Go to:
Gentoo Home
Documentation
Forums
Lists
Bugs
Planet
Store
Wiki
Get Gentoo!
Gentoo's Bugzilla – Attachment 67787 Details for
Bug 104533
bomberclone segfaults with libsdl-1.2.9
Home
|
New
–
[Ex]
|
Browse
|
Search
|
Privacy Policy
|
[?]
|
Reports
|
Requests
|
Help
|
New Account
|
Log In
[x]
|
Forgot Password
Login:
[x]
[patch]
hack
newdiff.patch (text/plain), 28.17 KB, created by
Aaron
on 2005-09-06 18:59:57 UTC
(
hide
)
Description:
hack
Filename:
MIME Type:
Creator:
Aaron
Created:
2005-09-06 18:59:57 UTC
Size:
28.17 KB
patch
obsolete
>--- SDL-1.2.9/src/video/SDL_blit_N.c 2005-04-19 22:57:39.000000000 -0700 >+++ SDL-1.2.8/src/video/SDL_blit_N.c 2004-02-18 10:22:04.000000000 -0700 >@@ -22,7 +22,7 @@ > > #ifdef SAVE_RCSID > static char rcsid = >- "@(#) $Id: SDL_blit_N.c,v 1.11 2005/04/20 05:57:39 icculus Exp $"; >+ "@(#) $Id: SDL_blit_N.c,v 1.9 2004/01/04 16:49:21 slouken Exp $"; > #endif > > #include <stdio.h> >@@ -35,655 +35,18 @@ > > /* Functions to blit from N-bit surfaces to other surfaces */ > >-#ifdef USE_ALTIVEC_BLITTERS >-#include <assert.h> >-#ifdef MACOSX >-#include <sys/sysctl.h> >-#include <stdlib.h> >-static size_t GetL3CacheSize( void ) >-{ >- const char key[] = "hw.l3cachesize"; >- u_int64_t result = 0; >- size_t typeSize = sizeof( result ); >- >- >- int err = sysctlbyname( key, &result, &typeSize, NULL, 0 ); >- if( 0 != err ) return 0; >- >- return result; >-} >-#else >-static size_t GetL3CacheSize( void ) >-{ >- /* XXX: Just guess G4 */ >- return 2097152; >-} >-#endif /* MACOSX */ >- >-#define UNALIGNED_PTR(x) (((size_t) x) & 0x0000000F) >-#define VSWIZZLE32(a,b,c,d) (vector unsigned char) \ >- ( 0x00+a, 0x00+b, 0x00+c, 0x00+d, \ >- 0x04+a, 0x04+b, 0x04+c, 0x04+d, \ >- 0x08+a, 0x08+b, 0x08+c, 0x08+d, \ >- 0x0C+a, 0x0C+b, 0x0C+c, 0x0C+d ) >- >-#define MAKE8888(dstfmt, r, g, b, a) \ >- ( ((r<<dstfmt->Rshift)&dstfmt->Rmask) | \ >- ((g<<dstfmt->Gshift)&dstfmt->Gmask) | \ >- ((b<<dstfmt->Bshift)&dstfmt->Bmask) | \ >- ((a<<dstfmt->Ashift)&dstfmt->Amask) ) >- >-/* >- * Data Stream Touch...Altivec cache prefetching. >- * >- * Don't use this on a G5...however, the speed boost is very significant >- * on a G4. >- */ >-#define DST_CHAN_SRC 1 >-#define DST_CHAN_DEST 2 >- >-/* macro to set DST control word value... */ >-#define DST_CTRL(size, count, stride) \ >- (((size) << 24) | ((count) << 16) | (stride)) >- >-#define VEC_ALIGNER(src) ((UNALIGNED_PTR(src)) \ >- ? vec_lvsl(0, src) \ >- : vec_add(vec_lvsl(8, src), vec_splat_u8(8))) >- >-/* Calculate the permute vector used for 32->32 swizzling */ >-static vector unsigned char calc_swizzle32(const SDL_PixelFormat *srcfmt, >- const SDL_PixelFormat *dstfmt) >-{ >- /* >- * We have to assume that the bits that aren't used by other >- * colors is alpha, and it's one complete byte, since some formats >- * leave alpha with a zero mask, but we should still swizzle the bits. >- */ >- /* ARGB */ >- const static struct SDL_PixelFormat default_pixel_format = { >- NULL, 0, 0, >- 0, 0, 0, 0, >- 16, 8, 0, 24, >- 0x00FF0000, 0x0000FF00, 0x000000FF, 0xFF000000, >- 0, 0}; >- if (!srcfmt) { >- srcfmt = &default_pixel_format; >- } >- if (!dstfmt) { >- dstfmt = &default_pixel_format; >- } >- vector unsigned char plus = (vector unsigned char)( 0x00, 0x00, 0x00, 0x00, >- 0x04, 0x04, 0x04, 0x04, >- 0x08, 0x08, 0x08, 0x08, >- 0x0C, 0x0C, 0x0C, 0x0C ); >- vector unsigned char vswiz; >- vector unsigned int srcvec; >-#define RESHIFT(X) (3 - ((X) >> 3)) >- Uint32 rmask = RESHIFT(srcfmt->Rshift) << (dstfmt->Rshift); >- Uint32 gmask = RESHIFT(srcfmt->Gshift) << (dstfmt->Gshift); >- Uint32 bmask = RESHIFT(srcfmt->Bshift) << (dstfmt->Bshift); >- Uint32 amask; >- /* Use zero for alpha if either surface doesn't have alpha */ >- if (dstfmt->Amask) { >- amask = ((srcfmt->Amask) ? RESHIFT(srcfmt->Ashift) : 0x10) << (dstfmt->Ashift); >- } else { >- amask = 0x10101010 & ((dstfmt->Rmask | dstfmt->Gmask | dstfmt->Bmask) ^ 0xFFFFFFFF); >- } >-#undef RESHIFT >- ((unsigned int *)&srcvec)[0] = (rmask | gmask | bmask | amask); >- vswiz = vec_add(plus, (vector unsigned char)vec_splat(srcvec, 0)); >- return(vswiz); >-} >- >-static void Blit_RGB888_RGB565(SDL_BlitInfo *info); >-static void Blit_RGB888_RGB565Altivec(SDL_BlitInfo *info) { >- int height = info->d_height; >- Uint8 *src = (Uint8 *) info->s_pixels; >- int srcskip = info->s_skip; >- Uint8 *dst = (Uint8 *) info->d_pixels; >- int dstskip = info->d_skip; >- SDL_PixelFormat *srcfmt = info->src; >- vector unsigned char valpha = vec_splat_u8(0); >- vector unsigned char vpermute = calc_swizzle32(srcfmt, NULL); >- vector unsigned char vgmerge = (vector unsigned char)( >- 0x00, 0x02, 0x00, 0x06, >- 0x00, 0x0a, 0x00, 0x0e, >- 0x00, 0x12, 0x00, 0x16, >- 0x00, 0x1a, 0x00, 0x1e); >- vector unsigned short v1 = vec_splat_u16(1); >- vector unsigned short v3 = vec_splat_u16(3); >- vector unsigned short v3f = (vector unsigned short)( >- 0x003f, 0x003f, 0x003f, 0x003f, >- 0x003f, 0x003f, 0x003f, 0x003f); >- vector unsigned short vfc = (vector unsigned short)( >- 0x00fc, 0x00fc, 0x00fc, 0x00fc, >- 0x00fc, 0x00fc, 0x00fc, 0x00fc); >- vector unsigned short vf800 = (vector unsigned short)vec_splat_u8(-7); >- vf800 = vec_sl(vf800, vec_splat_u16(8)); >- >- while (height--) { >- vector unsigned char valigner; >- vector unsigned char voverflow; >- vector unsigned char vsrc; >- >- int width = info->d_width; >- int extrawidth; >- >- /* do scalar until we can align... */ >-#define ONE_PIXEL_BLEND(condition, widthvar) \ >- while (condition) { \ >- Uint32 pixel; \ >- unsigned sR, sG, sB, sA; \ >- DISEMBLE_RGBA((Uint8 *)src, 4, srcfmt, pixel, \ >- sR, sG, sB, sA); \ >- *(Uint16 *)(dst) = (((sR << 8) & 0x0000F800) | \ >- ((sG << 3) & 0x000007E0) | \ >- ((sB >> 3) & 0x0000001F)); \ >- dst += 2; \ >- src += 4; \ >- widthvar--; \ >- } >- >- ONE_PIXEL_BLEND(((UNALIGNED_PTR(dst)) && (width)), width); >- >- /* After all that work, here's the vector part! */ >- extrawidth = (width % 8); /* trailing unaligned stores */ >- width -= extrawidth; >- vsrc = vec_ld(0, src); >- valigner = VEC_ALIGNER(src); >- >- while (width) { >- vector