Commit 1c122c24 authored by Thomas's avatar Thomas Committed by Yaowu Xu

Add quant and dequant functions for new quant matrices.

Change-Id: If0ba62428216fa343b9a37a3b349edba4103c00a
parent 95a89994
This diff is collapsed.
......@@ -712,8 +712,8 @@ static void dec_build_inter_predictors_sb(VP10Decoder *const pbi,
}
}
static INLINE TX_SIZE dec_get_uv_tx_size(const MB_MODE_INFO *mbmi, int n4_wl,
int n4_hl) {
static INLINE TX_SIZE
dec_get_uv_tx_size(const MB_MODE_INFO *mbmi, int n4_wl, int n4_hl) {
// get minimum log2 num4x4s dimension
const int x = VPXMIN(n4_wl, n4_hl);
return VPXMIN(mbmi->tx_size, x);
......@@ -1120,8 +1120,13 @@ static void setup_quantization(VP10_COMMON *const cm,
static void setup_segmentation_dequant(VP10_COMMON *const cm) {
// Build y/uv dequant values based on segmentation.
int i = 0;
#if CONFIG_AOM_QM
int lossless;
int j = 0;
int qmindex;
#endif
if (cm->seg.enabled) {
int i;
for (i = 0; i < MAX_SEGMENTS; ++i) {
const int qindex = vp10_get_qindex(&cm->seg, i, cm->base_qindex);
cm->y_dequant[i][0] =
......@@ -1131,6 +1136,19 @@ static void setup_segmentation_dequant(VP10_COMMON *const cm) {
vp10_dc_quant(qindex, cm->uv_dc_delta_q, cm->bit_depth);
cm->uv_dequant[i][1] =
vp10_ac_quant(qindex, cm->uv_ac_delta_q, cm->bit_depth);
#if CONFIG_AOM_QM
lossless = qindex == 0 && cm->y_dc_delta_q == 0 &&
cm->uv_dc_delta_q == 0 && cm->uv_ac_delta_q == 0;
// NB: depends on base index so there is only 1 set per frame
// No quant weighting when lossless
qmindex = lossless ? QINDEX_RANGE - 1 : cm->base_qindex;
for (j = 0; j < TX_SIZES; ++j) {
cm->y_iqmatrix[i][1][j] = aom_iqmatrix(cm, qmindex, 0, j, 1);
cm->y_iqmatrix[i][0][j] = aom_iqmatrix(cm, qmindex, 0, j, 0);
cm->uv_iqmatrix[i][1][j] = aom_iqmatrix(cm, qmindex, 1, j, 1);
cm->uv_iqmatrix[i][0][j] = aom_iqmatrix(cm, qmindex, 1, j, 0);
}
#endif
}
} else {
const int qindex = cm->base_qindex;
......@@ -1143,6 +1161,18 @@ static void setup_segmentation_dequant(VP10_COMMON *const cm) {
vp10_dc_quant(qindex, cm->uv_dc_delta_q, cm->bit_depth);
cm->uv_dequant[0][1] =
vp10_ac_quant(qindex, cm->uv_ac_delta_q, cm->bit_depth);
#if CONFIG_AOM_QM
lossless = qindex == 0 && cm->y_dc_delta_q == 0 && cm->uv_dc_delta_q == 0 &&
cm->uv_ac_delta_q == 0;
// No quant weighting when lossless
qmindex = lossless ? QINDEX_RANGE - 1 : cm->base_qindex;
for (j = 0; j < TX_SIZES; ++j) {
cm->y_iqmatrix[i][1][j] = aom_iqmatrix(cm, qmindex, 0, j, 1);
cm->y_iqmatrix[i][0][j] = aom_iqmatrix(cm, qmindex, 0, j, 0);
cm->uv_iqmatrix[i][1][j] = aom_iqmatrix(cm, qmindex, 1, j, 1);
cm->uv_iqmatrix[i][0][j] = aom_iqmatrix(cm, qmindex, 1, j, 0);
}
#endif
}
}
......
......@@ -109,6 +109,10 @@ VP10Decoder *vp10_decoder_create(BufferPool *const pool) {
vp10_loop_filter_init(cm);
#if CONFIG_AOM_QM
aom_qm_init(cm);
#endif
cm->error.setjmp = 0;
vpx_get_worker_interface()->init(&pbi->lf_worker);
......
