neon IsFlat

Move IsFlat to its own header. This allows it to continue to be
inlined. Using the RTCD and creating a distinct function slows down arm
builds.

   flower   mug
C    3.59  2.12
NEON 3.47  2.01

BUG=b/118740850

Change-Id: Id77e8f76d9e9790c498806e7070bbe37c10bc2e9

3 files changed, 72 insertions, 14 deletions

diff --git a/src/dsp/Makefile.am b/src/dsp/Makefile.am
index 8cd3e53c..9f67f5b3 100644
--- a/src/dsp/Makefile.am
+++ b/src/dsp/Makefile.am
@@ -40,6 +40,7 @@ ENC_SOURCES =
 ENC_SOURCES += cost.c
 ENC_SOURCES += enc.c
 ENC_SOURCES += lossless_enc.c
+ENC_SOURCES += quant.h
 ENC_SOURCES += ssim.c
 
 libwebpdspdecode_sse41_la_SOURCES =
diff --git a/src/dsp/quant.h b/src/dsp/quant.h
new file mode 100644
index 00000000..5ba6f9c3
--- /dev/null
+++ b/src/dsp/quant.h
@@ -0,0 +1,70 @@
+// Copyright 2018 Google Inc. All Rights Reserved.
+//
+// Use of this source code is governed by a BSD-style license
+// that can be found in the COPYING 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 WEBP_DSP_QUANT_H_
+#define WEBP_DSP_QUANT_H_
+
+#include "src/dsp/dsp.h"
+#include "src/webp/types.h"
+
+#if defined(WEBP_USE_NEON) && !defined(WEBP_ANDROID_NEON) && \
+    !defined(WEBP_HAVE_NEON_RTCD)
+#include <arm_neon.h>
+
+#define IsFlat IsFlat_NEON
+
+static uint32x2_t horizontal_add_uint32x4(const uint32x4_t a) {
+  const uint64x2_t b = vpaddlq_u32(a);
+  return vadd_u32(vreinterpret_u32_u64(vget_low_u64(b)),
+                  vreinterpret_u32_u64(vget_high_u64(b)));
+}
+
+static WEBP_INLINE int IsFlat(const int16_t* levels, int num_blocks,
+                              int thresh) {
+  const int16x8_t tst_ones = vdupq_n_s16(-1);
+  uint32x4_t sum = vdupq_n_u32(0);
+
+  for (int i = 0; i < num_blocks; ++i) {
+    // Set DC to zero.
+    const int16x8_t a_0 = vsetq_lane_s16(0, vld1q_s16(levels), 0);
+    const int16x8_t a_1 = vld1q_s16(levels + 8);
+
+    const uint16x8_t b_0 = vshrq_n_u16(vtstq_s16(a_0, tst_ones), 15);
+    const uint16x8_t b_1 = vshrq_n_u16(vtstq_s16(a_1, tst_ones), 15);
+
+    sum = vpadalq_u16(sum, b_0);
+    sum = vpadalq_u16(sum, b_1);
+
+    levels += 16;
+  }
+  return thresh >= (int32_t)vget_lane_u32(horizontal_add_uint32x4(sum), 0);
+}
+
+#else
+
+#define IsFlat IsFlat_C
+
+static WEBP_INLINE int IsFlat(const int16_t* levels, int num_blocks,
+                              int thresh) {
+  int score = 0;
+  while (num_blocks-- > 0) {      // TODO(skal): refine positional scoring?
+    int i;
+    for (i = 1; i < 16; ++i) {    // omit DC, we're only interested in AC
+      score += (levels[i] != 0);
+      if (score > thresh) return 0;
+    }
+    levels += 16;
+  }
+  return 1;
+}
+
+#endif  // defined(WEBP_USE_NEON) && !defined(WEBP_ANDROID_NEON) &&
+        // !defined(WEBP_HAVE_NEON_RTCD)
+
+#endif  // WEBP_DSP_QUANT_H_
diff --git a/src/enc/quant_enc.c b/src/enc/quant_enc.c
index a02ff405..03c682e3 100644
--- a/src/enc/quant_enc.c
+++ b/src/enc/quant_enc.c
@@ -15,6 +15,7 @@
 #include <math.h>
 #include <stdlib.h>  // for abs()
 
+#include "src/dsp/quant.h"
 #include "src/enc/vp8i_enc.h"
 #include "src/enc/cost_enc.h"
 
@@ -977,20 +978,6 @@ static void SwapOut(VP8EncIterator* const it) {
   SwapPtr(&it->yuv_out_, &it->yuv_out2_);
 }
 
-static WEBP_INLINE int IsFlat(const int16_t* levels, int num_blocks,
-                              int thresh) {
-  int score = 0;
-  while (num_blocks-- > 0) {      // TODO(skal): refine positional scoring?
-    int i;
-    for (i = 1; i < 16; ++i) {    // omit DC, we're only interested in AC
-      score += (levels[i] != 0);
-      if (score > thresh) return 0;
-    }
-    levels += 16;
-  }
-  return 1;
-}
-
 static void PickBestIntra16(VP8EncIterator* const it, VP8ModeScore* rd) {
   const int kNumBlocks = 16;
   VP8SegmentInfo* const dqm = &it->enc_->dqm_[it->mb_->segment_];