unsigned short vpixel, vrpixel, vgpixel, vbpixel; >- vector unsigned int vsrc1, vsrc2; >- vector unsigned char vdst; >- >- voverflow = vec_ld(15, src); >- vsrc = vec_perm(vsrc, voverflow, valigner); >- vsrc1 = (vector unsigned int)vec_perm(vsrc, valpha, vpermute); >- src += 16; >- vsrc = voverflow; >- voverflow = vec_ld(15, src); >- vsrc = vec_perm(vsrc, voverflow, valigner); >- vsrc2 = (vector unsigned int)vec_perm(vsrc, valpha, vpermute); >- /* 1555 */ >- vpixel = (vector unsigned short)vec_packpx(vsrc1, vsrc2); >- vgpixel = (vector unsigned short)vec_perm(vsrc1, vsrc2, vgmerge); >- vgpixel = vec_and(vgpixel, vfc); >- vgpixel = vec_sl(vgpixel, v3); >- vrpixel = vec_sl(vpixel, v1); >- vrpixel = vec_and(vrpixel, vf800); >- vbpixel = vec_and(vpixel, v3f); >- vdst = vec_or((vector unsigned char)vrpixel, (vector unsigned char)vgpixel); >- /* 565 */ >- vdst = vec_or(vdst, (vector unsigned char)vbpixel); >- vec_st(vdst, 0, dst); >- >- width -= 8; >- src += 16; >- dst += 16; >- vsrc = voverflow; >- } >- >- assert(width == 0); >- >- >- /* do scalar until we can align... */ >- ONE_PIXEL_BLEND((extrawidth), extrawidth); >-#undef ONE_PIXEL_BLEND >- >- src += srcskip; /* move to next row, accounting for pitch. */ >- dst += dstskip; >- } >- >- >-} >- >-static void Blit_RGB565_32Altivec(SDL_BlitInfo *info) { >- int height = info->d_height; >- Uint8 *src = (Uint8 *) info->s_pixels; >- int srcskip = info->s_skip; >- Uint8 *dst = (Uint8 *) info->d_pixels; >- int dstskip = info->d_skip; >- SDL_PixelFormat *srcfmt = info->src; >- SDL_PixelFormat *dstfmt = info->dst; >- unsigned alpha; >- vector unsigned char valpha; >- vector unsigned char vpermute; >- vector unsigned short vf800; >- vector unsigned int v8 = vec_splat_u32(8); >- vector unsigned int v16 = vec_add(v8, v8); >- vector unsigned short v2 = vec_splat_u16(2); >- vector unsigned short v3 = vec_splat_u16(3); >- /* >- 0x10 - 0x1f is the alpha >- 0x00 - 0x0e evens are the red >- 0x01 - 0x0f odds are zero >- */ >- vector unsigned char vredalpha1 = (vector unsigned char)( >- 0x10, 0x00, 0x01, 0x01, >- 0x10, 0x02, 0x01, 0x01, >- 0x10, 0x04, 0x01, 0x01, >- 0x10, 0x06, 0x01, 0x01 >- ); >- vector unsigned char vredalpha2 = (vector unsigned char)( >- vec_add((vector unsigned int)vredalpha1, vec_sl(v8, v16)) >- ); >- /* >- 0x00 - 0x0f is ARxx ARxx ARxx ARxx >- 0x11 - 0x0f odds are blue >- */ >- vector unsigned char vblue1 = (vector unsigned char)( >- 0x00, 0x01, 0x02, 0x11, >- 0x04, 0x05, 0x06, 0x13, >- 0x08, 0x09, 0x0a, 0x15, >- 0x0c, 0x0d, 0x0e, 0x17 >- ); >- vector unsigned char vblue2 = (vector unsigned char)( >- vec_add((vector unsigned int)vblue1, v8) >- ); >- /* >- 0x00 - 0x0f is ARxB ARxB ARxB ARxB >- 0x10 - 0x0e evens are green >- */ >- vector unsigned char vgreen1 = (vector unsigned char)( >- 0x00, 0x01, 0x10, 0x03, >- 0x04, 0x05, 0x12, 0x07, >- 0x08, 0x09, 0x14, 0x0b, >- 0x0c, 0x0d, 0x16, 0x0f >- ); >- vector unsigned char vgreen2 = (vector unsigned char)( >- vec_add((vector unsigned int)vgreen1, vec_sl(v8, v8)) >- ); >- >- >- assert(srcfmt->BytesPerPixel == 2); >- assert(dstfmt->BytesPerPixel == 4); >- >- vf800 = (vector unsigned short)vec_splat_u8(-7); >- vf800 = vec_sl(vf800, vec_splat_u16(8)); >- >- if (dstfmt->Amask && srcfmt->alpha) { >- ((unsigned char *)&valpha)[0] = alpha = srcfmt->alpha; >- valpha = vec_splat(valpha, 0); >- } else { >- alpha = 0; >- valpha = vec_splat_u8(0); >- } >- >- vpermute = calc_swizzle32(NULL, dstfmt); >- while (height--) { >- vector unsigned char valigner; >- vector unsigned char voverflow; >- vector unsigned char vsrc; >- >- int width = info->d_width; >- int extrawidth; >- >- /* do scalar until we can align... */ >-#define ONE_PIXEL_BLEND(condition, widthvar) \ >- while (condition) { \ >- unsigned sR, sG, sB; \ >- unsigned short pixel = *((unsigned short *)src); \ >- sR = (pixel >> 8) & 0xf8; \ >- sG = (pixel >> 3) & 0xfc; \ >- sB = (pixel << 3) & 0xf8; \ >- ASSEMBLE_RGBA(dst, 4, dstfmt, sR, sG, sB, alpha); \ >- src += 2; \ >- dst += 4; \ >- widthvar--; \ >- } >- ONE_PIXEL_BLEND(((UNALIGNED_PTR(dst)) && (width)), width); >- >- /* After all that work, here's the vector part! */ >- extrawidth = (width % 8); /* trailing unaligned stores */ >- width -= extrawidth; >- vsrc = vec_ld(0, src); >- valigner = VEC_ALIGNER(src); >- >- while (width) { >- vector unsigned short vR, vG, vB; >- vector unsigned char vdst1, vdst2; >- >- voverflow = vec_ld(15, src); >- vsrc = vec_perm(vsrc, voverflow, valigner); >- >- vR = vec_and((vector unsigned short)vsrc, vf800); >- vB = vec_sl((vector unsigned short)vsrc, v3); >- vG = vec_sl(vB, v2); >- >- vdst1 = (vector unsigned char)vec_perm((vector unsigned char)vR, valpha, vredalpha1); >- vdst1 = vec_perm(vdst1, (vector unsigned char)vB, vblue1); >- vdst1 = vec_perm(vdst1, (vector unsigned char)vG, vgreen1); >- vdst1 = vec_perm(vdst1, valpha, vpermute); >- vec_st(vdst1, 0, dst); >- >- vdst2 = (vector unsigned char)vec_perm((vector unsigned char)vR, valpha, vredalpha2); >- vdst2 = vec_perm(vdst2, (vector unsigned char)vB, vblue2); >- vdst2 = vec_perm(vdst2, (vector unsigned char)vG, vgreen2); >- vdst2 = vec_perm(vdst2, valpha, vpermute); >- vec_st(vdst2, 16, dst); >- >- width -= 8; >- dst += 32; >- src += 16; >- vsrc = voverflow; >- } >- >- assert(width == 0); >- >- >- /* do scalar until we can align... */ >- ONE_PIXEL_BLEND((extrawidth), extrawidth); >-#undef ONE_PIXEL_BLEND >- >- src += srcskip; /* move to next row, accounting for pitch. */ >- dst += dstskip; >- } >- >-} >- >-static void BlitNtoNKey(SDL_BlitInfo *info); >-static void BlitNtoNKeyCopyAlpha(SDL_BlitInfo *info); >-static void Blit32to32KeyAltivec(SDL_BlitInfo *info) >-{ >- int height = info->d_height; >- Uint32 *srcp = (Uint32 *) info->s_pixels; >- int srcskip = info->s_skip; >- Uint32 *dstp = (Uint32 *) info->d_pixels; >- int dstskip = info->d_skip; >- SDL_PixelFormat *srcfmt = info->src; >- int srcbpp = srcfmt->BytesPerPixel; >- SDL_PixelFormat *dstfmt = info->dst; >- int dstbpp = dstfmt->BytesPerPixel; >- int copy_alpha = (srcfmt->Amask && dstfmt->Amask); >- unsigned alpha = dstfmt->Amask ? srcfmt->alpha : 0; >- Uint32 rgbmask = srcfmt->Rmask | srcfmt->Gmask | srcfmt->Bmask; >- Uint32 ckey = info->src->colorkey; >- vector unsigned int valpha; >- vector unsigned char vpermute; >- vector unsigned char vzero; >- vector unsigned int vckey; >- vector unsigned int vrgbmask; >- vpermute = calc_swizzle32(srcfmt, dstfmt); >- if (info->d_width < 16) { >- if(copy_alpha) { >- return BlitNtoNKeyCopyAlpha(info); >- } else { >- return BlitNtoNKey(info); >- } >- } >- vzero = vec_splat_u8(0); >- if (alpha) { >- ((unsigned char *)&valpha)[0] = (unsigned char)alpha; >- valpha = (vector unsigned int)vec_splat((vector unsigned char)valpha, 0); >- } else { >- valpha = (vector unsigned int)vzero; >- } >- ckey &= rgbmask; >- ((unsigned int *)&vckey)[0] = ckey; >- vckey = vec_splat(vckey, 0); >- ((unsigned int *)&vrgbmask)[0] = rgbmask; >- vrgbmask = vec_splat(vrgbmask, 0); >- >- while (height--) { >-#define ONE_PIXEL_BLEND(condition, widthvar) \ >- if (copy_alpha) { \ >- while (condition) { \ >- Uint32 pixel; \ >- unsigned sR, sG, sB, sA; \ >- DISEMBLE_RGBA((Uint8 *)srcp, srcbpp, srcfmt, pixel, \ >- sR, sG, sB, sA); \ >- if ( (pixel & rgbmask) != ckey ) { \ >- ASSEMBLE_RGBA((Uint8 *)dstp, dstbpp, dstfmt, \ >- sR, sG, sB, sA); \ >- } \ >- ((Uint8 *)dstp) += dstbpp; \ >- ((Uint8 *)srcp) += srcbpp; \ >- widthvar--; \ >- } \ >- } else { \ >- while (condition) { \ >- Uint32 pixel; \ >- unsigned sR, sG, sB; \ >- RETRIEVE_RGB_PIXEL((Uint8 *)srcp, srcbpp, pixel); \ >- if ( pixel != ckey ) { \ >- RGB_FROM_PIXEL(pixel, srcfmt, sR, sG, sB); \ >- ASSEMBLE_RGBA((Uint8 *)dstp, dstbpp, dstfmt, \ >- sR, sG, sB, alpha); \ >- } \ >- ((Uint8 *)dstp) += dstbpp; \ >- ((Uint8 *)srcp) += srcbpp; \ >- widthvar--; \ >- } \ >- } >- int width = info->d_width; >- ONE_PIXEL_BLEND((UNALIGNED_PTR(dstp)) && (width), width); >- assert(width > 0); >- if (width > 0) { >- int extrawidth = (width % 4); >- vector unsigned char valigner = VEC_ALIGNER(srcp); >- vector unsigned int vs = vec_ld(0, srcp); >- width -= extrawidth; >- assert(width >= 4); >- while (width) { >- vector unsigned char vsel; >- vector unsigned int vd; >- vector unsigned int voverflow = vec_ld(15, srcp); >- /* load the source vec */ >- vs = vec_perm(vs, voverflow, valigner); >- /* vsel is set for items that match the key */ >- vsel = (vector unsigned char)vec_and(vs, vrgbmask); >- vsel = (vector unsigned char)vec_cmpeq(vs, vckey); >- /* permute the src vec to the dest format */ >- vs = vec_perm(vs, valpha, vpermute); >- /* load the destination vec */ >- vd = vec_ld(0, dstp); >- /* select the source and dest into vs */ >- vd = (vector unsigned int)vec_sel((vector unsigned char)vs, (vector unsigned char)vd, vsel); >- >- vec_st(vd, 0, dstp); >- srcp += 4; >- width -= 4; >- dstp += 4; >- vs = voverflow; >- } >- ONE_PIXEL_BLEND((extrawidth), extrawidth); >-#undef ONE_PIXEL_BLEND >- srcp += srcskip >> 2; >- dstp += dstskip >> 2; >- } >- } >-} >- >-/* Altivec code to swizzle one 32-bit surface to a different 32-bit format. */ >-/* Use this on a G5 */ >-static void ConvertAltivec32to32_noprefetch(SDL_BlitInfo *info) >-{ >- int height = info->d_height; >- Uint32 *src = (Uint32 *) info->s_pixels; >- int srcskip = info->s_skip; >- Uint32 *dst = (Uint32 *) info->d_pixels; >- int dstskip = info->d_skip; >- SDL_PixelFormat *srcfmt = info->src; >- int srcbpp = srcfmt->BytesPerPixel; >- SDL_PixelFormat *dstfmt = info->dst; >- int dstbpp = dstfmt->BytesPerPixel; >- vector unsigned int vzero = vec_splat_u32(0); >- vector unsigned char vpermute = calc_swizzle32(srcfmt, dstfmt); >- if (dstfmt->Amask && !srcfmt->Amask) { >- if (srcfmt->alpha) { >- vector unsigned char valpha; >- ((unsigned char *)&valpha)[0] = srcfmt->alpha; >- vzero = (vector unsigned int)vec_splat(valpha, 0); >- } >- } >- >- assert(srcbpp == 4); >- assert(dstbpp == 4); >- >- while (height--) { >- vector unsigned char valigner; >- vector unsigned int vbits; >- vector unsigned int voverflow; >- Uint32 bits; >- Uint8 r, g, b, a; >- >- int width = info->d_width; >- int extrawidth; >- >- /* do scalar until we can align... */ >- while ((UNALIGNED_PTR(dst)) && (width)) { >- bits = *(src++); >- RGBA_FROM_8888(bits, srcfmt, r, g, b, a); >- *(dst++) = MAKE8888(dstfmt, r, g, b, a); >- width--; >- } >- >- /* After all that work, here's the vector part! */ >- extrawidth = (width % 4); >- width -= extrawidth; >- valigner = VEC_ALIGNER(src); >- vbits = vec_ld(0, src); >- >- while (width) { >- voverflow = vec_ld(15, src); >- src += 4; >- width -= 4; >- vbits = vec_perm(vbits, voverflow, valigner); /* src is ready. */ >- vbits = vec_perm(vbits, vzero, vpermute); /* swizzle it. */ >- vec_st(vbits, 0, dst); /* store it back out. */ >- dst += 4; >- vbits = voverflow; >- } >- >- assert(width == 0); >- >- /* cover pixels at the end of the row that didn't fit in 16 bytes. */ >- while (extrawidth) { >- bits = *(src++); /* max 7 pixels, don't bother with prefetch. */ >- RGBA_FROM_8888(bits, srcfmt, r, g, b, a); >- *(dst++) = MAKE8888(dstfmt, r, g, b, a); >- extrawidth--; >- } >- >- src += srcskip >> 2; /* move to next row, accounting for pitch. */ >- dst += dstskip >> 2; >- } >- >-} >- >-/* Altivec code to swizzle one 32-bit surface to a different 32-bit format. */ >-/* Use this on a G4 */ >-static void ConvertAltivec32to32_prefetch(SDL_BlitInfo *info) >-{ >- const int scalar_dst_lead = sizeof (Uint32) * 4; >- const int vector_dst_lead = sizeof (Uint32) * 16; >+#ifdef USE_ASMBLIT > >- int height = info->d_height; >- Uint32 *src = (Uint32 *) info->s_pixels; >- int srcskip = info->s_skip; >- Uint32 *dst = (Uint32 *) info->d_pixels; >- int dstskip = info->d_skip; >- SDL_PixelFormat *srcfmt = info->src; >- int srcbpp = srcfmt->BytesPerPixel; >- SDL_PixelFormat *dstfmt = info->dst; >- int dstbpp = dstfmt->BytesPerPixel; >- vector unsigned int vzero = vec_splat_u32(0); >- vector unsigned char vpermute = calc_swizzle32(srcfmt, dstfmt); >- if (dstfmt->Amask && !srcfmt->Amask) { >- if (srcfmt->alpha) { >- vector unsigned char valpha; >- ((unsigned char *)&valpha)[0] = srcfmt->alpha; >- vzero = (vector