[AArch64] Adding Neon Sm3 & Sm4 Intrinsics

[rsanthir.quic]

Summary

This adds SM3 and SM4 Intrinsics support for AArch64, specifically:

vsm3ss1q_u32
vsm3tt1aq_u32
vsm3tt1bq_u32
vsm3tt2aq_u32
vsm3tt2bq_u32
vsm3partw1q_u32
vsm3partw2q_u32
vsm4eq_u32
vsm4ekeyq_u32

[rsanthir.quic]

This is the first in a series of patches that will address the following:
https://bugs.llvm.org/show_bug.cgi?id=47828

[labrinea]

llvm/test/CodeGen/AArch64/neon-sm4-sm3.ll

24

The forth argument (<4 x i32> %d) is redundant.

77

Shouldn't the registers be the other way around: sm4e v0.4s, v1.4s ? I believe the reason this happens is because of how CryptoRRTied is defined in llvm/lib/Target/AArch64/AArch64InstrFormats.td:

class CryptoRRTied<bits<1>op0, bits<2>op1, string asm, string asmops>
  : BaseCryptoV82<(outs V128:$Vd), (ins V128:$Vn, V128:$Vm), asm, asmops,
                  "$Vm = $Vd", []> {

Vd be should be the first source register (as well as destination register) and Vn should be the second source register.

[rsanthir.quic]

Thank you for taking a look at this @labrinea !

llvm/test/CodeGen/AArch64/neon-sm4-sm3.ll

77

I see what you mean, this has the added effect of correcting SHA512SU0 as well

[rsanthir.quic]

Corrected register ordering for sm4e and removed redundant argument in sm3ss1 test

[labrinea]

LGTM, thanks @rsanthir.quic !

[rsanthir.quic]

@t.p.northover Could you take a look before I ask for this to be merged?

[apazos]

clang/test/CodeGen/aarch64-neon-sm4-sm3-invalid.c

9

Consider renaming the test with range-check suffix so it is clear it is checking for invalid range for the immediate
In addition, consider rewriting it with //expected-error checks

clang/test/CodeGen/aarch64-neon-sm4-sm3.c

6

The test is written for aarch64 triple, in addition @t.p.northover might want to add arm64 triple checks. Tim what do you think?

[t.p.northover]

clang/test/CodeGen/aarch64-neon-sm4-sm3.c

6

Thanks for thinking of us, but generally I don't think it's worth duplicating tests along those lines, it's really difficult to write something that cares about the difference between arm64 and aarch64.

linux-gnu vs apple-ios (or others, these days) is more likely to be an issue, but even there I tend to think it's fine to just check one unless there's a specific reason to think it'll affect the targets differently.

[rsanthir.quic]

Updated test name and checks performed

[rsanthir.quic]

@labrinea could you commit this when you have some time please?

clang/include/clang/Basic/arm_neon.td

 
 def SHA1C : SInst<"vsha1c", "..1.", "QUi">;
 def SHA1P : SInst<"vsha1p", "..1.", "QUi">;
 def SHA1M : SInst<"vsha1m", "..1.", "QUi">;
 def SHA1SU0 : SInst<"vsha1su0", "....", "QUi">;
 def SHA256H : SInst<"vsha256h", "....", "QUi">;
 def SHA256H2 : SInst<"vsha256h2", "....", "QUi">;
 def SHA256SU1 : SInst<"vsha256su1", "....", "QUi">;
+
+def SM3SS1 : SInst<"vsm3ss1", "....", "QUi">;
+def SM3TT1A : SInst<"vsm3tt1a", "....I", "QUi">;
+def SM3TT1B : SInst<"vsm3tt1b", "....I", "QUi">;
+def SM3TT2A : SInst<"vsm3tt2a", "....I", "QUi">;
+def SM3TT2B : SInst<"vsm3tt2b", "....I", "QUi">;
+def SM3PARTW1 : SInst<"vsm3partw1", "....", "QUi">;
+def SM3PARTW2 : SInst<"vsm3partw2", "....", "QUi">;
+
+def SM4E : SInst<"vsm4e", "...", "QUi">;
+def SM4EKEY : SInst<"vsm4ekey", "...", "QUi">;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
 // Float -> Int conversions with explicit rounding mode
 
 let ArchGuard = "__ARM_ARCH >= 8" in {
 def FCVTNS_S32 : SInst<"vcvtn_s32", "S.", "fQf">;
 def FCVTNU_S32 : SInst<"vcvtn_u32", "U.", "fQf">;

clang/lib/CodeGen/CGBuiltin.cpp

static const ARMVectorIntrinsicInfo ARMSIMDIntrinsicMap [] = {
   NEONMAP1(vusmmlaq_v, arm_neon_usmmla, 0),
   NEONMAP0(vuzp_v),
   NEONMAP0(vuzpq_v),
   NEONMAP0(vzip_v),
   NEONMAP0(vzipq_v)
 };
 
 static const ARMVectorIntrinsicInfo AArch64SIMDIntrinsicMap[] = {
   NEONMAP1(__a64_vcvtq_low_bf16_v, aarch64_neon_bfcvtn, 0),
   NEONMAP0(splat_lane_v),
   NEONMAP0(splat_laneq_v),
   NEONMAP0(splatq_lane_v),
   NEONMAP0(splatq_laneq_v),
   NEONMAP1(vabs_v, aarch64_neon_abs, 0),
   NEONMAP1(vabsq_v, aarch64_neon_abs, 0),
   NEONMAP0(vaddhn_v),
   NEONMAP1(vaesdq_v, aarch64_crypto_aesd, 0),
static const ARMVectorIntrinsicInfo AArch64SIMDIntrinsicMap[] = {
   NEONMAP0(vshl_n_v),
   NEONMAP2(vshl_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts),
   NEONMAP0(vshll_n_v),
   NEONMAP0(vshlq_n_v),
   NEONMAP2(vshlq_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts),
   NEONMAP0(vshr_n_v),
   NEONMAP0(vshrn_n_v),
   NEONMAP0(vshrq_n_v),
+  NEONMAP1(vsm3partw1q_v, aarch64_crypto_sm3partw1, 0),
+  NEONMAP1(vsm3partw2q_v, aarch64_crypto_sm3partw2, 0),
+  NEONMAP1(vsm3ss1q_v, aarch64_crypto_sm3ss1, 0),
+  NEONMAP1(vsm3tt1aq_v, aarch64_crypto_sm3tt1a, 0),
+  NEONMAP1(vsm3tt1bq_v, aarch64_crypto_sm3tt1b, 0),
+  NEONMAP1(vsm3tt2aq_v, aarch64_crypto_sm3tt2a, 0),
+  NEONMAP1(vsm3tt2bq_v, aarch64_crypto_sm3tt2b, 0),
+  NEONMAP1(vsm4ekeyq_v, aarch64_crypto_sm4ekey, 0),
+  NEONMAP1(vsm4eq_v, aarch64_crypto_sm4e, 0),
   NEONMAP1(vst1_x2_v, aarch64_neon_st1x2, 0),
   NEONMAP1(vst1_x3_v, aarch64_neon_st1x3, 0),
   NEONMAP1(vst1_x4_v, aarch64_neon_st1x4, 0),
   NEONMAP1(vst1q_x2_v, aarch64_neon_st1x2, 0),
   NEONMAP1(vst1q_x3_v, aarch64_neon_st1x3, 0),
   NEONMAP1(vst1q_x4_v, aarch64_neon_st1x4, 0),
   NEONMAP0(vsubhn_v),
   NEONMAP0(vtst_v),
Value *CodeGenFunction::EmitCommonNeonBuiltinExpr(
   case NEON::BI__builtin_neon_vst3_lane_v:
   case NEON::BI__builtin_neon_vst3q_lane_v:
   case NEON::BI__builtin_neon_vst4_lane_v:
   case NEON::BI__builtin_neon_vst4q_lane_v: {
     llvm::Type *Tys[] = {Int8PtrTy, Ty};
     Ops.push_back(getAlignmentValue32(PtrOp0));
     return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "");
   }
+  case NEON::BI__builtin_neon_vsm3partw1q_v:
+  case NEON::BI__builtin_neon_vsm3partw2q_v:
+  case NEON::BI__builtin_neon_vsm3ss1q_v:
+  case NEON::BI__builtin_neon_vsm4ekeyq_v:
+  case NEON::BI__builtin_neon_vsm4eq_v: {
+    Function *F = CGM.getIntrinsic(Int);
+    return EmitNeonCall(F, Ops, "");
+  }
+  case NEON::BI__builtin_neon_vsm3tt1aq_v:
+  case NEON::BI__builtin_neon_vsm3tt1bq_v:
+  case NEON::BI__builtin_neon_vsm3tt2aq_v:
+  case NEON::BI__builtin_neon_vsm3tt2bq_v: {
+    Function *F = CGM.getIntrinsic(Int);
+    Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty);
+    return EmitNeonCall(F, Ops, "");
+  }
   case NEON::BI__builtin_neon_vst1_x2_v:
   case NEON::BI__builtin_neon_vst1q_x2_v:
   case NEON::BI__builtin_neon_vst1_x3_v:
   case NEON::BI__builtin_neon_vst1q_x3_v:
   case NEON::BI__builtin_neon_vst1_x4_v:
   case NEON::BI__builtin_neon_vst1q_x4_v: {
     llvm::Type *PTy = llvm::PointerType::getUnqual(VTy->getElementType());
     // TODO: Currently in AArch32 mode the pointer operand comes first, whereas

clang/test/CodeGen/aarch64-neon-sm4-sm3-invalid.c

+
+// RUN: not %clang_cc1 -triple aarch64-linux-gnu -target-feature +neon \
+// RUN:  -target-feature +crypto \
+// RUN: -S -emit-llvm -o - %s 2>&1 | FileCheck --check-prefix=CHECK-RANGE %s
+
+#include <arm_neon.h>
+
+void test_range_check_vsm3tt1a(uint32x4_t a, uint32x4_t b, uint32x4_t c) {
+  // CHECK-RANGE: error: argument value 5 is outside the valid range [0, 3]
+  vsm3tt1aq_u32(a, b, c, 5);
+  // CHECK-RANGE: error: argument value -1 is outside the valid range [0, 3]
+  vsm3tt1aq_u32(a, b, c, -1);
+}
+
+void test_range_check_vsm3tt1b(uint32x4_t a, uint32x4_t b, uint32x4_t c) {
+  // CHECK-RANGE: error: argument value 5 is outside the valid range [0, 3]
+  vsm3tt1bq_u32(a, b, c, 5);
+  // CHECK-RANGE: error: argument value -1 is outside the valid range [0, 3]
+  vsm3tt1bq_u32(a, b, c, -1);
+}
+
+void test_range_check_vsm3tt2a(uint32x4_t a, uint32x4_t b, uint32x4_t c) {
+  // CHECK-RANGE: error: argument value 5 is outside the valid range [0, 3]
+  vsm3tt2aq_u32(a, b, c, 5);
+  // CHECK-RANGE: error: argument value -1 is outside the valid range [0, 3]
+  vsm3tt2aq_u32(a, b, c, -1);
+}
+
+void test_range_check_vsm3tt2b(uint32x4_t a, uint32x4_t b, uint32x4_t c) {
+  // CHECK-RANGE: error: argument value 5 is outside the valid range [0, 3]
+  vsm3tt2bq_u32(a, b, c, 5);
+  // CHECK-RANGE: error: argument value -1 is outside the valid range [0, 3]
+  vsm3tt2bq_u32(a, b, c, -1);
+}

