[X86][AVX10.2] Support AVX10.2-SATCVT-DS new instructions. #102592
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
llvmbot
added
clang
|
@llvm/pr-subscribers-llvm-ir @llvm/pr-subscribers-backend-x86 Author: Malay Sanghi (MalaySanghi) ChangesRef.: https://cdrdv2.intel.com/v1/dl/getContent/828965 Patch is 494.14 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/102592.diff 31 Files Affected:
diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def
index a696cf117908e2..a8639c341d0a43 100644
--- a/clang/include/clang/Basic/BuiltinsX86.def
+++ b/clang/include/clang/Basic/BuiltinsX86.def
@@ -2122,6 +2122,36 @@ TARGET_BUILTIN(__builtin_ia32_vpdpwuud256, "V8iV8iV8iV8i", "nV:256:", "avxvnniin
TARGET_BUILTIN(__builtin_ia32_vpdpwuuds128, "V4iV4iV4iV4i", "nV:128:", "avxvnniint16|avx10.2-256")
TARGET_BUILTIN(__builtin_ia32_vpdpwuuds256, "V8iV8iV8iV8i", "nV:256:", "avxvnniint16|avx10.2-256")
+// AVX10.2 SATCVT-DS
+TARGET_BUILTIN(__builtin_ia32_vcvttssd2si32, "iV2dIi", "ncV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttssd2usi32, "UiV2dIi", "ncV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttsss2si32, "iV4fIi", "ncV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttsss2usi32, "UiV4fIi", "ncV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2dqs128_mask, "V4iV2dV4iUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2dqs256_round_mask, "V4iV4dV4iUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2dqs512_round_mask, "V8iV8dV8iUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2udqs128_mask, "V4iV2dV4iUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2udqs256_round_mask, "V4iV4dV4iUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2udqs512_round_mask, "V8iV8dV8iUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2qqs128_mask, "V2OiV2dV2OiUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2qqs256_round_mask, "V4OiV4dV4OiUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2qqs512_round_mask, "V8OiV8dV8OiUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2uqqs128_mask, "V2OiV2dV2OiUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2uqqs256_round_mask, "V4OiV4dV4OiUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2uqqs512_round_mask, "V8OiV8dV8OiUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2dqs128_mask, "V4iV4fV4iUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2dqs256_round_mask, "V8iV8fV8iUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2dqs512_round_mask, "V16iV16fV16iUsIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2udqs128_mask, "V4iV4fV4iUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2udqs256_round_mask, "V8iV8fV8iUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2udqs512_round_mask, "V16iV16fV16iUsIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2qqs128_mask, "V2OiV4fV2OiUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2qqs256_round_mask, "V4OiV4fV4OiUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2qqs512_round_mask, "V8OiV8fV8OiUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2uqqs128_mask, "V2OiV4fV2OiUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2uqqs256_round_mask, "V4OiV4fV4OiUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2uqqs512_round_mask, "V8OiV8fV8OiUcIi", "nV:512:", "avx10.2-512")
+
// AVX-NE-CONVERT
TARGET_BUILTIN(__builtin_ia32_vbcstnebf162ps128, "V4fyC*", "nV:128:", "avxneconvert")
TARGET_BUILTIN(__builtin_ia32_vbcstnebf162ps256, "V8fyC*", "nV:256:", "avxneconvert")
diff --git a/clang/include/clang/Basic/BuiltinsX86_64.def b/clang/include/clang/Basic/BuiltinsX86_64.def
index 5e00916d4b25ae..ed9b17b8bd7b8e 100644
--- a/clang/include/clang/Basic/BuiltinsX86_64.def
+++ b/clang/include/clang/Basic/BuiltinsX86_64.def
@@ -99,6 +99,12 @@ TARGET_BUILTIN(__builtin_ia32_vcvttsh2si64, "OiV8xIi", "ncV:128:", "avx512fp16")
TARGET_BUILTIN(__builtin_ia32_vcvttsh2usi64, "UOiV8xIi", "ncV:128:", "avx512fp16")
TARGET_BUILTIN(__builtin_ia32_directstore_u64, "vULi*ULi", "n", "movdiri")
+// AVX10.2 SATCVT-DS
+TARGET_BUILTIN(__builtin_ia32_vcvttssd2si64, "OiV2dIi", "ncV:128:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttssd2usi64, "UOiV2dIi", "ncV:128:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttsss2si64, "OiV4fIi", "ncV:128:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttsss2usi64, "UOiV4fIi", "ncV:128:", "avx10.2-512")
+
// UINTR
TARGET_BUILTIN(__builtin_ia32_clui, "v", "n", "uintr")
TARGET_BUILTIN(__builtin_ia32_stui, "v", "n", "uintr")
diff --git a/clang/lib/Headers/CMakeLists.txt b/clang/lib/Headers/CMakeLists.txt
index b61aeca6bbc910..9981290628697c 100644
--- a/clang/lib/Headers/CMakeLists.txt
+++ b/clang/lib/Headers/CMakeLists.txt
@@ -150,9 +150,11 @@ set(x86_files
avx10_2_512minmaxintrin.h
avx10_2_512niintrin.h
avx10_2_512satcvtintrin.h
+ avx10_2_512satcvtdsintrin.h
avx10_2minmaxintrin.h
avx10_2niintrin.h
avx10_2satcvtintrin.h
+ avx10_2satcvtdsintrin.h
avx2intrin.h
avx512bf16intrin.h
avx512bitalgintrin.h
diff --git a/clang/lib/Headers/avx10_2_512satcvtdsintrin.h b/clang/lib/Headers/avx10_2_512satcvtdsintrin.h
new file mode 100644
index 00000000000000..e8b815653c3d6e
--- /dev/null
+++ b/clang/lib/Headers/avx10_2_512satcvtdsintrin.h
@@ -0,0 +1,302 @@
+/*===----- avx10_2_512satcvtdsintrin.h - AVX10_2_512SATCVTDS intrinsics ----===
+ *
+ * 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_512satcvtdsintrin.h> directly; include <immintrin.h> instead."
