Address review: safer cross-compile detection, optimize NEON byte_merge

Makefile.common:
- Restrict armv7/armhf NEON auto-selection to HAVE_NEON=1 only
- Guard uname -m fallback to native-build platforms (unix/osx/win)
- Add -msse2 flag consistently for all 32-bit x86 selection paths

blitter_simd_neon.c:
- Build neon_byte_merge mask in registers via vcreate_u64 instead of
  stack array + vld1_u8 memory load

Co-Authored-By: Claude Opus 4.6 <[email protected]>

Author: JoeMatt

Makefile.common

@@ -65,49 +65,57 @@ else ifeq ($(BLITTER_SIMD),neon)
 else ifeq ($(BLITTER_SIMD),scalar)
    BLITTER_SIMD_SRC := $(CORE_DIR)/src/blitter_simd_scalar.c
 else
-# ARM targets: prefer NEON when the target configuration says it is available.
+# ARM targets: prefer NEON when guaranteed or explicitly enabled.
 ifeq ($(HAVE_NEON), 1)
    BLITTER_SIMD_SRC := $(CORE_DIR)/src/blitter_simd_neon.c
 endif
 ifneq (,$(filter ios-arm64 tvos-arm64,$(platform)))
    BLITTER_SIMD_SRC := $(CORE_DIR)/src/blitter_simd_neon.c
 endif
-# Only aarch64/arm64 guaranteed to have NEON; plain 'arm' may lack it
+# aarch64/arm64 always have NEON; plain 'arm' may lack it.
 ifneq (,$(filter aarch64 arm64,$(ARCH)))
    BLITTER_SIMD_SRC := $(CORE_DIR)/src/blitter_simd_neon.c
 endif
-ifneq (,$(filter arm64 armv8% armv7% armhf,$(platform)))
+# armv8+ implies NEON; armv7/armhf only use NEON if HAVE_NEON was set above.
+ifneq (,$(filter arm64 armv8%,$(platform)))
    BLITTER_SIMD_SRC := $(CORE_DIR)/src/blitter_simd_neon.c
 endif
 
 # x86/x64 targets: use SSE2.
 ifneq (,$(filter x86_64 x86 i686 i386,$(ARCH)))
    BLITTER_SIMD_SRC := $(CORE_DIR)/src/blitter_simd_sse2.c
-   # 32-bit x86 may need explicit SSE2 enable
-   ifneq (,$(filter i686 i386 x86,$(ARCH)))
-      CFLAGS += -msse2
-   endif
 endif
 ifneq (,$(filter x86_64 x86 i686 i386 win-x64 win32,$(platform)))
    BLITTER_SIMD_SRC := $(CORE_DIR)/src/blitter_simd_sse2.c
 endif
-# MSYS2/MinGW on x86_64
-ifneq (,$(filter MINGW64%,$(MSYSTEM)))
+# MSYS2/MinGW
+ifneq (,$(filter MINGW64% MINGW32%,$(MSYSTEM)))
    BLITTER_SIMD_SRC := $(CORE_DIR)/src/blitter_simd_sse2.c
 endif
+# 32-bit x86 needs explicit -msse2 (x86_64 has it baseline).
+ifeq ($(BLITTER_SIMD_SRC),$(CORE_DIR)/src/blitter_simd_sse2.c)
+ifneq (,$(filter i686 i386 x86 win32,$(ARCH) $(platform)))
+   CFLAGS += -msse2
+endif
+ifneq (,$(filter MINGW32%,$(MSYSTEM)))
+   CFLAGS += -msse2
+endif
+endif
 
-# Native build fallback: auto-detect from host when no SIMD was selected above.
+# Native build fallback: auto-detect from host architecture, but only for
+# native-build platforms (unix/osx/win). Cross-compile targets (vita, ps3,
+# libnx, etc.) set platform explicitly and should not use host detection.
 ifeq ($(BLITTER_SIMD_SRC),)
-ifneq (,$(filter x86_64 x86 i686 i386,$(shell uname -m 2>/dev/null)))
+ifneq (,$(filter unix osx win,$(platform)))
+ifneq (,$(filter x86_64 i686 i386,$(shell uname -m 2>/dev/null)))
    BLITTER_SIMD_SRC := $(CORE_DIR)/src/blitter_simd_sse2.c
 endif
-endif
-ifeq ($(BLITTER_SIMD_SRC),)
 ifneq (,$(filter aarch64 arm64,$(shell uname -m 2>/dev/null)))
    BLITTER_SIMD_SRC := $(CORE_DIR)/src/blitter_simd_neon.c
 endif
 endif
 endif
+endif
 
 # Fall back to scalar if no SIMD was selected (e.g., exotic platforms)
 ifeq ($(BLITTER_SIMD_SRC),)

src/blitter_simd_neon.c

@@ -108,18 +108,20 @@ static uint8_t neon_zcomp(uint64_t srcz, uint64_t dstz, uint8_t zmode)
  */
 static uint64_t neon_byte_merge(uint64_t src, uint64_t dst, uint16_t mask)
 {
-   /* Build 8-byte selection mask */
-   uint8_t sel[8];
-   sel[0] = (uint8_t)(mask & 0xFF);
-   sel[1] = (mask & 0x0100) ? 0xFF : 0x00;
-   sel[2] = (mask & 0x0200) ? 0xFF : 0x00;
-   sel[3] = (mask & 0x0400) ? 0xFF : 0x00;
-   sel[4] = (mask & 0x0800) ? 0xFF : 0x00;
-   sel[5] = (mask & 0x1000) ? 0xFF : 0x00;
-   sel[6] = (mask & 0x2000) ? 0xFF : 0x00;
-   sel[7] = (mask & 0x4000) ? 0xFF : 0x00;
-
-   uint8x8_t vmask = vld1_u8(sel);
+   /* Build 8-byte selection mask entirely in registers.
+    * Byte 0 uses the low 8 bits directly; bytes 1-7 are all-ones or
+    * all-zero based on mask bits 8-14. */
+   uint64_t sel =
+      ((uint64_t)(mask & 0x00FF)) |
+      ((uint64_t)((mask & 0x0100) ? 0xFF : 0x00) <<  8) |
+      ((uint64_t)((mask & 0x0200) ? 0xFF : 0x00) << 16) |
+      ((uint64_t)((mask & 0x0400) ? 0xFF : 0x00) << 24) |
+      ((uint64_t)((mask & 0x0800) ? 0xFF : 0x00) << 32) |
+      ((uint64_t)((mask & 0x1000) ? 0xFF : 0x00) << 40) |
+      ((uint64_t)((mask & 0x2000) ? 0xFF : 0x00) << 48) |
+      ((uint64_t)((mask & 0x4000) ? 0xFF : 0x00) << 56);
+
+   uint8x8_t vmask = vreinterpret_u8_u64(vcreate_u64(sel));
    uint8x8_t vsrc  = vreinterpret_u8_u64(vcreate_u64(src));
    uint8x8_t vdst  = vreinterpret_u8_u64(vcreate_u64(dst));