VFP support

config.h.in

@@ -57,6 +57,9 @@
 /* Define to 1 if you have the <unistd.h> header file. */
 #undef HAVE_UNISTD_H
 
+/* Define to FFT with ARM VFP. */
+#undef HAVE_VFP
+
 /* Define to 1 if the system has the type `_Bool'. */
 #undef HAVE__BOOL
 

configure

@@ -635,6 +635,8 @@ ac_subst_vars='am__EXEEXT_FALSE
 am__EXEEXT_TRUE
 LTLIBOBJS
 LIBOBJS
+HAVE_VFP_FALSE
+HAVE_VFP_TRUE
 HAVE_NEON_FALSE
 HAVE_NEON_TRUE
 HAVE_SSE_FALSE
@@ -776,6 +778,7 @@ enable_dynamic_code
 enable_single
 enable_sse
 enable_neon
+enable_vfp
 '
       ac_precious_vars='build_alias
 host_alias
@@ -1423,6 +1426,7 @@ Optional Features:
   --enable-single         compile single-precision library
   --enable-sse            enable SSE extensions
   --enable-neon           enable NEON extensions
+  --enable-vfp            enable VFP extensions
 
 Optional Packages:
   --with-PACKAGE[=ARG]    use PACKAGE [ARG=yes]
@@ -15428,6 +15432,30 @@ else
 fi
 
 
+# Check whether --enable-vfp was given.
+if test "${enable_vfp+set}" = set; then :
+  enableval=$enable_vfp; have_vfp=$enableval
+else
+  have_vfp=no
+fi
+
+if test "$have_vfp" = "yes"; then
+	if test "$SIMD" != "sse"; then
+		as_fn_error $? "conflicting SIMD extensisons specified" "$LINENO" 5
+	fi
+
+$as_echo "#define HAVE_VFP 1" >>confdefs.h
+
+fi
+ if test "$have_vfp" = "yes"; then
+  HAVE_VFP_TRUE=
+  HAVE_VFP_FALSE='#'
+else
+  HAVE_VFP_TRUE='#'
+  HAVE_VFP_FALSE=
+fi
+
+
 #AC_CANONICAL_HOST
 { $as_echo "$as_me:${as_lineno-$LINENO}: host is \"${host}\"" >&5
 $as_echo "$as_me: host is \"${host}\"" >&6;}
@@ -16195,6 +16223,10 @@ if test -z "${HAVE_NEON_TRUE}" && test -z "${HAVE_NEON_FALSE}"; then
   as_fn_error $? "conditional \"HAVE_NEON\" was never defined.
 Usually this means the macro was only invoked conditionally." "$LINENO" 5
 fi
+if test -z "${HAVE_VFP_TRUE}" && test -z "${HAVE_VFP_FALSE}"; then
+  as_fn_error $? "conditional \"HAVE_VFP\" was never defined.
+Usually this means the macro was only invoked conditionally." "$LINENO" 5
+fi
 
 : "${CONFIG_STATUS=./config.status}"
 ac_write_fail=0

configure.ac

@@ -49,6 +49,15 @@ if test "$have_neon" = "yes"; then
 fi
 AM_CONDITIONAL(HAVE_NEON, test "$have_neon" = "yes")
 
+AC_ARG_ENABLE(vfp, [AC_HELP_STRING([--enable-vfp],[enable VFP extensions])], have_vfp=$enableval, have_vfp=no)
+if test "$have_vfp" = "yes"; then
+	if test "$SIMD" != "sse"; then
+		AC_MSG_ERROR([conflicting SIMD extensisons specified])
+	fi
+	AC_DEFINE(HAVE_VFP,1,[Define to FFT with ARM VFP.])
+fi
+AM_CONDITIONAL(HAVE_VFP, test "$have_vfp" = "yes")
+
 #AC_CANONICAL_HOST
 AC_MSG_NOTICE([host is "${host}"])
 case "${host}" in

src/Makefile.am

@@ -13,16 +13,21 @@ endif
 libffts_includedir=$(includedir)/ffts
 libffts_include_HEADERS = ../include/ffts.h
 
