clang/test/CodeGen/X86/avx512dq-builtins-constrained.c

   1 // REQUIRES: x86-registered-target
   2 // RUN: %clang_cc1 -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512dq -emit-llvm -o - -Wall -Werror | FileCheck %s --check-prefix=UNCONSTRAINED --check-prefix=COMMON --check-prefix=COMMONIR
   3 // RUN: %clang_cc1 -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512dq -ffp-exception-behavior=maytrap -DSTRICT=1 -emit-llvm -o - -Wall -Werror | FileCheck %s --check-prefix=CONSTRAINED --check-prefix=COMMON --check-prefix=COMMONIR --implicit-check-not=fpexcept.maytrap
   4 // RUN: %clang_cc1 -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512dq -S -o - -Wall -Werror | FileCheck %s --check-prefix=CHECK-ASM --check-prefix=COMMON
   5 // RUN: %clang_cc1 -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-apple-darwin -target-feature +avx512dq -ffp-exception-behavior=maytrap -DSTRICT=1 -S -o - -Wall -Werror | FileCheck %s --check-prefix=CHECK-ASM --check-prefix=COMMON --implicit-check-not=fpexcept.maytrap
   6
   7 // Any cases of "fpexcept.maytrap" in this test are clang bugs.
   8
   9 #ifdef STRICT
  10 // Test that the constrained intrinsics are picking up the exception
  11 // metadata from the AST instead of the global default from the command line.
  12
  13 #pragma float_control(except, on)
  14 #endif
  15
  16
  17 #include <immintrin.h>
  18
  19 __m512d test_mm512_cvtepi64_pd(__m512i __A) {
  20   // COMMON-LABEL: test_mm512_cvtepi64_pd
  21   // UNCONSTRAINED: sitofp <8 x i64> %{{.*}} to <8 x double>
  22   // CONSTRAINED: call <8 x double> @llvm.experimental.constrained.sitofp.v8f64.v8i64(<8 x i64> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
  23   // CHECK-ASM: vcvtqq2pd
  24   return _mm512_cvtepi64_pd(__A);
  25 }
  26
  27 __m512d test_mm512_mask_cvtepi64_pd(__m512d __W, __mmask8 __U, __m512i __A) {
  28   // COMMON-LABEL: test_mm512_mask_cvtepi64_pd
  29   // UNCONSTRAINED: sitofp <8 x i64> %{{.*}} to <8 x double>
  30   // CONSTRAINED: call <8 x double> @llvm.experimental.constrained.sitofp.v8f64.v8i64(<8 x i64> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
  31   // COMMONIR: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
  32   // CHECK-ASM: vcvtqq2pd
  33   return _mm512_mask_cvtepi64_pd(__W, __U, __A);
  34 }
  35
  36 __m512d test_mm512_maskz_cvtepi64_pd(__mmask8 __U, __m512i __A) {
  37   // COMMON-LABEL: test_mm512_maskz_cvtepi64_pd
  38   // UNCONSTRAINED: sitofp <8 x i64> %{{.*}} to <8 x double>
  39   // CONSTRAINED: call <8 x double> @llvm.experimental.constrained.sitofp.v8f64.v8i64(<8 x i64> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
  40   // COMMONIR: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
  41   // CHECK-ASM: vcvtqq2pd
  42   return _mm512_maskz_cvtepi64_pd(__U, __A);
  43 }
  44
  45 __m512d test_mm512_cvt_roundepi64_pd(__m512i __A) {
  46   // COMMON-LABEL: test_mm512_cvt_roundepi64_pd
  47   // COMMONIR: @llvm.x86.avx512.sitofp.round.v8f64.v8i64
  48   // CHECK-ASM: vcvtqq2pd
  49   return _mm512_cvt_roundepi64_pd(__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
  50 }
  51
  52 __m512d test_mm512_mask_cvt_roundepi64_pd(__m512d __W, __mmask8 __U, __m512i __A) {
  53   // COMMON-LABEL: test_mm512_mask_cvt_roundepi64_pd
  54   // COMMONIR: @llvm.x86.avx512.sitofp.round.v8f64.v8i64
  55   // COMMONIR: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
  56   // CHECK-ASM: vcvtqq2pd
  57   return _mm512_mask_cvt_roundepi64_pd(__W, __U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
  58 }
  59
  60 __m512d test_mm512_maskz_cvt_roundepi64_pd(__mmask8 __U, __m512i __A) {
  61   // COMMON-LABEL: test_mm512_maskz_cvt_roundepi64_pd
  62   // COMMONIR: @llvm.x86.avx512.sitofp.round.v8f64.v8i64
  63   // COMMONIR: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
  64   // CHECK-ASM: vcvtqq2pd
