mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-01-18 22:44:02 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Change the lsx to baseline features.

Browse Source 
This patch change lsx to baseline feature, and lasx to dispatch
feature. Additionally, the runtime detection methods for lasx and
lsx have been modified.
This commit is contained in:
Hao Chen 2023-11-08 09:07:19 +08:00
parent 81907af74c
commit c19adb4953
5 changed files with 49 additions and 52 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
cmake/OpenCVCompilerOptimizations.cmake
View File

@ -403,11 +403,8 @@ elseif(LOONGARCH64)
ocv_update(CPU_KNOWN_OPTIMIZATIONS "LSX;LASX")
ocv_update(CPU_LSX_FLAGS_ON "-mlsx")
ocv_update(CPU_LASX_FLAGS_ON "-mlasx")
if("${CPU_BASELINE_DISABLE}" STREQUAL "LASX")
set(CPU_BASELINE "LSX" CACHE STRING "${HELP_CPU_BASELINE}")
else()
set(CPU_BASELINE "LASX" CACHE STRING "${HELP_CPU_BASELINE}")
endif()
set(CPU_BASELINE "LSX" CACHE STRING "${HELP_CPU_BASELINE}")
set(CPU_DISPATCH "LASX" CACHE STRING "${HELP_CPU_DISPATCH}")
endif()

26
modules/core/CMakeLists.txt
View File

@ -1,21 +1,21 @@
set(the_description "The Core Functionality")
ocv_add_dispatched_file(mathfuncs_core SSE2 AVX AVX2)
ocv_add_dispatched_file(stat SSE4_2 AVX2)
ocv_add_dispatched_file(arithm SSE2 SSE4_1 AVX2 VSX3)
ocv_add_dispatched_file(convert SSE2 AVX2 VSX3)
ocv_add_dispatched_file(convert_scale SSE2 AVX2)
ocv_add_dispatched_file(count_non_zero SSE2 AVX2)
ocv_add_dispatched_file(has_non_zero SSE2 AVX2)
ocv_add_dispatched_file(matmul SSE2 SSE4_1 AVX2 AVX512_SKX NEON_DOTPROD)
ocv_add_dispatched_file(mean SSE2 AVX2)
ocv_add_dispatched_file(merge SSE2 AVX2)
ocv_add_dispatched_file(split SSE2 AVX2)
ocv_add_dispatched_file(sum SSE2 AVX2)
ocv_add_dispatched_file(mathfuncs_core SSE2 AVX AVX2 LASX)
ocv_add_dispatched_file(stat SSE4_2 AVX2 LASX)
ocv_add_dispatched_file(arithm SSE2 SSE4_1 AVX2 VSX3 LASX)
ocv_add_dispatched_file(convert SSE2 AVX2 VSX3 LASX)
ocv_add_dispatched_file(convert_scale SSE2 AVX2 LASX)
ocv_add_dispatched_file(count_non_zero SSE2 AVX2 LASX)
ocv_add_dispatched_file(has_non_zero SSE2 AVX2 LASX )
ocv_add_dispatched_file(matmul SSE2 SSE4_1 AVX2 AVX512_SKX NEON_DOTPROD LASX)
ocv_add_dispatched_file(mean SSE2 AVX2 LASX)
ocv_add_dispatched_file(merge SSE2 AVX2 LASX)
ocv_add_dispatched_file(split SSE2 AVX2 LASX)
ocv_add_dispatched_file(sum SSE2 AVX2 LASX)
# dispatching for accuracy tests
ocv_add_dispatched_file_force_all(test_intrin128 TEST SSE2 SSE3 SSSE3 SSE4_1 SSE4_2 AVX FP16 AVX2 AVX512_SKX)
ocv_add_dispatched_file_force_all(test_intrin256 TEST AVX2 AVX512_SKX)
ocv_add_dispatched_file_force_all(test_intrin256 TEST AVX2 AVX512_SKX LASX)
ocv_add_dispatched_file_force_all(test_intrin512 TEST AVX512_SKX)

6
modules/core/include/opencv2/core/hal/intrin.hpp
View File

@ -246,12 +246,6 @@ using namespace CV_CPU_OPTIMIZATION_HAL_NAMESPACE;
#include "opencv2/core/hal/intrin_lsx.hpp"
#elif CV_LASX
#if !defined(CV_FORCE_SIMD128_CPP)
#define CV_FORCE_SIMD128_CPP 1
#endif
#include "opencv2/core/hal/intrin_cpp.hpp"
#else
#include "opencv2/core/hal/intrin_cpp.hpp"

47
modules/core/include/opencv2/core/hal/intrin_lasx.hpp
View File

@ -1419,20 +1419,6 @@ inline v_uint32x8 v_popcount(const v_int32x8& a)
inline v_uint64x4 v_popcount(const v_int64x4& a)
{ return v_popcount(v_reinterpret_as_u64(a)); }
/** Mask **/
#define OPENCV_HAL_IMPL_REINTERPRET_INT(ft, tt) \
inline tt reinterpret_int(ft x) { union { ft l; tt i; } v; v.l = x; return v.i; }
OPENCV_HAL_IMPL_REINTERPRET_INT(uchar, schar)
OPENCV_HAL_IMPL_REINTERPRET_INT(schar, schar)
OPENCV_HAL_IMPL_REINTERPRET_INT(ushort, short)
OPENCV_HAL_IMPL_REINTERPRET_INT(short, short)
OPENCV_HAL_IMPL_REINTERPRET_INT(unsigned, int)
OPENCV_HAL_IMPL_REINTERPRET_INT(int, int)
OPENCV_HAL_IMPL_REINTERPRET_INT(float, int)
OPENCV_HAL_IMPL_REINTERPRET_INT(uint64, int64)
OPENCV_HAL_IMPL_REINTERPRET_INT(int64, int64)
OPENCV_HAL_IMPL_REINTERPRET_INT(double, int64)
inline int v_signmask(const v_int8x32& a)
{
__m256i result = __lasx_xvmskltz_b(a.val);
@ -2151,7 +2137,8 @@ template<int n> inline
void v_rshr_pack_store(uchar* ptr, const v_uint16x16& a)
{
__m256i res = __lasx_xvssrlrni_bu_h(a.val, a.val, n);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
template<int n> inline
@ -2165,7 +2152,8 @@ template<int n> inline
void v_rshr_pack_u_store(uchar* ptr, const v_int16x16& a)
{
__m256i res = __lasx_xvssrarni_bu_h(a.val, a.val, n);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
template<int n> inline
@ -2179,7 +2167,8 @@ template<int n> inline
void v_rshr_pack_store(schar* ptr, const v_int16x16& a)
{
__m256i res = __lasx_xvssrarni_b_h(a.val, a.val, n);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
// 32
@ -2198,7 +2187,8 @@ inline void v_pack_store(short* ptr, const v_int32x8& a)
inline void v_pack_store(ushort* ptr, const v_uint32x8& a)
{
__m256i res = __lasx_xvssrlrni_hu_w(a.val, a.val, 0);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
inline void v_pack_u_store(ushort* ptr, const v_int32x8& a)
@ -2212,7 +2202,8 @@ template<int n> inline
void v_rshr_pack_store(ushort* ptr, const v_uint32x8& a)
{
__m256i res = __lasx_xvssrlrni_hu_w(a.val, a.val, n);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
template<int n> inline
@ -2223,7 +2214,8 @@ template<int n> inline
void v_rshr_pack_u_store(ushort* ptr, const v_int32x8& a)
{
__m256i res = __lasx_xvssrarni_hu_w(a.val, a.val, n);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
template<int n> inline
@ -2234,7 +2226,8 @@ template<int n> inline
void v_rshr_pack_store(short* ptr, const v_int32x8& a)
{
__m256i res = __lasx_xvssrarni_h_w(a.val, a.val, n);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
// 64
@ -2263,7 +2256,11 @@ v_uint32x8 v_rshr_pack(const v_uint64x4& a, const v_uint64x4& b)
template<int n> inline
void v_rshr_pack_store(unsigned* ptr, const v_uint64x4& a)
{ __lsx_vst(_v256_shuffle_odd_64(__lasx_xvsrlrni_w_d(a.val, a.val, n)), ptr, 0); }
{
__m256i res = __lasx_xvsrlrni_w_d(a.val, a.val, n);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
template<int n> inline
v_int32x8 v_rshr_pack(const v_int64x4& a, const v_int64x4& b)
@ -2271,7 +2268,11 @@ v_int32x8 v_rshr_pack(const v_int64x4& a, const v_int64x4& b)
template<int n> inline
void v_rshr_pack_store(int* ptr, const v_int64x4& a)
{ __lsx_vst(_v256_shuffle_odd_64(__lasx_xvsrarni_w_d(a.val, a.val, n)), ptr, 0); }
{
__m256i res = __lasx_xvsrarni_w_d(a.val, a.val, n);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
// pack boolean
inline v_uint8x32 v_pack_b(const v_uint16x16& a, const v_uint16x16& b)

15
modules/core/src/system.cpp
View File

@ -154,6 +154,12 @@ void* allocSingletonNewBuffer(size_t size) { return malloc(size); }
# endif
#endif
#if defined __loongarch64
#include "sys/auxv.h"
#define LA_HWCAP_LSX (1<<4)
#define LA_HWCAP_LASX (1<<5)
#endif
#if defined _WIN32 || defined WINCE
#ifndef _WIN32_WINNT // This is needed for the declaration of TryEnterCriticalSection in winbase.h with Visual Studio 2005 (and older?)
#define _WIN32_WINNT 0x0400 // http://msdn.microsoft.com/en-us/library/ms686857(VS.85).aspx
@ -704,12 +710,11 @@ struct HWFeatures
have[CV_CPU_RVV] = true;
#endif
#if defined __loongarch_sx
have[CV_CPU_LSX] = true;
#endif
#if defined __loongarch64 && defined __linux__
int flag = (int)getauxval(AT_HWCAP);
#if defined __loongarch_asx
have[CV_CPU_LASX] = true;
have[CV_CPU_LSX] = (flag & LA_HWCAP_LSX) != 0;
have[CV_CPU_LASX] = (flag & LA_HWCAP_LASX) != 0;
#endif
bool skip_baseline_check = false;