Factor inverse transform functions into vpx_dsp

This commit moves the module inverse transform functions from vp9
to vpx_dsp folder. The hybrid transform wrapper functions stay in
the vp9 folder, since it involves codec-specific data structures.

Change-Id: Ib066367c953d3d024c73ba65157bbd70a95c9ef8

test/idct8x8_test.cc

@@ -14,8 +14,7 @@
 
 #include "third_party/googletest/src/include/gtest/gtest.h"
 
-#include "./vp9_rtcd.h"
-
+#include "./vpx_dsp_rtcd.h"
 #include "test/acm_random.h"
 #include "vpx/vpx_integer.h"
 

vp9/common/vp9_idct.h

@@ -14,63 +14,16 @@
 #include <assert.h>
 
 #include "./vpx_config.h"
-#include "vpx_dsp/txfm_common.h"
-#if CONFIG_VP9_HIGHBITDEPTH
-#include "vpx_dsp/vpx_dsp_common.h"
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-#include "vpx_ports/mem.h"
 #include "vp9/common/vp9_common.h"
 #include "vp9/common/vp9_enums.h"
+#include "vpx_dsp/inv_txfm.h"
+#include "vpx_dsp/txfm_common.h"
+#include "vpx_ports/mem.h"
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-static INLINE tran_low_t check_range(tran_high_t input) {
-#if CONFIG_COEFFICIENT_RANGE_CHECKING
-  // For valid VP9 input streams, intermediate stage coefficients should always
-  // stay within the range of a signed 16 bit integer. Coefficients can go out
-  // of this range for invalid/corrupt VP9 streams. However, strictly checking
-  // this range for every intermediate coefficient can burdensome for a decoder,
-  // therefore the following assertion is only enabled when configured with
-  // --enable-coefficient-range-checking.
-  assert(INT16_MIN <= input);
-  assert(input <= INT16_MAX);
-#endif  // CONFIG_COEFFICIENT_RANGE_CHECKING
-  return (tran_low_t)input;
-}
-
-static INLINE tran_low_t dct_const_round_shift(tran_high_t input) {
-  tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
-  return check_range(rv);
-}
-
-#if CONFIG_VP9_HIGHBITDEPTH
-static INLINE tran_low_t highbd_check_range(tran_high_t input,
-                                            int bd) {
-#if CONFIG_COEFFICIENT_RANGE_CHECKING
-  // For valid highbitdepth VP9 streams, intermediate stage coefficients will
-  // stay within the ranges:
-  // - 8 bit: signed 16 bit integer
-  // - 10 bit: signed 18 bit integer
-  // - 12 bit: signed 20 bit integer
-  const int32_t int_max = (1 << (7 + bd)) - 1;
-  const int32_t int_min = -int_max - 1;
-  assert(int_min <= input);
-  assert(input <= int_max);
-  (void) int_min;
-#endif  // CONFIG_COEFFICIENT_RANGE_CHECKING
-  (void) bd;
-  return (tran_low_t)input;
-}
-
-static INLINE tran_low_t highbd_dct_const_round_shift(tran_high_t input,
-                                                      int bd) {
-  tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
-  return highbd_check_range(rv, bd);
-}
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-
 typedef void (*transform_1d)(const tran_low_t*, tran_low_t*);
 
 typedef struct {
@@ -85,28 +38,6 @@ typedef struct {
 } highbd_transform_2d;
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 
-#if CONFIG_EMULATE_HARDWARE
-// When CONFIG_EMULATE_HARDWARE is 1 the transform performs a
-// non-normative method to handle overflows. A stream that causes
-// overflows  in the inverse transform is considered invalid in VP9,
-// and a hardware implementer is free to choose any reasonable
-// method to handle overflows. However to aid in hardware
-// verification they can use a specific implementation of the
-// WRAPLOW() macro below that is identical to their intended
-// hardware implementation (and also use configure options to trigger
-// the C-implementation of the transform).
-//
-// The particular WRAPLOW implementation below performs strict
-// overflow wrapping to match common hardware implementations.
-// bd of 8 uses trans_low with 16bits, need to remove 16bits
-// bd of 10 uses trans_low with 18bits, need to remove 14bits
-// bd of 12 uses trans_low with 20bits, need to remove 12bits
-// bd of x uses trans_low with 8+x bits, need to remove 24-x bits
-#define WRAPLOW(x, bd) ((((int32_t)(x)) << (24 - bd)) >> (24 - bd))
-#else
-#define WRAPLOW(x, bd) ((int32_t)(x))
-#endif  // CONFIG_EMULATE_HARDWARE
-
 void vp9_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
                      int eob);
 void vp9_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
@@ -126,9 +57,6 @@ void vp9_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input, uint8_t *dest,
                       int stride, int eob);
 
 #if CONFIG_VP9_HIGHBITDEPTH
-void vp9_highbd_idct4(const tran_low_t *input, tran_low_t *output, int bd);
-void vp9_highbd_idct8(const tran_low_t *input, tran_low_t *output, int bd);
-void vp9_highbd_idct16(const tran_low_t *input, tran_low_t *output, int bd);
 void vp9_highbd_iwht4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
                             int eob, int bd);
 void vp9_highbd_idct4x4_add(const tran_low_t *input, uint8_t *dest, int stride,
@@ -145,11 +73,6 @@ void vp9_highbd_iht8x8_add(TX_TYPE tx_type, const tran_low_t *input,
                            uint8_t *dest, int stride, int eob, int bd);
 void vp9_highbd_iht16x16_add(TX_TYPE tx_type, const tran_low_t *input,
                              uint8_t *dest, int stride, int eob, int bd);
-static INLINE uint16_t highbd_clip_pixel_add(uint16_t dest, tran_high_t trans,
-                                             int bd) {
-  trans = WRAPLOW(trans, bd);
-  return clip_pixel_highbd(WRAPLOW(dest + trans, bd), bd);
-}
 #endif  // CONFIG_VP9_HIGHBITDEPTH
 #ifdef __cplusplus
 }  // extern "C"

vp9/encoder/x86/vp9_dct_ssse3.c

@@ -17,7 +17,7 @@
 #include <tmmintrin.h>  // SSSE3
 
 #include "./vp9_rtcd.h"
-#include "vp9/common/x86/vp9_idct_intrin_sse2.h"
+#include "vpx_dsp/x86/inv_txfm_sse2.h"
 #include "vpx_dsp/x86/txfm_common_sse2.h"
 
 void vp9_fdct8x8_quant_ssse3(const int16_t *input, int stride,

vp9/vp9_common.mk

@@ -66,7 +66,6 @@ VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_postproc.h
 VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_postproc.c
 VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_mfqe.h
 VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_mfqe.c
-VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_idct_sse2.asm
 ifeq ($(CONFIG_VP9_POSTPROC),yes)
 VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_mfqe_sse2.asm
 VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_postproc_sse2.asm
@@ -96,13 +95,6 @@ VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_mfqe_msa.c
 endif
 
 VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_idct_intrin_sse2.c
-VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_idct_intrin_sse2.h
-
-ifeq ($(ARCH_X86_64), yes)
-ifeq ($(CONFIG_USE_X86INC),yes)
-VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_idct_ssse3_x86_64.asm
-endif
-endif
 
 VP9_COMMON_SRCS-$(HAVE_NEON_ASM) += common/arm/neon/vp9_save_reg_neon$(ASM)
 
@@ -111,30 +103,4 @@ VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_iht4x4_add_neon.c
 VP9_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/vp9_iht8x8_add_neon.c
 endif
 
-# neon with assembly and intrinsics implementations. If both are available
-# prefer assembly.
-ifeq ($(HAVE_NEON_ASM), yes)
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_1_add_neon_asm$(ASM)
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_add_neon_asm$(ASM)
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct32x32_1_add_neon_asm$(ASM)
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct32x32_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_idct4x4_add_neon_asm$(ASM)
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_1_add_neon_asm$(ASM)
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_add_neon_asm$(ASM)
-else
-ifeq ($(HAVE_NEON), yes)
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_1_add_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_add_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct16x16_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct32x32_1_add_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct32x32_add_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct4x4_1_add_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct4x4_add_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_1_add_neon.c
-VP9_COMMON_SRCS-yes += common/arm/neon/vp9_idct8x8_add_neon.c
-endif  # HAVE_NEON
-endif  # HAVE_NEON_ASM
-
 $(eval $(call rtcd_h_template,vp9_rtcd,vp9/common/vp9_rtcd_defs.pl))

vp9/common/arm/neon/vp9_idct16x16_1_add_neon.c → vpx_dsp/arm/idct16x16_1_add_neon.c

@@ -10,7 +10,7 @@
 
 #include <arm_neon.h>
 
-#include "vp9/common/vp9_idct.h"
+#include "vpx_dsp/inv_txfm.h"
 #include "vpx_ports/mem.h"
 
 void vp9_idct16x16_1_add_neon(

vp9/common/arm/neon/vp9_idct16x16_neon.c → vpx_dsp/arm/idct16x16_neon.c

@@ -8,7 +8,6 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-#include "./vp9_rtcd.h"
 #include "vp9/common/vp9_common.h"
 
 void vp9_idct16x16_256_add_neon_pass1(const int16_t *input,

vp9/common/arm/neon/vp9_idct32x32_1_add_neon.c → vpx_dsp/arm/idct32x32_1_add_neon.c

@@ -12,7 +12,7 @@
 
 #include "./vpx_config.h"
 
-#include "vp9/common/vp9_idct.h"
+#include "vpx_dsp/inv_txfm.h"
 #include "vpx_ports/mem.h"
 
 static INLINE void LD_16x8(

vp9/common/arm/neon/vp9_idct4x4_1_add_neon.c → vpx_dsp/arm/idct4x4_1_add_neon.c

@@ -10,7 +10,7 @@
 
 #include <arm_neon.h>
 
-#include "vp9/common/vp9_idct.h"
+#include "vpx_dsp/inv_txfm.h"
 #include "vpx_ports/mem.h"
 
 void vp9_idct4x4_1_add_neon(

vp9/common/arm/neon/vp9_idct8x8_1_add_neon.c → vpx_dsp/arm/idct8x8_1_add_neon.c

@@ -10,7 +10,7 @@
 
 #include <arm_neon.h>
 
-#include "vp9/common/vp9_idct.h"
+#include "vpx_dsp/inv_txfm.h"
 #include "vpx_ports/mem.h"
 
 void vp9_idct8x8_1_add_neon(

vpx_dsp/inv_txfm.h

+/*
+ *  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.
+ */
+
+#ifndef VPX_DSP_INV_TXFM_H_
+#define VPX_DSP_INV_TXFM_H_
+
+#include <assert.h>
+
+#include "./vpx_config.h"
+#include "vpx_dsp/txfm_common.h"
+#include "vpx_ports/mem.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+static INLINE tran_low_t check_range(tran_high_t input) {
+#if CONFIG_COEFFICIENT_RANGE_CHECKING
+  // For valid VP9 input streams, intermediate stage coefficients should always
+  // stay within the range of a signed 16 bit integer. Coefficients can go out
+  // of this range for invalid/corrupt VP9 streams. However, strictly checking
+  // this range for every intermediate coefficient can burdensome for a decoder,
+  // therefore the following assertion is only enabled when configured with
+  // --enable-coefficient-range-checking.
+  assert(INT16_MIN <= input);
+  assert(input <= INT16_MAX);
+#endif  // CONFIG_COEFFICIENT_RANGE_CHECKING
+  return (tran_low_t)input;
+}
+
+static INLINE tran_low_t dct_const_round_shift(tran_high_t input) {
+  tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
+  return check_range(rv);
+}
+
+#if CONFIG_VP9_HIGHBITDEPTH
+static INLINE tran_low_t highbd_check_range(tran_high_t input,
+                                            int bd) {
+#if CONFIG_COEFFICIENT_RANGE_CHECKING
+  // For valid highbitdepth VP9 streams, intermediate stage coefficients will
+  // stay within the ranges:
+  // - 8 bit: signed 16 bit integer
+  // - 10 bit: signed 18 bit integer
+  // - 12 bit: signed 20 bit integer
+  const int32_t int_max = (1 << (7 + bd)) - 1;
+  const int32_t int_min = -int_max - 1;
+  assert(int_min <= input);
+  assert(input <= int_max);
+  (void) int_min;
+#endif  // CONFIG_COEFFICIENT_RANGE_CHECKING
+  (void) bd;
+  return (tran_low_t)input;
+}
+
+static INLINE tran_low_t highbd_dct_const_round_shift(tran_high_t input,
+                                                      int bd) {
+  tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
+  return highbd_check_range(rv, bd);
+}
+#endif  // CONFIG_VP9_HIGHBITDEPTH
+
+#if CONFIG_EMULATE_HARDWARE
+// When CONFIG_EMULATE_HARDWARE is 1 the transform performs a
+// non-normative method to handle overflows. A stream that causes
+// overflows  in the inverse transform is considered invalid in VP9,
+// and a hardware implementer is free to choose any reasonable
+// method to handle overflows. However to aid in hardware
+// verification they can use a specific implementation of the
+// WRAPLOW() macro below that is identical to their intended
+// hardware implementation (and also use configure options to trigger
+// the C-implementation of the transform).
+//
+// The particular WRAPLOW implementation below performs strict
+// overflow wrapping to match common hardware implementations.
+// bd of 8 uses trans_low with 16bits, need to remove 16bits
+// bd of 10 uses trans_low with 18bits, need to remove 14bits
+// bd of 12 uses trans_low with 20bits, need to remove 12bits
+// bd of x uses trans_low with 8+x bits, need to remove 24-x bits
+#define WRAPLOW(x, bd) ((((int32_t)(x)) << (24 - bd)) >> (24 - bd))
+#else
+#define WRAPLOW(x, bd) ((int32_t)(x))
+#endif  // CONFIG_EMULATE_HARDWARE
+
+void idct4_c(const tran_low_t *input, tran_low_t *output);
+void idct8_c(const tran_low_t *input, tran_low_t *output);
+void idct16_c(const tran_low_t *input, tran_low_t *output);
+void idct32_c(const tran_low_t *input, tran_low_t *output);
+void iadst4_c(const tran_low_t *input, tran_low_t *output);
+void iadst8_c(const tran_low_t *input, tran_low_t *output);
+void iadst16_c(const tran_low_t *input, tran_low_t *output);
+
+#if CONFIG_VP9_HIGHBITDEPTH
+void vp9_highbd_idct4_c(const tran_low_t *input, tran_low_t *output, int bd);
+void vp9_highbd_idct8_c(const tran_low_t *input, tran_low_t *output, int bd);
+void vp9_highbd_idct16_c(const tran_low_t *input, tran_low_t *output, int bd);
+void highbd_idct32_c(const tran_low_t *input, tran_low_t *output, int bd);
+
+void highbd_iadst4_c(const tran_low_t *input, tran_low_t *output, int bd);
+void highbd_iadst8_c(const tran_low_t *input, tran_low_t *output, int bd);
+void highbd_iadst16_c(const tran_low_t *input, tran_low_t *output, int bd);
+
+static INLINE uint16_t highbd_clip_pixel_add(uint16_t dest, tran_high_t trans,
+                                             int bd) {
+  trans = WRAPLOW(trans, bd);
+  return clip_pixel_highbd(WRAPLOW(dest + trans, bd), bd);
+}
+#endif
+
+static INLINE uint8_t clip_pixel_add(uint8_t dest, tran_high_t trans) {
+  trans = WRAPLOW(trans, 8);
+  return clip_pixel(WRAPLOW(dest + trans, 8));
+}
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif  // VPX_DSP_INV_TXFM_H_

vpx_dsp/vpx_dsp.mk

@@ -169,6 +169,43 @@ DSP_SRCS-$(HAVE_MSA)    += mips/fwd_txfm_msa.c
 DSP_SRCS-$(HAVE_MSA)    += mips/fwd_dct32x32_msa.c
 endif  # CONFIG_VP9_ENCODER
 
+# inverse transform
+ifeq ($(CONFIG_VP9),yes)
+DSP_SRCS-yes            += inv_txfm.h
+DSP_SRCS-yes            += inv_txfm.c
+DSP_SRCS-$(HAVE_SSE2)   += x86/inv_txfm_sse2.h
+DSP_SRCS-$(HAVE_SSE2)   += x86/inv_txfm_sse2.c
+DSP_SRCS-$(HAVE_SSE2)   += x86/inv_txfm_sse2.asm
+ifeq ($(ARCH_X86_64),yes)
+ifeq ($(CONFIG_USE_X86INC),yes)
+DSP_SRCS-$(HAVE_SSSE3)  += x86/inv_txfm_ssse3_x86_64.asm
+endif  # CONFIG_USE_X86INC
+endif  # ARCH_X86_64
+
+ifeq ($(HAVE_NEON_ASM),yes)
+DSP_SRCS-yes  += arm/idct4x4_1_add_neon$(ASM)
+DSP_SRCS-yes  += arm/idct4x4_add_neon$(ASM)
+DSP_SRCS-yes  += arm/idct8x8_1_add_neon$(ASM)
+DSP_SRCS-yes  += arm/idct8x8_add_neon$(ASM)
+DSP_SRCS-yes  += arm/idct16x16_1_add_neon$(ASM)
+DSP_SRCS-yes  += arm/idct16x16_add_neon$(ASM)
+DSP_SRCS-yes  += arm/idct32x32_1_add_neon$(ASM)
+DSP_SRCS-yes  += arm/idct32x32_add_neon$(ASM)
+else
+ifeq ($(HAVE_NEON),yes)
+DSP_SRCS-yes  += arm/idct4x4_1_add_neon.c
+DSP_SRCS-yes  += arm/idct4x4_add_neon.c
+DSP_SRCS-yes  += arm/idct8x8_1_add_neon.c
+DSP_SRCS-yes  += arm/idct8x8_add_neon.c
+DSP_SRCS-yes  += arm/idct16x16_1_add_neon.c
+DSP_SRCS-yes  += arm/idct16x16_add_neon.c
+DSP_SRCS-yes  += arm/idct32x32_1_add_neon.c
+DSP_SRCS-yes  += arm/idct32x32_add_neon.c
+endif  # HAVE_NEON
+endif  # HAVE_NEON_ASM
+DSP_SRCS-$(HAVE_NEON)  += arm/idct16x16_neon.c
+endif  # CONFIG_VP9
+
 # quantization
 ifeq ($(CONFIG_VP9_ENCODER),yes)
 DSP_SRCS-yes            += quantize.c

vpx_dsp/vpx_dsp_rtcd_defs.pl

@@ -592,6 +592,193 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
 }  # CONFIG_VP9_HIGHBITDEPTH
 }  # CONFIG_VP9_ENCODER
 
+#
+# Inverse transform
+if (vpx_config("CONFIG_VP9") eq "yes") {
+if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
+  # Note as optimized versions of these functions are added we need to add a check to ensure
+  # that when CONFIG_EMULATE_HARDWARE is on, it defaults to the C versions only.
+  add_proto qw/void vp9_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct4x4_1_add/;
+
+  add_proto qw/void vp9_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct4x4_16_add/;
+
+  add_proto qw/void vp9_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct8x8_1_add/;
+
+  add_proto qw/void vp9_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct8x8_64_add/;
+
+  add_proto qw/void vp9_idct8x8_12_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct8x8_12_add/;
+
+  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/;
+
+  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/;
+
+  add_proto qw/void vp9_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct16x16_10_add/;
+
+  add_proto qw/void vp9_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct32x32_1024_add/;
+
+  add_proto qw/void vp9_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct32x32_34_add/;
+
+  add_proto qw/void vp9_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_idct32x32_1_add/;
+
+  add_proto qw/void vp9_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_iwht4x4_1_add/;
+
+  add_proto qw/void vp9_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+  specialize qw/vp9_iwht4x4_16_add/;
+
+  add_proto qw/void vp9_highbd_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_highbd_idct4x4_1_add/;
+
+  add_proto qw/void vp9_highbd_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_highbd_idct8x8_1_add/;
+
+  add_proto qw/void vp9_highbd_idct16x16_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_highbd_idct16x16_1_add/;
+
+  add_proto qw/void vp9_highbd_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_highbd_idct32x32_1024_add/;
+
+  add_proto qw/void vp9_highbd_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_highbd_idct32x32_34_add/;
+
+  add_proto qw/void vp9_highbd_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_highbd_idct32x32_1_add/;
+
+  add_proto qw/void vp9_highbd_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_highbd_iwht4x4_1_add/;
+
+  add_proto qw/void vp9_highbd_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+  specialize qw/vp9_highbd_iwht4x4_16_add/;
+
+  # Force C versions if CONFIG_EMULATE_HARDWARE is 1
+  if (vpx_config("CONFIG_EMULATE_HARDWARE") eq "yes") {
+    add_proto qw/void vp9_highbd_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+    specialize qw/vp9_highbd_idct4x4_16_add/;
+
+    add_proto qw/void vp9_highbd_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+    specialize qw/vp9_highbd_idct8x8_64_add/;
+
+    add_proto qw/void vp9_highbd_idct8x8_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+    specialize qw/vp9_highbd_idct8x8_10_add/;
+
+    add_proto qw/void vp9_highbd_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+    specialize qw/vp9_highbd_idct16x16_256_add/;
+
+    add_proto qw/void vp9_highbd_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+    specialize qw/vp9_highbd_idct16x16_10_add/;
+  } else {
+    add_proto qw/void vp9_highbd_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+    specialize qw/vp9_highbd_idct4x4_16_add sse2/;
+
+    add_proto qw/void vp9_highbd_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+    specialize qw/vp9_highbd_idct8x8_64_add sse2/;
+
+    add_proto qw/void vp9_highbd_idct8x8_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+    specialize qw/vp9_highbd_idct8x8_10_add sse2/;
+
+    add_proto qw/void vp9_highbd_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+    specialize qw/vp9_highbd_idct16x16_256_add sse2/;
+
+    add_proto qw/void vp9_highbd_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
+    specialize qw/vp9_highbd_idct16x16_10_add sse2/;
+  }  # CONFIG_EMULATE_HARDWARE
+} else {
+  # Force C versions if CONFIG_EMULATE_HARDWARE is 1
+  if (vpx_config("CONFIG_EMULATE_HARDWARE") eq "yes") {
+    add_proto qw/void vp9_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct4x4_1_add/;
+
+    add_proto qw/void vp9_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct4x4_16_add/;
+
+    add_proto qw/void vp9_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct8x8_1_add/;
+
+    add_proto qw/void vp9_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct8x8_64_add/;
+
+    add_proto qw/void vp9_idct8x8_12_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct8x8_12_add/;
+
+    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/;
+
+    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/;
+
+    add_proto qw/void vp9_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct16x16_10_add/;
+
+    add_proto qw/void vp9_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct32x32_1024_add/;
+
+    add_proto qw/void vp9_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct32x32_34_add/;
+
+    add_proto qw/void vp9_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct32x32_1_add/;
+
+    add_proto qw/void vp9_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_iwht4x4_1_add/;
+
+    add_proto qw/void vp9_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_iwht4x4_16_add/;
+  } else {
+    add_proto qw/void vp9_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct4x4_1_add sse2 neon dspr2 msa/;
+
+    add_proto qw/void vp9_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct4x4_16_add sse2 neon dspr2 msa/;
+
+    add_proto qw/void vp9_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct8x8_1_add sse2 neon dspr2 msa/;
+
+    add_proto qw/void vp9_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct8x8_64_add sse2 neon dspr2 msa/, "$ssse3_x86_64_x86inc";
+
+    add_proto qw/void vp9_idct8x8_12_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct8x8_12_add sse2 neon dspr2 msa/, "$ssse3_x86_64_x86inc";
+
+    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 dspr2 msa/;
+
+    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 neon dspr2 msa/;
+
+    add_proto qw/void vp9_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct16x16_10_add sse2 neon dspr2 msa/;
+
+    add_proto qw/void vp9_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct32x32_1024_add sse2 neon dspr2 msa/;
+
+    add_proto qw/void vp9_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct32x32_34_add sse2 neon_asm dspr2 msa/;
+    # Need to add 34 eob idct32x32 neon implementation.
+    $vp9_idct32x32_34_add_neon_asm=vp9_idct32x32_1024_add_neon;
+
+    add_proto qw/void vp9_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_idct32x32_1_add sse2 neon dspr2 msa/;
+
+    add_proto qw/void vp9_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_iwht4x4_1_add msa/;
+
+    add_proto qw/void vp9_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
+    specialize qw/vp9_iwht4x4_16_add msa/, "$sse2_x86inc";
+  }  # CONFIG_EMULATE_HARDWARE
+}  # CONFIG_VP9_HIGHBITDEPTH
+}  # CONFIG_VP9
+
 #
 # Quantization
 #

vp9/common/x86/vp9_idct_intrin_sse2.h → vpx_dsp/x86/inv_txfm_sse2.h

 /*
- *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *  Copyright (c) 2015 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
@@ -8,12 +8,13 @@
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
-#include <assert.h>
+#ifndef VPX_DSP_X86_INV_TXFM_SSE2_H_
+#define VPX_DSP_X86_INV_TXFM_SSE2_H_
+
 #include <emmintrin.h>  // SSE2
 #include "./vpx_config.h"
 #include "vpx/vpx_integer.h"
-#include "vp9/common/vp9_common.h"
-#include "vp9/common/vp9_idct.h"
+#include "vpx_dsp/inv_txfm.h"
 
 // perform 8x8 transpose
 static INLINE void array_transpose_8x8(__m128i *in, __m128i *res) {
@@ -172,3 +173,12 @@ static INLINE void write_buffer_8x16(uint8_t *dest, __m128i *in, int stride) {
   RECON_AND_STORE(dest + 14 * stride, in[14]);
   RECON_AND_STORE(dest + 15 * stride, in[15]);
 }
+
+void idct4_sse2(__m128i *in);
+void idct8_sse2(__m128i *in);
+void idct16_sse2(__m128i *in0, __m128i *in1);
+void iadst4_sse2(__m128i *in);
+void iadst8_sse2(__m128i *in);
+void iadst16_sse2(__m128i *in0, __m128i *in1);
+
+#endif  // VPX_DSP_X86_INV_TXFM_SSE2_H_