[X86] Support SM4 EVEX version intrinsics/instructions. (#113402)

Ref.: https://cdrdv2.intel.com/v1/dl/getContent/671368

Author: FreddyLeaf

clang/docs/ReleaseNotes.rst

@@ -628,6 +628,10 @@ X86 Support
   * Supported MINMAX intrinsics of ``*_(mask(z)))_minmax(ne)_p[s|d|h|bh]`` and
   ``*_(mask(z)))_minmax_s[s|d|h]``.
 
+- Supported intrinsics for ``SM4 and AVX10.2``.
+  * Supported SM4 intrinsics of ``_mm512_sm4key4_epi32`` and
+  ``_mm512_sm4rnds4_epi32``.
+
 - All intrinsics in adcintrin.h can now be used in constant expressions.
 
 - All intrinsics in adxintrin.h can now be used in constant expressions.

clang/include/clang/Basic/BuiltinsX86.def

@@ -2179,6 +2179,10 @@ TARGET_BUILTIN(__builtin_ia32_vsm4key4256, "V8UiV8UiV8Ui", "nV:256:", "sm4")
 TARGET_BUILTIN(__builtin_ia32_vsm4rnds4128, "V4UiV4UiV4Ui", "nV:128:", "sm4")
 TARGET_BUILTIN(__builtin_ia32_vsm4rnds4256, "V8UiV8UiV8Ui", "nV:256:", "sm4")
 
+// SM4_EVEX
+TARGET_BUILTIN(__builtin_ia32_vsm4key4512, "V16UiV16UiV16Ui", "nV:512:", "avx10.2-512,sm4")
+TARGET_BUILTIN(__builtin_ia32_vsm4rnds4512, "V16UiV16UiV16Ui", "nV:512:", "avx10.2-512,sm4")
+
 // AVX10 MINMAX
 TARGET_BUILTIN(__builtin_ia32_vminmaxnepbf16128, "V8yV8yV8yIi", "nV:128:", "avx10.2-256")
 TARGET_BUILTIN(__builtin_ia32_vminmaxnepbf16256, "V16yV16yV16yIi", "nV:256:", "avx10.2-256")

clang/lib/Headers/CMakeLists.txt

@@ -243,6 +243,7 @@ set(x86_files
   shaintrin.h
   sm3intrin.h
   sm4intrin.h
+  sm4evexintrin.h
   smmintrin.h
   tbmintrin.h
   tmmintrin.h

clang/lib/Headers/immintrin.h

@@ -677,6 +677,11 @@ _storebe_i64(void * __P, long long __D) {
 #include <avx10_2_512satcvtintrin.h>
 #endif
 
+#if !defined(__SCE__) || __has_feature(modules) ||                             \
+    (defined(__AVX10_2_512__) && defined(__SM4__))
+#include <sm4evexintrin.h>
+#endif
+
 #if !defined(__SCE__) || __has_feature(modules) || defined(__ENQCMD__)
 #include <enqcmdintrin.h>
 #endif

clang/lib/Headers/sm4evexintrin.h

@@ -0,0 +1,32 @@
+/*===--------------- sm4evexintrin.h - SM4 EVEX 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 <sm4evexintrin.h> directly; include <immintrin.h> instead."
+#endif // __IMMINTRIN_H
+
+#ifndef __SM4EVEXINTRIN_H
+#define __SM4EVEXINTRIN_H
+
+#define __DEFAULT_FN_ATTRS512                                                  \
+  __attribute__((__always_inline__, __nodebug__,                               \
+                 __target__("sm4,avx10.2-512"), __min_vector_width__(512)))
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS512
+_mm512_sm4key4_epi32(__m512i __A, __m512i __B) {
+  return (__m512i)__builtin_ia32_vsm4key4512((__v16su)__A, (__v16su)__B);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS512
+_mm512_sm4rnds4_epi32(__m512i __A, __m512i __B) {
+  return (__m512i)__builtin_ia32_vsm4rnds4512((__v16su)__A, (__v16su)__B);
+}
+
+#undef __DEFAULT_FN_ATTRS512
+
+#endif // __SM4EVEXINTRIN_H

clang/test/CodeGen/X86/sm4-evex-builtins.c

@@ -0,0 +1,19 @@
+// RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-- -target-feature +sm4 \
+// RUN: -target-feature +avx10.2-512 -emit-llvm -o - -Wall -Werror | FileCheck %s
+// RUN: %clang_cc1 %s -ffreestanding -triple=i386-- -target-feature +sm4 \
+// RUN: -target-feature +avx10.2-512 -emit-llvm -o - -Wall -Werror | FileCheck %s
+
+#include <immintrin.h>
+#include <stddef.h>
+
+__m512i test_mm512_sm4key4_epi32(__m512i __A, __m512i __B) {
+  // CHECK-LABEL: @test_mm512_sm4key4_epi32(
+  // CHECK: call <16 x i32> @llvm.x86.vsm4key4512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+  return _mm512_sm4key4_epi32(__A, __B);
+}
+
+__m512i test_mm512_sm4rnds4_epi32(__m512i __A, __m512i __B) {
+  // CHECK-LABEL: @test_mm512_sm4rnds4_epi32(
+  // CHECK: call <16 x i32> @llvm.x86.vsm4rnds4512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}})
+  return _mm512_sm4rnds4_epi32(__A, __B);
+}

llvm/docs/ReleaseNotes.md

@@ -219,6 +219,8 @@ Changes to the X86 Backend
 
 * Supported instructions of `MOVRS AND AVX10.2`
 
+* Supported ISA of `SM4(EVEX)`.
+
 Changes to the OCaml bindings
 -----------------------------
 

llvm/include/llvm/IR/IntrinsicsX86.td

@@ -6099,6 +6099,11 @@ let TargetPrefix = "x86" in {
         DefaultAttrsIntrinsic<[llvm_v8i32_ty],
         [llvm_v8i32_ty, llvm_v8i32_ty],
         [IntrNoMem]>;
+  def int_x86_vsm4key4512
+      : ClangBuiltin<"__builtin_ia32_vsm4key4512">,
+        DefaultAttrsIntrinsic<[llvm_v16i32_ty],
+        [llvm_v16i32_ty, llvm_v16i32_ty],
+        [IntrNoMem]>;
   def int_x86_vsm4rnds4128
       : ClangBuiltin<"__builtin_ia32_vsm4rnds4128">,
         DefaultAttrsIntrinsic<[llvm_v4i32_ty],
@@ -6109,6 +6114,11 @@ let TargetPrefix = "x86" in {
         DefaultAttrsIntrinsic<[llvm_v8i32_ty],
         [llvm_v8i32_ty, llvm_v8i32_ty],
         [IntrNoMem]>;
+  def int_x86_vsm4rnds4512
+      : ClangBuiltin<"__builtin_ia32_vsm4rnds4512">,
+        DefaultAttrsIntrinsic<[llvm_v16i32_ty],
+        [llvm_v16i32_ty, llvm_v16i32_ty],
+        [IntrNoMem]>;
 }
 //===----------------------------------------------------------------------===//
 // RAO-INT intrinsics

llvm/lib/Target/X86/X86InstrAVX10.td

@@ -1675,3 +1675,17 @@ defm VMOVRSD : vmovrs_p_vl<0x6f, "vmovrsd", avx512vl_i32_info>,
                           T_MAP5, XS, EVEX_CD8<32, CD8VF>, Sched<[WriteVecLoad]>;
 defm VMOVRSQ : vmovrs_p_vl<0x6f, "vmovrsq", avx512vl_i64_info>,
                           T_MAP5, XS, REX_W, EVEX_CD8<64, CD8VF>, Sched<[WriteVecLoad]>;
+
+// SM4(EVEX)
+multiclass avx10_sm4_base<string OpStr> {
+  // SM4_Base is in X86InstrSSE.td.
+  let Predicates = [HasSM4, HasAVX10_2], AddedComplexity = 1 in {
+    defm Z128 : SM4_Base<OpStr, VR128X, "128", loadv4i32, i128mem>, EVEX_V128;
+    defm Z256 : SM4_Base<OpStr, VR256X, "256", loadv8i32, i256mem>, EVEX_V256;
+  }
+  let Predicates = [HasSM4, HasAVX10_2_512] in
+    defm Z : SM4_Base<OpStr, VR512, "512", loadv16i32, i512mem>, EVEX_V512;
+}
+
+defm VSM4KEY4 : avx10_sm4_base<"vsm4key4">, T8, XS, EVEX, VVVV;
+defm VSM4RNDS4 : avx10_sm4_base<"vsm4rnds4">, T8, XD, EVEX, VVVV;

llvm/test/CodeGen/X86/sm4-evex-intrinsics.ll

@@ -0,0 +1,64 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -verify-machineinstrs -mtriple=x86_64-- --show-mc-encoding -mattr=+sm4,+avx10.2-512 | FileCheck %s
+; RUN: llc < %s -verify-machineinstrs -mtriple=i686-- --show-mc-encoding -mattr=+sm4,+avx10.2-512 | FileCheck %s
+
+define <4 x i32> @test_int_x86_vsm4key4128(<4 x i32> %A, <4 x i32> %B) {
+; CHECK-LABEL: test_int_x86_vsm4key4128:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsm4key4 %xmm1, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7a,0xda,0xc1]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <4 x i32> @llvm.x86.vsm4key4128(<4 x i32> %A, <4 x i32> %B)
+  ret <4 x i32> %ret
+}
+declare <4 x i32> @llvm.x86.vsm4key4128(<4 x i32> %A, <4 x i32> %B)
+
+define <8 x i32> @test_int_x86_vsm4key4256(<8 x i32> %A, <8 x i32> %B) {
+; CHECK-LABEL: test_int_x86_vsm4key4256:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsm4key4 %ymm1, %ymm0, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7e,0xda,0xc1]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <8 x i32> @llvm.x86.vsm4key4256(<8 x i32> %A, <8 x i32> %B)
+  ret <8 x i32> %ret
+}
+declare <8 x i32> @llvm.x86.vsm4key4256(<8 x i32> %A, <8 x i32> %B)
+
+define <16 x i32> @test_int_x86_vsm4key4512(<16 x i32> %A, <16 x i32> %B) {
+; CHECK-LABEL: test_int_x86_vsm4key4512:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsm4key4 %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf2,0x7e,0x48,0xda,0xc1]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <16 x i32> @llvm.x86.vsm4key4512(<16 x i32> %A, <16 x i32> %B)
+  ret <16 x i32> %ret
+}
+declare <16 x i32> @llvm.x86.vsm4key4512(<16 x i32> %A, <16 x i32> %B)
+
+define <4 x i32> @test_int_x86_vsm4rnds4128(<4 x i32> %A, <4 x i32> %B) {
+; CHECK-LABEL: test_int_x86_vsm4rnds4128:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsm4rnds4 %xmm1, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7b,0xda,0xc1]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <4 x i32> @llvm.x86.vsm4rnds4128(<4 x i32> %A, <4 x i32> %B)
+  ret <4 x i32> %ret
+}
+declare <4 x i32> @llvm.x86.vsm4rnds4128(<4 x i32> %A, <4 x i32> %B)
+
+define <8 x i32> @test_int_x86_vsm4rnds4256(<8 x i32> %A, <8 x i32> %B) {
+; CHECK-LABEL: test_int_x86_vsm4rnds4256:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsm4rnds4 %ymm1, %ymm0, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7f,0xda,0xc1]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <8 x i32> @llvm.x86.vsm4rnds4256(<8 x i32> %A, <8 x i32> %B)
+  ret <8 x i32> %ret
+}
+declare <8 x i32> @llvm.x86.vsm4rnds4256(<8 x i32> %A, <8 x i32> %B)
+
+define <16 x i32> @test_int_x86_vsm4rnds4512(<16 x i32> %A, <16 x i32> %B) {
+; CHECK-LABEL: test_int_x86_vsm4rnds4512:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsm4rnds4 %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf2,0x7f,0x48,0xda,0xc1]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %ret = call <16 x i32> @llvm.x86.vsm4rnds4512(<16 x i32> %A, <16 x i32> %B)
+  ret <16 x i32> %ret
+}
+declare <16 x i32> @llvm.x86.vsm4rnds4512(<16 x i32> %A, <16 x i32> %B)
+