ShuriZma
/
RetroArch
mirror of https://github.com/libretro/RetroArch.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

(PS3) Libfreevec/Altivec implementations of memcpy/memset - 

use redefine macros in RGL and RetroArch for them - nice speedup
This commit is contained in:
twinaphex 2013-06-02 18:35:13 +02:00
parent 86458420cf
commit a581c9ac86
6 changed files with 703 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
griffin/griffin.c
View File

@ -22,6 +22,10 @@
#define HAVE_MENU #define HAVE_MENU
#endif #endif
#ifdef __CELLOS_LV2__
#include "../ps3/altivec_mem.c"
#endif
/*============================================================ /*============================================================
CONSOLE EXTENSIONS CONSOLE EXTENSIONS
============================================================ */ ============================================================ */

218
ps3/altivec_mem.c Normal file
View File

@ -0,0 +1,218 @@
/*********************************************************************************
* Copyright (C) 2008-2010 by Konstantinos Margaritis <markos@codex.gr> *
* All rights reserved. *
* *
* Redistribution and use in source and binary forms, with or without *
* modification, are permitted provided that the following conditions are met: *
* 1. Redistributions of source code must retain the above copyright *
* notice, this list of conditions and the following disclaimer. *
* 2. Redistributions in binary form must reproduce the above copyright *
* notice, this list of conditions and the following disclaimer in the *
* documentation and/or other materials provided with the distribution. *
* 3. Neither the name of the Codex nor the *
* names of its contributors may be used to endorse or promote products *
* derived from this software without specific prior written permission. *
* *
* THIS SOFTWARE IS PROVIDED BY CODEX ''AS IS'' AND ANY *
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED *
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE *
* DISCLAIMED. IN NO EVENT SHALL CODEX BE LIABLE FOR ANY *
* DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES *
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; *
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND *
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT *
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS *
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. *
*********************************************************************************/
#include <stdio.h>
#include <limits.h>
#include "altivec_mem.h"
void *vec_memcpy(void *dstpp, const void *srcpp, size_t len)
{
const uint8_t *src = srcpp;
uint8_t *dst = dstpp;
if (len >= sizeof(word_t))
{
// Prefetch some stuff
READ_PREFETCH_START1(src);
WRITE_PREFETCH_START2(dst);
// Copy until dst is word aligned
int al = copy_fwd_until_dst_word_aligned(dst, src), l;
if (al)
{
src += sizeof(word_t) - al;
dst += sizeof(word_t) - al;
len -= sizeof(word_t) - al;
}
// Now dst is word aligned. We'll continue by word copying, but
// for this we have to know the word-alignment of src also.
int srcoffset = ((word_t)(src) % sizeof(word_t)), sh_l, sh_r;
sh_l = srcoffset * CHAR_BIT;
sh_r = CHAR_BIT * sizeof(word_t) - sh_l;
// Take the word-aligned long pointers of src and dest.
word_t *dstl = (word_t *)(dst);
const word_t *srcl = (word_t *)(src - srcoffset);
if (len >= SIMD_PACKETSIZE)
{
// While we're not 16-byte aligned, move in 4-byte long steps.
al = (word_t)dstl % SIMD_PACKETSIZE;
if (al)
{
copy_fwd_until_dst_simd_aligned(dstl, srcl, srcoffset, al, sh_l, sh_r);
srcl += (SIMD_PACKETSIZE - al)/WORDS_IN_PACKET;
src = (uint8_t *) srcl + srcoffset;
dstl += (SIMD_PACKETSIZE - al)/WORDS_IN_PACKET;
len -= SIMD_PACKETSIZE - al;
}
// Now, dst is 16byte aligned. We can use SIMD if len >= 16
l = len / SIMD_PACKETSIZE;
len -= l * SIMD_PACKETSIZE;
if (((word_t)(src) % SIMD_PACKETSIZE) == 0)
copy_fwd_rest_blocks_aligned(dstl, src, l);
else
copy_fwd_rest_blocks_unaligned(dstl, src, srcoffset, sh_l, sh_r, l);
src += l*SIMD_PACKETSIZE;
dstl += l * WORDS_IN_PACKET;
srcl = (word_t *)(src - srcoffset);
}
// Stop the prefetching
PREFETCH_STOP1;
PREFETCH_STOP2;
//#endif
// Copy the remaining bytes using word-copying
// Handle alignment as appropriate
l = len / sizeof(word_t);
len -= l * sizeof(word_t);
if (srcoffset == 0)
{
copy_fwd_rest_words_aligned(dstl, srcl, l);
srcl += l;
src = (uint8_t *) srcl;
}
else
{
copy_fwd_rest_words_unaligned(dstl, srcl, sh_l, sh_r, l);
srcl += l;
src = (uint8_t *) srcl + srcoffset;
}
dstl += l;
// For the end copy we have to use char * pointers.
dst = (uint8_t *) dstl;
}
// Copy the remaining bytes
copy_fwd_rest_bytes(dst, src, len);
return dstpp;
}
void *vec_memcpy_aligned(void *dstpp, const void *srcpp, size_t len)
{
const uint8_t *src = srcpp;
uint8_t *dst = dstpp;
if (len >= sizeof(word_t))
{
// Prefetch some stuff
READ_PREFETCH_START1(src);
WRITE_PREFETCH_START2(dst);
// Take the word-aligned long pointers of src and dest.
word_t *dstl = (word_t *)(dst);
const word_t *srcl = (word_t *)(src);
int l;
#ifdef LIBFREEVEC_SIMD_ENGINE
if (len >= SIMD_PACKETSIZE)
{
l = len / SIMD_PACKETSIZE;
len -= l * SIMD_PACKETSIZE;
// Now, dst is 16byte aligned. We can use SIMD if len >= 16
copy_fwd_rest_blocks_aligned(dstl, src, l);
}
#endif
// Copy the remaining bytes using word-copying
// Handle alignment as appropriate
l = len / sizeof(word_t);
copy_fwd_rest_words_aligned(dstl, srcl, l);
srcl += l;
dstl += l;
len -= l * sizeof(word_t);
// For the end copy we have to use char * pointers.
src = (uint8_t *) srcl;
dst = (uint8_t *) dstl;
}
// Stop the prefetching
PREFETCH_STOP1;
PREFETCH_STOP2;
// Copy the remaining bytes
copy_fwd_rest_bytes(dst, src, len);
return dstpp;
}
void *vec_memset(void *s, int p, size_t len)
{
uint8_t* ptr = s;
uint8_t __attribute__ ((aligned(16))) P = p;
if (len >= sizeof(word_t))
{
word_t pw = charmask(P);
size_t al = ((size_t)ptr) % sizeof(word_t);
if (al)
{
memset_fwd_until_dst_word_aligned(ptr, P, al);
ptr += sizeof(word_t) - al;
len -= sizeof(word_t) - al;
}
int l;
word_t *ptr_w = (word_t *)(ptr);
if (len >= SIMD_PACKETSIZE)
{
// ptr is now word (32/64bit) aligned, memset until ptr is SIMD aligned
al = (word_t) ptr_w % SIMD_PACKETSIZE;
if (al)
{
memset_fwd_until_simd_aligned(ptr_w, pw, al);
ptr_w += (SIMD_PACKETSIZE - al)/WORDS_IN_PACKET;
len -= SIMD_PACKETSIZE - al;
}
// ptr is now 128-bit aligned
// perform set using SIMD
l = len / SIMD_PACKETSIZE;
len -= l * SIMD_PACKETSIZE;
memset_set_blocks(ptr_w, pw, P, l);
ptr_w += l * WORDS_IN_PACKET;
}
// memset the remaining words
l = len / sizeof(word_t);
len -= l * sizeof(word_t);
memset_rest_words(ptr_w, pw, l);
ptr_w += l;
ptr = (uint8_t *)ptr_w;
}
// Handle the remaining bytes
memset_rest_bytes(ptr, P, len);
return s;
}

