kbz_8/VulkanDriver
1
1
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

improving blitter
Build / build (push) Successful in 54s
Details
Test / build_and_test (push) Successful in 41m32s
Details

Browse Source
This commit is contained in:
kbz_8
2026-04-22 03:13:15 +02:00
parent aaeca6f854
commit 9b7728503b
3 changed files with 270 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/soft/device/blitter.zig
+95 -35
View File
@@ -17,6 +17,7 @@ const State = struct {
allow_srgb_conversion: bool,
clamp_to_edge: bool,
dst_samples: usize,
filter_3D: bool,
};
const BlitData = struct {
@@ -184,7 +185,7 @@ fn sample(src: []const u8, pos: zm.F32x4, dim: zm.F32x4, slice_bytes: usize, pit
var color: zm.F32x4 = .{ 0.0, 0.0, 0.0, 1.0 };
const src_texel_size = base.format.texelSize(state.src_format);
if (state.filter != .linear or base.format.isUint(state.src_format)) {
if (state.filter == .nearest or base.format.isUint(state.src_format)) {
var x: usize = @intFromFloat(pos[0]);
var y: usize = @intFromFloat(pos[1]);
var z: usize = @intFromFloat(pos[2]);
@@ -218,24 +219,56 @@ fn sample(src: []const u8, pos: zm.F32x4, dim: zm.F32x4, slice_bytes: usize, pit
const iz0: usize = @intCast(@max(@as(i32, @intFromFloat(fz0)), 0));
const ix1 = if (ix0 + 1 >= @as(usize, @intFromFloat(dim[0]))) ix0 else ix0 + 1;
const iy1 = if (iy0 + 1 >= @as(usize, @intFromFloat(dim[0]))) iy0 else iy0 + 1;
const iy1 = if (iy0 + 1 >= @as(usize, @intFromFloat(dim[1]))) iy0 else iy0 + 1;
const sample_0_0 = src[computeOffset3D(ix0, iy0, iz0, slice_bytes, pitch_bytes, src_texel_size)..];
const sample_0_1 = src[computeOffset3D(ix1, iy0, iz0, slice_bytes, pitch_bytes, src_texel_size)..];
const sample_1_0 = src[computeOffset3D(ix0, iy1, iz0, slice_bytes, pitch_bytes, src_texel_size)..];
const sample_1_1 = src[computeOffset3D(ix1, iy1, iz0, slice_bytes, pitch_bytes, src_texel_size)..];
if (state.filter_3D) {
const iz1 = if (iz0 + 1 >= @as(usize, @intFromFloat(dim[2]))) iz0 else iz0 + 1;
const pixel_0_0 = readFloat4(sample_0_0, state);
const pixel_0_1 = readFloat4(sample_0_1, state);
const pixel_1_0 = readFloat4(sample_1_0, state);
const pixel_1_1 = readFloat4(sample_1_1, state);
const sample_0_0_0 = src[computeOffset3D(ix0, iy0, iz0, slice_bytes, pitch_bytes, src_texel_size)..];
const sample_0_1_0 = src[computeOffset3D(ix1, iy0, iz0, slice_bytes, pitch_bytes, src_texel_size)..];
const sample_1_0_0 = src[computeOffset3D(ix0, iy1, iz0, slice_bytes, pitch_bytes, src_texel_size)..];
const sample_1_1_0 = src[computeOffset3D(ix1, iy1, iz0, slice_bytes, pitch_bytes, src_texel_size)..];
const sample_0_0_1 = src[computeOffset3D(ix0, iy0, iz1, slice_bytes, pitch_bytes, src_texel_size)..];
const sample_0_1_1 = src[computeOffset3D(ix1, iy0, iz1, slice_bytes, pitch_bytes, src_texel_size)..];
const sample_1_0_1 = src[computeOffset3D(ix0, iy1, iz1, slice_bytes, pitch_bytes, src_texel_size)..];
const sample_1_1_1 = src[computeOffset3D(ix1, iy1, iz1, slice_bytes, pitch_bytes, src_texel_size)..];
const fx = zm.f32x4s(fx0 - @as(f32, @floatFromInt(ix0)));
const fy = zm.f32x4s(fy0 - @as(f32, @floatFromInt(iy0)));
const ix = zm.f32x4s(1.0) - fx;
const iy = zm.f32x4s(1.0) - fy;
const pixel_0_0_0 = readFloat4(sample_0_0_0, state);
const pixel_0_1_0 = readFloat4(sample_0_1_0, state);
const pixel_1_0_0 = readFloat4(sample_1_0_0, state);
const pixel_1_1_0 = readFloat4(sample_1_1_0, state);
const pixel_0_0_1 = readFloat4(sample_0_0_1, state);
const pixel_0_1_1 = readFloat4(sample_0_1_1, state);
const pixel_1_0_1 = readFloat4(sample_1_0_1, state);
const pixel_1_1_1 = readFloat4(sample_1_1_1, state);
color = (pixel_0_0 * ix + pixel_0_1 * fx) * iy + (pixel_1_0 * ix + pixel_1_1 * fx) * fy;
const fx = zm.f32x4s(fx0 - @as(f32, @floatFromInt(ix0)));
const fy = zm.f32x4s(fy0 - @as(f32, @floatFromInt(iy0)));
const fz = zm.f32x4s(fz0 - @as(f32, @floatFromInt(iz0)));
const ix = zm.f32x4s(1.0) - fx;
const iy = zm.f32x4s(1.0) - fy;
const iz = zm.f32x4s(1.0) - fz;
color = ((pixel_0_0_0 * ix + pixel_0_1_0 * fx) * iy + (pixel_1_0_0 * ix + pixel_1_1_0 * fx) * fy) * iz +
((pixel_0_0_1 * ix + pixel_0_1_1 * fx) * iy + (pixel_1_0_1 * ix + pixel_1_1_1 * fx) * fy) * fz;
} else {
const sample_0_0 = src[computeOffset3D(ix0, iy0, iz0, slice_bytes, pitch_bytes, src_texel_size)..];
const sample_0_1 = src[computeOffset3D(ix1, iy0, iz0, slice_bytes, pitch_bytes, src_texel_size)..];
const sample_1_0 = src[computeOffset3D(ix0, iy1, iz0, slice_bytes, pitch_bytes, src_texel_size)..];
const sample_1_1 = src[computeOffset3D(ix1, iy1, iz0, slice_bytes, pitch_bytes, src_texel_size)..];
const pixel_0_0 = readFloat4(sample_0_0, state);
const pixel_0_1 = readFloat4(sample_0_1, state);
const pixel_1_0 = readFloat4(sample_1_0, state);
const pixel_1_1 = readFloat4(sample_1_1, state);
const fx = zm.f32x4s(fx0 - @as(f32, @floatFromInt(ix0)));
const fy = zm.f32x4s(fy0 - @as(f32, @floatFromInt(iy0)));
const ix = zm.f32x4s(1.0) - fx;
const iy = zm.f32x4s(1.0) - fy;
color = (pixel_0_0 * ix + pixel_0_1 * fx) * iy + (pixel_1_0 * ix + pixel_1_1 * fx) * fy;
}
}
return applyScaleAndClamp(color, state);
@@ -310,8 +343,13 @@ pub fn blitRegion(src: *const SoftImage, dst: *SoftImage, region: vk.ImageBlit,
.dst_format = dst_format,
.filter = filter,
.allow_srgb_conversion = allow_srgb_conversion,
.clamp_to_edge = false,
.clamp_to_edge = src_offset_0.x < 0 or
src_offset_0.y < 0 or
@as(u32, @intCast(src_offset_1.x)) > src_extent.width or
@as(u32, @intCast(src_offset_1.y)) > src_extent.height or
(filter != .nearest and ((x0 < 0.5) or (y0 < 0.5))),
.dst_samples = dst.interface.samples.toInt(),
.filter_3D = (src_offset_1.z - src_offset_0.z) != (dst_offset_1.z - dst_offset_0.z),
};
while (dst_subresource.array_layer <= last_layer) : ({
@@ -353,11 +391,6 @@ fn blit(state: State, data: BlitData) void {
const is_dst_int = base.format.isUint(state.dst_format) or base.format.isSint(state.dst_format);
const are_both_int = is_src_int and is_dst_int;
if (are_both_int) {
base.unsupported("Blit of only integer type images are not supported yet", .{});
return;
}
for (@intCast(data.dst_offset_0.z)..@intCast(data.dst_offset_1.z)) |k| {
const z = data.pos[2] + @as(f32, @floatFromInt(k)) * data.depth_ratio;
var dst_slice = data.dst_map[(k * data.dst_slice_pitch_bytes)..];
@@ -371,16 +404,27 @@ fn blit(state: State, data: BlitData) void {
var dst_pixel = dst_line[(i * base.format.texelSize(state.dst_format))..];
if (are_both_int) {
// TODO
var ix: usize = @intFromFloat(x);
var iy: usize = @intFromFloat(y);
var iz: usize = @intFromFloat(z);
if (state.clamp_to_edge) {
ix = std.math.clamp(ix, 0, @as(usize, @intFromFloat(data.dim[0])) - 1);
iy = std.math.clamp(iy, 0, @as(usize, @intFromFloat(data.dim[1])) - 1);
iz = std.math.clamp(iz, 0, @as(usize, @intFromFloat(data.dim[2])) - 1);
}
const src_map = data.src_map[computeOffset3D(ix, iy, iz, data.src_slice_pitch_bytes, data.src_row_pitch_bytes, base.format.texelSize(state.src_format))..];
const color = readFloat4(src_map, state);
for (0..state.dst_samples) |_| {
writeFloat4(color, dst_pixel, state);
if (dst_pixel.len < data.dst_slice_pitch_bytes)
break;
dst_pixel = dst_pixel[data.dst_slice_pitch_bytes..];
}
} else {
const color = sample(
data.src_map,
.{ x, y, z, 0.0 },
data.dim,
data.src_slice_pitch_bytes,
data.src_row_pitch_bytes,
state,
);
const color = sample(data.src_map, .{ x, y, z, 0.0 }, data.dim, data.src_slice_pitch_bytes, data.src_row_pitch_bytes, state);
for (0..state.dst_samples) |_| {
writeFloat4(color, dst_pixel, state);
if (dst_pixel.len < data.dst_slice_pitch_bytes)
@@ -396,11 +440,20 @@ fn blit(state: State, data: BlitData) void {
fn applyScaleAndClamp(base_color: zm.F32x4, state: State) zm.F32x4 {
var color: zm.F32x4 = base_color;
const unscale = base.format.getScale(state.src_format);
const scale = base.format.getScale(state.dst_format);
if (std.simd.firstTrue(unscale != scale) != null) {
color *= zm.f32x4(scale[0] / unscale[0], scale[1] / unscale[1], scale[2] / unscale[2], scale[3] / unscale[3]);
if (base.format.isFloat(state.src_format) and !base.format.isFloat(state.dst_format)) {
color = @min(color, scale);
color = @max(color, zm.f32x4(
if (base.format.isUnsignedComponent(state.dst_format, 0)) 0.0 else -scale[0],
if (base.format.isUnsignedComponent(state.dst_format, 1)) 0.0 else -scale[1],
if (base.format.isUnsignedComponent(state.dst_format, 2)) 0.0 else -scale[2],
if (base.format.isUnsignedComponent(state.dst_format, 3)) 0.0 else -scale[3],
));
}
if (!base.format.isUnsigned(state.src_format) and base.format.isUnsigned(state.dst_format)) {
color = @max(color, zm.f32x4s(0.0));
}
return color;
@@ -422,7 +475,12 @@ fn readFloat4(map: []const u8, state: State) zm.F32x4 {
c[3] = @as(f32, @floatFromInt(map[3])) / 255.0;
},
else => base.unsupported("Blitter source format {any}", .{state.src_format}),
.r32_uint => {
c[0] = std.mem.bytesToValue(f32, map);
c[3] = @as(f32, @floatFromInt(0xFFFFFFFF));
},
else => base.unsupported("Blitter: read from source format {any}", .{state.src_format}),
}
return c;
@@ -452,6 +510,8 @@ fn writeFloat4(color: zm.F32x4, map: []u8, state: State) void {
map[2] = @intFromFloat(color[2] * 255.0);
map[3] = @intFromFloat(color[3] * 255.0);
},
else => base.unsupported("Blitter dstination format {any}", .{state.src_format}),
.r32_sfloat => std.mem.bytesAsValue(f32, map).* = color[0],
.r32_uint => std.mem.bytesAsValue(u32, map).* = @intFromFloat(color[0]),
else => base.unsupported("Blitter: write to destination format {any}", .{state.dst_format}),
}
}