+#endif
+
+#ifndef __AVX10_2_512SATCVTDSINTRIN_H
+#define __AVX10_2_512SATCVTDSINTRIN_H
+
+/* Define the default attributes for the functions in this file. */
+#define __DEFAULT_FN_ATTRS \
+ __attribute__((__always_inline__, __nodebug__, __target__("avx10.2-512"), \
+ __min_vector_width__(512)))
+
+// 512 bit : Double -> Int
+static __inline__ __m256i __DEFAULT_FN_ATTRS _mm512_cvttspd_epi32(__m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask(
+ (__v8df)A, (__v8si)_mm256_undefined_si256(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttspd_epi32(__m256i W, __mmask8 U, __m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask(
+ (__v8df)A, (__v8si)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttspd_epi32(__mmask8 U, __m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask(
+ (__v8df)A, (__v8si)_mm256_setzero_si256(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundpd_epi32(A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)_mm256_undefined_si256(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundpd_epi32(W, U, A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)(__m256i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundpd_epi32(U, A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)_mm256_setzero_si256(), (__mmask8)(U), \
+ (const int)(R)))
+
+// 512 bit : Double -> uInt
+static __inline__ __m256i __DEFAULT_FN_ATTRS _mm512_cvttspd_epu32(__m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask(
+ (__v8df)A, (__v8si)_mm256_undefined_si256(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttspd_epu32(__m256i W, __mmask8 U, __m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask(
+ (__v8df)A, (__v8si)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttspd_epu32(__mmask8 U, __m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask(
+ (__v8df)A, (__v8si)_mm256_setzero_si256(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundpd_epu32(A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)_mm256_undefined_si256(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundpd_epu32(W, U, A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)(__m256i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundpd_epu32(U, A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)_mm256_setzero_si256(), (__mmask8)(U), \
+ (const int)(R)))
+
+#ifdef __x86_64__
+// 512 bit : Double -> Long
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttspd_epi64(__m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask(
+ (__v8df)A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttspd_epi64(__m512i W, __mmask8 U, __m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask(
+ (__v8df)A, (__v8di)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttspd_epi64(__mmask8 U, __m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask(
+ (__v8df)A, (__v8di)_mm512_setzero_si512(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundpd_epi64(A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)_mm512_undefined_epi32(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundpd_epi64(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)(__m512i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundpd_epi64(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)_mm512_setzero_si512(), (__mmask8)(U), \
+ (const int)(R)))
+
+// 512 bit : Double -> ULong
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttspd_epu64(__m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask(
+ (__v8df)A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttspd_epu64(__m512i W, __mmask8 U, __m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask(
+ (__v8df)A, (__v8di)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttspd_epu64(__mmask8 U, __m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask(
+ (__v8df)A, (__v8di)_mm512_setzero_si512(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundpd_epu64(A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)_mm512_undefined_epi32(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundpd_epu64(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)(__m512i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundpd_epu64(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)_mm512_setzero_si512(), (__mmask8)(U), \
+ (const int)(R)))
+
+#endif
+
+// 512 bit: Float -> int
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttsps_epi32(__m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask(
+ (__v16sf)(A), (__v16si)_mm512_undefined_epi32(), (__mmask16)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttsps_epi32(__m512i W, __mmask16 U, __m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask(
+ (__v16sf)(A), (__v16si)(W), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttsps_epi32(__mmask16 U, __m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask(
+ (__v16sf)(A), (__v16si)_mm512_setzero_si512(), U,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundps_epi32(A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)_mm512_undefined_epi32(), \
+ (__mmask16) - 1, (const int)(R)))
+
+#define _mm512_mask_cvtts_roundps_epi32(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)(__m512i)(W), (__mmask16)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundps_epi32(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)_mm512_setzero_si512(), (__mmask16)(U), \
+ (const int)(R)))
+
+// 512 bit: Float -> uint
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttsps_epu32(__m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask(
+ (__v16sf)(A), (__v16si)_mm512_undefined_epi32(), (__mmask16)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttsps_epu32(__m512i W, __mmask16 U, __m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask(
+ (__v16sf)(A), (__v16si)(W), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttsps_epu32(__mmask16 U, __m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask(
+ (__v16sf)(A), (__v16si)_mm512_setzero_si512(), U,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundps_epu32(A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)_mm512_undefined_epi32(), \
+ (__mmask16) - 1, (const int)(R)))
+
+#define _mm512_mask_cvtts_roundps_epu32(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)(__m512i)(W), (__mmask16)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundps_epu32(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)_mm512_setzero_si512(), (__mmask16)(U), \
+ (const int)(R)))
+
+#ifdef __x86_64__
+// 512 bit : float -> long
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttsps_epi64(__m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask(
+ (__v8sf)A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttsps_epi64(__m512i W, __mmask8 U, __m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask(
+ (__v8sf)A, (__v8di)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttsps_epi64(__mmask8 U, __m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask(
+ (__v8sf)A, (__v8di)_mm512_setzero_si512(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundps_epi64(A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)_mm512_undefined_epi32(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundps_epi64(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)(__m512i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundps_epi64(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)_mm512_setzero_si512(), (__mmask8)(U), \
+ (const int)(R)))
+
+// 512 bit : float -> ulong
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttsps_epu64(__m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask(
+ (__v8sf)A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttsps_epu64(__m512i W, __mmask8 U, __m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask(
+ (__v8sf)A, (__v8di)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttsps_epu64(__mmask8 U, __m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask(
+ (__v8sf)A, (__v8di)_mm512_setzero_si512(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundps_epu64(A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)_mm512_undefined_epi32(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundps_epu64(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)(__m512i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundps_epu64(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)_mm512_setzero_si512(), (__mmask8)(U), \
+ (const int)(R)))
+#endif
+
+#undef __DEFAULT_FN_ATTRS
+#endif // __AVX10_2_512SATCVTDSINTRIN_H
diff --git a/clang/lib/Headers/avx10_2satcvtdsintrin.h b/clang/lib/Headers/avx10_2satcvtdsintrin.h
new file mode 100644
index 00000000000000..5588c9ccfa4319
--- /dev/null
+++ b/clang/lib/Headers/avx10_2satcvtdsintrin.h
@@ -0,0 +1,453 @@
+/*===----------- avx10_2satcvtdsintrin.h - AVX512SATCVTDS intrinsics --------===
+ *
+ * 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_2satcvtdsintrin.h> directly; include <immintrin.h> instead."
+#endif // __IMMINTRIN_H
+
+#ifndef __AVX10_2SATCVTDSINTRIN_H
+#define __AVX10_2SATCVTDSINTRIN_H
+
+/* Define the default attributes for the functions in this file. */
+#define __DEFAULT_FN_ATTRS \
+ __attribute__((__always_inline__, __nodebug__, __target__("avx10.2-256"), \
+ __min_vector_width__(256)))
+
+#define _mm_cvtt_roundssd_i32(A, R) \
+ ((int)__builtin_ia32_vcvttssd2si32((__v2df)(__m128)(A), (const int)(R)))
+
+#define _mm_cvtt_roundssd_si32(A, R) \
+ ((int)__builtin_ia32_vcvttssd2si32...
[truncated]
|
|
@llvm/pr-subscribers-clang Author: Malay Sanghi (MalaySanghi) ChangesRef.: https://cdrdv2.intel.com/v1/dl/getContent/828965 Patch is 494.14 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/102592.diff 31 Files Affected:
diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def
index a696cf117908e2..a8639c341d0a43 100644
--- a/clang/include/clang/Basic/BuiltinsX86.def
+++ b/clang/include/clang/Basic/BuiltinsX86.def
@@ -2122,6 +2122,36 @@ TARGET_BUILTIN(__builtin_ia32_vpdpwuud256, "V8iV8iV8iV8i", "nV:256:", "avxvnniin
TARGET_BUILTIN(__builtin_ia32_vpdpwuuds128, "V4iV4iV4iV4i", "nV:128:", "avxvnniint16|avx10.2-256")
TARGET_BUILTIN(__builtin_ia32_vpdpwuuds256, "V8iV8iV8iV8i", "nV:256:", "avxvnniint16|avx10.2-256")
+// AVX10.2 SATCVT-DS
+TARGET_BUILTIN(__builtin_ia32_vcvttssd2si32, "iV2dIi", "ncV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttssd2usi32, "UiV2dIi", "ncV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttsss2si32, "iV4fIi", "ncV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttsss2usi32, "UiV4fIi", "ncV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2dqs128_mask, "V4iV2dV4iUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2dqs256_round_mask, "V4iV4dV4iUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2dqs512_round_mask, "V8iV8dV8iUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2udqs128_mask, "V4iV2dV4iUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2udqs256_round_mask, "V4iV4dV4iUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2udqs512_round_mask, "V8iV8dV8iUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2qqs128_mask, "V2OiV2dV2OiUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2qqs256_round_mask, "V4OiV4dV4OiUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2qqs512_round_mask, "V8OiV8dV8OiUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2uqqs128_mask, "V2OiV2dV2OiUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2uqqs256_round_mask, "V4OiV4dV4OiUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2uqqs512_round_mask, "V8OiV8dV8OiUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2dqs128_mask, "V4iV4fV4iUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2dqs256_round_mask, "V8iV8fV8iUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2dqs512_round_mask, "V16iV16fV16iUsIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2udqs128_mask, "V4iV4fV4iUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2udqs256_round_mask, "V8iV8fV8iUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2udqs512_round_mask, "V16iV16fV16iUsIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2qqs128_mask, "V2OiV4fV2OiUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2qqs256_round_mask, "V4OiV4fV4OiUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2qqs512_round_mask, "V8OiV8fV8OiUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2uqqs128_mask, "V2OiV4fV2OiUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2uqqs256_round_mask, "V4OiV4fV4OiUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2uqqs512_round_mask, "V8OiV8fV8OiUcIi", "nV:512:", "avx10.2-512")
+
// AVX-NE-CONVERT
TARGET_BUILTIN(__builtin_ia32_vbcstnebf162ps128, "V4fyC*", "nV:128:", "avxneconvert")
TARGET_BUILTIN(__builtin_ia32_vbcstnebf162ps256, "V8fyC*", "nV:256:", "avxneconvert")
diff --git a/clang/include/clang/Basic/BuiltinsX86_64.def b/clang/include/clang/Basic/BuiltinsX86_64.def
index 5e00916d4b25ae..ed9b17b8bd7b8e 100644
--- a/clang/include/clang/Basic/BuiltinsX86_64.def
+++ b/clang/include/clang/Basic/BuiltinsX86_64.def
@@ -99,6 +99,12 @@ TARGET_BUILTIN(__builtin_ia32_vcvttsh2si64, "OiV8xIi", "ncV:128:", "avx512fp16")
TARGET_BUILTIN(__builtin_ia32_vcvttsh2usi64, "UOiV8xIi", "ncV:128:", "avx512fp16")
TARGET_BUILTIN(__builtin_ia32_directstore_u64, "vULi*ULi", "n", "movdiri")
+// AVX10.2 SATCVT-DS
+TARGET_BUILTIN(__builtin_ia32_vcvttssd2si64, "OiV2dIi", "ncV:128:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttssd2usi64, "UOiV2dIi", "ncV:128:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttsss2si64, "OiV4fIi", "ncV:128:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttsss2usi64, "UOiV4fIi", "ncV:128:", "avx10.2-512")
+
// UINTR
TARGET_BUILTIN(__builtin_ia32_clui, "v", "n", "uintr")
TARGET_BUILTIN(__builtin_ia32_stui, "v", "n", "uintr")
diff --git a/clang/lib/Headers/CMakeLists.txt b/clang/lib/Headers/CMakeLists.txt
index b61aeca6bbc910..9981290628697c 100644
--- a/clang/lib/Headers/CMakeLists.txt
+++ b/clang/lib/Headers/CMakeLists.txt
@@ -150,9 +150,11 @@ set(x86_files
avx10_2_512minmaxintrin.h
avx10_2_512niintrin.h
avx10_2_512satcvtintrin.h
+ avx10_2_512satcvtdsintrin.h
avx10_2minmaxintrin.h
avx10_2niintrin.h
avx10_2satcvtintrin.h
+ avx10_2satcvtdsintrin.h
avx2intrin.h
avx512bf16intrin.h
avx512bitalgintrin.h
diff --git a/clang/lib/Headers/avx10_2_512satcvtdsintrin.h b/clang/lib/Headers/avx10_2_512satcvtdsintrin.h
new file mode 100644
index 00000000000000..e8b815653c3d6e
--- /dev/null
+++ b/clang/lib/Headers/avx10_2_512satcvtdsintrin.h
@@ -0,0 +1,302 @@
+/*===----- avx10_2_512satcvtdsintrin.h - AVX10_2_512SATCVTDS intrinsics ----===
+ *
+ * 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_512satcvtdsintrin.h> directly; include <immintrin.h> instead."
+#endif
+
+#ifndef __AVX10_2_512SATCVTDSINTRIN_H
+#define __AVX10_2_512SATCVTDSINTRIN_H
+
+/* Define the default attributes for the functions in this file. */
+#define __DEFAULT_FN_ATTRS \
+ __attribute__((__always_inline__, __nodebug__, __target__("avx10.2-512"), \
+ __min_vector_width__(512)))
+
+// 512 bit : Double -> Int
+static __inline__ __m256i __DEFAULT_FN_ATTRS _mm512_cvttspd_epi32(__m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask(
+ (__v8df)A, (__v8si)_mm256_undefined_si256(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttspd_epi32(__m256i W, __mmask8 U, __m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask(
+ (__v8df)A, (__v8si)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttspd_epi32(__mmask8 U, __m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask(
+ (__v8df)A, (__v8si)_mm256_setzero_si256(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundpd_epi32(A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)_mm256_undefined_si256(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundpd_epi32(W, U, A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)(__m256i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundpd_epi32(U, A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)_mm256_setzero_si256(), (__mmask8)(U), \
+ (const int)(R)))
+
+// 512 bit : Double -> uInt
+static __inline__ __m256i __DEFAULT_FN_ATTRS _mm512_cvttspd_epu32(__m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask(
+ (__v8df)A, (__v8si)_mm256_undefined_si256(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttspd_epu32(__m256i W, __mmask8 U, __m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask(
+ (__v8df)A, (__v8si)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttspd_epu32(__mmask8 U, __m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask(
+ (__v8df)A, (__v8si)_mm256_setzero_si256(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundpd_epu32(A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)_mm256_undefined_si256(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundpd_epu32(W, U, A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)(__m256i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundpd_epu32(U, A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)_mm256_setzero_si256(), (__mmask8)(U), \
+ (const int)(R)))
+
+#ifdef __x86_64__
+// 512 bit : Double -> Long
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttspd_epi64(__m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask(
+ (__v8df)A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttspd_epi64(__m512i W, __mmask8 U, __m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask(
+ (__v8df)A, (__v8di)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttspd_epi64(__mmask8 U, __m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask(
+ (__v8df)A, (__v8di)_mm512_setzero_si512(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundpd_epi64(A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)_mm512_undefined_epi32(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundpd_epi64(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)(__m512i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundpd_epi64(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)_mm512_setzero_si512(), (__mmask8)(U), \
+ (const int)(R)))
+
+// 512 bit : Double -> ULong
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttspd_epu64(__m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask(
+ (__v8df)A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttspd_epu64(__m512i W, __mmask8 U, __m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask(
+ (__v8df)A, (__v8di)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttspd_epu64(__mmask8 U, __m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask(
+ (__v8df)A, (__v8di)_mm512_setzero_si512(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundpd_epu64(A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)_mm512_undefined_epi32(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundpd_epu64(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)(__m512i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundpd_epu64(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)_mm512_setzero_si512(), (__mmask8)(U), \
+ (const int)(R)))
+
+#endif
+
+// 512 bit: Float -> int