clang/test/CodeGen/aarch64-neon-sm4-sm3.c

+// RUN: %clang_cc1 -triple aarch64-linux-gnu -target-feature +neon \
+// RUN:  -target-feature +crypto -S -emit-llvm -o - %s \
+// RUN:  | FileCheck %s
+
+// RUN: not %clang_cc1 -triple aarch64-linux-gnu -target-feature +neon \
+// RUN: -S -emit-llvm -o - %s 2>&1 | FileCheck --check-prefix=CHECK-NO-CRYPTO %s
+
+//The front-end only requires the addition of a +crypto in the command line,
+//  however the back-end requires +crypto +sm4 (includes sm4&sm3)
+
+#include <arm_neon.h>
+
+void test_vsm3partw1(uint32x4_t a, uint32x4_t b, uint32x4_t c) {
+  // CHECK-LABEL: @test_vsm3partw1(
+  // CHECK-NO-CRYPTO: warning: implicit declaration of function 'vsm3partw1q_u32' is invalid in C99
+  // CHECK: call <4 x i32> @llvm.aarch64.crypto.sm3partw1
+  uint32x4_t result = vsm3partw1q_u32(a, b, c);
+}
+
+void test_vsm3partw2(uint32x4_t a, uint32x4_t b, uint32x4_t c) {
+  // CHECK-LABEL: @test_vsm3partw2(
+  // CHECK: call <4 x i32> @llvm.aarch64.crypto.sm3partw2
+  uint32x4_t result = vsm3partw2q_u32(a, b, c);
+}
+
+void test_vsm3ss1(uint32x4_t a, uint32x4_t b, uint32x4_t c) {
+  // CHECK-LABEL: @test_vsm3ss1(
+  // CHECK: call <4 x i32> @llvm.aarch64.crypto.sm3ss1
+  uint32x4_t result = vsm3ss1q_u32(a, b, c);
+}
+
+void test_vsm3tt1a(uint32x4_t a, uint32x4_t b, uint32x4_t c) {
+  // CHECK-LABEL: @test_vsm3tt1a(
+  // CHECK: call <4 x i32> @llvm.aarch64.crypto.sm3tt1a
+  uint32x4_t result = vsm3tt1aq_u32(a, b, c, 2);
+}
+
+void test_vsm3tt1b(uint32x4_t a, uint32x4_t b, uint32x4_t c) {
+  // CHECK-LABEL: @test_vsm3tt1b(
+  // CHECK: call <4 x i32> @llvm.aarch64.crypto.sm3tt1b
+  uint32x4_t result = vsm3tt1bq_u32(a, b, c, 2);
+}
+
+void test_vsm3tt2a(uint32x4_t a, uint32x4_t b, uint32x4_t c) {
+  // CHECK-LABEL: @test_vsm3tt2a(
+  // CHECK: call <4 x i32> @llvm.aarch64.crypto.sm3tt2a
+  uint32x4_t result = vsm3tt2aq_u32(a, b, c, 2);
+}
+
+void test_vsm3tt2b(uint32x4_t a, uint32x4_t b, uint32x4_t c) {
+  // CHECK-LABEL: @test_vsm3tt2b(
+  // CHECK: call <4 x i32> @llvm.aarch64.crypto.sm3tt2b
+  uint32x4_t result = vsm3tt2bq_u32(a, b, c, 2);
+}
+
+void test_vsm4e(uint32x4_t a, uint32x4_t b) {
+  // CHECK-LABEL: @test_vsm4e(
+  // CHECK: call <4 x i32> @llvm.aarch64.crypto.sm4e
+  uint32x4_t result = vsm4eq_u32(a, b);
+}
+
+void test_vsm4ekey(uint32x4_t a, uint32x4_t b) {
+  // CHECK-LABEL: @test_vsm4ekey(
+  // CHECK: call  <4 x i32> @llvm.aarch64.crypto.sm4ekey
+  uint32x4_t result = vsm4ekeyq_u32(a, b);
+}

llvm/include/llvm/IR/IntrinsicsAArch64.td

let TargetPrefix = "aarch64" in {
   class Crypto_SHA_12Schedule_Intrinsic
     : DefaultAttrsIntrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty],
                 [IntrNoMem]>;
 
   // SHA intrinsic taking 8 words of the hash and 4 of the schedule.
   class Crypto_SHA_8Hash4Schedule_Intrinsic
     : DefaultAttrsIntrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty],
                 [IntrNoMem]>;
+
+  class Crypto_SM3_3Vector_Intrinsic
+    : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty],
+                [IntrNoMem]>;
+
+  class Crypto_SM3_3VectorIndexed_Intrinsic
+    : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty, llvm_i64_ty],
+                [IntrNoMem, ImmArg<ArgIndex<3>>]>;
+
+  class Crypto_SM4_2Vector_Intrinsic
+    : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty], [IntrNoMem]>;
 }
 
