This commit is contained in:
Nitload
2025-08-25 09:50:30 +08:00
parent bd0e31e922
commit 1359ff4ecc
1630 changed files with 738486 additions and 2399 deletions
+73
View File
@@ -0,0 +1,73 @@
#ifdef _MSC_VER
#pragma once
#endif
#ifndef SSC
#define SSC
#include "ssc_internal.h"
#define STBIDEF SSCAPI
#include "stb_image.h"
#include "gl.h"
#include "wgl.h"
#include "glfw3.h"
#include <immintrin.h>
#include <assert.h>
#ifndef SSCAPI_DEFINE
#ifdef _WIN32
#define SSCAPI __declspec(dllexport)
#elif defined _LINUX_
#define SSCAPI __attribute__((dllexport))
#elif defined _APPLE_
#define SSCAPI __attribute__((visibility("default")))
#else
#define SSCAPI
#endif
#endif // ! SSCAPI_DEFINE
#if __cplusplus
extern "C" {
#endif
#define byte char
#define cs_char uint16_t
#define ssc_null_assert(ptr) assert(ptr != NULL)
#define ALIGN(simdVec) _mm_loadu_ps(&simdVec)
#define ALIGN256(simdVec) _mm256_loadu_ps(&simdVec)
#define ALIGN512(simdVec) _mm512_loadu_ps(&simdVec)
typedef struct Image {
int width;
int height;
int channel;
char *pixelPtr;
} Image;
typedef char *(*imageLoader)(char const *filename, int *x, int *y, int *channels_in_file,
int desired_channels);
SSCAPI char infolog[512];
SSCAPI int SSCDECL glCheckShaderStat(GladGLContext *context, uint64_t shaderHandle, int key,
char **logRef);
SSCAPI void SSCDECL initGL(int majorVersion, int minorVersion);
SSCAPI unsigned int SSCDECL linkShaderProgram(GladGLContext *context, GLuint shadertype);
SSCAPI void SSCDECL loadImageData(char *location, Image *out, imageLoader loader);
SSCAPI GLFWglproc SSCDECL loadGlfuncDefault(char *procName);
SSCAPI void SSCDECL unloadImageData(Image *out);
SSCAPI char *SSCDECL readLog(void);
#pragma endregion
SSCAPI int SSCDECL load_intrinsic_function(void *intrinsic_context);
#ifdef __cplusplus
}
#endif
#endif
File diff suppressed because it is too large Load Diff
+632
View File
@@ -0,0 +1,632 @@
/*************************************************************************
* GLFW 3.4 - www.glfw->org
* A library for OpenGL, window and input
*------------------------------------------------------------------------
* Copyright (c) 2002-2006 Marcus Geelnard
* Copyright (c) 2006-2018 Camilla Löwy <elmindreda@glfw->org>
*
* This software is provided 'as-is', without any express or implied
* warranty. In no event will the authors be held liable for any damages
* arising from the use of this software.
*
* Permission is granted to anyone to use this software for any purpose,
* including commercial applications, and to alter it and redistribute it
* freely, subject to the following restrictions:
*
* 1. The origin of this software must not be misrepresented; you must not
* claim that you wrote the original software. If you use this software
* in a product, an acknowledgment in the product documentation would
* be appreciated but is not required.
*
* 2. Altered source versions must be plainly marked as such, and must not
* be misrepresented as being the original software.
*
* 3. This notice may not be removed or altered from any source
* distribution.
*
*************************************************************************/
#ifndef _glfw3_native_h_
#define _glfw3_native_h_
#ifdef __cplusplus
extern "C" {
#endif
/*************************************************************************
* Doxygen documentation
*************************************************************************/
/*! @file glfw3native.h
* @brief The header of the native access functions.
*
* This is the header file of the native access functions. See @ref native for
* more information.
*/
/*! @defgroup native Native access
* @brief Functions related to accessing native handles.
*
* **By using the native access functions you assert that you know what you're
* doing and how to fix problems caused by using them. If you don't, you
* shouldn't be using them.**
*
* Before the inclusion of @ref glfw3native.h, you may define zero or more
* window system API macro and zero or more context creation API macros.
*
* The chosen backends must match those the library was compiled for. Failure
* to do this will cause a link-time error.
*
* The available window API macros are:
* * `GLFW_EXPOSE_NATIVE_WIN32`
* * `GLFW_EXPOSE_NATIVE_COCOA`
* * `GLFW_EXPOSE_NATIVE_X11`
* * `GLFW_EXPOSE_NATIVE_WAYLAND`
*
* The available context API macros are:
* * `GLFW_EXPOSE_NATIVE_WGL`
* * `GLFW_EXPOSE_NATIVE_NSGL`
* * `GLFW_EXPOSE_NATIVE_GLX`
* * `GLFW_EXPOSE_NATIVE_EGL`
* * `GLFW_EXPOSE_NATIVE_OSMESA`
*
* These macros select which of the native access functions that are declared
* and which platform-specific headers to include. It is then up your (by
* definition platform-specific) code to handle which of these should be
* defined.