unsigned int)vec_splat(valpha, 0); >- } >- } >- >- assert(srcbpp == 4); >- assert(dstbpp == 4); >- >- while (height--) { >- vector unsigned char valigner; >- vector unsigned int vbits; >- vector unsigned int voverflow; >- Uint32 bits; >- Uint8 r, g, b, a; >- >- int width = info->d_width; >- int extrawidth; >- >- /* do scalar until we can align... */ >- while ((UNALIGNED_PTR(dst)) && (width)) { >- vec_dstt(src+scalar_dst_lead, DST_CTRL(2,32,1024), DST_CHAN_SRC); >- vec_dstst(dst+scalar_dst_lead, DST_CTRL(2,32,1024), DST_CHAN_DEST); >- bits = *(src++); >- RGBA_FROM_8888(bits, srcfmt, r, g, b, a); >- *(dst++) = MAKE8888(dstfmt, r, g, b, a); >- width--; >- } >- >- /* After all that work, here's the vector part! */ >- extrawidth = (width % 4); >- width -= extrawidth; >- valigner = VEC_ALIGNER(src); >- vbits = vec_ld(0, src); >- >- while (width) { >- vec_dstt(src+vector_dst_lead, DST_CTRL(2,32,1024), DST_CHAN_SRC); >- vec_dstst(dst+vector_dst_lead, DST_CTRL(2,32,1024), DST_CHAN_DEST); >- voverflow = vec_ld(15, src); >- src += 4; >- width -= 4; >- vbits = vec_perm(vbits, voverflow, valigner); /* src is ready. */ >- vbits = vec_perm(vbits, vzero, vpermute); /* swizzle it. */ >- vec_st(vbits, 0, dst); /* store it back out. */ >- dst += 4; >- vbits = voverflow; >- } >- >- assert(width == 0); >- >- /* cover pixels at the end of the row that didn't fit in 16 bytes. */ >- while (extrawidth) { >- bits = *(src++); /* max 7 pixels, don't bother with prefetch. */ >- RGBA_FROM_8888(bits, srcfmt, r, g, b, a); >- *(dst++) = MAKE8888(dstfmt, r, g, b, a); >- extrawidth--; >- } >- >- src += srcskip >> 2; /* move to next row, accounting for pitch. */ >- dst += dstskip >> 2; >- } >+/* Heheheh, we coerce Hermes into using SDL blit information */ >+#define X86_ASSEMBLER >+#define HermesConverterInterface SDL_BlitInfo >+#define HermesClearInterface void >+#define STACKCALL > >- vec_dss(DST_CHAN_SRC); >- vec_dss(DST_CHAN_DEST); >-} >+#include "HeadMMX.h" >+#include "HeadX86.h" > >-static Uint32 GetBlitFeatures( void ) >-{ >- static Uint32 features = 0xffffffff; >- if (features == 0xffffffff) { >- /* Provide an override for testing .. */ >- char *override = getenv("SDL_ALTIVEC_BLIT_FEATURES"); >- if (override) { >- features = 0; >- sscanf(override, "%u", &features); >- } else { >- features = ( 0 >- /* Feature 1 is has-MMX */ >- | ((SDL_HasMMX()) ? 1 : 0) >- /* Feature 2 is has-AltiVec */ >- | ((SDL_HasAltiVec()) ? 2 : 0) >- /* Feature 4 is dont-use-prefetch */ >- | ((GetL3CacheSize() == 0) ? 4 : 0) >- ); >- } >- } >- return features; >-} > #else >-/* Feature 1 is has-MMX */ >-#define GetBlitFeatures() ((Uint32)(SDL_HasMMX() ? 