From e19b546b40a3a954f8dcf9c9e1dbb84b3f381deb Mon Sep 17 00:00:00 2001
From: Ben Avison <bavison@riscosopen.org>
Date: Wed, 10 Oct 2018 19:30:32 +0100
Subject: [PATCH 08/11] ARM: Create configure option --enable-arm-neon to
 govern assembly optimizations

---
 configure.in              | 39 +++++++++++++++++++++++++++++++++++++++
 include/SDL_config.h.in   |  1 +
 include/SDL_cpuinfo.h     |  3 +++
 src/cpuinfo/SDL_cpuinfo.c | 37 +++++++++++++++++++++++++++++++++++++
 4 files changed, 80 insertions(+)

diff --git a/configure.in b/configure.in
index 76a63108d..ff374c86a 100644
--- a/configure.in
+++ b/configure.in
@@ -922,6 +922,44 @@ dnl            SOURCES="$SOURCES $srcdir/src/video/arm/pixman-arm-simd*.c"
     fi
 }
 
+dnl Check for ARM NEON instruction support using gas syntax
+CheckNEON()
+{
+    AC_ARG_ENABLE(arm-neon,
+AC_HELP_STRING([--enable-arm-neon], [use NEON assembly blitters on ARM [[default=yes]]]),
+                  enable_arm_neon=$enableval, enable_arm_neon=yes)
+    if test x$enable_video = xyes -a x$enable_assembly = xyes -a x$enable_arm_neon = xyes; then
+        save_CFLAGS="$CFLAGS"
+        have_arm_neon=no
+        CFLAGS="-x assembler-with-cpp $CFLAGS"
+        
+        AC_MSG_CHECKING(for ARM NEON)
+        AC_COMPILE_IFELSE([AC_LANG_SOURCE([[
+        .text
+        .fpu neon
+        .arch armv7a
+        .object_arch armv4
+        .eabi_attribute 10, 0
+        .arm
+        .altmacro
+        #ifndef __ARM_EABI__
+        #error EABI is required (to be sure that calling conventions are compatible)
+        #endif
+        pld [r0]
+        vmovn.u16 d0, q0
+        ]])], have_arm_neon=yes)
+        AC_MSG_RESULT($have_arm_neon)
+        
+        CFLAGS="$save_CFLAGS"
+        
+        if test x$have_arm_neon = xyes; then
+            AC_DEFINE(SDL_ARM_NEON_BLITTERS)
+dnl            SOURCES="$SOURCES $srcdir/src/video/arm/pixman-arm-neon*.c"
+            SOURCES="$SOURCES $srcdir/src/video/arm/pixman-arm-neon*.S"
+        fi
+    fi
+}
+
 dnl See if GCC's -fvisibility=hidden is supported (gcc4 and later, usually).
 dnl  Details of this flag are here: http://gcc.gnu.org/wiki/Visibility
 CheckVisibilityHidden()
@@ -2466,6 +2504,7 @@ case "$host" in
         CheckNASM
         CheckAltivec
         CheckARM
+        CheckNEON
         CheckOSS
         CheckDMEDIA
         CheckMME
diff --git a/include/SDL_config.h.in b/include/SDL_config.h.in
index b319aaae9..9c5f45f8c 100644
--- a/include/SDL_config.h.in
+++ b/include/SDL_config.h.in
@@ -313,5 +313,6 @@
 #undef SDL_HERMES_BLITTERS
 #undef SDL_ALTIVEC_BLITTERS
 #undef SDL_ARM_SIMD_BLITTERS
+#undef SDL_ARM_NEON_BLITTERS
 
 #endif /* _SDL_config_h */
diff --git a/include/SDL_cpuinfo.h b/include/SDL_cpuinfo.h
index 9698e910d..1335b9823 100644
--- a/include/SDL_cpuinfo.h
+++ b/include/SDL_cpuinfo.h
@@ -63,6 +63,9 @@ extern DECLSPEC SDL_bool SDLCALL SDL_HasAltiVec(void);
 /** This function returns true if the CPU has ARM SIMD (ARMv6) features */
 extern DECLSPEC SDL_bool SDLCALL SDL_HasARMSIMD(void);
 
+/** This function returns true if the CPU has ARM NEON features */
+extern DECLSPEC SDL_bool SDLCALL SDL_HasARMNEON(void);
+
 /* Ends C function definitions when using C++ */
 #ifdef __cplusplus
 }
diff --git a/src/cpuinfo/SDL_cpuinfo.c b/src/cpuinfo/SDL_cpuinfo.c
index b6d39a739..8c1027249 100644
--- a/src/cpuinfo/SDL_cpuinfo.c
+++ b/src/cpuinfo/SDL_cpuinfo.c
@@ -46,6 +46,7 @@
 #define CPU_HAS_SSE2	0x00000080
 #define CPU_HAS_ALTIVEC	0x00000100
 #define CPU_HAS_ARM_SIMD 0x00000200
+#define CPU_HAS_ARM_NEON 0x00000400
 
 #if SDL_ALTIVEC_BLITTERS && HAVE_SETJMP && !__MACOSX__ && !__OpenBSD__
 /* This is the brute force way of detecting instruction sets...
@@ -422,6 +423,25 @@ static __inline__ int CPU_haveARMSIMD(void)
 	return arm_simd;
 }
 
+static __inline__ int CPU_haveARMNEON(void)
+{
+	int arm_neon = 0;
+	int fd;
+
+	fd = open("/proc/self/auxv", O_RDONLY);
+	if (fd >= 0)
+	{
+		Elf32_auxv_t aux;
+		while (read(fd, &aux, sizeof aux) == sizeof aux)
+		{
+			if (aux.a_type == AT_HWCAP)
+				arm_neon = (aux.a_un.a_val & 4096) != 0;
+		}
+		close(fd);
+	}
+	return arm_neon;
+}
+
 #else
 
 static __inline__ int CPU_haveARMSIMD(void)
@@ -429,6 +449,11 @@ static __inline__ int CPU_haveARMSIMD(void)
 	return 0;
 }
 
+static __inline__ int CPU_haveARMNEON(void)
+{
+	return 0;
+}
+
 #endif
 
 static Uint32 SDL_CPUFeatures = 0xFFFFFFFF;
@@ -464,6 +489,9 @@ static Uint32 SDL_GetCPUFeatures(void)
 		if ( CPU_haveARMSIMD() ) {
 			SDL_CPUFeatures |= CPU_HAS_ARM_SIMD;
 		}
+		if ( CPU_haveARMNEON() ) {
+			SDL_CPUFeatures |= CPU_HAS_ARM_NEON;
+		}
 	}
 	return SDL_CPUFeatures;
 }
@@ -540,6 +568,14 @@ SDL_bool SDL_HasARMSIMD(void)
 	return SDL_FALSE;
 }
 
+SDL_bool SDL_HasARMNEON(void)
+{
+	if ( SDL_GetCPUFeatures() & CPU_HAS_ARM_NEON ) {
+		return SDL_TRUE;
+	}
+	return SDL_FALSE;
+}
+
 #ifdef TEST_MAIN
 
 #include <stdio.h>
@@ -555,6 +591,7 @@ int main()
 	printf("SSE2: %d\n", SDL_HasSSE2());
 	printf("AltiVec: %d\n", SDL_HasAltiVec());
 	printf("ARM SIMD: %d\n", SDL_HasARMSIMD());
+	printf("ARM NEON: %d\n", SDL_HasARMNEON());
 	return 0;
 }
 
-- 
2.11.0