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

[SYCL] Add tests for some half builtins #880

Merged
merged 8 commits into from
Mar 28, 2022
Merged

[SYCL] Add tests for some half builtins #880

Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
b7aa4c9
[SYCL] added tests for some half builtins
t4c1 Mar 2, 2022
2e797c1
disabled for OpenCL backend on CPU
t4c1 Mar 3, 2022
02f653c
format
t4c1 Mar 3, 2022
ac29a61
actually check relative error
t4c1 Mar 7, 2022
2f7fc03
format
t4c1 Mar 7, 2022
20dd111
addressed review comments and refactored into a single kernel
t4c1 Mar 21, 2022
f0dbb4a
Update SYCL/Basic/half_builtins.cpp
t4c1 Mar 21, 2022
bf1912d
format
t4c1 Mar 21, 2022
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
197 changes: 197 additions & 0 deletions SYCL/Basic/half_builtins.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,197 @@
// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
// RUN: %HOST_RUN_PLACEHOLDER %t.out
// RUN: %CPU_RUN_PLACEHOLDER %t.out
// RUN: %GPU_RUN_PLACEHOLDER %t.out
// RUN: %ACC_RUN_PLACEHOLDER %t.out

// OpenCL CPU driver does not support cl_khr_fp16 extension
// UNSUPPORTED: cpu && opencl
Copy link

Choose a reason for hiding this comment

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

NIT: would it be better to test for the extension support in the source code itself?


#include <sycl/sycl.hpp>

#include <cmath>
#include <limits>

using namespace sycl;

constexpr int SZ_max = 16;

bool check(float a, float b) {
return fabs(2 * (a - b) / (a + b)) < std::numeric_limits<half>::epsilon() ||
a < std::numeric_limits<half>::min();
}

template <int N> bool check(vec<float, N> a, vec<float, N> b) {
for (int i = 0; i < N; i++) {
if (!check(a[i], b[i])) {
return false;
}
}
return true;
}

#define TEST_BUILTIN_1_VEC_IMPL(NAME, SZ) \
{ \
float##SZ *a = (float##SZ *)&A[0]; \
float##SZ *b = (float##SZ *)&B[0]; \
if (i < SZ_max / SZ) { \
if (!check(NAME(a[i]), NAME(a[i].convert<half>()).convert<float>())) { \
err[0] = 1; \
} \
} \
}

// vectors of size 3 need separate test, as they actually have the size of 4
// elements
#define TEST_BUILTIN_1_VEC3_IMPL(NAME) \
{ \
float3 *a = (float3 *)&A[0]; \
float3 *b = (float3 *)&B[0]; \
if (i < SZ_max / 4) { \
if (!check(NAME(a[i]), NAME(a[i].convert<half>()).convert<float>())) { \
err[0] = 1; \
} \
} \
}

#define TEST_BUILTIN_1_SCAL_IMPL(NAME) \
{ \
float *a = (float *)&A[0]; \
float *b = (float *)&B[0]; \
if (!check(NAME(a[i]), (float)NAME((half)a[i]))) { \
err[0] = 1; \
} \
}

#define TEST_BUILTIN_1(NAME) \
TEST_BUILTIN_1_SCAL_IMPL(NAME) \
TEST_BUILTIN_1_VEC_IMPL(NAME, 2) \
TEST_BUILTIN_1_VEC3_IMPL(NAME) \
TEST_BUILTIN_1_VEC_IMPL(NAME, 4) \
TEST_BUILTIN_1_VEC_IMPL(NAME, 8) \
TEST_BUILTIN_1_VEC_IMPL(NAME, 16)

#define TEST_BUILTIN_2_VEC_IMPL(NAME, SZ) \
{ \
float##SZ *a = (float##SZ *)&A[0]; \
float##SZ *b = (float##SZ *)&B[0]; \
if (i < SZ_max / SZ) { \
if (!check(NAME(a[i], b[i]), \
NAME(a[i].convert<half>(), b[i].convert<half>()) \
.convert<float>())) { \
err[0] = 1; \
} \
} \
}

#define TEST_BUILTIN_2_VEC3_IMPL(NAME) \
{ \
float3 *a = (float3 *)&A[0]; \
float3 *b = (float3 *)&B[0]; \
if (i < SZ_max / 4) { \
if (!check(NAME(a[i], b[i]), \
NAME(a[i].convert<half>(), b[i].convert<half>()) \
.convert<float>())) { \
err[0] = 1; \
} \
} \
}

#define TEST_BUILTIN_2_SCAL_IMPL(NAME) \
{ \
float *a = (float *)&A[0]; \
float *b = (float *)&B[0]; \
if (!check(NAME(a[i], b[i]), (float)NAME((half)a[i], (half)b[i]))) { \
err[0] = 1; \
} \
}

#define TEST_BUILTIN_2(NAME) \
TEST_BUILTIN_2_SCAL_IMPL(NAME) \
TEST_BUILTIN_2_VEC_IMPL(NAME, 2) \
TEST_BUILTIN_2_VEC3_IMPL(NAME) \
TEST_BUILTIN_2_VEC_IMPL(NAME, 4) \
TEST_BUILTIN_2_VEC_IMPL(NAME, 8) \
TEST_BUILTIN_2_VEC_IMPL(NAME, 16)

#define TEST_BUILTIN_3_VEC_IMPL(NAME, SZ) \
{ \
float##SZ *a = (float##SZ *)&A[0]; \
float##SZ *b = (float##SZ *)&B[0]; \
float##SZ *c = (float##SZ *)&C[0]; \
if (i < SZ_max / SZ) { \
if (!check(NAME(a[i], b[i], c[i]), \
NAME(a[i].convert<half>(), b[i].convert<half>(), \
c[i].convert<half>()) \
.convert<float>())) { \
err[0] = 1; \
} \
} \
}

#define TEST_BUILTIN_3_VEC3_IMPL(NAME) \
{ \
float3 *a = (float3 *)&A[0]; \
float3 *b = (float3 *)&B[0]; \
float3 *c = (float3 *)&C[0]; \
if (i < SZ_max / 4) { \
if (!check(NAME(a[i], b[i], c[i]), \
NAME(a[i].convert<half>(), b[i].convert<half>(), \
c[i].convert<half>()) \
.convert<float>())) { \
err[0] = 1; \
} \
} \
}

#define TEST_BUILTIN_3_SCAL_IMPL(NAME) \
{ \
float *a = (float *)&A[0]; \
float *b = (float *)&B[0]; \
float *c = (float *)&C[0]; \
if (!check(NAME(a[i], b[i], c[i]), \
(float)NAME((half)a[i], (half)b[i], (half)c[i]))) { \
err[0] = 1; \
} \
}

#define TEST_BUILTIN_3(NAME) \
TEST_BUILTIN_3_SCAL_IMPL(NAME) \
TEST_BUILTIN_3_VEC_IMPL(NAME, 2) \
TEST_BUILTIN_3_VEC3_IMPL(NAME) \
TEST_BUILTIN_3_VEC_IMPL(NAME, 4) \
TEST_BUILTIN_3_VEC_IMPL(NAME, 8) \
TEST_BUILTIN_3_VEC_IMPL(NAME, 16)

int main() {
queue q;
float16 a, b, c, d;
for (int i = 0; i < SZ_max; i++) {
a[i] = i / (float)SZ_max;
b[i] = (SZ_max - i) / (float)SZ_max;
c[i] = (float)(3 * i);
}
int err = 0;
{
buffer<float16> a_buf(&a, 1);
buffer<float16> b_buf(&b, 1);
buffer<float16> c_buf(&c, 1);
buffer<int> err_buf(&err, 1);
q.submit([&](handler &cgh) {
auto A = a_buf.get_access<access::mode::read>(cgh);
auto B = b_buf.get_access<access::mode::read>(cgh);
auto C = c_buf.get_access<access::mode::read>(cgh);
auto err = err_buf.get_access<access::mode::write>(cgh);
cgh.parallel_for(SZ_max, [=](item<1> index) {
size_t i = index.get_id(0);
TEST_BUILTIN_1(fabs);
TEST_BUILTIN_2(fmin);
TEST_BUILTIN_2(fmax);
TEST_BUILTIN_3(fma);
});
});
}
assert(err == 0);

return 0;
}