introduce GetLargeValue() to slim-fast GetCoeffs().

GetCoeffs is (by far) the most consuming function of the decoder.
No speed change (unfortunately), but the main loop is somehow clearer.

Change-Id: I78f1c10cadc2c8696c041f5cbda86cab92cc6598
This commit is contained in:
skal 2012-11-28 08:24:23 +01:00
parent d5838cd598
commit c34a3758ad

View File

@ -458,7 +458,7 @@ int VP8GetHeaders(VP8Decoder* const dec, VP8Io* const io) {
//------------------------------------------------------------------------------
// Residual decoding (Paragraph 13.2 / 13.3)
static const uint8_t kBands[16 + 1] = {
static const int kBands[16 + 1] = {
0, 1, 2, 3, 6, 4, 5, 6, 6, 6, 6, 6, 6, 6, 6, 7,
0 // extra entry as sentinel
};
@ -474,26 +474,11 @@ static const uint8_t kZigzag[16] = {
};
typedef const uint8_t (*ProbaArray)[NUM_CTX][NUM_PROBAS]; // for const-casting
typedef const uint8_t (*ProbaCtxArray)[NUM_PROBAS];
// Returns the position of the last non-zero coeff plus one
// (and 0 if there's no coeff at all)
static int GetCoeffs(VP8BitReader* const br, ProbaArray prob,
int ctx, const quant_t dq, int n, int16_t* out) {
// n is either 0 or 1 here. kBands[n] is not necessary for extracting '*p'.
const uint8_t* p = prob[n][ctx];
if (!VP8GetBit(br, p[0])) { // first EOB is more a 'CBP' bit.
return 0;
}
while (1) {
++n;
if (!VP8GetBit(br, p[1])) {
p = prob[kBands[n]][0];
} else { // non zero coeff
int v, j;
if (!VP8GetBit(br, p[2])) {
p = prob[kBands[n]][1];
v = 1;
} else {
// See section 13-2: http://tools.ietf.org/html/rfc6386#section-13.2
static int GetLargeValue(VP8BitReader* const br, const uint8_t* const p) {
int v;
if (!VP8GetBit(br, p[3])) {
if (!VP8GetBit(br, p[4])) {
v = 2;
@ -520,19 +505,39 @@ static int GetCoeffs(VP8BitReader* const br, ProbaArray prob,
v += 3 + (8 << cat);
}
}
p = prob[kBands[n]][2];
return v;
}
// Returns the position of the last non-zero coeff plus one
// (and 0 if there's no coeff at all)
static int GetCoeffs(VP8BitReader* const br, ProbaArray prob,
int ctx, const quant_t dq, int n, int16_t* out) {
// n is either 0 or 1 here. kBands[n] is not necessary for extracting '*p'.
const uint8_t* p = prob[n][ctx];
if (!VP8GetBit(br, p[0])) { // first EOB is more a 'CBP' bit.
return 0;
}
for (; n < 16; ++n) {
const ProbaCtxArray p_ctx = prob[kBands[n + 1]];
if (!VP8GetBit(br, p[1])) {
p = p_ctx[0];
} else { // non zero coeff
int v;
if (!VP8GetBit(br, p[2])) {
v = 1;
p = p_ctx[1];
} else {
v = GetLargeValue(br, p);
p = p_ctx[2];
}
out[kZigzag[n]] = VP8GetSigned(br, v) * dq[n > 0];
if (n < 15 && !VP8GetBit(br, p[0])) { // EOB
return n + 1;
}
j = kZigzag[n - 1];
out[j] = VP8GetSigned(br, v) * dq[j > 0];
if (n == 16 || !VP8GetBit(br, p[0])) { // EOB
return n;
}
}
if (n == 16) {
return 16;
}
}
}
// Alias-safe way of converting 4bytes to 32bits.
typedef union {