libwebp/src/dsp/cost_neon.c

123 lines
4.2 KiB
C
Raw Normal View History

// 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.
// -----------------------------------------------------------------------------
//
// ARM NEON version of cost functions
#include "src/dsp/dsp.h"
#if defined(WEBP_USE_NEON)
#include "src/dsp/neon.h"
#include "src/enc/cost_enc.h"
static const uint8_t position[16] = { 1, 2, 3, 4, 5, 6, 7, 8,
9, 10, 11, 12, 13, 14, 15, 16 };
static void SetResidualCoeffs_NEON(const int16_t* const coeffs,
VP8Residual* const res) {
const int16x8_t minus_one = vdupq_n_s16(-1);
const int16x8_t coeffs_0 = vld1q_s16(coeffs);
const int16x8_t coeffs_1 = vld1q_s16(coeffs + 8);
const uint16x8_t eob_0 = vtstq_s16(coeffs_0, minus_one);
const uint16x8_t eob_1 = vtstq_s16(coeffs_1, minus_one);
const uint8x16_t eob = vcombine_u8(vqmovn_u16(eob_0), vqmovn_u16(eob_1));
const uint8x16_t masked = vandq_u8(eob, vld1q_u8(position));
#ifdef __aarch64__
res->last = vmaxvq_u8(masked) - 1;
#else
const uint8x8_t eob_8x8 = vmax_u8(vget_low_u8(masked), vget_high_u8(masked));
const uint16x8_t eob_16x8 = vmovl_u8(eob_8x8);
const uint16x4_t eob_16x4 =
vmax_u16(vget_low_u16(eob_16x8), vget_high_u16(eob_16x8));
const uint32x4_t eob_32x4 = vmovl_u16(eob_16x4);
uint32x2_t eob_32x2 =
vmax_u32(vget_low_u32(eob_32x4), vget_high_u32(eob_32x4));
eob_32x2 = vpmax_u32(eob_32x2, eob_32x2);
vst1_lane_s32(&res->last, vreinterpret_s32_u32(eob_32x2), 0);
--res->last;
#endif // __aarch64__
res->coeffs = coeffs;
}
static int GetResidualCost_NEON(int ctx0, const VP8Residual* const res) {
uint8_t levels[16], ctxs[16];
uint16_t abs_levels[16];
int n = res->first;
// should be prob[VP8EncBands[n]], but it's equivalent for n=0 or 1
const int p0 = res->prob[n][ctx0][0];
CostArrayPtr const costs = res->costs;
const uint16_t* t = costs[n][ctx0];
// bit_cost(1, p0) is already incorporated in t[] tables, but only if ctx != 0
// (as required by the syntax). For ctx0 == 0, we need to add it here or it'll
// be missing during the loop.
int cost = (ctx0 == 0) ? VP8BitCost(1, p0) : 0;
if (res->last < 0) {
return VP8BitCost(0, p0);
}
{ // precompute clamped levels and contexts, packed to 8b.
const uint8x16_t kCst2 = vdupq_n_u8(2);
const uint8x16_t kCst67 = vdupq_n_u8(MAX_VARIABLE_LEVEL);
const int16x8_t c0 = vld1q_s16(res->coeffs);
const int16x8_t c1 = vld1q_s16(res->coeffs + 8);
const uint16x8_t E0 = vreinterpretq_u16_s16(vabsq_s16(c0));
const uint16x8_t E1 = vreinterpretq_u16_s16(vabsq_s16(c1));
const uint8x16_t F = vcombine_u8(vqmovn_u16(E0), vqmovn_u16(E1));
const uint8x16_t G = vminq_u8(F, kCst2); // context = 0,1,2
const uint8x16_t H = vminq_u8(F, kCst67); // clamp_level in [0..67]
vst1q_u8(ctxs, G);
vst1q_u8(levels, H);
vst1q_u16(abs_levels, E0);
vst1q_u16(abs_levels + 8, E1);
}
for (; n < res->last; ++n) {
const int ctx = ctxs[n];
const int level = levels[n];
const int flevel = abs_levels[n]; // full level
cost += VP8LevelFixedCosts[flevel] + t[level]; // simplified VP8LevelCost()
t = costs[n + 1][ctx];
}
// Last coefficient is always non-zero
{
const int level = levels[n];
const int flevel = abs_levels[n];
assert(flevel != 0);
cost += VP8LevelFixedCosts[flevel] + t[level];
if (n < 15) {
const int b = VP8EncBands[n + 1];
const int ctx = ctxs[n];
const int last_p0 = res->prob[b][ctx][0];
cost += VP8BitCost(0, last_p0);
}
}
return cost;
}
//------------------------------------------------------------------------------
// Entry point
extern void VP8EncDspCostInitNEON(void);
WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspCostInitNEON(void) {
VP8SetResidualCoeffs = SetResidualCoeffs_NEON;
VP8GetResidualCost = GetResidualCost_NEON;
}
#else // !WEBP_USE_NEON
WEBP_DSP_INIT_STUB(VP8EncDspCostInitNEON)
#endif // WEBP_USE_NEON