*
* If you do not want the platform-specific headers to be included, define
* `GLFW_NATIVE_INCLUDE_NONE` before including the @ref glfw3native.h header.
*
* @code
* #define GLFW_EXPOSE_NATIVE_WIN32
* #define GLFW_EXPOSE_NATIVE_WGL
* #define GLFW_NATIVE_INCLUDE_NONE
* #include <GLFW/glfw3native.h>
* @endcode
*/
/*************************************************************************
* System headers and types
*************************************************************************/
#if !defined(GLFW_NATIVE_INCLUDE_NONE)
#if defined(GLFW_EXPOSE_NATIVE_WIN32) || defined(GLFW_EXPOSE_NATIVE_WGL)
/* This is a workaround for the fact that glfw3.h needs to export APIENTRY (for
* example to allow applications to correctly declare a GL_KHR_debug callback)
* but windows.h assumes no one will define APIENTRY before it does
*/
#if defined(GLFW_APIENTRY_DEFINED)
#undef APIENTRY
#undef GLFW_APIENTRY_DEFINED
#endif
#include <windows.h>
#endif
#if defined(GLFW_EXPOSE_NATIVE_COCOA) || defined(GLFW_EXPOSE_NATIVE_NSGL)
#if defined(__OBJC__)
#import <Cocoa/Cocoa.h>
#else
#include <ApplicationServices/ApplicationServices.h>
#include <objc/objc.h>
#endif
#endif
#if defined(GLFW_EXPOSE_NATIVE_X11) || defined(GLFW_EXPOSE_NATIVE_GLX)
#include <X11/Xlib.h>
#include <X11/extensions/Xrandr.h>
#endif
#if defined(GLFW_EXPOSE_NATIVE_WAYLAND)
#include <wayland-client.h>
#endif
#if defined(GLFW_EXPOSE_NATIVE_WGL)
/* WGL is declared by windows.h */
#endif
#if defined(GLFW_EXPOSE_NATIVE_NSGL)
/* NSGL is declared by Cocoa.h */
#endif
#if defined(GLFW_EXPOSE_NATIVE_GLX)
/* This is a workaround for the fact that glfw3.h defines GLAPIENTRY because by
* default it also acts as an OpenGL header
* However, glx.h will include gl.h, which will define it unconditionally
*/
#if defined(GLFW_GLAPIENTRY_DEFINED)
#undef GLAPIENTRY
#undef GLFW_GLAPIENTRY_DEFINED
#endif
#include <GL/glx.h>
#endif
#if defined(GLFW_EXPOSE_NATIVE_EGL)
#include <EGL/egl.h>
#endif
#if defined(GLFW_EXPOSE_NATIVE_OSMESA)
/* This is a workaround for the fact that glfw3.h defines GLAPIENTRY because by
* default it also acts as an OpenGL header
* However, osmesa.h will include gl.h, which will define it unconditionally
*/
#if defined(GLFW_GLAPIENTRY_DEFINED)
#undef GLAPIENTRY
#undef GLFW_GLAPIENTRY_DEFINED
#endif
#include <GL/osmesa.h>
#endif
#endif /*GLFW_NATIVE_INCLUDE_NONE*/
/*************************************************************************
* Functions
*************************************************************************/
#if defined(GLFW_EXPOSE_NATIVE_WIN32)
/*! @brief Returns the adapter device name of the specified monitor.
*
* @return The UTF-8 encoded adapter device name (for example `\\.\DISPLAY1`)
* of the specified monitor, or `NULL` if an [error](@ref error_handling)
* occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.1.
*
* @ingroup native
*/
SSCAPI const char *glfwGetWin32Adapter(GLFWmonitor *monitor);
/*! @brief Returns the display device name of the specified monitor.
*
* @return The UTF-8 encoded display device name (for example
* `\\.\DISPLAY1\Monitor0`) of the specified monitor, or `NULL` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.1.
*
* @ingroup native
*/
SSCAPI const char *glfwGetWin32Monitor(GLFWmonitor *monitor);
/*! @brief Returns the `HWND` of the specified window.
*
* @return The `HWND` of the specified window, or `NULL` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED.
*
* @remark The `HDC` associated with the window can be queried with the
* [GetDC](https://docs.microsoft.com/en-us/windows/win32/api/winuser/nf-winuser-getdc)
* function.
* @code
* HDC dc = GetDC(glfwGetWin32Window(window));
* @endcode
* This DC is private and does not need to be released.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.0.
*
* @ingroup native
*/
SSCAPI HWND glfwGetWin32Window(GLFWwindow *window);
#endif
#if defined(GLFW_EXPOSE_NATIVE_WGL)
/*! @brief Returns the `HGLRC` of the specified window.
*
* @return The `HGLRC` of the specified window, or `NULL` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NO_WINDOW_CONTEXT and @ref
* GLFW_NOT_INITIALIZED.
*
* @remark The `HDC` associated with the window can be queried with the
* [GetDC](https://docs.microsoft.com/en-us/windows/win32/api/winuser/nf-winuser-getdc)
* function.
* @code
* HDC dc = GetDC(glfwGetWin32Window(window));
* @endcode
* This DC is private and does not need to be released.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.0.
*
* @ingroup native
*/
SSCAPI HGLRC glfwGetWGLContext(GLFWwindow *window);
#endif
#if defined(GLFW_EXPOSE_NATIVE_COCOA)
/*! @brief Returns the `CGDirectDisplayID` of the specified monitor.
*
* @return The `CGDirectDisplayID` of the specified monitor, or
* `kCGNullDirectDisplay` if an [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.1.
*
* @ingroup native
*/
SSCAPI CGDirectDisplayID glfwGetCocoaMonitor(GLFWmonitor *monitor);
/*! @brief Returns the `NSWindow` of the specified window.
*
* @return The `NSWindow` of the specified window, or `nil` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.0.
*
* @ingroup native
*/
SSCAPI id glfwGetCocoaWindow(GLFWwindow *window);
#endif
#if defined(GLFW_EXPOSE_NATIVE_NSGL)
/*! @brief Returns the `NSOpenGLContext` of the specified window.
*
* @return The `NSOpenGLContext` of the specified window, or `nil` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NO_WINDOW_CONTEXT and @ref
* GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.0.
*
* @ingroup native
*/
SSCAPI id glfwGetNSGLContext(GLFWwindow *window);
#endif
#if defined(GLFW_EXPOSE_NATIVE_X11)
/*! @brief Returns the `Display` used by GLFW.
*
* @return The `Display` used by GLFW, or `NULL` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.0.
*
* @ingroup native
*/
SSCAPI Display *glfwGetX11Display(void);
/*! @brief Returns the `RRCrtc` of the specified monitor.
*
* @return The `RRCrtc` of the specified monitor, or `None` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.1.
*
* @ingroup native
*/
SSCAPI RRCrtc glfwGetX11Adapter(GLFWmonitor *monitor);
/*! @brief Returns the `RROutput` of the specified monitor.
*
* @return The `RROutput` of the specified monitor, or `None` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.1.
*
* @ingroup native
*/
SSCAPI RROutput glfwGetX11Monitor(GLFWmonitor *monitor);
/*! @brief Returns the `Window` of the specified window.
*
* @return The `Window` of the specified window, or `None` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.0.
*
* @ingroup native
*/
SSCAPI Window glfwGetX11Window(GLFWwindow *window);
/*! @brief Sets the current primary selection to the specified string.
*
* @param[in] string A UTF-8 encoded string.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED and @ref
* GLFW_PLATFORM_ERROR.
*
* @pointer_lifetime The specified string is copied before this function
* returns.
*
* @thread_safety This function must only be called from the main thread.
*
* @sa @ref clipboard
* @sa glfwGetX11SelectionString
* @sa glfwSetClipboardString
*
* @since Added in version 3.3.
*
* @ingroup native
*/
SSCAPI void glfwSetX11SelectionString(const char *string);
/*! @brief Returns the contents of the current primary selection as a string.
*
* If the selection is empty or if its contents cannot be converted, `NULL`
* is returned and a @ref GLFW_FORMAT_UNAVAILABLE error is generated.
*
* @return The contents of the selection as a UTF-8 encoded string, or `NULL`
* if an [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED and @ref
* GLFW_PLATFORM_ERROR.
*
* @pointer_lifetime The returned string is allocated and freed by GLFW. You
* should not free it yourself. It is valid until the next call to @ref
* glfwGetX11SelectionString or @ref glfwSetX11SelectionString, or until the
* library is terminated.
*
* @thread_safety This function must only be called from the main thread.
*
* @sa @ref clipboard
* @sa glfwSetX11SelectionString
* @sa glfwGetClipboardString
*
* @since Added in version 3.3.
*
* @ingroup native
*/
SSCAPI const char *glfwGetX11SelectionString(void);
#endif
#if defined(GLFW_EXPOSE_NATIVE_GLX)
/*! @brief Returns the `GLXContext` of the specified window.
*
* @return The `GLXContext` of the specified window, or `NULL` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NO_WINDOW_CONTEXT and @ref
* GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.0.
*
* @ingroup native
*/
SSCAPI GLXContext glfwGetGLXContext(GLFWwindow *window);
/*! @brief Returns the `GLXWindow` of the specified window.
