potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 23:33:05 +04:00
Code Packages Releases Wiki Activity

core: Rename cv::float16_t to cv::hfloat (#25217)

Browse Source 
* rename cv::float16_t to cv::fp16_t

* add typedef fp16_t float16_t

* remove zero(), bits() from fp16_t class

* fp16_t -> hfloat

* remove cv::float16_t::fromBits; add hfloatFromBits

* undo changes in conv_winograd_f63.simd.hpp and conv_block.simd.hpp

* undo some changes in dnn
This commit is contained in:
Yuantao Feng
2024-03-22 04:44:19 +08:00
committed by GitHub
parent 3aefd4862c
commit 3afe8ddaf8
31 changed files with 156 additions and 166 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/include/opencv2/core/cvdef.h
+28 -38
View File
@@ -809,40 +809,20 @@ using std::uint64_t;
namespace cv
{
class float16_t
class hfloat
{
public:
#if CV_FP16_TYPE
float16_t() : h(0) {}
explicit float16_t(float x) { h = (__fp16)x; }
hfloat() : h(0) {}
explicit hfloat(float x) { h = (__fp16)x; }
operator float() const { return (float)h; }
static float16_t fromBits(ushort w)
{
Cv16suf u;
u.u = w;
float16_t result;
result.h = u.h;
return result;
}
static float16_t zero()
{
float16_t result;
result.h = (__fp16)0;
return result;
}
ushort bits() const
{
Cv16suf u;
u.h = h;
return u.u;
}
protected:
__fp16 h;
#else
float16_t() : w(0) {}
explicit float16_t(float x)
hfloat() : w(0) {}
explicit hfloat(float x)
{
#if CV_FP16 && CV_AVX2
__m128 v = _mm_load_ss(&x);
@@ -893,25 +873,35 @@ protected:
#endif
}
static float16_t fromBits(ushort b)
{
float16_t result;
result.w = b;
return result;
}
static float16_t zero()
{
float16_t result;
result.w = (ushort)0;
return result;
}
ushort bits() const { return w; }
protected:
ushort w;
#endif
};
inline hfloat hfloatFromBits(ushort w) {
#if CV_FP16_TYPE
Cv16suf u;
u.u = w;
hfloat res(float(u.h));
return res;
#else
Cv32suf out;
unsigned t = ((w & 0x7fff) << 13) + 0x38000000;
unsigned sign = (w & 0x8000) << 16;
unsigned e = w & 0x7c00;
out.u = t + (1 << 23);
out.u = (e >= 0x7c00 ? t + 0x38000000 :
e == 0 ? (static_cast<void>(out.f -= 6.103515625e-05f), out.u) : t) | sign;
hfloat res(out.f);
return res;
#endif
}
typedef hfloat float16_t;
}
#endif
modules/core/include/opencv2/core/hal/hal.hpp
+2 -2
View File
@@ -195,8 +195,8 @@ CV_EXPORTS void addWeighted32s( const int* src1, size_t step1, const int* src2,
CV_EXPORTS void addWeighted32f( const float* src1, size_t step1, const float* src2, size_t step2, float* dst, size_t step, int width, int height, void* scalars );
CV_EXPORTS void addWeighted64f( const double* src1, size_t step1, const double* src2, size_t step2, double* dst, size_t step, int width, int height, void* scalars );
CV_EXPORTS void cvt16f32f( const float16_t* src, float* dst, int len );
CV_EXPORTS void cvt32f16f( const float* src, float16_t* dst, int len );
CV_EXPORTS void cvt16f32f( const hfloat* src, float* dst, int len );
CV_EXPORTS void cvt32f16f( const float* src, hfloat* dst, int len );
CV_EXPORTS void addRNGBias32f( float* arr, const float* scaleBiasPairs, int len );
CV_EXPORTS void addRNGBias64f( double* arr, const double* scaleBiasPairs, int len );
