std/vp8: implement IDCT/IWHT

This commit, part of a series, is inspired by pull request PR #168 by
Damian Kaczmarek (with AI assistance), with some modifications.

Updates PR #168
This commit is contained in:
Nigel Tao committed 2026-06-17 16:11:36 +10:00
1 parent bb3c85e2ef
commit 68d13b8740
9 files changed
+1249 -15

No files matched your search

+27 -2
View File
@@ -17,10 +17,14 @@ pri func decoder.decode_coefficients!(
var bc : base.u32
var by : base.u32
var bx : base.u32
var b : base.u32[..= 23]
var plane : base.u32[..= 3]
var lnz : base.u32[..= 1]
var tnz : base.u32[..= 1]
var bit : base.u32[..= 1]
this.mb_dc_nz = 0
this.mb_ac_nz = 0
this.initialize_mb_coeffs!()
// Handle "intra16x16 prediction".
@@ -42,6 +46,8 @@ pri func decoder.decode_coefficients!(
this.mb_states_top[args.mbx] |= lnz << 23
this.mb_states_left |= lnz << 23
this.inverse_wht!()
plane = 0 // Plane 0: Y beginning at coefficient 1.
}
@@ -59,18 +65,28 @@ pri func decoder.decode_coefficients!(
tnz = 1 & (this.mb_states_top[args.mbx] >> (24 + bx))
this.mb_states_top[args.mbx] &= 0xFFFF_FFFF ^ ((1 as base.u32) << (24 + bx))
b = (4 * by) + bx
lnz = this.decode_block_coefficients!(
workbuf: args.workbuf,
mby: args.mby,
seg: args.seg,
b: (4 * by) + bx,
b: b,
plane: plane,
context: lnz + tnz)
bit = 0
if this.mb_coeffs[b][0] <> 0 {
bit = 1
}
this.mb_dc_nz |= bit << b
this.mb_ac_nz |= lnz << b
this.mb_states_top[args.mbx] |= lnz << (24 + bx)
bx += 1
}
this.mb_states_left |= lnz << (24 + by)
by += 1
}
@@ -93,18 +109,27 @@ pri func decoder.decode_coefficients!(
tnz = 1 & (this.mb_states_top[args.mbx] >> (28 + bc + bx))
this.mb_states_top[args.mbx] &= 0xFFFF_FFFF ^ ((1 as base.u32) << (28 + bc + bx))
b = (2 * bc) + (2 * by) + bx + 16
lnz = this.decode_block_coefficients!(
workbuf: args.workbuf,
mby: args.mby,
seg: args.seg,
b: (2 * bc) + (2 * by) + bx + 16,
b: b,
plane: 2, // Plane 2: U or V.
context: lnz + tnz)
this.mb_states_top[args.mbx] |= lnz << (28 + bc + bx)
bit = 0
if this.mb_coeffs[b][0] <> 0 {
bit = 1
}
this.mb_dc_nz |= bit << b
this.mb_ac_nz |= lnz << b
bx += 1
}
this.mb_states_left |= lnz << (28 + bc + by)
by += 1
}
bc += 2
+486
View File
@@ -0,0 +1,486 @@
// Copyright 2026 The Wuffs Authors.
//
// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
// https://www.apache.org/licenses/LICENSE-2.0> or the MIT license
// <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your
// option. This file may not be copied, modified, or distributed
// except according to those terms.
