Skip to content
This repository was archived by the owner on Mar 28, 2023. It is now read-only.

[SYCL][ESIMD] Add a runtime test for BFN function #1671

Merged
merged 6 commits into from
Mar 24, 2023
Merged
Changes from all commits
Commits
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
259 changes: 259 additions & 0 deletions SYCL/ESIMD/bfn.cpp
Original file line number Diff line number Diff line change
@@ -0,0 +1,259 @@
//==---------------- bfn.cpp - DPC++ ESIMD binary function test ------------==//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM
// Exceptions. See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
// REQUIRES: intel-gpu-dg2 || intel-gpu-pvc
// RUN: %clangxx -fsycl-device-code-split=per_kernel -fsycl %s -o %t.out
// RUN: %GPU_RUN_PLACEHOLDER %t.out

// This test checks binary function (bfn) operations. Combinations of
// - argument type - uint16_t, uint32_t.
// - binary function - several binary functins with three operands (~, &, |, ^).

#include "esimd_test_utils.hpp"

#include <sycl/ext/intel/esimd.hpp>
#include <sycl/sycl.hpp>

#include <iostream>

using namespace sycl;
using namespace sycl::ext::intel;

// --- Initialization function for source operands of binary functions.

template <class T> struct InitOps {
void operator()(T *In0, T *In1, T *In2, T *Out, size_t Size) const {
for (auto I = 0; I < Size; ++I) {
In0[I] = I * 3;
In1[I] = I * 3 + 1;
In2[I] = I * 3 + 2;
Out[I] = (T)0;
}
}
};

// --- Test boolean control functions.

using bfn_t = experimental::esimd::bfn_t;

constexpr experimental::esimd::bfn_t F1 = bfn_t::x | bfn_t::y | bfn_t::z;
constexpr experimental::esimd::bfn_t F2 = bfn_t::x & bfn_t::y & bfn_t::z;
constexpr experimental::esimd::bfn_t F3 = ~bfn_t::x | bfn_t::y ^ bfn_t::z;

// --- Template functions calculating given boolean operation on host and device

enum ArgKind {
AllVec,
AllSca,
};

template <class T, experimental::esimd::bfn_t Op> struct HostFunc;

#define DEFINE_HOST_OP(FUNC_CTRL) \
template <class T> struct HostFunc<T, FUNC_CTRL> { \
T operator()(T X0, T X1, T X2) { \
T res = 0; \
for (unsigned i = 0; i < sizeof(X0) * 8; i++) { \
T mask = 0x1UL << i; \
res = (res & ~mask) | \
((static_cast<uint8_t>(FUNC_CTRL) >> \
((((X0 >> i) & 0x1UL)) + (((X1 >> i) & 0x1UL) << 1) + \
(((X2 >> i) & 0x1UL) << 2)) & \
0x1UL) \
<< i); \
} \
return res; \
} \
};

DEFINE_HOST_OP(F1);
DEFINE_HOST_OP(F2);
DEFINE_HOST_OP(F3);

// --- Specializations per each boolean operation.

template <class T, int N, experimental::esimd::bfn_t Op, int Args = AllVec>
struct ESIMDf;

#define DEFINE_ESIMD_DEVICE_OP(FUNC_CTRL) \
template <class T, int N> struct ESIMDf<T, N, FUNC_CTRL, AllVec> { \
esimd::simd<T, N> \
operator()(esimd::simd<T, N> X0, esimd::simd<T, N> X1, \
esimd::simd<T, N> X2) const SYCL_ESIMD_FUNCTION { \
return experimental::esimd::bfn<FUNC_CTRL, T, N>(X0, X1, X2); \
} \
}; \
template <class T, int N> struct ESIMDf<T, N, FUNC_CTRL, AllSca> { \
esimd::simd<T, N> operator()(T X0, T X1, T X2) const SYCL_ESIMD_FUNCTION { \
return experimental::esimd::bfn<FUNC_CTRL, T, N>(X0, X1, X2); \
} \
};

DEFINE_ESIMD_DEVICE_OP(F1);
DEFINE_ESIMD_DEVICE_OP(F2);
DEFINE_ESIMD_DEVICE_OP(F3);

// --- Generic kernel calculating a binary function operation on array elements.

template <class T, int N, experimental::esimd::bfn_t Op,
template <class, int, experimental::esimd::bfn_t, int> class Kernel>
struct DeviceFunc {
const T *In0, *In1, *In2;
T *Out;

DeviceFunc(const T *In0, const T *In1, const T *In2, T *Out)
: In0(In0), In1(In1), In2(In2), Out(Out) {}

void operator()(id<1> I) const SYCL_ESIMD_KERNEL {
unsigned int Offset = I * N;
esimd::simd<T, N> V0;
esimd::simd<T, N> V1;
esimd::simd<T, N> V2;
V0.copy_from(In0 + Offset);
V1.copy_from(In1 + Offset);
V2.copy_from(In2 + Offset);

if (I.get(0) % 2 == 0) {
for (int J = 0; J < N; J++) {
Kernel<T, N, Op, AllSca> DevF{};
T Val0 = V0[J];
T Val1 = V1[J];
T Val2 = V2[J];
esimd::simd<T, N> V = DevF(Val0, Val1, Val2); // scalar arg
V0[J] = V[J];
}
} else {
Kernel<T, N, Op, AllVec> DevF{};
V0 = DevF(V0, V1, V2); // vector arg
}
V0.copy_to(Out + Offset);
};
};

// --- Generic test function for boolean function.

template <class T, int N, experimental::esimd::bfn_t Op, int Range,
template <class, int, experimental::esimd::bfn_t, int> class Kernel,
typename InitF = InitOps<T>>
bool test(queue &Q, const std::string &Name, InitF Init = InitOps<T>{}) {
constexpr size_t Size = Range * N;

auto UA = esimd_test::usm_malloc_shared<T>(Q, Size);
T *A = UA.get();
auto UB = esimd_test::usm_malloc_shared<T>(Q, Size);
T *B = UB.get();
auto UC = esimd_test::usm_malloc_shared<T>(Q, Size);
T *C = UC.get();
auto UD = esimd_test::usm_malloc_shared<T>(Q, Size);
T *D = UD.get();
Init(A, B, C, D, Size);

std::cout << " " << Name << " test"
<< "...\n";

try {
// number of workgroups
sycl::range<1> GlobalRange{Range};

// threads (workitems) in each workgroup
sycl::range<1> LocalRange{1};

auto E = Q.submit([=](handler &CGH) {
DeviceFunc<T, N, Op, Kernel> F(A, B, C, D);
CGH.parallel_for(nd_range<1>{GlobalRange, LocalRange}, F);
});
E.wait();
} catch (sycl::exception &Exc) {
std::cout << " *** ERROR. SYCL exception caught: << " << Exc.what()
<< "\n";
return false;
}

int ErrCnt = 0;

for (unsigned I = 0; I < Size; ++I) {
T Gold;

Gold = HostFunc<T, Op>{}((T)A[I], (T)B[I], (T)C[I]);
T Test = D[I];

if (Test != Gold) {
if (++ErrCnt < 10) {
std::cout << "\tfailed at index " << I << ", " << std::hex << Test
<< " != " << Gold << " (gold); "
<< "Input was: " << (T)A[I] << ", " << (T)B[I] << ", "
<< (T)C[I] << "; "
<< "FuncCtrl: " << int(Op) << std::dec << "\n";
}
}
}

if (ErrCnt > 0) {
std::cout << " pass rate: "
<< ((float)(Size - ErrCnt) / (float)Size) * 100.0f << "% ("
<< (Size - ErrCnt) << "/" << Size << ")\n";
}

std::cout << (ErrCnt > 0 ? " FAILED\n" : " Passed\n");
return ErrCnt == 0;
}

// --- Tests all boolean operations with given vector length.

template <class T, int N, int Range> bool testESIMD(queue &Q) {
bool Pass = true;

std::cout << "--- TESTING ESIMD functions, T=" << typeid(T).name()
<< ", N = " << N << ", Range: " << Range << "...\n";

Pass &= test<T, N, F1, Range, ESIMDf>(Q, "F1");
Pass &= test<T, N, F2, Range, ESIMDf>(Q, "F2");
Pass &= test<T, N, F3, Range, ESIMDf>(Q, "F3");
return Pass;
}

template <class T, int N> bool testESIMDRanges(queue &Q) {
bool Pass = true;
// Test vector API.
Pass &= testESIMD<T, N, 128>(Q);
// Test scalar API with odd size.
Pass &= testESIMD<T, N, 101>(Q);
return Pass;
}

template <class T> bool testESIMDGroup(queue &Q) {
bool Pass = true;
Pass &= testESIMDRanges<T, 1>(Q);
Pass &= testESIMDRanges<T, 5>(Q);
Pass &= testESIMDRanges<T, 8>(Q);
Pass &= testESIMDRanges<T, 16>(Q);
Pass &= testESIMDRanges<T, 32>(Q);
return Pass;
}

// --- The entry point.

int main(void) {
queue Q(esimd_test::ESIMDSelector, esimd_test::createExceptionHandler());
auto Dev = Q.get_device();
std::cout << "Running on " << Dev.get_info<sycl::info::device::name>()
<< "\n";
bool Pass = true;

Pass &= testESIMDGroup<uint16_t>(Q);
Pass &= testESIMDGroup<uint32_t>(Q);
Pass &= testESIMDGroup<int16_t>(Q);
Pass &= testESIMDGroup<int32_t>(Q);

Pass &= testESIMDGroup<uint8_t>(Q);
Pass &= testESIMDGroup<int8_t>(Q);
Pass &= testESIMDGroup<uint64_t>(Q);
Pass &= testESIMDGroup<int64_t>(Q);

std::cout << (Pass ? "Test Passed\n" : "Test FAILED\n");
return Pass ? 0 : 1;
}