fix: massive PNG save performance regression (2.6x speedup)

.gitignore

@@ -1,3 +1,4 @@
 .zig-cache/
 node_modules/
 zig-out/
+.benchtree/

bench.sh

@@ -0,0 +1,160 @@
+#!/usr/bin/env bash
+
+# builds the last 8 commits and compares the performance of odiff
+# against the current git HEAD
+
+function print_help() {
+	echo "usage: ./bench.sh [-n <num>] [--unstaged] [--help]"
+	echo ""
+	echo "-c, --clean : cleans the build directory (default: false)"
+	echo "-s, --skip-build : skips building a commit if it already exists (default: true)"
+	echo "-f, --from <commit> : compares the changes from the given commit (default: HEAD)"
+	echo "-n <num> : number of commits to compare (default: 8)"
+	echo "-u, --unstaged : also builds and compares with unstaged changes (default: false)"
+	echo "-r, --runs <num> : number of runs (default: 10)"
+	echo "-h, --help : prints this help message"
+}
+
+# must be in sync with build.zig
+exe_name="odiff"
+function build_project() {
+	local id="$1"
+	echo "Building $id"
+	zig build -Doptimize=ReleaseFast --prefix-exe-dir "bench/$id"
+	if [[ $? -ne 0 ]]; then
+		echo "Failed to build $id"
+		return 1
+	fi
+	echo "Successfully built $id"
+}
+
+function checkout_and_build() {
+	local commit_id="$1"
+	local skip_if_exists="$2"
+
+	echo "Checking out $commit_id"
+	if [[ $commit_id != "unstaged" ]]; then
+		if [[ $skip_if_exists == true ]]; then
+			if [[ -d "zig-out/bench/$commit_id" ]]; then
+				echo "Skipping build of $commit_id"
+				return
+			fi
+		fi
+
+		if [[ ! -d ".benchtree" ]]; then
+			git worktree add ".benchtree" "$commit_id"
+		fi
+		pushd ".benchtree"
+		git checkout "$commit_id"
+		build_project "$commit_id"
+		if [[ $? -ne 0 ]]; then
+			popd
+			return 1
+		fi
+		popd
+		mkdir -p "zig-out/bench/$commit_id"
+		mv ".benchtree/zig-out/bench/$commit_id/$exe_name" "zig-out/bench/$commit_id/$exe_name"
+		echo "Successfully built zig-out/bench/$commit_id/$exe_name"
+	else
+		build_project "$commit_id"
+	fi
+}
+
+function get_commit_ids() {
+	local from="$1"
+	local commit_ids=()
+
+	if $unstaged && [[ -n $(git diff --name-only) ]]; then
+		commit_ids+=("unstaged")
+	fi
+
+	commit_ids+=($(git rev-parse "$from"))
+	if [[ $num_commits -gt 1 ]]; then
+		for i in $(seq 1 $(($num_commits-1))); do
+			commit_ids+=($(git rev-parse "$from"~"$i"))
+		done
+	fi
+	echo "${commit_ids[@]}"
+}
+
+args=$(getopt -o "n:uh" --long "unstaged,help,runs,from" -n "bench.sh" -- "$@")
+
+num_commits=8
+unstaged=false
+clean=false
+skip_if_exists=true
+runs=10
+from="HEAD"
+while true; do
+	case "$1" in
+		-n)
+			num_commits="$2"
+			shift 2
+			;;
+		-c | --clean)
+			clean=true
+			shift
+			;;
+		-s | --skip-build)
+			skip_if_exists=true
+			shift
+			;;
+		-u | --unstaged)
+			unstaged=true
+			shift
+			;;
+		-r | --runs)
+			runs="$2"
+			shift 2
+			;;
+		-f | --from)
+			from="$2"
+			shift 2
+			;;
+		-h | --help)
+			print_help
+			exit 0
+			;;
+		--)
+			shift
+			break
+			;;
+		"")
+			break
+			;;
+		*)
+			print_help
+			exit 1
+			;;
+	esac
+done
+
+if $clean; then
+	echo "Cleaning output directory"
+	rm -rf "zig-out/bench"
+fi
+
+commit_ids=($(get_commit_ids "$from"))
+runnable_commit_ids=()
+echo "Building $num_commits commits"
+for commit_id in "${commit_ids[@]}"; do
+	checkout_and_build "$commit_id" "$skip_if_exists"
+	if [[ $? -eq 0 ]]; then
+		runnable_commit_ids+=("$commit_id")
+	fi
+done
+
+# ensure that hyperfine is installed
+if ! command -v hyperfine &> /dev/null; then
+	echo "hyperfine is not installed."
+	echo "Please install hyperfine using the following command:"
+	echo "cargo install hyperfine"
+	exit 1
+fi
+
+bench_cmd="hyperfine -i -N --warmup 1 --runs $runs "
+options="./images/www.cypress.io.png ./images/www.cypress.io-1.png ./images/www.cypress-diff.png"
+for commit_id in "${runnable_commit_ids[@]}"; do
+	bench_cmd+="\"$PWD/zig-out/bench/$commit_id/$exe_name $options\" "
+done
+eval "$bench_cmd"

