diff --git a/Makefile.am b/Makefile.am
index 3c3671620..05e534feb 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -81,12 +81,8 @@ libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES = png.c pngerror.c\
 	png.h pngconf.h pngdebug.h pnginfo.h pngpriv.h pngstruct.h pngusr.dfa
 
 if PNG_ARM_NEON
-  libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += arm/arm_init.c \
-  if PNG_ARM_NEON_INTRINSICS
-        arm/filter_neon_intrinsics.c
-  else
-	arm/filter_neon.S
-  endif
+  libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES += arm/arm_init.c\
+	arm/filter_neon.S arm/filter_neon_intrinsics.c
 endif
 
 nodist_libpng@PNGLIB_MAJOR@@PNGLIB_MINOR@_la_SOURCES = pnglibconf.h
diff --git a/arm/filter_neon.S b/arm/filter_neon.S
index b8aef1053..d77ba59ca 100644
--- a/arm/filter_neon.S
+++ b/arm/filter_neon.S
@@ -20,6 +20,12 @@
 .section .note.GNU-stack,"",%progbits /* mark stack as non-executable */
 #endif
 
+/* Assembler NEON support - only works for 32-bit ARM (i.e. it does not work for
+ * ARM64).  The code in arm/filter_neon_intrinsics.c supports ARM64, however it
+ * only works if -mfpu=neon is specified on the GCC command line.
+ */
+#ifndef __ARM_NEON__ /* else use arm/filter_neon_intrinsics.c */
+
 #ifdef PNG_READ_SUPPORTED
 #if PNG_ARM_NEON_OPT > 0
 
@@ -235,3 +241,4 @@ func    png_read_filter_row_paeth3_neon, export=1
 endfunc
 #endif /* PNG_ARM_NEON_OPT > 0 */
 #endif /* PNG_READ_SUPPORTED */
+#endif /* !__ARM_NEON__ */
diff --git a/arm/filter_neon_intrinsics.c b/arm/filter_neon_intrinsics.c
index 334c1f5c1..9f15cd574 100644
--- a/arm/filter_neon_intrinsics.c
+++ b/arm/filter_neon_intrinsics.c
@@ -14,11 +14,30 @@
 
 #include "pngpriv.h"
 
-#ifdef PNG_READ_SUPPORTED
-#if PNG_ARM_NEON_OPT > 0
+/* This code requires -mfpu=neon on the command line: */
+#ifdef __ARM_NEON__ /* else use arm/filter_neon.S */
 
 #include <arm_neon.h>
 
+/* libpng row pointers are not necessarily aligned to any particular boundary,
+ * however this code will only work with appropriate alignment.  arm/arm_init.c
+ * checks for this (and will not compile unless it is done), this code uses
+ * variants of png_aligncast to avoid compiler warnings.
+ */
+#define png_ptr(type,pointer) png_aligncast(type *,pointer)
+#define png_ptrc(type,pointer) png_aligncastconst(const type *,pointer)
+
+/* The following relies on a variable 'temp_pointer' being declared with type
+ * 'type'.  This is written this way just to hide the GCC strict aliasing
+ * warning; note that the code is safe because there never is an alias between
+ * the input and output pointers.
+ */
+#define png_ldr(type,pointer)\
+   (temp_pointer = png_ptr(type,pointer), *temp_pointer)
+
+#ifdef PNG_READ_SUPPORTED
+#if PNG_ARM_NEON_OPT > 0
+
 void
 png_read_filter_row_up_neon(png_row_infop row_info, png_bytep row,
    png_const_bytep prev_row)
@@ -27,10 +46,10 @@ png_read_filter_row_up_neon(png_row_infop row_info, png_bytep row,
    png_bytep rp_stop = row + row_info->rowbytes;
    png_const_bytep pp = prev_row;
 
-   uint8x16_t qrp, qpp;
-
    for (; rp != rp_stop; rp += 16, pp += 16)
    {
+      uint8x16_t qrp, qpp;
+
       qrp = vld1q_u8(rp);
       qpp = vld1q_u8(pp);
       qrp = vaddq_u8(qrp, qpp);
@@ -38,24 +57,25 @@ png_read_filter_row_up_neon(png_row_infop row_info, png_bytep row,
    }
 }
 
-void inline
+void
 png_read_filter_row_sub3_neon(png_row_infop row_info, png_bytep row,
    png_const_bytep prev_row)
 {
    png_bytep rp = row;
    png_bytep rp_stop = row + row_info->rowbytes;
 
-   PNG_UNUSED(prev_row)
+   uint8x16_t vtmp = vld1q_u8(rp);
+   uint8x8x2_t *vrpt = png_ptr(uint8x8x2_t, &vtmp);
+   uint8x8x2_t vrp = *vrpt;
 
-   uint8x8_t vtmp1, vtmp2;
    uint8x8x4_t vdest;
    vdest.val[3] = vdup_n_u8(0);
 
-   uint8x16_t vtmp = vld1q_u8(rp);
-   uint8x8x2_t vrp = *((uint8x8x2_t *)(&vtmp));
-
    for (; rp != rp_stop;)
    {
+      uint8x8_t vtmp1, vtmp2;
+      uint32x2_t *temp_pointer;
+
       vtmp1 = vext_u8(vrp.val[0], vrp.val[1], 3);
       vdest.val[0] = vadd_u8(vdest.val[3], vrp.val[0]);
       vtmp2 = vext_u8(vrp.val[0], vrp.val[1], 6);
@@ -66,17 +86,20 @@ png_read_filter_row_sub3_neon(png_row_infop row_info, png_bytep row,
       vdest.val[3] = vadd_u8(vdest.val[2], vtmp1);
 
       vtmp = vld1q_u8(rp + 12);
-      vrp = *((uint8x8x2_t *)(&vtmp));
+      vrpt = png_ptr(uint8x8x2_t, &vtmp);
+      vrp = *vrpt;
 
-      vst1_lane_u32((uint32_t *)rp, *((uint32x2_t *)(&vdest.val[0])), 0);
+      vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[0]), 0);
       rp += 3;
-      vst1_lane_u32((uint32_t *)rp, *((uint32x2_t *)(&vdest.val[1])), 0);
+      vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[1]), 0);
       rp += 3;
-      vst1_lane_u32((uint32_t *)rp, *((uint32x2_t *)(&vdest.val[2])), 0);
+      vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[2]), 0);
       rp += 3;
-      vst1_lane_u32((uint32_t *)rp, *((uint32x2_t *)(&vdest.val[3])), 0);
+      vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[3]), 0);
       rp += 3;
    }
+
+   PNG_UNUSED(prev_row)
 }
 
 void
@@ -86,22 +109,24 @@ png_read_filter_row_sub4_neon(png_row_infop row_info, png_bytep row,
    png_bytep rp = row;
    png_bytep rp_stop = row + row_info->rowbytes;
 
-   PNG_UNUSED(prev_row)
-
    uint8x8x4_t vdest;
    vdest.val[3] = vdup_n_u8(0);
 
    for (; rp != rp_stop; rp += 16)
    {
-      uint32x2x4_t vtmp = vld4_u32((uint32_t *)rp);
-      uint8x8x4_t vrp = *((uint8x8x4_t *)(&vtmp));
+      uint32x2x4_t vtmp = vld4_u32(png_ptr(uint32_t,rp));
+      uint8x8x4_t *vrpt = png_ptr(uint8x8x4_t,&vtmp);
+      uint8x8x4_t vrp = *vrpt;
+      uint32x2x4_t *temp_pointer;
 
       vdest.val[0] = vadd_u8(vdest.val[3], vrp.val[0]);
       vdest.val[1] = vadd_u8(vdest.val[0], vrp.val[1]);
       vdest.val[2] = vadd_u8(vdest.val[1], vrp.val[2]);
       vdest.val[3] = vadd_u8(vdest.val[2], vrp.val[3]);
-      vst4_lane_u32((uint32_t *)rp, *((uint32x2x4_t *)(&vdest)), 0);
+      vst4_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2x4_t,&vdest), 0);
    }
+
+   PNG_UNUSED(prev_row)
 }
 
 void
@@ -113,18 +138,27 @@ png_read_filter_row_avg3_neon(png_row_infop row_info, png_bytep row,
    png_bytep rp_stop = row + row_info->rowbytes;
 
    uint8x16_t vtmp;
-   uint8x8x2_t vrp, vpp;
-   uint8x8_t vtmp1, vtmp2, vtmp3;
+   uint8x8x2_t *vrpt;
+   uint8x8x2_t vrp;
    uint8x8x4_t vdest;
    vdest.val[3] = vdup_n_u8(0);
 
    vtmp = vld1q_u8(rp);
-   vrp = *((uint8x8x2_t *)(&vtmp));
+   vrpt = png_ptr(uint8x8x2_t,&vtmp);
+   vrp = *vrpt;
 
    for (; rp != rp_stop; pp += 12)
    {
+      uint8x8_t vtmp1, vtmp2, vtmp3;
+
+      uint8x8x2_t *vppt;
+      uint8x8x2_t vpp;
+
+      uint32x2_t *temp_pointer;
+
       vtmp = vld1q_u8(pp);
-      vpp = *((uint8x8x2_t *)(&vtmp));
+      vppt = png_ptr(uint8x8x2_t,&vtmp);
+      vpp = *vppt;
 
       vtmp1 = vext_u8(vrp.val[0], vrp.val[1], 3);
       vdest.val[0] = vhadd_u8(vdest.val[3], vpp.val[0]);
@@ -139,7 +173,8 @@ png_read_filter_row_avg3_neon(png_row_infop row_info, png_bytep row,
       vtmp1 = vext_u8(vrp.val[1], vrp.val[1], 1);
 
       vtmp = vld1q_u8(rp + 12);
-      vrp = *((uint8x8x2_t *)(&vtmp));
+      vrpt = png_ptr(uint8x8x2_t,&vtmp);
+      vrp = *vrpt;
 
       vdest.val[2] = vhadd_u8(vdest.val[1], vtmp2);
       vdest.val[2] = vadd_u8(vdest.val[2], vtmp3);
@@ -149,13 +184,13 @@ png_read_filter_row_avg3_neon(png_row_infop row_info, png_bytep row,
       vdest.val[3] = vhadd_u8(vdest.val[2], vtmp2);
       vdest.val[3] = vadd_u8(vdest.val[3], vtmp1);
 
-      vst1_lane_u32((uint32_t *)rp, *((uint32x2_t *)(&vdest.val[0])), 0);
+      vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[0]), 0);
       rp += 3;
-      vst1_lane_u32((uint32_t *)rp, *((uint32x2_t *)(&vdest.val[1])), 0);
+      vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[1]), 0);
       rp += 3;
-      vst1_lane_u32((uint32_t *)rp, *((uint32x2_t *)(&vdest.val[2])), 0);
+      vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[2]), 0);
       rp += 3;
-      vst1_lane_u32((uint32_t *)rp, *((uint32x2_t *)(&vdest.val[3])), 0);
+      vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[3]), 0);
       rp += 3;
    }
 }
@@ -168,17 +203,22 @@ png_read_filter_row_avg4_neon(png_row_infop row_info, png_bytep row,
    png_bytep rp_stop = row + row_info->rowbytes;
    png_const_bytep pp = prev_row;
 
-   uint32x2x4_t vtmp;
-   uint8x8x4_t vrp, vpp;
    uint8x8x4_t vdest;
    vdest.val[3] = vdup_n_u8(0);
 
    for (; rp != rp_stop; rp += 16, pp += 16)
    {
-      vtmp = vld4_u32((uint32_t *)rp);
-      vrp = *((uint8x8x4_t *)(&vtmp));
-      vtmp = vld4_u32((uint32_t *)pp);
-      vpp = *((uint8x8x4_t *)(&vtmp));
+      uint32x2x4_t vtmp;
+      uint8x8x4_t *vrpt, *vppt;
+      uint8x8x4_t vrp, vpp;
+      uint32x2x4_t *temp_pointer;
+
+      vtmp = vld4_u32(png_ptr(uint32_t,rp));
+      vrpt = png_ptr(uint8x8x4_t,&vtmp);
+      vrp = *vrpt;
+      vtmp = vld4_u32(png_ptrc(uint32_t,pp));
+      vppt = png_ptr(uint8x8x4_t,&vtmp);
+      vpp = *vppt;
 
       vdest.val[0] = vhadd_u8(vdest.val[3], vpp.val[0]);
       vdest.val[0] = vadd_u8(vdest.val[0], vrp.val[0]);
@@ -189,11 +229,11 @@ png_read_filter_row_avg4_neon(png_row_infop row_info, png_bytep row,
       vdest.val[3] = vhadd_u8(vdest.val[2], vpp.val[3]);
       vdest.val[3] = vadd_u8(vdest.val[3], vrp.val[3]);
 
-      vst4_lane_u32((uint32_t *)rp, *((uint32x2x4_t *)(&vdest)), 0);
+      vst4_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2x4_t,&vdest), 0);
    }
 }
 
-uint8x8_t
+static uint8x8_t
 paeth(uint8x8_t a, uint8x8_t b, uint8x8_t c)
 {
    uint8x8_t d, e;
@@ -229,19 +269,26 @@ png_read_filter_row_paeth3_neon(png_row_infop row_info, png_bytep row,
    png_bytep rp_stop = row + row_info->rowbytes;
 
    uint8x16_t vtmp;
-   uint8x8x2_t vrp, vpp;
-   uint8x8_t vtmp1, vtmp2, vtmp3;
+   uint8x8x2_t *vrpt;
+   uint8x8x2_t vrp;
    uint8x8_t vlast = vdup_n_u8(0);
    uint8x8x4_t vdest;
    vdest.val[3] = vdup_n_u8(0);
 
    vtmp = vld1q_u8(rp);
-   vrp = *((uint8x8x2_t *)(&vtmp));
+   vrpt = png_ptr(uint8x8x2_t,&vtmp);
+   vrp = *vrpt;
 
    for (; rp != rp_stop; pp += 12)
    {
+      uint8x8x2_t *vppt;
+      uint8x8x2_t vpp;
+      uint8x8_t vtmp1, vtmp2, vtmp3;
+      uint32x2_t *temp_pointer;
+
       vtmp = vld1q_u8(pp);
-      vpp = *((uint8x8x2_t *)(&vtmp));
+      vppt = png_ptr(uint8x8x2_t,&vtmp);
+      vpp = *vppt;
 
       vdest.val[0] = paeth(vdest.val[3], vpp.val[0], vlast);
       vdest.val[0] = vadd_u8(vdest.val[0], vrp.val[0]);
@@ -260,20 +307,21 @@ png_read_filter_row_paeth3_neon(png_row_infop row_info, png_bytep row,
       vtmp2 = vext_u8(vpp.val[1], vpp.val[1], 1);
 
       vtmp = vld1q_u8(rp + 12);
-      vrp = *((uint8x8x2_t *)(&vtmp));
+      vrpt = png_ptr(uint8x8x2_t,&vtmp);
+      vrp = *vrpt;
 
       vdest.val[3] = paeth(vdest.val[2], vtmp2, vtmp3);
       vdest.val[3] = vadd_u8(vdest.val[3], vtmp1);
 
       vlast = vtmp2;
 
-      vst1_lane_u32((uint32_t *)rp, *((uint32x2_t *)(&vdest.val[0])), 0);
+      vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[0]), 0);
       rp += 3;
-      vst1_lane_u32((uint32_t *)rp, *((uint32x2_t *)(&vdest.val[1])), 0);
+      vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[1]), 0);
       rp += 3;
-      vst1_lane_u32((uint32_t *)rp, *((uint32x2_t *)(&vdest.val[2])), 0);
+      vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[2]), 0);
       rp += 3;
-      vst1_lane_u32((uint32_t *)rp, *((uint32x2_t *)(&vdest.val[3])), 0);
+      vst1_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2_t,&vdest.val[3]), 0);
       rp += 3;
    }
 }
@@ -286,18 +334,23 @@ png_read_filter_row_paeth4_neon(png_row_infop row_info, png_bytep row,
    png_bytep rp_stop = row + row_info->rowbytes;
    png_const_bytep pp = prev_row;
 
