Portable Single Instruction Multiple Data (SIMD) Intrinsics Wrapper. More...

#include "align.h"
#include <math.h>
#include <stdbool.h>
#include <stdint.h>
#include <string.h>

Include dependency graph for simd.h:

This graph shows which files directly or indirectly include this file:

Macros
#define	simd_swizzle(v, x, y, z, w) ((simd_vec_t){{(v).f[(x)], (v).f[(y)], (v).f[(z)], (v).f[(w)]}})
	Rearranges vector components.

#define	simd_transpose4(r0, r1, r2, r3)
	Transposes a 4x4 matrix defined by 4 row/col vectors in-place.

Detailed Description

Portable Single Instruction Multiple Data (SIMD) Intrinsics Wrapper.

This header provides a unified abstraction layer over hardware-accelerated vector operations. It automatically detects the target architecture and maps abstract vector types to the underlying hardware registers.

Supported Backends:

x86/x64: Uses SSE (Streaming SIMD Extensions). Baseline is SSE2.
- Automatic optimizations for SSE4.1(via compiler flags).
ARM: Uses NEON.
- Specific paths for ARMv7 (32-bit) and AArch64 (64-bit).
Scalar: Fallback for architectures without supported SIMD units.

Design Philosophy:

Vertical Operations: Primary focus. Operations like add, sub, mul happen component-wise ($x+x, y+y, \dots$). These are the fastest instructions.
Horizontal Operations: Operations like dot products or reductions ($x+y+z+w$) are generally slower than vertical ones and are implemented using shuffles/adds.
SOA (Structure of Arrays): This library is best used with Data-Oriented Design where data is laid out as xxxx... yyyy... zzzz... rather than xyz xyz....

Usage Example:

simd_vec_t pos_a = simd_set(10.0f, 20.0f, 30.0f, 1.0f);
simd_vec_t pos_b = simd_set(5.0f, 5.0f, 5.0f, 0.0f);
 
// Vertical addition (happens in one CPU cycle equivalent)
simd_vec_t result = simd_add(pos_a, pos_b);
 
// Dot product
float dist_sq = simd_dot3(result, result);

Definition in file simd.h.

Macro Definition Documentation

◆ simd_swizzle

#define simd_swizzle	(	v,
		x,
		y,
		z,
		w
	)	((simd_vec_t){{(v).f[(x)], (v).f[(y)], (v).f[(z)], (v).f[(w)]}})

Rearranges vector components.

Usage: simd_vec_t result = simd_swizzle(v, SIMD_Y, SIMD_X, SIMD_W, SIMD_Z);

Note: This MUST be a macro because x86 intrinsics require compile-time constants.

Definition at line 1059 of file simd.h.

Macros

Detailed Description

Macro Definition Documentation

◆ simd_swizzle