Merge "Refactor 1D transforms" into nextgenv2
diff --git a/test/borders_test.cc b/test/borders_test.cc
index 6592375..ff3812c 100644
--- a/test/borders_test.cc
+++ b/test/borders_test.cc
@@ -52,7 +52,7 @@
// extend into the border and test the border condition.
cfg_.g_lag_in_frames = 25;
cfg_.rc_2pass_vbr_minsection_pct = 5;
- cfg_.rc_2pass_vbr_minsection_pct = 2000;
+ cfg_.rc_2pass_vbr_maxsection_pct = 2000;
cfg_.rc_target_bitrate = 2000;
cfg_.rc_max_quantizer = 10;
diff --git a/test/cpu_speed_test.cc b/test/cpu_speed_test.cc
index 8baa2f9..6a938a0 100644
--- a/test/cpu_speed_test.cc
+++ b/test/cpu_speed_test.cc
@@ -74,7 +74,7 @@
// the encoder to producing lots of big partitions which will likely
// extend into the border and test the border condition.
cfg_.rc_2pass_vbr_minsection_pct = 5;
- cfg_.rc_2pass_vbr_minsection_pct = 2000;
+ cfg_.rc_2pass_vbr_maxsection_pct = 2000;
cfg_.rc_target_bitrate = 400;
cfg_.rc_max_quantizer = 0;
cfg_.rc_min_quantizer = 0;
@@ -92,7 +92,7 @@
::libvpx_test::Y4mVideoSource video("screendata.y4m", 0, 25);
cfg_.g_timebase = video.timebase();
cfg_.rc_2pass_vbr_minsection_pct = 5;
- cfg_.rc_2pass_vbr_minsection_pct = 2000;
+ cfg_.rc_2pass_vbr_maxsection_pct = 2000;
cfg_.rc_target_bitrate = 400;
cfg_.rc_max_quantizer = 0;
cfg_.rc_min_quantizer = 0;
@@ -109,7 +109,7 @@
// the encoder to producing lots of big partitions which will likely
// extend into the border and test the border condition.
cfg_.rc_2pass_vbr_minsection_pct = 5;
- cfg_.rc_2pass_vbr_minsection_pct = 2000;
+ cfg_.rc_2pass_vbr_maxsection_pct = 2000;
cfg_.rc_target_bitrate = 12000;
cfg_.rc_max_quantizer = 10;
cfg_.rc_min_quantizer = 0;
@@ -125,7 +125,7 @@
// when passing in a very high min q. This pushes the encoder to producing
// lots of small partitions which might will test the other condition.
cfg_.rc_2pass_vbr_minsection_pct = 5;
- cfg_.rc_2pass_vbr_minsection_pct = 2000;
+ cfg_.rc_2pass_vbr_maxsection_pct = 2000;
cfg_.rc_target_bitrate = 200;
cfg_.rc_min_quantizer = 40;
diff --git a/test/vp9_ethread_test.cc b/test/vp9_ethread_test.cc
index 29a653f..3445bf2 100644
--- a/test/vp9_ethread_test.cc
+++ b/test/vp9_ethread_test.cc
@@ -48,7 +48,7 @@
cfg_.g_lag_in_frames = 3;
cfg_.rc_end_usage = VPX_VBR;
cfg_.rc_2pass_vbr_minsection_pct = 5;
- cfg_.rc_2pass_vbr_minsection_pct = 2000;
+ cfg_.rc_2pass_vbr_maxsection_pct = 2000;
} else {
cfg_.g_lag_in_frames = 0;
cfg_.rc_end_usage = VPX_CBR;
diff --git a/vp10/common/blockd.h b/vp10/common/blockd.h
index d5139f7..016fc75 100644
--- a/vp10/common/blockd.h
+++ b/vp10/common/blockd.h
@@ -380,7 +380,7 @@
#if CONFIG_EXT_TX
#define ALLOW_INTRA_EXT_TX 1
// whether masked transforms are used for 32X32
-#define USE_MSKTX_FOR_32X32 1
+#define USE_MSKTX_FOR_32X32 0
static const int num_ext_tx_set_inter[EXT_TX_SETS_INTER] = {
1, 19, 12, 2
@@ -447,7 +447,7 @@
{ 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
{ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1},
{ 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1},
- { 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1},
+ { 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0},
};
static INLINE int get_ext_tx_types(TX_SIZE tx_size, BLOCK_SIZE bs,
diff --git a/vp10/common/common.h b/vp10/common/common.h
index 4abcbf6..54c7b89 100644
--- a/vp10/common/common.h
+++ b/vp10/common/common.h
@@ -33,12 +33,12 @@
// Use this for variably-sized arrays.
#define vp10_copy_array(dest, src, n) { \
- assert(sizeof(*dest) == sizeof(*src)); \
- memcpy(dest, src, n * sizeof(*src)); \
+ assert(sizeof(*(dest)) == sizeof(*(src))); \
+ memcpy(dest, src, n * sizeof(*(src))); \
}
#define vp10_zero(dest) memset(&(dest), 0, sizeof(dest))
-#define vp10_zero_array(dest, n) memset(dest, 0, n * sizeof(*dest))
+#define vp10_zero_array(dest, n) memset(dest, 0, n * sizeof(*(dest)))
static INLINE int get_unsigned_bits(unsigned int num_values) {
return num_values > 0 ? get_msb(num_values) + 1 : 0;
diff --git a/vp10/encoder/rdopt.c b/vp10/encoder/rdopt.c
index c7a2702..6543b33 100644
--- a/vp10/encoder/rdopt.c
+++ b/vp10/encoder/rdopt.c
@@ -6075,28 +6075,6 @@
!is_comp_interintra_pred &&
#endif // CONFIG_EXT_INTER
is_obmc_allowed(mbmi);
- int best_obmc_flag = 0;
-#if CONFIG_VP9_HIGHBITDEPTH
- DECLARE_ALIGNED(16, uint16_t, tmp_buf1_16[MAX_MB_PLANE * CU_SIZE * CU_SIZE]);
- DECLARE_ALIGNED(16, uint16_t, tmp_buf2_16[MAX_MB_PLANE * CU_SIZE * CU_SIZE]);
- uint8_t *tmp_buf1, *tmp_buf2;
- uint8_t *obmc_tmp_buf1[3];
- uint8_t *obmc_tmp_buf2[3];
-#else
- DECLARE_ALIGNED(16, uint8_t, tmp_buf1[MAX_MB_PLANE * CU_SIZE * CU_SIZE]);
- DECLARE_ALIGNED(16, uint8_t, tmp_buf2[MAX_MB_PLANE * CU_SIZE * CU_SIZE]);
- uint8_t *obmc_tmp_buf1[3] = {tmp_buf1, tmp_buf1 + CU_SIZE * CU_SIZE,
- tmp_buf1 + CU_SIZE * CU_SIZE * 2};
- uint8_t *obmc_tmp_buf2[3] = {tmp_buf2, tmp_buf2 + CU_SIZE * CU_SIZE,
- tmp_buf2 + CU_SIZE * CU_SIZE * 2};
-#endif // CONFIG_VP9_HIGHBITDEPTH
- int obmc_tmp_stride[3] = {CU_SIZE, CU_SIZE, CU_SIZE};
-
- uint8_t skip_txfm_bestfilter[2][MAX_MB_PLANE << 2] = {{0}, {0}};
- int64_t bsse_bestfilter[2][MAX_MB_PLANE << 2] = {{0}, {0}};
- int skip_txfm_sb_bestfilter[2] = {0};
- int64_t skip_sse_sb_bestfilter[2] = {INT64_MAX};
-
int rate2_nocoeff, best_rate2 = INT_MAX,
best_skippable, best_xskip, best_disable_skip = 0;
#if CONFIG_SUPERTX
@@ -6154,25 +6132,9 @@
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
tmp_buf = CONVERT_TO_BYTEPTR(tmp_buf16);
-#if CONFIG_OBMC
- tmp_buf1 = CONVERT_TO_BYTEPTR(tmp_buf1_16);
- tmp_buf2 = CONVERT_TO_BYTEPTR(tmp_buf2_16);
-#endif // CONFIG_OBMC
} else {
tmp_buf = (uint8_t *)tmp_buf16;
-#if CONFIG_OBMC
- tmp_buf1 = (uint8_t *)tmp_buf1_16;
- tmp_buf2 = (uint8_t *)tmp_buf2_16;
-#endif // CONFIG_OBMC
}
-#if CONFIG_OBMC
- obmc_tmp_buf1[0] = tmp_buf1;
- obmc_tmp_buf1[1] = tmp_buf1 + 4096;
- obmc_tmp_buf1[2] = tmp_buf1 + 8192;
- obmc_tmp_buf2[0] = tmp_buf2;
- obmc_tmp_buf2[1] = tmp_buf2 + 4096;
- obmc_tmp_buf2[2] = tmp_buf2 + 8192;
-#endif // CONFIG_OBMC
