Support architectures other than x86

Theoretically, at least.

Makefile.am

@@ -1,6 +1,11 @@
-AM_CFLAGS = -O2 -msse2 -mfpmath=sse -Wall -Wextra -Wno-unused-parameter
+warning_flags = -Wall -Wextra -Wno-unused-parameter -Wshadow
+common_cflags = -O2 $(MFLAGS) $(warning_flags)
+AM_CFLAGS = $(common_cflags)
+AM_CXXFLAGS = -std=c++11 $(common_cflags)
 
-AM_CPPFLAGS = $(VapourSynth_CFLAGS)
+AM_CPPFLAGS = $(VapourSynth_CFLAGS) -DNNEDI3_DATADIR='"$(pkgdatadir)"'
+
+dist_pkgdata_DATA = src/nnedi3\ weights.bin
 
 lib_LTLIBRARIES = libnnedi3.la
 
@@ -11,11 +16,13 @@ yasm_verbose_0 = @echo "  YASM    " $@;
 .asm.lo:
 	$(yasm_verbose)$(LIBTOOL) $(AM_V_lt) --mode=compile --tag=CC $(AS) $(ASFLAGS) -o $@ $< -prefer-non-pic
 
-libnnedi3_la_SOURCES = src/cpufeatures.c \
-					   src/cpufeatures.h \
-					   src/nnedi3.c \
-					   src/asm/binary1.asm \
-					   src/asm/cpu-a.asm \
-					   src/asm/nnedi3.asm
+libnnedi3_la_SOURCES = src/nnedi3.cpp \
+					   src/cpufeatures.cpp \
+					   src/cpufeatures.h
+
+if NNEDI3_X86
+libnnedi3_la_SOURCES += src/asm/cpu-a.asm \
+						src/asm/nnedi3.asm
+endif
 
 libnnedi3_la_LDFLAGS = -no-undefined -avoid-version $(PLUGINLDFLAGS)

configure.ac

@@ -1,6 +1,6 @@
-AC_INIT([nnedi3], [1], [https://github.com/dubhater/vapoursynth-nnedi3/issues], [nnedi3], [https://github.com/dubhater/vapoursynth-nnedi3/])
+AC_INIT([nnedi3], [3], [https://github.com/dubhater/vapoursynth-nnedi3/issues], [nnedi3], [https://github.com/dubhater/vapoursynth-nnedi3/])
 
-: ${CFLAGS=""}
+: ${CXXFLAGS=""}
 
 AM_INIT_AUTOMAKE([foreign no-dist-gzip dist-xz subdir-objects no-define])
 AM_SILENT_RULES([yes])
@@ -9,53 +9,83 @@ LT_INIT([disable-static win32-dll])
 
 AC_CANONICAL_HOST
 
-AC_PROG_CC_C99
+AC_PROG_CXX
 
 AC_SEARCH_LIBS([sqrt], [m], [], [
    AC_MSG_ERROR([unable to find the sqrt() function])
 ])
 
-AC_CHECK_PROGS([YASM], [yasm])
-AS_IF(
-   [test "x$YASM" = "x"],
-   [AC_MSG_ERROR([yasm required but not found])],
-   [AS="$YASM"]
+
+X86="false"
+PPC="false"
+ARM="false"
+
+AS_CASE(
+        [$host_cpu],
+        [i?86], [BITS="32" ASFLAGS="$ASFLAGS -DARCH_X86_64=0" X86="true"],
+        [x86_64], [BITS="64" ASFLAGS="$ASFLAGS -DARCH_X86_64=1 -DPIC -m amd64" X86="true"],
+        [powerpc*], [PPC="true"],
+        [arm*], [ARM="true"] # Maybe doesn't work for all arm systems?
 )
 
 AS_CASE(
-  [$host_cpu],
-  [i?86],   [BITS="32" ASFLAGS="$ASFLAGS -DARCH_X86_64=0"],
-  [x86_64], [BITS="64" ASFLAGS="$ASFLAGS -DARCH_X86_64=1 -DPIC -m amd64"],
-  [AC_MSG_ERROR([Currently nnedi3 can only be compiled for Intel CPUs.])]
+        [$host_os],
+        [darwin*],
+        [
+         ASFLAGS="$ASFLAGS -f macho$BITS -DPREFIX"
+        ],
+        [*linux*|gnu*|dragonfly*|*bsd*], # The BSDs are close enough, right?
+        [
+         ASFLAGS="$ASFLAGS -f elf"
+        ],
+        [cygwin*|mingw*],
+        [
+         ASFLAGS="$ASFLAGS -f win32"
+         AS_IF(
+               [test "x$BITS" = "x32"],
+               [
+                ASFLAGS="$ASFLAGS -DPREFIX"
+                AC_SUBST([PLUGINLDFLAGS], ["-Wl,--kill-at"])
+               ]
+         )
+        ],
+        [AC_MSG_ERROR(["Unknown host OS: $host_os"])]
 )
 
-ASFLAGS="$ASFLAGS -Dprivate_prefix=nnedi3"
+AS_IF(
+      [test "x$X86" = "xtrue"],
+      [
+       AC_DEFINE([NNEDI3_X86])
 
-AS_CASE(
-   [$host_os],
-   [darwin*],
-   [
-      ASFLAGS="$ASFLAGS -f macho$BITS -DPREFIX"
-   ],
-   [*linux*|gnu*|dragonfly*|*bsd*],
-   [
-      ASFLAGS="$ASFLAGS -f elf"
-   ],
-   [cygwin*|mingw*],
-   [
-      ASFLAGS="$ASFLAGS -f win32"
-      AS_IF(
-         [test "x$BITS" = "x32"],
-         [
-          ASFLAGS="$ASFLAGS -DPREFIX"
-          AC_SUBST([PLUGINLDFLAGS], ["-Wl,--kill-at"])
-         ]
-      )
-   ],
-   [AC_MSG_ERROR([Unknown host OS])]
+       ASFLAGS="$ASFLAGS -Dprivate_prefix=nnedi3"
+
+       AC_SUBST([MFLAGS], ["-mfpmath=sse -msse2"])
+
+       AC_CHECK_PROGS([YASM], [yasm])
+       AS_IF(
+             [test "x$YASM" = "x"],
+             [AC_MSG_ERROR([yasm required but not found])],
+             [AS="$YASM"]
+       )
+      ]
+)
+
+AS_IF(
+      [test "x$PPC" = "xtrue"],
+      [AC_DEFINE([NNEDI3_POWERPC])]
+)
+
+AS_IF(
+      [test "x$ARM" = "xtrue"],
+      [AC_DEFINE([NNEDI3_ARM])]
 )
+
 AC_SUBST([ASFLAGS])
 
+
+AM_CONDITIONAL([NNEDI3_X86], [test "x$X86" = "xtrue"])
+
+
 PKG_CHECK_MODULES([VapourSynth], [vapoursynth])
 
 AC_CONFIG_FILES([Makefile])

readme.rst

@@ -9,6 +9,8 @@ This is a port of tritical's nnedi3 filter.
 Usage
 =====
 
+The file ``nnedi3 weights.bin`` is required. In Windows, it must be located in the same folder as ``libnnedi3.dll``. Everywhere else it can be located either in the same folder as ``libnnedi3.so``/``libnnedi3.dylib``, or in ``$prefix/share/nnedi3/``. The build system installs it at the latter location automatically.
+
 ::
 
    nnedi3.nnedi3(clip clip, int field[, bint dh=False, bint Y=True, bint U=True, bint V=True, int nsize=6, int nns=1, int qual=1, int etype=0, int pscrn=2, bint opt=True, int fapprox=15])
@@ -63,7 +65,7 @@ Compilation
    ./configure
    make
 
-yasm is currently not optional.
+On x86, yasm is currently not optional.
 
 DLLs can be found in the "releases" section.
 

src/cpufeatures.c → src/cpufeatures.cpp

@@ -23,10 +23,12 @@
 
 #include "cpufeatures.h"
 
-#ifdef VS_TARGET_CPU_X86
+#ifdef NNEDI3_X86
+extern "C" {
 extern void nnedi3_cpu_cpuid(uint32_t index, uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx);
 extern void nnedi3_cpu_xgetbv(uint32_t op, uint32_t *eax, uint32_t *edx);
 extern void nnedi3_cpu_cpuid_test(void);
+}
 
 void getCPUFeatures(CPUFeatures *cpuFeatures) {
     memset(cpuFeatures, 0, sizeof(CPUFeatures));
@@ -62,32 +64,30 @@ void getCPUFeatures(CPUFeatures *cpuFeatures) {
     }
 
 }
-#elif defined(VS_TARGET_OS_LINUX)
+#else
 #include <sys/auxv.h>
 
 void getCPUFeatures(CPUFeatures *cpuFeatures) {
+    memset(cpuFeatures, 0, sizeof(CPUFeatures));
+
     unsigned long long hwcap = getauxval(AT_HWCAP);
 
     cpuFeatures->can_run_vs = 1;
 
-#ifdef VS_TARGET_CPU_ARM
+#ifdef NNEDI3_ARM
     cpuFeatures->half_fp = !!(hwcap & HWCAP_ARM_HALF);
     cpuFeatures->edsp = !!(hwcap & HWCAP_ARM_EDSP);
     cpuFeatures->iwmmxt = !!(hwcap & HWCAP_ARM_IWMMXT);
     cpuFeatures->neon = !!(hwcap & HWCAP_ARM_NEON);
     cpuFeatures->fast_mult = !!(hwcap & HWCAP_ARM_FAST_MULT);
     cpuFeatures->idiv_a = !!(hwcap & HWCAP_ARM_IDIVA);
-#elif defined(VS_TARGET_CPU_POWERPC)
+#elif defined(NNEDI3_POWERPC)
     cpuFeatures->altivec = !!(hwcap & PPC_FEATURE_HAS_ALTIVEC);
     cpuFeatures->spe = !!(hwcap & PPC_FEATURE_HAS_SPE);
     cpuFeatures->efp_single = !!(hwcap & PPC_FEATURE_HAS_EFP_SINGLE);
     cpuFeatures->efp_double = !!(hwcap & PPC_FEATURE_HAS_EFP_DOUBLE);
     cpuFeatures->dfp = !!(hwcap & PPC_FEATURE_HAS_DFP);
     cpuFeatures->vsx = !!(hwcap & PPC_FEATURE_HAS_VSX);
-#else
-#error Do not know how to get CPU features on Linux.
 #endif
 }
-#else
-#error Do not know how to get CPU features.
 #endif

src/cpufeatures.h

@@ -18,13 +18,12 @@
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */
 
-#define VS_TARGET_CPU_X86
 
 typedef struct CPUFeatures {
     // This is to determine if the cpu is up to the minimum requirements in terms of supported instructions
     // that the VapourSynth core uses.
     char can_run_vs;
-#ifdef VS_TARGET_CPU_X86
+#ifdef NNEDI3_X86
     // On x86, all features up to sse2 are required.
     char sse3;
     char ssse3;
@@ -34,31 +33,24 @@ typedef struct CPUFeatures {
     char fma4;
     char avx;
     char avx2;
-#elif defined(VS_TARGET_CPU_ARM)
+#elif defined(NNEDI3_ARM)
     // On ARM, VFP-D16+ (16 double registers or more) is required.
     char half_fp;
     char edsp;
     char iwmmxt;
     char neon;
     char fast_mult;
     char idiv_a;
-#elif defined(VS_TARGET_CPU_POWERPC)
+#elif defined(NNEDI3_POWERPC)
     // On PowerPC, FPU and MMU are required.
     char altivec;
     char spe;
     char efp_single;
     char efp_double;
     char dfp;
     char vsx;
-#else
-#error No VS_TARGET_CPU_* defined/handled!
 #endif
 } CPUFeatures;
 
-#ifdef __cplusplus
-#define CPU_FEATURES_EXTERN_C extern "C"
-#else
-#define CPU_FEATURES_EXTERN_C
-#endif
 
-CPU_FEATURES_EXTERN_C void getCPUFeatures(CPUFeatures *cpuFeatures);
+void getCPUFeatures(CPUFeatures *cpuFeatures);