474
ps3/altivec_mem.h Normal file
View File

@ -0,0 +1,474 @@
/*********************************************************************************
* Copyright (C) 2008-2010 by Konstantinos Margaritis <markos@codex.gr> *
* All rights reserved. *
* *
* Redistribution and use in source and binary forms, with or without *
* modification, are permitted provided that the following conditions are met: *
* 1. Redistributions of source code must retain the above copyright *
* notice, this list of conditions and the following disclaimer. *
* 2. Redistributions in binary form must reproduce the above copyright *
* notice, this list of conditions and the following disclaimer in the *
* documentation and/or other materials provided with the distribution. *
* 3. Neither the name of the Codex nor the *
* names of its contributors may be used to endorse or promote products *
* derived from this software without specific prior written permission. *
* *
* THIS SOFTWARE IS PROVIDED BY CODEX ''AS IS'' AND ANY *
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED *
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE *
* DISCLAIMED. IN NO EVENT SHALL CODEX BE LIABLE FOR ANY *
* DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES *
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; *
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND *
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT *
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS *
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. *
*********************************************************************************/
/* $Id$ */
#ifndef ALTIVEC_MEM_H
#define ALTIVEC_MEM_H
#include <sys/types.h>
#include <stdint.h>
#include <stddef.h>
#ifdef __cplusplus
extern "C" {
#endif
#if defined(__SNC__)
#include <ppu_altivec_internals.h>
#else
#include <altivec.h>
#endif
#define himagic 0x80808080L
#define lomagic 0x01010101L
#define magic_bits32 0x07efefeff
#define magic_bits64 (((unsigned long int) 0x7efefefe << 32) | 0xfefefeff)
#define charmask8(c) ((uint8_t)(c & 0xff))
#define charmask16(c) (uint16_t)((charmask8(c)) | (charmask8(c) << 8))
#define charmask32(c) (uint32_t)((charmask16(c)) | (charmask16(c) << 16))
#define charmask64(c) (uint64_t)((charmask32(c)) | (charmask32(c) << 32))
#define QMAKESTR(x) #x
#define MAKESTR(x) QMAKESTR(x)
#define SMASH(x,y) x/y
#define MAKEINC(x) SMASH(x,MACROFILE)
#define LIBFREEVEC_SIMD_MACROS_INC MAKEINC(LIBFREEVEC_SIMD_ENGINE)
#define ptrdiff_t(a, b) ((word_t)(a)-(word_t)(b))
#define CMP_LT_OR_GT(a, b) ((a) - (b))
#define MIN(a,b) ((a) <= (b) ? (a) : (b))
#define DIFF(a, b) ((a)-(b))
#ifdef LSB_FIRST
#define MERGE_SHIFTED_WORDS(a, b, sl, sr) ((a) >> sl) | ((b) << sr)
#else
#define MERGE_SHIFTED_WORDS(a, b, sl, sr) ((a) << sl) | ((b) >> sr)
#endif
#define DST_CHAN_1 1
#define DST_CHAN_2 2
#define DST_CTRL(size, count, stride) (((size) << 24) | ((count) << 16) | (stride))
//#define READ_PREFETCH_START1(addr) vec_dstt(addr, DST_CTRL(2,2,32), DST_CHAN_1)
//#define READ_PREFETCH_START2(addr) vec_dstt(addr, DST_CTRL(2,2,32), DST_CHAN_2)
//#define WRITE_PREFETCH_START1(addr) vec_dststt(addr, DST_CTRL(2,2,32), DST_CHAN_1)
//#define WRITE_PREFETCH_START2(addr) vec_dststt(addr, DST_CTRL(2,2,32), DST_CHAN_2)
