add isa check

This commit is contained in:
Nitload
2026-01-04 07:56:37 +08:00
parent 53085d35d2
commit e1d2a8707d
13 changed files with 123 additions and 533 deletions
+7
View File
@@ -61,14 +61,20 @@
<PropertyGroup Label="UserMacros" />
<PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
<OutDir>$(SolutionDir)build</OutDir>
<RunCodeAnalysis>true</RunCodeAnalysis>
<EnableClangTidyCodeAnalysis>false</EnableClangTidyCodeAnalysis>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
<OutDir>$(SolutionDir)build</OutDir>
<GenerateManifest>false</GenerateManifest>
<RunCodeAnalysis>true</RunCodeAnalysis>
<EnableClangTidyCodeAnalysis>false</EnableClangTidyCodeAnalysis>
</PropertyGroup>
<PropertyGroup Condition="'$(Configuration)|$(Platform)'=='PerformanceDebug|x64'">
<OutDir>$(SolutionDir)build</OutDir>
<GenerateManifest>false</GenerateManifest>
<RunCodeAnalysis>true</RunCodeAnalysis>
<EnableClangTidyCodeAnalysis>false</EnableClangTidyCodeAnalysis>
</PropertyGroup>
<ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
<ClCompile>
@@ -175,6 +181,7 @@
<ClInclude Include="sn-head\sn_internal.h" />
<ClInclude Include="sn-head\sn_intrinsic.h" />
<ClInclude Include="sn-head\sn_matkernel.h" />
<ClInclude Include="sn-head\sn_target.h" />
<ClInclude Include="sn-head\StgSharpNative.h" />
<ClInclude Include="sn-head\wgl.h" />
</ItemGroup>
@@ -92,6 +92,9 @@
<ClInclude Include="lib\stbi\stb_image.h">
<Filter>头文件</Filter>
</ClInclude>
<ClInclude Include="sn-head\sn_target.h">
<Filter>头文件</Filter>
</ClInclude>
</ItemGroup>
<ItemGroup>
<Library Include="lib\glfw\glfw3.lib">
+35 -1
View File
@@ -1,4 +1,4 @@
#ifndef SSC_INTRIN
#ifndef SSC_INTRIN
#define SSC_INTRIN
#define SSC_INTRIN_HEAD
@@ -328,4 +328,38 @@ typedef enum most_advanced_instruction {
#endif
} most_advanced_instruction;
#include "sn_target.h"
#if SN_IS_ARCH(SN_ARCH_X86_64)
typedef enum intrinsic_level {
// No SIMD support, will cause error if used in matrix operations
SIMD0 = 0,
// SSE4.2 support
SIMD1 = 1,
// AVX2 support
SIMD2 = 2,
// AVX2 and AVX2-FMA support
SIMD3 = 3,
// AVX2-16C support based on SIMD3
SIMD4 = 4,
// AVX-512-dual support
SIMD5 = 5,
// AVX-512-native support
SIMD6 = 6,
// AVX-512 with 8/16-wide support
SIMD7 = 7,
// AMX support
SIMD8 = 8,
// AVX-10 support
SIMD9 = 9,
// future use
SIMD_MAX = 10
} intrinsic_level;
#else
typedef enum intrinsic_level {
} intrinsic_level;
#endif
#endif
+61
View File
@@ -0,0 +1,61 @@
#ifndef SN_TARGET
#define SN_TARGET
typedef enum SN_OS_T {
SN_OS_WINDOWS,
SN_OS_LINUX,
SN_OS_MACOS,
SN_OS_IOS,
SN_OS_ANDROID,
SN_OS_UNKNOWN
} SN_OS_T;
typedef enum SN_ARCH_T {
SN_ARCH_X86_64,
SN_ARCH_ARM64,
SN_ARCH_RISCV64,
SN_ARCH_PPC64le,
} SN_ARCH_T;
/*
* Target selection policy:
* - Default (no SN_EXTERN_TARGET): fixed Windows + x86_64.
* - External (SN_EXTERN_TARGET): build system must provide SN_OS and SN_ARCH.
*/
#ifndef SN_EXTERN_TARGET
#define SN_OS_TARGET SN_OS_WINDOWS
#define SN_ARCH_TARGET SN_ARCH_X86_64
#else
#ifndef SN_OS_TARGET
#error "SN_EXTERN_TARGET is defined but SN_OS is not provided."
#endif
#ifndef SN_ARCH_TARGET
#error "SN_EXTERN_TARGET is defined but SN_ARCH is not provided."
#endif
/* Validate externally supplied values are within known enum constants. */
#if (SN_OS_TARGET != SN_OS_WINDOWS) && (SN_OS_TARGET != SN_OS_LINUX) && (SN_OS_TARGET != SN_OS_MACOS) && \
(SN_OS_TARGET != SN_OS_IOS) && (SN_OS_TARGET != SN_OS_ANDROID) && (SN_OS_TARGET != SN_OS_UNKNOWN)
#error "SN_OS has an invalid value. Use SN_OS_<...> enum constants."
#endif
#if (SN_ARCH_TARGET != SN_ARCH_X86_64) && (SN_ARCH_TARGET != SN_ARCH_ARM64) && (SN_ARCH_TARGET != SN_ARCH_RISCV64) && \
(SN_ARCH_TARGET != SN_ARCH_PPC64le)
#error "SN_ARCH has an invalid value. Use SN_ARCH_<...> enum constants."
#endif
#endif
#ifndef SN_IS_OS
#define SN_IS_OS(os) ((SN_OS_TARGET) == (os))
#endif
#ifndef SN_IS_ARCH
#define SN_IS_ARCH(arch) ((SN_ARCH_TARGET) == (arch))
#endif
#endif // SN_TARGET
-32
View File
@@ -1,32 +0,0 @@
#ifdef _MSC_VER
#pragma once
#endif
#include <xmmintrin.h>
#ifndef SSC_INTERNAL
#define SSC_INTERNAL
#ifdef _MSC_VER
#define _CRT_SECURE_NO_WARNINGS 1
#define WIN32_LEAN_AND_MEAN
#define SSCDECL __cdecl
#define SSCAPI __declspec(dllexport)
#define FORCEINLINE __forceinline
#include <windows.h>
#elif defined(__clang__) || defined(__GNUC__)
#define SSCDECL __attribute__((cdecl))
#define SSCAPI __attribute__((dllexport))
#define FORCEINLINE __attribute__((always_inline))
#else
#define SSCAPI
#define SSCDECL
#define FORCEINLINE
#endif
#define SSCAPI_DEFINE
#define PRIVATE static
#define INTERNAL extern
#endif //
-176
View File
@@ -1,176 +0,0 @@
#ifdef _MSC_VER
#pragma once
#endif
#ifndef SSC_INTRIN
#define SSC_INTRIN
#include "StgSharpC.h"
#include <immintrin.h>
#include <smmintrin.h>
#ifdef _MSC_VER
#include <intrin.h>
#include <isa_availability.h>
#endif
#define LASTTWO(x) (x & 3)
#define LAST(x) (x & 3)
#define get_align_begin(x) ((x + 15) & (~15))
#define SSE_INSERT(source, target, reset3, reset2, reset1, reset0) \
(LASTTWO(source - 1) << 6) | (LASTTWO(target - 1) << 4) | (LAST(reset3) << 3) | \
(LAST(reset2) << 2) | (LAST(reset1) << 1) | LAST(reset0)
#define SSE_INSERT_SIMPLE(source, target) SSE_INSERT(source, target, 0, 0, 0, 0)
#define M256_PERMUTE(s3, s2, s1, s0) (LASTTWO(s3) << 6) | (LASTTWO(s2) << 4) | (LASTTWO(s1) << 2) | (LASTTWO(s0)
#define zero_vec _mm_setzero_ps()
#define zero_vec_256 _mm256_setzero_ps()
typedef union uint64_u {
uint64_t u64;
uint32_t u32[2];
} uint64_u;
typedef union mat_kernel{
__m128 xmm[4];
__m256 ymm[2];
__m512 zmm[1];
float m[4][4];
}mat_kernel;
typedef union column_2 {
__m256 stream;
__m128 column[2];
float m[4][2];
} __2_columnset;
typedef union column_3 {
__m256 part;
__m128 column[3];
float m[4][3];
} __3_columnset;
typedef union column_4 {
__m512 stream;
__m256 half[2];
__m128 column[4];
float m[4][4];
} __4_columnset;
typedef void(SSCDECL *VECTORNORMALIZEPROC)(__m128 *source, __m128 *target);
typedef void(SSCDECL *DOTPROC)(void *transpose, __m128 *vector, __m128 *ans);
typedef void(SSCDECL *MATRIXTRANSPOSEPROC)(void *source, void *target);
typedef void(SSCDECL *MATRIXDETPROC)(void *mat, void *transpose);
typedef void(SSCDECL *MATARITHMETIC)(void *left, void *right, void *ans);
typedef void(SSCDECL *SCALERMATARITHMETIC)(void *left, void *right, void *ans);
typedef uint64_t(SSCDECL *FACTORIALROC)(int n);
#pragma region matix function
INTERNAL void normalize(__m128 *source, __m128 *target);
INTERNAL void SSCDECL addmatrix2_sse(__2_columnset *left, __2_columnset *right, __2_columnset *ans);
INTERNAL void SSCDECL addmatrix2_avx(__2_columnset *left, __2_columnset *right, __2_columnset *ans);
INTERNAL void SSCDECL submatrix2_sse(__2_columnset *left, __2_columnset *right, __2_columnset *ans);
INTERNAL void SSCDECL addmatrix2_avx(__2_columnset *left, __2_columnset *right, __2_columnset *ans);
INTERNAL void SSCDECL transpose23(__2_columnset const *source, __3_columnset *target);
INTERNAL void SSCDECL transpose24(__2_columnset const *source, __4_columnset *target);
INTERNAL void SSCDECL addmatrix3_sse(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SSCDECL addmatrix3_avx(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SSCDECL addmatrix3_512(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SSCDECL submatrix3_sse(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SSCDECL submatrix3_avx(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SSCDECL submatrix3_512(__3_columnset *left, __3_columnset *right, __3_columnset *ans);
