From a37a7b00d5416a25ed33b2a3e83c85b4aeda43cc Mon Sep 17 00:00:00 2001
From: Scott LaVarnway <slavarnway@google.com>
Date: Wed, 28 Jun 2017 05:09:14 -0700
Subject: [PATCH] wasm: Add DC16*, VE16, and HE16 functions

BUG=webp:352

Change-Id: Ia003257d00c2c2ea16a6e6344671237e78c0eac6
---
 src/dsp/dec_wasm.c | 112 +++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 112 insertions(+)

diff --git a/src/dsp/dec_wasm.c b/src/dsp/dec_wasm.c
index 57879136..e13e7446 100644
--- a/src/dsp/dec_wasm.c
+++ b/src/dsp/dec_wasm.c
@@ -9,17 +9,129 @@
 //
 // WebAssembly (WASM) version of some decoding functions.
 //
+// Based on dec_sse2.c
 
 #include "./dsp.h"
+#include "../dec/vp8i_dec.h"
 
 #if defined(WEBP_USE_WASM)
 
+typedef int32_t int32x4 __attribute__((__vector_size__(16)));
+typedef uint32_t uint32x4 __attribute__((__vector_size__(16)));
+typedef int16_t int16x8 __attribute__((__vector_size__(16)));
+typedef uint16_t uint16x8 __attribute__((__vector_size__(16)));
+typedef int8_t  int8x16 __attribute__((__vector_size__(16)));
+typedef uint8_t  uint8x16 __attribute__((__vector_size__(16)));
+
+//------------------------------------------------------------------------------
+//
+
+static WEBP_INLINE uint8x16 get_16_bytes(uint8_t* dst) {
+  uint8x16 a;
+  memcpy(&a, dst, 16);
+  return a;
+}
+
+static WEBP_INLINE uint8x16 splat_uint8(uint32_t val) {
+  uint8x16 a;
+  a[0] = val;
+  a = (uint8x16)__builtin_shufflevector(
+      a, a, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
+  return a;
+}
+
+//------------------------------------------------------------------------------
+// Luma 16x16
+
+static WEBP_INLINE void Put16(uint8_t v, uint8_t* dst) {
+  int j;
+  const uint8x16 values = splat_uint8(v);
+  for (j = 0; j < 16; ++j) {
+    memcpy(dst, &values, 16);
+    dst += BPS;
+  }
+}
+
+static void VE16(uint8_t* dst) {
+  const uint8x16 top = get_16_bytes(dst - BPS);
+  int j;
+  for (j = 0; j < 16; ++j) {
+    memcpy(dst + j * BPS, &top, 16);
+  }
+}
+
+static void HE16(uint8_t* dst) {     // horizontal
+  int j;
+  for (j = 16; j > 0; --j) {
+    const uint8x16 values = splat_uint8(dst[-1]);
+    memcpy(dst, &values, 16);
+    dst += BPS;
+  }
+}
+
+static WEBP_INLINE uint32_t add_horizontal_16(uint8_t* dst) {
+  const uint8x16 zero = (uint8x16){0};
+  const uint8x16 a = get_16_bytes(dst);
+  const uint16x8 _a_lbw = (uint16x8)__builtin_shufflevector(
+      a, zero, 0, 16, 1, 16, 2, 16, 3, 16, 4, 16, 5, 16, 6, 16, 7, 16);
+  const uint16x8 _a_hbw = (uint16x8)__builtin_shufflevector(
+      a, zero, 8, 16, 9, 16, 10, 16, 11, 16, 12, 16, 13, 16, 14, 16, 15, 16);
+  const uint16x8 sum_a = _a_lbw + _a_hbw;
+  const uint16x8 sum_b = (uint16x8)__builtin_shufflevector(
+      sum_a, sum_a, 4, 5, 6, 7, 4, 5, 6, 7);
+  const uint16x8 sum_c = sum_a + sum_b;
+  const uint16x8 sum_d = (uint16x8)__builtin_shufflevector(
+      sum_c, sum_c, 2, 3, 2, 3, 2, 3, 2, 3);
+  const uint16x8 sum_e = sum_c + sum_d;
+  const uint16x8 sum_f = (uint16x8)__builtin_shufflevector(
+      sum_e, sum_e, 1, 1, 1, 1, 1, 1, 1, 1);
+  const uint16x8 sum_g = sum_e + sum_f;
+  return sum_g[0] & 0xffff;
+}
+
+static void DC16(uint8_t* dst) {    // DC
+  const uint32_t sum = add_horizontal_16(dst - BPS);
+  int left = 0;
+  int j;
+  for (j = 0; j < 16; ++j) {
+    left += dst[-1 + j * BPS];
+  }
+  {
+    const int DC = sum + left + 16;
+    Put16(DC >> 5, dst);
+  }
+}
+
+static void DC16NoTop(uint8_t* dst) {   // DC with top samples not available
+  int DC = 8;
+  int j;
+  for (j = 0; j < 16; ++j) {
+    DC += dst[-1 + j * BPS];
+  }
+  Put16(DC >> 4, dst);
+}
+
+static void DC16NoLeft(uint8_t* dst) {  // DC with left samples not available
+  const int DC = 8 + add_horizontal_16(dst - BPS);
+  Put16(DC >> 4, dst);
+}
+
+static void DC16NoTopLeft(uint8_t* dst) {  // DC with no top and left samples
+  Put16(0x80, dst);
+}
+
 //------------------------------------------------------------------------------
 // Entry point
 
 extern void VP8DspInitWASM(void);
 
 WEBP_TSAN_IGNORE_FUNCTION void VP8DspInitWASM(void) {
+  VP8PredLuma16[0] = DC16;
+  VP8PredLuma16[2] = VE16;
+  VP8PredLuma16[3] = HE16;
+  VP8PredLuma16[4] = DC16NoTop;
+  VP8PredLuma16[5] = DC16NoLeft;
+  VP8PredLuma16[6] = DC16NoTopLeft;
 }
 
 #else  // !WEBP_USE_WASM