Add macOS and Windows release scripts

.gitignore

@@ -25,3 +25,6 @@ test/samples/
 # Internal design notes and AI planning artifacts - not for distribution
 funnelcake.md
 docs/superpowers/
+.claude/settings.local.json
+*.profdata
+*.profraw

CHANGELOG.md

@@ -0,0 +1,71 @@
+# Changelog
+
+All notable changes to funnelcake are recorded here. Format follows
+[Keep a Changelog](https://keepachangelog.com/en/1.1.0/); the project does
+not yet follow a tagged release cadence, so changes accumulate under
+**Unreleased** until a tag is cut.
+
+## [Unreleased]
+
+### Added
+- `FUSED_ERR_OUT_OF_MEMORY` (-5) hard-error code returned by
+  `fused_scaler_init` and `fused_hdr_init` when allocation of the internal
+  state struct fails. Previously these paths returned the misleading
+  `FUSED_ERR_NO_STEPS`.
+- `FUSED_LOG_INFO` (2) log level for low-frequency status / diagnostic
+  messages. Routed through the existing `log_warnings` config so callers
+  that install a `FUSED_LOG_CALLBACK` can filter info out by inspecting
+  the `level` argument while still receiving warnings.
+
+### Changed
+- The "tone map LUTs generated" diagnostic in `fused_tonemap_generate_luts`
+  now logs at `FUSED_LOG_INFO` instead of `FUSED_LOG_WARN`. Callers that
+  previously suppressed it via `log_warnings = FUSED_LOG_SUPPRESS` keep
+  the same behavior; callback-based loggers can now keep warnings while
+  dropping info.
+- The "no SIMD support detected" notice from both init functions now
+  routes through `fused_log(&ctx->log_warnings, FUSED_LOG_WARN, …)`
+  instead of writing to `stderr` directly, so users with a configured
+  log target see it where they expect.
+- `fused_scaler_free` and `fused_hdr_free` now also reset
+  `effective_width` and `effective_height` on the context, matching the
+  reset of the other result fields.
+
+### Fixed
+- **HDR init memory leak**: `fused_hdr_init` could leak `state->sdr_temp[i]`
+  buffers if SDR-only steps had been allocated successfully and a later
+  step (the 1:1 tonemap output, or the "no valid steps" check) failed.
+  The error paths called `fused_hdr_free(ctx)` while `ctx->_internal` was
+  still NULL, skipping the cleanup of `state`'s sdr_temp pointers, and
+  then `free(state)` released the struct without freeing those buffers.
+  Init now attaches `state` to `ctx->_internal` immediately after
+  allocation so any subsequent error path goes through `fused_hdr_free`
+  and releases everything.
+- The misaligned-source warning emitted by `fused_scaler_run` and
+  `fused_hdr_run` used a process-wide `static int warned` flag, so the
+  first context to encounter a misaligned source silenced the warning
+  for every other context in the process (and the flag was not
+  thread-safe). Each context now owns its own `src_misaligned_warned`
+  flag inside its internal state.
+- The misaligned-source warning was missing a trailing newline.
+
+### Documentation
+- `docs/API.md`: documented `FUSED_LOG_INFO`, `FUSED_ERR_OUT_OF_MEMORY`,
+  and the relationship between log levels and the routing config.
+
+---
+
+## Conventions
+
+- **Added** — new public API surface (functions, constants, struct fields).
+- **Changed** — non-breaking behavioral changes to existing API.
+- **Deprecated** — APIs scheduled for removal.
+- **Removed** — APIs that have been deleted.
+- **Fixed** — bug fixes that don't change documented behavior.
+- **Security** — vulnerability fixes.
+- **Performance** — measurable speed/memory wins, with a one-line summary
+  of the workload and the delta.
+- **Documentation** — doc-only changes worth noting.
+
+Group breaking changes under their own **Breaking** subsection and call
+out the impact on callers.

INSTALL.md

@@ -125,3 +125,59 @@ Artifacts are written to `dist/` as per-architecture tarballs containing
 release archives are built with `CC=clang LTO=0` so the static library contains
 standard object files suitable for downstream linkers that do not understand
 Clang LTO bitcode.
+
+## macOS Release Artifacts
+
+To build a native macOS release archive:
+
+    ./scripts/build-macos.sh
+
+The script must be run on macOS with Xcode command line tools installed. It
+writes `dist/funnelcake-macos-<arch>.tar.gz` containing `libfunnelcake.a`,
+`include/funnelcake.h`, the README, install notes, and `BUILD_INFO`.
+
+## Windows Release Artifacts
+
+To build Windows release archives:
+
+    ./scripts/build-windows.sh             # bash / MSYS2 / Git Bash
+    scripts\build-windows.ps1              # native PowerShell 5.1+
+
+By default the script builds every Windows target whose toolchain is available.
+MinGW-w64 artifacts contain `libfunnelcake.a`; MSVC artifacts contain
+`funnelcake.lib`. Both package layouts include `include/funnelcake.h`, the
+README, install notes, and `BUILD_INFO`.
+
+For MinGW-w64 only:
+
+    ./scripts/build-windows.sh --mingw
+    scripts\build-windows.ps1 -Mingw
+
+For `x86_64` MinGW, install tools that provide `x86_64-w64-mingw32-gcc` and
+`x86_64-w64-mingw32-ar`. For Windows on ARM64 MinGW, install tools that provide
+`aarch64-w64-mingw32-gcc` and `aarch64-w64-mingw32-ar`.
+
+For MSVC only, run from a Visual Studio developer shell where `cl.exe` and
+`lib.exe` are in `PATH`:
+
+    ./scripts/build-windows.sh --msvc
+    scripts\build-windows.ps1 -Msvc
+
+Both MinGW and MSVC builds use the normal source selection (AVX2 on `x86_64`,
+NEON on `aarch64`/ARM64). The NEON kernels guard on `__aarch64__ || _M_ARM64`,
+so MSVC ARM64 picks up the same SIMD coverage as the MinGW cross-compile.
+
+### Windows ARM64 only
+
+For a Windows-on-ARM64 build without touching the x86_64 paths, use the
+dedicated PowerShell driver:
+
+    scripts\build-windows-arm64.ps1                # MinGW + MSVC, whichever is available
+    scripts\build-windows-arm64.ps1 -Mingw         # cross-compile via aarch64-w64-mingw32-gcc
+    scripts\build-windows-arm64.ps1 -Msvc          # native ARM64 MSVC
+
+The `-Msvc` path requires an "ARM64 Native Tools Command Prompt for VS" or
+an equivalent Developer PowerShell with `VSCMD_ARG_TGT_ARCH=arm64`; the
+script refuses to run if the shell is not configured for ARM64. Both
+artifact layouts mirror `build-windows.ps1`: a per-toolchain `dist/`
+package plus a `.zip.sha256`.

Makefile

@@ -15,6 +15,10 @@ TEST_OPT = -O2
 UNAME_M := $(shell uname -m)
 UNAME_S := $(shell uname -s)
 
+ifeq ($(UNAME_S),Windows_NT)
+  CFLAGS_BASE += -D__USE_MINGW_ANSI_STDIO=1
+endif
+
 # Normalize FreeBSD's "amd64" to "x86_64" so the SIMD-selection blocks
 # below match. FreeBSD/arm64 already reports "aarch64".
 ifeq ($(UNAME_S),FreeBSD)

README.md

@@ -620,7 +620,8 @@ benchmark comparison; without it the library and headers install but
 
 The scalar fallback is correct on all platforms but significantly slower.
 On hardware without AVX2, NEON, or RVV, the library logs a one-time notice
-to stderr at first init.
+through the configured `log_warnings` channel at first init (default:
+stderr).
 