modules/core/include/opencv2/core/hal/intrin.hpp
+1 -1
View File
@@ -708,7 +708,7 @@ namespace CV__SIMD_NAMESPACE {
inline v_int32 vx_load_expand(const short* ptr) { return VXPREFIX(_load_expand)(ptr); }
inline v_int64 vx_load_expand(const int* ptr) { return VXPREFIX(_load_expand)(ptr); }
inline v_uint64 vx_load_expand(const unsigned* ptr) { return VXPREFIX(_load_expand)(ptr); }
inline v_float32 vx_load_expand(const float16_t * ptr) { return VXPREFIX(_load_expand)(ptr); }
inline v_float32 vx_load_expand(const hfloat * ptr) { return VXPREFIX(_load_expand)(ptr); }
//! @}
//! @name Wide load with quad expansion
modules/core/include/opencv2/core/hal/intrin_avx.hpp
+3 -3
View File
@@ -3137,7 +3137,7 @@ OPENCV_HAL_IMPL_AVX_LOADSTORE_INTERLEAVE(v_float64x4, double, f64, v_uint64x4, u
// FP16
//
inline v_float32x8 v256_load_expand(const float16_t* ptr)
inline v_float32x8 v256_load_expand(const hfloat* ptr)
{
#if CV_FP16
return v_float32x8(_mm256_cvtph_ps(_mm_loadu_si128((const __m128i*)ptr)));
@@ -3149,7 +3149,7 @@ inline v_float32x8 v256_load_expand(const float16_t* ptr)
#endif
}
inline void v_pack_store(float16_t* ptr, const v_float32x8& a)
inline void v_pack_store(hfloat* ptr, const v_float32x8& a)
{
#if CV_FP16
__m128i ah = _mm256_cvtps_ph(a.val, 0);
@@ -3158,7 +3158,7 @@ inline void v_pack_store(float16_t* ptr, const v_float32x8& a)
float CV_DECL_ALIGNED(32) buf[8];
v_store_aligned(buf, a);
for (int i = 0; i < 8; i++)
ptr[i] = float16_t(buf[i]);
ptr[i] = hfloat(buf[i]);
#endif
}
modules/core/include/opencv2/core/hal/intrin_avx512.hpp
+2 -2
View File
@@ -506,12 +506,12 @@ inline v_float64x8 v_reinterpret_as_f64(const v_float32x16& a)
{ return v_float64x8(_mm512_castps_pd(a.val)); }
// FP16
inline v_float32x16 v512_load_expand(const float16_t* ptr)
inline v_float32x16 v512_load_expand(const hfloat* ptr)
{
return v_float32x16(_mm512_cvtph_ps(_mm256_loadu_si256((const __m256i*)ptr)));
}
inline void v_pack_store(float16_t* ptr, const v_float32x16& a)
inline void v_pack_store(hfloat* ptr, const v_float32x16& a)
{
__m256i ah = _mm512_cvtps_ph(a.val, 0);
_mm256_storeu_si256((__m256i*)ptr, ah);
modules/core/include/opencv2/core/hal/intrin_cpp.hpp
+5 -5
View File
@@ -3251,7 +3251,7 @@ template<int n> inline v_reg<double, n/2> v_dotprod_expand_fast(const v_reg<int,
////// FP16 support ///////
inline v_reg<float, simd128_width / sizeof(float)>
v_load_expand(const float16_t* ptr)
v_load_expand(const hfloat* ptr)
{
v_reg<float, simd128_width / sizeof(float)> v;
for( int i = 0; i < v.nlanes; i++ )
@@ -3262,7 +3262,7 @@ v_load_expand(const float16_t* ptr)
}
#if CV_SIMD256
inline v_reg<float, simd256_width / sizeof(float)>
v256_load_expand(const float16_t* ptr)
v256_load_expand(const hfloat* ptr)
{
v_reg<float, simd256_width / sizeof(float)> v;
for (int i = 0; i < v.nlanes; i++)
@@ -3274,7 +3274,7 @@ v256_load_expand(const float16_t* ptr)
#endif
#if CV_SIMD512
inline v_reg<float, simd512_width / sizeof(float)>
v512_load_expand(const float16_t* ptr)
v512_load_expand(const hfloat* ptr)
{
v_reg<float, simd512_width / sizeof(float)> v;
for (int i = 0; i < v.nlanes; i++)
@@ -3286,11 +3286,11 @@ v512_load_expand(const float16_t* ptr)
#endif
template<int n> inline void
v_pack_store(float16_t* ptr, const v_reg<float, n>& v)
v_pack_store(hfloat* ptr, const v_reg<float, n>& v)
{
for( int i = 0; i < v.nlanes; i++ )
{
ptr[i] = float16_t(v.s[i]);
ptr[i] = hfloat(v.s[i]);
}
}
modules/core/include/opencv2/core/hal/intrin_lasx.hpp
+3 -3
View File
@@ -2983,7 +2983,7 @@ OPENCV_HAL_IMPL_LASX_LOADSTORE_INTERLEAVE(v_float64x4, double, f64, v_uint64x4,
// FP16
//
inline v_float32x8 v256_load_expand(const float16_t* ptr)
inline v_float32x8 v256_load_expand(const hfloat* ptr)
{
#if CV_FP16
//1-load128, 2-permi, 3-cvt
@@ -2996,7 +2996,7 @@ inline v_float32x8 v256_load_expand(const float16_t* ptr)
#endif
}
inline void v_pack_store(float16_t* ptr, const v_float32x8& a)
inline void v_pack_store(hfloat* ptr, const v_float32x8& a)
{
#if CV_FP16
__m256i ah = __lasx_xvfcvt_h_s(a.val, a.val);
@@ -3005,7 +3005,7 @@ inline void v_pack_store(float16_t* ptr, const v_float32x8& a)
float CV_DECL_ALIGNED(32) buf[8];
v_store_aligned(buf, a);
for (int i = 0; i < 8; i++)
ptr[i] = float16_t(buf[i]);
ptr[i] = hfloat(buf[i]);
#endif
}
modules/core/include/opencv2/core/hal/intrin_lsx.hpp
+3 -3
View File
@@ -2498,7 +2498,7 @@ OPENCV_HAL_IMPL_LSX_LOADSTORE_INTERLEAVE(v_float64x2, double, f64, v_uint64x2, u
// FP16
//
inline v_float32x4 v_load_expand(const float16_t* ptr)
inline v_float32x4 v_load_expand(const hfloat* ptr)
{
#if CV_FP16
return v_float32x4(__lsx_vfcvtl_s_h((__m128)__lsx_vld(ptr, 0)));
@@ -2510,7 +2510,7 @@ inline v_float32x4 v_load_expand(const float16_t* ptr)
#endif
}
inline void v_pack_store(float16_t* ptr, const v_float32x4& a)
inline void v_pack_store(hfloat* ptr, const v_float32x4& a)
{
#if CV_FP16
__m128i res = (__m218i)__lsx_vfcvt_h_s(a.val, a.val);
@@ -2519,7 +2519,7 @@ inline void v_pack_store(float16_t* ptr, const v_float32x4& a)
float CV_DECL_ALIGNED(32) buf[4];
v_store_aligned(buf, a);
for (int i = 0; i < 4; i++)
ptr[i] = float16_t(buf[i]);
ptr[i] = hfloat(buf[i]);
#endif
}
modules/core/include/opencv2/core/hal/intrin_msa.hpp
+5 -5
View File
@@ -1838,7 +1838,7 @@ inline v_float32x4 v_broadcast_element(const v_float32x4& a)
////// FP16 support ///////
#if CV_FP16
inline v_float32x4 v_load_expand(const float16_t* ptr)
inline v_float32x4 v_load_expand(const hfloat* ptr)
{
#ifndef msa_ld1_f16
v4f16 v = (v4f16)msa_ld1_s16((const short*)ptr);
@@ -1848,7 +1848,7 @@ inline v_float32x4 v_load_expand(const float16_t* ptr)
return v_float32x4(msa_cvt_f32_f16(v));
}
inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
inline void v_pack_store(hfloat* ptr, const v_float32x4& v)
{
v4f16 hv = msa_cvt_f16_f32(v.val);
@@ -1859,7 +1859,7 @@ inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
#endif
}
#else
inline v_float32x4 v_load_expand(const float16_t* ptr)
inline v_float32x4 v_load_expand(const hfloat* ptr)
{
float buf[4];
for( int i = 0; i < 4; i++ )
@@ -1867,12 +1867,12 @@ inline v_float32x4 v_load_expand(const float16_t* ptr)
return v_load(buf);
}
inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
inline void v_pack_store(hfloat* ptr, const v_float32x4& v)
{
float buf[4];
v_store(buf, v);
for( int i = 0; i < 4; i++ )