 // AES
 def int_aarch64_crypto_aese   : Crypto_AES_DataKey_Intrinsic;
 def int_aarch64_crypto_aesd   : Crypto_AES_DataKey_Intrinsic;
 def int_aarch64_crypto_aesmc  : Crypto_AES_Data_Intrinsic;
 def int_aarch64_crypto_aesimc : Crypto_AES_Data_Intrinsic;
 
 // SHA1
 def int_aarch64_crypto_sha1c  : Crypto_SHA_5Hash4Schedule_Intrinsic;
 def int_aarch64_crypto_sha1p  : Crypto_SHA_5Hash4Schedule_Intrinsic;
 def int_aarch64_crypto_sha1m  : Crypto_SHA_5Hash4Schedule_Intrinsic;
 def int_aarch64_crypto_sha1h  : Crypto_SHA_1Hash_Intrinsic;
 
 def int_aarch64_crypto_sha1su0 : Crypto_SHA_12Schedule_Intrinsic;
 def int_aarch64_crypto_sha1su1 : Crypto_SHA_8Schedule_Intrinsic;
 
 // SHA256
 def int_aarch64_crypto_sha256h   : Crypto_SHA_8Hash4Schedule_Intrinsic;
 def int_aarch64_crypto_sha256h2  : Crypto_SHA_8Hash4Schedule_Intrinsic;
 def int_aarch64_crypto_sha256su0 : Crypto_SHA_8Schedule_Intrinsic;
 def int_aarch64_crypto_sha256su1 : Crypto_SHA_12Schedule_Intrinsic;
 
+//SM3 & SM4
+def int_aarch64_crypto_sm3partw1 : Crypto_SM3_3Vector_Intrinsic;
+def int_aarch64_crypto_sm3partw2 : Crypto_SM3_3Vector_Intrinsic;
+def int_aarch64_crypto_sm3ss1    : Crypto_SM3_3Vector_Intrinsic;
+def int_aarch64_crypto_sm3tt1a   : Crypto_SM3_3VectorIndexed_Intrinsic;
+def int_aarch64_crypto_sm3tt1b   : Crypto_SM3_3VectorIndexed_Intrinsic;
+def int_aarch64_crypto_sm3tt2a   : Crypto_SM3_3VectorIndexed_Intrinsic;
+def int_aarch64_crypto_sm3tt2b   : Crypto_SM3_3VectorIndexed_Intrinsic;
+def int_aarch64_crypto_sm4e      : Crypto_SM4_2Vector_Intrinsic;
+def int_aarch64_crypto_sm4ekey   : Crypto_SM4_2Vector_Intrinsic;
+
 //===----------------------------------------------------------------------===//
 // CRC32
 
 let TargetPrefix = "aarch64" in {
 
 def int_aarch64_crc32b  : DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
     [IntrNoMem]>;
 def int_aarch64_crc32cb : DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],