 
 ## HDR10 support

docs/API.md

@@ -221,9 +221,9 @@ struct means write to stderr.
 
 | Field | Type | Description |
 |-------|------|-------------|
-| `target` | `int` | One of the `FUSED_LOG_*` constants. |
+| `target` | `int` | One of the `FUSED_LOG_*` target constants. |
 | `file` | `FILE *` | Used when `target == FUSED_LOG_FILE`. Must be a valid open file. |
-| `callback` | `void (*)(int level, const char *msg, void *ctx)` | Used when `target == FUSED_LOG_CALLBACK`. `level` is `FUSED_LOG_ERROR` or `FUSED_LOG_WARN`. |
+| `callback` | `void (*)(int level, const char *msg, void *ctx)` | Used when `target == FUSED_LOG_CALLBACK`. `level` is one of `FUSED_LOG_ERROR`, `FUSED_LOG_WARN`, `FUSED_LOG_INFO`. |
 | `callback_ctx` | `void *` | Passed through opaquely as the `ctx` argument to `callback`. |
 
 Log target constants:
@@ -236,6 +236,15 @@ Log target constants:
 | `FUSED_LOG_SUPPRESS` | 3 | Discard all messages |
 | `FUSED_LOG_CALLBACK` | 4 | Call `config.callback` |
 
+Log level constants (passed to callbacks; stderr/stdout/file targets emit
+every message regardless of level):
+
+| Constant | Value | Meaning |
+|----------|-------|---------|
+| `FUSED_LOG_ERROR` | 0 | Hard error — init failed, no resources allocated. Routed via `log_errors`. |
+| `FUSED_LOG_WARN`  | 1 | Partial success or fallback — request still produced output. Routed via `log_warnings`. |
+| `FUSED_LOG_INFO`  | 2 | Low-frequency status / diagnostic. Routed via `log_warnings`; filter on `level` in a callback to drop. |
+
 
 ## Scale Step Flags
 
@@ -515,6 +524,7 @@ are valid, and `fused_scaler_run` must not be called.
 | `FUSED_ERR_NO_STEPS` | -2 | No valid step flags remain after filtering (all were rejected or none were set). |
 | `FUSED_ERR_BAD_DIMENSIONS` | -3 | `src_width` or `src_height` is <= 0, or too small for the requested steps. |
 | `FUSED_ERR_BAD_ALIGNMENT` | -4 | `src_y_stride` or `src_uv_stride` is not 32-byte aligned. |
+| `FUSED_ERR_OUT_OF_MEMORY` | -5 | Allocation of internal state failed; output buffers (if any were allocated earlier in init) have already been released. |
 
 
 ## Alignment Requirements
@@ -678,9 +688,12 @@ scaler.log_warnings.callback     = my_log;
 scaler.log_warnings.callback_ctx = my_logger_instance;
 ```
 
-The `level` argument to the callback is `FUSED_LOG_ERROR` (0) or
-`FUSED_LOG_WARN` (1). The `msg` string is a complete formatted message;
-do not call `fused_scaler_*` functions from within the callback.
+The `level` argument to the callback is `FUSED_LOG_ERROR` (0),
+`FUSED_LOG_WARN` (1), or `FUSED_LOG_INFO` (2). Info-level messages
+(e.g. "tone map LUTs generated") share the `log_warnings` config — to
+keep warnings but drop info, install a callback and filter on `level`.
+The `msg` string is a complete formatted message; do not call
+`fused_scaler_*` functions from within the callback.
 
 
 ## HDR10 API Reference

funnelcake.pc

@@ -0,0 +1,11 @@
+prefix=/usr/local
+exec_prefix=${prefix}
+libdir=/usr/local/lib
+includedir=/usr/local/include
+
+Name: funnelcake
+Description: SIMD YUV scaler with HDR/SDR tonemapping
+Version: 0.1.0
+Cflags: -I${includedir}
+Libs: -L${libdir} -lfunnelcake
+Libs.private: -lm

include/funnelcake.h

@@ -140,14 +140,22 @@ extern "C" {
 #define FUSED_ERR_NO_STEPS       (-2)  /* no valid step flags set after filtering   */
 #define FUSED_ERR_BAD_DIMENSIONS (-3)  /* src_width/height <= 0 or too small        */
 #define FUSED_ERR_BAD_ALIGNMENT  (-4)  /* strides not 32-byte aligned               */
+#define FUSED_ERR_OUT_OF_MEMORY  (-5)  /* allocation of internal state failed       */
 
 
 /* --------------------------------------------------------------------------
  * Log levels
+ *
+ * Levels are passed to FUSED_LOG_CALLBACK callbacks (which can filter on
+ * them); stderr/stdout/file targets emit every message regardless of level.
+ * Routing of info-level diagnostics shares the warnings logger config, so
+ * callers that want to drop info but keep warnings should install a callback
+ * and filter by level.
  * -------------------------------------------------------------------------- */
 
 #define FUSED_LOG_ERROR     0
 #define FUSED_LOG_WARN      1
+#define FUSED_LOG_INFO      2   /* low-frequency diagnostic / status messages */
 
 
 /* --------------------------------------------------------------------------

scripts/build-macos.sh

@@ -0,0 +1,89 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+REPO_ROOT="$(cd "${SCRIPT_DIR}/.." && pwd)"
+DIST_DIR="${REPO_ROOT}/dist"
+BUILD_DATE="$(date -u +%Y%m%dT%H%M%SZ)"
+HOST_OS="$(uname -s)"
+HOST_ARCH="$(uname -m)"
+
+usage() {
+  cat <<'EOF'
+Usage: scripts/build-macos.sh
+
+Build a native macOS release archive. This script must run on macOS with
+Xcode command line tools installed.
+EOF
+}
+
+if [ "${1:-}" = "-h" ] || [ "${1:-}" = "--help" ]; then
+  usage
+  exit 0
+fi
+
+if [ "$#" -gt 0 ]; then
+  echo "error: unknown option: $1" >&2
+  usage >&2
+  exit 1
+fi
+
+require_tool() {
+  if ! command -v "$1" >/dev/null 2>&1; then
+    echo "error: required tool not found: $1" >&2
+    exit 1
+  fi
+}
+
+if [ "${HOST_OS}" != "Darwin" ]; then
+  echo "error: macOS artifacts require a Darwin host" >&2
+  exit 1
+fi
+
+require_tool clang
+require_tool make
+require_tool shasum
+require_tool tar
+
+mkdir -p "${DIST_DIR}"
+
+package_dir="funnelcake-macos-${HOST_ARCH}"
+
+echo "==> Building macOS ${HOST_ARCH} artifact"
+(
+  cd "${REPO_ROOT}"
+  make clean
+  make lib CC=clang LTO=0 UNAME_S=Darwin UNAME_M="${HOST_ARCH}"
+)
+
+rm -rf "${DIST_DIR}/${package_dir}"
+mkdir -p "${DIST_DIR}/${package_dir}/include"
+cp "${REPO_ROOT}/libfunnelcake.a" "${DIST_DIR}/${package_dir}/"
+cp "${REPO_ROOT}/include/funnelcake.h" "${DIST_DIR}/${package_dir}/include/"
+cp "${REPO_ROOT}/README.md" "${REPO_ROOT}/INSTALL.md" "${DIST_DIR}/${package_dir}/"
+
+{
+  printf '%s\n' "name=funnelcake"
+  printf '%s\n' "target_os=macos"
+  printf '%s\n' "target_arch=${HOST_ARCH}"
+  printf '%s\n' "compiler=clang"
+  printf '%s\n' "lto=0"
+  printf '%s\n' "build_date=${BUILD_DATE}"
+} > "${DIST_DIR}/${package_dir}/BUILD_INFO"
+
+rm -f "${DIST_DIR}/${package_dir}.tar.gz" "${DIST_DIR}/${package_dir}.tar.gz.sha256"
+(
+  cd "${DIST_DIR}"
+  tar -czf "${package_dir}.tar.gz" "${package_dir}"
+  shasum -a 256 "${package_dir}.tar.gz" > "${package_dir}.tar.gz.sha256"
+)
+
+(
+  cd "${REPO_ROOT}"
+  make clean
+)
+
+echo ""
+echo "Artifacts written to ${DIST_DIR}:"
+echo "  ${package_dir}.tar.gz"
+echo "  ${package_dir}.tar.gz.sha256"