......@@ -43,14 +43,25 @@ static INLINE int read_coeff(const vpx_prob *probs, int n, vpx_reader *r) {
return val;
}
#if CONFIG_AOM_QM
static int decode_coefs(const MACROBLOCKD *xd, PLANE_TYPE type,
tran_low_t *dqcoeff, TX_SIZE tx_size, const int16_t *dq,
int ctx, const int16_t *scan, const int16_t *nb,
vpx_reader *r) {
vpx_reader *r, const qm_val_t *iqm[2][TX_SIZES])
#else
static int decode_coefs(const MACROBLOCKD *xd, PLANE_TYPE type,
tran_low_t *dqcoeff, TX_SIZE tx_size, const int16_t *dq,
int ctx, const int16_t *scan, const int16_t *nb,
vpx_reader *r)
#endif
{
FRAME_COUNTS *counts = xd->counts;
const int max_eob = 16 << (tx_size << 1);
const FRAME_CONTEXT *const fc = xd->fc;
const int ref = is_inter_block(&xd->mi[0]->mbmi);
#if CONFIG_AOM_QM
const qm_val_t *iqmatrix = iqm[!ref][tx_size];
#endif
int band, c = 0;
const vpx_prob(*coef_probs)[COEFF_CONTEXTS][UNCONSTRAINED_NODES] =
fc->coef_probs[tx_size][type][ref];
......@@ -183,6 +194,10 @@ static int decode_coefs(const MACROBLOCKD *xd, PLANE_TYPE type,
}
}
}
#if CONFIG_AOM_QM
dqv = ((iqmatrix[scan[c]] * (int)dqv) + (1 << (AOM_QM_BITS - 1))) >>
AOM_QM_BITS;
#endif
v = (val * dqv) >> dq_shift;
#if CONFIG_COEFFICIENT_RANGE_CHECKING
#if CONFIG_VPX_HIGHBITDEPTH
......@@ -249,8 +264,16 @@ int vp10_decode_block_tokens(MACROBLOCKD *xd, int plane, const scan_order *sc,
const int16_t *const dequant = pd->seg_dequant[seg_id];
const int ctx =
get_entropy_context(tx_size, pd->above_context + x, pd->left_context + y);
#if CONFIG_AOM_QM
const int eob =
decode_coefs(xd, pd->plane_type, pd->dqcoeff, tx_size, dequant, ctx,
sc->scan, sc->neighbors, r, pd->seg_iqmatrix[seg_id]);
#else
const int eob = decode_coefs(xd, pd->plane_type, pd->dqcoeff, tx_size,
dequant, ctx, sc->scan, sc->neighbors, r);
#endif
dec_set_contexts(xd, pd, tx_size, eob > 0, x, y);
return eob;
}
......@@ -1021,7 +1021,12 @@ void vp10_fdct8x8_quant_c(const int16_t *input, int stride,
const int16_t *quant_shift_ptr,
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
const int16_t *dequant_ptr, uint16_t *eob_ptr,
const int16_t *scan, const int16_t *iscan) {
const int16_t *scan, const int16_t *iscan
#if CONFIG_AOM_QM
,
const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr
#endif
) {
int eob = -1;
int i, j;
......@@ -1107,16 +1112,29 @@ void vp10_fdct8x8_quant_c(const int16_t *input, int stride,
for (i = 0; i < n_coeffs; i++) {
const int rc = scan[i];
const int coeff = coeff_ptr[rc];
#if CONFIG_AOM_QM
const qm_val_t wt = qm_ptr[rc];
const qm_val_t iwt = iqm_ptr[rc];
const int dequant =
(dequant_ptr[rc != 0] * iwt + (1 << (AOM_QM_BITS - 1))) >>
AOM_QM_BITS;
#endif
const int coeff_sign = (coeff >> 31);
const int abs_coeff = (coeff ^ coeff_sign) - coeff_sign;
int tmp = clamp(abs_coeff + round_ptr[rc != 0], INT16_MIN, INT16_MAX);
tmp = (tmp * quant_ptr[rc != 0]) >> 16;
qcoeff_ptr[rc] = (tmp ^ coeff_sign) - coeff_sign;
int64_t tmp = clamp(abs_coeff + round_ptr[rc != 0], INT16_MIN, INT16_MAX);
int tmp32;
#if CONFIG_AOM_QM
tmp32 = (tmp * quant_ptr[rc != 0] * wt) >> (16 + AOM_QM_BITS);
qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant;
#else
tmp32 = (tmp * quant_ptr[rc != 0]) >> 16;
qcoeff_ptr[rc] = (tmp32 ^ coeff_sign) - coeff_sign;
dqcoeff_ptr[rc] = qcoeff_ptr[rc] * dequant_ptr[rc != 0];
#endif
if (tmp) eob = i;
if (tmp32) eob = i;
}
}
*eob_ptr = eob + 1;
......
This diff is collapsed.
......@@ -1613,6 +1613,9 @@ VP10_COMP *vp10_create_compressor(VP10EncoderConfig *oxcf,
* vp10_init_quantizer() for every frame.
*/
vp10_init_quantizer(cpi);
#if CONFIG_AOM_QM
aom_qm_init(cm);
#endif
vp10_loop_filter_init(cm);
......
This diff is collapsed.
This diff is collapsed.