//
// SPDX-License-Identifier: Apache-2.0 OR MIT
pri func decoder.inverse_dct_dc_only!(
cachex: base.u32[..= 28],
cachey: base.u32[..= 22],
b: base.u32[..= 23]) {
var dc : base.u16
var i : base.u32
dc = this.util.sign_extend_rshift_u16(a: this.mb_coeffs[args.b][0] ~mod+ 4, n: 3)
i = 0
while i < 16 {
this.yuv_cache[args.cachey + (i >> 2)][args.cachex + (i & 3)] = CLAMP[1023 & (dc ~mod+
(this.yuv_cache[args.cachey + (i >> 2)][args.cachex + (i & 3)] as base.u16))]
i += 1
}
}
pri func decoder.inverse_dct_full!(
cachex: base.u32[..= 28],
cachey: base.u32[..= 22],
b: base.u32[..= 23]) {
var a00 : base.u32
var a01 : base.u32
var a02 : base.u32
var a03 : base.u32
var a10 : base.u32
var a11 : base.u32
var a12 : base.u32
var a13 : base.u32
var a20 : base.u32
var a21 : base.u32
var a22 : base.u32
var a23 : base.u32
var a30 : base.u32
var a31 : base.u32
var a32 : base.u32
var a33 : base.u32
var b00 : base.u32
var b01 : base.u32
var b02 : base.u32
var b03 : base.u32
var b10 : base.u32
var b11 : base.u32
var b12 : base.u32
var b13 : base.u32
var b20 : base.u32
var b21 : base.u32
var b22 : base.u32
var b23 : base.u32
var b30 : base.u32
var b31 : base.u32
var b32 : base.u32
var b33 : base.u32
var m00 : base.u32
var m01 : base.u32
var m02 : base.u32
var m03 : base.u32
var m10 : base.u32
var m11 : base.u32
var m12 : base.u32
var m13 : base.u32
var m20 : base.u32
var m21 : base.u32
var m22 : base.u32
var m23 : base.u32
var m30 : base.u32
var m31 : base.u32
var m32 : base.u32
var m33 : base.u32
var dc0 : base.u32
var dc1 : base.u32
var dc2 : base.u32
var dc3 : base.u32
var p00 : base.u32
var p01 : base.u32
var p02 : base.u32
var p03 : base.u32
var p10 : base.u32
var p11 : base.u32
var p12 : base.u32
var p13 : base.u32
var p20 : base.u32
var p21 : base.u32
var p22 : base.u32
var p23 : base.u32
var p30 : base.u32
var p31 : base.u32
var p32 : base.u32
var p33 : base.u32
var q00 : base.u32
var q01 : base.u32
var q02 : base.u32
var q03 : base.u32
var q10 : base.u32
var q11 : base.u32
var q12 : base.u32
var q13 : base.u32
var q20 : base.u32
var q21 : base.u32
var q22 : base.u32
var q23 : base.u32
var q30 : base.u32
var q31 : base.u32
var q32 : base.u32
var q33 : base.u32
// 85627 is approximately 65536 * cos(pi/8) * sqrt(2).
// 35468 is approximately 65536 * sin(pi/8) * sqrt(2).
// --------
a00 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0x0])
a01 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0x4])
a02 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0x8])
a03 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0xC])
b00 = a00 ~mod+ a02
b01 = a00 ~mod- a02
b02 = this.util.sign_extend_rshift_u32(a: a01 ~mod* 35468, n: 16) ~mod-
this.util.sign_extend_rshift_u32(a: a03 ~mod* 85627, n: 16)
b03 = this.util.sign_extend_rshift_u32(a: a01 ~mod* 85627, n: 16) ~mod+
this.util.sign_extend_rshift_u32(a: a03 ~mod* 35468, n: 16)
m00 = b00 ~mod+ b03
m01 = b01 ~mod+ b02
m02 = b01 ~mod- b02
m03 = b00 ~mod- b03
// ----
a10 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0x1])
a11 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0x5])
a12 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0x9])
a13 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0xD])
b10 = a10 ~mod+ a12
b11 = a10 ~mod- a12
b12 = this.util.sign_extend_rshift_u32(a: a11 ~mod* 35468, n: 16) ~mod-
this.util.sign_extend_rshift_u32(a: a13 ~mod* 85627, n: 16)
b13 = this.util.sign_extend_rshift_u32(a: a11 ~mod* 85627, n: 16) ~mod+
this.util.sign_extend_rshift_u32(a: a13 ~mod* 35468, n: 16)
m10 = b10 ~mod+ b13
m11 = b11 ~mod+ b12
m12 = b11 ~mod- b12
m13 = b10 ~mod- b13
// ----
a20 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0x2])
a21 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0x6])
a22 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0xA])
a23 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0xE])
b20 = a20 ~mod+ a22
b21 = a20 ~mod- a22
b22 = this.util.sign_extend_rshift_u32(a: a21 ~mod* 35468, n: 16) ~mod-
this.util.sign_extend_rshift_u32(a: a23 ~mod* 85627, n: 16)
b23 = this.util.sign_extend_rshift_u32(a: a21 ~mod* 85627, n: 16) ~mod+
this.util.sign_extend_rshift_u32(a: a23 ~mod* 35468, n: 16)
m20 = b20 ~mod+ b23
m21 = b21 ~mod+ b22
m22 = b21 ~mod- b22
m23 = b20 ~mod- b23
// ----
a30 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0x3])
a31 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0x7])
a32 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0xB])
a33 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[args.b][0xF])
b30 = a30 ~mod+ a32
b31 = a30 ~mod- a32
b32 = this.util.sign_extend_rshift_u32(a: a31 ~mod* 35468, n: 16) ~mod-
this.util.sign_extend_rshift_u32(a: a33 ~mod* 85627, n: 16)
b33 = this.util.sign_extend_rshift_u32(a: a31 ~mod* 85627, n: 16) ~mod+
this.util.sign_extend_rshift_u32(a: a33 ~mod* 35468, n: 16)
m30 = b30 ~mod+ b33
m31 = b31 ~mod+ b32
m32 = b31 ~mod- b32
m33 = b30 ~mod- b33
// --------
dc0 = m00 ~mod+ 4
p00 = dc0 ~mod+ m20
p01 = dc0 ~mod- m20
p02 = this.util.sign_extend_rshift_u32(a: m10 ~mod* 35468, n: 16) ~mod-
this.util.sign_extend_rshift_u32(a: m30 ~mod* 85627, n: 16)
p03 = this.util.sign_extend_rshift_u32(a: m10 ~mod* 85627, n: 16) ~mod+
this.util.sign_extend_rshift_u32(a: m30 ~mod* 35468, n: 16)
q00 = this.util.sign_extend_rshift_u32(a: p00 ~mod+ p03, n: 3)
q01 = this.util.sign_extend_rshift_u32(a: p01 ~mod+ p02, n: 3)
q02 = this.util.sign_extend_rshift_u32(a: p01 ~mod- p02, n: 3)
q03 = this.util.sign_extend_rshift_u32(a: p00 ~mod- p03, n: 3)
this.yuv_cache[args.cachey + 0][args.cachex + 0] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 0][args.cachex + 0] as base.u32) ~mod+ q00)]
this.yuv_cache[args.cachey + 0][args.cachex + 1] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 0][args.cachex + 1] as base.u32) ~mod+ q01)]
this.yuv_cache[args.cachey + 0][args.cachex + 2] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 0][args.cachex + 2] as base.u32) ~mod+ q02)]
this.yuv_cache[args.cachey + 0][args.cachex + 3] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 0][args.cachex + 3] as base.u32) ~mod+ q03)]
// ----
dc1 = m01 ~mod+ 4
p10 = dc1 ~mod+ m21
p11 = dc1 ~mod- m21
p12 = this.util.sign_extend_rshift_u32(a: m11 ~mod* 35468, n: 16) ~mod-
this.util.sign_extend_rshift_u32(a: m31 ~mod* 85627, n: 16)
p13 = this.util.sign_extend_rshift_u32(a: m11 ~mod* 85627, n: 16) ~mod+
this.util.sign_extend_rshift_u32(a: m31 ~mod* 35468, n: 16)
q10 = this.util.sign_extend_rshift_u32(a: p10 ~mod+ p13, n: 3)
q11 = this.util.sign_extend_rshift_u32(a: p11 ~mod+ p12, n: 3)
q12 = this.util.sign_extend_rshift_u32(a: p11 ~mod- p12, n: 3)
q13 = this.util.sign_extend_rshift_u32(a: p10 ~mod- p13, n: 3)
this.yuv_cache[args.cachey + 1][args.cachex + 0] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 1][args.cachex + 0] as base.u32) ~mod+ q10)]
this.yuv_cache[args.cachey + 1][args.cachex + 1] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 1][args.cachex + 1] as base.u32) ~mod+ q11)]
this.yuv_cache[args.cachey + 1][args.cachex + 2] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 1][args.cachex + 2] as base.u32) ~mod+ q12)]
this.yuv_cache[args.cachey + 1][args.cachex + 3] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 1][args.cachex + 3] as base.u32) ~mod+ q13)]
// ----
dc2 = m02 ~mod+ 4
p20 = dc2 ~mod+ m22
p21 = dc2 ~mod- m22
p22 = this.util.sign_extend_rshift_u32(a: m12 ~mod* 35468, n: 16) ~mod-
this.util.sign_extend_rshift_u32(a: m32 ~mod* 85627, n: 16)
p23 = this.util.sign_extend_rshift_u32(a: m12 ~mod* 85627, n: 16) ~mod+
this.util.sign_extend_rshift_u32(a: m32 ~mod* 35468, n: 16)
q20 = this.util.sign_extend_rshift_u32(a: p20 ~mod+ p23, n: 3)
q21 = this.util.sign_extend_rshift_u32(a: p21 ~mod+ p22, n: 3)
q22 = this.util.sign_extend_rshift_u32(a: p21 ~mod- p22, n: 3)
q23 = this.util.sign_extend_rshift_u32(a: p20 ~mod- p23, n: 3)
this.yuv_cache[args.cachey + 2][args.cachex + 0] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 2][args.cachex + 0] as base.u32) ~mod+ q20)]
this.yuv_cache[args.cachey + 2][args.cachex + 1] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 2][args.cachex + 1] as base.u32) ~mod+ q21)]
this.yuv_cache[args.cachey + 2][args.cachex + 2] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 2][args.cachex + 2] as base.u32) ~mod+ q22)]
this.yuv_cache[args.cachey + 2][args.cachex + 3] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 2][args.cachex + 3] as base.u32) ~mod+ q23)]
// ----
dc3 = m03 ~mod+ 4
p30 = dc3 ~mod+ m23
p31 = dc3 ~mod- m23
p32 = this.util.sign_extend_rshift_u32(a: m13 ~mod* 35468, n: 16) ~mod-
this.util.sign_extend_rshift_u32(a: m33 ~mod* 85627, n: 16)
p33 = this.util.sign_extend_rshift_u32(a: m13 ~mod* 85627, n: 16) ~mod+
this.util.sign_extend_rshift_u32(a: m33 ~mod* 35468, n: 16)
q30 = this.util.sign_extend_rshift_u32(a: p30 ~mod+ p33, n: 3)
q31 = this.util.sign_extend_rshift_u32(a: p31 ~mod+ p32, n: 3)
q32 = this.util.sign_extend_rshift_u32(a: p31 ~mod- p32, n: 3)
q33 = this.util.sign_extend_rshift_u32(a: p30 ~mod- p33, n: 3)
this.yuv_cache[args.cachey + 3][args.cachex + 0] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 3][args.cachex + 0] as base.u32) ~mod+ q30)]
this.yuv_cache[args.cachey + 3][args.cachex + 1] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 3][args.cachex + 1] as base.u32) ~mod+ q31)]
this.yuv_cache[args.cachey + 3][args.cachex + 2] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 3][args.cachex + 2] as base.u32) ~mod+ q32)]
this.yuv_cache[args.cachey + 3][args.cachex + 3] = CLAMP[1023 & ((
this.yuv_cache[args.cachey + 3][args.cachex + 3] as base.u32) ~mod+ q33)]
}
pri func decoder.inverse_wht!() {
var a00 : base.u32
var a01 : base.u32
var a02 : base.u32
var a03 : base.u32
var a10 : base.u32
var a11 : base.u32
var a12 : base.u32
var a13 : base.u32
var a20 : base.u32
var a21 : base.u32
var a22 : base.u32
var a23 : base.u32
var a30 : base.u32
var a31 : base.u32
var a32 : base.u32
var a33 : base.u32
var b00 : base.u32
var b01 : base.u32
var b02 : base.u32
var b03 : base.u32
var b10 : base.u32
var b11 : base.u32
var b12 : base.u32
var b13 : base.u32
var b20 : base.u32
var b21 : base.u32
var b22 : base.u32
var b23 : base.u32
var b30 : base.u32
var b31 : base.u32
var b32 : base.u32
var b33 : base.u32
var m00 : base.u32
var m01 : base.u32
var m02 : base.u32
var m03 : base.u32
var m10 : base.u32
var m11 : base.u32
var m12 : base.u32
var m13 : base.u32
var m20 : base.u32
var m21 : base.u32
var m22 : base.u32
var m23 : base.u32
var m30 : base.u32
var m31 : base.u32
var m32 : base.u32
var m33 : base.u32
var p00 : base.u32
var p01 : base.u32
var p02 : base.u32
var p03 : base.u32
var p10 : base.u32
var p11 : base.u32
var p12 : base.u32
var p13 : base.u32
var p20 : base.u32
var p21 : base.u32
var p22 : base.u32
var p23 : base.u32
var p30 : base.u32
var p31 : base.u32
var p32 : base.u32
var p33 : base.u32
// --------
a00 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0x0])
a01 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0x4])
a02 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0x8])
a03 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0xC])
b00 = a00 ~mod+ a03
b01 = a01 ~mod+ a02
b02 = a01 ~mod- a02
b03 = a00 ~mod- a03
m00 = b00 ~mod+ b01
m01 = b03 ~mod+ b02
m02 = b00 ~mod- b01
m03 = b03 ~mod- b02
// ----
a10 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0x1])
a11 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0x5])
a12 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0x9])
a13 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0xD])
b10 = a10 ~mod+ a13
b11 = a11 ~mod+ a12
b12 = a11 ~mod- a12
b13 = a10 ~mod- a13
m10 = b10 ~mod+ b11
m11 = b13 ~mod+ b12
m12 = b10 ~mod- b11
m13 = b13 ~mod- b12
// ----
a20 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0x2])
a21 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0x6])
a22 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0xA])
a23 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0xE])
b20 = a20 ~mod+ a23
b21 = a21 ~mod+ a22
b22 = a21 ~mod- a22
b23 = a20 ~mod- a23
m20 = b20 ~mod+ b21
m21 = b23 ~mod+ b22
m22 = b20 ~mod- b21
m23 = b23 ~mod- b22
// ----
a30 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0x3])
a31 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0x7])
a32 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0xB])
a33 = this.util.sign_extend_convert_u16_u32(a: this.mb_coeffs[24][0xF])
b30 = a30 ~mod+ a33
b31 = a31 ~mod+ a32
b32 = a31 ~mod- a32
b33 = a30 ~mod- a33
m30 = b30 ~mod+ b31
m31 = b33 ~mod+ b32
m32 = b30 ~mod- b31
m33 = b33 ~mod- b32
// --------
p00 = (m00 ~mod+ m30) ~mod+ 3
p01 = (m10 ~mod+ m20)
p02 = (m10 ~mod- m20)
p03 = (m00 ~mod- m30) ~mod+ 3
this.mb_coeffs[0x0][0] = (this.util.sign_extend_rshift_u32(a: p00 ~mod+ p01, n: 3) & 0xFFFF) as base.u16
this.mb_coeffs[0x1][0] = (this.util.sign_extend_rshift_u32(a: p03 ~mod+ p02, n: 3) & 0xFFFF) as base.u16
this.mb_coeffs[0x2][0] = (this.util.sign_extend_rshift_u32(a: p00 ~mod- p01, n: 3) & 0xFFFF) as base.u16
this.mb_coeffs[0x3][0] = (this.util.sign_extend_rshift_u32(a: p03 ~mod- p02, n: 3) & 0xFFFF) as base.u16
// ----
p10 = (m01 ~mod+ m31) ~mod+ 3
p11 = (m11 ~mod+ m21)
p12 = (m11 ~mod- m21)
p13 = (m01 ~mod- m31) ~mod+ 3
this.mb_coeffs[0x4][0] = (this.util.sign_extend_rshift_u32(a: p10 ~mod+ p11, n: 3) & 0xFFFF) as base.u16
this.mb_coeffs[0x5][0] = (this.util.sign_extend_rshift_u32(a: p13 ~mod+ p12, n: 3) & 0xFFFF) as base.u16
this.mb_coeffs[0x6][0] = (this.util.sign_extend_rshift_u32(a: p10 ~mod- p11, n: 3) & 0xFFFF) as base.u16
this.mb_coeffs[0x7][0] = (this.util.sign_extend_rshift_u32(a: p13 ~mod- p12, n: 3) & 0xFFFF) as base.u16
// ----
p20 = (m02 ~mod+ m32) ~mod+ 3
p21 = (m12 ~mod+ m22)
p22 = (m12 ~mod- m22)
p23 = (m02 ~mod- m32) ~mod+ 3
this.mb_coeffs[0x8][0] = (this.util.sign_extend_rshift_u32(a: p20 ~mod+ p21, n: 3) & 0xFFFF) as base.u16
this.mb_coeffs[0x9][0] = (this.util.sign_extend_rshift_u32(a: p23 ~mod+ p22, n: 3) & 0xFFFF) as base.u16
this.mb_coeffs[0xA][0] = (this.util.sign_extend_rshift_u32(a: p20 ~mod- p21, n: 3) & 0xFFFF) as base.u16
this.mb_coeffs[0xB][0] = (this.util.sign_extend_rshift_u32(a: p23 ~mod- p22, n: 3) & 0xFFFF) as base.u16
// ----
p30 = (m03 ~mod+ m33) ~mod+ 3
p31 = (m13 ~mod+ m23)
p32 = (m13 ~mod- m23)
p33 = (m03 ~mod- m33) ~mod+ 3
this.mb_coeffs[0xC][0] = (this.util.sign_extend_rshift_u32(a: p30 ~mod+ p31, n: 3) & 0xFFFF) as base.u16
this.mb_coeffs[0xD][0] = (this.util.sign_extend_rshift_u32(a: p33 ~mod+ p32, n: 3) & 0xFFFF) as base.u16
this.mb_coeffs[0xE][0] = (this.util.sign_extend_rshift_u32(a: p30 ~mod- p31, n: 3) & 0xFFFF) as base.u16
this.mb_coeffs[0xF][0] = (this.util.sign_extend_rshift_u32(a: p33 ~mod- p32, n: 3) & 0xFFFF) as base.u16
}
+47
View File
@@ -78,6 +78,8 @@ pri func decoder.decode_one_macroblock!(workbuf: slice base.u8, mbx: base.u32[..
// Clear the HNZC bits.
this.mb_states_left &= 0x0000_FFFF
this.mb_states_top[args.mbx] &= 0x0000_FFFF
this.mb_dc_nz = 0
this.mb_ac_nz = 0
} else {
this.decode_coefficients!(
@@ -88,6 +90,9 @@ pri func decoder.decode_one_macroblock!(workbuf: slice base.u8, mbx: base.u32[..
this.reconstruct!(
workbuf: args.workbuf, mbx: args.mbx, mby: args.mby,
luma_mode: luma_mode, chroma_mode: chroma_mode)
this.copy_from_yuv_cache!(
workbuf: args.workbuf, mbx: args.mbx, mby: args.mby)
}
// RFC 6386 Section 11.2. Luma Modes.
@@ -257,3 +262,45 @@ pri func decoder.decode_one_subblock_mode!(
return 9 // 9=B_HU_PRED.
}
pri func decoder.copy_from_yuv_cache!(
workbuf: slice base.u8,
mbx: base.u32[..= 0x3FF],
mby: base.u32[..= 0x3FF]) {
var w : base.u64
var i : base.u32
var dst : slice base.u8
w = ((((args.mby * this.workbuf_yuv_y_stride) + args.mbx) * 16) as base.u64) + 0
i = 0
while i < 16 {
if w < args.workbuf.length() {
dst = args.workbuf[w ..]
dst.copy_from_slice!(s: this.yuv_cache[i + 0x01][0x08 .. 0x18])
}
w ~mod+= this.workbuf_yuv_y_stride as base.u64
i += 1
}
w = ((((args.mby * this.workbuf_yuv_uv_stride) + args.mbx) * 8) as base.u64) + this.workbuf_yuv_y_end
i = 0
while i < 8 {
if w < args.workbuf.length() {
dst = args.workbuf[w ..]
dst.copy_from_slice!(s: this.yuv_cache[i + 0x12][0x08 .. 0x10])
}
w ~mod+= this.workbuf_yuv_uv_stride as base.u64
i += 1
}
w = ((((args.mby * this.workbuf_yuv_uv_stride) + args.mbx) * 8) as base.u64) + this.workbuf_yuv_u_end
i = 0
while i < 8 {
if w < args.workbuf.length() {
dst = args.workbuf[w ..]
dst.copy_from_slice!(s: this.yuv_cache[i + 0x12][0x18 .. 0x20])
}
w ~mod+= this.workbuf_yuv_uv_stride as base.u64
i += 1
}
}
+59 -4
View File
@@ -16,26 +16,81 @@ pri func decoder.reconstruct!(
chroma_mode: base.u32[..= 3]) {
var mode : base.u32[..= 12]
var b : base.u32
var mask : base.u32
this.prepare_yuv_cache!(workbuf: args.workbuf, mbx: args.mbx, mby: args.mby)
if args.luma_mode < 4 {
mode = this.substitute_dc_top_left(mode: args.luma_mode, mbx: args.mbx, mby: args.mby)
this.predict_y16!(mode: mode)
}
} else {
b = 0
while b < 16 {
b = 0
while b < 16 {
if args.luma_mode >= 4 {
this.predict_y4!(b: b)
b += 1
}
mask = (1 as base.u32) << (b + 0x00)
if (this.mb_ac_nz & mask) <> 0 {
this.inverse_dct_full!(
cachex: ((b & 0x3) * 4) + 0x08,
cachey: ((b & 0xC) * 1) + 0x01,
b: (b + 0x00))
} else if (this.mb_dc_nz & mask) <> 0 {
this.inverse_dct_dc_only!(
cachex: ((b & 0x3) * 4) + 0x08,
cachey: ((b & 0xC) * 1) + 0x01,
b: (b + 0x00))
}
b += 1
}
mode = this.substitute_dc_top_left(mode: args.chroma_mode, mbx: args.mbx, mby: args.mby)
this.predict_uv8!(b: 0, mode: mode)
if (this.mb_ac_nz & 0x0F_0000) <> 0 {
b = 0
while b < 4 {
this.inverse_dct_full!(
cachex: ((b & 0x1) * 4) + 0x08,
cachey: ((b & 0x2) * 2) + 0x12,
b: (b + 0x10))
b += 1
}
} else if (this.mb_dc_nz & 0x0F_0000) <> 0 {
b = 0
while b < 4 {
this.inverse_dct_dc_only!(
cachex: ((b & 0x1) * 4) + 0x08,
cachey: ((b & 0x2) * 2) + 0x12,
b: (b + 0x10))
b += 1
}
}
this.predict_uv8!(b: 1, mode: mode)
if (this.mb_ac_nz & 0xF0_0000) <> 0 {
b = 0
while b < 4 {
this.inverse_dct_full!(
cachex: ((b & 0x1) * 4) + 0x18,
cachey: ((b & 0x2) * 2) + 0x12,
b: (b + 0x14))
b += 1
}
} else if (this.mb_dc_nz & 0xF0_0000) <> 0 {
b = 0
while b < 4 {
this.inverse_dct_dc_only!(
cachex: ((b & 0x1) * 4) + 0x18,
cachey: ((b & 0x2) * 2) + 0x12,
b: (b + 0x14))
b += 1
}
}
}
pri func decoder.substitute_dc_top_left(
+11
View File
@@ -100,6 +100,17 @@ pub struct decoder? implements base.image_decoder(
mb_subblock_modes : array[16] base.u8[..= 9],
// Two u32 values whose bits are packed like this:
//
// - 24 ..= 31 are unused.
// - 20 ..= 23 are chroma (V) HZNC bits.
// - 16 ..= 19 are chroma (U) HZNC bits.
// - 0 ..= 15 are luma (Y) HNZC bits.
//
// HNZC stands for has-non-zero-coefficients.
mb_dc_nz : base.u32,
mb_ac_nz : base.u32,
yuv_cache : array[26] array[32] base.u8,
dst_x : base.u32,