ARM64: Various optimizations for symbol decode
Changes stem from redesigning the reduction stage of the multisymbol decode function. * No longer use adapt4 for 5 possible symbol values * Specialize reduction for 4/8/16 decode functions * Modify control flow +------------------------+--------------+--------------+---------------+ | | Neoverse V1 | Neoverse N1 | Cortex A72 | | | (Graviton 3) | (Graviton 2) | (Graviton 1) | +------------------------+-------+------+-------+------+-------+-------+ | | Old | New | Old | New | Old | New | +------------------------+-------+------+-------+------+-------+-------+ | decode_bool_neon | 13.0 | 12.9 | 14.9 | 14.0 | 39.3 | 29.0 | +------------------------+-------+------+-------+------+-------+-------+ | decode_bool_adapt_neon | 15.4 | 15.6 | 17.5 | 16.8 | 41.6 | 33.5 | +------------------------+-------+------+-------+------+-------+-------+ | decode_bool_equi_neon | 11.3 | 12.0 | 14.0 | 12.2 | 35.0 | 26.3 | +------------------------+-------+------+-------+------+-------+-------+ | decode_hi_tok_c | 73.7 | 57.8 | 73.4 | 60.5 | 130.1 | 103.9 | +------------------------+-------+------+-------+------+-------+-------+ | decode_hi_tok_neon | 63.3 | 48.2 | 65.2 | 51.2 | 119.0 | 105.3 | +------------------------+-------+------+-------+------+-------+-------+ | decode_symbol_\ | 28.6 | 22.5 | 28.4 | 23.5 | 67.8 | 55.1 | | adapt4_neon | | | | | | | +------------------------+-------+------+-------+------+-------+-------+ | decode_symbol_\ | 29.5 | 26.6 | 29.0 | 28.8 | 76.6 | 74.0 | | adapt8_neon | | | | | | | +------------------------+-------+------+-------+------+-------+-------+ | decode_symbol_\ | 31.6 | 31.2 | 33.3 | 33.0 | 77.5 | 68.1 | | adapt16_neon | | | | | | | +------------------------+-------+------+-------+------+-------+-------+
This commit is contained in:
Kyle Siefring
+180
-62
@@ -35,9 +35,14 @@
|
||||
#define CNT 28
|
||||
#define ALLOW_UPDATE_CDF 32
|
||||
|
||||
#define COEFFS_BASE_OFFSET 30
|
||||
#define MASKS8_OFFSET (64-COEFFS_BASE_OFFSET)
|
||||
|
||||
const coeffs
|
||||
.short 60, 56, 52, 48, 44, 40, 36, 32, 28, 24, 20, 16, 12, 8, 4, 0
|
||||
.short 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
|
||||
// masks8
|
||||
.short -0x202, -0x202, -0x202, -0x202, -0x202, -0x202, -0x202, 0xF0E
|
||||
endconst
|
||||
|
||||
.macro ld1_n d0, d1, src, sz, n
|
||||
@@ -117,41 +122,70 @@ endconst
|
||||
|
||||
function msac_decode_symbol_adapt4_neon, export=1
|
||||
.macro decode_update sz, szb, n
|
||||
.if \n == 16
|
||||
sub sp, sp, #48
|
||||
.endif
|
||||
add x8, x0, #RNG
|
||||
ld1_n v0, v1, x1, \sz, \n // cdf
|
||||
ld1r {v4\sz}, [x8] // rng
|
||||
movrel x9, coeffs, 30
|
||||
ld1r {v29\sz}, [x8] // rng
|
||||
movrel x9, coeffs, COEFFS_BASE_OFFSET
|
||||
movi v31\sz, #0x7f, lsl #8 // 0x7f00
|
||||
sub x9, x9, x2, lsl #1
|
||||
sub x10, x9, x2, lsl #1
|
||||
mvni v30\sz, #0x3f // 0xffc0
|
||||
and v7\szb, v4\szb, v31\szb // rng & 0x7f00
|
||||
str h4, [sp, #14] // store original u = s->rng
|
||||
and v7\szb, v29\szb, v31\szb // rng & 0x7f00
|
||||
.if \n == 16
|
||||
str h29, [sp, #14] // store original u = s->rng
|
||||
.endif
|
||||
and_n v2, v3, v0, v1, v30, v30, \szb, \n // cdf & 0xffc0
|
||||
|
||||
ld1_n v4, v5, x9, \sz, \n // EC_MIN_PROB * (n_symbols - ret)
|
||||
ld1_n v4, v5, x10, \sz, \n // EC_MIN_PROB * (n_symbols - ret)
|
||||
sqdmulh_n v6, v7, v2, v3, v7, v7, \sz, \n // ((cdf >> EC_PROB_SHIFT) * (r - 128)) >> 1
|
||||
add x8, x0, #DIF + 6
|
||||
ldr d28, [x0, #DIF]
|
||||
|
||||
add_n v4, v5, v2, v3, v4, v5, \sz, \n // v = cdf + EC_MIN_PROB * (n_symbols - ret)
|
||||
add_n v4, v5, v6, v7, v4, v5, \sz, \n // v = ((cdf >> EC_PROB_SHIFT) * r) >> 1 + EC_MIN_PROB * (n_symbols - ret)
|
||||
|
||||
ld1r {v6\sz}, [x8] // dif >> (EC_WIN_SIZE - 16)
|
||||
dup v30\sz, v28.h[3] // dif >> (EC_WIN_SIZE - 16)
|
||||
.if \n == 8
|
||||
ldr q31, [x9, #MASKS8_OFFSET]
|
||||
.elseif \n == 16
|
||||
str_n q4, q5, sp, #16, \n // store v values to allow indexed access
|
||||
|
||||
cmhs_n v2, v3, v6, v6, v4, v5, \sz, \n // c >= v
|
||||
|
||||
.if \n == 16
|
||||
add v6\sz, v2\sz, v3\sz
|
||||
addv h6, v6\sz // -n + ret
|
||||
.else
|
||||
addv h6, v2\sz // -n + ret
|
||||
.endif
|
||||
ldr w4, [x0, #ALLOW_UPDATE_CDF]
|
||||
smov w15, v6.h[0]
|
||||
add w15, w15, #\n // ret
|
||||
|
||||
cbz w4, L(renorm)
|
||||
// After the condition starts being true it continues, such that the vector looks like:
|
||||
// 0, 0, 0 ... -1, -1
|
||||
cmhs_n v2, v3, v30, v30, v4, v5, \sz, \n // c >= v
|
||||
.if \n == 4
|
||||
ext v29\szb, v29\szb, v4\szb, #6 // u
|
||||
umov x15, v2.d[0]
|
||||
ldr w4, [x0, #ALLOW_UPDATE_CDF]
|
||||
rev x15, x15
|
||||
sub v29\sz, v29\sz, v4\sz // rng = u-v
|
||||
// rev + clz = count trailing zeros
|
||||
clz x15, x15 // 16*ret
|
||||
.elseif \n == 8
|
||||
// The final short of the compare is always set.
|
||||
// Using addv, subtract -0x202*ret from this value to create a lookup table for a short.
|
||||
// For n == 8:
|
||||
// -0x202 + -0x202 + ... + 0xF0E
|
||||
// (0x202*7) | (1 << 8)
|
||||
// ^-------offset for second byte of the short
|
||||
and v31\szb, v31\szb, v2\szb
|
||||
ext v29\szb, v29\szb, v4\szb, #14 // u
|
||||
addv h31, v31\sz // ((2*ret + 1) << 8) | (2*ret)
|
||||
ldr w4, [x0, #ALLOW_UPDATE_CDF]
|
||||
sub v30\sz, v30\sz, v4\sz // (dif >> 48) - v
|
||||
smov w15, v31.b[0] // 2*ret
|
||||
sub v29\sz, v29\sz, v4\sz // rng = u-v
|
||||
.elseif \n == 16
|
||||
add v6\sz, v2\sz, v3\sz
|
||||
addv h31, v6\sz // -n + ret
|
||||
ldr w4, [x0, #ALLOW_UPDATE_CDF]
|
||||
smov w15, v31.h[0]
|
||||
.endif
|
||||
|
||||
cbz w4, 0f
|
||||
|
||||
// update_cdf
|
||||
ldrh w3, [x1, x2, lsl #1] // count = cdf[n_symbols]
|
||||
.if \n == 16
|
||||
@@ -181,27 +215,88 @@ function msac_decode_symbol_adapt4_neon, export=1
|
||||
add_n v0, v1, v16, v17, v2, v3, \sz, \n // cdf + (32768 - cdf[i]) >> rate
|
||||
st1_n v0, v1, x1, \sz, \n
|
||||
strh w3, [x1, x2, lsl #1]
|
||||
.endm
|
||||
|
||||
decode_update .4h, .8b, 4
|
||||
0:
|
||||
// renorm
|
||||
.if \n == 4
|
||||
ldr w6, [x0, #CNT]
|
||||
ldr x7, [x0, #DIF]
|
||||
mov x4, v29.d[0] // rng (packed)
|
||||
mov x3, v4.d[0] // v (packed)
|
||||
|
||||
L(renorm):
|
||||
add x8, sp, w15, uxtw #1
|
||||
ldrh w3, [x8, #16] // v
|
||||
ldurh w4, [x8, #14] // u
|
||||
// Shift 'v'/'rng' for ret into the 16 least sig bits. There is
|
||||
// garbage in the remaining bits, but we can work around this.
|
||||
lsr x4, x4, x15 // rng
|
||||
lsr x3, x3, x15 // v
|
||||
lsl w5, w4, #16 // rng << 16
|
||||
sub x7, x7, x3, lsl #48 // dif - (v << 48)
|
||||
clz w5, w5 // d = clz(rng << 16)
|
||||
lsl w4, w4, w5 // rng << d
|
||||
subs w6, w6, w5 // cnt -= d
|
||||
lsl x7, x7, x5 // (dif - (v << 48)) << d
|
||||
strh w4, [x0, #RNG]
|
||||
b.lo 1f
|
||||
str w6, [x0, #CNT]
|
||||
str x7, [x0, #DIF]
|
||||
lsr w0, w15, #4
|
||||
ret
|
||||
1:
|
||||
lsr w15, w15, #4
|
||||
b L(refill)
|
||||
.elseif \n == 8
|
||||
ldr w6, [x0, #CNT]
|
||||
tbl v30.8b, {v30.16b}, v31.8b
|
||||
tbl v29.8b, {v29.16b}, v31.8b
|
||||
ins v28.h[3], v30.h[0] // dif - (v << 48)
|
||||
clz v0.4h, v29.4h // d = clz(rng)
|
||||
umov w5, v0.h[0]
|
||||
ushl v29.4h, v29.4h, v0.4h // rng << d
|
||||
|
||||
// The vec for clz(rng) is filled with garbage after the first short,
|
||||
// but ushl/sshl conveniently uses only the first byte for the shift
|
||||
// amount.