INTERNAL void SSCDECL transpose32(__3_columnset const *source, __2_columnset *target);
INTERNAL void SSCDECL transpose33(__3_columnset const *source, __3_columnset *target);
INTERNAL void SSCDECL transpose34(__3_columnset const *source, __4_columnset *target);
INTERNAL float SSCDECL det_mat3(__3_columnset *matrix, __3_columnset *tranpose);
INTERNAL void SSCDECL addmatrix4_sse(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SSCDECL addmatrix4_avx(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SSCDECL addmatrix4_512(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SSCDECL submatrix4_sse(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SSCDECL submatrix4_avx(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SSCDECL submatrix4_512(__4_columnset *left, __4_columnset *right, __4_columnset *ans);
INTERNAL void SSCDECL transpose42(__4_columnset const *source, __2_columnset *target);
INTERNAL void SSCDECL transpose43(__4_columnset const *source, __3_columnset *target);
INTERNAL void SSCDECL transpose44(__4_columnset const *source, __4_columnset *target);
INTERNAL float SSCDECL det_mat4(__4_columnset *matrix, __4_columnset *transpose);
#pragma endregion
#pragma region dot
INTERNAL void FORCEINLINE SSCDECL dot_41_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void FORCEINLINE SSCDECL dot_31_sse(__3_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void FORCEINLINE SSCDECL dot_41_avx(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void FORCEINLINE SSCDECL dot_31_avx(__3_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_32_sse(__3_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_32_avx(__3_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_42_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_42_avx(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_43_sse(__4_columnset *transpose, __m128 *vector, __m128 *ans);
INTERNAL void SSCDECL dot_43_avx(__4_columnset *transpose, __m128 *vector, __m128 *ans);
//INTERNAL void SSCDECL dot_43_512(__4_columnset *transpose, __m128 *vector, __m128 *ans);
#pragma endregion
#pragma region scaler_simd
INTERNAL uint64_t SSCDECL factorial_simd_sse(int n);
#pragma endregion
#pragma region string
INTERNAL int SSCDECL city_hash_simplify_sse(byte const *str, int length);
INTERNAL int SSCDECL index_pair_sse(short const *str, uint32_t target, int length);
#pragma endregion
typedef struct ssc_intrinsic {
MATARITHMETIC add_m2;
MATARITHMETIC add_m3;
MATARITHMETIC add_m4;
int (*city_hash_simplify)(byte *str, int length);
MATRIXDETPROC det_matrix33;
MATRIXDETPROC det_matrix44;
DOTPROC dot_31;
DOTPROC dot_32;
DOTPROC dot_41;
DOTPROC dot_42;
DOTPROC dot_43;
FACTORIALROC factorial_simd;
int (*index_pair)(short const *str, uint32_t target, int length);
VECTORNORMALIZEPROC normalize_3;
MATARITHMETIC sub_m2;
MATARITHMETIC sub_m3;
MATARITHMETIC sub_m4;
MATRIXTRANSPOSEPROC transpose_23;
MATRIXTRANSPOSEPROC transpose_24;
MATRIXTRANSPOSEPROC transpose_32;
MATRIXTRANSPOSEPROC transpose_33;
MATRIXTRANSPOSEPROC transpose_34;
MATRIXTRANSPOSEPROC transpose_42;
MATRIXTRANSPOSEPROC transpose_43;
MATRIXTRANSPOSEPROC transpose_44;
} ssc_intrinsic;
typedef enum most_advanced_instruction {
#ifdef _MSC_VER
SSE = __IA_SUPPORT_SSE42,
AVX256 = __IA_SUPPORT_VECTOR256,
AVX512 = __IA_SUPPORT_VECTOR512,
NEON
#else
SSE,
AVX256,
AVX512,
NEON
#endif
} most_advanced_instruction;
#endif // !SSC_INTRIN
+6
View File
@@ -1,3 +1,7 @@
#include "sn_target.h"
#if SN_IS_ARCH(SN_ARCH_X86_64)
#include "sn_intrinsic.h"
DECLARE_BUF_PROC_ANS_SCALAR(float, sse, fill)
@@ -315,3 +319,5 @@ DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(double, 512, scalar_mul)
ans[i].f64_z[1] = c1;
}
}
#endif
+1
View File
@@ -12,6 +12,7 @@ PRIVATE void load_intrinsic_sse(sn_intrinsic *context);
PRIVATE void load_intrinsic_avx(sn_intrinsic *context);
PRIVATE void load_intrinsic_512(sn_intrinsic *context);
PRIVATE int check_isa(most_advanced_instruction instruction_family)
{
if (instruction_family != NEON) {