From 8cf352abac23e655d5c40df273ab47e2c8395944 Mon Sep 17 00:00:00 2001 From: Jim Bankoski Date: Tue, 19 Nov 2013 14:31:38 -0800 Subject: [PATCH] entropy code speedup Change-Id: Ic316d3374ff9a2b43897272260947d56765a0fdd --- vp9/common/vp9_entropy.c | 265 ++++++++++++++++++++++++++++++++++- vp9/common/vp9_entropy.h | 4 + vp9/decoder/vp9_detokenize.c | 38 ++--- 3 files changed, 287 insertions(+), 20 deletions(-) diff --git a/vp9/common/vp9_entropy.c b/vp9/common/vp9_entropy.c index 1cf9dbcbb..82aa77eef 100644 --- a/vp9/common/vp9_entropy.c +++ b/vp9/common/vp9_entropy.c @@ -15,7 +15,6 @@ #include "vpx_mem/vpx_mem.h" #include "vpx/vpx_integer.h" -#define MODEL_NODES (ENTROPY_NODES - UNCONSTRAINED_NODES) DECLARE_ALIGNED(16, const uint8_t, vp9_norm[256]) = { 0, 7, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, @@ -158,6 +157,8 @@ const vp9_tree_index vp9_coefmodel_tree[6] = { // the probabilities for the rest of the nodes. // beta = 8 + + static const vp9_prob pareto8_probs[COEFPROB_MODELS][MODEL_NODES] = { { 3, 86, 128, 6, 86, 23, 88, 29}, { 9, 86, 129, 17, 88, 61, 94, 76}, @@ -289,6 +290,268 @@ static const vp9_prob pareto8_probs[COEFPROB_MODELS][MODEL_NODES] = { {255, 246, 247, 255, 239, 255, 253, 255} }; +// This table is an expansion of the table : modelcoefprobs_pareto8 +// to all 255 probabilities using the code as follows to do the expansion: +// tree_probs[i] = (model[l][i - UNCONSTRAINED_NODES] + +// model[l + 1][i - UNCONSTRAINED_NODES]) >> 1; +const vp9_prob vp9_pareto8_full[255][MODEL_NODES] = { + { 3, 86, 128, 6, 86, 23, 88, 29}, + { 6, 86, 128, 11, 87, 42, 91, 52}, + { 9, 86, 129, 17, 88, 61, 94, 76}, + { 12, 86, 129, 22, 88, 77, 97, 93}, + { 15, 87, 129, 28, 89, 93, 100, 110}, + { 17, 87, 129, 33, 90, 105, 103, 123}, + { 20, 88, 130, 38, 91, 118, 106, 136}, + { 23, 88, 130, 43, 91, 128, 108, 146}, + { 26, 89, 131, 48, 92, 139, 111, 156}, + { 28, 89, 131, 53, 93, 147, 114, 163}, + { 31, 90, 131, 58, 94, 156, 117, 171}, + { 34, 90, 131, 62, 94, 163, 119, 177}, + { 37, 90, 132, 66, 95, 171, 122, 184}, + { 39, 90, 132, 70, 96, 177, 124, 189}, + { 42, 91, 132, 75, 97, 183, 127, 194}, + { 44, 91, 132, 79, 97, 188, 129, 198}, + { 47, 92, 133, 83, 98, 193, 132, 202}, + { 49, 92, 133, 86, 99, 197, 134, 205}, + { 52, 93, 133, 90, 100, 201, 137, 208}, + { 54, 93, 133, 94, 100, 204, 139, 211}, + { 57, 94, 134, 98, 101, 208, 142, 214}, + { 59, 94, 134, 101, 102, 211, 144, 216}, + { 62, 94, 135, 105, 103, 214, 146, 218}, + { 64, 94, 135, 108, 103, 216, 148, 220}, + { 66, 95, 135, 111, 104, 219, 151, 222}, + { 68, 95, 135, 114, 105, 221, 153, 223}, + { 71, 96, 136, 117, 106, 224, 155, 225}, + { 73, 96, 136, 120, 106, 225, 157, 226}, + { 76, 97, 136, 123, 107, 227, 159, 228}, + { 78, 97, 136, 126, 108, 229, 160, 229}, + { 80, 98, 137, 129, 109, 231, 162, 231}, + { 82, 98, 137, 131, 109, 232, 164, 232}, + { 84, 98, 138, 134, 110, 234, 166, 233}, + { 86, 98, 138, 137, 111, 235, 168, 234}, + { 89, 99, 138, 140, 112, 236, 170, 235}, + { 91, 99, 138, 142, 112, 237, 171, 235}, + { 93, 100, 139, 145, 113, 238, 173, 236}, + { 95, 100, 139, 147, 114, 239, 174, 237}, + { 97, 101, 140, 149, 115, 240, 176, 238}, + { 99, 101, 140, 151, 115, 241, 177, 238}, + {101, 102, 140, 154, 116, 242, 179, 239}, + {103, 102, 140, 156, 117, 242, 180, 239}, + {105, 103, 141, 158, 118, 243, 182, 240}, + {107, 103, 141, 160, 118, 243, 183, 240}, + {109, 104, 141, 162, 119, 244, 185, 241}, + {111, 104, 141, 164, 119, 244, 186, 241}, + {113, 104, 142, 166, 120, 245, 187, 242}, + {114, 104, 142, 168, 121, 245, 188, 242}, + {116, 105, 143, 170, 122, 246, 190, 243}, + {118, 105, 143, 171, 122, 246, 191, 243}, + {120, 106, 143, 173, 123, 247, 192, 244}, + {121, 106, 143, 175, 124, 247, 193, 244}, + {123, 107, 144, 177, 125, 248, 195, 244}, + {125, 107, 144, 178, 125, 248, 196, 244}, + {127, 108, 145, 180, 126, 249, 197, 245}, + {128, 108, 145, 181, 127, 249, 198, 245}, + {130, 109, 145, 183, 128, 249, 199, 245}, + {132, 109, 145, 184, 128, 249, 200, 245}, + {134, 110, 146, 186, 129, 250, 201, 246}, + {135, 110, 146, 187, 130, 250, 202, 246}, + {137, 111, 147, 189, 131, 251, 203, 246}, + {138, 111, 147, 190, 131, 251, 204, 246}, + {140, 112, 147, 192, 132, 251, 205, 247}, + {141, 112, 147, 193, 132, 251, 206, 247}, + {143, 113, 148, 194, 133, 251, 207, 247}, + {144, 113, 148, 195, 134, 251, 207, 247}, + {146, 114, 149, 197, 135, 252, 208, 248}, + {147, 114, 149, 198, 135, 252, 209, 248}, + {149, 115, 149, 199, 136, 252, 210, 248}, + {150, 115, 149, 200, 137, 252, 210, 248}, + {152, 115, 150, 201, 138, 252, 211, 248}, + {153, 115, 150, 202, 138, 252, 212, 248}, + {155, 116, 151, 204, 139, 253, 213, 249}, + {156, 116, 151, 205, 139, 253, 213, 249}, + {158, 117, 151, 206, 140, 253, 214, 249}, + {159, 117, 151, 207, 141, 253, 215, 249}, + {161, 118, 152, 208, 142, 253, 216, 249}, + {162, 118, 152, 209, 142, 253, 216, 249}, + {163, 119, 153, 210, 143, 253, 217, 249}, + {164, 119, 153, 211, 143, 253, 217, 249}, + {166, 120, 153, 212, 144, 254, 218, 250}, + {167, 120, 153, 212, 145, 254, 219, 250}, + {168, 121, 154, 213, 146, 254, 220, 250}, + {169, 121, 154, 214, 146, 254, 220, 250}, + {171, 122, 155, 215, 147, 254, 221, 250}, + {172, 122, 155, 216, 147, 254, 221, 250}, + {173, 123, 155, 217, 148, 254, 222, 250}, + {174, 123, 155, 217, 149, 254, 222, 250}, + {176, 124, 156, 218, 150, 254, 223, 250}, + {177, 124, 156, 219, 150, 254, 223, 250}, + {178, 125, 157, 220, 151, 254, 224, 251}, + {179, 125, 157, 220, 151, 254, 224, 251}, + {180, 126, 157, 221, 152, 254, 225, 251}, + {181, 126, 157, 221, 152, 254, 225, 251}, + {183, 127, 158, 222, 153, 254, 226, 251}, + {184, 127, 158, 223, 154, 254, 226, 251}, + {185, 128, 159, 224, 155, 255, 227, 251}, + {186, 128, 159, 224, 155, 255, 227, 251}, + {187, 129, 160, 225, 156, 255, 228, 251}, + {188, 130, 160, 225, 156, 255, 228, 251}, + {189, 131, 160, 226, 157, 255, 228, 251}, + {190, 131, 160, 226, 158, 255, 228, 251}, + {191, 132, 161, 227, 159, 255, 229, 251}, + {192, 132, 161, 227, 159, 255, 229, 251}, + {193, 133, 162, 228, 160, 255, 230, 252}, + {194, 133, 162, 229, 160, 255, 230, 252}, + {195, 134, 163, 230, 161, 255, 231, 252}, + {196, 134, 163, 230, 161, 255, 231, 252}, + {197, 135, 163, 231, 162, 255, 231, 252}, + {198, 135, 163, 231, 162, 255, 231, 252}, + {199, 136, 164, 232, 163, 255, 232, 252}, + {200, 136, 164, 232, 164, 255, 232, 252}, + {201, 137, 165, 233, 165, 255, 233, 252}, + {201, 137, 165, 233, 165, 255, 233, 252}, + {202, 138, 166, 233, 166, 255, 233, 252}, + {203, 138, 166, 233, 166, 255, 233, 252}, + {204, 139, 166, 234, 167, 255, 234, 252}, + {205, 139, 166, 234, 167, 255, 234, 252}, + {206, 140, 167, 235, 168, 255, 235, 252}, + {206, 140, 167, 235, 168, 255, 235, 252}, + {207, 141, 168, 236, 169, 255, 235, 252}, + {208, 141, 168, 236, 170, 255, 235, 252}, + {209, 142, 169, 237, 171, 255, 236, 252}, + {209, 143, 169, 237, 171, 255, 236, 252}, + {210, 144, 169, 237, 172, 255, 236, 252}, + {211, 144, 169, 237, 172, 255, 236, 252}, + {212, 145, 170, 238, 173, 255, 237, 252}, + {213, 145, 170, 238, 173, 255, 237, 252}, + {214, 146, 171, 239, 174, 255, 237, 253}, + {214, 146, 171, 239, 174, 255, 237, 253}, + {215, 147, 172, 240, 175, 255, 238, 253}, + {215, 147, 172, 240, 175, 255, 238, 253}, + {216, 148, 173, 240, 176, 255, 238, 253}, + {217, 148, 173, 240, 176, 255, 238, 253}, + {218, 149, 173, 241, 177, 255, 239, 253}, + {218, 149, 173, 241, 178, 255, 239, 253}, + {219, 150, 174, 241, 179, 255, 239, 253}, + {219, 151, 174, 241, 179, 255, 239, 253}, + {220, 152, 175, 242, 180, 255, 240, 253}, + {221, 152, 175, 242, 180, 255, 240, 253}, + {222, 153, 176, 242, 181, 255, 240, 253}, + {222, 153, 176, 242, 181, 255, 240, 253}, + {223, 154, 177, 243, 182, 255, 240, 253}, + {223, 154, 177, 243, 182, 255, 240, 253}, + {224, 155, 178, 244, 183, 255, 241, 253}, + {224, 155, 178, 244, 183, 255, 241, 253}, + {225, 156, 178, 244, 184, 255, 241, 253}, + {225, 157, 178, 244, 184, 255, 241, 253}, + {226, 158, 179, 244, 185, 255, 242, 253}, + {227, 158, 179, 244, 185, 255, 242, 253}, + {228, 159, 180, 245, 186, 255, 242, 253}, + {228, 159, 180, 245, 186, 255, 242, 253}, + {229, 160, 181, 245, 187, 255, 242, 253}, + {229, 160, 181, 245, 187, 255, 242, 253}, + {230, 161, 182, 246, 188, 255, 243, 253}, + {230, 162, 182, 246, 188, 255, 243, 253}, + {231, 163, 183, 246, 189, 255, 243, 253}, + {231, 163, 183, 246, 189, 255, 243, 253}, + {232, 164, 184, 247, 190, 255, 243, 253}, + {232, 164, 184, 247, 190, 255, 243, 253}, + {233, 165, 185, 247, 191, 255, 244, 253}, + {233, 165, 185, 247, 191, 255, 244, 253}, + {234, 166, 185, 247, 192, 255, 244, 253}, + {234, 167, 185, 247, 192, 255, 244, 253}, + {235, 168, 186, 248, 193, 255, 244, 253}, + {235, 168, 186, 248, 193, 255, 244, 253}, + {236, 169, 187, 248, 194, 255, 244, 253}, + {236, 169, 187, 248, 194, 255, 244, 253}, + {236, 170, 188, 248, 195, 255, 245, 253}, + {236, 170, 188, 248, 195, 255, 245, 253}, + {237, 171, 189, 249, 196, 255, 245, 254}, + {237, 172, 189, 249, 196, 255, 245, 254}, + {238, 173, 190, 249, 197, 255, 245, 254}, + {238, 173, 190, 249, 197, 255, 245, 254}, + {239, 174, 191, 249, 198, 255, 245, 254}, + {239, 174, 191, 249, 198, 255, 245, 254}, + {240, 175, 192, 249, 199, 255, 246, 254}, + {240, 176, 192, 249, 199, 255, 246, 254}, + {240, 177, 193, 250, 200, 255, 246, 254}, + {240, 177, 193, 250, 200, 255, 246, 254}, + {241, 178, 194, 250, 201, 255, 246, 254}, + {241, 178, 194, 250, 201, 255, 246, 254}, + {242, 179, 195, 250, 202, 255, 246, 254}, + {242, 180, 195, 250, 202, 255, 246, 254}, + {242, 181, 196, 250, 203, 255, 247, 254}, + {242, 181, 196, 250, 203, 255, 247, 254}, + {243, 182, 197, 251, 204, 255, 247, 254}, + {243, 183, 197, 251, 204, 255, 247, 254}, + {244, 184, 198, 251, 205, 255, 247, 254}, + {244, 184, 198, 251, 205, 255, 247, 254}, + {244, 185, 199, 251, 206, 255, 247, 254}, + {244, 185, 199, 251, 206, 255, 247, 254}, + {245, 186, 200, 251, 207, 255, 247, 254}, + {245, 187, 200, 251, 207, 255, 247, 254}, + {246, 188, 201, 252, 207, 255, 248, 254}, + {246, 188, 201, 252, 207, 255, 248, 254}, + {246, 189, 202, 252, 208, 255, 248, 254}, + {246, 190, 202, 252, 208, 255, 248, 254}, + {247, 191, 203, 252, 209, 255, 248, 254}, + {247, 191, 203, 252, 209, 255, 248, 254}, + {247, 192, 204, 252, 210, 255, 248, 254}, + {247, 193, 204, 252, 210, 255, 248, 254}, + {248, 194, 205, 252, 211, 255, 248, 254}, + {248, 194, 205, 252, 211, 255, 248, 254}, + {248, 195, 206, 252, 212, 255, 249, 254}, + {248, 196, 206, 252, 212, 255, 249, 254}, + {249, 197, 207, 253, 213, 255, 249, 254}, + {249, 197, 207, 253, 213, 255, 249, 254}, + {249, 198, 208, 253, 214, 255, 249, 254}, + {249, 199, 209, 253, 214, 255, 249, 254}, + {250, 200, 210, 253, 215, 255, 249, 254}, + {250, 200, 210, 253, 215, 255, 249, 254}, + {250, 201, 211, 253, 215, 255, 249, 254}, + {250, 202, 211, 253, 215, 255, 249, 254}, + {250, 203, 212, 253, 216, 255, 249, 254}, + {250, 203, 212, 253, 216, 255, 249, 254}, + {251, 204, 213, 253, 217, 255, 250, 254}, + {251, 205, 213, 253, 217, 255, 250, 254}, + {251, 206, 214, 254, 218, 255, 250, 254}, + {251, 206, 215, 254, 218, 255, 250, 254}, + {252, 207, 216, 254, 219, 255, 250, 254}, + {252, 208, 216, 254, 219, 255, 250, 254}, + {252, 209, 217, 254, 220, 255, 250, 254}, + {252, 210, 217, 254, 220, 255, 250, 254}, + {252, 211, 218, 254, 221, 255, 250, 254}, + {252, 212, 218, 254, 221, 255, 250, 254}, + {253, 213, 219, 254, 222, 255, 250, 254}, + {253, 213, 220, 254, 222, 255, 250, 254}, + {253, 214, 221, 254, 223, 255, 250, 254}, + {253, 215, 221, 254, 223, 255, 250, 254}, + {253, 216, 222, 254, 224, 255, 251, 254}, + {253, 217, 223, 254, 224, 255, 251, 254}, + {253, 218, 224, 254, 225, 255, 251, 254}, + {253, 219, 224, 254, 225, 255, 251, 254}, + {254, 220, 225, 254, 225, 255, 251, 254}, + {254, 221, 226, 254, 225, 255, 251, 254}, + {254, 222, 227, 255, 226, 255, 251, 254}, + {254, 223, 227, 255, 226, 255, 251, 254}, + {254, 224, 228, 255, 227, 255, 251, 254}, + {254, 225, 229, 255, 227, 255, 251, 254}, + {254, 226, 230, 255, 228, 255, 251, 254}, + {254, 227, 230, 255, 229, 255, 251, 254}, + {255, 228, 231, 255, 230, 255, 251, 254}, + {255, 229, 232, 255, 230, 255, 251, 254}, + {255, 230, 233, 255, 231, 255, 252, 254}, + {255, 231, 234, 255, 231, 255, 252, 254}, + {255, 232, 235, 255, 232, 255, 252, 254}, + {255, 233, 236, 255, 232, 255, 252, 254}, + {255, 235, 237, 255, 233, 255, 252, 254}, + {255, 236, 238, 255, 234, 255, 252, 254}, + {255, 238, 240, 255, 235, 255, 252, 255}, + {255, 239, 241, 255, 235, 255, 252, 254}, + {255, 241, 243, 255, 236, 255, 252, 254}, + {255, 243, 245, 255, 237, 255, 252, 254}, + {255, 246, 247, 255, 239, 255, 253, 255}, +}; + static void extend_to_full_distribution(vp9_prob *probs, vp9_prob p) { const int l = (p - 1) / 2; if (p & 1) { diff --git a/vp9/common/vp9_entropy.h b/vp9/common/vp9_entropy.h index e133d6509..9ab918fa3 100644 --- a/vp9/common/vp9_entropy.h +++ b/vp9/common/vp9_entropy.h @@ -17,6 +17,7 @@ #include "vp9/common/vp9_common.h" #include "vp9/common/vp9_scan.h" #include "vp9/common/vp9_treecoder.h" +#include "vp9/common/vp9_entropymode.h" #define DIFF_UPDATE_PROB 252 @@ -141,6 +142,9 @@ static const uint8_t *get_band_translate(TX_SIZE tx_size) { #define PIVOT_NODE 2 // which node is pivot +#define MODEL_NODES (ENTROPY_NODES - UNCONSTRAINED_NODES) +extern const vp9_prob vp9_pareto8_full[255][MODEL_NODES]; + typedef vp9_prob vp9_coeff_probs_model[REF_TYPES][COEF_BANDS] [PREV_COEF_CONTEXTS] [UNCONSTRAINED_NODES]; diff --git a/vp9/decoder/vp9_detokenize.c b/vp9/decoder/vp9_detokenize.c index f6b97bb77..522d66602 100644 --- a/vp9/decoder/vp9_detokenize.c +++ b/vp9/decoder/vp9_detokenize.c @@ -23,14 +23,14 @@ #define EOB_CONTEXT_NODE 0 #define ZERO_CONTEXT_NODE 1 #define ONE_CONTEXT_NODE 2 -#define LOW_VAL_CONTEXT_NODE 3 -#define TWO_CONTEXT_NODE 4 -#define THREE_CONTEXT_NODE 5 -#define HIGH_LOW_CONTEXT_NODE 6 -#define CAT_ONE_CONTEXT_NODE 7 -#define CAT_THREEFOUR_CONTEXT_NODE 8 -#define CAT_THREE_CONTEXT_NODE 9 -#define CAT_FIVE_CONTEXT_NODE 10 +#define LOW_VAL_CONTEXT_NODE 0 +#define TWO_CONTEXT_NODE 1 +#define THREE_CONTEXT_NODE 2 +#define HIGH_LOW_CONTEXT_NODE 3 +#define CAT_ONE_CONTEXT_NODE 4 +#define CAT_THREEFOUR_CONTEXT_NODE 5 +#define CAT_THREE_CONTEXT_NODE 6 +#define CAT_FIVE_CONTEXT_NODE 7 #define CAT1_MIN_VAL 5 #define CAT2_MIN_VAL 7 @@ -76,14 +76,16 @@ static const int token_to_counttoken[MAX_ENTROPY_TOKENS] = { #define WRITE_COEF_CONTINUE(val, token) \ { \ - dqcoeff_ptr[scan[c]] = (vp9_read_bit(r) ? -val : val) * \ - dq[c > 0] / (1 + (tx_size == TX_32X32)); \ + v = (val * dqv) >> dq_shift; \ + dqcoeff_ptr[scan[c]] = (vp9_read_bit(r) ? -v : v); \ INCREMENT_COUNT(token); \ token_cache[scan[c]] = vp9_pt_energy_class[token]; \ ++c; \ + dqv = dq[1]; \ continue; \ } + #define ADJUST_COEF(prob, bits_count) \ do { \ val += (vp9_read(r, prob) << bits_count); \ @@ -100,8 +102,6 @@ static int decode_coefs(VP9_COMMON *cm, const MACROBLOCKD *xd, int band, c = 0; const vp9_prob (*coef_probs)[PREV_COEF_CONTEXTS][UNCONSTRAINED_NODES] = fc->coef_probs[tx_size][type][ref]; - vp9_prob coef_probs_full[COEF_BANDS][PREV_COEF_CONTEXTS][ENTROPY_NODES]; - uint8_t load_map[COEF_BANDS][PREV_COEF_CONTEXTS] = { { 0 } }; const vp9_prob *prob; unsigned int (*coef_counts)[PREV_COEF_CONTEXTS][UNCONSTRAINED_NODES + 1] = counts->coef[tx_size][type][ref]; @@ -110,9 +110,12 @@ static int decode_coefs(VP9_COMMON *cm, const MACROBLOCKD *xd, const int16_t *scan, *nb; const uint8_t *cat6; const uint8_t *band_translate = get_band_translate(tx_size); + const int dq_shift = (tx_size == TX_32X32); const MODE_INFO *const mi = xd->mi_8x8[0]; const MB_MODE_INFO *const mbmi = &mi->mbmi; scan_order const *so; + int v; + int16_t dqv = dq[0]; if (mbmi->ref_frame[0] > 0 || type != PLANE_TYPE_Y_WITH_DC || xd->lossless) { so = &inter_scan_orders[tx_size]; @@ -148,6 +151,7 @@ static int decode_coefs(VP9_COMMON *cm, const MACROBLOCKD *xd, if (!vp9_read(r, prob[ZERO_CONTEXT_NODE])) { INCREMENT_COUNT(ZERO_TOKEN); token_cache[scan[c]] = vp9_pt_energy_class[ZERO_TOKEN]; + dqv = dq[1]; \ ++c; goto SKIP_START; } @@ -156,13 +160,9 @@ static int decode_coefs(VP9_COMMON *cm, const MACROBLOCKD *xd, if (!vp9_read(r, prob[ONE_CONTEXT_NODE])) { WRITE_COEF_CONTINUE(1, ONE_TOKEN); } - // Load full probabilities if not already loaded - if (!load_map[band][pt]) { - vp9_model_to_full_probs(coef_probs[band][pt], - coef_probs_full[band][pt]); - load_map[band][pt] = 1; - } - prob = coef_probs_full[band][pt]; + + prob = vp9_pareto8_full[coef_probs[band][pt][PIVOT_NODE]-1]; + // LOW_VAL_CONTEXT_NODE_0_ if (!vp9_read(r, prob[LOW_VAL_CONTEXT_NODE])) { if (!vp9_read(r, prob[TWO_CONTEXT_NODE])) { -- 2.40.0