86 lines · c
1// REQUIRES: x86-registered-target2// RUN: %clang_cc1 -x c -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +avx512vl -target-feature +avx512fp16 -emit-llvm -o - -Wall -Werror | FileCheck %s --check-prefix=UNCONSTRAINED --check-prefix=COMMON3// RUN: %clang_cc1 -x c -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +avx512vl -target-feature +avx512fp16 -ffp-exception-behavior=maytrap -DSTRICT=1 -emit-llvm -o - -Wall -Werror | FileCheck %s --check-prefix=CONSTRAINED --check-prefix=COMMON4// RUN: %clang_cc1 -x c -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +avx512vl -target-feature +avx512fp16 -S -o - -Wall -Werror | FileCheck %s --check-prefix=CHECK-ASM --check-prefix=COMMON5// RUN: %clang_cc1 -x c -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +avx512vl -target-feature +avx512fp16 -ffp-exception-behavior=maytrap -DSTRICT=1 -S -o - -Wall -Werror | FileCheck %s --check-prefix=CHECK-ASM --check-prefix=COMMON6// RUN: %clang_cc1 -x c++ -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +avx512vl -target-feature +avx512fp16 -emit-llvm -o - -Wall -Werror | FileCheck %s --check-prefix=UNCONSTRAINED --check-prefix=COMMON7// RUN: %clang_cc1 -x c++ -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +avx512vl -target-feature +avx512fp16 -ffp-exception-behavior=maytrap -DSTRICT=1 -emit-llvm -o - -Wall -Werror | FileCheck %s --check-prefix=CONSTRAINED --check-prefix=COMMON8// RUN: %clang_cc1 -x c++ -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +avx512vl -target-feature +avx512fp16 -S -o - -Wall -Werror | FileCheck %s --check-prefix=CHECK-ASM --check-prefix=COMMON9// RUN: %clang_cc1 -x c++ -ffreestanding %s -triple=x86_64-unknown-linux-gnu -target-feature +avx512vl -target-feature +avx512fp16 -ffp-exception-behavior=maytrap -DSTRICT=1 -S -o - -Wall -Werror | FileCheck %s --check-prefix=CHECK-ASM --check-prefix=COMMON10 11#ifdef STRICT12// Test that the constrained intrinsics are picking up the exception13// metadata from the AST instead of the global default from the command line.14 15#pragma float_control(except, on)16#endif17 18 19#include <immintrin.h>20 21__m128h test_mm_sqrt_sh(__m128h x, __m128h y) {22 // COMMON-LABEL: test_mm_sqrt_sh23 // UNCONSTRAINED: call {{.*}}half @llvm.sqrt.f16(half {{.*}})24 // CONSTRAINED: call {{.*}}half @llvm.experimental.constrained.sqrt.f16(half {{.*}}, metadata !{{.*}})25 // CHECK-ASM: vsqrtsh %xmm{{.*}},26 return _mm_sqrt_sh(x, y);27}28 29__m128h test_mm_mask_sqrt_sh(__m128h __W, __mmask8 __U, __m128h __A, __m128h __B){30 // COMMON-LABEL: test_mm_mask_sqrt_sh31 // COMMONIR: extractelement <8 x half> %{{.*}}, i64 032 // UNCONSTRAINED: call {{.*}}half @llvm.sqrt.f16(half %{{.*}})33 // CONSTRAINED: call {{.*}}half @llvm.experimental.constrained.sqrt.f16(half %{{.*}}, metadata !{{.*}})34 // CHECK-ASM: vsqrtsh %xmm{{.*}},35 // COMMONIR-NEXT: extractelement <8 x half> %{{.*}}, i64 036 // COMMONIR-NEXT: bitcast i8 %{{.*}} to <8 x i1>37 // COMMONIR-NEXT: extractelement <8 x i1> %{{.*}}, i64 038 // COMMONIR-NEXT: select i1 {{.*}}, half {{.*}}, half {{.*}}39 // COMMONIR-NEXT: insertelement <8 x half> %{{.*}}, half {{.*}}, i64 040 return _mm_mask_sqrt_sh(__W,__U,__A,__B);41}42 43__m128h test_mm_maskz_sqrt_sh(__mmask8 __U, __m128h __A, __m128h __B){44 // COMMON-LABEL: test_mm_maskz_sqrt_sh45 // COMMONIR: extractelement <2 x half> %{{.*}}, i64 046 // UNCONSTRAINED: call {{.*}}half @llvm.sqrt.f16(half %{{.*}})47 // CONSTRAINED: call {{.*}}half @llvm.experimental.constrained.sqrt.f16(half %{{.*}}, metadata !{{.*}})48 // CHECK-ASM: vsqrtsh %xmm{{.*}},49 // COMMONIR-NEXT: extractelement <2 x half> %{{.*}}, i64 050 // COMMONIR-NEXT: bitcast i8 %{{.*}} to <8 x i1>51 // COMMONIR-NEXT: extractelement <8 x i1> %{{.*}}, i64 052 // COMMONIR-NEXT: select i1 {{.*}}, half {{.*}}, half {{.*}}53 // COMMONIR-NEXT: insertelement <2 x half> %{{.*}}, half {{.*}}, i64 054 return _mm_maskz_sqrt_sh(__U,__A,__B);55}56 57__m512h test_mm512_sqrt_ph(__m512h x) {58 // COMMON-LABEL: test_mm512_sqrt_ph59 // UNCONSTRAINED: call {{.*}}<32 x half> @llvm.sqrt.v32f16(<32 x half> {{.*}})60 // CONSTRAINED: call {{.*}}<32 x half> @llvm.experimental.constrained.sqrt.v32f16(<32 x half> {{.*}}, metadata !{{.*}})61 // CHECK-ASM: vsqrtph %zmm{{.*}},62 return _mm512_sqrt_ph(x);63}64 65__m512h test_mm512_mask_sqrt_ph (__m512h __W, __mmask32 __U, __m512h __A)66{67 // COMMON-LABEL: test_mm512_mask_sqrt_ph68 // UNCONSTRAINED: call {{.*}}<32 x half> @llvm.sqrt.v32f16(<32 x half> %{{.*}})69 // CONSTRAINED: call {{.*}}<32 x half> @llvm.experimental.constrained.sqrt.v32f16(<32 x half> %{{.*}}, metadata !{{.*}})70 // CHECK-ASM: vsqrtph %zmm{{.*}},71 // COMMONIR: bitcast i32 %{{.*}} to <32 x i1>72 // COMMONIR: select <32 x i1> %{{.*}}, <32 x half> %{{.*}}, <32 x half> %{{.*}}73 return _mm512_mask_sqrt_ph (__W,__U,__A);74}75 76__m512h test_mm512_maskz_sqrt_ph (__mmask32 __U, __m512h __A)77{78 // COMMON-LABEL: test_mm512_maskz_sqrt_ph79 // UNCONSTRAINED: call {{.*}}<32 x half> @llvm.sqrt.v32f16(<32 x half> %{{.*}})80 // CONSTRAINED: call {{.*}}<32 x half> @llvm.experimental.constrained.sqrt.v32f16(<32 x half> %{{.*}}, metadata !{{.*}})81 // CHECK-ASM: vsqrtph %zmm{{.*}},82 // COMMONIR: bitcast i32 %{{.*}} to <32 x i1>83 // COMMONIR: select <32 x i1> %{{.*}}, <32 x half> %{{.*}}, <32 x half> {{.*}}84 return _mm512_maskz_sqrt_ph (__U,__A);85}86