arm: add support for detecting SIMD (NEON) availability (WIP)

Requires getauxval() which is available at least in Linux/Android
with recent versions of the libc and that is therefore behind a
configure like macro.

A similar function from FreeBSD>=12, Windows and NetBSD is used
for each case.

While at it, consolidate the code to use the same externally
visible flag that is used in x86 for SSE2.

API_CHANGES

@@ -1,5 +1,8 @@
 This file is the short summary of the API changes:
 
+10.12.2022 - Non-backward compatible
+    SLJIT_HAS_SSE2 renamed to SLJIT_HAS_SIMD
+
 10.11.2022 - Non-backward compatible
     Extract the pre/post update operations from
     sljit_emit_mem to sljit_emit_mem_update

sljit_src/sljitConfigInternal.h

@@ -580,10 +580,12 @@ typedef double sljit_f64;
 
 #endif /* !SLJIT_FPU_UNALIGNED */
 
-#if (defined SLJIT_CONFIG_X86_32 && SLJIT_CONFIG_X86_32)
-/* Auto detect SSE2 support using CPUID.
-   On 64 bit x86 cpus, sse2 must be present. */
-#define SLJIT_DETECT_SSE2 1
+#if (defined SLJIT_CONFIG_X86_32 && SLJIT_CONFIG_X86_32) || \
+	(defined SLJIT_CONFIG_ARM_V7 && SLJIT_CONFIG_ARM_V7)
+/* Auto detect availability of SSE2 (using CPUID) or NEON.
+   Mandatory on 64 bit x86 or ARM (aarch64) cpus.
+   Optionally available for 32 bit since pentium or armv7. */
+#define SLJIT_DETECT_SIMD 1
 #endif
 
 /*****************************************************************************************/

sljit_src/sljitLir.h

@@ -632,9 +632,11 @@ static SLJIT_INLINE sljit_uw sljit_get_generated_code_size(struct sljit_compiler
 /* [Emulated] Prefetch instruction is available (emulated as a nop). */
 #define SLJIT_HAS_PREFETCH		7
 
-#if (defined SLJIT_CONFIG_X86 && SLJIT_CONFIG_X86)
+#if (defined SLJIT_CONFIG_X86 && SLJIT_CONFIG_X86) \
+	|| (defined SLJIT_CONFIG_ARM && SLJIT_CONFIG_ARM)
 /* [Not emulated] SSE2 support is available on x86. */
-#define SLJIT_HAS_SSE2			100
+/* [Not emulated] NEON support is available on ARM. */
+#define SLJIT_HAS_SIMD			100
 #endif
 
 SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_has_cpu_feature(sljit_s32 feature_type);

sljit_src/sljitNativeARM_32.c

@@ -41,6 +41,69 @@ SLJIT_API_FUNC_ATTRIBUTE const char* sljit_get_platform_name(void)
 #endif
 }
 
+#define CPU_FEATURE_NEON	(1UL << 12)
+static unsigned long cpu_feature_list;
+
+#if (defined SLJIT_DETECT_SIMD && SLJIT_DETECT_SIMD)
+#if defined(HAVE_GETAUXVAL) || defined(HAVE_ELF_AUX_INFO)
+#include <sys/auxv.h>
+#endif
+
+#ifdef __NetBSD__
+#include <sys/param.h>
+#include <sys/sysctl.h>
+#endif
+
+static void get_cpu_features(void)
+{
+	if (cpu_feature_list)
+		return;
+
+#if defined(__ARM_ARCH) && __ARM_ARCH == 8
+	/* TODO: confirm if optional with armv9 */
+	/* mandatory for armv8 */
+	cpu_feature_list = CPU_FEATURE_NEON;
+#elif defined(HAVE_GETAUXVAL)
+	cpu_feature_list = getauxval(AT_HWCAP);
+	if (errno == ENOENT)
+		cpu_feature_list = 1;
+#elif defined(__OpenBSD__)
+	/* required feature */
+	cpu_feature_list = CPU_FEATURE_NEON;
+#elif defined(__APPLE__) && defined (__ARM_NEON__)
+	cpu_feature_list = CPU_FEATURE_NEON;
+#elif defined(_WIN32)
+#ifndef FP_ARM_NEON_INSTRUCTIONS_AVAILABLE
+#define FP_ARM_NEON_INSTRUCTIONS_AVAILABLE 19
+#endif
+	if (IsProcessorFeaturePresent(FP_ARM_NEON_INSTRUCTIONS_AVAILABLE))
+		cpu_feature_list = CPU_FEATURE_NEON;
+#elif defined(__FreeBSD__) && defined(HAVE_ELF_AUX_INFO)
+	unsigned long buf;
+
+	if (elf_aux_info(AT_HWCAP, (void *)&buf, (int)sizeof(buf)))
+		cpu_feature_list = 1;
+		return;
+	}
+
+	if (buf & CPU_FEATURE_NEON)
+		cpu_feature_list = buf;
+#elif defined(__NetBSD__) || defined(__FreeBSD__)
+	int neon;
+	size_t len = sizeof(int);
+
+	if (sysctlbyname("machdep.neon_present", &neon, &len, NULL, 0) < 0) {
+		cpu_feature_list = 1;
+		return;
+	}
+
+	if (neon)
+		cpu_feature_list = CPU_FEATURE_NEON;
+#endif
+}
+
+#endif /* SLJIT_DETECT_SIMD */
+
 /* Last register + 1. */
 #define TMP_REG1	(SLJIT_NUMBER_OF_REGISTERS + 2)
 #define TMP_REG2	(SLJIT_NUMBER_OF_REGISTERS + 3)
@@ -973,6 +1036,13 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_has_cpu_feature(sljit_s32 feature_type)
 		return 2;
 #endif
 
+	case SLJIT_HAS_SIMD:
+#if (defined SLJIT_CONFIG_ARM_V7 && SLJIT_CONFIG_ARM_V7) && \
+      (defined SLJIT_DETECT_SIMD && SLJIT_DETECT_SIMD)
+		if (!cpu_feature_list)
+			get_cpu_features();
+#endif /* SLJIT_CONFIG_ARM_V7 && SLJIT_DETECT_SIMD */
+		return (cpu_feature_list & CPU_FEATURE_NEON) != 0;
 	default:
 		return 0;
 	}

sljit_src/sljitNativeARM_64.c

@@ -392,6 +392,7 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_has_cpu_feature(sljit_s32 feature_type)
 		return 1;
 #endif
 
+	case SLJIT_HAS_SIMD:
 	case SLJIT_HAS_CLZ:
 	case SLJIT_HAS_CTZ:
 	case SLJIT_HAS_ROT:

sljit_src/sljitNativeX86_common.c

@@ -294,7 +294,7 @@ static const sljit_u8 freg_lmap[SLJIT_NUMBER_OF_FLOAT_REGISTERS + 1] = {
    built-in CPU features. Therefore they can be overwritten by different threads
    if they detect the CPU features in the same time. */
 #define CPU_FEATURE_DETECTED		0x001
-#if (defined SLJIT_DETECT_SSE2 && SLJIT_DETECT_SSE2)
+#if (defined SLJIT_DETECT_SIMD && SLJIT_DETECT_SIMD)
 #define CPU_FEATURE_SSE2		0x002
 #endif
 #define CPU_FEATURE_LZCNT		0x004
@@ -444,7 +444,7 @@ static void get_cpu_features(void)
 
