add runtime NEON detection

configure gets 2 new options:
--enable-neon / --enable-neon-rtcd

the NEON modules are split to their own convenience lib and built with
auto-detected flags if none are given via CFLAGS.

the /proc/cpuinfo check will only be used for armv7 targets whose
toolchain does not enable NEON by default or didn't have NEON forced by
the CFLAGS from the environment.

Change-Id: I2755bc1d065d5d6ee6143b44978c2082f8bef1c5

src/dsp/Makefile.am

@@ -1,6 +1,7 @@
 noinst_LTLIBRARIES = libwebpdsp.la libwebpdsp_avx2.la
 noinst_LTLIBRARIES += libwebpdsp_sse2.la libwebpdspdecode_sse2.la
 noinst_LTLIBRARIES += libwebpdsp_sse41.la libwebpdspdecode_sse41.la
+noinst_LTLIBRARIES += libwebpdsp_neon.la libwebpdspdecode_neon.la
 
 if BUILD_LIBWEBPDECODER
   noinst_LTLIBRARIES += libwebpdspdecode.la
@@ -18,23 +19,18 @@ COMMON_SOURCES += dec.c
 COMMON_SOURCES += dec_clip_tables.c
 COMMON_SOURCES += dec_mips32.c
 COMMON_SOURCES += dec_mips_dsp_r2.c
-COMMON_SOURCES += dec_neon.c
 COMMON_SOURCES += dsp.h
 COMMON_SOURCES += filters.c
 COMMON_SOURCES += filters_mips_dsp_r2.c
 COMMON_SOURCES += lossless.c
 COMMON_SOURCES += lossless.h
 COMMON_SOURCES += lossless_mips_dsp_r2.c
-COMMON_SOURCES += lossless_neon.c
 COMMON_SOURCES += mips_macro.h
-COMMON_SOURCES += neon.h
 COMMON_SOURCES += rescaler.c
 COMMON_SOURCES += rescaler_mips32.c
 COMMON_SOURCES += rescaler_mips_dsp_r2.c
-COMMON_SOURCES += rescaler_neon.c
 COMMON_SOURCES += upsampling.c
 COMMON_SOURCES += upsampling_mips_dsp_r2.c
-COMMON_SOURCES += upsampling_neon.c
 COMMON_SOURCES += yuv.c
 COMMON_SOURCES += yuv.h
 COMMON_SOURCES += yuv_mips32.c
@@ -49,11 +45,9 @@ ENC_SOURCES += cost_mips_dsp_r2.c
 ENC_SOURCES += enc.c
 ENC_SOURCES += enc_mips32.c
 ENC_SOURCES += enc_mips_dsp_r2.c
-ENC_SOURCES += enc_neon.c
 ENC_SOURCES += lossless_enc.c
 ENC_SOURCES += lossless_enc_mips32.c
 ENC_SOURCES += lossless_enc_mips_dsp_r2.c
-ENC_SOURCES += lossless_enc_neon.c
 
 libwebpdsp_avx2_la_SOURCES =
 libwebpdsp_avx2_la_SOURCES += enc_avx2.c
@@ -77,6 +71,15 @@ libwebpdspdecode_sse2_la_SOURCES += yuv_sse2.c
 libwebpdspdecode_sse2_la_CPPFLAGS = $(libwebpdsp_sse2_la_CPPFLAGS)
 libwebpdspdecode_sse2_la_CFLAGS = $(libwebpdsp_sse2_la_CFLAGS)
 
+libwebpdspdecode_neon_la_SOURCES =
+libwebpdspdecode_neon_la_SOURCES += dec_neon.c
+libwebpdspdecode_neon_la_SOURCES += lossless_neon.c
+libwebpdspdecode_neon_la_SOURCES += neon.h
+libwebpdspdecode_neon_la_SOURCES += rescaler_neon.c
+libwebpdspdecode_neon_la_SOURCES += upsampling_neon.c
+libwebpdspdecode_neon_la_CPPFLAGS = $(libwebpdsp_neon_la_CPPFLAGS)
+libwebpdspdecode_neon_la_CFLAGS = $(libwebpdsp_neon_la_CFLAGS)
+
 libwebpdsp_sse2_la_SOURCES =
 libwebpdsp_sse2_la_SOURCES += argb_sse2.c
 libwebpdsp_sse2_la_SOURCES += cost_sse2.c
@@ -93,6 +96,13 @@ libwebpdsp_sse41_la_CPPFLAGS = $(libwebpdsp_la_CPPFLAGS)
 libwebpdsp_sse41_la_CFLAGS = $(AM_CFLAGS) $(SSE41_FLAGS)
 libwebpdsp_sse41_la_LIBADD = libwebpdspdecode_sse41.la
 
+libwebpdsp_neon_la_SOURCES =
+libwebpdsp_neon_la_SOURCES += enc_neon.c
+libwebpdsp_neon_la_SOURCES += lossless_enc_neon.c
+libwebpdsp_neon_la_CPPFLAGS = $(libwebpdsp_la_CPPFLAGS)
+libwebpdsp_neon_la_CFLAGS = $(AM_CFLAGS) $(NEON_FLAGS)
+libwebpdsp_neon_la_LIBADD = libwebpdspdecode_neon.la
+
 libwebpdsp_la_SOURCES = $(COMMON_SOURCES) $(ENC_SOURCES)
 
 noinst_HEADERS =
@@ -106,6 +116,7 @@ libwebpdsp_la_LDFLAGS = -lm
 libwebpdsp_la_LIBADD =
 libwebpdsp_la_LIBADD += libwebpdsp_avx2.la libwebpdsp_sse2.la
 libwebpdsp_la_LIBADD += libwebpdsp_sse41.la
+libwebpdsp_la_LIBADD += libwebpdsp_neon.la
 
 if BUILD_LIBWEBPDECODER
   libwebpdspdecode_la_SOURCES = $(COMMON_SOURCES)
@@ -115,4 +126,5 @@ if BUILD_LIBWEBPDECODER
   libwebpdspdecode_la_LIBADD =
   libwebpdspdecode_la_LIBADD += libwebpdspdecode_sse2.la
   libwebpdspdecode_la_LIBADD += libwebpdspdecode_sse41.la
+  libwebpdspdecode_la_LIBADD += libwebpdspdecode_neon.la
 endif

src/dsp/cpu.c

@@ -13,6 +13,11 @@
 
 #include "./dsp.h"
 
+#if defined(WEBP_HAVE_NEON_RTCD)
+#include <stdio.h>
+#include <string.h>
+#endif
+
 #if defined(WEBP_ANDROID_NEON)
 #include <cpu-features.h>
 #endif
@@ -142,8 +147,27 @@ VP8CPUInfo VP8GetCPUInfo = AndroidCPUInfo;
 // define a dummy function to enable turning off NEON at runtime by setting
 // VP8DecGetCPUInfo = NULL
 static int armCPUInfo(CPUFeature feature) {
-  (void)feature;
+  if (feature != kNEON) return 0;
+#if defined(__linux__) && defined(WEBP_HAVE_NEON_RTCD)
+  {
+    int has_neon = 0;
+    char line[200];
+    FILE* const cpuinfo = fopen("/proc/cpuinfo", "r");
+    if (cpuinfo == NULL) return 0;
+    while (fgets(line, sizeof(line), cpuinfo)) {
+      if (!strncmp(line, "Features", 8)) {
+        if (strstr(line, " neon ") != NULL) {
+          has_neon = 1;
+          break;
+        }
+      }
+    }
+    fclose(cpuinfo);
+    return has_neon;
+  }
+#else
   return 1;
+#endif
 }
 VP8CPUInfo VP8GetCPUInfo = armCPUInfo;
 #elif defined(WEBP_USE_MIPS32) || defined(WEBP_USE_MIPS_DSP_R2)

src/dsp/dsp.h

@@ -72,7 +72,8 @@ extern "C" {
 // The intrinsics currently cause compiler errors with arm-nacl-gcc and the
 // inline assembly would need to be modified for use with Native Client.
 #if (defined(__ARM_NEON__) || defined(WEBP_ANDROID_NEON) || \
-     defined(__aarch64__)) && !defined(__native_client__)
+     defined(__aarch64__) || defined(WEBP_HAVE_NEON)) && \
+    !defined(__native_client__)
 #define WEBP_USE_NEON
 #endif