Skip to content

[Headers][X86] Allow AVX512 reduction intrinsics to be used in constexpr - #152363

Merged
RKSimon merged 4 commits into
llvm:mainfrom
pedroclobo:x86-constexpr-avx512
Aug 7, 2025
Merged

[Headers][X86] Allow AVX512 reduction intrinsics to be used in constexpr#152363
RKSimon merged 4 commits into
llvm:mainfrom
pedroclobo:x86-constexpr-avx512

Conversation

@pedroclobo

@pedroclobo pedroclobo commented Aug 6, 2025

Copy link
Copy Markdown
Member

Closes #152324.
Part of #30794.

This PR adds constexpr support for the following AVX512 integer reduction intrinsics:

  • _mm512_reduce_add_epi32
  • _mm512_reduce_add_epi64
  • _mm512_reduce_mul_epi32
  • _mm512_reduce_mul_epi64
  • _mm512_reduce_and_epi32
  • _mm512_reduce_and_epi64
  • _mm512_reduce_or_epi32
  • _mm512_reduce_or_epi64
  • _mm512_reduce_max_epi32
  • _mm512_reduce_max_epi64
  • _mm512_reduce_min_epi32
  • _mm512_reduce_min_epi64
  • _mm512_reduce_max_epu32
  • _mm512_reduce_max_epu64
  • _mm512_reduce_min_epu32
  • _mm512_reduce_min_epu64

@pedroclobo
[Headers][X86] Allow AVX512 reduction intrinsics to be used in constexpr
b3c0c47 
This PR adds `constexpr` support for the following AVX512 integer
reduction intrinsics:

- `_mm512_reduce_add_epi32`
- `_mm512_reduce_add_epi64`
- `_mm512_reduce_mul_epi32`
- `_mm512_reduce_mul_epi64`
- `_mm512_reduce_and_epi32`
- `_mm512_reduce_and_epi64`
- `_mm512_reduce_or_epi32`
- `_mm512_reduce_or_epi64`
- `_mm512_reduce_max_epi32`
- `_mm512_reduce_max_epi64`
- `_mm512_reduce_min_epi32`
- `_mm512_reduce_min_epi64`
- `_mm512_reduce_max_epu32`
- `_mm512_reduce_max_epu64`
- `_mm512_reduce_min_epu32`
- `_mm512_reduce_min_epu64`
@llvmbot llvmbot added clang Clang issues not falling into any other category backend:X86 clang:headers Headers provided by Clang, e.g. for intrinsics labels Aug 6, 2025
@llvmbot

llvmbot commented Aug 6, 2025

Copy link
Copy Markdown
Member

@llvm/pr-subscribers-clang

Author: Pedro Lobo (pedroclobo)

Changes

Closes #152324.
Part of #30794.

This PR adds constexpr support for the following AVX512 integer reduction intrinsics:

  • _mm512_reduce_add_epi32
  • _mm512_reduce_add_epi64
  • _mm512_reduce_mul_epi32
  • _mm512_reduce_mul_epi64
  • _mm512_reduce_and_epi32
  • _mm512_reduce_and_epi64
  • _mm512_reduce_or_epi32
  • _mm512_reduce_or_epi64
  • _mm512_reduce_max_epi32
  • _mm512_reduce_max_epi64
  • _mm512_reduce_min_epi32
  • _mm512_reduce_min_epi64
  • _mm512_reduce_max_epu32
  • _mm512_reduce_max_epu64
  • _mm512_reduce_min_epu32
  • _mm512_reduce_min_epu64

Full diff: https://github.com/llvm/llvm-project/pull/152363.diff

1 Files Affected:

  • (modified) clang/lib/Headers/avx512fintrin.h (+16-16)
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 45e7eeb5327d0..2f218150ca867 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -9337,19 +9337,19 @@ _mm512_mask_abs_pd(__m512d __W, __mmask8 __K, __m512d __A)
  * This takes log2(n) steps where n is the number of elements in the vector.
  */
 
-static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_add_epi64(__m512i __W) {
+static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_add_epi64(__m512i __W) {
   return __builtin_reduce_add((__v8di)__W);
 }
 
-static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_mul_epi64(__m512i __W) {
+static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_mul_epi64(__m512i __W) {
   return __builtin_reduce_mul((__v8di)__W);
 }
 
-static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_and_epi64(__m512i __W) {
+static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_and_epi64(__m512i __W) {
   return __builtin_reduce_and((__v8di)__W);
 }
 
-static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_or_epi64(__m512i __W) {
+static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_or_epi64(__m512i __W) {
   return __builtin_reduce_or((__v8di)__W);
 }
 
@@ -9400,22 +9400,22 @@ _mm512_mask_reduce_mul_pd(__mmask8 __M, __m512d __W) {
   return __builtin_ia32_reduce_fmul_pd512(1.0, __W);
 }
 