ptr[i] = (float16_t)buf[i];
ptr[i] = (hfloat)buf[i];
}
#endif
modules/core/include/opencv2/core/hal/intrin_neon.hpp
+5 -5
View File
@@ -2605,7 +2605,7 @@ inline void v_lut_deinterleave(const double* tab, const v_int32x4& idxvec, v_flo
////// FP16 support ///////
#if CV_FP16
inline v_float32x4 v_load_expand(const float16_t* ptr)
inline v_float32x4 v_load_expand(const hfloat* ptr)
{
float16x4_t v =
#ifndef vld1_f16 // APPLE compiler defines vld1_f16 as macro
@@ -2616,7 +2616,7 @@ inline v_float32x4 v_load_expand(const float16_t* ptr)
return v_float32x4(vcvt_f32_f16(v));
}
inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
inline void v_pack_store(hfloat* ptr, const v_float32x4& v)
{
float16x4_t hv = vcvt_f16_f32(v.val);
@@ -2627,7 +2627,7 @@ inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
#endif
}
#else
inline v_float32x4 v_load_expand(const float16_t* ptr)
inline v_float32x4 v_load_expand(const hfloat* ptr)
{
const int N = 4;
float buf[N];
@@ -2635,12 +2635,12 @@ inline v_float32x4 v_load_expand(const float16_t* ptr)
return v_load(buf);
}
inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
inline void v_pack_store(hfloat* ptr, const v_float32x4& v)
{
const int N = 4;
float buf[N];
v_store(buf, v);
for( int i = 0; i < N; i++ ) ptr[i] = float16_t(buf[i]);
for( int i = 0; i < N; i++ ) ptr[i] = hfloat(buf[i]);
}
#endif
modules/core/include/opencv2/core/hal/intrin_rvv.hpp
+5 -5
View File
@@ -2873,17 +2873,17 @@ inline v_float32x4 v_pack_triplets(const v_float32x4& vec) { return vec; }
////// FP16 support ///////
#if CV_FP16
inline v_float32x4 v_load_expand(const float16_t* ptr)
inline v_float32x4 v_load_expand(const hfloat* ptr)
{
return v_float32x4(vfwcvt_f_f_v_f32m1(vle16_v_f16mf2(ptr, 4), 4));
}
inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
inline void v_pack_store(hfloat* ptr, const v_float32x4& v)
{
vse16_v_f16mf2(ptr, vfncvt_f_f_w_f16mf2(v, 4), 4);
}
#else
inline v_float32x4 v_load_expand(const float16_t* ptr)
inline v_float32x4 v_load_expand(const hfloat* ptr)
{
const int N = 4;
float buf[N];
@@ -2891,12 +2891,12 @@ inline v_float32x4 v_load_expand(const float16_t* ptr)
return v_load(buf);
}
inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
inline void v_pack_store(hfloat* ptr, const v_float32x4& v)
{
const int N = 4;
float buf[N];
v_store(buf, v);
for( int i = 0; i < N; i++ ) ptr[i] = float16_t(buf[i]);
for( int i = 0; i < N; i++ ) ptr[i] = hfloat(buf[i]);
}
#endif
modules/core/include/opencv2/core/hal/intrin_rvv071.hpp
+4 -4
View File
@@ -2858,14 +2858,14 @@ inline v_float64x2 v_dotprod_expand_fast(const v_int32x4& a, const v_int32x4& b,
#endif
////// FP16 support ///////
#if __riscv_v == 7000
inline v_float32x4 v_load_expand(const float16_t* ptr)
inline v_float32x4 v_load_expand(const hfloat* ptr)
{
vfloat16m1_t v = vle16_v_f16m1((__fp16*)ptr, 4);
vfloat32m2_t v32 = vfwcvt_f_f_v_f32m2(v, 4);
return v_float32x4(vget_v_f32m2_f32m1(v32, 0));
}
inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
inline void v_pack_store(hfloat* ptr, const v_float32x4& v)
{
vfloat32m2_t v32 = vundefined_f32m2();
v32 = vset_v_f32m1_f32m2(v32, 0, v.val);
@@ -2873,14 +2873,14 @@ inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
vse16_v_f16m1((__fp16*)ptr, hv, 4);
}
#else
inline v_float32x4 v_load_expand(const float16_t* ptr)
inline v_float32x4 v_load_expand(const hfloat* ptr)
{
vfloat16mf2_t v = vle16_v_f16mf2((__fp16*)ptr, 4);
vfloat32m1_t v32 = vfwcvt_f_f_v_f32m1(v, 4);
return v_float32x4(v32);
}
inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
inline void v_pack_store(hfloat* ptr, const v_float32x4& v)
{
//vfloat32m2_t v32 = vundefined_f32m2();
//v32 = vset_f32m2(v32, 0, v.val);
modules/core/include/opencv2/core/hal/intrin_rvv_scalable.hpp
+5 -5
View File
@@ -1810,28 +1810,28 @@ OPENCV_HAL_IMPL_RVV_PACK_TRIPLETS(v_float64, vlmul_trunc_u8mf8)
////// FP16 support ///////
#if defined(__riscv_zfh) && __riscv_zfh
inline v_float32 v_load_expand(const float16_t* ptr)
inline v_float32 v_load_expand(const hfloat* ptr)
{
return vfwcvt_f(vle16_v_f16mf2((_Float16*)ptr, VTraits<v_float32>::vlanes()) ,VTraits<v_float32>::vlanes());;
}
inline void v_pack_store(float16_t* ptr, const v_float32& v)
inline void v_pack_store(hfloat* ptr, const v_float32& v)
{
vse16_v_f16mf2((_Float16*)ptr, vfncvt_f_f_w_f16mf2(v, VTraits<v_float32>::vlanes()), VTraits<v_float32>::vlanes());
}
#else
inline v_float32 v_load_expand(const float16_t* ptr)
inline v_float32 v_load_expand(const hfloat* ptr)
{
float buf[32];
for( int i = 0; i < VTraits<v_float32>::vlanes(); i++ ) buf[i] = (float)ptr[i];
return v_load(buf);
}
inline void v_pack_store(float16_t* ptr, const v_float32& v)
inline void v_pack_store(hfloat* ptr, const v_float32& v)
{
float buf[32];
v_store(buf, v);
for( int i = 0; i < VTraits<v_float32>::vlanes(); i++ ) ptr[i] = float16_t(buf[i]);
for( int i = 0; i < VTraits<v_float32>::vlanes(); i++ ) ptr[i] = hfloat(buf[i]);
}
#endif
////////////// Rounding //////////////
modules/core/include/opencv2/core/hal/intrin_sse.hpp
+2 -2
View File
@@ -3407,7 +3407,7 @@ inline v_float32x4 v_broadcast_element(const v_float32x4& v)
////////////// FP16 support ///////////////////////////
inline v_float32x4 v_load_expand(const float16_t* ptr)
inline v_float32x4 v_load_expand(const hfloat* ptr)
{
#if CV_FP16
return v_float32x4(_mm_cvtph_ps(_mm_loadu_si128((const __m128i*)ptr)));
@@ -3427,7 +3427,7 @@ inline v_float32x4 v_load_expand(const float16_t* ptr)
#endif
}
inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
inline void v_pack_store(hfloat* ptr, const v_float32x4& v)
{
#if CV_FP16
__m128i fp16_value = _mm_cvtps_ph(v.val, 0);
modules/core/include/opencv2/core/hal/intrin_vsx.hpp
+2 -2
View File
@@ -1361,7 +1361,7 @@ inline v_float32x4 v_pack_triplets(const v_float32x4& vec)
/////// FP16 support ////////
inline v_float32x4 v_load_expand(const float16_t* ptr)
inline v_float32x4 v_load_expand(const hfloat* ptr)
{
vec_ushort8 vf16 = vec_ld_l8((const ushort*)ptr);
#if CV_VSX3 && defined(vec_extract_fp_from_shorth)
@@ -1388,7 +1388,7 @@ inline v_float32x4 v_load_expand(const float16_t* ptr)
#endif
}
inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
inline void v_pack_store(hfloat* ptr, const v_float32x4& v)
{
// fixme: Is there any builtin op or intrinsic that cover "xvcvsphp"?
#if CV_VSX3 && !defined(CV_COMPILER_VSX_BROKEN_ASM)
modules/core/include/opencv2/core/hal/intrin_wasm.hpp
+6 -6
View File
@@ -2754,7 +2754,7 @@ inline v_float32x4 v_broadcast_element(const v_float32x4& a)
////////////// FP16 support ///////////////////////////
inline v_float32x4 v_load_expand(const float16_t* ptr)
inline v_float32x4 v_load_expand(const hfloat* ptr)
{
float a[4];
for (int i = 0; i < 4; i++)
@@ -2762,14 +2762,14 @@ inline v_float32x4 v_load_expand(const float16_t* ptr)
return v_float32x4(wasm_v128_load(a));
}
inline void v_pack_store(float16_t* ptr, const v_float32x4& v)
inline void v_pack_store(hfloat* ptr, const v_float32x4& v)
{
double v_[4];
wasm_v128_store(v_, v.val);
ptr[0] = float16_t(v_[0]);
ptr[1] = float16_t(v_[1]);
ptr[2] = float16_t(v_[2]);
ptr[3] = float16_t(v_[3]);
ptr[0] = hfloat(v_[0]);
ptr[1] = hfloat(v_[1]);
ptr[2] = hfloat(v_[2]);
ptr[3] = hfloat(v_[3]);
}
inline void v_cleanup() {}
modules/core/include/opencv2/core/saturate.hpp
+11 -11
View File
@@ -158,20 +158,20 @@ template<> inline uint64 saturate_cast<uint64>(int64 v) { return (uint64)st
template<> inline int64 saturate_cast<int64>(uint64 v) { return (int64)std::min(v, (uint64)LLONG_MAX); }
/** @overload */
template<typename _Tp> static inline _Tp saturate_cast(float16_t v) { return saturate_cast<_Tp>((float)v); }
template<typename _Tp> static inline _Tp saturate_cast(hfloat v) { return saturate_cast<_Tp>((float)v); }
// in theory, we could use a LUT for 8u/8s->16f conversion,
// but with hardware support for FP32->FP16 conversion the current approach is preferable
template<> inline float16_t saturate_cast<float16_t>(uchar v) { return float16_t((float)v); }
template<> inline float16_t saturate_cast<float16_t>(schar v) { return float16_t((float)v); }
template<> inline float16_t saturate_cast<float16_t>(ushort v) { return float16_t((float)v); }
template<> inline float16_t saturate_cast<float16_t>(short v) { return float16_t((float)v); }
template<> inline float16_t saturate_cast<float16_t>(unsigned v){ return float16_t((float)v); }
template<> inline float16_t saturate_cast<float16_t>(int v) { return float16_t((float)v); }
template<> inline float16_t saturate_cast<float16_t>(uint64 v) { return float16_t((float)v); }
template<> inline float16_t saturate_cast<float16_t>(int64 v) { return float16_t((float)v); }
template<> inline float16_t saturate_cast<float16_t>(float v) { return float16_t(v); }
template<> inline float16_t saturate_cast<float16_t>(double v) { return float16_t((float)v); }
template<> inline hfloat saturate_cast<hfloat>(uchar v) { return hfloat((float)v); }
template<> inline hfloat saturate_cast<hfloat>(schar v) { return hfloat((float)v); }
template<> inline hfloat saturate_cast<hfloat>(ushort v) { return hfloat((float)v); }
template<> inline hfloat saturate_cast<hfloat>(short v) { return hfloat((float)v); }
template<> inline hfloat saturate_cast<hfloat>(unsigned v){ return hfloat((float)v); }
template<> inline hfloat saturate_cast<hfloat>(int v) { return hfloat((float)v); }
template<> inline hfloat saturate_cast<hfloat>(uint64 v) { return hfloat((float)v); }
template<> inline hfloat saturate_cast<hfloat>(int64 v) { return hfloat((float)v); }
template<> inline hfloat saturate_cast<hfloat>(float v) { return hfloat(v); }
template<> inline hfloat saturate_cast<hfloat>(double v) { return hfloat((float)v); }
//! @}
modules/core/include/opencv2/core/traits.hpp
+3 -3
View File
@@ -261,10 +261,10 @@ public:
};
};
template<> class DataType<float16_t>
template<> class DataType<hfloat>
{
public:
typedef float16_t value_type;
typedef hfloat value_type;
typedef float work_type;
typedef value_type channel_type;
typedef value_type vec_type;
@@ -347,7 +347,7 @@ template<> class TypeDepth<CV_64F>
template<> class TypeDepth<CV_16F>
{
enum { depth = CV_16F };
typedef float16_t value_type;
typedef hfloat value_type;
};
#endif