......@@ -18,6 +18,47 @@
extern "C" {
#endif
#if CONFIG_AOM_QM
void vpx_quantize_dc(const tran_low_t *coeff_ptr, int n_coeffs, int skip_block,
const int16_t *round_ptr, const int16_t quant_ptr,
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
const int16_t dequant_ptr, uint16_t *eob_ptr,
const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr);
void vpx_quantize_dc_32x32(const tran_low_t *coeff_ptr, int skip_block,
const int16_t *round_ptr, const int16_t quant_ptr,
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
const int16_t dequant_ptr, uint16_t *eob_ptr,
const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr);
void vpx_quantize_b_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr,
uint16_t *eob_ptr, const int16_t *scan,
const int16_t *iscan, const qm_val_t *qm_ptr,
const qm_val_t *iqm_ptr);
#if CONFIG_VPX_HIGHBITDEPTH
void vpx_highbd_quantize_dc(const tran_low_t *coeff_ptr, int n_coeffs,
int skip_block, const int16_t *round_ptr,
const int16_t quant_ptr, tran_low_t *qcoeff_ptr,
tran_low_t *dqcoeff_ptr, const int16_t dequant_ptr,
uint16_t *eob_ptr, const qm_val_t *qm_ptr,
const qm_val_t *iqm_ptr);
void vpx_highbd_quantize_dc_32x32(
const tran_low_t *coeff_ptr, int skip_block, const int16_t *round_ptr,
const int16_t quant_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
const int16_t dequant_ptr, uint16_t *eob_ptr, const qm_val_t *qm_ptr,
const qm_val_t *iqm_ptr);
void vpx_highbd_quantize_b_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr,
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
const int16_t *dequant_ptr, uint16_t *eob_ptr,
const int16_t *scan, const int16_t *iscan,
const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr);
#endif
#else
void vpx_quantize_dc(const tran_low_t *coeff_ptr, int n_coeffs, int skip_block,
const int16_t *round_ptr, const int16_t quant_ptr,
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
......@@ -26,7 +67,13 @@ void vpx_quantize_dc_32x32(const tran_low_t *coeff_ptr, int skip_block,
const int16_t *round_ptr, const int16_t quant_ptr,
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
const int16_t dequant_ptr, uint16_t *eob_ptr);
void vpx_quantize_b_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr,
uint16_t *eob_ptr, const int16_t *scan,
const int16_t *iscan);
#if CONFIG_VPX_HIGHBITDEPTH
void vpx_highbd_quantize_dc(const tran_low_t *coeff_ptr, int n_coeffs,
int skip_block, const int16_t *round_ptr,
......@@ -39,6 +86,14 @@ void vpx_highbd_quantize_dc_32x32(const tran_low_t *coeff_ptr, int skip_block,
tran_low_t *qcoeff_ptr,
tran_low_t *dqcoeff_ptr,
const int16_t dequant_ptr, uint16_t *eob_ptr);
void vpx_highbd_quantize_b_c(const tran_low_t *coeff_ptr, intptr_t n_coeffs,
int skip_block, const int16_t *zbin_ptr,
const int16_t *round_ptr, const int16_t *quant_ptr,
const int16_t *quant_shift_ptr,
tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr,
const int16_t *dequant_ptr, uint16_t *eob_ptr,
const int16_t *scan, const int16_t *iscan);
#endif
#endif
#ifdef __cplusplus
......
......@@ -931,21 +931,35 @@ if (vpx_config("CONFIG_VPX_HIGHBITDEPTH") eq "yes") {
#
# Quantization
#
if (vpx_config("CONFIG_VP10_ENCODER") eq "yes") {
add_proto qw/void vpx_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vpx_quantize_b sse2/, "$ssse3_x86_64_x86inc", "$avx_x86_64_x86inc";
if (vpx_config("CONFIG_AOM_QM") eq "yes") {
if (vpx_config("CONFIG_VP10_ENCODER") eq "yes") {
add_proto qw/void vpx_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr";
add_proto qw/void vpx_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vpx_quantize_b_32x32/, "$ssse3_x86_64_x86inc", "$avx_x86_64_x86inc";
add_proto qw/void vpx_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr";
if (vpx_config("CONFIG_VPX_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vpx_highbd_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vpx_highbd_quantize_b sse2/;
if (vpx_config("CONFIG_VPX_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vpx_highbd_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr";
add_proto qw/void vpx_highbd_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vpx_highbd_quantize_b_32x32 sse2/;
} # CONFIG_VPX_HIGHBITDEPTH
} # CONFIG_VP10_ENCODER
add_proto qw/void vpx_highbd_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr";
} # CONFIG_VPX_HIGHBITDEPTH
} # CONFIG_VP10_ENCODER
} else {
if (vpx_config("CONFIG_VP10_ENCODER") eq "yes") {
add_proto qw/void vpx_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vpx_quantize_b sse2/, "$ssse3_x86_64_x86inc", "$avx_x86_64_x86inc";
add_proto qw/void vpx_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vpx_quantize_b_32x32/, "$ssse3_x86_64_x86inc", "$avx_x86_64_x86inc";
if (vpx_config("CONFIG_VPX_HIGHBITDEPTH") eq "yes") {
add_proto qw/void vpx_highbd_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vpx_highbd_quantize_b sse2/;
add_proto qw/void vpx_highbd_quantize_b_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
specialize qw/vpx_highbd_quantize_b_32x32 sse2/;
} # CONFIG_VPX_HIGHBITDEPTH
} # CONFIG_VP10_ENCODER
} # CONFIG_AOM_QM
if (vpx_config("CONFIG_ENCODERS") eq "yes") {
#
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment