Skip to content

Apply upstream Eigen commits e246f9cb and 43a01f06 #152

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
eddelbuettel merged 2 commits into from
Oct 7, 2026
+111 −12
Merged

Apply upstream Eigen commits e246f9cb and 43a01f06 #152

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
872bc3a
Apply upstream Eigen commits e246f9cb and 43a01f06
SteveBronder Oct 7, 2026
5bd787e
Merge branch 'fix/stanheaders' into patch/stanheaders
eddelbuettel Oct 7, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 10 additions & 1 deletion ChangeLog
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,10 +1,19 @@

2026-10-07 Steve Bronder <sbronder@flatironinstitute.org>

* inst/include/Eigen/src/Core/arch/AVX/TypeCasting.h: Apply upstream
commit 43a01f06 to support gcc < 10.1 and clang < 10
* inst/include/Eigen/src/Core/arch/AVX512/PacketMath.h: Idem
* inst/include/Eigen/src/Core/arch/AVX512/TypeCasting.h: Idem

2026-10-06 Dirk Eddelbuettel <edd@debian.org>

* patches/eigen-5.0.1-stan.diff: Document previous commit as diff

2026-10-01 Dirk Eddelbuettel <edd@debian.org>

* inst/include/Eigen/src/Core/Fill.h: Apply patch to aid StanHeaders
* inst/include/Eigen/src/Core/Fill.h: Apply upstream commit e246f9cb
to use memset only if !NumTraits<Scalar>::RequireInitialization

2026-07-27 Dirk Eddelbuettel <edd@debian.org>

Expand Down
40 changes: 37 additions & 3 deletions inst/include/Eigen/src/Core/arch/AVX/TypeCasting.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,6 +56,40 @@ struct type_casting_traits<int64_t, double> : vectorized_type_casting_traits<int
#endif
#endif

EIGEN_STRONG_INLINE __m256 _eigen_mm256_set_m128(__m128 hi, __m128 lo) {
#if EIGEN_COMP_GNUC && (EIGEN_COMP_CLANG < 1000 || EIGEN_COMP_GNUC < 810)
__m256 result = _mm256_castps128_ps256(lo);
return _mm256_insertf128_ps(result, hi, 1);
#else
return _mm256_set_m128(hi, lo);
#endif
}

EIGEN_STRONG_INLINE __m256d _eigen_mm256_set_m128d(__m128d hi, __m128d lo) {
#if EIGEN_COMP_GNUC && (EIGEN_COMP_CLANG < 1000 || EIGEN_COMP_GNUC < 810)
__m256d result = _mm256_castpd128_pd256(lo);
return _mm256_insertf128_pd(result, hi, 1);
#else
return _mm256_set_m128d(hi, lo);
#endif
}

EIGEN_STRONG_INLINE __m256i _eigen_mm256_set_m128i(__m128i hi, __m128i lo) {
#if EIGEN_COMP_GNUC && (EIGEN_COMP_CLANG < 1000 || EIGEN_COMP_GNUC < 810)
#if defined(EIGEN_VECTORIZE_AVX2)
__m256i result = _mm256_castsi128_si256(lo);
return _mm256_inserti128_si256(result, hi, 1);
#else
EIGEN_ALIGN32 int32_t tmp[8];
_mm_storeu_si128(reinterpret_cast<__m128i*>(tmp), lo);
_mm_storeu_si128(reinterpret_cast<__m128i*>(tmp + 4), hi);
return _mm256_loadu_si256(reinterpret_cast<const __m256i*>(tmp));
#endif
#else
return _mm256_set_m128i(hi, lo);
#endif
}

template <>
EIGEN_STRONG_INLINE Packet16b pcast<Packet8f, Packet16b>(const Packet8f& a, const Packet8f& b) {
__m256 nonzero_a = _mm256_cmp_ps(a, pzero(a), _CMP_NEQ_UQ);
Expand Down Expand Up @@ -109,7 +143,7 @@ EIGEN_STRONG_INLINE Packet8i pcast<Packet8f, Packet8i>(const Packet8f& a) {

template <>
EIGEN_STRONG_INLINE Packet8i pcast<Packet4d, Packet8i>(const Packet4d& a, const Packet4d& b) {
return _mm256_set_m128i(_mm256_cvttpd_epi32(b), _mm256_cvttpd_epi32(a));
return _eigen_mm256_set_m128i(_mm256_cvttpd_epi32(b), _mm256_cvttpd_epi32(a));
}

template <>
Expand All @@ -124,7 +158,7 @@ EIGEN_STRONG_INLINE Packet8f pcast<Packet8i, Packet8f>(const Packet8i& a) {

template <>
EIGEN_STRONG_INLINE Packet8f pcast<Packet4d, Packet8f>(const Packet4d& a, const Packet4d& b) {
return _mm256_set_m128(_mm256_cvtpd_ps(b), _mm256_cvtpd_ps(a));
return _eigen_mm256_set_m128(_mm256_cvtpd_ps(b), _mm256_cvtpd_ps(a));
}

template <>
Expand Down Expand Up @@ -249,7 +283,7 @@ EIGEN_STRONG_INLINE Packet4d pcast<Packet4l, Packet4d>(const Packet4l& a) {

template <>
EIGEN_STRONG_INLINE Packet4d pcast<Packet2l, Packet4d>(const Packet2l& a, const Packet2l& b) {
return _mm256_set_m128d((pcast<Packet2l, Packet2d>(b)), (pcast<Packet2l, Packet2d>(a)));
return _eigen_mm256_set_m128d((pcast<Packet2l, Packet2d>(b)), (pcast<Packet2l, Packet2d>(a)));
}

template <>
Expand Down
62 changes: 55 additions & 7 deletions inst/include/Eigen/src/Core/arch/AVX512/PacketMath.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,54 @@ typedef eigen_packet_wrapper<__m512i, 6> Packet32s;
typedef eigen_packet_wrapper<__m256i, 6> Packet16s;
typedef eigen_packet_wrapper<__m128i, 6> Packet8s;

EIGEN_STRONG_INLINE Packet16i _eigen_mm512_loadu_epi32(const int* from) {
#if EIGEN_COMP_GNUC && (EIGEN_COMP_CLANG < 1000 || EIGEN_COMP_GNUC < 1010)
return _mm512_loadu_si512(reinterpret_cast<const void*>(from));
#else
return _mm512_loadu_epi32(from);
#endif
}

EIGEN_STRONG_INLINE Packet16i _eigen_mm512_loadu_epi64(const int64_t* from) {
#if EIGEN_COMP_GNUC && (EIGEN_COMP_CLANG < 1000 || EIGEN_COMP_GNUC < 1010)
return _mm512_loadu_si512(reinterpret_cast<const void*>(from));
#else
return _mm512_loadu_epi64(from);
#endif
}

EIGEN_STRONG_INLINE void _eigen_mm512_storeu_epi32(void* to, const Packet16i& from) {
#if EIGEN_COMP_GNUC && (EIGEN_COMP_CLANG < 1000 || EIGEN_COMP_GNUC < 1010)
_mm512_storeu_si512(to, from);
#else
_mm512_storeu_epi32(to, from);
#endif
}

EIGEN_STRONG_INLINE void _eigen_mm512_storeu_epi64(void* to, const Packet16i& from) {
#if EIGEN_COMP_GNUC && (EIGEN_COMP_CLANG < 1000 || EIGEN_COMP_GNUC < 1010)
_mm512_storeu_si512(to, from);
#else
_mm512_storeu_epi64(to, from);
#endif
}

EIGEN_STRONG_INLINE void _eigen_mm256_storeu_epi32(void* to, const __m256i& from) {
#if EIGEN_COMP_GNUC && (EIGEN_COMP_CLANG < 1000 || EIGEN_COMP_GNUC < 1010)
_mm256_storeu_si256(reinterpret_cast<__m256i*>(to), from);
#else
_mm256_storeu_epi32(to, from);
#endif
}

EIGEN_STRONG_INLINE void _eigen_mm_storeu_epi32(void* to, const __m128i& from) {
#if EIGEN_COMP_GNUC && (EIGEN_COMP_CLANG < 1000 || EIGEN_COMP_GNUC < 1010)
_mm_storeu_si128(reinterpret_cast<__m128i*>(to), from);
#else
_mm_storeu_epi32(to, from);
#endif
}

template <>
struct is_arithmetic<__m512> {
enum { value = true };
Expand Down Expand Up @@ -1033,11 +1081,11 @@ EIGEN_STRONG_INLINE Packet8d ploadu<Packet8d>(const double* from) {
}
template <>
EIGEN_STRONG_INLINE Packet16i ploadu<Packet16i>(const int* from) {
EIGEN_DEBUG_UNALIGNED_LOAD return _mm512_loadu_epi32(from);
EIGEN_DEBUG_UNALIGNED_LOAD return _eigen_mm512_loadu_epi32(from);
}
template <>
EIGEN_STRONG_INLINE Packet8l ploadu<Packet8l>(const int64_t* from) {
EIGEN_DEBUG_UNALIGNED_LOAD return _mm512_loadu_epi64(from);
EIGEN_DEBUG_UNALIGNED_LOAD return _eigen_mm512_loadu_epi64(from);
}

template <>
Expand Down Expand Up @@ -1158,11 +1206,11 @@ EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const Packet8d& from) {
}
template <>
EIGEN_STRONG_INLINE void pstoreu<int>(int* to, const Packet16i& from) {
EIGEN_DEBUG_UNALIGNED_STORE _mm512_storeu_epi32(to, from);
EIGEN_DEBUG_UNALIGNED_STORE _eigen_mm512_storeu_epi32(to, from);
}
template <>
EIGEN_STRONG_INLINE void pstoreu<int64_t>(int64_t* to, const Packet8l& from) {
EIGEN_DEBUG_UNALIGNED_STORE _mm512_storeu_epi64(to, from);
EIGEN_DEBUG_UNALIGNED_STORE _eigen_mm512_storeu_epi64(to, from);
}
template <>
EIGEN_STRONG_INLINE void pstoreu<float>(float* to, const Packet16f& from, uint16_t umask) {
Expand Down Expand Up @@ -3019,19 +3067,19 @@ EIGEN_STRONG_INLINE void pstore<numext::int16_t, Packet8s>(numext::int16_t* out,
template <>
EIGEN_STRONG_INLINE void pstoreu<numext::int16_t, Packet32s>(numext::int16_t* out, const Packet32s& x) {
EIGEN_DEBUG_UNALIGNED_STORE
_mm512_storeu_epi32(out, x);
_eigen_mm512_storeu_epi32(out, x);
}

template <>
EIGEN_STRONG_INLINE void pstoreu<numext::int16_t, Packet16s>(numext::int16_t* out, const Packet16s& x) {
EIGEN_DEBUG_UNALIGNED_STORE
_mm256_storeu_epi32(out, x);
_eigen_mm256_storeu_epi32(out, x);
}

template <>
EIGEN_STRONG_INLINE void pstoreu<numext::int16_t, Packet8s>(numext::int16_t* out, const Packet8s& x) {
EIGEN_DEBUG_UNALIGNED_STORE
_mm_storeu_epi32(out, x);
_eigen_mm_storeu_epi32(out, x);
}

template <>
Expand Down
10 changes: 9 additions & 1 deletion inst/include/Eigen/src/Core/arch/AVX512/TypeCasting.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,9 +52,17 @@ struct type_casting_traits<bfloat16, float> : vectorized_type_casting_traits<bfl
template <>
struct type_casting_traits<float, bfloat16> : vectorized_type_casting_traits<float, bfloat16> {};

EIGEN_STRONG_INLINE __mmask16 _eigen_mm512_cmpneq_ps_mask(__m512 a, __m512 b) {
#if EIGEN_COMP_GNUC && (EIGEN_COMP_CLANG < 1000 || EIGEN_COMP_GNUC < 810)
return _mm512_cmp_ps_mask(a, b, _CMP_NEQ_UQ);
#else
return _mm512_cmpneq_ps_mask(a, b);
#endif
}

template <>
EIGEN_STRONG_INLINE Packet16b pcast<Packet16f, Packet16b>(const Packet16f& a) {
__mmask16 mask = _mm512_cmpneq_ps_mask(a, pzero(a));
__mmask16 mask = _eigen_mm512_cmpneq_ps_mask(a, pzero(a));
return _mm512_maskz_cvtepi32_epi8(mask, _mm512_set1_epi32(1));
}

Expand Down
Loading