divide SIMD codes for compiling them with differnt compiler options

julius-speech · Sep 7, 2016 · 1df3fde · 1df3fde
1 parent 7d31ba5
commit 1df3fde
Show file tree

Hide file tree

Showing 11 changed files with 363 additions and 242 deletions.
diff --git a/libsent/Makefile.in b/libsent/Makefile.in
@@ -109,6 +109,10 @@ src/phmm/addlog.o \
 src/phmm/mkwhmm.o \
 src/phmm/vsegment.o \
 src/phmm/calc_dnn.o \
+src/phmm/calc_dnn_fma.o \
+src/phmm/calc_dnn_avx.o \
+src/phmm/calc_dnn_sse.o \
+src/phmm/calc_dnn_neon.o \
 src/util/endian.o \
 src/util/jlog.o \
 src/util/mymalloc.o \
@@ -142,8 +146,17 @@ $(TARGET): $(OBJ)
 	$(AR) $@ $?
 	$(RANLIB) $@
 
-src/phmm/calc_dnn.o: src/phmm/calc_dnn.c
-	$(CC) $(CFLAGS) $(CPPFLAGS) @SIMD_CFLAGS@ -o $@ -c $<
+src/phmm/calc_dnn_fma.o: src/phmm/calc_dnn_fma.c
+	$(CC) $(CFLAGS) $(CPPFLAGS) @SIMD_FMA_CFLAGS@ -o $@ -c $<
+
+src/phmm/calc_dnn_avx.o: src/phmm/calc_dnn_avx.c
+	$(CC) $(CFLAGS) $(CPPFLAGS) @SIMD_AVX_CFLAGS@ -o $@ -c $<
+
+src/phmm/calc_dnn_sse.o: src/phmm/calc_dnn_sse.c
+	$(CC) $(CFLAGS) $(CPPFLAGS) @SIMD_SSE_CFLAGS@ -o $@ -c $<
+
+src/phmm/calc_dnn_neon.o: src/phmm/calc_dnn_neon.c
+	$(CC) $(CFLAGS) $(CPPFLAGS) @SIMD_NEON_CFLAGS@ -o $@ -c $<
 
 
 ############################################################

diff --git a/libsent/configure b/libsent/configure
@@ -646,7 +646,10 @@ RM
 INSTALL_DATA
 INSTALL_SCRIPT
 INSTALL_PROGRAM
-SIMD_CFLAGS
+SIMD_NEON_CFLAGS
+SIMD_SSE_CFLAGS
+SIMD_AVX_CFLAGS
+SIMD_FMA_CFLAGS
 CPP
 OBJEXT
 EXEEXT
@@ -3310,6 +3313,7 @@ _ACEOF
 if ac_fn_c_try_compile "$LINENO"; then :
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
 $as_echo "yes" >&6; }
+  SIMD_FMA_CFLAGS=""
   xxxxFMA=yes
 else
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
@@ -3337,7 +3341,7 @@ _ACEOF
 if ac_fn_c_try_compile "$LINENO"; then :
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
 $as_echo "yes" >&6; }
-    SIMD_CFLAGS="-mfma"
+    SIMD_FMA_CFLAGS="-mfma"
     xxxxFMA=yes
 else
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
@@ -3346,14 +3350,15 @@ fi
 rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
   CFLAGS="$xxxCFLAGS"
 fi
-
 if test "$xxxxFMA" = yes; then
-  xxxxAVX=yes
-else
-  xxxxAVX=no
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking for SIMD AVX instruction" >&5
+  $as_echo "#define HAS_SIMD_FMA 1" >>confdefs.h
+
+fi
+
+xxxxAVX=no
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for SIMD AVX instruction" >&5
 $as_echo_n "checking for SIMD AVX instruction... " >&6; }
-  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
 /* end confdefs.h.  */
 #include <immintrin.h>
 
@@ -3368,19 +3373,20 @@ _ACEOF
 if ac_fn_c_try_compile "$LINENO"; then :
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
 $as_echo "yes" >&6; }
-    xxxxAVX=yes
+  SIMD_AVX_CFLAGS=""
+  xxxxAVX=yes
 else
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
 $as_echo "no" >&6; }
 
 fi
 rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
-  if test "$xxxxAVX" = no; then
-        { $as_echo "$as_me:${as_lineno-$LINENO}: checking for SIMD AVX instruction with -mavx" >&5
+if test "$xxxxAVX" = no; then
+    { $as_echo "$as_me:${as_lineno-$LINENO}: checking for SIMD AVX instruction with -mavx" >&5
 $as_echo_n "checking for SIMD AVX instruction with -mavx... " >&6; }
-    xxxCFLAGS="$CFLAGS"
-    CFLAGS="$CFLAGS -mavx"
-    cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+  xxxCFLAGS="$CFLAGS"
+  CFLAGS="$CFLAGS -mavx"
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
 /* end confdefs.h.  */
 #include <immintrin.h>
 
@@ -3395,24 +3401,24 @@ _ACEOF
 if ac_fn_c_try_compile "$LINENO"; then :
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
 $as_echo "yes" >&6; }
-      SIMD_CFLAGS="-mavx"
-      xxxxAVX=yes
+    SIMD_AVX_CFLAGS="-mavx"
+    xxxxAVX=yes
 else
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
 $as_echo "no" >&6; }
 fi
 rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
-    CFLAGS="$xxxCFLAGS"
-  fi
+  CFLAGS="$xxxCFLAGS"
 fi
-
 if test "$xxxxAVX" = yes; then
-  xxxxSSE=yes
-else
-  xxxxSSE=no
-  { $as_echo "$as_me:${as_lineno-$LINENO}: checking for SIMD SSE instruction" >&5
+  $as_echo "#define HAS_SIMD_AVX 1" >>confdefs.h
+
+fi
+
+xxxxSSE=no
+{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for SIMD SSE instruction" >&5
 $as_echo_n "checking for SIMD SSE instruction... " >&6; }
-  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
 /* end confdefs.h.  */
 #include <immintrin.h>
 
@@ -3427,19 +3433,20 @@ _ACEOF
 if ac_fn_c_try_compile "$LINENO"; then :
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
 $as_echo "yes" >&6; }
-    xxxxSSE=yes
+  SIMD_SSE_CFLAGS=""
+  xxxxSSE=yes
 else
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
 $as_echo "no" >&6; }
 
 fi
 rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
-  if test "$xxxxSSE" = no; then
-        { $as_echo "$as_me:${as_lineno-$LINENO}: checking for SIMD SSE instruction with -msse" >&5
+if test "$xxxxSSE" = no; then
+    { $as_echo "$as_me:${as_lineno-$LINENO}: checking for SIMD SSE instruction with -msse" >&5
 $as_echo_n "checking for SIMD SSE instruction with -msse... " >&6; }
-    xxxCFLAGS="$CFLAGS"
-    CFLAGS="$CFLAGS -msse"
-    cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+  xxxCFLAGS="$CFLAGS"
+  CFLAGS="$CFLAGS -msse"
+  cat confdefs.h - <<_ACEOF >conftest.$ac_ext
 /* end confdefs.h.  */
 #include <immintrin.h>
 
@@ -3454,19 +3461,22 @@ _ACEOF
 if ac_fn_c_try_compile "$LINENO"; then :
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
 $as_echo "yes" >&6; }
-      SIMD_CFLAGS="-msse"
-      xxxxSSE=yes
+    SIMD_SSE_CFLAGS="-msse"
+    xxxxSSE=yes
 else
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
 $as_echo "no" >&6; }
 fi
 rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
-    CFLAGS="$xxxCFLAGS"
-  fi
+  CFLAGS="$xxxCFLAGS"
+fi
+if test "$xxxxSSE" = yes; then
+  $as_echo "#define HAS_SIMD_SSE 1" >>confdefs.h
+
 fi
 
 if test "$use_neon" = yes; then
-  xxxxNEON=yes
+  xxxxNEON=no
   { $as_echo "$as_me:${as_lineno-$LINENO}: checking for NEON SIMD instruction for ARM" >&5
 $as_echo_n "checking for NEON SIMD instruction for ARM... " >&6; }
   cat confdefs.h - <<_ACEOF >conftest.$ac_ext
@@ -3484,10 +3494,11 @@ _ACEOF
 if ac_fn_c_try_compile "$LINENO"; then :
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
 $as_echo "yes" >&6; }
+    SIMD_NEON_CFLAGS=""
+    xxxxNEON=yes
 else
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
 $as_echo "no" >&6; }
-    xxxxNEON=no
 
 fi
 rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
@@ -3511,22 +3522,26 @@ _ACEOF
 if ac_fn_c_try_compile "$LINENO"; then :
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: yes" >&5
 $as_echo "yes" >&6; }
-      SIMD_CFLAGS="-mfpu=neon"
+      SIMD_NEON_CFLAGS="-mfpu=neon"
       xxxxNEON=yes
 else
   { $as_echo "$as_me:${as_lineno-$LINENO}: result: no" >&5
 $as_echo "no" >&6; }
+
 fi
 rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext
     CFLAGS="$xxxCFLAGS"
   fi
   if test "$xxxxNEON" = yes; then
-    $as_echo "#define USE_ARM_NEON 1" >>confdefs.h
+    $as_echo "#define HAS_SIMD_NEON 1" >>confdefs.h
 
   fi
 fi
 
 
+
+
+
 # Find a good install program.  We prefer a C program (faster),
 # so one script is as good as another.  But avoid the broken or
 # incompatible versions:

diff --git a/libsent/configure.in b/libsent/configure.in
@@ -139,6 +139,7 @@ AC_MSG_CHECKING([for SIMD FMA instruction])
 AC_TRY_COMPILE([#include <immintrin.h>
   ],[__m256 v1, v2; __m256 x = _mm256_fmadd_ps(v1, v2, x);],
   AC_MSG_RESULT([yes])
+  SIMD_FMA_CFLAGS=""
   xxxxFMA=yes,
   AC_MSG_RESULT([no])
 )
@@ -150,72 +151,76 @@ if test "$xxxxFMA" = no; then
   AC_TRY_COMPILE([#include <immintrin.h>
     ],[__m256 v1, v2; __m256 x = _mm256_fmadd_ps(v1, v2, x);],
     AC_MSG_RESULT([yes])
-    SIMD_CFLAGS="-mfma"
+    SIMD_FMA_CFLAGS="-mfma"
     xxxxFMA=yes,
     AC_MSG_RESULT([no]))
   CFLAGS="$xxxCFLAGS"
 fi
-
 if test "$xxxxFMA" = yes; then
-  xxxxAVX=yes
-else
-  xxxxAVX=no
-  AC_MSG_CHECKING([for SIMD AVX instruction])
+  AC_DEFINE(HAS_SIMD_FMA)
+fi
+
+xxxxAVX=no
+AC_MSG_CHECKING([for SIMD AVX instruction])
+AC_TRY_COMPILE([#include <immintrin.h>
+  ],[__m256 v = _mm256_setzero_ps();],
+  AC_MSG_RESULT([yes])
+  SIMD_AVX_CFLAGS=""
+  xxxxAVX=yes,
+  AC_MSG_RESULT([no])
+)
+if test "$xxxxAVX" = no; then
+  dnl retry with "-mavx" option
+  AC_MSG_CHECKING([for SIMD AVX instruction with -mavx])
+  xxxCFLAGS="$CFLAGS"
+  CFLAGS="$CFLAGS -mavx"
   AC_TRY_COMPILE([#include <immintrin.h>
     ],[__m256 v = _mm256_setzero_ps();],
     AC_MSG_RESULT([yes])
+    SIMD_AVX_CFLAGS="-mavx"
     xxxxAVX=yes,
-    AC_MSG_RESULT([no])
-  )
-  if test "$xxxxAVX" = no; then
-    dnl retry with "-mavx" option
-    AC_MSG_CHECKING([for SIMD AVX instruction with -mavx])
-    xxxCFLAGS="$CFLAGS"
-    CFLAGS="$CFLAGS -mavx"
-    AC_TRY_COMPILE([#include <immintrin.h>
-      ],[__m256 v = _mm256_setzero_ps();],
-      AC_MSG_RESULT([yes])
-      SIMD_CFLAGS="-mavx"
-      xxxxAVX=yes,
-      AC_MSG_RESULT([no]))
-    CFLAGS="$xxxCFLAGS"
-  fi
+    AC_MSG_RESULT([no]))
+  CFLAGS="$xxxCFLAGS"
 fi
-
 if test "$xxxxAVX" = yes; then
-  xxxxSSE=yes
-else
-  xxxxSSE=no
-  AC_MSG_CHECKING([for SIMD SSE instruction])
+  AC_DEFINE(HAS_SIMD_AVX)
+fi
+
+xxxxSSE=no
+AC_MSG_CHECKING([for SIMD SSE instruction])
+AC_TRY_COMPILE([#include <immintrin.h>
+  ],[__m128 x = _mm_setzero_ps();],
+  AC_MSG_RESULT([yes])
+  SIMD_SSE_CFLAGS=""
+  xxxxSSE=yes,
+  AC_MSG_RESULT([no])
+)
+if test "$xxxxSSE" = no; then
+  dnl retry with "-msse" option
+  AC_MSG_CHECKING([for SIMD SSE instruction with -msse])
+  xxxCFLAGS="$CFLAGS"
+  CFLAGS="$CFLAGS -msse"
   AC_TRY_COMPILE([#include <immintrin.h>
     ],[__m128 x = _mm_setzero_ps();],
     AC_MSG_RESULT([yes])
+    SIMD_SSE_CFLAGS="-msse"
     xxxxSSE=yes,
-    AC_MSG_RESULT([no])
-  )
-  if test "$xxxxSSE" = no; then
-    dnl retry with "-msse" option
-    AC_MSG_CHECKING([for SIMD SSE instruction with -msse])
-    xxxCFLAGS="$CFLAGS"
-    CFLAGS="$CFLAGS -msse"
-    AC_TRY_COMPILE([#include <immintrin.h>
-      ],[__m128 x = _mm_setzero_ps();],
-      AC_MSG_RESULT([yes])
-      SIMD_CFLAGS="-msse"
-      xxxxSSE=yes,
-      AC_MSG_RESULT([no]))
-    CFLAGS="$xxxCFLAGS"
-  fi
+    AC_MSG_RESULT([no]))
+  CFLAGS="$xxxCFLAGS"
+fi
+if test "$xxxxSSE" = yes; then
+  AC_DEFINE(HAS_SIMD_SSE)
 fi
 
 if test "$use_neon" = yes; then
-  xxxxNEON=yes
+  xxxxNEON=no
   AC_MSG_CHECKING([for NEON SIMD instruction for ARM])
   AC_TRY_COMPILE([#include <arm_neon.h>
     ],[float32x4_t x = vdupq_n_f32(0);],
-    AC_MSG_RESULT([yes]),
+    AC_MSG_RESULT([yes])
+    SIMD_NEON_CFLAGS=""
+    xxxxNEON=yes,
     AC_MSG_RESULT([no])
-    xxxxNEON=no
   )
   if test "$xxxxNEON" = no; then
     dnl retry with "-mfpu=neon" option
@@ -225,16 +230,20 @@ if test "$use_neon" = yes; then
     AC_TRY_COMPILE([#include <arm_neon.h>
       ],[float32x4_t x = vdupq_n_f32(0);],
       AC_MSG_RESULT([yes])
-      SIMD_CFLAGS="-mfpu=neon"
+      SIMD_NEON_CFLAGS="-mfpu=neon"
       xxxxNEON=yes,
-      AC_MSG_RESULT([no]))
+      AC_MSG_RESULT([no])
+    )
     CFLAGS="$xxxCFLAGS"
   fi
   if test "$xxxxNEON" = yes; then
-    AC_DEFINE(USE_ARM_NEON)
+    AC_DEFINE(HAS_SIMD_NEON)
   fi
 fi
-AC_SUBST(SIMD_CFLAGS)
+AC_SUBST(SIMD_FMA_CFLAGS)
+AC_SUBST(SIMD_AVX_CFLAGS)
+AC_SUBST(SIMD_SSE_CFLAGS)
+AC_SUBST(SIMD_NEON_CFLAGS)
 
 dnl Checks for programs.
 AC_PROG_INSTALL