//#define PREFETCH_STOP1 vec_dss(DST_CHAN_1)
//#define PREFETCH_STOP2 vec_dss(DST_CHAN_2)
#define READ_PREFETCH_START1(addr)
#define READ_PREFETCH_START2(addr)
#define WRITE_PREFETCH_START1(addr)
#define WRITE_PREFETCH_START2(addr)
#define PREFETCH_STOP1
#define PREFETCH_STOP2
//#define LINUX64
#ifdef LINUX64
#define word_t uint64_t
#define LIBFREEVEC_SCALAR_MACROS_INC MAKEINC(scalar64)
#define charmask(c) charmask64(c)
#define SIMD_PACKETSIZE 16
#define WORDS_IN_PACKET 2
static inline int copy_fwd_until_dst_word_aligned(uint8_t *d, const uint8_t *s)
{
int dstal = ((word_t)d) % sizeof(word_t);
switch (dstal)
{
case 1:
*d++ = *s++;
case 2:
*d++ = *s++;
case 3:
*d++ = *s++;
case 4:
*d++ = *s++;
case 5:
*d++ = *s++;
case 6:
*d++ = *s++;
case 7:
*d = *s;
}
return dstal;
}
static inline void copy_fwd_rest_bytes(uint8_t *d, const uint8_t *s, size_t len)
{
switch (len)
{
case 7:
*d++ = *s++;
case 6:
*d++ = *s++;
case 5:
*d++ = *s++;
case 4:
*d++ = *s++;
case 3:
*d++ = *s++;
case 2:
*d++ = *s++;
case 1:
*d = *s;
}
}
static inline void copy_fwd_rest_words_aligned(word_t *d, const word_t *s, size_t l)
{
while (l > 0)
{
*d++ = *s++;
l--;
}
}
static inline void copy_fwd_rest_words_unaligned(word_t *d, const word_t *s, int sl, int sr, size_t l)
{
while (l > 0)
{
*d++ = MERGE_SHIFTED_WORDS(*(s), *(s + 1), sl, sr); s++;
l--;
}
}
static inline void copy_fwd_until_dst_simd_aligned(word_t *d, const word_t *s,
int srcoffset4, size_t dstal, int sl, int sr)
{
if (srcoffset4 == 0)
{
if (dstal == 8)
*d++ = *s++;
}
else
{
if (dstal == 8)
{
*d = MERGE_SHIFTED_WORDS(*(s), *(s + 1), sl, sr);
d++; s++;
}
}
}
static inline void memset_fwd_until_dst_word_aligned(uint8_t *ptr, uint8_t c, size_t al)
{
switch(al)
{
case 1:
*ptr++ = c;
case 2:
*ptr++ = c;
case 3:
*ptr++ = c;
}
}
static inline void memset_fwd_until_simd_aligned(word_t *ptr_w, word_t w, size_t al)
{
switch (al)
{
case 4:
*ptr_w++ = w;
case 8:
*ptr_w++ = w;
case 12:
*ptr_w++ = w;
}
}
static inline void memset_rest_words(word_t *ptr_w, word_t w, size_t l)
{
while (l--)
*ptr_w++ = w;
}
static inline int memset_rest_bytes(uint8_t *ptr, uint8_t c, size_t len)
{
switch(len)
{
case 3:
*ptr++ = c;
case 2:
*ptr++ = c;
case 1:
*ptr++ = c;
}
}
#else
#define word_t uint32_t
#define LIBFREEVEC_SCALAR_MACROS_INC MAKEINC(scalar32)
#define charmask(c) charmask32(c)
#define SIMD_PACKETSIZE 16
#define WORDS_IN_PACKET 4
static inline int copy_fwd_until_dst_word_aligned(uint8_t *d, const uint8_t *s)
{
size_t dstal = ((size_t)d) % sizeof(word_t);
switch (dstal)
{
case 1:
*d++ = *s++;
case 2:
*d++ = *s++;
case 3:
*d = *s;
}
return dstal;
}
static inline void copy_fwd_rest_bytes(uint8_t *d, const uint8_t *s, size_t len)
{
switch (len)
{
case 3:
*d++ = *s++;
case 2:
*d++ = *s++;
case 1:
*d++ = *s++;
}
}
static inline void copy_fwd_rest_words_aligned(word_t *d, const word_t *s, size_t l)
{
while (l > 0)
{
*d++ = *s++;
l--;
}
}
static inline void copy_fwd_rest_words_unaligned(word_t *d, const word_t *s, int sl, int sr, size_t l)
{
while (l > 0)
{
*d++ = MERGE_SHIFTED_WORDS(*(s), *(s + 1), sl, sr); s++;
l--;
}
}
static inline void copy_fwd_until_dst_simd_aligned(word_t *d, const word_t *s,
int srcoffset4, size_t dstal, int sl, int sr)
