From f48abe8308ba7a67e443ad0911e06d62fd47ba91 Mon Sep 17 00:00:00 2001
From: Viswanath Puttagunta <viswanath.puttagunta@linaro.org>
Date: Fri, 15 May 2015 12:42:19 -0500
Subject: armv7(float): Optimize encode usecase using NE10 library

Optimize opus encode (float only) usecase using ARM NE10
library. Mainly effects opus_fft and ctl_mdct_forward
and related functions.

This optimization can be used for ARM CPUs that have NEON
VFP unit. This patch only enables optimizations for ARMv7.

Official ARM NE10 library page available at
http://projectne10.github.io/Ne10/

To enable this optimization, use
--enable-intrinsics --with-NE10=<install_prefix>
or
--enable-intrinsics --with-NE10-libraries=<NE10_lib_dir> --with-NE10-includes=<NE10_includes_dir>

Compile time checks made during configure process to make sure
optimization option available only when compiler supports NEON
instrinsics.

Runtime checks made to make sure optimized functions only called
on appropriate hardware.

Signed-off-by: Timothy B. Terriberry <tterribe@xiph.org>
---
 configure.ac | 87 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 86 insertions(+), 1 deletion(-)

(limited to 'configure.ac')

diff --git a/configure.ac b/configure.ac
index de75c094..d94fc7e2 100644
--- a/configure.ac
+++ b/configure.ac
@@ -378,6 +378,80 @@ AS_VAR_SET_IF([X86_SSE2_CFLAGS], [], [AS_VAR_SET([X86_SSE2_CFLAGS], "DEFAULT_X86
 AS_VAR_SET_IF([X86_SSE4_1_CFLAGS], [], [AS_VAR_SET([X86_SSE4_1_CFLAGS], "DEFAULT_X86_SSE4_1_CFLAGS")])
 AS_VAR_SET_IF([ARM_NEON_INTR_CFLAGS], [], [AS_VAR_SET([ARM_NEON_INTR_CFLAGS], ["$RESOLVED_DEFAULT_ARM_NEON_INTR_CFLAGS"])])
 
+AC_DEFUN([OPUS_PATH_NE10],
+   [
+      AC_ARG_WITH(NE10,
+                  AC_HELP_STRING([--with-NE10=PFX],[Prefix where libNE10 is installed (optional)]),
+                  NE10_prefix="$withval", NE10_prefix="")
+      AC_ARG_WITH(NE10-libraries,
+                  AC_HELP_STRING([--with-NE10-libraries=DIR],
+                        [Directory where libNE10 library is installed (optional)]),
+                  NE10_libraries="$withval", NE10_libraries="")
+      AC_ARG_WITH(NE10-includes,
+                  AC_HELP_STRING([--with-NE10-includes=DIR],
+                                 [Directory where libNE10 header files are installed (optional)]),
+                  NE10_includes="$withval", NE10_includes="")
+
+      if test "x$NE10_libraries" != "x" ; then
+         NE10_LIBS="-L$NE10_libraries"
+      elif test "x$NE10_prefix" = "xno" || test "x$NE10_prefix" = "xyes" ; then
+         NE10_LIBS=""
+      elif test "x$NE10_prefix" != "x" ; then
+         NE10_LIBS="-L$NE10_prefix/lib"
+      elif test "x$prefix" != "xNONE" ; then
+         NE10_LIBS="-L$prefix/lib"
+      fi
+
+      if test "x$NE10_prefix" != "xno" ; then
+         NE10_LIBS="$NE10_LIBS -lNE10"
+      fi
+
+      if test "x$NE10_includes" != "x" ; then
+         NE10_CFLAGS="-I$NE10_includes"
+      elif test "x$NE10_prefix" = "xno" || test "x$NE10_prefix" = "xyes" ; then
+         NE10_CFLAGS=""
+      elif test "x$ogg_prefix" != "x" ; then
+         NE10_CFLAGS="-I$NE10_prefix/include"
+      elif test "x$prefix" != "xNONE"; then
+         NE10_CFLAGS="-I$prefix/include"
+      fi
+
+      AC_MSG_CHECKING(for NE10)
+      save_CFLAGS="$CFLAGS"; CFLAGS="$NE10_CFLAGS"
+      save_LIBS="$LIBS"; LIBS="$NE10_LIBS $LIBM"
+      AC_LINK_IFELSE(
+         [
+            AC_LANG_PROGRAM(
+               [[#include <NE10_init.h>
+               ]],
+               [[
+                  ne10_fft_cfg_float32_t cfg;
+                  cfg = ne10_fft_alloc_c2c_float32_neon(480);
+               ]]
+            )
+         ],[
+            HAVE_ARM_NE10=1
+            AC_MSG_RESULT([yes])
+         ],[
+            HAVE_ARM_NE10=0
+            AC_MSG_RESULT([no])
+            NE10_CFLAGS=""
+            NE10_LIBS=""
+         ]
+      )
+      CFLAGS="$save_CFLAGS"; LIBS="$save_LIBS"
+      #Now we know if libNE10 is installed or not
+      AS_IF([test x"$HAVE_ARM_NE10" = x"1"],
+         [
+            AC_DEFINE([HAVE_ARM_NE10], 1, [NE10 library is installed on host. Make sure it is on target!])
+            AC_SUBST(HAVE_ARM_NE10)
+            AC_SUBST(NE10_CFLAGS)
+            AC_SUBST(NE10_LIBS)
+         ],[]
+      )
+   ]
+)
+
 AS_IF([test x"$enable_intrinsics" = x"yes"],[
    intrinsics_support=""
    AS_CASE([$host_cpu],
@@ -417,7 +491,16 @@ AS_IF([test x"$enable_intrinsics" = x"yes"],[
             AS_IF([test x"$OPUS_ARM_PRESUME_NEON_INTR" = x"1"],
                [AC_DEFINE([OPUS_ARM_PRESUME_NEON_INTR], 1, [Define if binary requires NEON intrinsics support])])
 
-			AS_IF([test x"$rtcd_support" = x""],
+            OPUS_PATH_NE10()
+            AS_IF([test x"$NE10_LIBS" != x""],
+            [
+                 intrinsics_support="$intrinsics_support (NE10)"
+                 AS_IF([test x"enable_rtcd" != x"" \
+                  && test x"$OPUS_ARM_PRESUME_NEON_INTR" != x"1"],
+                    [rtcd_support="$rtcd_support (NE10)"],[])
+            ])
+
+            AS_IF([test x"$rtcd_support" = x""],
                [rtcd_support=no])
 
             AS_IF([test x"$intrinsics_support" = x""],
@@ -588,6 +671,8 @@ AS_IF([test x"$enable_intrinsics" = x"yes"],[
 AM_CONDITIONAL([CPU_ARM], [test "$cpu_arm" = "yes"])
 AM_CONDITIONAL([OPUS_ARM_NEON_INTR],
     [test x"$OPUS_ARM_MAY_HAVE_NEON_INTR" = x"1"])
+AM_CONDITIONAL([HAVE_ARM_NE10],
+    [test x"$HAVE_ARM_NE10" = x"1"])
 AM_CONDITIONAL([HAVE_SSE],
     [test x"$OPUS_X86_MAY_HAVE_SSE" = x"1"])
 AM_CONDITIONAL([HAVE_SSE2],
-- 
cgit v1.2.3