+
+if HAVE_VFP
+libffts_la_SOURCES += vfp.s 
+else
 if HAVE_NEON
 
 if DYNAMIC_DISABLED
 libffts_la_SOURCES += neon_static_f.s neon_static_i.s
 else
-libffts_la_SOURCES += neon.s vfp.s 
+libffts_la_SOURCES += neon.s
 endif
 
 else 
 if HAVE_SSE
 libffts_la_SOURCES += sse.s
 endif
 endif
+endif

src/Makefile.in

@@ -53,9 +53,10 @@ build_triplet = @build@
 host_triplet = @host@
 @DYNAMIC_DISABLED_TRUE@am__append_1 = ffts_static.c
 @DYNAMIC_DISABLED_FALSE@am__append_2 = codegen.c
-@DYNAMIC_DISABLED_TRUE@@HAVE_NEON_TRUE@am__append_3 = neon_static_f.s neon_static_i.s
-@DYNAMIC_DISABLED_FALSE@@HAVE_NEON_TRUE@am__append_4 = neon.s vfp.s 
-@HAVE_NEON_FALSE@@HAVE_SSE_TRUE@am__append_5 = sse.s
+@HAVE_VFP_TRUE@am__append_3 = vfp.s 
+@DYNAMIC_DISABLED_TRUE@@HAVE_NEON_TRUE@@HAVE_VFP_FALSE@am__append_4 = neon_static_f.s neon_static_i.s
+@DYNAMIC_DISABLED_FALSE@@HAVE_NEON_TRUE@@HAVE_VFP_FALSE@am__append_5 = neon.s
+@HAVE_NEON_FALSE@@HAVE_SSE_TRUE@@HAVE_VFP_FALSE@am__append_6 = sse.s
 subdir = src
 DIST_COMMON = $(libffts_include_HEADERS) $(srcdir)/Makefile.am \
 	$(srcdir)/Makefile.in $(top_srcdir)/depcomp
@@ -99,18 +100,20 @@ am__installdirs = "$(DESTDIR)$(libdir)" \
 LTLIBRARIES = $(lib_LTLIBRARIES)
 libffts_la_LIBADD =
 am__libffts_la_SOURCES_DIST = ffts.c ffts_nd.c ffts_real.c \
-	ffts_real_nd.c patterns.c ffts_static.c codegen.c \
-	neon_static_f.s neon_static_i.s neon.s vfp.s sse.s
+	ffts_real_nd.c patterns.c ffts_static.c codegen.c vfp.s \
+	neon_static_f.s neon_static_i.s neon.s sse.s
 @DYNAMIC_DISABLED_TRUE@am__objects_1 = ffts_static.lo
 @DYNAMIC_DISABLED_FALSE@am__objects_2 = codegen.lo
-@DYNAMIC_DISABLED_TRUE@@HAVE_NEON_TRUE@am__objects_3 =  \
-@DYNAMIC_DISABLED_TRUE@@HAVE_NEON_TRUE@	neon_static_f.lo \
-@DYNAMIC_DISABLED_TRUE@@HAVE_NEON_TRUE@	neon_static_i.lo
-@DYNAMIC_DISABLED_FALSE@@HAVE_NEON_TRUE@am__objects_4 = neon.lo vfp.lo
-@HAVE_NEON_FALSE@@HAVE_SSE_TRUE@am__objects_5 = sse.lo
+@HAVE_VFP_TRUE@am__objects_3 = vfp.lo
+@DYNAMIC_DISABLED_TRUE@@HAVE_NEON_TRUE@@HAVE_VFP_FALSE@am__objects_4 = neon_static_f.lo \
+@DYNAMIC_DISABLED_TRUE@@HAVE_NEON_TRUE@@HAVE_VFP_FALSE@	neon_static_i.lo
+@DYNAMIC_DISABLED_FALSE@@HAVE_NEON_TRUE@@HAVE_VFP_FALSE@am__objects_5 = neon.lo
+@HAVE_NEON_FALSE@@HAVE_SSE_TRUE@@HAVE_VFP_FALSE@am__objects_6 =  \
+@HAVE_NEON_FALSE@@HAVE_SSE_TRUE@@HAVE_VFP_FALSE@	sse.lo
 am_libffts_la_OBJECTS = ffts.lo ffts_nd.lo ffts_real.lo \
 	ffts_real_nd.lo patterns.lo $(am__objects_1) $(am__objects_2) \
-	$(am__objects_3) $(am__objects_4) $(am__objects_5)
+	$(am__objects_3) $(am__objects_4) $(am__objects_5) \
+	$(am__objects_6)
 libffts_la_OBJECTS = $(am_libffts_la_OBJECTS)
 DEFAULT_INCLUDES = -I.@am__isrc@ -I$(top_builddir)
 depcomp = $(SHELL) $(top_srcdir)/depcomp
@@ -263,7 +266,7 @@ top_srcdir = @top_srcdir@
 lib_LTLIBRARIES = libffts.la
 libffts_la_SOURCES = ffts.c ffts_nd.c ffts_real.c ffts_real_nd.c \
 	patterns.c $(am__append_1) $(am__append_2) $(am__append_3) \
-	$(am__append_4) $(am__append_5)
+	$(am__append_4) $(am__append_5) $(am__append_6)
 libffts_includedir = $(includedir)/ffts
 libffts_include_HEADERS = ../include/ffts.h
 all: all-am

src/codegen.c

@@ -42,10 +42,11 @@
 #include <sys/types.h>
 #include <sys/mman.h>
 
-#ifdef __ARM_NEON__
-	#include "codegen_neon.h"
-//	#include "neon_float.h"
+#ifdef HAVE_NEON
+	#include "codegen_arm.h"
 	#include "neon.h"
+#elif HAVE_VFP
+	#include "codegen_arm.h"
 	#include "vfp.h"
 #else
 	#include "codegen_sse.h"
@@ -99,15 +100,15 @@ uint32_t LUT_offset(size_t N, size_t leafN) {
 		for(i=0;i<n_luts-1;i++) {
 			p_lut_size = lut_size;
 			if(!i || hardcoded) {
-			#ifdef __ARM_NEON__
+			#ifdef __arm__
 				if(N <= 32) lut_size += n/4 * 2 * sizeof(cdata_t);
 			else lut_size += n/4 * sizeof(cdata_t);
 			#else
 				lut_size += n/4 * 2 * sizeof(cdata_t);
 			#endif
 			//	n *= 2;
 			} else {
-			#ifdef __ARM_NEON__
+			#ifdef __arm__ 
 				lut_size += n/8 * 3 * sizeof(cdata_t);
 			#else
 				lut_size += n/8 * 3 * 2 * sizeof(cdata_t);
@@ -118,7 +119,7 @@ uint32_t LUT_offset(size_t N, size_t leafN) {
 	return lut_size;
 }
 
-#ifdef __ARM_NEON__
+#ifdef __arm__ 
 	typedef uint32_t insns_t;
 #else
 	typedef uint8_t insns_t;
@@ -147,7 +148,7 @@ void insert_nops(uint8_t **p, uint32_t count) {
 
 
 void align_mem16(uint8_t **p, uint32_t offset) {
-#ifdef __ARM_NEON__
+#ifdef __x86_64__
 	int r = (16 - (offset & 0xf)) - ((uint32_t)(*p) & 0xf);
 	r = (16 + r) & 0xf;
 	insert_nops(p, r);	
@@ -170,7 +171,7 @@ void ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN, int sign) {
 
 	pps = ps;
 
-#ifdef __ARM_NEON__
+#ifdef __arm__ 
 	if(N < 8192) p->transform_size = 8192;
 	else p->transform_size = N;
 #else
@@ -203,7 +204,7 @@ void ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN, int sign) {
 
 	insns_t *x_8_addr = fp;
 #ifdef __arm__
-#ifdef __ARM_NEON__
+#ifdef HAVE_NEON
 	memcpy(fp, neon_x8, neon_x8_t - neon_x8);
 	if(sign < 0) {
 		fp[31] ^= 0x00200000; fp[32] ^= 0x00200000; fp[33] ^= 0x00200000; fp[34] ^= 0x00200000;
@@ -229,7 +230,7 @@ void ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN, int sign) {
 	insns_t *x_4_addr = fp;
 #ifdef __arm__
 
-#ifdef __ARM_NEON__
+#ifdef HAVE_NEON
 	memcpy(fp, neon_x4, neon_x8 - neon_x4);
 	if(sign < 0) {
 		fp[26] ^= 0x00200000; fp[28] ^= 0x00200000; fp[31] ^= 0x00200000; fp[32] ^= 0x00200000;
@@ -248,7 +249,7 @@ void ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN, int sign) {
 #endif
 	insns_t *start = fp;
 
-#ifdef __ARM_NEON__
+#ifdef __arm__ 
 	*fp = PUSH_LR(); fp++;
 	*fp = 0xed2d8b10; fp++;
 
@@ -271,7 +272,7 @@ void ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN, int sign) {
 
 #ifdef __arm__
 		*fp = LDRI(2, 1, ((uint32_t)&p->ee_ws) - ((uint32_t)p)); fp++; 
-  	#ifdef __ARM_NEON__
+  	#ifdef HAVE_NEON
   		MOVI(&fp, 11, p->i0);
   	#else 
   		MOVI(&fp, 11, p->i0);
@@ -291,7 +292,7 @@ void ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN, int sign) {
 #endif
 	//fp++;
 #ifdef __arm__
-#ifdef __ARM_NEON__
+#ifdef HAVE_NEON
 	memcpy(fp, neon_ee, neon_oo - neon_ee);
 	if(sign < 0) {
 		fp[33] ^= 0x00200000; fp[37] ^= 0x00200000; fp[38] ^= 0x00200000; fp[39] ^= 0x00200000;
@@ -425,14 +426,14 @@ void ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN, int sign) {
 
 
   	if(pps[0] == 2*leafN) {
-   //   CALL(&fp, x_4_addr);
+      CALL(&fp, x_4_addr);
 //  	}else if(!pps[2]){
 //	  //uint32_t *x_8_t_addr = fp;
 //		memcpy(fp, neon_x8_t, neon_ee - neon_x8_t);
 //		fp += (neon_ee - neon_x8_t) / 4;
 //		//*fp++ = BL(fp+2, x_8_t_addr);
   	}else{
-    //		CALL(&fp, x_8_addr);
+    		CALL(&fp, x_8_addr);
   	}
 
 		pAddr = pps[1] * 4;
@@ -445,7 +446,7 @@ void ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN, int sign) {
 	}
 #endif
 #ifdef __arm__
-#ifdef __ARM_NEON__
+#ifdef HAVE_NEON
 	if(__builtin_ctzl(N) & 1){
 		ADDI(&fp, 2, 7, 0);
 		ADDI(&fp, 7, 9, 0);
@@ -612,7 +613,7 @@ void ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN, int sign) {
       *fp = BL(fp+2, x_4_addr); fp++;
   	}else if(!pps[2]){
   	  //uint32_t *x_8_t_addr = fp;
-#ifdef __ARM_NEON__
+#ifdef HAVE_NEON
   		memcpy(fp, neon_x8_t, neon_ee - neon_x8_t);
   		if(sign < 0) {
   			fp[31] ^= 0x00200000; fp[32] ^= 0x00200000; fp[33] ^= 0x00200000; fp[34] ^= 0x00200000;

src/codegen_neon.h → src/codegen_arm.h

@@ -31,8 +31,8 @@
 
 */
 
-#ifndef __CODEGEN_NEON_H__
-#define __CODEGEN_NEON_H__
+#ifndef __CODEGEN_ARM_H__
+#define __CODEGEN_ARM_H__
 
 
 

src/ffts.c

@@ -105,7 +105,7 @@ ffts_plan_t *ffts_init_1d(size_t N, int sign) {
 	if(N >= 32) {
 		ffts_init_offsets(p, N, leafN);
 #ifdef __arm__
-#ifdef __ARM_NEON__
+#ifdef HAVE_NEON
 		ffts_init_is(p, N, leafN, 1);
 #else
 		ffts_init_is(p, N, leafN, 1);
@@ -120,7 +120,7 @@ ffts_plan_t *ffts_init_1d(size_t N, int sign) {
 		p->i2 = N/leafN/3;
 
 	#ifdef __arm__	
-	#ifdef __ARM_NEON__
+	#ifdef HAVE_NEON
 	p->i0/=2;
 	p->i1/=2;
 	#endif
@@ -164,15 +164,15 @@ ffts_plan_t *ffts_init_1d(size_t N, int sign) {
 
 		for(i=0;i<n_luts;i++) {
 			if(!i || hardcoded) {
-			#ifdef __ARM_NEON__
+			#ifdef HAVE_NEON
 				if(N <= 32) lut_size += n/4 * 2 * sizeof(cdata_t);
 				else lut_size += n/4 * sizeof(cdata_t);
 			#else
 				lut_size += n/4 * 2 * sizeof(cdata_t);
 			#endif
 				n *= 2;
 			} else {
-			#ifdef __ARM_NEON__
+			#ifdef HAVE_NEON
 				lut_size += n/8 * 3 * sizeof(cdata_t);
 			#else
 				lut_size += n/8 * 3 * 2 * sizeof(cdata_t);
@@ -196,7 +196,7 @@ ffts_plan_t *ffts_init_1d(size_t N, int sign) {
 		n = leafN*2;
 		if(hardcoded) n = 8;
 
-		#ifdef __ARM_NEON__
+		#ifdef HAVE_NEON
 			V neg = (sign < 0) ? VLIT4(0.0f, 0.0f, 0.0f, 0.0f) : VLIT4(-0.0f, -0.0f, -0.0f, -0.0f);
 		#endif
 
@@ -234,7 +234,7 @@ ffts_plan_t *ffts_init_1d(size_t N, int sign) {
 					//w = FFTS_MALLOC(n/4 * sizeof(cdata_t), 32);
 					float *fw = (float *)w;
 					VS temp0, temp1, temp2;
-  				#ifdef __ARM_NEON__
+  				#ifdef HAVE_NEON
 					for(j=0;j<n/4;j+=4) {
   					temp0 = VLD2(fw0 + j*2);
   					temp0.val[1] = VXOR(temp0.val[1], neg);
@@ -289,7 +289,7 @@ ffts_plan_t *ffts_init_1d(size_t N, int sign) {
 				//w = FFTS_MALLOC(n/8 * 3 * sizeof(cdata_t), 32);
 				float *fw = (float *)w;
 				VS temp0, temp1, temp2;
-				#ifdef __ARM_NEON__	
+				#ifdef HAVE_NEON	
   			for(j=0;j<n/8;j+=4) {
   				temp0 = VLD2(fw0 + j*2);
   				temp0.val[1] = VXOR(temp0.val[1], neg);