llvm/lib/Target/AArch64/AArch64InstrInfo.td

 def SM3TT1B   : CryptoRRRi2Tied<0b0, 0b01, "sm3tt1b">;
 def SM3TT2A   : CryptoRRRi2Tied<0b0, 0b10, "sm3tt2a">;
 def SM3TT2B   : CryptoRRRi2Tied<0b0, 0b11, "sm3tt2b">;
 def SM3SS1    : CryptoRRRR_4S<0b10, "sm3ss1">;
 def SM3PARTW1 : CryptoRRRTied_4S<0b1, 0b00, "sm3partw1">;
 def SM3PARTW2 : CryptoRRRTied_4S<0b1, 0b01, "sm3partw2">;
 def SM4ENCKEY : CryptoRRR_4S<0b1, 0b10, "sm4ekey">;
 def SM4E      : CryptoRRTied_4S<0b0, 0b01, "sm4e">;
+
+def : Pat<(v4i32 (int_aarch64_crypto_sm3ss1 (v4i32 V128:$Vn), (v4i32 V128:$Vm), (v4i32 V128:$Va))),
+          (SM3SS1 (v4i32 V128:$Vn), (v4i32 V128:$Vm), (v4i32 V128:$Va))>;
+
+class SM3PARTW_pattern<Instruction INST, Intrinsic OpNode>
+  : Pat<(v4i32 (OpNode (v4i32 V128:$Vd), (v4i32 V128:$Vn), (v4i32 V128:$Vm))),
+        (INST (v4i32 V128:$Vd), (v4i32 V128:$Vn), (v4i32 V128:$Vm))>;
+
+class SM3TT_pattern<Instruction INST, Intrinsic OpNode>
+  : Pat<(v4i32 (OpNode (v4i32 V128:$Vd), (v4i32 V128:$Vn), (v4i32 V128:$Vm), (i64 VectorIndexS_timm:$imm) )),
+        (INST (v4i32 V128:$Vd), (v4i32 V128:$Vn), (v4i32 V128:$Vm), (VectorIndexS_timm:$imm))>;
+
+class SM4_pattern<Instruction INST, Intrinsic OpNode>
+  : Pat<(v4i32 (OpNode (v4i32 V128:$Vn), (v4i32 V128:$Vm))),
+        (INST (v4i32 V128:$Vn), (v4i32 V128:$Vm))>;
+
+def : SM3PARTW_pattern<SM3PARTW1, int_aarch64_crypto_sm3partw1>;
+def : SM3PARTW_pattern<SM3PARTW2, int_aarch64_crypto_sm3partw2>;
+
+def : SM3TT_pattern<SM3TT1A, int_aarch64_crypto_sm3tt1a>;
+def : SM3TT_pattern<SM3TT1B, int_aarch64_crypto_sm3tt1b>;
+def : SM3TT_pattern<SM3TT2A, int_aarch64_crypto_sm3tt2a>;
+def : SM3TT_pattern<SM3TT2B, int_aarch64_crypto_sm3tt2b>;
+
+def : SM4_pattern<SM4ENCKEY, int_aarch64_crypto_sm4ekey>;
+def : SM4_pattern<SM4E, int_aarch64_crypto_sm4e>;
 } // HasSM4
 
 let Predicates = [HasRCPC] in {
   // v8.3 Release Consistent Processor Consistent support, optional in v8.2.
   def LDAPRB  : RCPCLoad<0b00, "ldaprb", GPR32>;
   def LDAPRH  : RCPCLoad<0b01, "ldaprh", GPR32>;
   def LDAPRW  : RCPCLoad<0b10, "ldapr", GPR32>;
   def LDAPRX  : RCPCLoad<0b11, "ldapr", GPR64>;