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttsps_epi32(__m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask(
+ (__v16sf)(A), (__v16si)_mm512_undefined_epi32(), (__mmask16)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttsps_epi32(__m512i W, __mmask16 U, __m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask(
+ (__v16sf)(A), (__v16si)(W), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttsps_epi32(__mmask16 U, __m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask(
+ (__v16sf)(A), (__v16si)_mm512_setzero_si512(), U,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundps_epi32(A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)_mm512_undefined_epi32(), \
+ (__mmask16) - 1, (const int)(R)))
+
+#define _mm512_mask_cvtts_roundps_epi32(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)(__m512i)(W), (__mmask16)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundps_epi32(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)_mm512_setzero_si512(), (__mmask16)(U), \
+ (const int)(R)))
+
+// 512 bit: Float -> uint
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttsps_epu32(__m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask(
+ (__v16sf)(A), (__v16si)_mm512_undefined_epi32(), (__mmask16)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttsps_epu32(__m512i W, __mmask16 U, __m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask(
+ (__v16sf)(A), (__v16si)(W), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttsps_epu32(__mmask16 U, __m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask(
+ (__v16sf)(A), (__v16si)_mm512_setzero_si512(), U,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundps_epu32(A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)_mm512_undefined_epi32(), \
+ (__mmask16) - 1, (const int)(R)))
+
+#define _mm512_mask_cvtts_roundps_epu32(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)(__m512i)(W), (__mmask16)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundps_epu32(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)_mm512_setzero_si512(), (__mmask16)(U), \
+ (const int)(R)))
+
+#ifdef __x86_64__
+// 512 bit : float -> long
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttsps_epi64(__m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask(
+ (__v8sf)A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttsps_epi64(__m512i W, __mmask8 U, __m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask(
+ (__v8sf)A, (__v8di)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttsps_epi64(__mmask8 U, __m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask(
+ (__v8sf)A, (__v8di)_mm512_setzero_si512(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundps_epi64(A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)_mm512_undefined_epi32(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundps_epi64(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)(__m512i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundps_epi64(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)_mm512_setzero_si512(), (__mmask8)(U), \
+ (const int)(R)))
+
+// 512 bit : float -> ulong
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttsps_epu64(__m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask(
+ (__v8sf)A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttsps_epu64(__m512i W, __mmask8 U, __m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask(
+ (__v8sf)A, (__v8di)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttsps_epu64(__mmask8 U, __m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask(
+ (__v8sf)A, (__v8di)_mm512_setzero_si512(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundps_epu64(A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)_mm512_undefined_epi32(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundps_epu64(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)(__m512i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundps_epu64(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)_mm512_setzero_si512(), (__mmask8)(U), \
+ (const int)(R)))
+#endif
+
+#undef __DEFAULT_FN_ATTRS
+#endif // __AVX10_2_512SATCVTDSINTRIN_H
diff --git a/clang/lib/Headers/avx10_2satcvtdsintrin.h b/clang/lib/Headers/avx10_2satcvtdsintrin.h
new file mode 100644
index 00000000000000..5588c9ccfa4319
--- /dev/null
+++ b/clang/lib/Headers/avx10_2satcvtdsintrin.h
@@ -0,0 +1,453 @@
+/*===----------- avx10_2satcvtdsintrin.h - AVX512SATCVTDS intrinsics --------===
+ *
+ * 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_2satcvtdsintrin.h> directly; include <immintrin.h> instead."
+#endif // __IMMINTRIN_H
+
+#ifndef __AVX10_2SATCVTDSINTRIN_H
+#define __AVX10_2SATCVTDSINTRIN_H
+
+/* Define the default attributes for the functions in this file. */
+#define __DEFAULT_FN_ATTRS \
+ __attribute__((__always_inline__, __nodebug__, __target__("avx10.2-256"), \
+ __min_vector_width__(256)))
+
+#define _mm_cvtt_roundssd_i32(A, R) \
+ ((int)__builtin_ia32_vcvttssd2si32((__v2df)(__m128)(A), (const int)(R)))
+
+#define _mm_cvtt_roundssd_si32(A, R) \
+ ((int)__builtin_ia32_vcvttssd2si32...
[truncated]
|
|
@llvm/pr-subscribers-mc Author: Malay Sanghi (MalaySanghi) ChangesRef.: https://cdrdv2.intel.com/v1/dl/getContent/828965 Patch is 494.14 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/102592.diff 31 Files Affected:
diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def
index a696cf117908e2..a8639c341d0a43 100644
--- a/clang/include/clang/Basic/BuiltinsX86.def
+++ b/clang/include/clang/Basic/BuiltinsX86.def
@@ -2122,6 +2122,36 @@ TARGET_BUILTIN(__builtin_ia32_vpdpwuud256, "V8iV8iV8iV8i", "nV:256:", "avxvnniin
TARGET_BUILTIN(__builtin_ia32_vpdpwuuds128, "V4iV4iV4iV4i", "nV:128:", "avxvnniint16|avx10.2-256")
TARGET_BUILTIN(__builtin_ia32_vpdpwuuds256, "V8iV8iV8iV8i", "nV:256:", "avxvnniint16|avx10.2-256")
+// AVX10.2 SATCVT-DS
+TARGET_BUILTIN(__builtin_ia32_vcvttssd2si32, "iV2dIi", "ncV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttssd2usi32, "UiV2dIi", "ncV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttsss2si32, "iV4fIi", "ncV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttsss2usi32, "UiV4fIi", "ncV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2dqs128_mask, "V4iV2dV4iUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2dqs256_round_mask, "V4iV4dV4iUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2dqs512_round_mask, "V8iV8dV8iUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2udqs128_mask, "V4iV2dV4iUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2udqs256_round_mask, "V4iV4dV4iUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2udqs512_round_mask, "V8iV8dV8iUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2qqs128_mask, "V2OiV2dV2OiUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2qqs256_round_mask, "V4OiV4dV4OiUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2qqs512_round_mask, "V8OiV8dV8OiUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2uqqs128_mask, "V2OiV2dV2OiUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2uqqs256_round_mask, "V4OiV4dV4OiUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttpd2uqqs512_round_mask, "V8OiV8dV8OiUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2dqs128_mask, "V4iV4fV4iUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2dqs256_round_mask, "V8iV8fV8iUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2dqs512_round_mask, "V16iV16fV16iUsIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2udqs128_mask, "V4iV4fV4iUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2udqs256_round_mask, "V8iV8fV8iUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2udqs512_round_mask, "V16iV16fV16iUsIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2qqs128_mask, "V2OiV4fV2OiUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2qqs256_round_mask, "V4OiV4fV4OiUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2qqs512_round_mask, "V8OiV8fV8OiUcIi", "nV:512:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2uqqs128_mask, "V2OiV4fV2OiUc", "nV:128:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2uqqs256_round_mask, "V4OiV4fV4OiUcIi", "nV:256:", "avx10.2-256")
+TARGET_BUILTIN(__builtin_ia32_vcvttps2uqqs512_round_mask, "V8OiV8fV8OiUcIi", "nV:512:", "avx10.2-512")
+
// AVX-NE-CONVERT
TARGET_BUILTIN(__builtin_ia32_vbcstnebf162ps128, "V4fyC*", "nV:128:", "avxneconvert")
TARGET_BUILTIN(__builtin_ia32_vbcstnebf162ps256, "V8fyC*", "nV:256:", "avxneconvert")
diff --git a/clang/include/clang/Basic/BuiltinsX86_64.def b/clang/include/clang/Basic/BuiltinsX86_64.def
index 5e00916d4b25ae..ed9b17b8bd7b8e 100644
--- a/clang/include/clang/Basic/BuiltinsX86_64.def
+++ b/clang/include/clang/Basic/BuiltinsX86_64.def
@@ -99,6 +99,12 @@ TARGET_BUILTIN(__builtin_ia32_vcvttsh2si64, "OiV8xIi", "ncV:128:", "avx512fp16")
TARGET_BUILTIN(__builtin_ia32_vcvttsh2usi64, "UOiV8xIi", "ncV:128:", "avx512fp16")
TARGET_BUILTIN(__builtin_ia32_directstore_u64, "vULi*ULi", "n", "movdiri")
+// AVX10.2 SATCVT-DS
+TARGET_BUILTIN(__builtin_ia32_vcvttssd2si64, "OiV2dIi", "ncV:128:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttssd2usi64, "UOiV2dIi", "ncV:128:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttsss2si64, "OiV4fIi", "ncV:128:", "avx10.2-512")
+TARGET_BUILTIN(__builtin_ia32_vcvttsss2usi64, "UOiV4fIi", "ncV:128:", "avx10.2-512")
+
// UINTR
TARGET_BUILTIN(__builtin_ia32_clui, "v", "n", "uintr")
TARGET_BUILTIN(__builtin_ia32_stui, "v", "n", "uintr")
diff --git a/clang/lib/Headers/CMakeLists.txt b/clang/lib/Headers/CMakeLists.txt
index b61aeca6bbc910..9981290628697c 100644
--- a/clang/lib/Headers/CMakeLists.txt
+++ b/clang/lib/Headers/CMakeLists.txt
@@ -150,9 +150,11 @@ set(x86_files
avx10_2_512minmaxintrin.h
avx10_2_512niintrin.h
avx10_2_512satcvtintrin.h
+ avx10_2_512satcvtdsintrin.h
avx10_2minmaxintrin.h
avx10_2niintrin.h
avx10_2satcvtintrin.h
+ avx10_2satcvtdsintrin.h
avx2intrin.h
avx512bf16intrin.h
avx512bitalgintrin.h
diff --git a/clang/lib/Headers/avx10_2_512satcvtdsintrin.h b/clang/lib/Headers/avx10_2_512satcvtdsintrin.h
new file mode 100644
index 00000000000000..e8b815653c3d6e
--- /dev/null
+++ b/clang/lib/Headers/avx10_2_512satcvtdsintrin.h
@@ -0,0 +1,302 @@
+/*===----- avx10_2_512satcvtdsintrin.h - AVX10_2_512SATCVTDS intrinsics ----===
+ *
+ * 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_512satcvtdsintrin.h> directly; include <immintrin.h> instead."