-static __inline__ int __DEFAULT_FN_ATTRS512
+static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_add_epi32(__m512i __W) {
   return __builtin_reduce_add((__v16si)__W);
 }
 
-static __inline__ int __DEFAULT_FN_ATTRS512
+static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_mul_epi32(__m512i __W) {
   return __builtin_reduce_mul((__v16si)__W);
 }
 
-static __inline__ int __DEFAULT_FN_ATTRS512
+static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_and_epi32(__m512i __W) {
   return __builtin_reduce_and((__v16si)__W);
 }
 
-static __inline__ int __DEFAULT_FN_ATTRS512
+static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_or_epi32(__m512i __W) {
   return __builtin_reduce_or((__v16si)__W);
 }
@@ -9466,22 +9466,22 @@ _mm512_mask_reduce_mul_ps(__mmask16 __M, __m512 __W) {
   return __builtin_ia32_reduce_fmul_ps512(1.0f, __W);
 }
 
-static __inline__ long long __DEFAULT_FN_ATTRS512
+static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_max_epi64(__m512i __V) {
   return __builtin_reduce_max((__v8di)__V);
 }
 
-static __inline__ unsigned long long __DEFAULT_FN_ATTRS512
+static __inline__ unsigned long long __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_max_epu64(__m512i __V) {
   return __builtin_reduce_max((__v8du)__V);
 }
 
-static __inline__ long long __DEFAULT_FN_ATTRS512
+static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_min_epi64(__m512i __V) {
   return __builtin_reduce_min((__v8di)__V);
 }
 
-static __inline__ unsigned long long __DEFAULT_FN_ATTRS512
+static __inline__ unsigned long long __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_min_epu64(__m512i __V) {
   return __builtin_reduce_min((__v8du)__V);
 }
@@ -9509,22 +9509,22 @@ _mm512_mask_reduce_min_epu64(__mmask8 __M, __m512i __V) {
   __V = _mm512_mask_mov_epi64(_mm512_set1_epi64(-1LL), __M, __V);
   return __builtin_reduce_min((__v8du)__V);
 }
-static __inline__ int __DEFAULT_FN_ATTRS512
+static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_max_epi32(__m512i __V) {
   return __builtin_reduce_max((__v16si)__V);
 }
 
-static __inline__ unsigned int __DEFAULT_FN_ATTRS512
+static __inline__ unsigned int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_max_epu32(__m512i __V) {
   return __builtin_reduce_max((__v16su)__V);
 }
 
-static __inline__ int __DEFAULT_FN_ATTRS512
+static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_min_epi32(__m512i __V) {
   return __builtin_reduce_min((__v16si)__V);
 }
 
-static __inline__ unsigned int __DEFAULT_FN_ATTRS512
+static __inline__ unsigned int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_min_epu32(__m512i __V) {
   return __builtin_reduce_min((__v16su)__V);
 }

@llvmbot

llvmbot commented Aug 6, 2025

Copy link
Copy Markdown
Member

@llvm/pr-subscribers-backend-x86

Author: Pedro Lobo (pedroclobo)

Changes

Closes #152324.
Part of #30794.

This PR adds constexpr support for the following AVX512 integer reduction intrinsics:

  • _mm512_reduce_add_epi32
  • _mm512_reduce_add_epi64
  • _mm512_reduce_mul_epi32
  • _mm512_reduce_mul_epi64
  • _mm512_reduce_and_epi32
  • _mm512_reduce_and_epi64
  • _mm512_reduce_or_epi32
  • _mm512_reduce_or_epi64
  • _mm512_reduce_max_epi32
  • _mm512_reduce_max_epi64
  • _mm512_reduce_min_epi32
  • _mm512_reduce_min_epi64
  • _mm512_reduce_max_epu32
  • _mm512_reduce_max_epu64
  • _mm512_reduce_min_epu32
  • _mm512_reduce_min_epu64

Full diff: https://github.com/llvm/llvm-project/pull/152363.diff

1 Files Affected:

  • (modified) clang/lib/Headers/avx512fintrin.h (+16-16)
diff --git a/clang/lib/Headers/avx512fintrin.h b/clang/lib/Headers/avx512fintrin.h
index 45e7eeb5327d0..2f218150ca867 100644
--- a/clang/lib/Headers/avx512fintrin.h
+++ b/clang/lib/Headers/avx512fintrin.h
@@ -9337,19 +9337,19 @@ _mm512_mask_abs_pd(__m512d __W, __mmask8 __K, __m512d __A)
  * This takes log2(n) steps where n is the number of elements in the vector.
  */
 
-static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_add_epi64(__m512i __W) {
+static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_add_epi64(__m512i __W) {
   return __builtin_reduce_add((__v8di)__W);
 }
 
-static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_mul_epi64(__m512i __W) {
+static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_mul_epi64(__m512i __W) {
   return __builtin_reduce_mul((__v8di)__W);
 }
 
-static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_and_epi64(__m512i __W) {
+static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_and_epi64(__m512i __W) {
   return __builtin_reduce_and((__v8di)__W);
 }
 
