #include "vpx/vpx_encoder.h"
@@ -138,7 +138,7 @@ class Encoder {
     const vpx_codec_err_t res = vpx_codec_control_(&encoder_, ctrl_id, arg);
     ASSERT_EQ(VPX_CODEC_OK, res) << EncoderError();
   }
-#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER
+#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER || CONFIG_VP10_ENCODER
   void Control(int ctrl_id, vpx_active_map_t *arg) {
     const vpx_codec_err_t res = vpx_codec_control_(&encoder_, ctrl_id, arg);
     ASSERT_EQ(VPX_CODEC_OK, res) << EncoderError();

tools_common.c

@@ -16,11 +16,11 @@
 
 #include "./tools_common.h"
 
-#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER
+#if CONFIG_VP8_ENCODER || CONFIG_VP9_ENCODER || CONFIG_VP10_ENCODER
 #include "vpx/vp8cx.h"
 #endif
 
-#if CONFIG_VP8_DECODER || CONFIG_VP9_DECODER
+#if CONFIG_VP8_DECODER || CONFIG_VP9_DECODER || CONFIG_VP10_DECODER
 #include "vpx/vp8dx.h"
 #endif
 
@@ -140,6 +140,10 @@ static const VpxInterface vpx_encoders[] = {
 #if CONFIG_VP9_ENCODER
   {"vp9", VP9_FOURCC, &vpx_codec_vp9_cx},
 #endif
+
+#if CONFIG_VP10_ENCODER
+  {"vp10", VP10_FOURCC, &vpx_codec_vp10_cx},
+#endif
 };
 
 int get_vpx_encoder_count(void) {
@@ -174,6 +178,10 @@ static const VpxInterface vpx_decoders[] = {
 #if CONFIG_VP9_DECODER
   {"vp9", VP9_FOURCC, &vpx_codec_vp9_dx},
 #endif
+
+#if CONFIG_VP10_DECODER
+  {"vp10", VP10_FOURCC, &vpx_codec_vp10_dx},
+#endif
 };
 
 int get_vpx_decoder_count(void) {

tools_common.h

@@ -62,6 +62,7 @@
 
 #define VP8_FOURCC 0x30385056
 #define VP9_FOURCC 0x30395056
+#define VP10_FOURCC 0x303a5056
 
 enum VideoFileType {
   FILE_TYPE_RAW,

vp10/common/arm/neon/vp9_iht4x4_add_neon.c

+/*
+ *  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 <assert.h>
+
+#include "./vp10_rtcd.h"
+#include "./vpx_config.h"
+#include "vp10/common/vp9_common.h"
+
+static int16_t sinpi_1_9 = 0x14a3;
+static int16_t sinpi_2_9 = 0x26c9;
+static int16_t sinpi_3_9 = 0x3441;
+static int16_t sinpi_4_9 = 0x3b6c;
+static int16_t cospi_8_64 = 0x3b21;
+static int16_t cospi_16_64 = 0x2d41;
+static int16_t cospi_24_64 = 0x187e;
+
+static INLINE void TRANSPOSE4X4(
+        int16x8_t *q8s16,
+        int16x8_t *q9s16) {
+    int32x4_t q8s32, q9s32;
+    int16x4x2_t d0x2s16, d1x2s16;
+    int32x4x2_t q0x2s32;
+
+    d0x2s16 = vtrn_s16(vget_low_s16(*q8s16), vget_high_s16(*q8s16));
+    d1x2s16 = vtrn_s16(vget_low_s16(*q9s16), vget_high_s16(*q9s16));
+
+    q8s32 = vreinterpretq_s32_s16(vcombine_s16(d0x2s16.val[0], d0x2s16.val[1]));
+    q9s32 = vreinterpretq_s32_s16(vcombine_s16(d1x2s16.val[0], d1x2s16.val[1]));
+    q0x2s32 = vtrnq_s32(q8s32, q9s32);
+
+    *q8s16 = vreinterpretq_s16_s32(q0x2s32.val[0]);
+    *q9s16 = vreinterpretq_s16_s32(q0x2s32.val[1]);
+    return;
+}
+
+static INLINE void GENERATE_COSINE_CONSTANTS(
+        int16x4_t *d0s16,
+        int16x4_t *d1s16,
+        int16x4_t *d2s16) {
+    *d0s16 = vdup_n_s16(cospi_8_64);
+    *d1s16 = vdup_n_s16(cospi_16_64);
+    *d2s16 = vdup_n_s16(cospi_24_64);
+    return;
+}
+
+static INLINE void GENERATE_SINE_CONSTANTS(
+        int16x4_t *d3s16,
+        int16x4_t *d4s16,
+        int16x4_t *d5s16,
+        int16x8_t *q3s16) {
+    *d3s16 = vdup_n_s16(sinpi_1_9);
+    *d4s16 = vdup_n_s16(sinpi_2_9);
+    *q3s16 = vdupq_n_s16(sinpi_3_9);
+    *d5s16 = vdup_n_s16(sinpi_4_9);
+    return;
+}
+
+static INLINE void IDCT4x4_1D(
+        int16x4_t *d0s16,
+        int16x4_t *d1s16,
+        int16x4_t *d2s16,
+        int16x8_t *q8s16,
+        int16x8_t *q9s16) {
+    int16x4_t d16s16, d17s16, d18s16, d19s16, d23s16, d24s16;
+    int16x4_t d26s16, d27s16, d28s16, d29s16;
+    int32x4_t q10s32, q13s32, q14s32, q15s32;
+    int16x8_t q13s16, q14s16;
+
+    d16s16 = vget_low_s16(*q8s16);
+    d17s16 = vget_high_s16(*q8s16);
+    d18s16 = vget_low_s16(*q9s16);
+    d19s16 = vget_high_s16(*q9s16);
+
+    d23s16 = vadd_s16(d16s16, d18s16);
+    d24s16 = vsub_s16(d16s16, d18s16);
+
+    q15s32 = vmull_s16(d17s16, *d2s16);
+    q10s32 = vmull_s16(d17s16, *d0s16);
+    q13s32 = vmull_s16(d23s16, *d1s16);
+    q14s32 = vmull_s16(d24s16, *d1s16);
+    q15s32 = vmlsl_s16(q15s32, d19s16, *d0s16);
+    q10s32 = vmlal_s16(q10s32, d19s16, *d2s16);
+
+    d26s16 = vqrshrn_n_s32(q13s32, 14);
+    d27s16 = vqrshrn_n_s32(q14s32, 14);
+    d29s16 = vqrshrn_n_s32(q15s32, 14);
+    d28s16 = vqrshrn_n_s32(q10s32, 14);
+
+    q13s16 = vcombine_s16(d26s16, d27s16);
+    q14s16 = vcombine_s16(d28s16, d29s16);
+    *q8s16 = vaddq_s16(q13s16, q14s16);
+    *q9s16 = vsubq_s16(q13s16, q14s16);
+    *q9s16 = vcombine_s16(vget_high_s16(*q9s16),
+                          vget_low_s16(*q9s16));  // vswp
+    return;
+}
+
+static INLINE void IADST4x4_1D(
+        int16x4_t *d3s16,
+        int16x4_t *d4s16,
+        int16x4_t *d5s16,
+        int16x8_t *q3s16,
+        int16x8_t *q8s16,
+        int16x8_t *q9s16) {
+    int16x4_t d6s16, d16s16, d17s16, d18s16, d19s16;
+    int32x4_t q8s32, q9s32, q10s32, q11s32, q12s32, q13s32, q14s32, q15s32;
+
+    d6s16 = vget_low_s16(*q3s16);
+
+    d16s16 = vget_low_s16(*q8s16);
+    d17s16 = vget_high_s16(*q8s16);
+    d18s16 = vget_low_s16(*q9s16);
+    d19s16 = vget_high_s16(*q9s16);
+
+    q10s32 = vmull_s16(*d3s16, d16s16);
+    q11s32 = vmull_s16(*d4s16, d16s16);
+    q12s32 = vmull_s16(d6s16, d17s16);
+    q13s32 = vmull_s16(*d5s16, d18s16);
+    q14s32 = vmull_s16(*d3s16, d18s16);
+    q15s32 = vmovl_s16(d16s16);
+    q15s32 = vaddw_s16(q15s32, d19s16);
+    q8s32  = vmull_s16(*d4s16, d19s16);
+    q15s32 = vsubw_s16(q15s32, d18s16);
+    q9s32  = vmull_s16(*d5s16, d19s16);
+
+    q10s32 = vaddq_s32(q10s32, q13s32);
+    q10s32 = vaddq_s32(q10s32, q8s32);
+    q11s32 = vsubq_s32(q11s32, q14s32);
+    q8s32  = vdupq_n_s32(sinpi_3_9);
+    q11s32 = vsubq_s32(q11s32, q9s32);
+    q15s32 = vmulq_s32(q15s32, q8s32);
+
+    q13s32 = vaddq_s32(q10s32, q12s32);
+    q10s32 = vaddq_s32(q10s32, q11s32);
+    q14s32 = vaddq_s32(q11s32, q12s32);
+    q10s32 = vsubq_s32(q10s32, q12s32);
+
+    d16s16 = vqrshrn_n_s32(q13s32, 14);
+    d17s16 = vqrshrn_n_s32(q14s32, 14);
+    d18s16 = vqrshrn_n_s32(q15s32, 14);
+    d19s16 = vqrshrn_n_s32(q10s32, 14);
+
+    *q8s16 = vcombine_s16(d16s16, d17s16);
+    *q9s16 = vcombine_s16(d18s16, d19s16);
+    return;
+}
+
+void vp10_iht4x4_16_add_neon(const tran_low_t *input, uint8_t *dest,
+                            int dest_stride, int tx_type) {
+    uint8x8_t d26u8, d27u8;
+    int16x4_t d0s16, d1s16, d2s16, d3s16, d4s16, d5s16;
+    uint32x2_t d26u32, d27u32;
+    int16x8_t q3s16, q8s16, q9s16;
+    uint16x8_t q8u16, q9u16;
+
+    d26u32 = d27u32 = vdup_n_u32(0);
+
+    q8s16 = vld1q_s16(input);
+    q9s16 = vld1q_s16(input + 8);
+
+    TRANSPOSE4X4(&q8s16, &q9s16);
+
+    switch (tx_type) {
+      case 0:  // idct_idct is not supported. Fall back to C
+        vp10_iht4x4_16_add_c(input, dest, dest_stride, tx_type);
+        return;
+        break;
+      case 1:  // iadst_idct
+        // generate constants
+        GENERATE_COSINE_CONSTANTS(&d0s16, &d1s16, &d2s16);
+        GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16);
+
+        // first transform rows
+        IDCT4x4_1D(&d0s16, &d1s16, &d2s16, &q8s16, &q9s16);
+
+        // transpose the matrix
+        TRANSPOSE4X4(&q8s16, &q9s16);
+
+        // then transform columns
+        IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
+        break;
+      case 2:  // idct_iadst
+        // generate constantsyy
+        GENERATE_COSINE_CONSTANTS(&d0s16, &d1s16, &d2s16);
+        GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16);
+
+        // first transform rows
+        IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
+
+        // transpose the matrix
+        TRANSPOSE4X4(&q8s16, &q9s16);
+
+        // then transform columns
+        IDCT4x4_1D(&d0s16, &d1s16, &d2s16, &q8s16, &q9s16);
+        break;
+      case 3:  // iadst_iadst
+        // generate constants
+        GENERATE_SINE_CONSTANTS(&d3s16, &d4s16, &d5s16, &q3s16);
+
+        // first transform rows
+        IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
+
+        // transpose the matrix
+        TRANSPOSE4X4(&q8s16, &q9s16);
+
+        // then transform columns
+        IADST4x4_1D(&d3s16, &d4s16, &d5s16, &q3s16, &q8s16, &q9s16);
+        break;
+      default:  // iadst_idct
+        assert(0);
+        break;
+    }
+
+    q8s16 = vrshrq_n_s16(q8s16, 4);
+    q9s16 = vrshrq_n_s16(q9s16, 4);
+
+    d26u32 = vld1_lane_u32((const uint32_t *)dest, d26u32, 0);
+    dest += dest_stride;
+    d26u32 = vld1_lane_u32((const uint32_t *)dest, d26u32, 1);
+    dest += dest_stride;
+    d27u32 = vld1_lane_u32((const uint32_t *)dest, d27u32, 0);
+    dest += dest_stride;
+    d27u32 = vld1_lane_u32((const uint32_t *)dest, d27u32, 1);
+
+    q8u16 = vaddw_u8(vreinterpretq_u16_s16(q8s16), vreinterpret_u8_u32(d26u32));
+    q9u16 = vaddw_u8(vreinterpretq_u16_s16(q9s16), vreinterpret_u8_u32(d27u32));
+
+    d26u8 = vqmovun_s16(vreinterpretq_s16_u16(q8u16));
+    d27u8 = vqmovun_s16(vreinterpretq_s16_u16(q9u16));
+
+    vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d27u8), 1);
+    dest -= dest_stride;
+    vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d27u8), 0);
+    dest -= dest_stride;
+    vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d26u8), 1);
+    dest -= dest_stride;
+    vst1_lane_u32((uint32_t *)dest, vreinterpret_u32_u8(d26u8), 0);
+    return;
+}

vp10/common/mips/dspr2/vp9_itrans16_dspr2.c

+/*
+ *  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+#include <stdio.h>
+
+#include "./vpx_config.h"
+#include "./vp10_rtcd.h"
+#include "vp10/common/vp9_common.h"
+#include "vp10/common/vp9_blockd.h"
+#include "vp10/common/vp9_idct.h"
+#include "vpx_dsp/mips/inv_txfm_dspr2.h"
+#include "vpx_dsp/txfm_common.h"
+#include "vpx_ports/mem.h"
+
+#if HAVE_DSPR2
+void vp10_iht16x16_256_add_dspr2(const int16_t *input, uint8_t *dest,
+                                int pitch, int tx_type) {
+  int i, j;
+  DECLARE_ALIGNED(32, int16_t,  out[16 * 16]);
+  int16_t *outptr = out;
+  int16_t temp_out[16];
+  uint32_t pos = 45;
+
+  /* bit positon for extract from acc */
+  __asm__ __volatile__ (
+    "wrdsp    %[pos],    1    \n\t"
+    :
+    : [pos] "r" (pos)
+  );
+
+  switch (tx_type) {
+    case DCT_DCT:     // DCT in both horizontal and vertical
+      idct16_rows_dspr2(input, outptr, 16);
+      idct16_cols_add_blk_dspr2(out, dest, pitch);
+      break;
+    case ADST_DCT:    // ADST in vertical, DCT in horizontal
+      idct16_rows_dspr2(input, outptr, 16);
+
+      outptr = out;
+
+      for (i = 0; i < 16; ++i) {
+        iadst16_dspr2(outptr, temp_out);
+
+        for (j = 0; j < 16; ++j)
+          dest[j * pitch + i] =
+                    clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6)
+                                      + dest[j * pitch + i]);
+        outptr += 16;
+      }
+      break;
+    case DCT_ADST:    // DCT in vertical, ADST in horizontal
+    {
+      int16_t temp_in[16 * 16];
+
+      for (i = 0; i < 16; ++i) {
+        /* prefetch row */
+        prefetch_load((const uint8_t *)(input + 16));
+
+        iadst16_dspr2(input, outptr);
+        input += 16;
+        outptr += 16;
+      }