+#endif
+
+#ifndef __AVX10_2_512SATCVTDSINTRIN_H
+#define __AVX10_2_512SATCVTDSINTRIN_H
+
+/* Define the default attributes for the functions in this file. */
+#define __DEFAULT_FN_ATTRS \
+ __attribute__((__always_inline__, __nodebug__, __target__("avx10.2-512"), \
+ __min_vector_width__(512)))
+
+// 512 bit : Double -> Int
+static __inline__ __m256i __DEFAULT_FN_ATTRS _mm512_cvttspd_epi32(__m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask(
+ (__v8df)A, (__v8si)_mm256_undefined_si256(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttspd_epi32(__m256i W, __mmask8 U, __m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask(
+ (__v8df)A, (__v8si)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttspd_epi32(__mmask8 U, __m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask(
+ (__v8df)A, (__v8si)_mm256_setzero_si256(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundpd_epi32(A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)_mm256_undefined_si256(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundpd_epi32(W, U, A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)(__m256i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundpd_epi32(U, A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2dqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)_mm256_setzero_si256(), (__mmask8)(U), \
+ (const int)(R)))
+
+// 512 bit : Double -> uInt
+static __inline__ __m256i __DEFAULT_FN_ATTRS _mm512_cvttspd_epu32(__m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask(
+ (__v8df)A, (__v8si)_mm256_undefined_si256(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttspd_epu32(__m256i W, __mmask8 U, __m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask(
+ (__v8df)A, (__v8si)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttspd_epu32(__mmask8 U, __m512d A) {
+ return ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask(
+ (__v8df)A, (__v8si)_mm256_setzero_si256(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundpd_epu32(A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)_mm256_undefined_si256(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundpd_epu32(W, U, A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)(__m256i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundpd_epu32(U, A, R) \
+ ((__m256i)__builtin_ia32_vcvttpd2udqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8si)_mm256_setzero_si256(), (__mmask8)(U), \
+ (const int)(R)))
+
+#ifdef __x86_64__
+// 512 bit : Double -> Long
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttspd_epi64(__m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask(
+ (__v8df)A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttspd_epi64(__m512i W, __mmask8 U, __m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask(
+ (__v8df)A, (__v8di)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttspd_epi64(__mmask8 U, __m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask(
+ (__v8df)A, (__v8di)_mm512_setzero_si512(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundpd_epi64(A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)_mm512_undefined_epi32(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundpd_epi64(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)(__m512i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundpd_epi64(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2qqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)_mm512_setzero_si512(), (__mmask8)(U), \
+ (const int)(R)))
+
+// 512 bit : Double -> ULong
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttspd_epu64(__m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask(
+ (__v8df)A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttspd_epu64(__m512i W, __mmask8 U, __m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask(
+ (__v8df)A, (__v8di)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttspd_epu64(__mmask8 U, __m512d A) {
+ return ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask(
+ (__v8df)A, (__v8di)_mm512_setzero_si512(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundpd_epu64(A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)_mm512_undefined_epi32(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundpd_epu64(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)(__m512i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundpd_epu64(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttpd2uqqs512_round_mask( \
+ (__v8df)(__m512d)(A), (__v8di)_mm512_setzero_si512(), (__mmask8)(U), \
+ (const int)(R)))
+
+#endif
+
+// 512 bit: Float -> int
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttsps_epi32(__m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask(
+ (__v16sf)(A), (__v16si)_mm512_undefined_epi32(), (__mmask16)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttsps_epi32(__m512i W, __mmask16 U, __m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask(
+ (__v16sf)(A), (__v16si)(W), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttsps_epi32(__mmask16 U, __m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask(
+ (__v16sf)(A), (__v16si)_mm512_setzero_si512(), U,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundps_epi32(A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)_mm512_undefined_epi32(), \
+ (__mmask16) - 1, (const int)(R)))
+
+#define _mm512_mask_cvtts_roundps_epi32(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)(__m512i)(W), (__mmask16)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundps_epi32(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2dqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)_mm512_setzero_si512(), (__mmask16)(U), \
+ (const int)(R)))
+
+// 512 bit: Float -> uint
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttsps_epu32(__m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask(
+ (__v16sf)(A), (__v16si)_mm512_undefined_epi32(), (__mmask16)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttsps_epu32(__m512i W, __mmask16 U, __m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask(
+ (__v16sf)(A), (__v16si)(W), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttsps_epu32(__mmask16 U, __m512 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask(
