Fixed bug 3816 - asm code in video/SDL_stretch.c

Ozkan Sezer

- adds MSVC __declspec(align(x)) support,
- disables asm if PAGE_ALIGNED no macro is defined,
- still disables asm for gcc < 4.6, need more info,
- drops Watcom support.
This commit is contained in:
Sam Lantinga 2021-01-27 11:39:03 -08:00
parent 1e5f0073c6
commit 06ad887f44

View File

@ -34,54 +34,58 @@
into the general blitting mechanism. into the general blitting mechanism.
*/ */
#if ((defined(_MSC_VER) && defined(_M_IX86)) || \ #if ((defined(_MSC_VER) && defined(_M_IX86)) || \
(defined(__WATCOMC__) && defined(__386__)) || \
(defined(__GNUC__) && defined(__i386__))) && SDL_ASSEMBLY_ROUTINES (defined(__GNUC__) && defined(__i386__))) && SDL_ASSEMBLY_ROUTINES
#define USE_ASM_STRETCH
#endif
/* There's a bug with gcc 4.4.1 and -O2 where srcp doesn't get the correct /* There's a bug with gcc 4.4.1 and -O2 where srcp doesn't get the correct
* value after the first scanline. */ * value after the first scanline. */
/* This bug seems fixed, at least with gcc >= 4.6 */ /* This bug seems fixed, at least with gcc >= 4.6 */
# define USE_ASM_STRETCH #if defined(USE_ASM_STRETCH) && \
# if defined(__GNUC__) defined(__GNUC__) && (__GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 6))
# if (__GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 6)) #undef USE_ASM_STRETCH
# undef USE_ASM_STRETCH #endif
# endif
# endif /* And it doesn't work if mprotect isn't available */
/* But it doesn't work if mprotect isn't available */ #if defined(USE_ASM_STRETCH) && \
# if !defined(HAVE_MPROTECT) && defined(USE_ASM_STRETCH) !defined(HAVE_MPROTECT) && !defined(__WIN32__)
# undef USE_ASM_STRETCH #undef USE_ASM_STRETCH
# endif
#endif #endif
#ifdef USE_ASM_STRETCH #ifdef USE_ASM_STRETCH
#ifdef HAVE_MPROTECT #ifdef __WIN32__
#define WIN32_LEAN_AND_MEAN
#include <windows.h>
#elif defined(HAVE_MPROTECT)
#include <sys/types.h> #include <sys/types.h>
#include <sys/mman.h> #include <sys/mman.h>
#endif #endif
#ifdef __GNUC__ #ifdef __GNUC__
#define PAGE_ALIGNED __attribute__((__aligned__(4096))) #define PAGE_ALIGNED __attribute__((__aligned__(4096)))
#elif defined(_MSC_VER)
#define PAGE_ALIGNED __declspec(align(4096))
#else #else
#define PAGE_ALIGNED #undef USE_ASM_STRETCH
#endif #endif
#endif /**/
#ifdef USE_ASM_STRETCH
#if defined(_M_IX86) || defined(__i386__) || defined(__386__)
#define PREFIX16 0x66 #define PREFIX16 0x66
#define STORE_BYTE 0xAA #define STORE_BYTE 0xAA
#define STORE_WORD 0xAB #define STORE_WORD 0xAB
#define LOAD_BYTE 0xAC #define LOAD_BYTE 0xAC
#define LOAD_WORD 0xAD #define LOAD_WORD 0xAD
#define RETURN 0xC3 #define RETURN 0xC3
#else
#error Need assembly opcodes for this architecture
#endif
static unsigned char copy_row[4096] PAGE_ALIGNED; static PAGE_ALIGNED unsigned char copy_row[4096];
static int static int
generate_rowbytes(int src_w, int dst_w, int bpp) generate_rowbytes(int src_w, int dst_w, int bpp)
{ {
static struct static struct {
{
int bpp; int bpp;
int src_w; int src_w;
int dst_w; int dst_w;
@ -92,6 +96,9 @@ generate_rowbytes(int src_w, int dst_w, int bpp)
int pos, inc; int pos, inc;
unsigned char *eip, *fence; unsigned char *eip, *fence;
unsigned char load, store; unsigned char load, store;
#ifdef __WIN32__
DWORD oldprot;
#endif
/* See if we need to regenerate the copy buffer */ /* See if we need to regenerate the copy buffer */
if ((src_w == last.src_w) && (dst_w == last.dst_w) && (bpp == last.bpp)) { if ((src_w == last.src_w) && (dst_w == last.dst_w) && (bpp == last.bpp)) {
@ -115,8 +122,12 @@ generate_rowbytes(int src_w, int dst_w, int bpp)
default: default:
return SDL_SetError("ASM stretch of %d bytes isn't supported", bpp); return SDL_SetError("ASM stretch of %d bytes isn't supported", bpp);
} }
#ifdef HAVE_MPROTECT
/* Make the code writeable */ /* Make the code writeable */
#ifdef __WIN32__
if (!VirtualProtect(copy_row, sizeof(copy_row), PAGE_READWRITE, &oldprot)) {
return SDL_SetError("Couldn't make copy buffer writeable");
}
#elif defined(HAVE_MPROTECT)
if (mprotect(copy_row, sizeof(copy_row), PROT_READ | PROT_WRITE) < 0) { if (mprotect(copy_row, sizeof(copy_row), PROT_READ | PROT_WRITE) < 0) {
return SDL_SetError("Couldn't make copy buffer writeable"); return SDL_SetError("Couldn't make copy buffer writeable");
} }
@ -147,8 +158,12 @@ generate_rowbytes(int src_w, int dst_w, int bpp)
} }
*eip++ = RETURN; *eip++ = RETURN;
#ifdef HAVE_MPROTECT
/* Make the code executable but not writeable */ /* Make the code executable but not writeable */
#ifdef __WIN32__
if (!VirtualProtect(copy_row, sizeof(copy_row), PAGE_EXECUTE_READ, &oldprot)) {
return SDL_SetError("Couldn't make copy buffer executable");
}
#elif defined(HAVE_MPROTECT)
if (mprotect(copy_row, sizeof(copy_row), PROT_READ | PROT_EXEC) < 0) { if (mprotect(copy_row, sizeof(copy_row), PROT_READ | PROT_EXEC) < 0) {
return SDL_SetError("Couldn't make copy buffer executable"); return SDL_SetError("Couldn't make copy buffer executable");
} }
@ -156,7 +171,6 @@ generate_rowbytes(int src_w, int dst_w, int bpp)
last.status = 0; last.status = 0;
return (0); return (0);
} }
#endif /* USE_ASM_STRETCH */ #endif /* USE_ASM_STRETCH */
#define DEFINE_COPY_ROW(name, type) \ #define DEFINE_COPY_ROW(name, type) \