|
||||
ushl d28, d28, d0 // (dif - (v << 48)) << d
|
||||
|
||||
subs w6, w6, w5 // cnt -= d
|
||||
str h29, [x0, #RNG]
|
||||
b.lo 1f
|
||||
str w6, [x0, #CNT]
|
||||
str d28, [x0, #DIF]
|
||||
lsr w0, w15, #1 // ret
|
||||
ret
|
||||
1:
|
||||
lsr w15, w15, #1 // ret
|
||||
mov x7, v28.d[0]
|
||||
b L(refill)
|
||||
.elseif \n == 16
|
||||
add x8, sp, w15, sxtw #1
|
||||
ldrh w3, [x8, #48] // v
|
||||
ldurh w4, [x8, #46] // u
|
||||
ldr w6, [x0, #CNT]
|
||||
ldr x7, [x0, #DIF]
|
||||
sub w4, w4, w3 // rng = u - v
|
||||
clz w5, w4 // clz(rng)
|
||||
eor w5, w5, #16 // d = clz(rng) ^ 16
|
||||
sub x7, x7, x3, lsl #48 // dif - (v << 48)
|
||||
L(renorm2):
|
||||
lsl w4, w4, w5 // rng << d
|
||||
subs w6, w6, w5 // cnt -= d
|
||||
lsl x7, x7, x5 // (dif - (v << 48)) << d
|
||||
str w4, [x0, #RNG]
|
||||
b.hs 4f
|
||||
add sp, sp, #48
|
||||
b.lo 1f
|
||||
str w6, [x0, #CNT]
|
||||
str x7, [x0, #DIF]
|
||||
add w0, w15, #\n // ret
|
||||
ret
|
||||
1:
|
||||
add w15, w15, #\n // ret
|
||||
b L(refill)
|
||||
.endif
|
||||
.endm
|
||||
|
||||
decode_update .4h, .8b, 4
|
||||
|
||||
L(refill):
|
||||
// refill
|
||||
ldp x3, x4, [x0] // BUF_POS, BUF_END
|
||||
add x5, x3, #8
|
||||
@@ -229,7 +324,6 @@ L(renorm2):
|
||||
str x7, [x0, #DIF]
|
||||
|
||||
mov w0, w15
|
||||
add sp, sp, #48
|
||||
ret
|
||||
|
||||
5: // pad_with_ones
|
||||
@@ -258,29 +352,26 @@ endfunc
|
||||
|
||||
function msac_decode_symbol_adapt8_neon, export=1
|
||||
decode_update .8h, .16b, 8
|
||||
b L(renorm)
|
||||
endfunc
|
||||
|
||||
function msac_decode_symbol_adapt16_neon, export=1
|
||||
decode_update .8h, .16b, 16
|
||||
b L(renorm)
|
||||
endfunc
|
||||
|
||||
function msac_decode_hi_tok_neon, export=1
|
||||
ld1 {v0.4h}, [x1] // cdf
|
||||
add x16, x0, #RNG
|
||||
movi v31.4h, #0x7f, lsl #8 // 0x7f00
|
||||
movrel x17, coeffs, 30-2*3
|
||||
movrel x17, coeffs, COEFFS_BASE_OFFSET-2*3
|
||||
mvni v30.4h, #0x3f // 0xffc0
|
||||
ldrh w9, [x1, #6] // count = cdf[n_symbols]
|
||||
ld1r {v3.4h}, [x16] // rng
|
||||
ld1 {v29.4h}, [x17] // EC_MIN_PROB * (n_symbols - ret)
|
||||
add x17, x0, #DIF + 6
|
||||
mov w13, #-24
|
||||
mov w13, #-24*8
|
||||
and v17.8b, v0.8b, v30.8b // cdf & 0xffc0
|
||||
ldr w10, [x0, #ALLOW_UPDATE_CDF]
|
||||
ld1r {v1.8h}, [x17] // dif >> (EC_WIN_SIZE - 16)
|
||||
sub sp, sp, #48
|
||||
ldr w6, [x0, #CNT]
|
||||
ldr x7, [x0, #DIF]
|
||||
1:
|
||||
@@ -288,13 +379,14 @@ function msac_decode_hi_tok_neon, export=1
|
||||
sqdmulh v6.4h, v17.4h, v7.4h // ((cdf >> EC_PROB_SHIFT) * (r - 128)) >> 1
|
||||
add v4.4h, v17.4h, v29.4h // v = cdf + EC_MIN_PROB * (n_symbols - ret)
|
||||
add v4.4h, v6.4h, v4.4h // v = ((cdf >> EC_PROB_SHIFT) * r) >> 1 + EC_MIN_PROB * (n_symbols - ret)
|
||||
str h3, [sp, #14] // store original u = s->rng
|
||||
cmhs v2.4h, v1.4h, v4.4h // c >= v
|
||||
str q4, [sp, #16] // store v values to allow indexed access
|
||||
addv h6, v2.4h // -4 + ret
|
||||
add w13, w13, #5
|
||||
smov w15, v6.h[0]
|
||||
add w15, w15, #4 // ret
|
||||
add w13, w13, #5*8
|
||||
ext v18.8b, v3.8b, v4.8b, #6 // u
|
||||
umov x15, v2.d[0]
|
||||
rev x15, x15
|
||||
sub v18.4h, v18.4h, v4.4h // rng = u-v
|
||||
// rev + clz = count trailing zeros
|
||||
clz x15, x15 // 16*ret
|
||||
|
||||
cbz w10, 2f
|
||||
// update_cdf
|
||||
@@ -302,29 +394,32 @@ function msac_decode_hi_tok_neon, export=1
|
||||
mov w4, #-5
|
||||
orr v2.4h, #0x80, lsl #8 // i >= val ? -1 : 32768
|
||||
sub w4, w4, w9, lsr #4 // -((count >> 4) + 5)
|
||||
sub v4.4h, v2.4h, v0.4h // (32768 - cdf[i]) or (-1 - cdf[i])
|
||||
sub v2.4h, v2.4h, v0.4h // (32768 - cdf[i]) or (-1 - cdf[i])
|
||||
dup v6.4h, w4 // -rate
|
||||
|
||||
sub w9, w9, w9, lsr #5 // count - (count == 32)
|
||||
sshl v4.4h, v4.4h, v6.4h // ({32768,-1} - cdf[i]) >> rate
|
||||
sshl v2.4h, v2.4h, v6.4h // ({32768,-1} - cdf[i]) >> rate
|
||||
add w9, w9, #1 // count + (count < 32)
|
||||
add v0.4h, v5.4h, v4.4h // cdf[i] + (32768 - cdf[i]) >> rate
|
||||
add v0.4h, v5.4h, v2.4h // cdf[i] + (32768 - cdf[i]) >> rate
|
||||
st1 {v0.4h}, [x1]
|
||||
and v17.8b, v0.8b, v30.8b // cdf & 0xffc0
|
||||
strh w9, [x1, #6]
|
||||
|
||||
2:
|
||||
add x8, sp, w15, uxtw #1
|
||||
ldrh w3, [x8, #16] // v
|
||||
ldurh w4, [x8, #14] // u
|
||||
sub w4, w4, w3 // rng = u - v
|
||||
clz w5, w4 // clz(rng)
|
||||
eor w5, w5, #16 // d = clz(rng) ^ 16
|
||||
mov x4, v18.d[0] // rng (packed)
|
||||
mov x3, v4.d[0] // v (packed)
|
||||
|
||||
// Shift 'v'/'rng' for ret into the 16 least sig bits. There is
|
||||
// garbage in the remaining bits, but we can work around this.
|
||||
lsr x4, x4, x15 // rng
|
||||
lsr x3, x3, x15 // v
|
||||
lsl w5, w4, #16 // rng << 16
|
||||
sub x7, x7, x3, lsl #48 // dif - (v << 48)
|
||||
clz w5, w5 // d = clz(rng << 16)
|
||||
lsl w4, w4, w5 // rng << d
|
||||
subs w6, w6, w5 // cnt -= d
|
||||
lsl x7, x7, x5 // (dif - (v << 48)) << d
|
||||
str w4, [x0, #RNG]
|
||||
strh w4, [x0, #RNG]
|
||||
dup v3.4h, w4
|
||||
b.hs 5f
|
||||
|
||||
@@ -351,17 +446,15 @@ function msac_decode_hi_tok_neon, export=1
|
||||
orr x7, x7, x8 // dif |= next_bits
|
||||
|
||||
5: // end
|
||||
lsl w15, w15, #1
|
||||
sub w15, w15, #5
|
||||
sub w15, w15, #5*8
|
||||
lsr x12, x7, #48
|
||||
adds w13, w13, w15 // carry = tok_br < 3 || tok == 15
|
||||
dup v1.8h, w12
|
||||
b.cc 1b // loop if !carry
|
||||
add w13, w13, #30
|
||||
add w13, w13, #30*8
|
||||
str w6, [x0, #CNT]
|
||||
add sp, sp, #48
|
||||
str x7, [x0, #DIF]
|
||||
lsr w0, w13, #1
|
||||
lsr w0, w13, #4
|
||||
ret
|
||||
|
||||
6: // pad_with_ones
|
||||
@@ -390,7 +483,6 @@ endfunc
|
||||
|
||||
function msac_decode_bool_equi_neon, export=1
|
||||
ldp w5, w6, [x0, #RNG] // + CNT
|
||||
sub sp, sp, #48
|
||||
ldr x7, [x0, #DIF]
|
||||
bic w4, w5, #0xff // r &= 0xff00
|
||||
add w4, w4, #8
|
||||
@@ -403,12 +495,20 @@ function msac_decode_bool_equi_neon, export=1
|
||||
|
||||
clz w5, w4 // clz(rng)
|
||||
eor w5, w5, #16 // d = clz(rng) ^ 16
|
||||
b L(renorm2)
|
||||
lsl w4, w4, w5 // rng << d
|
||||
subs w6, w6, w5 // cnt -= d
|
||||
lsl x7, x7, x5 // (dif - (v << 48)) << d
|
||||
str w4, [x0, #RNG]
|
||||
b.lo L(refill)
|
||||
|
||||
str w6, [x0, #CNT]
|
||||
str x7, [x0, #DIF]
|
||||
mov w0, w15
|
||||
ret
|
||||
endfunc
|
||||
|
||||
function msac_decode_bool_neon, export=1
|
||||
ldp w5, w6, [x0, #RNG] // + CNT
|
||||
sub sp, sp, #48
|
||||
ldr x7, [x0, #DIF]
|
||||
lsr w4, w5, #8 // r >> 8
|
||||
bic w1, w1, #0x3f // f &= ~63
|
||||
@@ -423,13 +523,21 @@ function msac_decode_bool_neon, export=1
|
||||
|
||||
clz w5, w4 // clz(rng)
|
||||
eor w5, w5, #16 // d = clz(rng) ^ 16
|
||||
b L(renorm2)
|
||||
lsl w4, w4, w5 // rng << d
|
||||
subs w6, w6, w5 // cnt -= d
|
||||
lsl x7, x7, x5 // (dif - (v << 48)) << d
|
||||
str w4, [x0, #RNG]
|
||||
b.lo L(refill)
|
||||
|
||||
str w6, [x0, #CNT]
|
||||
str x7, [x0, #DIF]
|
||||
mov w0, w15
|
||||
ret
|
||||
endfunc
|
||||
|
||||
function msac_decode_bool_adapt_neon, export=1
|
||||
ldr w9, [x1] // cdf[0-1]
|
||||
ldp w5, w6, [x0, #RNG] // + CNT
|
||||
sub sp, sp, #48
|
||||
ldr x7, [x0, #DIF]
|
||||
lsr w4, w5, #8 // r >> 8
|
||||
and w2, w9, #0xffc0 // f &= ~63
|
||||
@@ -447,7 +555,7 @@ function msac_decode_bool_adapt_neon, export=1
|
||||
clz w5, w4 // clz(rng)
|
||||
eor w5, w5, #16 // d = clz(rng) ^ 16
|
||||
|
||||
cbz w10, L(renorm2)
|
||||
cbz w10, 1f
|
||||
|
||||
lsr w2, w9, #16 // count = cdf[1]
|
||||
and w9, w9, #0xffff // cdf[0]
|
||||
@@ -465,5 +573,15 @@ function msac_decode_bool_adapt_neon, export=1
|
||||
strh w9, [x1]
|
||||
strh w10, [x1, #2]
|
||||
|
||||
b L(renorm2)
|
||||
1:
|
||||
lsl w4, w4, w5 // rng << d
|
||||
subs w6, w6, w5 // cnt -= d
|
||||
lsl x7, x7, x5 // (dif - (v << 48)) << d
|
||||
str w4, [x0, #RNG]
|
||||
b.lo L(refill)
|
||||
|
||||
str w6, [x0, #CNT]
|
||||
str x7, [x0, #DIF]
|
||||
mov w0, w15
|
||||
ret
|
||||
endfunc
|
||||
|
||||
+1
-1
@@ -1162,7 +1162,7 @@ static int decode_b(Dav1dTaskContext *const t,
|
||||
ts->cdf.m.use_filter_intra[bs]);
|
||||
if (is_filter) {
|
||||
b->y_mode = FILTER_PRED;
|
||||
b->y_angle = dav1d_msac_decode_symbol_adapt4(&ts->msac,
|
||||
b->y_angle = dav1d_msac_decode_symbol_adapt8(&ts->msac,
|
||||
ts->cdf.m.filter_intra, 4);
|
||||
}
|
||||
if (DEBUG_BLOCK_INFO)
|
||||
|
||||
+1
-1
@@ -68,7 +68,7 @@ unsigned dav1d_msac_decode_bool_c(MsacContext *s, unsigned f);
|
||||
unsigned dav1d_msac_decode_hi_tok_c(MsacContext *s, uint16_t *cdf);
|
||||
int dav1d_msac_decode_subexp(MsacContext *s, int ref, int n, unsigned k);
|
||||
|
||||
/* Supported n_symbols ranges: adapt4: 1-4, adapt8: 1-7, adapt16: 3-15 */
|
||||
/* Supported n_symbols ranges: adapt4: 1-3, adapt8: 1-7, adapt16: 3-15 */
|
||||
#ifndef dav1d_msac_decode_symbol_adapt4
|
||||
#define dav1d_msac_decode_symbol_adapt4 dav1d_msac_decode_symbol_adapt_c
|
||||
#endif
|
||||
|
||||
+2
-2
@@ -369,7 +369,7 @@ static int decode_coefs(Dav1dTaskContext *const t,
|
||||
const enum IntraPredMode y_mode_nofilt = b->y_mode == FILTER_PRED ?
|
||||
dav1d_filter_mode_to_y_mode[b->y_angle] : b->y_mode;
|
||||
if (f->frame_hdr->reduced_txtp_set || t_dim->min == TX_16X16) {
|
||||
idx = dav1d_msac_decode_symbol_adapt4(&ts->msac,
|
||||
idx = dav1d_msac_decode_symbol_adapt8(&ts->msac,
|
||||
ts->cdf.m.txtp_intra2[t_dim->min][y_mode_nofilt], 4);
|
||||
*txtp = dav1d_tx_types_per_set[idx + 0];
|
||||
} else {
|
||||
@@ -412,7 +412,7 @@ static int decode_coefs(Dav1dTaskContext *const t,
|
||||
eob_bin = dav1d_msac_decode_symbol_adapt##ns(&ts->msac, eob_bin_cdf, 4 + sz); \
|
||||
break; \
|
||||
}
|
||||
case_sz(0, 16, 4, [is_1d]);
|
||||
case_sz(0, 16, 8, [is_1d]);
|
||||
case_sz(1, 32, 8, [is_1d]);
|
||||
case_sz(2, 64, 8, [is_1d]);
|
||||
case_sz(3, 128, 8, [is_1d]);
|
||||
|
||||
@@ -144,7 +144,7 @@ static void check_decode_symbol(MsacDSPContext *const c, uint8_t *const buf) {
|
||||
MsacContext s_c, s_a;
|
||||
|
||||
declare_func(unsigned, MsacContext *s, uint16_t *cdf, size_t n_symbols);
|
||||
CHECK_SYMBOL_ADAPT( 4, 1, 4);
|
||||
CHECK_SYMBOL_ADAPT( 4, 1, 3);
|
||||
CHECK_SYMBOL_ADAPT( 8, 1, 7);
|
||||
CHECK_SYMBOL_ADAPT(16, 3, 15);
|
||||
report("decode_symbol");
|
||||
|
||||
Reference in New Issue
Block a user