  65   return _mm512_maskz_cvt_roundepi64_pd(__U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
  66 }
  67
  68 __m256 test_mm512_cvtepi64_ps(__m512i __A) {
  69   // COMMON-LABEL: test_mm512_cvtepi64_ps
  70   // UNCONSTRAINED: sitofp <8 x i64> %{{.*}} to <8 x float>
  71   // CONSTRAINED: call <8 x float> @llvm.experimental.constrained.sitofp.v8f32.v8i64(<8 x i64> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
  72   // CHECK-ASM: vcvtqq2ps
  73   return _mm512_cvtepi64_ps(__A);
  74 }
  75
  76 __m256 test_mm512_mask_cvtepi64_ps(__m256 __W, __mmask8 __U, __m512i __A) {
  77   // COMMON-LABEL: test_mm512_mask_cvtepi64_ps
  78   // UNCONSTRAINED: sitofp <8 x i64> %{{.*}} to <8 x float>
  79   // CONSTRAINED: call <8 x float> @llvm.experimental.constrained.sitofp.v8f32.v8i64(<8 x i64> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
  80   // COMMONIR: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
  81   // CHECK-ASM: vcvtqq2ps
  82   return _mm512_mask_cvtepi64_ps(__W, __U, __A);
  83 }
  84
  85 __m256 test_mm512_maskz_cvtepi64_ps(__mmask8 __U, __m512i __A) {
  86   // COMMON-LABEL: test_mm512_maskz_cvtepi64_ps
  87   // UNCONSTRAINED: sitofp <8 x i64> %{{.*}} to <8 x float>
  88   // CONSTRAINED: call <8 x float> @llvm.experimental.constrained.sitofp.v8f32.v8i64(<8 x i64> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
  89   // COMMONIR: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
  90   // CHECK-ASM: vcvtqq2ps
  91   return _mm512_maskz_cvtepi64_ps(__U, __A);
  92 }
  93
  94 __m256 test_mm512_cvt_roundepi64_ps(__m512i __A) {
  95   // COMMON-LABEL: test_mm512_cvt_roundepi64_ps
  96   // COMMONIR: @llvm.x86.avx512.sitofp.round.v8f32.v8i64
  97   // CHECK-ASM: vcvtqq2ps
  98   return _mm512_cvt_roundepi64_ps(__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
  99 }
 100
 101 __m256 test_mm512_mask_cvt_roundepi64_ps(__m256 __W, __mmask8 __U, __m512i __A) {
 102   // COMMON-LABEL: test_mm512_mask_cvt_roundepi64_ps
 103   // COMMONIR: @llvm.x86.avx512.sitofp.round.v8f32.v8i64
 104   // COMMONIR: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
 105   // CHECK-ASM: vcvtqq2ps
 106   return _mm512_mask_cvt_roundepi64_ps(__W, __U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 107 }
 108
 109 __m256 test_mm512_maskz_cvt_roundepi64_ps(__mmask8 __U, __m512i __A) {
 110   // COMMON-LABEL: test_mm512_maskz_cvt_roundepi64_ps
 111   // COMMONIR: @llvm.x86.avx512.sitofp.round.v8f32.v8i64
 112   // COMMONIR: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
 113   // CHECK-ASM: vcvtqq2ps
 114   return _mm512_maskz_cvt_roundepi64_ps(__U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 115 }
 116
 117 __m512d test_mm512_cvtepu64_pd(__m512i __A) {
 118   // COMMON-LABEL: test_mm512_cvtepu64_pd
 119   // UNCONSTRAINED: uitofp <8 x i64> %{{.*}} to <8 x double>
 120   // CONSTRAINED: call <8 x double> @llvm.experimental.constrained.uitofp.v8f64.v8i64(<8 x i64> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 121   // CHECK-ASM: vcvtuqq2pd
 122   return _mm512_cvtepu64_pd(__A);
 123 }
 124
 125 __m512d test_mm512_mask_cvtepu64_pd(__m512d __W, __mmask8 __U, __m512i __A) {
 126   // COMMON-LABEL: test_mm512_mask_cvtepu64_pd
 127   // UNCONSTRAINED: uitofp <8 x i64> %{{.*}} to <8 x double>
 128   // CONSTRAINED: call <8 x double> @llvm.experimental.constrained.uitofp.v8f64.v8i64(<8 x i64> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 129   // COMMONIR: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
 130   // CHECK-ASM: vcvtuqq2pd
 131   return _mm512_mask_cvtepu64_pd(__W, __U, __A);
 132 }
 133
 134 __m512d test_mm512_maskz_cvtepu64_pd(__mmask8 __U, __m512i __A) {
 135   // COMMON-LABEL: test_mm512_maskz_cvtepu64_pd
 136   // UNCONSTRAINED: uitofp <8 x i64> %{{.*}} to <8 x double>
 137   // CONSTRAINED: call <8 x double> @llvm.experimental.constrained.uitofp.v8f64.v8i64(<8 x i64> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 138   // COMMONIR: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