#endif // CONFIG_VP9_HIGHBITDEPTH
if (is_comp_pred) {
@@ -6436,15 +6398,6 @@
int64_t rs_rd;
int tmp_skip_sb = 0;
int64_t tmp_skip_sse = INT64_MAX;
-#if CONFIG_OBMC
- int obmc_flag = 0;
- int tmp_skip_sb_obmc = 0;
- int64_t tmp_skip_sse_obmc = INT64_MAX;
- int64_t rdobmc = INT64_MAX;
- uint8_t *obmc_tmp_buf[3];
- uint8_t tmp_skip_txfm[MAX_MB_PLANE << 2] = {0};
- int64_t tmp_bsse[MAX_MB_PLANE << 2] = {0};
-#endif // CONFIG_OBMC
mbmi->interp_filter = i;
rs = vp10_get_switchable_rate(cpi, xd);
@@ -6457,21 +6410,11 @@
VPXMIN(filter_cache[SWITCHABLE_FILTERS], rd + rs_rd);
if (cm->interp_filter == SWITCHABLE)
rd += rs_rd;
-#if CONFIG_OBMC
- if (allow_obmc) {
- obmc_flag = best_obmc_flag;
- rd += RDCOST(x->rdmult, x->rddiv,
- cpi->obmc_cost[bsize][obmc_flag], 0);
- }
-#endif // CONFIG_OBMC
*mask_filter = VPXMAX(*mask_filter, rd);
} else {
int rate_sum = 0;
int64_t dist_sum = 0;
-#if CONFIG_OBMC
- int rate_sum_obmc = 0;
- int64_t dist_sum_obmc = 0;
-#endif // CONFIG_OBMC
+
if (i > 0 && cpi->sf.adaptive_interp_filter_search &&
(cpi->sf.interp_filter_search_mask & (1 << i))) {
rate_sum = INT_MAX;
@@ -6488,18 +6431,10 @@
(cm->interp_filter == mbmi->interp_filter ||
(i == 0 && intpel_mv && IsInterpolatingFilter(i))))) {
restore_dst_buf(xd, orig_dst, orig_dst_stride);
-#if CONFIG_OBMC
- for (j = 0; j < MAX_MB_PLANE; j++) {
- obmc_tmp_buf[j] = obmc_tmp_buf1[j];
- }
-#endif // CONFIG_OBMC
} else {
for (j = 0; j < MAX_MB_PLANE; j++) {
xd->plane[j].dst.buf = tmp_buf + j * 64 * 64;
xd->plane[j].dst.stride = 64;
-#if CONFIG_OBMC
- obmc_tmp_buf[j] = obmc_tmp_buf2[j];
-#endif // CONFIG_OBMC
}
}
vp10_build_inter_predictors_sb(xd, mi_row, mi_col, bsize);
@@ -6507,34 +6442,6 @@
&tmp_skip_sb, &tmp_skip_sse);
rd = RDCOST(x->rdmult, x->rddiv, rate_sum, dist_sum);
-#if CONFIG_OBMC
- if (allow_obmc) {
- rd += RDCOST(x->rdmult, x->rddiv, cpi->obmc_cost[bsize][0], 0);
- memcpy(tmp_skip_txfm, x->skip_txfm, sizeof(tmp_skip_txfm));
- memcpy(tmp_bsse, x->bsse, sizeof(tmp_bsse));
-
- vp10_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, 1,
- obmc_tmp_buf, obmc_tmp_stride,
- dst_buf1, dst_stride1,
- dst_buf2, dst_stride2);
- for (j = 0; j < MAX_MB_PLANE; ++j) {
- xd->plane[j].dst.buf = obmc_tmp_buf[j];
- xd->plane[j].dst.stride = obmc_tmp_stride[j];
- }
- model_rd_for_sb(cpi, bsize, x, xd, &rate_sum_obmc, &dist_sum_obmc,
- &tmp_skip_sb_obmc, &tmp_skip_sse_obmc);
- rdobmc = RDCOST(x->rdmult, x->rddiv,
- rate_sum_obmc + cpi->obmc_cost[bsize][1],
- dist_sum_obmc);
-
- if ((double)rdobmc <= 0.99 * (double)rd) {
- obmc_flag = 1;
