From: Andrei Tatar <andrei@unikraft.io>
Date: Mon, 17 Jul 2023 15:19:57 +0000 (+0200)
Subject: Add compatibility with clang >= 16
X-Git-Tag: RELEASE-0.14.0~10
X-Git-Url: http://xenbits.xensource.com/gitweb?a=commitdiff_plain;h=0e54e4dd238d75ed3e9f1c607ddccb9ffc0012af;p=unikraft%2Flibs%2Fintel-intrinsics.git

Add compatibility with clang >= 16

Implementations based off headers in LLVM 16.0.0.

Signed-off-by: Andrei Tatar <andrei@unikraft.io>
Reviewed-by: Maria Sfiraiala <maria.sfiraiala@gmail.com>
Reviewed-by: Radu Nichita <radunichita99@gmail.com>
Approved-by: Razvan Deaconescu <razvand@unikraft.io>
Tested-by: Unikraft CI <monkey@unikraft.io>
GitHub-Closes: #3
---

diff --git a/include/avx512bf16intrin.h b/include/avx512bf16intrin.h
index 0965373..eef0fc3 100644
--- a/include/avx512bf16intrin.h
+++ b/include/avx512bf16intrin.h
@@ -13,9 +13,15 @@
 #ifndef __AVX512BF16INTRIN_H
 #define __AVX512BF16INTRIN_H
 
+#if (__clang_major__ > 15)
+typedef __bf16 __v32bf __attribute__((__vector_size__(64), __aligned__(64)));
+typedef __bf16 __m512bh __attribute__((__vector_size__(64), __aligned__(64)));
+typedef __bf16 __bfloat16;
+#else
 typedef short __m512bh __attribute__((__vector_size__(64), __aligned__(64)));
 typedef short __m256bh __attribute__((__vector_size__(32), __aligned__(32)));
 typedef unsigned short __bfloat16;
+#endif
 
 #define __DEFAULT_FN_ATTRS512 \
   __attribute__((__always_inline__, __nodebug__, __target__("avx512bf16"), \
diff --git a/include/avx512vlbf16intrin.h b/include/avx512vlbf16intrin.h
index 6a5a860..adc43c1 100644
--- a/include/avx512vlbf16intrin.h
+++ b/include/avx512vlbf16intrin.h
@@ -13,7 +13,9 @@
 #ifndef __AVX512VLBF16INTRIN_H
 #define __AVX512VLBF16INTRIN_H
 
+#if (__clang_major__ <= 15)
 typedef short __m128bh __attribute__((__vector_size__(16), __aligned__(16)));
+#endif
 
 #define __DEFAULT_FN_ATTRS128 \
   __attribute__((__always_inline__, __nodebug__, \
@@ -415,9 +417,15 @@ _mm256_maskz_dpbf16_ps(__mmask8 __U, __m256 __D, __m256bh __A, __m256bh __B) {
 ///    and fraction field is truncated to 7 bits.
 static __inline__ __bfloat16 __DEFAULT_FN_ATTRS128 _mm_cvtness_sbh(float __A) {
   __v4sf __V = {__A, 0, 0, 0};
+#if (__clang_major__ > 15)
+  __v8bf __R = __builtin_ia32_cvtneps2bf16_128_mask(
+      (__v4sf)__V, (__v8bf)_mm_undefined_si128(), (__mmask8)-1);
+  return (__bf16)__R[0];
+#else
   __v8hi __R = __builtin_ia32_cvtneps2bf16_128_mask(
       (__v4sf)__V, (__v8hi)_mm_undefined_si128(), (__mmask8)-1);
   return __R[0];
+#endif
 }
 
 /// Convert Packed BF16 Data to Packed float Data.
diff --git a/include/avxintrin.h b/include/avxintrin.h
index 17fe636..2f2a159 100644
--- a/include/avxintrin.h
+++ b/include/avxintrin.h
@@ -39,6 +39,18 @@ typedef float __m256_u __attribute__ ((__vector_size__ (32), __aligned__(1)));
 typedef double __m256d_u __attribute__((__vector_size__(32), __aligned__(1)));
 typedef long long __m256i_u __attribute__((__vector_size__(32), __aligned__(1)));
 
+#if (__clang_major__ > 15)
+#ifdef __SSE2__
+/* Both _Float16 and __bf16 require SSE2 being enabled. */
+typedef _Float16 __v16hf __attribute__((__vector_size__(32), __aligned__(32)));
+typedef _Float16 __m256h __attribute__((__vector_size__(32), __aligned__(32)));
+typedef _Float16 __m256h_u __attribute__((__vector_size__(32), __aligned__(1)));
+
+typedef __bf16 __v16bf __attribute__((__vector_size__(32), __aligned__(32)));
+typedef __bf16 __m256bh __attribute__((__vector_size__(32), __aligned__(32)));
+#endif
+#endif
+
 /* Define the default attributes for the functions in this file. */
 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx"), __min_vector_width__(256)))
 #define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("avx"), __min_vector_width__(128)))
diff --git a/include/emmintrin.h b/include/emmintrin.h
index c2a8d4a..c25ce30 100644
--- a/include/emmintrin.h
+++ b/include/emmintrin.h
@@ -37,6 +37,18 @@ typedef unsigned char __v16qu __attribute__((__vector_size__(16)));
  * appear in the interface though. */
 typedef signed char __v16qs __attribute__((__vector_size__(16)));
 
+#if (__clang_major__ > 15)
+#ifdef __SSE2__
+/* Both _Float16 and __bf16 require SSE2 being enabled. */
+typedef _Float16 __v8hf __attribute__((__vector_size__(16), __aligned__(16)));
+typedef _Float16 __m128h __attribute__((__vector_size__(16), __aligned__(16)));
+typedef _Float16 __m128h_u __attribute__((__vector_size__(16), __aligned__(1)));
+
+typedef __bf16 __v8bf __attribute__((__vector_size__(16), __aligned__(16)));
+typedef __bf16 __m128bh __attribute__((__vector_size__(16), __aligned__(16)));
+#endif
+#endif
+
 /* Define the default attributes for the functions in this file. */
 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("sse2"), __min_vector_width__(128)))
 #define __DEFAULT_FN_ATTRS_MMX __attribute__((__always_inline__, __nodebug__, __target__("mmx,sse2"), __min_vector_width__(64)))