std/vp8: simplify swizzle_one_row_of_macroblocks

Prior to 381e186f "std/vp8: swizzle_ycck once, not once per MB row", we
swizzled one row of macroblocks at a time. After 381e186f, we swizzled
the whole image in one pass, but kept the option of, in the future,
amending the swizzle_ycck implementation to allow ((y_min_incl > 0) and
(ycc_upsampling == FANCY_ETC)) at the same time.

However, 381e186f didn't show a performance loss so large that it seems
worth complicating the swizzle_ycck implementation in that way. So we
just drop, in this commit, the option of that amendment.
This commit is contained in:
Nigel Tao committed 2026-07-03 21:58:33 +10:00
1 parent b2a2486096
commit 67591a3570
2 files changed
+33 -99

No files matched your search

+20 -43
View File
@@ -82114,11 +82114,10 @@ wuffs_vp8__decoder__copy_from_yuv_cache(
WUFFS_BASE__GENERATED_C_CODE
static wuffs_base__status
wuffs_vp8__decoder__swizzle_one_row_of_macroblocks(
wuffs_vp8__decoder__swizzle(
wuffs_vp8__decoder* self,
wuffs_base__pixel_buffer* a_dst,
wuffs_base__slice_u8 a_workbuf,
uint32_t a_mby);
wuffs_base__slice_u8 a_workbuf);
WUFFS_BASE__GENERATED_C_CODE
static wuffs_base__empty_struct
@@ -83995,14 +83994,13 @@ wuffs_vp8__decoder__decode_macroblocks(
v_mby += 1u;
}
if (self->private_impl.f_mbh > 0u) {
v_mby = (self->private_impl.f_mbh - 1u);
if (self->private_impl.f_filt_level <= 0u) {
} else if (self->private_impl.f_filt_simple) {
wuffs_vp8__decoder__filter_simple(self, a_workbuf, v_mby);
wuffs_vp8__decoder__filter_simple(self, a_workbuf, (self->private_impl.f_mbh - 1u));
} else {
wuffs_vp8__decoder__filter_normal(self, a_workbuf, v_mby);
wuffs_vp8__decoder__filter_normal(self, a_workbuf, (self->private_impl.f_mbh - 1u));
}
v_status = wuffs_vp8__decoder__swizzle_one_row_of_macroblocks(self, a_dst, a_workbuf, v_mby);
v_status = wuffs_vp8__decoder__swizzle(self, a_dst, a_workbuf);
if ( ! wuffs_base__status__is_ok(&v_status)) {
return wuffs_private_impl__status__ensure_not_a_suspension(v_status);
}
@@ -84263,62 +84261,41 @@ wuffs_vp8__decoder__copy_from_yuv_cache(
return wuffs_base__make_empty_struct();
}
// -------- func vp8.decoder.swizzle_one_row_of_macroblocks
// -------- func vp8.decoder.swizzle
WUFFS_BASE__GENERATED_C_CODE
static wuffs_base__status
wuffs_vp8__decoder__swizzle_one_row_of_macroblocks(
wuffs_vp8__decoder__swizzle(
wuffs_vp8__decoder* self,
wuffs_base__pixel_buffer* a_dst,
wuffs_base__slice_u8 a_workbuf,
uint32_t a_mby) {
uint32_t v_y_min_incl = 0;
uint32_t v_y_max_excl = 0;
uint64_t v_o = 0;
wuffs_base__slice_u8 v_src0 = {0};
wuffs_base__slice_u8 v_src1 = {0};
wuffs_base__slice_u8 v_src2 = {0};
wuffs_base__slice_u8 a_workbuf) {
wuffs_base__status v_status = wuffs_base__make_status(NULL);
if ((self->private_impl.f_workbuf_yuv_y_end > self->private_impl.f_workbuf_yuv_u_end) || (self->private_impl.f_workbuf_yuv_u_end > self->private_impl.f_workbuf_yuv_v_end) || (self->private_impl.f_workbuf_yuv_v_end > ((uint64_t)(a_workbuf.len)))) {
return wuffs_base__make_status(wuffs_base__error__bad_workbuf_length);
}
v_y_min_incl = wuffs_base__u32__sat_sub((a_mby * 16u), 6u);
v_y_min_incl = 0u;
v_y_max_excl = ((a_mby * 16u) + 10u);
if ((a_mby + 1u) >= self->private_impl.f_mbh) {
v_y_max_excl = self->private_impl.f_height;
}
v_o = ((uint64_t)((v_y_min_incl * self->private_impl.f_workbuf_yuv_y_stride)));
if (v_o < self->private_impl.f_workbuf_yuv_y_end) {
v_src0 = wuffs_base__slice_u8__subslice_ij(a_workbuf, v_o, self->private_impl.f_workbuf_yuv_y_end);
}
v_o = (((uint64_t)(((v_y_min_incl / 2u) * self->private_impl.f_workbuf_yuv_uv_stride))) + self->private_impl.f_workbuf_yuv_y_end);
if (v_o < self->private_impl.f_workbuf_yuv_u_end) {
v_src1 = wuffs_base__slice_u8__subslice_ij(a_workbuf, v_o, self->private_impl.f_workbuf_yuv_u_end);
}
v_o = (((uint64_t)(((v_y_min_incl / 2u) * self->private_impl.f_workbuf_yuv_uv_stride))) + self->private_impl.f_workbuf_yuv_u_end);
if (v_o < self->private_impl.f_workbuf_yuv_v_end) {
v_src2 = wuffs_base__slice_u8__subslice_ij(a_workbuf, v_o, self->private_impl.f_workbuf_yuv_v_end);
}
v_status = wuffs_base__pixel_swizzler__swizzle_ycck(&self->private_impl.f_swizzler,
a_dst,
wuffs_base__pixel_buffer__palette(a_dst),
0u,
self->private_impl.f_width,
v_y_min_incl,
v_y_max_excl,
v_src0,
v_src1,
v_src2,
0u,
self->private_impl.f_height,
wuffs_base__slice_u8__subslice_j(a_workbuf, self->private_impl.f_workbuf_yuv_y_end),
wuffs_base__slice_u8__subslice_ij(a_workbuf,
self->private_impl.f_workbuf_yuv_y_end,
self->private_impl.f_workbuf_yuv_u_end),
wuffs_base__slice_u8__subslice_ij(a_workbuf,
self->private_impl.f_workbuf_yuv_u_end,
self->private_impl.f_workbuf_yuv_v_end),
wuffs_base__utility__empty_slice_u8(),
(self->private_impl.f_mbw * 16u),
(self->private_impl.f_mbw * 8u),
(self->private_impl.f_mbw * 8u),
0u,
wuffs_base__u32__sat_sub((self->private_impl.f_mbh * 16u), v_y_min_incl),
wuffs_base__u32__sat_sub((self->private_impl.f_mbh * 8u), (v_y_min_incl / 2u)),
wuffs_base__u32__sat_sub((self->private_impl.f_mbh * 8u), (v_y_min_incl / 2u)),
(self->private_impl.f_mbh * 16u),
(self->private_impl.f_mbh * 8u),
(self->private_impl.f_mbh * 8u),
0u,
(self->private_impl.f_mbw * 16u),
(self->private_impl.f_mbw * 8u),
+13 -56
View File
@@ -47,29 +47,21 @@ pri func decoder.decode_macroblocks!(dst: ptr base.pixel_buffer, workbuf: slice
} else {
this.filter_normal!(workbuf: args.workbuf, mby: mby - 1)
}
// TODO(§): re-enable this when swizzle_ycck can take y_min_incl > 0.
//
// status = this.swizzle_one_row_of_macroblocks!(dst: args.dst, workbuf: args.workbuf, mby: mby - 1)
// if not status.is_ok() {
// return status
// }
}
mby += 1
}
if this.mbh > 0 {
mby = this.mbh - 1
if this.filt_level <= 0 {
// No-op.
} else if this.filt_simple {
this.filter_simple!(workbuf: args.workbuf, mby: mby)
this.filter_simple!(workbuf: args.workbuf, mby: this.mbh - 1)
} else {
this.filter_normal!(workbuf: args.workbuf, mby: mby)
this.filter_normal!(workbuf: args.workbuf, mby: this.mbh - 1)
}
status = this.swizzle_one_row_of_macroblocks!(dst: args.dst, workbuf: args.workbuf, mby: mby)
status = this.swizzle!(dst: args.dst, workbuf: args.workbuf)
if not status.is_ok() {
return status
}
@@ -349,18 +341,9 @@ pri func decoder.copy_from_yuv_cache!(
}
}
pri func decoder.swizzle_one_row_of_macroblocks!(
pri func decoder.swizzle!(
dst: ptr base.pixel_buffer,
workbuf: slice base.u8,
mby: base.u32[..= 0x3FF]) base.status {
var y_min_incl : base.u32[..= 0x3FFF]
var y_max_excl : base.u32[..= 0x3FFF]
var o : base.u64
var src0 : slice base.u8
var src1 : slice base.u8
var src2 : slice base.u8
workbuf: slice base.u8) base.status {
var status : base.status
if (this.workbuf_yuv_y_end > this.workbuf_yuv_u_end) or
@@ -371,49 +354,24 @@ pri func decoder.swizzle_one_row_of_macroblocks!(
assert this.workbuf_yuv_u_end <= args.workbuf.length() via "a <= b: a <= c; c <= b"(c: this.workbuf_yuv_v_end)
assert this.workbuf_yuv_y_end <= args.workbuf.length() via "a <= b: a <= c; c <= b"(c: this.workbuf_yuv_u_end)
// There are 16 rows of pixels per macroblock. However, the bottom 6 rows
// of pixels (the bottom 3 rows of samples, at 4:2:0 chroma subsampling)
// can be modified by the the macroblock below's loop filtering (which
// happens after decoder.swizzle_one_row_of_macroblocks returns), unless
// we're swizzling the bottom-most row of macroblocks.
y_min_incl = (args.mby * 16) ~sat- 6
y_min_incl = 0 // TODO(§): delete this.
y_max_excl = (args.mby * 16) + 10
if (args.mby + 1) >= this.mbh {
y_max_excl = this.height
}
o = (y_min_incl * this.workbuf_yuv_y_stride) as base.u64
if o < this.workbuf_yuv_y_end {
src0 = args.workbuf[o .. this.workbuf_yuv_y_end]
}
o = (((y_min_incl / 2) * this.workbuf_yuv_uv_stride) as base.u64) + this.workbuf_yuv_y_end
if o < this.workbuf_yuv_u_end {
src1 = args.workbuf[o .. this.workbuf_yuv_u_end]
}
o = (((y_min_incl / 2) * this.workbuf_yuv_uv_stride) as base.u64) + this.workbuf_yuv_u_end
if o < this.workbuf_yuv_v_end {
src2 = args.workbuf[o .. this.workbuf_yuv_v_end]
}
status = this.swizzler.swizzle_ycck!(
dst: args.dst,
dst_palette: args.dst.palette(),
x_min_incl: 0,
x_max_excl: this.width,
y_min_incl: y_min_incl,
y_max_excl: y_max_excl,
src0: src0,
src1: src1,
src2: src2,
y_min_incl: 0,
y_max_excl: this.height,
src0: args.workbuf[.. this.workbuf_yuv_y_end],
src1: args.workbuf[this.workbuf_yuv_y_end .. this.workbuf_yuv_u_end],
src2: args.workbuf[this.workbuf_yuv_u_end .. this.workbuf_yuv_v_end],
src3: this.util.empty_slice_u8(),
width0: this.mbw * 16,
width1: this.mbw * 8,
width2: this.mbw * 8,
width3: 0,
height0: (this.mbh * 16) ~sat- y_min_incl,
height1: (this.mbh * 8) ~sat- (y_min_incl / 2),
height2: (this.mbh * 8) ~sat- (y_min_incl / 2),
height0: this.mbh * 16,
height1: this.mbh * 8,
height2: this.mbh * 8,
height3: 0,
stride0: this.mbw * 16,
stride1: this.mbw * 8,
@@ -428,7 +386,6 @@ pri func decoder.swizzle_one_row_of_macroblocks!(
v2: 1,
v3: 0,
ycc_model: base.YCC_MODEL__BT_601_STUDIO_RANGE as base.u8,
// TODO(§): let fancy upsampling work with y_min_incl > 0.
ycc_upsampling: base.YCC_UPSAMPLING__FANCY_LIKE_LIBWEBP as base.u8,
scratch_buffer_2k: this.swizzle_ycck_scratch_buffer_2k[..])
return status