*
* @return The `GLXWindow` of the specified window, or `None` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NO_WINDOW_CONTEXT and @ref
* GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.2.
*
* @ingroup native
*/
SSCAPI GLXWindow glfwGetGLXWindow(GLFWwindow *window);
#endif
#if defined(GLFW_EXPOSE_NATIVE_WAYLAND)
/*! @brief Returns the `struct wl_display*` used by GLFW.
*
* @return The `struct wl_display*` used by GLFW, or `NULL` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.2.
*
* @ingroup native
*/
SSCAPI struct wl_display *glfwGetWaylandDisplay(void);
/*! @brief Returns the `struct wl_output*` of the specified monitor.
*
* @return The `struct wl_output*` of the specified monitor, or `NULL` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.2.
*
* @ingroup native
*/
SSCAPI struct wl_output *glfwGetWaylandMonitor(GLFWmonitor *monitor);
/*! @brief Returns the main `struct wl_surface*` of the specified window.
*
* @return The main `struct wl_surface*` of the specified window, or `NULL` if
* an [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.2.
*
* @ingroup native
*/
SSCAPI struct wl_surface *glfwGetWaylandWindow(GLFWwindow *window);
#endif
#if defined(GLFW_EXPOSE_NATIVE_EGL)
/*! @brief Returns the `EGLDisplay` used by GLFW.
*
* @return The `EGLDisplay` used by GLFW, or `EGL_NO_DISPLAY` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NOT_INITIALIZED.
*
* @remark Because EGL is initialized on demand, this function will return
* `EGL_NO_DISPLAY` until the first context has been created via EGL.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.0.
*
* @ingroup native
*/
SSCAPI EGLDisplay glfwGetEGLDisplay(void);
/*! @brief Returns the `EGLContext` of the specified window.
*
* @return The `EGLContext` of the specified window, or `EGL_NO_CONTEXT` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NO_WINDOW_CONTEXT and @ref
* GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.0.
*
* @ingroup native
*/
SSCAPI EGLContext glfwGetEGLContext(GLFWwindow *window);
/*! @brief Returns the `EGLSurface` of the specified window.
*
* @return The `EGLSurface` of the specified window, or `EGL_NO_SURFACE` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NO_WINDOW_CONTEXT and @ref
* GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.0.
*
* @ingroup native
*/
SSCAPI EGLSurface glfwGetEGLSurface(GLFWwindow *window);
#endif
#if defined(GLFW_EXPOSE_NATIVE_OSMESA)
/*! @brief Retrieves the color buffer associated with the specified window.
*
* @param[in] window The window whose color buffer to retrieve.
* @param[out] width Where to store the width of the color buffer, or `NULL`.
* @param[out] height Where to store the height of the color buffer, or `NULL`.
* @param[out] format Where to store the OSMesa pixel format of the color
* buffer, or `NULL`.
* @param[out] buffer Where to store the address of the color buffer, or
* `NULL`.
* @return `GLFW_TRUE` if successful, or `GLFW_FALSE` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NO_WINDOW_CONTEXT and @ref
* GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.3.
*
* @ingroup native
*/
SSCAPI int glfwGetOSMesaColorBuffer(GLFWwindow *window, int *width, int *height, int *format,
void **buffer);
/*! @brief Retrieves the depth buffer associated with the specified window.
*
* @param[in] window The window whose depth buffer to retrieve.
* @param[out] width Where to store the width of the depth buffer, or `NULL`.
* @param[out] height Where to store the height of the depth buffer, or `NULL`.
* @param[out] bytesPerValue Where to store the number of bytes per depth
* buffer element, or `NULL`.
* @param[out] buffer Where to store the address of the depth buffer, or
* `NULL`.
* @return `GLFW_TRUE` if successful, or `GLFW_FALSE` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NO_WINDOW_CONTEXT and @ref
* GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.3.
*
* @ingroup native
*/
SSCAPI int glfwGetOSMesaDepthBuffer(GLFWwindow *window, int *width, int *height, int *bytesPerValue,
void **buffer);
/*! @brief Returns the `OSMesaContext` of the specified window.
*
* @return The `OSMesaContext` of the specified window, or `NULL` if an
* [error](@ref error_handling) occurred.
*
* @errors Possible errors include @ref GLFW_NO_WINDOW_CONTEXT and @ref
* GLFW_NOT_INITIALIZED.
*
* @thread_safety This function may be called from any thread. Access is not
* synchronized.
*
* @since Added in version 3.3.
*
* @ingroup native
*/
SSCAPI OSMesaContext glfwGetOSMesaContext(GLFWwindow *window);
#endif
#ifdef __cplusplus
}
#endif
#endif /* _glfw3_native_h_ */
+32
View File
@@ -0,0 +1,32 @@
#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
@@ -0,0 +1,176 @@
#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