wasm: Add DC8uv*, VE8uv, and HE8uv functions
BUG=webp:352
Change-Id: Ia0b2b0d5007c12fff201ac94673312420da42a53
diff --git a/src/dsp/dec_wasm.c b/src/dsp/dec_wasm.c
index e13e744..7f2ab94 100644
--- a/src/dsp/dec_wasm.c
+++ b/src/dsp/dec_wasm.c
@@ -32,6 +32,12 @@
return a;
}
+static WEBP_INLINE uint8x16 get_8_bytes(uint8_t* dst) {
+ uint8x16 a;
+ memcpy(&a, dst, 8);
+ return a;
+}
+
static WEBP_INLINE uint8x16 splat_uint8(uint32_t val) {
uint8x16 a;
a[0] = val;
@@ -121,6 +127,83 @@
}
//------------------------------------------------------------------------------
+// Chroma
+
+static WEBP_INLINE uint32_t add_horizontal_8(uint8_t* dst) {
+ const uint8x16 zero = (uint8x16){0};
+ const uint8x16 a = get_8_bytes(dst);
+ const uint16x8 _a_lbw = (uint16x8)__builtin_shufflevector(
+ a, zero, 0, 16, 1, 16, 2, 16, 3, 16, 16, 16, 16, 16, 16, 16, 16, 16);
+ const uint16x8 _a_hbw = (uint16x8)__builtin_shufflevector(
+ a, zero, 4, 16, 5, 16, 6, 16, 7, 16, 16, 16, 16, 16, 16, 16, 16, 16);
+ const uint16x8 sum_a = _a_lbw + _a_hbw;
+ const uint16x8 sum_b = (uint16x8)__builtin_shufflevector(
+ sum_a, sum_a, 2, 3, 2, 3, 2, 3, 2, 3);
+ const uint16x8 sum_c = sum_a + sum_b;
+ const uint16x8 sum_d = (uint16x8)__builtin_shufflevector(
+ sum_c, sum_c, 1, 1, 1, 1, 1, 1, 1, 1);
+ const uint16x8 sum_e = sum_c + sum_d;
+ return sum_e[0] & 0xffff;
+}
+
+static void VE8uv(uint8_t* dst) { // vertical
+ const uint8x16 top = get_8_bytes(dst - BPS);
+ int j;
+ for (j = 0; j < 8; ++j) {
+ memcpy(dst + j * BPS, &top, 8);
+ }
+}
+
+static void HE8uv(uint8_t* dst) { // horizontal
+ int j;
+ for (j = 8; j > 0; --j) {
+ const uint8x16 values = splat_uint8(dst[-1]);
+ memcpy(dst, &values, 8);
+ dst += BPS;
+ }
+}
+
+// helper for chroma-DC predictions
+static WEBP_INLINE void Put8x8uv(uint8_t v, uint8_t* dst) {
+ int j;
+ const uint8x16 values = splat_uint8(v);
+ for (j = 0; j < 8; ++j) {
+ memcpy(dst + j * BPS, &values, 8);
+ }
+}
+
+static void DC8uv(uint8_t* dst) { // DC
+ int left = 0;
+ int j;
+ const uint32_t sum = add_horizontal_8(dst - BPS);
+ for (j = 0; j < 8; ++j) {
+ left += dst[-1 + j * BPS];
+ }
+ {
+ const int DC = sum + left + 8;
+ Put8x8uv(DC >> 4, dst);
+ }
+}
+
+static void DC8uvNoLeft(uint8_t* dst) { // DC with no left samples
+ const uint32_t DC = 4 + add_horizontal_8(dst - BPS);
+ Put8x8uv(DC >> 3, dst);
+}
+
+static void DC8uvNoTop(uint8_t* dst) { // DC with no top samples
+ int dc0 = 4;
+ int i;
+ for (i = 0; i < 8; ++i) {
+ dc0 += dst[-1 + i * BPS];
+ }
+ Put8x8uv(dc0 >> 3, dst);
+}
+
+static void DC8uvNoTopLeft(uint8_t* dst) { // DC with nothing
+ Put8x8uv(0x80, dst);
+}
+
+//------------------------------------------------------------------------------
// Entry point
extern void VP8DspInitWASM(void);
@@ -132,6 +215,13 @@
VP8PredLuma16[4] = DC16NoTop;
VP8PredLuma16[5] = DC16NoLeft;
VP8PredLuma16[6] = DC16NoTopLeft;
+
+ VP8PredChroma8[0] = DC8uv;
+ VP8PredChroma8[2] = VE8uv;
+ VP8PredChroma8[3] = HE8uv;
+ VP8PredChroma8[4] = DC8uvNoTop;
+ VP8PredChroma8[5] = DC8uvNoLeft;
+ VP8PredChroma8[6] = DC8uvNoTopLeft;
}
#else // !WEBP_USE_WASM