src/io/io.zig

@@ -170,13 +170,9 @@ pub fn saveImageWithFormat(img: Image, file_path: []const u8, format: ImageForma
         .truncate = true,
     });
     defer file.close();
-    var buffer: [1024 * 1024]u8 = undefined;
-    var file_writer = file.writer(&buffer);
 
     switch (format) {
-        .png => try png.save(img, &file_writer.interface),
+        .png => try png.save(img, file),
         else => return error.UnsupportedFormat,
     }
-
-    try file_writer.interface.flush();
 }

src/io/png.zig

@@ -38,7 +38,7 @@ pub fn load(allocator: std.mem.Allocator, data: []const u8) !Image {
     };
 }
 
-pub fn save(img: Image, writer: *std.Io.Writer) !void {
+pub fn save(img: Image, file: std.fs.File) !void {
     const ctx = c.spng_ctx_new(c.SPNG_CTX_ENCODER) orelse return error.OutOfMemory;
     defer c.spng_ctx_free(ctx);
 
@@ -53,6 +53,10 @@ pub fn save(img: Image, writer: *std.Io.Writer) !void {
     };
     if (c.spng_set_ihdr(ctx, &ihdr) != 0) return error.InvalidData;
 
+    if (c.spng_set_option(ctx, c.SPNG_FILTER_CHOICE, c.SPNG_DISABLE_FILTERING) != 0) return error.InvalidData;
+
+    var buffer: [4096]u8 = undefined;
+    var file_writer = file.writer(&buffer);
     if (c.spng_set_png_stream(
         ctx,
         struct {
@@ -66,11 +70,11 @@ pub fn save(img: Image, writer: *std.Io.Writer) !void {
                 return 0;
             }
         }.writeFn,
-        @ptrCast(@alignCast(writer)),
+        @ptrCast(@alignCast(&file_writer.interface)),
     ) != 0) return error.InvalidData;
 
-    const pixel_data = img.slice();
-    const res = c.spng_encode_image(ctx, @ptrCast(@alignCast(pixel_data.ptr)), pixel_data.len * @sizeOf(u32), c.SPNG_FMT_PNG, c.SPNG_ENCODE_FINALIZE);
+    const u8_slice: []u8 = @ptrCast(img.slice());
+    const res = c.spng_encode_image(ctx, u8_slice.ptr, u8_slice.len, c.SPNG_FMT_PNG, c.SPNG_ENCODE_FINALIZE);
     if (res != 0) {
         const err_msg = std.mem.span(c.spng_strerror(res));
         std.log.err("writePNG: failed to encode image {s}", .{err_msg});