- rd = rdobmc;
- rate_sum = rate_sum_obmc;
- dist_sum = dist_sum_obmc;
- }
- }
-#endif // CONFIG_OBMC
filter_cache[i] = rd;
filter_cache[SWITCHABLE_FILTERS] =
VPXMIN(filter_cache[SWITCHABLE_FILTERS], rd + rs_rd);
@@ -6557,10 +6464,6 @@
newbest = i == 0 || rd < best_rd;
if (newbest) {
-#if CONFIG_OBMC
- if (allow_obmc)
- best_obmc_flag = obmc_flag;
-#endif // CONFIG_OBMC
best_rd = rd;
best_filter = mbmi->interp_filter;
if (cm->interp_filter == SWITCHABLE && i &&
@@ -6574,31 +6477,6 @@
pred_exists = 1;
tmp_rd = best_rd;
-#if CONFIG_OBMC
- if (allow_obmc) {
- skip_txfm_sb_bestfilter[0] = tmp_skip_sb;
- skip_sse_sb_bestfilter[0] = tmp_skip_sse;
- memcpy(skip_txfm_bestfilter[0], tmp_skip_txfm, sizeof(skip_txfm));
- memcpy(bsse_bestfilter[0], tmp_bsse, sizeof(bsse));
-
- skip_txfm_sb_bestfilter[1] = tmp_skip_sb_obmc;
- skip_sse_sb_bestfilter[1] = tmp_skip_sse_obmc;
- memcpy(skip_txfm_bestfilter[1], x->skip_txfm, sizeof(skip_txfm));
- memcpy(bsse_bestfilter[1], x->bsse, sizeof(bsse));
- if (best_obmc_flag) {
- tmp_skip_sb = tmp_skip_sb_obmc;
- tmp_skip_sse = tmp_skip_sse_obmc;
- } else {
- memcpy(x->skip_txfm, tmp_skip_txfm, sizeof(tmp_skip_txfm));
- memcpy(x->bsse, tmp_bsse, sizeof(tmp_bsse));
- }
- } else {
- skip_txfm_sb_bestfilter[0] = tmp_skip_sb;
- skip_sse_sb_bestfilter[0] = tmp_skip_sse;
- memcpy(skip_txfm_bestfilter[0], x->skip_txfm, sizeof(skip_txfm));
- memcpy(bsse_bestfilter[0], x->bsse, sizeof(bsse));
- }
-#endif // CONFIG_OBMC
skip_txfm_sb = tmp_skip_sb;
skip_sse_sb = tmp_skip_sse;
memcpy(skip_txfm, x->skip_txfm, sizeof(skip_txfm));
@@ -6923,15 +6801,7 @@
#endif // CONFIG_EXT_INTERP
#endif // CONFIG_EXT_INTER
-#if CONFIG_OBMC
- if (allow_obmc)
- mbmi->obmc = best_obmc_flag;
- else
- mbmi->obmc = 0;
-#endif // CONFIG_OBMC
-
if (pred_exists) {
-#if !CONFIG_OBMC
if (best_needs_copy) {
// again temporarily set the buffers to local memory to prevent a memcpy
for (i = 0; i < MAX_MB_PLANE; i++) {
@@ -6939,20 +6809,11 @@
xd->plane[i].dst.stride = 64;
}
}
-#endif // !CONFIG_OBMC
- rd = tmp_rd + RDCOST(x->rdmult, x->rddiv, rs, 0);
-#if CONFIG_OBMC
- if (allow_obmc)
- rd += RDCOST(x->rdmult, x->rddiv,
- cpi->obmc_cost[bsize][mbmi->obmc], 0);
-#endif // CONFIG_OBMC
+ rd = tmp_rd;
} else {
int tmp_rate;
int64_t tmp_dist;
-#if CONFIG_OBMC
- int64_t rdobmc = INT64_MAX;
- restore_dst_buf(xd, orig_dst, orig_dst_stride);
-#endif // CONFIG_OBMC
+
// Handles the special case when a filter that is not in the
// switchable list (ex. bilinear) is indicated at the frame level, or
// skip condition holds.