-static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_or_epi64(__m512i __W) {
+static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR _mm512_reduce_or_epi64(__m512i __W) {
   return __builtin_reduce_or((__v8di)__W);
 }
 
@@ -9400,22 +9400,22 @@ _mm512_mask_reduce_mul_pd(__mmask8 __M, __m512d __W) {
   return __builtin_ia32_reduce_fmul_pd512(1.0, __W);
 }
 
-static __inline__ int __DEFAULT_FN_ATTRS512
+static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_add_epi32(__m512i __W) {
   return __builtin_reduce_add((__v16si)__W);
 }
 
-static __inline__ int __DEFAULT_FN_ATTRS512
+static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_mul_epi32(__m512i __W) {
   return __builtin_reduce_mul((__v16si)__W);
 }
 
-static __inline__ int __DEFAULT_FN_ATTRS512
+static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_and_epi32(__m512i __W) {
   return __builtin_reduce_and((__v16si)__W);
 }
 
-static __inline__ int __DEFAULT_FN_ATTRS512
+static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_or_epi32(__m512i __W) {
   return __builtin_reduce_or((__v16si)__W);
 }
@@ -9466,22 +9466,22 @@ _mm512_mask_reduce_mul_ps(__mmask16 __M, __m512 __W) {
   return __builtin_ia32_reduce_fmul_ps512(1.0f, __W);
 }
 
-static __inline__ long long __DEFAULT_FN_ATTRS512
+static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_max_epi64(__m512i __V) {
   return __builtin_reduce_max((__v8di)__V);
 }
 
-static __inline__ unsigned long long __DEFAULT_FN_ATTRS512
+static __inline__ unsigned long long __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_max_epu64(__m512i __V) {
   return __builtin_reduce_max((__v8du)__V);
 }
 
-static __inline__ long long __DEFAULT_FN_ATTRS512
+static __inline__ long long __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_min_epi64(__m512i __V) {
   return __builtin_reduce_min((__v8di)__V);
 }
 
-static __inline__ unsigned long long __DEFAULT_FN_ATTRS512
+static __inline__ unsigned long long __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_min_epu64(__m512i __V) {
   return __builtin_reduce_min((__v8du)__V);
 }
@@ -9509,22 +9509,22 @@ _mm512_mask_reduce_min_epu64(__mmask8 __M, __m512i __V) {
   __V = _mm512_mask_mov_epi64(_mm512_set1_epi64(-1LL), __M, __V);
   return __builtin_reduce_min((__v8du)__V);
 }
-static __inline__ int __DEFAULT_FN_ATTRS512
+static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_max_epi32(__m512i __V) {
   return __builtin_reduce_max((__v16si)__V);
 }
 
-static __inline__ unsigned int __DEFAULT_FN_ATTRS512
+static __inline__ unsigned int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_max_epu32(__m512i __V) {
   return __builtin_reduce_max((__v16su)__V);
 }
 
-static __inline__ int __DEFAULT_FN_ATTRS512
+static __inline__ int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_min_epi32(__m512i __V) {
   return __builtin_reduce_min((__v16si)__V);
 }
 
-static __inline__ unsigned int __DEFAULT_FN_ATTRS512
+static __inline__ unsigned int __DEFAULT_FN_ATTRS512_CONSTEXPR
 _mm512_reduce_min_epu32(__m512i __V) {
   return __builtin_reduce_min((__v16su)__V);
 }

@github-actions

github-actions Bot commented Aug 6, 2025 •
edited
Loading

Copy link
Copy Markdown

✅ With the latest revision this PR passed the C/C++ code formatter.

RKSimon
RKSimon reviewed Aug 6, 2025
View reviewed changes

@RKSimon RKSimon left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please can you add constexpr test coverage to avx512f-builtins.c

@pedroclobo

pedroclobo commented Aug 7, 2025

Copy link
Copy Markdown
Member Author

Please can you add constexpr test coverage to avx512f-builtins.c

I took the liberty of adding the constexpr tests alongside each intrinsic's lit test.
Let me know if you'd really prefer to have them in avx512f-builtins.c instead.

@pedroclobo
pedroclobo requested a review from RKSimon August 7, 2025 00:19
RKSimon
RKSimon approved these changes Aug 7, 2025
View reviewed changes

@RKSimon RKSimon left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM - cheers!

@RKSimon
RKSimon merged commit 5805e88 into llvm:main Aug 7, 2025
@pedroclobo
pedroclobo deleted the x86-constexpr-avx512 branch August 7, 2025 10:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@RKSimon RKSimon RKSimon approved these changes

Assignees

No one assigned

Labels

backend:X86 clang:headers Headers provided by Clang, e.g. for intrinsics clang Clang issues not falling into any other category

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

[Headers][X86] Allow AVX512 integer reduction intrinsics to be used in constexpr

3 participants

@pedroclobo @llvmbot @RKSimon