Merge changes Id5beb35d,I2945fe54,Ib0f3cfd6,I78a2eba8

* changes:
  partial fdct neon: add 32x32_1
  partial fdct neon: add 16x16_1
  partial fdct neon: add 4x4_1
  partial fdct neon: move 8x8_1 and enable hbd tests

test/dct_partial_test.cc

@@ -139,9 +139,23 @@ INSTANTIATE_TEST_CASE_P(
 #endif  // HAVE_SSE2
 
 #if HAVE_NEON
-INSTANTIATE_TEST_CASE_P(NEON, PartialFdctTest,
+#if CONFIG_VP9_HIGHBITDEPTH
-                        ::testing::Values(make_tuple(&vpx_fdct8x8_1_neon, 8,
+INSTANTIATE_TEST_CASE_P(
-                                                     VPX_BITS_8)));
+    NEON, PartialFdctTest,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_1_neon, 32, VPX_BITS_8),
+                      make_tuple(&vpx_fdct16x16_1_neon, 16, VPX_BITS_8),
+                      make_tuple(&vpx_fdct8x8_1_neon, 8, VPX_BITS_12),
+                      make_tuple(&vpx_fdct8x8_1_neon, 8, VPX_BITS_10),
+                      make_tuple(&vpx_fdct8x8_1_neon, 8, VPX_BITS_8),
+                      make_tuple(&vpx_fdct4x4_1_neon, 4, VPX_BITS_8)));
+#else
+INSTANTIATE_TEST_CASE_P(
+    NEON, PartialFdctTest,
+    ::testing::Values(make_tuple(&vpx_fdct32x32_1_neon, 32, VPX_BITS_8),
+                      make_tuple(&vpx_fdct16x16_1_neon, 16, VPX_BITS_8),
+                      make_tuple(&vpx_fdct8x8_1_neon, 8, VPX_BITS_8),
+                      make_tuple(&vpx_fdct4x4_1_neon, 4, VPX_BITS_8)));
+#endif  // CONFIG_VP9_HIGHBITDEPTH
 #endif  // HAVE_NEON
 
 #if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH

vpx_dsp/arm/fdct_partial_neon.c

@@ -0,0 +1,102 @@
+/*
+ *  Copyright (c) 2017 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 "./vpx_dsp_rtcd.h"
+#include "./vpx_config.h"
+#include "vpx_dsp/arm/mem_neon.h"
+
+static INLINE int32_t sum_int16x8(const int16x8_t a) {
+  const int32x4_t b = vpaddlq_s16(a);
+  const int64x2_t c = vpaddlq_s32(b);
+  const int32x2_t d = vadd_s32(vreinterpret_s32_s64(vget_low_s64(c)),
+                               vreinterpret_s32_s64(vget_high_s64(c)));
+  return vget_lane_s32(d, 0);
+}
+
+void vpx_fdct4x4_1_neon(const int16_t *input, tran_low_t *output, int stride) {
+  int16x4_t a0, a1, a2, a3;
+  int16x8_t b0, b1;
+  int16x8_t c;
+
+  a0 = vld1_s16(input);
+  input += stride;
+  a1 = vld1_s16(input);
+  input += stride;
+  a2 = vld1_s16(input);
+  input += stride;
+  a3 = vld1_s16(input);
+
+  b0 = vcombine_s16(a0, a1);
+  b1 = vcombine_s16(a2, a3);
+
+  c = vaddq_s16(b0, b1);
+
+  output[0] = (tran_low_t)(sum_int16x8(c) << 1);
+  output[1] = 0;
+}
+
+void vpx_fdct8x8_1_neon(const int16_t *input, tran_low_t *output, int stride) {
+  int r;
+  int16x8_t sum = vld1q_s16(&input[0]);
+  for (r = 1; r < 8; ++r) {
+    const int16x8_t input_00 = vld1q_s16(&input[r * stride]);
+    sum = vaddq_s16(sum, input_00);
+  }
+
+  output[0] = (tran_low_t)sum_int16x8(sum);
+  output[1] = 0;
+}
+
+void vpx_fdct16x16_1_neon(const int16_t *input, tran_low_t *output,
+                          int stride) {
+  int r;
+  int16x8_t left = vld1q_s16(input);
+  int16x8_t right = vld1q_s16(input + 8);
+  input += stride;
+  for (r = 1; r < 16; ++r) {
+    const int16x8_t a = vld1q_s16(input);
+    const int16x8_t b = vld1q_s16(input + 8);
+    input += stride;
+    left = vaddq_s16(left, a);
+    right = vaddq_s16(right, b);
+  }
+
+  output[0] = (tran_low_t)((sum_int16x8(left) + sum_int16x8(right)) >> 1);
+  output[1] = 0;
+}
+
+void vpx_fdct32x32_1_neon(const int16_t *input, tran_low_t *output,
+                          int stride) {
+  int r;
+  int16x8_t a0 = vld1q_s16(input);
+  int16x8_t a1 = vld1q_s16(input + 8);
+  int16x8_t a2 = vld1q_s16(input + 16);
+  int16x8_t a3 = vld1q_s16(input + 24);
+  input += stride;
+  for (r = 1; r < 32; ++r) {
+    const int16x8_t b0 = vld1q_s16(input);
+    const int16x8_t b1 = vld1q_s16(input + 8);
+    const int16x8_t b2 = vld1q_s16(input + 16);
+    const int16x8_t b3 = vld1q_s16(input + 24);
+    input += stride;
+    a0 = vaddq_s16(a0, b0);
+    a1 = vaddq_s16(a1, b1);
+    a2 = vaddq_s16(a2, b2);
+    a3 = vaddq_s16(a3, b3);
+  }
+
+  // TODO(johannkoenig): sum and shift the values in neon registers.
+  output[0] = (tran_low_t)(
+      (sum_int16x8(a0) + sum_int16x8(a1) + sum_int16x8(a2) + sum_int16x8(a3)) >>
+      3);
+  output[1] = 0;
+}

vpx_dsp/arm/fwd_txfm_neon.c

@@ -207,24 +207,3 @@ void vpx_fdct8x8_neon(const int16_t *input, tran_low_t *final_output,
     store_s16q_to_tran_low(final_output + 7 * 8, input_7);
   }
 }
-
-void vpx_fdct8x8_1_neon(const int16_t *input, tran_low_t *output, int stride) {
-  int r;
-  int16x8_t sum = vld1q_s16(&input[0]);
-  for (r = 1; r < 8; ++r) {
-    const int16x8_t input_00 = vld1q_s16(&input[r * stride]);
-    sum = vaddq_s16(sum, input_00);
-  }
-  {
-    const int32x4_t a = vpaddlq_s16(sum);
-    const int64x2_t b = vpaddlq_s32(a);
-    const int32x2_t c = vadd_s32(vreinterpret_s32_s64(vget_low_s64(b)),
-                                 vreinterpret_s32_s64(vget_high_s64(b)));
-#if CONFIG_VP9_HIGHBITDEPTH
-    output[0] = vget_lane_s32(c, 0);
-#else
-    output[0] = vget_lane_s16(vreinterpret_s16_s32(c), 0);
-#endif
-    output[1] = 0;
-  }
-}

vpx_dsp/vpx_dsp.mk

@@ -196,6 +196,7 @@ DSP_SRCS-$(HAVE_AVX2)   += x86/fwd_dct32x32_impl_avx2.h
 DSP_SRCS-$(HAVE_NEON)   += arm/fdct_neon.c
 DSP_SRCS-$(HAVE_NEON)   += arm/fdct16x16_neon.c
 DSP_SRCS-$(HAVE_NEON)   += arm/fdct32x32_neon.c
+DSP_SRCS-$(HAVE_NEON)   += arm/fdct_partial_neon.c
 DSP_SRCS-$(HAVE_NEON)   += arm/fwd_txfm_neon.c
 DSP_SRCS-$(HAVE_MSA)    += mips/fwd_txfm_msa.h
 DSP_SRCS-$(HAVE_MSA)    += mips/fwd_txfm_msa.c

vpx_dsp/vpx_dsp_rtcd_defs.pl

@@ -487,7 +487,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   specialize qw/vpx_fdct4x4 neon sse2/;
 
   add_proto qw/void vpx_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vpx_fdct4x4_1 sse2/;
+  specialize qw/vpx_fdct4x4_1 sse2 neon/;
 
   add_proto qw/void vpx_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
   specialize qw/vpx_fdct8x8 neon sse2/;
@@ -499,7 +499,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   specialize qw/vpx_fdct16x16 neon sse2/;
 
   add_proto qw/void vpx_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vpx_fdct16x16_1 sse2/;
+  specialize qw/vpx_fdct16x16_1 sse2 neon/;
 
   add_proto qw/void vpx_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
   specialize qw/vpx_fdct32x32 neon sse2/;
@@ -508,7 +508,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   specialize qw/vpx_fdct32x32_rd sse2/;
 
   add_proto qw/void vpx_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vpx_fdct32x32_1 sse2/;
+  specialize qw/vpx_fdct32x32_1 sse2 neon/;
 
   add_proto qw/void vpx_highbd_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
   specialize qw/vpx_highbd_fdct4x4 sse2/;
@@ -517,6 +517,8 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   specialize qw/vpx_highbd_fdct8x8 sse2/;
 
   add_proto qw/void vpx_highbd_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
+  specialize qw/vpx_highbd_fdct8x8_1 neon/;
+  $vpx_highbd_fdct8x8_1_neon=vpx_fdct8x8_1_neon;
 
   add_proto qw/void vpx_highbd_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
   specialize qw/vpx_highbd_fdct16x16 sse2/;
@@ -535,7 +537,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   specialize qw/vpx_fdct4x4 neon sse2 msa/;
 
   add_proto qw/void vpx_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vpx_fdct4x4_1 sse2/;
+  specialize qw/vpx_fdct4x4_1 sse2 neon/;
 
   add_proto qw/void vpx_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
   specialize qw/vpx_fdct8x8 sse2 neon msa/, "$ssse3_x86_64";
@@ -547,7 +549,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   specialize qw/vpx_fdct16x16 neon sse2 msa/;
 
   add_proto qw/void vpx_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vpx_fdct16x16_1 sse2 msa/;
+  specialize qw/vpx_fdct16x16_1 sse2 neon msa/;
 
   add_proto qw/void vpx_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
   specialize qw/vpx_fdct32x32 neon sse2 avx2 msa/;
@@ -556,7 +558,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
   specialize qw/vpx_fdct32x32_rd sse2 avx2 msa/;
 
   add_proto qw/void vpx_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
-  specialize qw/vpx_fdct32x32_1 sse2 msa/;
+  specialize qw/vpx_fdct32x32_1 sse2 neon msa/;
 }  # CONFIG_VP9_HIGHBITDEPTH
 }  # CONFIG_VP9_ENCODER