 #endif /* _MSC_VER && _MSC_VER >= 1400 */
 
-#if (defined SLJIT_DETECT_SSE2 && SLJIT_DETECT_SSE2)
+#if (defined SLJIT_DETECT_SIMD && SLJIT_DETECT_SIMD)
 	if (value & 0x4000000)
 		feature_list |= CPU_FEATURE_SSE2;
 #endif
@@ -738,13 +738,13 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_has_cpu_feature(sljit_s32 feature_type)
 	case SLJIT_HAS_FPU:
 #ifdef SLJIT_IS_FPU_AVAILABLE
 		return SLJIT_IS_FPU_AVAILABLE;
-#elif (defined SLJIT_DETECT_SSE2 && SLJIT_DETECT_SSE2)
+#elif (defined SLJIT_DETECT_SIMD && SLJIT_DETECT_SIMD)
 		if (cpu_feature_list == 0)
 			get_cpu_features();
 		return (cpu_feature_list & CPU_FEATURE_SSE2) != 0;
-#else /* SLJIT_DETECT_SSE2 */
+#else /* !SLJIT_DETECT_SIMD */
 		return 1;
-#endif /* SLJIT_DETECT_SSE2 */
+#endif /* SLJIT_IS_FPU_AVAILABLE */
 
 #if (defined SLJIT_CONFIG_X86_32 && SLJIT_CONFIG_X86_32)
 	case SLJIT_HAS_VIRTUAL_REGISTERS:
@@ -772,14 +772,14 @@ SLJIT_API_FUNC_ATTRIBUTE sljit_s32 sljit_has_cpu_feature(sljit_s32 feature_type)
 	case SLJIT_HAS_PREFETCH:
 		return 1;
 
-	case SLJIT_HAS_SSE2:
-#if (defined SLJIT_DETECT_SSE2 && SLJIT_DETECT_SSE2)
+	case SLJIT_HAS_SIMD:
+#if (defined SLJIT_DETECT_SIMD && SLJIT_DETECT_SIMD)
 		if (cpu_feature_list == 0)
 			get_cpu_features();
 		return (cpu_feature_list & CPU_FEATURE_SSE2) != 0;
-#else /* !SLJIT_DETECT_SSE2 */
+#else /* !SLJIT_DETECT_SIMD */
 		return 1;
-#endif /* SLJIT_DETECT_SSE2 */
+#endif /* SLJIT_DETECT_SIMD */
 
 	default:
 		return 0;

test_src/sljitTest.c

@@ -10636,10 +10636,18 @@ static void test85(void)
 
 int sljit_test(int argc, char* argv[])
 {
-	sljit_s32 has_arg = (argc >= 2 && argv[1][0] == '-' && argv[1][2] == '\0');
+	int fpu;
+	int simd = 0;
+	char features[24];
+	int has_arg = (argc >= 2 && argv[1][0] == '-' && argv[1][2] == '\0');
 	verbose = has_arg && argv[1][1] == 'v';
 	silent = has_arg && argv[1][1] == 's';
 
+#if (defined(SLJIT_CONFIG_ARM) && SLJIT_CONFIG_ARM) \
+	|| (defined SLJIT_CONFIG_X86 && SLJIT_CONFIG_X86)
+	simd = sljit_has_cpu_feature(SLJIT_HAS_SIMD);
+#endif
+
 	if (!verbose && !silent)
 		printf("Pass -v to enable verbose, -s to disable this hint.\n\n");
 
@@ -10743,7 +10751,16 @@ int sljit_test(int argc, char* argv[])
 		printf("all tests are " COLOR_GREEN "PASSED" COLOR_DEFAULT " ");
 	else
 		printf(COLOR_RED "%d" COLOR_DEFAULT " (" COLOR_RED "%d%%" COLOR_DEFAULT ") tests are " COLOR_RED "FAILED" COLOR_DEFAULT " ", TEST_COUNT - successful_tests, (TEST_COUNT - successful_tests) * 100 / TEST_COUNT);
-	printf("on " COLOR_ARCH "%s" COLOR_DEFAULT "%s\n", sljit_get_platform_name(), sljit_has_cpu_feature(SLJIT_HAS_FPU) ? " (with fpu)" : " (without fpu)");
+
+	fpu = sljit_has_cpu_feature(SLJIT_HAS_FPU);
+	if (simd && fpu)
+		strcpy(features, " (with: fpu, simd)");
+	else if (fpu)
+		strcpy(features, " (with fpu)");
+	else
+		strcpy(features, " (without fpu)");
+
+	printf("on " COLOR_ARCH "%s" COLOR_DEFAULT "%s\n", sljit_get_platform_name(), features);
 
 	return TEST_COUNT - successful_tests;