Skip to content

Conversation

@RKSimon
Copy link
Collaborator

@RKSimon RKSimon commented Nov 14, 2025

_mm_sqrt_sh / _mm512_sqrt_ph - these were missed from #167692

_mm_sqrt_sh / _mm512_sqrt_ph - these were missed from llvm#167692
@RKSimon RKSimon enabled auto-merge (squash) November 14, 2025 11:29
@llvmbot llvmbot added clang Clang issues not falling into any other category backend:X86 labels Nov 14, 2025
@llvmbot
Copy link
Member

llvmbot commented Nov 14, 2025

@llvm/pr-subscribers-clang

@llvm/pr-subscribers-backend-x86

Author: Simon Pilgrim (RKSimon)

Changes

_mm_sqrt_sh / _mm512_sqrt_ph - these were missed from #167692


Full diff: https://github.com/llvm/llvm-project/pull/168046.diff

1 Files Affected:

  • (added) clang/test/CodeGen/X86/avx512fp16-builtins-constrained.c (+85)
diff --git a/clang/test/CodeGen/X86/avx512fp16-builtins-constrained.c b/clang/test/CodeGen/X86/avx512fp16-builtins-constrained.c
new file mode 100644
index 0000000000000..cb029ccdcbf59
--- /dev/null
+++ b/clang/test/CodeGen/X86/avx512fp16-builtins-constrained.c
@@ -0,0 +1,85 @@
+// REQUIRES: x86-registered-target
+// 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=COMMON
+// 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=COMMON
+// 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=COMMON
+// 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=COMMON
+// 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=COMMON
+// 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=COMMON
+// 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=COMMON
+// 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=COMMON
+
+#ifdef STRICT
+// Test that the constrained intrinsics are picking up the exception
+// metadata from the AST instead of the global default from the command line.
+
+#pragma float_control(except, on)
+#endif
+
+
+#include <immintrin.h>
+
+__m128h test_mm_sqrt_sh(__m128h x, __m128h y) {
+  // COMMON-LABEL: test_mm_sqrt_sh
+  // UNCONSTRAINED: call {{.*}}half @llvm.sqrt.f16(half {{.*}})
+  // CONSTRAINED: call {{.*}}half @llvm.experimental.constrained.sqrt.f16(half {{.*}}, metadata !{{.*}})
+  // CHECK-ASM: vsqrtsh %xmm{{.*}},
+  return _mm_sqrt_sh(x, y);
+}
+
+__m128h test_mm_mask_sqrt_sh(__m128h __W, __mmask8 __U, __m128h __A, __m128h __B){
+  // COMMON-LABEL: test_mm_mask_sqrt_sh
+  // COMMONIR: extractelement <8 x half> %{{.*}}, i64 0
+  // UNCONSTRAINED: call {{.*}}half @llvm.sqrt.f16(half %{{.*}})
+  // CONSTRAINED: call {{.*}}half @llvm.experimental.constrained.sqrt.f16(half %{{.*}}, metadata !{{.*}})
+  // CHECK-ASM: vsqrtsh %xmm{{.*}},
+  // COMMONIR-NEXT: extractelement <8 x half> %{{.*}}, i64 0
+  // COMMONIR-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // COMMONIR-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // COMMONIR-NEXT: select i1 {{.*}}, half {{.*}}, half {{.*}}
+  // COMMONIR-NEXT: insertelement <8 x half> %{{.*}}, half {{.*}}, i64 0
+  return _mm_mask_sqrt_sh(__W,__U,__A,__B);
+}
+
+__m128h test_mm_maskz_sqrt_sh(__mmask8 __U, __m128h __A, __m128h __B){
+  // COMMON-LABEL: test_mm_maskz_sqrt_sh
+  // COMMONIR: extractelement <2 x half> %{{.*}}, i64 0
+  // UNCONSTRAINED: call {{.*}}half @llvm.sqrt.f16(half %{{.*}})
+  // CONSTRAINED: call {{.*}}half @llvm.experimental.constrained.sqrt.f16(half %{{.*}}, metadata !{{.*}})
+  // CHECK-ASM: vsqrtsh %xmm{{.*}},
+  // COMMONIR-NEXT: extractelement <2 x half> %{{.*}}, i64 0
+  // COMMONIR-NEXT: bitcast i8 %{{.*}} to <8 x i1>
+  // COMMONIR-NEXT: extractelement <8 x i1> %{{.*}}, i64 0
+  // COMMONIR-NEXT: select i1 {{.*}}, half {{.*}}, half {{.*}}
+  // COMMONIR-NEXT: insertelement <2 x half> %{{.*}}, half {{.*}}, i64 0
+  return _mm_maskz_sqrt_sh(__U,__A,__B);
+}
+
+__m512h test_mm512_sqrt_ph(__m512h x) {
+  // COMMON-LABEL: test_mm512_sqrt_ph
+  // UNCONSTRAINED: call {{.*}}<32 x half> @llvm.sqrt.v32f16(<32 x half> {{.*}})
+  // CONSTRAINED: call {{.*}}<32 x half> @llvm.experimental.constrained.sqrt.v32f16(<32 x half> {{.*}}, metadata !{{.*}})
+  // CHECK-ASM: vsqrtph %zmm{{.*}},
+  return _mm512_sqrt_ph(x);
+}
+
+__m512h test_mm512_mask_sqrt_ph (__m512h __W, __mmask32 __U, __m512h __A)
+{
+  // COMMON-LABEL: test_mm512_mask_sqrt_ph
+  // UNCONSTRAINED: call <32 x half> @llvm.sqrt.v32f16(<32 x half> %{{.*}})
+  // CONSTRAINED: call <32 x half> @llvm.experimental.constrained.sqrt.v32f16(<32 x half> %{{.*}}, metadata !{{.*}})
+  // CHECK-ASM: vsqrtph %zmm{{.*}},
+  // COMMONIR: bitcast i32 %{{.*}} to <32 x i1>
+  // COMMONIR: select <32 x i1> %{{.*}}, <32 x half> %{{.*}}, <32 x half> %{{.*}}
+  return _mm512_mask_sqrt_ph (__W,__U,__A);
+}
+
+__m512h test_mm512_maskz_sqrt_ph (__mmask32 __U, __m512h __A)
+{
+  // COMMON-LABEL: test_mm512_maskz_sqrt_ph
+  // UNCONSTRAINED: call <32 x half> @llvm.sqrt.v32f16(<32 x half> %{{.*}})
+  // CONSTRAINED: call <32 x half> @llvm.experimental.constrained.sqrt.v32f16(<32 x half> %{{.*}}, metadata !{{.*}})
+  // CHECK-ASM: vsqrtph %zmm{{.*}},
+  // COMMONIR: bitcast i32 %{{.*}} to <32 x i1>
+  // COMMONIR: select <32 x i1> %{{.*}}, <32 x half> %{{.*}}, <32 x half> {{.*}}
+  return _mm512_maskz_sqrt_ph (__U,__A);
+}