-   uint32x2x4_t vtmp;
-   uint8x8x4_t vrp, vpp;
    uint8x8_t vlast = vdup_n_u8(0);
    uint8x8x4_t vdest;
    vdest.val[3] = vdup_n_u8(0);
 
    for (; rp != rp_stop; rp += 16, pp += 16)
    {
-      vtmp = vld4_u32((uint32_t *)rp);
-      vrp = *((uint8x8x4_t *)(&vtmp));
-      vtmp = vld4_u32((uint32_t *)pp);
-      vpp = *((uint8x8x4_t *)(&vtmp));
+      uint32x2x4_t vtmp;
+      uint8x8x4_t *vrpt, *vppt;
+      uint8x8x4_t vrp, vpp;
+      uint32x2x4_t *temp_pointer;
+
+      vtmp = vld4_u32(png_ptr(uint32_t,rp));
+      vrpt = png_ptr(uint8x8x4_t,&vtmp);
+      vrp = *vrpt;
+      vtmp = vld4_u32(png_ptrc(uint32_t,pp));
+      vppt = png_ptr(uint8x8x4_t,&vtmp);
+      vpp = *vppt;
 
       vdest.val[0] = paeth(vdest.val[3], vpp.val[0], vlast);
       vdest.val[0] = vadd_u8(vdest.val[0], vrp.val[0]);
@@ -310,9 +363,10 @@ png_read_filter_row_paeth4_neon(png_row_infop row_info, png_bytep row,
 
       vlast = vpp.val[3];
 
-      vst4_lane_u32((uint32_t *)rp, *((uint32x2x4_t *)(&vdest)), 0);
+      vst4_lane_u32(png_ptr(uint32_t,rp), png_ldr(uint32x2x4_t,&vdest), 0);
    }
 }
 
 #endif /* PNG_ARM_NEON_OPT > 0 */
 #endif /* PNG_READ_SUPPORTED */
+#endif /* __ARM_NEON__ */
diff --git a/configure.ac b/configure.ac
index a7bcf3ba8..1a6019298 100644
--- a/configure.ac
+++ b/configure.ac
@@ -248,13 +248,13 @@ AC_ARG_ENABLE([arm-neon],
                    [Disable ARM Neon optimizations])
          # Prevent inclusion of the assembler files below:
          enable_arm_neon=no;;
-      check|check-intrinsics)
+      check)
          AC_DEFINE([PNG_ARM_NEON_CHECK_SUPPORTED], [],
                    [Check for ARM Neon support at run-time]);;
-      api|api-intrinsics)
+      api)
          AC_DEFINE([PNG_ARM_NEON_API_SUPPORTED], [],
                    [Turn on ARM Neon optimizations at run-time]);;
-      yes|on|yes-intrinsics|on-intrinsics)
+      yes|on)
          AC_DEFINE([PNG_ARM_NEON_OPT], [2],
                    [Enable ARM Neon optimizations])
          AC_MSG_WARN([--enable-arm-neon: please specify 'check' or 'api', if]
@@ -275,16 +275,6 @@ AM_CONDITIONAL([PNG_ARM_NEON],
       *)    test "$enable_arm_neon" != '';;
     esac])
 
-case "$enable_arm_neon" in
-    *-intrinsics)
-       use_arm_intrinsics='yes';;
-    *)
-       use_arm_intrinsics='no';;
-esac
-
-AM_CONDITIONAL([PNG_ARM_NEON_INTRINSICS],
-   [test "$use_arm_intrinsics" = 'yes'])
-
 AC_MSG_NOTICE([[Extra options for compiler: $PNG_COPTS]])
 
 # Config files, substituting as above