SDL  2.0
SDL_simd.h File Reference
#include "SDL.h"
#include "../SDL_internal.h"
+ Include dependency graph for SDL_simd.h:
+ This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Functions

size_t SDL_SIMDGetAlignment (void)
 Report the alignment this system needs for SIMD allocations. More...
 
voidSDL_SIMDAlloc (const size_t len)
 Allocate memory in a SIMD-friendly way. More...
 
void SDL_SIMDFree (void *ptr)
 Deallocate memory obtained from SDL_SIMDAlloc. More...
 

Function Documentation

§ SDL_SIMDAlloc()

void* SDL_SIMDAlloc ( const size_t  len)

Allocate memory in a SIMD-friendly way.

This will allocate a block of memory that is suitable for use with SIMD instructions. Specifically, it will be properly aligned and padded for the system's supported vector instructions.

The memory returned will be padded such that it is safe to read or write an incomplete vector at the end of the memory block. This can be useful so you don't have to drop back to a scalar fallback at the end of your SIMD processing loop to deal with the final elements without overflowing the allocated buffer.

You must free this memory with SDL_FreeSIMD(), not free() or SDL_free() or delete[], etc.

Note that SDL will only deal with SIMD instruction sets it is aware of; for example, SDL 2.0.8 knows that SSE wants 16-byte vectors (SDL_HasSSE()), and AVX2 wants 32 bytes (SDL_HasAVX2()), but doesn't know that AVX-512 wants 64. To be clear: if you can't decide to use an instruction set with an SDL_Has*() function, don't use that instruction set with memory allocated through here.

SDL_AllocSIMD(0) will return a non-NULL pointer, assuming the system isn't out of memory.

Parameters
lenThe length, in bytes, of the block to allocated. The actual allocated block might be larger due to padding, etc.
Returns
Pointer to newly-allocated block, NULL if out of memory.
See also
SDL_SIMDAlignment
SDL_SIMDFree

Definition at line 799 of file SDL_cpuinfo.c.

References NULL, retval, SDL_malloc, and SDL_SIMDGetAlignment().

800 {
801  const size_t alignment = SDL_SIMDGetAlignment();
802  const size_t padding = alignment - (len % alignment);
803  const size_t padded = (padding != alignment) ? (len + padding) : len;
804  Uint8 *retval = NULL;
805  Uint8 *ptr = (Uint8 *) SDL_malloc(padded + alignment + sizeof (void *));
806  if (ptr) {
807  /* store the actual malloc pointer right before our aligned pointer. */
808  retval = ptr + sizeof (void *);
809  retval += alignment - (((size_t) retval) % alignment);
810  *(((void **) retval) - 1) = ptr;
811  }
812  return retval;
813 }
unsigned int size_t
GLenum GLsizei len
SDL_bool retval
uint8_t Uint8
Definition: SDL_stdinc.h:179
#define NULL
Definition: begin_code.h:164
#define SDL_malloc
size_t SDL_SIMDGetAlignment(void)
Report the alignment this system needs for SIMD allocations.
Definition: SDL_cpuinfo.c:789

§ SDL_SIMDFree()

void SDL_SIMDFree ( void ptr)

Deallocate memory obtained from SDL_SIMDAlloc.

It is not valid to use this function on a pointer from anything but SDL_SIMDAlloc(). It can't be used on pointers from malloc, realloc, SDL_malloc, memalign, new[], etc.

However, SDL_SIMDFree(NULL) is a legal no-op.

See also
SDL_SIMDAlloc

Definition at line 816 of file SDL_cpuinfo.c.

References main, SDL_free, SDL_GetCPUCacheLineSize(), SDL_GetCPUCount(), SDL_GetCPUType(), SDL_GetSystemRAM(), SDL_Has3DNow(), SDL_HasAltiVec(), SDL_HasAVX(), SDL_HasAVX2(), SDL_HasAVX512F(), SDL_HasMMX(), SDL_HasNEON(), SDL_HasRDTSC(), SDL_HasSSE(), SDL_HasSSE2(), SDL_HasSSE3(), SDL_HasSSE41(), and SDL_HasSSE42().

817 {
818  if (ptr) {
819  void **realptr = (void **) ptr;
820  realptr--;
821  SDL_free(*(((void **) ptr) - 1));
822  }
823 }
#define SDL_free

§ SDL_SIMDGetAlignment()

size_t SDL_SIMDGetAlignment ( void  )

Report the alignment this system needs for SIMD allocations.

This will return the minimum number of bytes to which a pointer must be aligned to be compatible with SIMD instructions on the current machine. For example, if the machine supports SSE only, it will return 16, but if it supports AVX-512F, it'll return 64 (etc). This only reports values for instruction sets SDL knows about, so if your SDL build doesn't have SDL_HasAVX512F(), then it might return 16 for the SSE support it sees and not 64 for the AVX-512 instructions that exist but SDL doesn't know about. Plan accordingly.

Definition at line 789 of file SDL_cpuinfo.c.

References SDL_assert, SDL_GetCPUFeatures(), and SDL_SIMDAlignment.

Referenced by SDL_SIMDAlloc().

790 {
791  if (SDL_SIMDAlignment == 0xFFFFFFFF) {
792  SDL_GetCPUFeatures(); /* make sure this has been calculated */
793  }
795  return SDL_SIMDAlignment;
796 }
static Uint32 SDL_GetCPUFeatures(void)
Definition: SDL_cpuinfo.c:594
#define SDL_assert(condition)
Definition: SDL_assert.h:169
static Uint32 SDL_SIMDAlignment
Definition: SDL_cpuinfo.c:591