Skip to content
Snippets Groups Projects
Commit 8ddb0cb9 authored by Christos Anastopoulos's avatar Christos Anastopoulos Committed by Adam Edward Barton
Browse files

gcc11 supports the clang __builtin_convertvector for vectorized types. Add a helper in vec.h.

parent a1408770
9 merge requests!69091Fix correlated smearing bug in JER in JetUncertainties in 22.0,!58791DataQualityConfigurations: Modify L1Calo config for web display,!51674Fixing hotSpotInHIST for Run3 HIST,!50012RecExConfig: Adjust log message levels from GetRunNumber and GetLBNumber,!46784MuonCondInterface: Enable thread-safety checking.,!46776Updated LArMonitoring config file for WD to match new files produced using MT,!46538Draft: Added missing xAOD::TrigConfKeys from DESDM_MCP,!46514TGC Digitization: Implementation of signal propagation time between the sensor edge and ASD,!46147gcc11 supports the clang __builtin_convertvector for vectorized types. Add a helper in vec.h.
Hide whitespace changes
Inline Side-by-side
Control/CxxUtils/CxxUtils/features.h
+ 9
− 0
View file @ 8ddb0cb9
......@@ -76,6 +76,15 @@
#define HAVE_VECTOR_TERNARY_OPERATOR 0
#endif
// Do we additionally support the clang
// __builtin_convertvector
// GCC>11 does
#if HAVE_VECTOR_SIZE_ATTRIBUTE && (defined(__clang__) || (__GNUC__ >= 11))
#define HAVE_CONVERT_VECTOR 1
#else
#define HAVE_CONVERT_VECTOR 0
#endif
// Do we have mallinfo2? Present in glibc 2.33,
// in which mallinfo is deprecated.
#define HAVE_MALLINFO2 (__GLIBC_PREREQ(2, 33))
......
Control/CxxUtils/CxxUtils/vec.h
+ 30
− 6
View file @ 8ddb0cb9
......@@ -84,6 +84,11 @@
* indicates that the element number i-N
* from the second input vector should be placed
* in the corresponding position in the result vector.
* - @c CxxUtils::vconvert(VEC1& dst, const VEC2& src)
* Fills dst with the result of a
* static_cast of every element of src
* to the element type of dst.
* dst[i] = static_cast<vec_type_t<VEC1>>(src[i])
*
* In terms of expected performance it might be advantageous to
* use vector types that fit the size of the ISA.
......@@ -112,9 +117,10 @@
namespace CxxUtils {
// Define @c WANT_VECTOR_FALLBACK prior to including this file to always
// Define @c WANT_VECTOR_FALLBACK prior to including this file to
// make the fallback class @c vec_fb visible, even if we support the
// built-in type. Intended for testing.
// built-in type.
// Intended for testing.
#ifndef WANT_VECTOR_FALLBACK
# define WANT_VECTOR_FALLBACK 0
#endif
......@@ -526,10 +532,10 @@ vpermute(VEC& dst, const VEC& src)
constexpr size_t N = vec_size<VEC>();
static_assert((sizeof...(Indices) == N),
"Number of indices different than vector size");
"vpermute number of indices different than vector size");
static_assert(
bool_pack_helper::all_true<(Indices >= 0 && Indices < N)...>::value,
"permute indices outside allowed range");
"vpermute indices outside allowed range");
#if !HAVE_VECTOR_SIZE_ATTRIBUTE || WANT_VECTOR_FALLBACK
dst = VEC{ src[Indices]... };
......@@ -551,10 +557,10 @@ vblend(VEC& dst, const VEC& src1, const VEC& src2)
{
constexpr size_t N = vec_size<VEC>();
static_assert((sizeof...(Indices) == N),
"Number of indices different than vector size");
"vblend number of indices different than vector size");
static_assert(
bool_pack_helper::all_true<(Indices >= 0 && Indices < 2 * N)...>::value,
"blend indices outside allowed range");
"vblend indices outside allowed range");
#if !HAVE_VECTOR_SIZE_ATTRIBUTE || WANT_VECTOR_FALLBACK
size_t pos{ 0 };
......@@ -573,6 +579,24 @@ vblend(VEC& dst, const VEC& src1, const VEC& src2)
#endif
}
template<typename VEC1, typename VEC2>
inline void
vconvert(VEC1& dst, const VEC2& src)
{
static_assert((vec_size<VEC1>() == vec_size<VEC2>()),
"vconvert dst and src have different number of elements");
#if !HAVE_CONVERT_VECTOR || WANT_VECTOR_FALLBACK
typedef vec_type_t<VEC1> ELT;
constexpr size_t N = vec_size<VEC1>();
for (size_t i = 0; i < N; ++i) {
dst[i] = static_cast<ELT>(src[i]);
}
#else
dst = __builtin_convertvector(src, VEC1);
#endif
}
} // namespace CxxUtils
#endif // not CXXUTILS_VEC_H
Control/CxxUtils/test/vec_test_common.h
+ 50
− 19
View file @ 8ddb0cb9
......@@ -34,7 +34,7 @@ check(const VEC& v, const std::valarray<T>& a)
{
const size_t N = CxxUtils::vec_size<VEC>();
assert(a.size() == N);
for (size_t i = 0; i < N; i++) {
for (size_t i = 0; i < N; ++i) {
if (v[i] != a[i]) {
std::cerr << "Mismatch " << typeid(VEC).name() << " "
<< typeid(std::valarray<T>).name() << " " << i << "\n";
......@@ -52,7 +52,7 @@ check(const VEC& v, const std::valarray<bool>& a)
{
const size_t N = CxxUtils::vec_size<VEC>();
assert(a.size() == N);
for (size_t i = 0; i < N; i++) {
for (size_t i = 0; i < N; ++i) {
assert(bool(v[i]) == a[i]);
}
}
......@@ -66,7 +66,7 @@ NO_SANITIZE_UNDEFINED(const VEC& v1)
size_t N = CxxUtils::vec_size<VEC>();
std::valarray<T> a1(N);
for (size_t i = 0; i < N; i++)
for (size_t i = 0; i < N; ++i)
a1[i] = v1[i];
check(v1, a1);
......@@ -119,7 +119,7 @@ NO_SANITIZE_UNDEFINED(const VEC& v1)
size_t N = CxxUtils::vec_size<VEC>();
std::valarray<T> a1(N);
for (size_t i = 0; i < N; i++)
for (size_t i = 0; i < N; ++i)
a1[i] = v1[i];
check(v1, a1);
......@@ -183,16 +183,16 @@ test_relops(const VEC& v1)
size_t N = CxxUtils::vec_size<VEC>();
std::valarray<T> a1(N);
for (size_t i = 0; i < N; i++)
for (size_t i = 0; i < N; ++i)
a1[i] = v1[i];
VEC v2;
v2[0] = v1[0];
for (size_t i = 1; i < N; i++)
for (size_t i = 1; i < N; ++i)
v2[i] = v1[N - i];
std::valarray<T> a2(N);
for (size_t i = 0; i < N; i++)
for (size_t i = 0; i < N; ++i)
a2[i] = v2[i];
#define TEST(op) \
......@@ -220,15 +220,15 @@ test_logops(const VEC& v1)
size_t N = CxxUtils::vec_size<VEC>();
std::valarray<T> a1(N);
for (size_t i = 0; i < N; i++)
for (size_t i = 0; i < N; ++i)
a1[i] = v1[i];
VEC v2;
for (size_t i = 0; i < N; i++)
for (size_t i = 0; i < N; ++i)
v2[i] = v1[N - 1 - i];
std::valarray<T> a2(N);
for (size_t i = 0; i < N; i++)
for (size_t i = 0; i < N; ++i)
a2[i] = v2[i];
{
......@@ -277,7 +277,7 @@ test_broadcast(const VEC& v1)
CxxUtils::vbroadcast(v2, v1[0]);
size_t N = CxxUtils::vec_size<VEC>();
for (size_t i = 0; i < N; i++) {
for (size_t i = 0; i < N; ++i) {
assert(v1[i] == v2[i]);
}
}
......@@ -292,13 +292,13 @@ test_storeload(const VEC& v1)
CxxUtils::vstore(mem_addr, v1);
size_t N = CxxUtils::vec_size<VEC>();
for (size_t i = 0; i < N; i++) {
for (size_t i = 0; i < N; ++i) {
assert(v1[i] == mem_addr[i]);
}
VEC v2;
CxxUtils::vload(v2, mem_addr);
for (size_t i = 0; i < N; i++) {
for (size_t i = 0; i < N; ++i) {
assert(v2[i] == mem_addr[i]);
}
}
......@@ -312,7 +312,7 @@ test_min(const VEC& v1)
VEC min;
CxxUtils::vmin(min, v1, v2);
constexpr size_t N = CxxUtils::vec_size<VEC>();
for (size_t i = 0; i < N; i++) {
for (size_t i = 0; i < N; ++i) {
assert(min[i] == v1[i]);
}
}
......@@ -326,7 +326,7 @@ test_max(const VEC& v1)
VEC max;
CxxUtils::vmax(max, v1, v2);
constexpr size_t N = CxxUtils::vec_size<VEC>();
for (size_t i = 0; i < N; i++) {
for (size_t i = 0; i < N; ++i) {
assert(max[i] == v2[i]);
}
}
......@@ -340,7 +340,7 @@ test_select(const VEC& v1)
CxxUtils::mask_type_t<VEC> greater;
CxxUtils::mask_type_t<VEC> less;
constexpr size_t N = CxxUtils::vec_size<VEC>();
for (size_t i = 0; i < N; i++) {
for (size_t i = 0; i < N; ++i) {
greater[i] = v1[i] > v2[i];
less[i] = v1[i] < v2[i];
}
......@@ -350,7 +350,7 @@ test_select(const VEC& v1)
VEC selectLess;
CxxUtils::vselect(selectLess, v1, v2, less);
for (size_t i = 0; i < N; i++) {
for (size_t i = 0; i < N; ++i) {
assert(selectGreater[i] == v2[i]);
assert(selectLess[i] == v1[i]);
}
......@@ -374,7 +374,7 @@ test_permute(const VEC& v1)
CxxUtils::vpermute<15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0>(
v2, v1);
}
for (size_t i = 0; i < N; i++) {
for (size_t i = 0; i < N; ++i) {
assert(v2[i] == v1[(N-1)-i]);
}
}
......@@ -398,7 +398,7 @@ test_blend(const VEC& v1)
CxxUtils::vblend<0, 1, 2, 3, 4, 5, 6, 7, 16, 17, 18, 19, 20, 21, 22, 23>(
v3, v1, v2);
}
for (size_t i = 0; i < N; i++) {
for (size_t i = 0; i < N; ++i) {
if (i < N / 2) {
assert(v3[i] == v1[i]);
} else {
......@@ -407,6 +407,31 @@ test_blend(const VEC& v1)
}
}
template<class VEC>
void
test_convert_to_double(const VEC& v1)
{
constexpr size_t N = CxxUtils::vec_size<VEC>();
CxxUtils::vec<double, N> dst;
CxxUtils::vconvert(dst, v1);
for (size_t i = 0; i < N; ++i) {
assert(dst[i] == static_cast<double>(v1[i]));
}
}
template<class VEC>
void
test_convert_to_int(const VEC& v1)
{
constexpr size_t N = CxxUtils::vec_size<VEC>();
CxxUtils::vec<int, N> dst;
CxxUtils::vconvert(dst, v1);
for (size_t i = 0; i < N; ++i) {
assert(dst[i] == static_cast<int>(v1[i]));
}
}
template<template<class T, size_t N> class VEC>
void
testFloat1()
......@@ -438,6 +463,8 @@ testFloat1()
test_max(VEC<T, N> INITN(N, 1.5, 2.5, 3.5, 4.5, 5.5, 6.5, 7.5, 8.5)); \
test_permute(VEC<T, N> INITN(N, 1.5, 2.5, 3.5, 4.5, 5.5, 6.5, 7.5, 8.5)); \
test_blend(VEC<T, N> INITN(N, 1.5, 2.5, 3.5, 4.5, 5.5, 6.5, 7.5, 8.5)); \
test_convert_to_int( \
VEC<T, N> INITN(N, 1.5, 2.5, 3.5, 4.5, 5.5, 6.5, 7.5, 8.5)); \
} while (0)
TEST_FLOAT(float, 4); // 128 bit wide 4 floats
......@@ -450,6 +477,8 @@ testFloat1()
#undef INITN
#undef ELT
}
template<template<class T, size_t N> class VEC>
void
testInt1()
......@@ -488,6 +517,8 @@ testInt1()
N, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)); \
test_blend(VEC<T, N> INITN( \
N, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)); \
test_convert_to_double(VEC<T, N> INITN( \
N, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)); \
test_int(VEC<T, N> INITN( \
N, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)); \
test_logops( \
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment