Provide modeling and runtime detection of imm8 arm extension

include/xsimd/arch/xsimd_i8mm_neon64.hpp

@@ -0,0 +1,17 @@
+/***************************************************************************
+ * Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and         *
+ * Martin Renou                                                             *
+ * Copyright (c) QuantStack                                                 *
+ * Copyright (c) Serge Guelton                                              *
+ *                                                                          *
+ * Distributed under the terms of the BSD 3-Clause License.                 *
+ *                                                                          *
+ * The full license is in the file LICENSE, distributed with this software. *
+ ****************************************************************************/
+
+#ifndef XSIMD_I8MM_NEON64_HPP
+#define XSIMD_I8MM_NEON64_HPP
+
+#include "../types/xsimd_i8mm_neon64_register.hpp"
+
+#endif

include/xsimd/arch/xsimd_isa.hpp

@@ -104,6 +104,10 @@
 #include "./xsimd_neon64.hpp"
 #endif
 
+#if XSIMD_WITH_I8MM_NEON64
+#include "./xsimd_i8mm_neon64.hpp"
+#endif
+
 #if XSIMD_WITH_SVE
 #include "./xsimd_sve.hpp"
 #endif

include/xsimd/arch/xsimd_neon64.hpp

@@ -92,7 +92,7 @@ namespace xsimd
         template <class A, class T>
         inline batch<T, A> broadcast(T val, requires_arch<neon64>) noexcept
         {
-            return broadcast<neon64>(val, neon {});
+            return broadcast<A>(val, neon {});
         }
 
         template <class A>

include/xsimd/config/xsimd_arch.hpp

@@ -194,7 +194,7 @@ namespace xsimd
 
     using all_sve_architectures = arch_list<detail::sve<512>, detail::sve<256>, detail::sve<128>>;
     using all_rvv_architectures = arch_list<detail::rvv<512>, detail::rvv<256>, detail::rvv<128>>;
-    using all_arm_architectures = typename detail::join<all_sve_architectures, arch_list<neon64, neon>>::type;
+    using all_arm_architectures = typename detail::join<all_sve_architectures, arch_list<i8mm<neon64>, neon64, neon>>::type;
     using all_riscv_architectures = all_rvv_architectures;
     using all_wasm_architectures = arch_list<wasm>;
     using all_architectures = typename detail::join<all_riscv_architectures, all_wasm_architectures, all_arm_architectures, all_x86_architectures>::type;

include/xsimd/config/xsimd_config.hpp

@@ -349,6 +349,17 @@
 #define XSIMD_WITH_NEON64 0
 #endif
 
+/**
+ * @ingroup xsimd_config_macro
+ *
+ * Set to 1 if i8mm neon64 extension is available at compile-time, to 0 otherwise.
+ */
+#if defined(__ARM_FEATURE_MATMUL_INT8)
+#define XSIMD_WITH_I8MM_NEON64 1
+#else
+#define XSIMD_WITH_I8MM_NEON64 0
+#endif
+
 /**
  * @ingroup xsimd_config_macro
  *

include/xsimd/config/xsimd_cpuid.hpp

@@ -18,6 +18,11 @@
 #if defined(__linux__) && (defined(__ARM_NEON) || defined(_M_ARM) || defined(__riscv_vector))
 #include <asm/hwcap.h>
 #include <sys/auxv.h>
+
+#ifndef HWCAP2_I8MM
+#define HWCAP2_I8MM (1 << 13)
+#endif
+
 #endif
 
 #if defined(_MSC_VER)
@@ -66,6 +71,7 @@ namespace xsimd
             ARCH_FIELD_EX(avx512vnni<::xsimd::avx512vbmi>, avx512vnni_vbmi)
             ARCH_FIELD(neon)
             ARCH_FIELD(neon64)
+            ARCH_FIELD_EX(i8mm<::xsimd::neon64>, i8mm_neon64)
             ARCH_FIELD(sve)
             ARCH_FIELD(rvv)
             ARCH_FIELD(wasm)
@@ -83,6 +89,9 @@ namespace xsimd
 #if defined(__aarch64__) || defined(_M_ARM64)
                 neon = 1;
                 neon64 = 1;
+#if defined(__linux__) && (!defined(__ANDROID_API__) || __ANDROID_API__ >= 18)
+                i8mm_neon64 = bool(getauxval(AT_HWCAP2) & HWCAP2_I8MM);
+#endif
 #elif defined(__ARM_NEON) || defined(_M_ARM)
 
 #if defined(__linux__) && (!defined(__ANDROID_API__) || __ANDROID_API__ >= 18)

include/xsimd/types/xsimd_all_registers.hpp

@@ -36,6 +36,8 @@
 #include "xsimd_avx512dq_register.hpp"
 #include "xsimd_avx512f_register.hpp"
 
+#include "xsimd_i8mm_neon64_register.hpp"
+
 #include "xsimd_neon64_register.hpp"
 #include "xsimd_neon_register.hpp"
 

include/xsimd/types/xsimd_i8mm_neon64_register.hpp

@@ -0,0 +1,46 @@
+/***************************************************************************
+ * Copyright (c) Johan Mabille, Sylvain Corlay, Wolf Vollprecht and         *
+ * Martin Renou                                                             *
+ * Copyright (c) QuantStack                                                 *
+ * Copyright (c) Serge Guelton                                              *
+ *                                                                          *
+ * Distributed under the terms of the BSD 3-Clause License.                 *
+ *                                                                          *
+ * The full license is in the file LICENSE, distributed with this software. *
+ ****************************************************************************/
+
+#ifndef XSIMD_I8MM_NEON64_REGISTER_HPP
+#define XSIMD_I8MM_NEON64_REGISTER_HPP
+
+#include "./xsimd_neon64_register.hpp"
+
+namespace xsimd
+{
+    template <typename arch>
+    struct i8mm;
+
+    /**
+     * @ingroup architectures
+     *
+     * Neon64 + i8mm instructions
+     */
+    template <>
+    struct i8mm<neon64> : neon64
+    {
+        static constexpr bool supported() noexcept { return XSIMD_WITH_I8MM_NEON64; }
+        static constexpr bool available() noexcept { return true; }
+        static constexpr unsigned version() noexcept { return generic::version(8, 2, 0); }
+        static constexpr char const* name() noexcept { return "i8mm+neon64"; }
+    };
+
+#if XSIMD_WITH_I8MM_NEON64
+    namespace types
+    {
+
+        XSIMD_DECLARE_SIMD_REGISTER_ALIAS(i8mm<neon64>, neon64);
+
+    }
+#endif
+
+}
+#endif