llvm/test/CodeGen/AArch64/neon-sm4-sm3.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc %s -mtriple=aarch64 -mattr=+v8.3a,+sm4 -o - | FileCheck %s
+
+define <4 x i32> @test_vsm3partw1(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c) {
+; CHECK-LABEL: test_vsm3partw1:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    sm3partw1 v0.4s, v1.4s, v2.4s
+; CHECK-NEXT:    ret
+entry:
+  %vsm3partw1.i = tail call <4 x i32> @llvm.aarch64.crypto.sm3partw1(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c)
+  ret <4 x i32> %vsm3partw1.i
+}
+
+define <4 x i32> @test_vsm3partw2(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c) {
+; CHECK-LABEL: test_vsm3partw2:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    sm3partw2 v0.4s, v1.4s, v2.4s
+; CHECK-NEXT:    ret
+entry:
+  %vsm3partw2.i = tail call <4 x i32> @llvm.aarch64.crypto.sm3partw2(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c)
+  ret <4 x i32> %vsm3partw2.i
+}
+
+define <4 x i32> @test_vsm3ss1(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c, <4 x i32> %d) {
+; CHECK-LABEL: test_vsm3ss1:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    sm3ss1 v0.4s, v0.4s, v1.4s, v2.4s
+; CHECK-NEXT:    ret
+entry:
+  %vsm3ss1.i = tail call <4 x i32> @llvm.aarch64.crypto.sm3ss1(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c)
+  ret <4 x i32> %vsm3ss1.i
+}
+
+define <4 x i32> @test_vsm3tt1a(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c) {
+; CHECK-LABEL: test_vsm3tt1a:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    sm3tt1a v0.4s, v1.4s, v2.s[2]
+; CHECK-NEXT:    ret
+entry:
+  %vsm3tt1a.i = tail call <4 x i32> @llvm.aarch64.crypto.sm3tt1a(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c, i64 2)
+  ret <4 x i32> %vsm3tt1a.i
+}
+
+define <4 x i32> @test_vsm3tt1b(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c) {
+; CHECK-LABEL: test_vsm3tt1b:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    sm3tt1b v0.4s, v1.4s, v2.s[2]
+; CHECK-NEXT:    ret
+entry:
+  %vsm3tt1b.i = tail call <4 x i32> @llvm.aarch64.crypto.sm3tt1b(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c, i64 2)
+  ret <4 x i32> %vsm3tt1b.i
+}
+
+define <4 x i32> @test_vsm3tt2a(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c) {
+; CHECK-LABEL: test_vsm3tt2a:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    sm3tt2a v0.4s, v1.4s, v2.s[2]
+; CHECK-NEXT:    ret
+entry:
+  %vsm3tt2a.i = tail call <4 x i32> @llvm.aarch64.crypto.sm3tt2a(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c, i64 2)
+  ret <4 x i32> %vsm3tt2a.i
+}
+
+define <4 x i32> @test_vsm3tt2b(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c) {
+; CHECK-LABEL: test_vsm3tt2b:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    sm3tt2b v0.4s, v1.4s, v2.s[2]
+; CHECK-NEXT:    ret
+entry:
+  %vsm3tt2b.i = tail call <4 x i32> @llvm.aarch64.crypto.sm3tt2b(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c, i64 2)
+  ret <4 x i32> %vsm3tt2b.i
+}
+
+define <4 x i32> @test_vsm4e(<4 x i32> %a, <4 x i32> %b) {
+; CHECK-LABEL: test_vsm4e:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    sm4e v1.4s, v0.4s
+; CHECK-NEXT:    mov v0.16b, v1.16b
+; CHECK-NEXT:    ret
+entry:
+  %vsm4e.i = tail call <4 x i32> @llvm.aarch64.crypto.sm4e(<4 x i32> %a, <4 x i32> %b)
+  ret <4 x i32> %vsm4e.i
+}
+
+define <4 x i32> @test_vsm4ekey(<4 x i32> %a, <4 x i32> %b) {
+; CHECK-LABEL: test_vsm4ekey:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    sm4ekey v0.4s, v0.4s, v1.4s
+; CHECK-NEXT:    ret
+entry:
+  %vsm4ekey.i = tail call <4 x i32> @llvm.aarch64.crypto.sm4ekey(<4 x i32> %a, <4 x i32> %b)
+  ret <4 x i32> %vsm4ekey.i
+}
+
+declare <4 x i32> @llvm.aarch64.crypto.sm3partw1(<4 x i32>, <4 x i32>, <4 x i32>)
+declare <4 x i32> @llvm.aarch64.crypto.sm3partw2(<4 x i32>, <4 x i32>, <4 x i32>)
+declare <4 x i32> @llvm.aarch64.crypto.sm3ss1(<4 x i32>, <4 x i32>, <4 x i32>)
+declare <4 x i32> @llvm.aarch64.crypto.sm3tt1a(<4 x i32>, <4 x i32>, <4 x i32>, i64 immarg)
+declare <4 x i32> @llvm.aarch64.crypto.sm3tt2b(<4 x i32>, <4 x i32>, <4 x i32>, i64 immarg)
+declare <4 x i32> @llvm.aarch64.crypto.sm3tt2a(<4 x i32>, <4 x i32>, <4 x i32>, i64 immarg)
+declare <4 x i32> @llvm.aarch64.crypto.sm3tt1b(<4 x i32>, <4 x i32>, <4 x i32>, i64 immarg)
+declare <4 x i32> @llvm.aarch64.crypto.sm4e(<4 x i32>, <4 x i32>)
+declare <4 x i32> @llvm.aarch64.crypto.sm4ekey(<4 x i32>, <4 x i32>)