+ (__v16sf)(A), (__v16si)_mm512_setzero_si512(), U,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundps_epu32(A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)_mm512_undefined_epi32(), \
+ (__mmask16) - 1, (const int)(R)))
+
+#define _mm512_mask_cvtts_roundps_epu32(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)(__m512i)(W), (__mmask16)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundps_epu32(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2udqs512_round_mask( \
+ (__v16sf)(__m512)(A), (__v16si)_mm512_setzero_si512(), (__mmask16)(U), \
+ (const int)(R)))
+
+#ifdef __x86_64__
+// 512 bit : float -> long
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttsps_epi64(__m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask(
+ (__v8sf)A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttsps_epi64(__m512i W, __mmask8 U, __m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask(
+ (__v8sf)A, (__v8di)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttsps_epi64(__mmask8 U, __m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask(
+ (__v8sf)A, (__v8di)_mm512_setzero_si512(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundps_epi64(A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)_mm512_undefined_epi32(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundps_epi64(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)(__m512i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundps_epi64(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2qqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)_mm512_setzero_si512(), (__mmask8)(U), \
+ (const int)(R)))
+
+// 512 bit : float -> ulong
+static __inline__ __m512i __DEFAULT_FN_ATTRS _mm512_cvttsps_epu64(__m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask(
+ (__v8sf)A, (__v8di)_mm512_undefined_epi32(), (__mmask8)-1,
+ _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_cvttsps_epu64(__m512i W, __mmask8 U, __m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask(
+ (__v8sf)A, (__v8di)W, U, _MM_FROUND_CUR_DIRECTION));
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_cvttsps_epu64(__mmask8 U, __m256 A) {
+ return ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask(
+ (__v8sf)A, (__v8di)_mm512_setzero_si512(), U, _MM_FROUND_CUR_DIRECTION));
+}
+
+#define _mm512_cvtts_roundps_epu64(A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)_mm512_undefined_epi32(), (__mmask8) - 1, \
+ (const int)(R)))
+
+#define _mm512_mask_cvtts_roundps_epu64(W, U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)(__m512i)(W), (__mmask8)(U), \
+ (const int)(R)))
+
+#define _mm512_maskz_cvtts_roundps_epu64(U, A, R) \
+ ((__m512i)__builtin_ia32_vcvttps2uqqs512_round_mask( \
+ (__v8sf)(__m256)(A), (__v8di)_mm512_setzero_si512(), (__mmask8)(U), \
+ (const int)(R)))
+#endif
+
+#undef __DEFAULT_FN_ATTRS
+#endif // __AVX10_2_512SATCVTDSINTRIN_H
diff --git a/clang/lib/Headers/avx10_2satcvtdsintrin.h b/clang/lib/Headers/avx10_2satcvtdsintrin.h
new file mode 100644
index 00000000000000..5588c9ccfa4319
--- /dev/null
+++ b/clang/lib/Headers/avx10_2satcvtdsintrin.h
@@ -0,0 +1,453 @@
+/*===----------- avx10_2satcvtdsintrin.h - AVX512SATCVTDS intrinsics --------===
+ *
+ * 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_2satcvtdsintrin.h> directly; include <immintrin.h> instead."
+#endif // __IMMINTRIN_H
+
+#ifndef __AVX10_2SATCVTDSINTRIN_H
+#define __AVX10_2SATCVTDSINTRIN_H
+
+/* Define the default attributes for the functions in this file. */
+#define __DEFAULT_FN_ATTRS \
+ __attribute__((__always_inline__, __nodebug__, __target__("avx10.2-256"), \
+ __min_vector_width__(256)))
+
+#define _mm_cvtt_roundssd_i32(A, R) \
+ ((int)__builtin_ia32_vcvttssd2si32((__v2df)(__m128)(A), (const int)(R)))
+
+#define _mm_cvtt_roundssd_si32(A, R) \
+ ((int)__builtin_ia32_vcvttssd2si32...
[truncated]
|
|
@phoebewang @KanRobert please review |
phoebewang
reviewed
Aug 12, 2024
Comment on lines
+9
to
+10
| declare i32 @llvm.fptosi.sat.i32.f32 (float) | ||
| declare i64 @llvm.fptosi.sat.i64.f32 (float) |
There was a problem hiding this comment.
This will be a follow-up change. Mappings from public intrinsic to new ISA for vector types are not a part of this PR.
FreddyLeaf
reviewed
Aug 13, 2024
FreddyLeaf
changed the title
MalaySanghi
commented
Aug 27, 2024
| @@ -45,6 +45,14 @@ bool SemaX86::CheckBuiltinRoundingOrSAE(unsigned BuiltinID, CallExpr *TheCall) { | |||
| case X86::BI__builtin_ia32_vcvttsh2si64: | |||
There was a problem hiding this comment.
@phoebewang These don't have sis32/64 either.
I'm not sure what is the convention here.
There was a problem hiding this comment.
We don't have a VCVTTSH2SIS instruction. The general convention is to use the instruction name directly.
phoebewang
reviewed
Sep 4, 2024
phoebewang
reviewed
Sep 4, 2024
phoebewang
reviewed
Sep 4, 2024
phoebewang
reviewed
Sep 4, 2024
phoebewang
reviewed
Sep 4, 2024
phoebewang
reviewed
Sep 4, 2024
phoebewang
reviewed
Sep 5, 2024
phoebewang
reviewed
Sep 5, 2024
phoebewang
reviewed
Sep 5, 2024
phoebewang
reviewed
Sep 5, 2024
phoebewang
reviewed
Sep 5, 2024
phoebewang
reviewed
Sep 5, 2024
phoebewang
reviewed
Sep 5, 2024
phoebewang
reviewed
Sep 5, 2024
phoebewang
reviewed
Sep 5, 2024
phoebewang
reviewed
Sep 5, 2024
phoebewang
reviewed
Sep 5, 2024
phoebewang
reviewed
Sep 5, 2024
phoebewang
reviewed
Sep 9, 2024
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/13/builds/2175 Here is the relevant piece of the build log for the reference |
| @@ -0,0 +1,52 @@ | |||
| // RUN: %clang_cc1 -flax-vector-conversions=none -ffreestanding %s -triple=i386-unknown-unknown -target-feature +avx10.2-512 -emit-llvm -Wall -Werror -verify | |||
There was a problem hiding this comment.
Can we remove "-emit-llvm" from this test (and others in this patch) since noone seems to look at the produced output anyway?
See e.g. b7730a2
[test] Avoid writing to a potentially write-protected dir (#102073)
solving a similar problem.
There was a problem hiding this comment.
Hi,
Thanks for bringing this to my notice. I'll send a patch in some time
There was a problem hiding this comment.
created #108525
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/146/builds/1139 Here is the relevant piece of the build log for the reference |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Ref.: https://cdrdv2.intel.com/v1/dl/getContent/828965