@@ -6960,34 +6821,6 @@
model_rd_for_sb(cpi, bsize, x, xd, &tmp_rate, &tmp_dist,
&skip_txfm_sb, &skip_sse_sb);
rd = RDCOST(x->rdmult, x->rddiv, rs + tmp_rate, tmp_dist);
-#if CONFIG_OBMC
- skip_txfm_sb_bestfilter[0] = skip_txfm_sb;
- skip_sse_sb_bestfilter[0] = skip_sse_sb;
- memcpy(skip_txfm_bestfilter[0], x->skip_txfm, sizeof(skip_txfm));
- memcpy(bsse_bestfilter[0], x->bsse, sizeof(bsse));
- if (allow_obmc) {
- rd += RDCOST(x->rdmult, x->rddiv, cpi->obmc_cost[bsize][0], 0);
- vp10_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, 1,
- obmc_tmp_buf1, obmc_tmp_stride,
- dst_buf1, dst_stride1,
- dst_buf2, dst_stride2);
- for (i = 0; i < MAX_MB_PLANE; ++i) {
- xd->plane[i].dst.buf = obmc_tmp_buf1[i];
- xd->plane[i].dst.stride = obmc_tmp_stride[i];
- }
- model_rd_for_sb(cpi, bsize, x, xd, &tmp_rate, &tmp_dist,
- &skip_txfm_sb, &skip_sse_sb);
- rdobmc = RDCOST(x->rdmult, x->rddiv,
- rs + tmp_rate + cpi->obmc_cost[bsize][1], tmp_dist);
-
- skip_txfm_sb_bestfilter[1] = skip_txfm_sb;
- skip_sse_sb_bestfilter[1] = skip_sse_sb;
- memcpy(skip_txfm_bestfilter[1], x->skip_txfm, sizeof(skip_txfm));
- memcpy(bsse_bestfilter[1], x->bsse, sizeof(bsse));
- if ((double)rdobmc <= 0.99 * (double)rd)
- rd = rdobmc;
- }
-#endif // CONFIG_OBMC
memcpy(skip_txfm, x->skip_txfm, sizeof(skip_txfm));
memcpy(bsse, x->bsse, sizeof(bsse));
}
@@ -7078,40 +6911,16 @@
#if CONFIG_OBMC
best_rd = INT64_MAX;
for (mbmi->obmc = 0; mbmi->obmc <= allow_obmc; mbmi->obmc++) {
- int64_t tmp_rd;
+ int64_t tmp_rd, tmp_dist;
+ int tmp_rate;
- if (pred_exists) {
- if (best_needs_copy) {
- if (mbmi->obmc) {
- for (i = 0; i < MAX_MB_PLANE; i++) {
- xd->plane[i].dst.buf = obmc_tmp_buf2[i];
- xd->plane[i].dst.stride = 64;
- }
- } else {
- for (i = 0; i < MAX_MB_PLANE; i++) {
- xd->plane[i].dst.buf = tmp_buf + i * 64 * 64;
- xd->plane[i].dst.stride = 64;
- }
- }
- } else {
- if (mbmi->obmc) {
- for (i = 0; i < MAX_MB_PLANE; i++) {
- xd->plane[i].dst.buf = obmc_tmp_buf1[i];
- xd->plane[i].dst.stride = 64;
- }
- } else {
- restore_dst_buf(xd, orig_dst, orig_dst_stride);
- }
- }
- } else {
- if (mbmi->obmc) {
- for (i = 0; i < MAX_MB_PLANE; i++) {
- xd->plane[i].dst.buf = obmc_tmp_buf1[i];
- xd->plane[i].dst.stride = 64;
- }
- } else {
- restore_dst_buf(xd, orig_dst, orig_dst_stride);
- }
+ if (mbmi->obmc) {
+ vp10_build_obmc_inter_prediction(cm, xd, mi_row, mi_col, 0,
+ NULL, NULL,
+ dst_buf1, dst_stride1,
+ dst_buf2, dst_stride2);
+ model_rd_for_sb(cpi, bsize, x, xd, &tmp_rate, &tmp_dist,
+ &skip_txfm_sb, &skip_sse_sb);
}
#if CONFIG_VP9_HIGHBITDEPTH
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
@@ -7126,11 +6935,6 @@
vp10_get_sby_perpixel_variance(cpi, &xd->plane[0].dst, bsize);
#endif // CONFIG_VP9_HIGHBITDEPTH
- skip_txfm_sb = skip_txfm_sb_bestfilter[mbmi->obmc];
- skip_sse_sb = skip_sse_sb_bestfilter[mbmi->obmc];
- memcpy(x->skip_txfm, skip_txfm_bestfilter[mbmi->obmc],
- sizeof(skip_txfm));
- memcpy(x->bsse, bsse_bestfilter[mbmi->obmc], sizeof(bsse));
x->skip = 0;
*rate2 = rate2_nocoeff;
@@ -7231,10 +7035,7 @@
// The cost of skip bit needs to be added.
#if CONFIG_OBMC
- mbmi->skip = xd->lossless[mbmi->segment_id] ? 0 : 1;
- if (xd->lossless[mbmi->segment_id])
- *rate2 += vp10_cost_bit(vp10_get_skip_prob(cm, xd), 0);
- else
+ mbmi->skip = 0;
#endif // CONFIG_OBMC
*rate2 += vp10_cost_bit(vp10_get_skip_prob(cm, xd), 1);
diff --git a/vp8/common/common.h b/vp8/common/common.h
index e58a9cc..c42e875 100644
--- a/vp8/common/common.h
+++ b/vp8/common/common.h
@@ -32,13 +32,13 @@
/* Use this for variably-sized arrays. */
#define vp8_copy_array( Dest, Src, N) { \
- assert( sizeof( *Dest) == sizeof( *Src)); \
- memcpy( Dest, Src, N * sizeof( *Src)); \
+ assert( sizeof( *(Dest)) == sizeof( *(Src))); \
+ memcpy( Dest, Src, N * sizeof( *(Src))); \
}
-#define vp8_zero( Dest) memset( &Dest, 0, sizeof( Dest));
+#define vp8_zero( Dest) memset( &(Dest), 0, sizeof( Dest));
-#define vp8_zero_array( Dest, N) memset( Dest, 0, N * sizeof( *Dest));
+#define vp8_zero_array( Dest, N) memset( Dest, 0, N * sizeof( *(Dest)));
#ifdef __cplusplus
diff --git a/vp9/common/vp9_common.h b/vp9/common/vp9_common.h
index 76e7cd4..9d5dbc6 100644
--- a/vp9/common/vp9_common.h
+++ b/vp9/common/vp9_common.h
@@ -33,12 +33,12 @@
// Use this for variably-sized arrays.
#define vp9_copy_array(dest, src, n) { \
- assert(sizeof(*dest) == sizeof(*src)); \
- memcpy(dest, src, n * sizeof(*src)); \
+ assert(sizeof(*(dest)) == sizeof(*(src))); \
+ memcpy(dest, src, n * sizeof(*(src))); \
}
#define vp9_zero(dest) memset(&(dest), 0, sizeof(dest))
-#define vp9_zero_array(dest, n) memset(dest, 0, n * sizeof(*dest))
+#define vp9_zero_array(dest, n) memset(dest, 0, n * sizeof(*(dest)))
static INLINE int get_unsigned_bits(unsigned int num_values) {
return num_values > 0 ? get_msb(num_values) + 1 : 0;
