Merge "Isolate vp10's fwd_txfm from vp9"

vp10/common/vp10_fwd_txfm.h

+/*
+ *  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
+ *  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 VP10_COMMON_VP10_FWD_TXFM_H_
+#define VP10_COMMON_VP10_FWD_TXFM_H_
+
+#include "vpx_dsp/txfm_common.h"
+#include "vpx_dsp/fwd_txfm.h"
+
+void vp10_fdct32(const tran_high_t *input, tran_high_t *output, int round);
+#endif  // VP10_COMMON_VP10_FWD_TXFM_H_

vp10/common/vp10_rtcd_defs.pl

@@ -95,6 +95,57 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
 
   add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
   specialize qw/vp10_iht16x16_256_add/;
+
+  add_proto qw/void vp10_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_fdct4x4 sse2/;
+
+  add_proto qw/void vp10_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_fdct4x4_1 sse2/;
+
+  add_proto qw/void vp10_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_fdct8x8 sse2/;
+
+  add_proto qw/void vp10_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_fdct8x8_1 sse2/;
+
+  add_proto qw/void vp10_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_fdct16x16 sse2/;
+
+  add_proto qw/void vp10_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_fdct16x16_1 sse2/;
+
+  add_proto qw/void vp10_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_fdct32x32 sse2/;
+
+  add_proto qw/void vp10_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_fdct32x32_rd sse2/;
+
+  add_proto qw/void vp10_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_fdct32x32_1 sse2/;
+
+  add_proto qw/void vp10_highbd_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_highbd_fdct4x4 sse2/;
+
+  add_proto qw/void vp10_highbd_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_highbd_fdct8x8 sse2/;
+
+  add_proto qw/void vp10_highbd_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_highbd_fdct8x8_1/;
+
+  add_proto qw/void vp10_highbd_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_highbd_fdct16x16 sse2/;
+
+  add_proto qw/void vp10_highbd_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_highbd_fdct16x16_1/;
+
+  add_proto qw/void vp10_highbd_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_highbd_fdct32x32 sse2/;
+
+  add_proto qw/void vp10_highbd_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_highbd_fdct32x32_rd sse2/;
+
+  add_proto qw/void vp10_highbd_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vp10_highbd_fdct32x32_1/;
 } else {
   # Force C versions if CONFIG_EMULATE_HARDWARE is 1
   if (vpx_config("CONFIG_EMULATE_HARDWARE") eq "yes") {
@@ -106,6 +157,33 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
 
     add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
     specialize qw/vp10_iht16x16_256_add/;
+
+    add_proto qw/void vp10_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct4x4/;
+
+    add_proto qw/void vp10_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct4x4_1/;
+
+    add_proto qw/void vp10_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct8x8/;
+
+    add_proto qw/void vp10_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct8x8_1/;
+
+    add_proto qw/void vp10_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct16x16/;
+
+    add_proto qw/void vp10_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct16x16_1/;
+
+    add_proto qw/void vp10_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct32x32/;
+
+    add_proto qw/void vp10_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct32x32_rd/;
+
+    add_proto qw/void vp10_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct32x32_1/;
   } else {
     add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
     specialize qw/vp10_iht4x4_16_add sse2 neon dspr2 msa/;
@@ -115,6 +193,33 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
 
     add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
     specialize qw/vp10_iht16x16_256_add sse2 dspr2 msa/;
+
+    add_proto qw/void vp10_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct4x4 sse2/;
+
+    add_proto qw/void vp10_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct4x4_1 sse2/;
+
+    add_proto qw/void vp10_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct8x8 sse2/;
+
+    add_proto qw/void vp10_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct8x8_1 sse2/;
+
+    add_proto qw/void vp10_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct16x16 sse2/;
+
+    add_proto qw/void vp10_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct16x16_1 sse2/;
+
+    add_proto qw/void vp10_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct32x32 sse2/;
+
+    add_proto qw/void vp10_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct32x32_rd sse2/;
+
+    add_proto qw/void vp10_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
+    specialize qw/vp10_fdct32x32_1 sse2/;
   }
 }
 

vp10/common/x86/vp10_fwd_txfm_sse2.c

+/*
+ *  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
+ *  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 <emmintrin.h>  // SSE2
+
+#include "./vpx_config.h"
+#include "vpx_dsp/vpx_dsp_common.h"
+#include "vpx_dsp/x86/fwd_txfm_sse2.h"
+
+void vp10_fdct4x4_1_sse2(const int16_t *input, tran_low_t *output, int stride) {
+  __m128i in0, in1;
+  __m128i tmp;
+  const __m128i zero = _mm_setzero_si128();
+  in0  = _mm_loadl_epi64((const __m128i *)(input +  0 * stride));
+  in1  = _mm_loadl_epi64((const __m128i *)(input +  1 * stride));
+  in1  = _mm_unpacklo_epi64(in1, _mm_loadl_epi64((const __m128i *)
+         (input +  2 * stride)));
+  in0  = _mm_unpacklo_epi64(in0, _mm_loadl_epi64((const __m128i *)
+         (input +  3 * stride)));
+
+  tmp = _mm_add_epi16(in0, in1);
+  in0 = _mm_unpacklo_epi16(zero, tmp);
+  in1 = _mm_unpackhi_epi16(zero, tmp);
+  in0 = _mm_srai_epi32(in0, 16);
+  in1 = _mm_srai_epi32(in1, 16);
+
+  tmp = _mm_add_epi32(in0, in1);
+  in0 = _mm_unpacklo_epi32(tmp, zero);
+  in1 = _mm_unpackhi_epi32(tmp, zero);
+
+  tmp = _mm_add_epi32(in0, in1);
+  in0 = _mm_srli_si128(tmp, 8);
+
+  in1 = _mm_add_epi32(tmp, in0);
+  in0 = _mm_slli_epi32(in1, 1);
+  store_output(&in0, output);
+}
+
+void vp10_fdct8x8_1_sse2(const int16_t *input, tran_low_t *output, int stride) {
+  __m128i in0  = _mm_load_si128((const __m128i *)(input + 0 * stride));
+  __m128i in1  = _mm_load_si128((const __m128i *)(input + 1 * stride));
+  __m128i in2  = _mm_load_si128((const __m128i *)(input + 2 * stride));
+  __m128i in3  = _mm_load_si128((const __m128i *)(input + 3 * stride));
+  __m128i u0, u1, sum;
+
+  u0 = _mm_add_epi16(in0, in1);
+  u1 = _mm_add_epi16(in2, in3);
+
+  in0  = _mm_load_si128((const __m128i *)(input + 4 * stride));
+  in1  = _mm_load_si128((const __m128i *)(input + 5 * stride));
+  in2  = _mm_load_si128((const __m128i *)(input + 6 * stride));
+  in3  = _mm_load_si128((const __m128i *)(input + 7 * stride));
+
+  sum = _mm_add_epi16(u0, u1);
+
+  in0 = _mm_add_epi16(in0, in1);
+  in2 = _mm_add_epi16(in2, in3);
+  sum = _mm_add_epi16(sum, in0);
+
+  u0  = _mm_setzero_si128();
+  sum = _mm_add_epi16(sum, in2);
+
+  in0 = _mm_unpacklo_epi16(u0, sum);
+  in1 = _mm_unpackhi_epi16(u0, sum);
+  in0 = _mm_srai_epi32(in0, 16);
+  in1 = _mm_srai_epi32(in1, 16);
+
+  sum = _mm_add_epi32(in0, in1);
+  in0 = _mm_unpacklo_epi32(sum, u0);
+  in1 = _mm_unpackhi_epi32(sum, u0);
+
+  sum = _mm_add_epi32(in0, in1);
+  in0 = _mm_srli_si128(sum, 8);
+
+  in1 = _mm_add_epi32(sum, in0);
+  store_output(&in1, output);
+}
+
+void vp10_fdct16x16_1_sse2(const int16_t *input, tran_low_t *output,
+                          int stride) {
+  __m128i in0, in1, in2, in3;
+  __m128i u0, u1;
+  __m128i sum = _mm_setzero_si128();
+  int i;
+
+  for (i = 0; i < 2; ++i) {
+    input += 8 * i;
+    in0  = _mm_load_si128((const __m128i *)(input +  0 * stride));
+    in1  = _mm_load_si128((const __m128i *)(input +  1 * stride));
+    in2  = _mm_load_si128((const __m128i *)(input +  2 * stride));
+    in3  = _mm_load_si128((const __m128i *)(input +  3 * stride));
+
+    u0 = _mm_add_epi16(in0, in1);
+    u1 = _mm_add_epi16(in2, in3);
+    sum = _mm_add_epi16(sum, u0);
+
+    in0  = _mm_load_si128((const __m128i *)(input +  4 * stride));
+    in1  = _mm_load_si128((const __m128i *)(input +  5 * stride));
+    in2  = _mm_load_si128((const __m128i *)(input +  6 * stride));
+    in3  = _mm_load_si128((const __m128i *)(input +  7 * stride));
+
+    sum = _mm_add_epi16(sum, u1);
+    u0  = _mm_add_epi16(in0, in1);
+    u1  = _mm_add_epi16(in2, in3);
+    sum = _mm_add_epi16(sum, u0);
+
+    in0  = _mm_load_si128((const __m128i *)(input +  8 * stride));
+    in1  = _mm_load_si128((const __m128i *)(input +  9 * stride));
+    in2  = _mm_load_si128((const __m128i *)(input + 10 * stride));
+    in3  = _mm_load_si128((const __m128i *)(input + 11 * stride));
+
+    sum = _mm_add_epi16(sum, u1);
+    u0  = _mm_add_epi16(in0, in1);
+    u1  = _mm_add_epi16(in2, in3);
+    sum = _mm_add_epi16(sum, u0);
+
+    in0  = _mm_load_si128((const __m128i *)(input + 12 * stride));
+    in1  = _mm_load_si128((const __m128i *)(input + 13 * stride));
+    in2  = _mm_load_si128((const __m128i *)(input + 14 * stride));
+    in3  = _mm_load_si128((const __m128i *)(input + 15 * stride));
+
+    sum = _mm_add_epi16(sum, u1);
+    u0  = _mm_add_epi16(in0, in1);
+    u1  = _mm_add_epi16(in2, in3);
+    sum = _mm_add_epi16(sum, u0);
+
+    sum = _mm_add_epi16(sum, u1);
+  }
+
+  u0  = _mm_setzero_si128();
+  in0 = _mm_unpacklo_epi16(u0, sum);
+  in1 = _mm_unpackhi_epi16(u0, sum);
+  in0 = _mm_srai_epi32(in0, 16);
+  in1 = _mm_srai_epi32(in1, 16);
+
+  sum = _mm_add_epi32(in0, in1);
+  in0 = _mm_unpacklo_epi32(sum, u0);
+  in1 = _mm_unpackhi_epi32(sum, u0);
+
+  sum = _mm_add_epi32(in0, in1);
+  in0 = _mm_srli_si128(sum, 8);
+
+  in1 = _mm_add_epi32(sum, in0);
+  in1 = _mm_srai_epi32(in1, 1);
+  store_output(&in1, output);
+}
+
+void vp10_fdct32x32_1_sse2(const int16_t *input, tran_low_t *output,
+                          int stride) {
+  __m128i in0, in1, in2, in3;
+  __m128i u0, u1;
+  __m128i sum = _mm_setzero_si128();
+  int i;
+
+  for (i = 0; i < 8; ++i) {
+    in0  = _mm_load_si128((const __m128i *)(input +  0));
+    in1  = _mm_load_si128((const __m128i *)(input +  8));
+    in2  = _mm_load_si128((const __m128i *)(input + 16));
+    in3  = _mm_load_si128((const __m128i *)(input + 24));
+
+    input += stride;
+    u0 = _mm_add_epi16(in0, in1);
+    u1 = _mm_add_epi16(in2, in3);
+    sum = _mm_add_epi16(sum, u0);
+
+    in0  = _mm_load_si128((const __m128i *)(input +  0));
+    in1  = _mm_load_si128((const __m128i *)(input +  8));
+    in2  = _mm_load_si128((const __m128i *)(input + 16));
+    in3  = _mm_load_si128((const __m128i *)(input + 24));
+
+    input += stride;
+    sum = _mm_add_epi16(sum, u1);
+    u0  = _mm_add_epi16(in0, in1);
+    u1  = _mm_add_epi16(in2, in3);
+    sum = _mm_add_epi16(sum, u0);
+
+    in0  = _mm_load_si128((const __m128i *)(input +  0));
+    in1  = _mm_load_si128((const __m128i *)(input +  8));
+    in2  = _mm_load_si128((const __m128i *)(input + 16));
+    in3  = _mm_load_si128((const __m128i *)(input + 24));
+
+    input += stride;
+    sum = _mm_add_epi16(sum, u1);
+    u0  = _mm_add_epi16(in0, in1);
+    u1  = _mm_add_epi16(in2, in3);
+    sum = _mm_add_epi16(sum, u0);
+
+    in0  = _mm_load_si128((const __m128i *)(input +  0));
+    in1  = _mm_load_si128((const __m128i *)(input +  8));
+    in2  = _mm_load_si128((const __m128i *)(input + 16));
+    in3  = _mm_load_si128((const __m128i *)(input + 24));
+
+    input += stride;
+    sum = _mm_add_epi16(sum, u1);
+    u0  = _mm_add_epi16(in0, in1);
+    u1  = _mm_add_epi16(in2, in3);
+    sum = _mm_add_epi16(sum, u0);
+
+    sum = _mm_add_epi16(sum, u1);
+  }
+
+  u0  = _mm_setzero_si128();
+  in0 = _mm_unpacklo_epi16(u0, sum);
+  in1 = _mm_unpackhi_epi16(u0, sum);
+  in0 = _mm_srai_epi32(in0, 16);
+  in1 = _mm_srai_epi32(in1, 16);
+
+  sum = _mm_add_epi32(in0, in1);
+  in0 = _mm_unpacklo_epi32(sum, u0);
+  in1 = _mm_unpackhi_epi32(sum, u0);
+
+  sum = _mm_add_epi32(in0, in1);
+  in0 = _mm_srli_si128(sum, 8);
+
+  in1 = _mm_add_epi32(sum, in0);
+  in1 = _mm_srai_epi32(in1, 3);
+  store_output(&in1, output);
+}
+
+#define DCT_HIGH_BIT_DEPTH 0
+#define FDCT4x4_2D vp10_fdct4x4_sse2
+#define FDCT8x8_2D vp10_fdct8x8_sse2
+#define FDCT16x16_2D vp10_fdct16x16_sse2
+#include "vp10/common/x86/vp10_fwd_txfm_impl_sse2.h"
+#undef  FDCT4x4_2D
+#undef  FDCT8x8_2D
+#undef  FDCT16x16_2D
+
+#define FDCT32x32_2D vp10_fdct32x32_rd_sse2
+#define FDCT32x32_HIGH_PRECISION 0
+#include "vp10/common/x86/vp10_fwd_dct32x32_impl_sse2.h"
+#undef  FDCT32x32_2D
+#undef  FDCT32x32_HIGH_PRECISION
+
+#define FDCT32x32_2D vp10_fdct32x32_sse2
+#define FDCT32x32_HIGH_PRECISION 1
+#include "vp10/common/x86/vp10_fwd_dct32x32_impl_sse2.h"  // NOLINT
+#undef  FDCT32x32_2D
+#undef  FDCT32x32_HIGH_PRECISION
+#undef  DCT_HIGH_BIT_DEPTH
+
+#if CONFIG_VP9_HIGHBITDEPTH
+#define DCT_HIGH_BIT_DEPTH 1
+#define FDCT4x4_2D vp10_highbd_fdct4x4_sse2
+#define FDCT8x8_2D vp10_highbd_fdct8x8_sse2
+#define FDCT16x16_2D vp10_highbd_fdct16x16_sse2
+#include "vp10/common/x86/vp10_fwd_txfm_impl_sse2.h" // NOLINT
+#undef  FDCT4x4_2D
+#undef  FDCT8x8_2D
+#undef  FDCT16x16_2D
+
+#define FDCT32x32_2D vp10_highbd_fdct32x32_rd_sse2
+#define FDCT32x32_HIGH_PRECISION 0
+#include "vp10/common/x86/vp10_fwd_dct32x32_impl_sse2.h" // NOLINT
+#undef  FDCT32x32_2D
+#undef  FDCT32x32_HIGH_PRECISION
+
+#define FDCT32x32_2D vp10_highbd_fdct32x32_sse2
+#define FDCT32x32_HIGH_PRECISION 1
+#include "vp10/common/x86/vp10_fwd_dct32x32_impl_sse2.h" // NOLINT
+#undef  FDCT32x32_2D
+#undef  FDCT32x32_HIGH_PRECISION
+#undef  DCT_HIGH_BIT_DEPTH
+#endif  // CONFIG_VP9_HIGHBITDEPTH

vp10/vp10_common.mk

@@ -61,6 +61,8 @@ VP10_COMMON_SRCS-$(CONFIG_POSTPROC_VISUALIZER) += common/textblit.c
 VP10_COMMON_SRCS-yes += common/common_data.h
 VP10_COMMON_SRCS-yes += common/scan.c
 VP10_COMMON_SRCS-yes += common/scan.h
+VP10_COMMON_SRCS-yes += common/vp10_fwd_txfm.h
+VP10_COMMON_SRCS-yes += common/vp10_fwd_txfm.c
 
 VP10_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/postproc.h
 VP10_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/postproc.c
@@ -87,6 +89,9 @@ VP10_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/mfqe_msa.c
 endif
 
 VP10_COMMON_SRCS-$(HAVE_SSE2) += common/x86/idct_intrin_sse2.c
+VP10_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp10_fwd_txfm_sse2.c
+VP10_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp10_fwd_dct32x32_impl_sse2.h
+VP10_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp10_fwd_txfm_impl_sse2.h
 
 ifneq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
 VP10_COMMON_SRCS-$(HAVE_NEON) += common/arm/neon/iht4x4_add_neon.c