{
if (srcoffset4 == 0)
{
switch (dstal)
{
case 4:
*d++ = *s++;
case 8:
*d++ = *s++;
case 12:
*d++ = *s++;
}
}
else
{
switch (dstal)
{
case 4:
*d = MERGE_SHIFTED_WORDS(*(s), *(s + 1), sl, sr);
d++; s++;
case 8:
*d = MERGE_SHIFTED_WORDS(*(s), *(s + 1), sl, sr);
d++; s++;
case 12:
*d = MERGE_SHIFTED_WORDS(*(s), *(s + 1), sl, sr);
d++; s++;
}
}
}
static inline void memset_fwd_until_dst_word_aligned(uint8_t *ptr, uint8_t c, size_t al)
{
switch(al)
{
case 1:
*ptr++ = c;
case 2:
*ptr++ = c;
case 3:
*ptr++ = c;
}
}
static inline void memset_fwd_until_simd_aligned(word_t *ptr_w, word_t w, size_t al)
{
switch (al)
{
case 4:
*ptr_w++ = w;
case 8:
*ptr_w++ = w;
case 12:
*ptr_w++ = w;
}
}
static inline void memset_rest_words(word_t *ptr_w, word_t w, size_t l)
{
while (l--)
{
*ptr_w++ = w;
}
}
static inline int memset_rest_bytes(uint8_t *ptr, uint8_t c, size_t len)
{
switch(len)
{
case 3:
*ptr++ = c;
case 2:
*ptr++ = c;
case 1:
*ptr++ = c;
}
}
#endif
static inline void copy_fwd_rest_blocks_aligned(word_t *d, const uint8_t *s, size_t blocks)
{
__vector unsigned char v1, v2, v3, v4;
// Unroll blocks of 4 words
while (blocks > 4)
{
v1 = vec_ld(0, s);
v2 = vec_ld(16, s);
v3 = vec_ld(32, s);
v4 = vec_ld(48, s);
vec_st(v1, 0, (uint8_t *)d);
vec_st(v2, 16, (uint8_t *)d);
vec_st(v3, 32, (uint8_t *)d);
vec_st(v4, 48, (uint8_t *)d);
d += 16; s += 4 * SIMD_PACKETSIZE;
blocks -= 4;
}
while (blocks > 0)
{
v1 = vec_ld(0, s);
vec_st(v1, 0, (uint8_t *)d);
d += 4; s += SIMD_PACKETSIZE;
blocks--;
}
}
static inline void copy_fwd_rest_blocks_unaligned(word_t *d, const uint8_t *s, int srcoffset, int sl, int sr, size_t blocks)
{
__vector unsigned char mask, MSQ1, LSQ1, LSQ2, LSQ3, LSQ4;
mask = vec_lvsl(0, s);
// Unroll blocks of 4 words
while (blocks > 4)
{
MSQ1 = vec_ld(0, s);
LSQ1 = vec_ld(15, s);
LSQ2 = vec_ld(31, s);
LSQ3 = vec_ld(47, s);
LSQ4 = vec_ld(63, s);
vec_st(vec_perm(MSQ1, LSQ1, mask), 0, (uint8_t *)d);
vec_st(vec_perm(LSQ1, LSQ2, mask), 16, (uint8_t *)d);
vec_st(vec_perm(LSQ2, LSQ3, mask), 32, (uint8_t *)d);
vec_st(vec_perm(LSQ3, LSQ4, mask), 48, (uint8_t *)d);
d += 16; s += 4 * SIMD_PACKETSIZE;
blocks -= 4;
}
while (blocks > 0)
{
MSQ1 = vec_ld(0, s);
LSQ1 = vec_ld(15, s);
vec_st(vec_perm(MSQ1, LSQ1, mask), 0, (uint8_t *)d);
d += 4; s += SIMD_PACKETSIZE;
blocks--;
}
}
static inline __vector unsigned char simdpacket_set_from_byte(const uint8_t c)
{
__vector unsigned char v = vec_lde(0, &c);
return vec_splat(v, 0);
}
static inline void memset_set_blocks(word_t *ptr_w, word_t pw, uint8_t c, size_t blocks)
{
__vector unsigned char vc = simdpacket_set_from_byte(c);
while (blocks > 4)
{
vec_st(vc, 0, (uint8_t *)ptr_w);
vec_st(vc, 16, (uint8_t *)ptr_w);
vec_st(vc, 32, (uint8_t *)ptr_w);
vec_st(vc, 48, (uint8_t *)ptr_w);
ptr_w += 4 * WORDS_IN_PACKET;
blocks -= 4;
}
while (blocks--)
{
vec_st(vc, 0, (uint8_t *)ptr_w);
ptr_w += WORDS_IN_PACKET;
}
}
extern void *vec_memcpy(void *dstpp, const void *srcpp, size_t len);
extern void *vec_memcpy_aligned(void *dstpp, const void *srcpp, size_t len);
extern void *vec_memset(void *s, int p, size_t len);
#ifdef __cplusplus
}
#endif
#define memcpy(dest, src, n) vec_memcpy(dest, src, n)
#define memset(s, c, n) vec_memset(s, c, n)
#endif