 139   // CHECK-ASM: vcvtuqq2pd
 140   return _mm512_maskz_cvtepu64_pd(__U, __A);
 141 }
 142
 143 __m512d test_mm512_cvt_roundepu64_pd(__m512i __A) {
 144   // COMMON-LABEL: test_mm512_cvt_roundepu64_pd
 145   // COMMONIR: @llvm.x86.avx512.uitofp.round.v8f64.v8i64
 146   // CHECK-ASM: vcvtuqq2pd
 147   return _mm512_cvt_roundepu64_pd(__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 148 }
 149
 150 __m512d test_mm512_mask_cvt_roundepu64_pd(__m512d __W, __mmask8 __U, __m512i __A) {
 151   // COMMON-LABEL: test_mm512_mask_cvt_roundepu64_pd
 152   // COMMONIR: @llvm.x86.avx512.uitofp.round.v8f64.v8i64
 153   // COMMONIR: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
 154   // CHECK-ASM: vcvtuqq2pd
 155   return _mm512_mask_cvt_roundepu64_pd(__W, __U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 156 }
 157
 158 __m512d test_mm512_maskz_cvt_roundepu64_pd(__mmask8 __U, __m512i __A) {
 159   // COMMON-LABEL: test_mm512_maskz_cvt_roundepu64_pd
 160   // COMMONIR: @llvm.x86.avx512.uitofp.round.v8f64.v8i64
 161   // COMMONIR: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
 162   // CHECK-ASM: vcvtuqq2pd
 163   return _mm512_maskz_cvt_roundepu64_pd(__U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 164 }
 165
 166 __m256 test_mm512_cvtepu64_ps(__m512i __A) {
 167   // COMMON-LABEL: test_mm512_cvtepu64_ps
 168   // UNCONSTRAINED: uitofp <8 x i64> %{{.*}} to <8 x float>
 169   // CONSTRAINED: call <8 x float> @llvm.experimental.constrained.uitofp.v8f32.v8i64(<8 x i64> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 170   // CHECK-ASM: vcvtuqq2ps
 171   return _mm512_cvtepu64_ps(__A);
 172 }
 173
 174 __m256 test_mm512_mask_cvtepu64_ps(__m256 __W, __mmask8 __U, __m512i __A) {
 175   // COMMON-LABEL: test_mm512_mask_cvtepu64_ps
 176   // UNCONSTRAINED: uitofp <8 x i64> %{{.*}} to <8 x float>
 177   // CONSTRAINED: call <8 x float> @llvm.experimental.constrained.uitofp.v8f32.v8i64(<8 x i64> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 178   // COMMONIR: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
 179   // CHECK-ASM: vcvtuqq2ps
 180   return _mm512_mask_cvtepu64_ps(__W, __U, __A);
 181 }
 182
 183 __m256 test_mm512_maskz_cvtepu64_ps(__mmask8 __U, __m512i __A) {
 184   // COMMON-LABEL: test_mm512_maskz_cvtepu64_ps
 185   // UNCONSTRAINED: uitofp <8 x i64> %{{.*}} to <8 x float>
 186   // CONSTRAINED: call <8 x float> @llvm.experimental.constrained.uitofp.v8f32.v8i64(<8 x i64> %{{.*}}, metadata !"round.tonearest", metadata !"fpexcept.strict")
 187   // COMMONIR: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
 188   // CHECK-ASM: vcvtuqq2ps
 189   return _mm512_maskz_cvtepu64_ps(__U, __A);
 190 }
 191
 192 __m256 test_mm512_cvt_roundepu64_ps(__m512i __A) {
 193   // COMMON-LABEL: test_mm512_cvt_roundepu64_ps
 194   // COMMONIR: @llvm.x86.avx512.uitofp.round.v8f32.v8i64
 195   // CHECK-ASM: vcvtuqq2ps
 196   return _mm512_cvt_roundepu64_ps(__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 197 }
 198
 199 __m256 test_mm512_mask_cvt_roundepu64_ps(__m256 __W, __mmask8 __U, __m512i __A) {
 200   // COMMON-LABEL: test_mm512_mask_cvt_roundepu64_ps
 201   // COMMONIR: @llvm.x86.avx512.uitofp.round.v8f32.v8i64
 202   // COMMONIR: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
 203   // CHECK-ASM: vcvtuqq2ps
 204   return _mm512_mask_cvt_roundepu64_ps(__W, __U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 205 }
 206
 207 __m256 test_mm512_maskz_cvt_roundepu64_ps(__mmask8 __U, __m512i __A) {
 208   // COMMON-LABEL: test_mm512_maskz_cvt_roundepu64_ps
 209   // COMMONIR: @llvm.x86.avx512.uitofp.round.v8f32.v8i64
 210   // COMMONIR: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
 211   // CHECK-ASM: vcvtuqq2ps
 212   return _mm512_maskz_cvt_roundepu64_ps(__U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 213 }
 214