mirror of
https://github.com/Tencent/ncnn.git
synced 2026-10-19 14:32:15 +08:00
Lune
bedf9c6568
Add backend support for new UnaryOp operations across scalar, SIMD and Vulkan paths, including sign, sinh, cosh, asinh, acosh, atanh, expm1 and log1p. Wire the new ops through onnx2ncnn and pnnx expression conversion, update generated Python emission, and add regression coverage for the new conversion paths. Fix portability issues for expm1/log1p math declarations, NCNN_SIMPLEMATH builds, ONNX-enabled pnnx linking, and stabilize unary op tests across fp16, bf16 and torch exporter/version differences. Co-authored-by: nihui <nihuini@tencent.com>
131 lines
3.0 KiB
C++
131 lines
3.0 KiB
C++
// Copyright 2020 Tencent
|
|
// SPDX-License-Identifier: BSD-3-Clause
|
|
|
|
#include "testutil.h"
|
|
|
|
#define OP_TYPE_MAX 28
|
|
|
|
static int op_type = 0;
|
|
|
|
static int test_unaryop(const ncnn::Mat& _a)
|
|
{
|
|
ncnn::Mat a = _a;
|
|
if (op_type == 2 || op_type == 3)
|
|
{
|
|
// large dynamic range for floor ceil
|
|
for (int i = 0; i < a.total(); i++)
|
|
{
|
|
a[i] *= 1000;
|
|
}
|
|
}
|
|
if (op_type == 5 || op_type == 6 || op_type == 8 || op_type == 17)
|
|
{
|
|
// value must be positive for sqrt rsqrt log
|
|
Randomize(a, 0.001f, 2.f);
|
|
}
|
|
if (op_type == 25)
|
|
{
|
|
// value must be >= 1 for acosh
|
|
Randomize(a, 1.001f, 2.f);
|
|
}
|
|
if (op_type == 11 || op_type == 12 || op_type == 13)
|
|
{
|
|
// smaller range for tan asin acos
|
|
Randomize(a, -1.f, 1.f);
|
|
}
|
|
if (op_type == 26)
|
|
{
|
|
// keep away from +/-1 so bf16 input casting does not round into the singularity
|
|
Randomize(a, -0.99f, 0.99f);
|
|
}
|
|
if (op_type == 27)
|
|
{
|
|
// value must be > -1 for log1p
|
|
Randomize(a, -0.999f, 2.f);
|
|
}
|
|
#if __powerpc__
|
|
// nearbyintf produces wrong result in halfway cases, why ?
|
|
// too troublesome to resolve the compiler or qemu problem
|
|
// so just skip them --- nihui
|
|
if (op_type == 18)
|
|
{
|
|
// drop 0.4 ~ 0.6
|
|
for (int i = 0; i < a.total(); i++)
|
|
{
|
|
float v = a[i];
|
|
float vv = fabs(v - (int)v);
|
|
while (vv > 0.4f && vv < 0.6f)
|
|
{
|
|
v = RandomFloat(-15, 15);
|
|
vv = fabs(v - (int)v);
|
|
}
|
|
a[i] = v;
|
|
}
|
|
}
|
|
#endif // __powerpc__
|
|
|
|
ncnn::ParamDict pd;
|
|
pd.set(0, op_type);
|
|
|
|
std::vector<ncnn::Mat> weights(0);
|
|
|
|
int ret = test_layer("UnaryOp", pd, weights, a);
|
|
if (ret != 0)
|
|
{
|
|
fprintf(stderr, "test_unaryop failed a.dims=%d a=(%d %d %d %d) op_type=%d\n", a.dims, a.w, a.h, a.d, a.c, op_type);
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
static int test_unaryop_0()
|
|
{
|
|
return 0
|
|
|| test_unaryop(RandomMat(11, 3, 2, 16))
|
|
|| test_unaryop(RandomMat(10, 2, 2, 12))
|
|
|| test_unaryop(RandomMat(6, 1, 5, 13));
|
|
}
|
|
|
|
static int test_unaryop_1()
|
|
{
|
|
return 0
|
|
|| test_unaryop(RandomMat(11, 7, 16))
|
|
|| test_unaryop(RandomMat(10, 4, 12))
|
|
|| test_unaryop(RandomMat(6, 5, 13));
|
|
}
|
|
|
|
static int test_unaryop_2()
|
|
{
|
|
return 0
|
|
|| test_unaryop(RandomMat(12, 16))
|
|
|| test_unaryop(RandomMat(10, 12))
|
|
|| test_unaryop(RandomMat(14, 15));
|
|
}
|
|
|
|
static int test_unaryop_3()
|
|
{
|
|
return 0
|
|
|| test_unaryop(RandomMat(128))
|
|
|| test_unaryop(RandomMat(12))
|
|
|| test_unaryop(RandomMat(15));
|
|
}
|
|
|
|
int main()
|
|
{
|
|
SRAND(7767517);
|
|
|
|
for (op_type = 0; op_type < OP_TYPE_MAX; op_type++)
|
|
{
|
|
int ret = 0
|
|
|| test_unaryop_0()
|
|
|| test_unaryop_1()
|
|
|| test_unaryop_2()
|
|
|| test_unaryop_3();
|
|
|
|
if (ret != 0)
|
|
return ret;
|
|
}
|
|
|
|
return 0;
|
|
}
|