@RKSimon RKSimon merged commit 5829bb9 into llvm:main Nov 14, 2025
10 of 13 checks passed
@RKSimon RKSimon deleted the x86-sqrtfp16-constrained-tests branch November 14, 2025 12:15
@llvm-ci
Copy link
Collaborator

llvm-ci commented Nov 14, 2025

LLVM Buildbot has detected a new failure on builder ppc64le-flang-rhel-clang running on ppc64le-flang-rhel-test while building clang at step 3 "clean-build-dir".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/157/builds/42209

Here is the relevant piece of the build log for the reference
Step 3 (clean-build-dir) failure: Delete failed. (failure) (timed out)
Step 6 (test-build-unified-tree-check-flang) failure: 1200 seconds without output running [b'ninja', b'check-flang'], attempting to kill
...
PASS: Flang :: Transforms/OpenMP/lower-workdistribute-doloop.mlir (3989 of 3999)
PASS: Flang :: Parser/OpenMP/reduction-modifier.f90 (3990 of 3999)
PASS: Flang :: Lower/bbcnull.f90 (3991 of 3999)
PASS: Flang :: Driver/mcmodel.f90 (3992 of 3999)
PASS: Flang :: Fir/target-rewrite-triple.fir (3993 of 3999)
PASS: Flang :: Driver/darwin-version.f90 (3994 of 3999)
PASS: Flang :: Fir/target-rewrite-complex.fir (3995 of 3999)
PASS: Flang :: Driver/aarch64-sve-vector-bits.f90 (3996 of 3999)
PASS: Flang :: Driver/color-diagnostics-forwarding.f90 (3997 of 3999)
PASS: Flang :: Transforms/debug-dwarf-version.fir (3998 of 3999)
command timed out: 1200 seconds without output running [b'ninja', b'check-flang'], attempting to kill
process killed by signal 9
program finished with exit code -1
elapsedTime=1909.917873

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

backend:X86 clang Clang issues not falling into any other category

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants