std/etc2: support 11-bit source formats

This commit is contained in:
Nigel Tao committed 2025-06-10 16:07:43 +10:00
1 parent 5ec14737f3
commit 9833880b78
2 files changed
+522 -26

No files matched your search

+230 -13
View File
@@ -41,7 +41,6 @@
pub status "#bad header"
pub status "#truncated input"
pub status "#unsupported ETC2 file"
pub const DECODER_WORKBUF_LEN_MAX_INCL_WORST_CASE : base.u64 = 0
@@ -93,6 +92,7 @@ pub func decoder.decode_image_config?(dst: nptr base.image_config, src: base.io_
pri func decoder.do_decode_image_config?(dst: nptr base.image_config, src: base.io_reader) {
var c32 : base.u32
var i : base.u32
var rounded_up_width : base.u16
var rounded_up_height : base.u16
@@ -116,13 +116,34 @@ pri func decoder.do_decode_image_config?(dst: nptr base.image_config, src: base.
} else if (c32 == '20\x00\x04'le) or // ETC2 RGBA1.
(c32 == '20\x00\x0B'le) { // ETC2 sRGBA1.
this.pixfmt = base.PIXEL_FORMAT__BGRA_BINARY
} else if (c32 & 0xFFFF) == '20'le {
return "#unsupported ETC2 file"
} else if (c32 == '20\x00\x05'le) { // ETC2 R11U.
this.pixfmt = base.PIXEL_FORMAT__Y_16LE
choose from_colors_to_buffer = [from_colors_to_buffer_r11u]
} else if (c32 == '20\x00\x06'le) { // ETC2 RG11U.
this.pixfmt = base.PIXEL_FORMAT__BGRA_NONPREMUL_4X16LE
choose from_colors_to_buffer = [from_colors_to_buffer_rg11u]
} else if (c32 == '20\x00\x07'le) { // ETC2 R11S.
this.pixfmt = base.PIXEL_FORMAT__Y_16LE
choose from_colors_to_buffer = [from_colors_to_buffer_r11s]
} else if (c32 == '20\x00\x08'le) { // ETC2 RG11S.
this.pixfmt = base.PIXEL_FORMAT__BGRA_NONPREMUL_4X16LE
choose from_colors_to_buffer = [from_colors_to_buffer_rg11s]
} else {
return "#bad header"
}
this.srgb = (c32 >> 24) >= 0x09
if this.pixfmt == base.PIXEL_FORMAT__BGRA_NONPREMUL_4X16LE {
i = 0
while i <= (4096 - 8) {
this.buffer[i + 0] = 0x00
this.buffer[i + 1] = 0x00
this.buffer[i + 6] = 0xFF
this.buffer[i + 7] = 0xFF
i += 8
}
}
rounded_up_width = args.src.read_u16be?()
rounded_up_height = args.src.read_u16be?()
@@ -214,6 +235,7 @@ pub func decoder.decode_frame?(dst: ptr base.pixel_buffer, src: base.io_reader,
pri func decoder.do_decode_frame?(dst: ptr base.pixel_buffer, src: base.io_reader, blend: base.pixel_blend, workbuf: slice base.u8, opts: nptr base.decode_frame_options) {
var remaining : base.u32[..= 0x1000_0000]
var max_nbb : base.u32[..= 64]
var status : base.status
if this.call_sequence == 0x40 {
@@ -239,7 +261,12 @@ pri func decoder.do_decode_frame?(dst: ptr base.pixel_buffer, src: base.io_reade
remaining = ((this.width + 3) / 4) * ((this.height + 3) / 4)
while remaining > 0 {
this.num_buffered_blocks = remaining.min(no_more_than: 64)
max_nbb = 64
if ((this.pixfmt == base.PIXEL_FORMAT__Y_16LE)) or
(this.pixfmt == base.PIXEL_FORMAT__BGRA_NONPREMUL_4X16LE) {
max_nbb = 32
}
this.num_buffered_blocks = remaining.min(no_more_than: max_nbb)
if remaining < this.num_buffered_blocks {
return base."#too much data"
}
@@ -264,7 +291,8 @@ pri func decoder.from_src_to_colors?(src: base.io_reader) {
while bi < this.num_buffered_blocks {
assert bi < 64 via "a < b: a < c; c <= b"(c: this.num_buffered_blocks)
if this.pixfmt == base.PIXEL_FORMAT__BGRA_NONPREMUL {
if (this.pixfmt == base.PIXEL_FORMAT__BGRA_NONPREMUL) or
(this.pixfmt == base.PIXEL_FORMAT__BGRA_NONPREMUL_4X16LE) {
this.alphas[bi] = args.src.read_u64be?()
}
this.colors[bi] = args.src.read_u64be?()
@@ -272,7 +300,9 @@ pri func decoder.from_src_to_colors?(src: base.io_reader) {
}
}
pri func decoder.from_colors_to_buffer!() {
pri func decoder.from_colors_to_buffer!(),
choosy,
{
var bi : base.u32[..= 64]
var color : base.u64
var r0 : base.u32
@@ -781,7 +811,172 @@ pri func decoder.from_alphas_to_buffer!() {
}
}
pri func decoder.from_colors_to_buffer_r11u!() {
this.from_colors_to_buffer_unsigned!(input: 1, dst_bytes_per_pixel: 2, offset_adjustment: 0)
}
pri func decoder.from_colors_to_buffer_r11s!() {
this.from_colors_to_buffer_signed!(input: 1, dst_bytes_per_pixel: 2, offset_adjustment: 0)
}
pri func decoder.from_colors_to_buffer_rg11u!() {
this.from_colors_to_buffer_unsigned!(input: 0, dst_bytes_per_pixel: 8, offset_adjustment: 4)
this.from_colors_to_buffer_unsigned!(input: 1, dst_bytes_per_pixel: 8, offset_adjustment: 2)
}
pri func decoder.from_colors_to_buffer_rg11s!() {
this.from_colors_to_buffer_signed!(input: 0, dst_bytes_per_pixel: 8, offset_adjustment: 4)
this.from_colors_to_buffer_signed!(input: 1, dst_bytes_per_pixel: 8, offset_adjustment: 2)
}
pri func decoder.from_colors_to_buffer_unsigned!(input: base.u32[..= 1], dst_bytes_per_pixel: base.u32[..= 8], offset_adjustment: base.u32[..= 4]) {
var nbb : base.u32[..= 32]
var bi : base.u32[..= 32]
var color : base.u64
var c0 : base.u32
var multiplier : base.u32[..= 120]
var which : base.u32[..= 15]
var offset : base.u32[..= 996]
var y : base.u32
var x : base.u32
var shift : base.u32[..= 45]
var delta : base.u32
var c11 : base.u32
var c16 : base.u32
var o : base.u32[..= 4092]
nbb = this.num_buffered_blocks.min(no_more_than: 32)
while bi < nbb {
assert bi < 32 via "a < b: a < c; c <= b"(c: nbb)
if args.input == 0 {
color = this.alphas[bi]
} else {
color = this.colors[bi]
}
c0 = (((color >> 56) as base.u32) * 8) + 4
multiplier = ((color >> 52) & 0x0F) as base.u32
multiplier *= 8
if multiplier == 0 {
multiplier = 1
}
which = ((color >> 48) & 0x0F) as base.u32
offset = (4 * args.dst_bytes_per_pixel * bi) + args.offset_adjustment
y = 0
while y < 4,
inv bi < 32,
{
x = 0
while x < 4,
inv bi < 32,
inv y < 4,
{
shift = (((x ^ 3) * 4) | (y ^ 3)) * 3
delta = multiplier ~mod* ALPHA_MODIFIERS[which][7 & (color >> shift)]
c11 = c0 ~mod+ delta
if c11 >= 0x8000_0000 {
c11 = 0
assert c11 <= 2047
} else if c11 > 2047 {
c11 = 2047
assert c11 <= 2047
}
c16 = (c11 << 5) | (c11 >> 6)
o = offset + (x * args.dst_bytes_per_pixel) + (y * 1024)
this.buffer[o + 0] = ((c16 >> 0) & 0xFF) as base.u8
this.buffer[o + 1] = ((c16 >> 8) & 0xFF) as base.u8
x += 1
}
y += 1
}
bi += 1
}
}
pri func decoder.from_colors_to_buffer_signed!(input: base.u32[..= 1], dst_bytes_per_pixel: base.u32[..= 8], offset_adjustment: base.u32[..= 4]) {
var nbb : base.u32[..= 32]
var bi : base.u32[..= 32]
var color : base.u64
var c0 : base.u32
var multiplier : base.u32[..= 120]
var which : base.u32[..= 15]
var offset : base.u32[..= 996]
var y : base.u32
var x : base.u32
var shift : base.u32[..= 45]
var delta : base.u32
var c11 : base.u32
var c16 : base.u32
var o : base.u32[..= 4092]
nbb = this.num_buffered_blocks.min(no_more_than: 32)
while bi < nbb {
assert bi < 32 via "a < b: a < c; c <= b"(c: nbb)
if args.input == 0 {
color = this.alphas[bi]
} else {
color = this.colors[bi]
}
c0 = ((color >> 56) as base.u32) * 8
if c0 < (0x80 * 8) {
// No-op.
} else if c0 == (0x80 * 8) {
c0 = 0xFFFF_FC08
} else {
c0 ~mod-= (0x100 * 8)
}
multiplier = ((color >> 52) & 0x0F) as base.u32
multiplier *= 8
if multiplier == 0 {
multiplier = 1
}
which = ((color >> 48) & 0x0F) as base.u32
offset = (4 * args.dst_bytes_per_pixel * bi) + args.offset_adjustment
y = 0
while y < 4,
inv bi < 32,
{
x = 0
while x < 4,
inv bi < 32,
inv y < 4,
{
shift = (((x ^ 3) * 4) | (y ^ 3)) * 3
delta = multiplier ~mod* ALPHA_MODIFIERS[which][7 & (color >> shift)]
c11 = c0 ~mod+ delta
if c11 <= 0x0000_03FF { // 0x0000_03FF is +1023.
// No-op.
} else if c11 < 0x8000_0000 {
c11 = 0x0000_03FF
} else if c11 < 0xFFFF_FC01 { // 0xFFFF_FC01 is -1023.
c11 = 0xFFFF_FC01
}
if c11 < 0x8000_0000 {
c16 = (c11 ~mod<< 5) | (c11 >> 5)
} else {
c11 = 0 ~mod- c11
c16 = (c11 ~mod<< 5) | (c11 >> 5)
c16 = 0 ~mod- c16
}
c16 ^= 0x8000
o = offset + (x * args.dst_bytes_per_pixel) + (y * 1024)
this.buffer[o + 0] = ((c16 >> 0) & 0xFF) as base.u8
this.buffer[o + 1] = ((c16 >> 8) & 0xFF) as base.u8
x += 1
}
y += 1
}
bi += 1
}
}
pri func decoder.from_buffer_to_dst!(dst: ptr base.pixel_buffer) base.status {
var src_bytes_per_pixel : base.u32[..= 8]
var dst_pixfmt : base.pixel_format
var dst_bits_per_pixel : base.u32[..= 256]
var dst_bytes_per_pixel : base.u32[..= 32]
@@ -797,6 +992,17 @@ pri func decoder.from_buffer_to_dst!(dst: ptr base.pixel_buffer) base.status {
var i : base.u64
var num_src_pixels : base.u32
if this.pixfmt == base.PIXEL_FORMAT__Y_16LE {
src_bytes_per_pixel = 2
assert src_bytes_per_pixel > 0
} else if this.pixfmt == base.PIXEL_FORMAT__BGRA_NONPREMUL_4X16LE {
src_bytes_per_pixel = 8
assert src_bytes_per_pixel > 0
} else {
src_bytes_per_pixel = 4
assert src_bytes_per_pixel > 0
}
// TODO: the dst_pixfmt variable shouldn't be necessary. We should be able
// to chain the two calls: "args.dst.pixel_format().bits_per_pixel()".
dst_pixfmt = args.dst.pixel_format()
@@ -808,7 +1014,9 @@ pri func decoder.from_buffer_to_dst!(dst: ptr base.pixel_buffer) base.status {
dst_bytes_per_row = (this.width * dst_bytes_per_pixel) as base.u64
tab = args.dst.plane(p: 0)
while bi < this.num_buffered_blocks {
while bi < this.num_buffered_blocks,
inv src_bytes_per_pixel > 0,
{
assert bi < 64 via "a < b: a < c; c <= b"(c: this.num_buffered_blocks)
if this.width <= this.dst_x {
@@ -824,15 +1032,24 @@ pri func decoder.from_buffer_to_dst!(dst: ptr base.pixel_buffer) base.status {
dy = 0
while dy < 4,
inv src_bytes_per_pixel > 0,
inv bi < 64,
{
si = (1024 * dy) + (16 * bi)
sj = (1024 * dy) + (16 * 64)
if src_bytes_per_pixel == 4 {
si = (1024 * dy) + (16 * bi)
sj = (1024 * dy) + (16 * 64) // 64 is max_nbb.
} else if src_bytes_per_pixel < 4 {
si = (1024 * dy) + (8 * bi)
sj = (1024 * dy) + (8 * 32) // 32 is max_nbb.
} else {
si = (1024 * dy) + (32 * (bi & 31))
sj = (1024 * dy) + (32 * 32) // 32 is max_nbb.
}
if si < sj {
src = this.buffer[si .. sj]
}
if ((4 * rem_x) as base.u64) < src.length() {
src = src[.. (4 * rem_x) as base.u64]
if ((src_bytes_per_pixel * rem_x) as base.u64) < src.length() {
src = src[.. (src_bytes_per_pixel * rem_x) as base.u64]
}
if (this.dst_y ~mod+ dy) >= this.height {
@@ -854,9 +1071,9 @@ pri func decoder.from_buffer_to_dst!(dst: ptr base.pixel_buffer) base.status {
dy += 1
}
num_src_pixels = ((src.length() & 0xFFFF_FFFF) as base.u32) / 4
num_src_pixels = ((src.length() & 0xFFFF_FFFF) as base.u32) / src_bytes_per_pixel
this.dst_x ~mod+= num_src_pixels
bi ~mod+= (num_src_pixels + 3) / 4
bi ~mod+= (num_src_pixels ~mod+ 3) / 4
}
return ok