5
ps3/rgl/src/cg/cgbdefs.hpp
View File

@ -12,11 +12,6 @@ typedef unsigned int Elf32_Word;
typedef unsigned short Elf64_Half; typedef unsigned short Elf64_Half;
typedef short Elf64_Shalf; typedef short Elf64_Shalf;
#ifdef __CELLOS_LV2__
typedef size_t ptrdiff_t;
typedef size_t ptrdiff_t;
#endif
/////////////////////// ///////////////////////
// ELF Header Constants // ELF Header Constants

2
ps3/rgl/src/ps3/rgl_ps3.cpp
View File

@ -20,6 +20,8 @@
#include <sdk_version.h> #include <sdk_version.h>
#include "../../altivec_mem.h"
#include "include/GmmAlloc.h" #include "include/GmmAlloc.h"
#include "include/rgl-constants.h" #include "include/rgl-constants.h"
#include "include/rgl-typedefs.h" #include "include/rgl-typedefs.h"

5
ps3/rgl/src/ps3/rgl_ps3_raster.cpp
View File

@ -16,6 +16,11 @@
*/ */
#include "../rgl.h" #include "../rgl.h"
#include "../rglp.h"
#include <sdk_version.h>
#include "../../altivec_mem.h"
#include "include/GmmAlloc.h" #include "include/GmmAlloc.h"
#include "include/rgl-constants.h" #include "include/rgl-constants.h"