[X86] Add kadd intrinsics to match gcc and icc.

This adds the following intrinsics: _kadd_mask64 _kadd_mask32 _kadd_mask16 _kadd_mask8 These are missing from the Intel Intrinsics Guide, but are implemented by both gcc and icc. git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@340879 91177308-0d34-0410-b5e6-96231b3b80d8
vsujan · Aug 28, 2018 · 1d39f2e · 1d39f2e
1 parent bc0fa1d
commit 1d39f2e
Show file tree

Hide file tree

Showing 6 changed files with 101 additions and 1 deletion.
diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
@@ -1737,6 +1737,10 @@ TARGET_BUILTIN(__builtin_ia32_fpclassps512_mask, "UsV16fIiUs", "ncV:512:", "avx5
 TARGET_BUILTIN(__builtin_ia32_fpclasspd512_mask, "UcV8dIiUc", "ncV:512:", "avx512dq")
 TARGET_BUILTIN(__builtin_ia32_fpclasssd_mask, "UcV2dIiUc", "ncV:128:", "avx512dq")
 TARGET_BUILTIN(__builtin_ia32_fpclassss_mask, "UcV4fIiUc", "ncV:128:", "avx512dq")
+TARGET_BUILTIN(__builtin_ia32_kaddqi, "UcUcUc", "nc", "avx512dq")
+TARGET_BUILTIN(__builtin_ia32_kaddhi, "UsUsUs", "nc", "avx512dq")
+TARGET_BUILTIN(__builtin_ia32_kaddsi, "UiUiUi", "nc", "avx512bw")
+TARGET_BUILTIN(__builtin_ia32_kadddi, "ULLiULLiULLi", "nc", "avx512bw")
 TARGET_BUILTIN(__builtin_ia32_kandqi, "UcUcUc", "nc", "avx512dq")
 TARGET_BUILTIN(__builtin_ia32_kandhi, "UsUsUs", "nc", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_kandsi, "UiUiUi", "nc", "avx512bw")

diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
@@ -8613,7 +8613,7 @@ static Value *EmitX86MaskLogic(CodeGenFunction &CGF, Instruction::BinaryOps Opc,
     LHS = CGF.Builder.CreateNot(LHS);
 
   return CGF.Builder.CreateBitCast(CGF.Builder.CreateBinOp(Opc, LHS, RHS),
-                                  CGF.Builder.getIntNTy(std::max(NumElts, 8U)));
+                                   Ops[0]->getType());
 }
 
 static Value *EmitX86Select(CodeGenFunction &CGF,
@@ -10031,6 +10031,34 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
     return Builder.CreateZExt(Cmp, ConvertType(E->getType()));
   }
 
+  case X86::BI__builtin_ia32_kaddqi:
+  case X86::BI__builtin_ia32_kaddhi:
+  case X86::BI__builtin_ia32_kaddsi:
+  case X86::BI__builtin_ia32_kadddi: {
+    Intrinsic::ID IID;
+    switch (BuiltinID) {
+    default: llvm_unreachable("Unsupported intrinsic!");
+    case X86::BI__builtin_ia32_kaddqi:
+      IID = Intrinsic::x86_avx512_kadd_b;
+      break;
+    case X86::BI__builtin_ia32_kaddhi:
+      IID = Intrinsic::x86_avx512_kadd_w;
+      break;
+    case X86::BI__builtin_ia32_kaddsi:
+      IID = Intrinsic::x86_avx512_kadd_d;
+      break;
+    case X86::BI__builtin_ia32_kadddi:
+      IID = Intrinsic::x86_avx512_kadd_q;
+      break;
+    }
+
+    unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
+    Value *LHS = getMaskVecValue(*this, Ops[0], NumElts);
+    Value *RHS = getMaskVecValue(*this, Ops[1], NumElts);
+    Function *Intr = CGM.getIntrinsic(IID);
+    Value *Res = Builder.CreateCall(Intr, {LHS, RHS});
+    return Builder.CreateBitCast(Res, Ops[0]->getType());
+  }
   case X86::BI__builtin_ia32_kandqi:
   case X86::BI__builtin_ia32_kandhi:
   case X86::BI__builtin_ia32_kandsi:

diff --git a/lib/Headers/avx512bwintrin.h b/lib/Headers/avx512bwintrin.h
@@ -143,6 +143,18 @@ _kortest_mask64_u8(__mmask64 __A, __mmask64 __B, unsigned char *__C) {
   return (unsigned char)__builtin_ia32_kortestzdi(__A, __B);
 }
 
+static __inline__ __mmask32 __DEFAULT_FN_ATTRS
+_kadd_mask32(__mmask32 __A, __mmask32 __B)
+{
+  return (__mmask32)__builtin_ia32_kaddsi((__mmask32)__A, (__mmask32)__B);
+}
+
+static __inline__ __mmask64 __DEFAULT_FN_ATTRS
+_kadd_mask64(__mmask64 __A, __mmask64 __B)
+{
+  return (__mmask64)__builtin_ia32_kadddi((__mmask64)__A, (__mmask64)__B);
+}
+
 /* Integer compare */
 
 #define _mm512_cmp_epi8_mask(a, b, p) \

diff --git a/lib/Headers/avx512dqintrin.h b/lib/Headers/avx512dqintrin.h
@@ -86,6 +86,18 @@ _kortest_mask8_u8(__mmask8 __A, __mmask8 __B, unsigned char *__C) {
   return (unsigned char)__builtin_ia32_kortestzqi(__A, __B);
 }
 
+static __inline__ __mmask8 __DEFAULT_FN_ATTRS
+_kadd_mask8(__mmask8 __A, __mmask8 __B)
+{
+  return (__mmask8)__builtin_ia32_kaddqi((__mmask8)__A, (__mmask8)__B);
+}
+
+static __inline__ __mmask16 __DEFAULT_FN_ATTRS
+_kadd_mask16(__mmask16 __A, __mmask16 __B)
+{
+  return (__mmask16)__builtin_ia32_kaddhi((__mmask16)__A, (__mmask16)__B);
+}
+
 static __inline__ __m512i __DEFAULT_FN_ATTRS512
 _mm512_mullo_epi64 (__m512i __A, __m512i __B) {
   return (__m512i) ((__v8du) __A * (__v8du) __B);

diff --git a/test/CodeGen/avx512bw-builtins.c b/test/CodeGen/avx512bw-builtins.c
@@ -226,6 +226,28 @@ unsigned char test_kortest_mask64_u8(__m512i __A, __m512i __B, __m512i __C, __m5
                             _mm512_cmpneq_epu8_mask(__C, __D), CF);
 }
 
+__mmask32 test_kadd_mask32(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
+  // CHECK-LABEL: @test_kadd_mask32
+  // CHECK: [[LHS:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i32 %{{.*}} to <32 x i1>
+  // CHECK: [[RES:%.*]] = call <32 x i1> @llvm.x86.avx512.kadd.d(<32 x i1> [[LHS]], <32 x i1> [[RHS]])
+  // CHECK: bitcast <32 x i1> [[RES]] to i32
+  return _mm512_mask_cmpneq_epu16_mask(_kadd_mask32(_mm512_cmpneq_epu16_mask(__A, __B),
+                                                    _mm512_cmpneq_epu16_mask(__C, __D)),
+                                                    __E, __F);
+}
+
+__mmask64 test_kadd_mask64(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
+  // CHECK-LABEL: @test_kadd_mask64
+  // CHECK: [[LHS:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i64 %{{.*}} to <64 x i1>
+  // CHECK: [[RES:%.*]] = call <64 x i1> @llvm.x86.avx512.kadd.q(<64 x i1> [[LHS]], <64 x i1> [[RHS]])
+  // CHECK: bitcast <64 x i1> [[RES]] to i64
+  return _mm512_mask_cmpneq_epu8_mask(_kadd_mask64(_mm512_cmpneq_epu8_mask(__A, __B),
+                                                   _mm512_cmpneq_epu8_mask(__C, __D)),
+                                                   __E, __F);
+}
+
 __mmask64 test_mm512_cmpeq_epi8_mask(__m512i __a, __m512i __b) {
   // CHECK-LABEL: @test_mm512_cmpeq_epi8_mask
   // CHECK: icmp eq <64 x i8> %{{.*}}, %{{.*}}

diff --git a/test/CodeGen/avx512dq-builtins.c b/test/CodeGen/avx512dq-builtins.c
@@ -114,6 +114,28 @@ unsigned char test_kortest_mask8_u8(__m512i __A, __m512i __B, __m512i __C, __m51
                            _mm512_cmpneq_epu64_mask(__C, __D), CF);
 }
 
+__mmask8 test_kadd_mask8(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
+  // CHECK-LABEL: @test_kadd_mask8
+  // CHECK: [[LHS:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i8 %{{.*}} to <8 x i1>
+  // CHECK: [[RES:%.*]] = call <8 x i1> @llvm.x86.avx512.kadd.b(<8 x i1> [[LHS]], <8 x i1> [[RHS]])
+  // CHECK: bitcast <8 x i1> [[RES]] to i8
+  return _mm512_mask_cmpneq_epu64_mask(_kadd_mask8(_mm512_cmpneq_epu64_mask(__A, __B),
+                                                   _mm512_cmpneq_epu64_mask(__C, __D)),
+                                                   __E, __F);
+}
+
+__mmask16 test_kadd_mask16(__m512i __A, __m512i __B, __m512i __C, __m512i __D, __m512i __E, __m512i __F) {
+  // CHECK-LABEL: @test_kadd_mask16
+  // CHECK: [[LHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
+  // CHECK: [[RHS:%.*]] = bitcast i16 %{{.*}} to <16 x i1>
+  // CHECK: [[RES:%.*]] = call <16 x i1> @llvm.x86.avx512.kadd.w(<16 x i1> [[LHS]], <16 x i1> [[RHS]])
+  // CHECK: bitcast <16 x i1> [[RES]] to i16
+  return _mm512_mask_cmpneq_epu32_mask(_kadd_mask16(_mm512_cmpneq_epu32_mask(__A, __B),
+                                                    _mm512_cmpneq_epu32_mask(__C, __D)),
+                                                    __E, __F);
+}
+
 __m512i test_mm512_mullo_epi64 (__m512i __A, __m512i __B) {
   // CHECK-LABEL: @test_mm512_mullo_epi64
   // CHECK: mul <8 x i64>