[clang] [llvm] [X86][AVX10.2] Support AVX10.2-CONVERT new instructions. (PR #101600)
Phoebe Wang via cfe-commits
cfe-commits at lists.llvm.org
Wed Aug 14 18:57:34 PDT 2024
================
@@ -0,0 +1,286 @@
+/*===--------- avx10_2_512convertintrin.h - AVX10_2_512CONVERT -------------===
+ *
+ * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See https://llvm.org/LICENSE.txt for license information.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ *===-----------------------------------------------------------------------===
+ */
+#ifndef __IMMINTRIN_H
+#error \
+ "Never use <avx10_2_512convertintrin.h> directly; include <immintrin.h> instead."
+#endif // __IMMINTRIN_H
+
+#ifdef __SSE2__
+
+#ifndef __AVX10_2_512CONVERTINTRIN_H
+#define __AVX10_2_512CONVERTINTRIN_H
+
+/* Define the default attributes for the functions in this file. */
+#define __DEFAULT_FN_ATTRS512 \
+ __attribute__((__always_inline__, __nodebug__, __target__("avx10.2-512"), \
+ __min_vector_width__(512)))
+
+static __inline__ __m512h __DEFAULT_FN_ATTRS512 _mm512_cvtx2ps_ph(__m512 __A,
+ __m512 __B) {
+ return (__m512h)__builtin_ia32_vcvt2ps2phx512_mask(
+ (__v16sf)__A, (__v16sf)__B, (__v32hf)_mm512_setzero_ph(), (__mmask32)(-1),
+ _MM_FROUND_CUR_DIRECTION);
+}
+
+static __inline__ __m512h __DEFAULT_FN_ATTRS512
+_mm512_mask_cvtx2ps_ph(__m512h __W, __mmask32 __U, __m512 __A, __m512 __B) {
----------------
phoebewang wrote:
Missing `_mm512_maskz_cvtx2ps_ph`? Seems several zeroing mask intrinsics are missing.
https://github.com/llvm/llvm-project/pull/101600
More information about the cfe-commits
mailing list