1 : 0)) >-#endif > > /* This is now endian dependent */ > #if ( SDL_BYTEORDER == SDL_LIL_ENDIAN ) >@@ -694,19 +57,6 @@ > #define LO 1 > #endif > >-#ifdef USE_ASMBLIT >- >-/* Heheheh, we coerce Hermes into using SDL blit information */ >-#define X86_ASSEMBLER >-#define HermesConverterInterface SDL_BlitInfo >-#define HermesClearInterface void >-#define STACKCALL >- >-#include "HeadMMX.h" >-#include "HeadX86.h" >- >-#else >- > /* Special optimized blit for RGB 8-8-8 --> RGB 3-3-2 */ > #define RGB888_RGB332(dst, src) { \ > dst = (((src)&0x00E00000)>>16)| \ >@@ -1056,7 +406,11 @@ > > > /* Special optimized blit for RGB 5-6-5 --> 32-bit RGB surfaces */ >-#define RGB565_32(dst, src, map) (map[src[LO]*2] + map[src[HI]*2+1]) >+#if ( SDL_BYTEORDER == SDL_LIL_ENDIAN ) >+#define RGB565_32(dst, src, map) (map[src[0]*2] + map[src[1]*2+1]) >+#else /* ( SDL_BYTEORDER == SDL_BIG_ENDIAN ) */ >+#define RGB565_32(dst, src, map) (map[src[1]*2] + map[src[0]*2+1]) >+#endif > static void Blit_RGB565_32(SDL_BlitInfo *info, const Uint32 *map) > { > #ifndef USE_DUFFS_LOOP >@@ -2068,10 +1422,10 @@ > Uint32 srcR, srcG, srcB; > int dstbpp; > Uint32 dstR, dstG, dstB; >- Uint32 blit_features; >+ SDL_bool cpu_mmx; > void *aux_data; > SDL_loblit blitfunc; >- enum { NO_ALPHA=1, SET_ALPHA=2, COPY_ALPHA=4 } alpha; >+ enum { NO_ALPHA, SET_ALPHA, COPY_ALPHA } alpha; > }; > static const struct blit_table normal_blit_1[] = { > /* Default for 8-bit RGB source, an invalid combination */ >@@ -2086,11 +1440,6 @@ > { 0x0000F800,0x000007E0,0x0000001F, 2, 0x0000001F,0x000003E0,0x00007C00, > 0, ConvertX86p16_16BGR555, ConvertX86, NO_ALPHA }, > #endif >-#ifdef USE_ALTIVEC_BLITTERS >- /* has-altivec */ >- { 0x0000F800,0x000007E0,0x0000001F, 4, 0x00000000,0x00000000,0x00000000, >- 2, NULL, Blit_RGB565_32Altivec, NO_ALPHA | COPY_ALPHA | SET_ALPHA }, >-#endif > { 0x0000F800,0x000007E0,0x0000001F, 4, 0x00FF0000,0x0000FF00,0x000000FF, > 0, NULL, Blit_RGB565_ARGB8888, SET_ALPHA }, > { 0x0000F800,0x000007E0,0x0000001F, 4, 0x000000FF,0x0000FF00,0x00FF0000, >@@ -2136,17 +1485,6 @@ > { 0x00FF0000,0x0000FF00,0x000000FF, 4, 0x0000FF00,0x00FF0000,0xFF000000, > 0, ConvertX86p32_32BGRA888, ConvertX86, NO_ALPHA }, > #else >-#ifdef USE_ALTIVEC_BLITTERS >- /* has-altivec | dont-use-prefetch */ >- { 0x00000000,0x00000000,0x00000000, 4, 0x00000000,0x00000000,0x00000000, >- 6, NULL, ConvertAltivec32to32_noprefetch, NO_ALPHA | COPY_ALPHA | SET_ALPHA }, >- /* has-altivec */ >- { 0x00000000,0x00000000,0x00000000, 4, 0x00000000,0x00000000,0x00000000, >- 2, NULL, ConvertAltivec32to32_prefetch, NO_ALPHA | COPY_ALPHA | SET_ALPHA }, >- /* has-altivec */ >- { 0x00000000,0x00000000,0x00000000, 2, 0x0000F800,0x000007E0,0x0000001F, >- 2, NULL, Blit_RGB888_RGB565Altivec, NO_ALPHA }, >-#endif > { 0x00FF0000,0x0000FF00,0x000000FF, 2, 0x0000F800,0x000007E0,0x0000001F, > 0, NULL, Blit_RGB888_RGB565, NO_ALPHA }, > { 0x00FF0000,0x0000FF00,0x000000FF, 2, 0x00007C00,0x000003E0,0x0000001F, >@@ -2159,9 +1497,6 @@ > normal_blit_1, normal_blit_2, normal_blit_3, normal_blit_4 > }; > >-/* Mask matches table, or table entry is zero */ >-#define MASKOK(x, y) (((x) == (y)) || ((y) == 0x00000000)) >- > SDL_loblit SDL_CalculateBlitN(SDL_Surface *surface, int blit_index) > { > struct private_swaccel *sdata; >@@ -2197,12 +1532,6 @@ > else if(dstfmt->BytesPerPixel == 1) > return BlitNto1Key; > else { >-#ifdef USE_ALTIVEC_BLITTERS >- if((srcfmt->BytesPerPixel == 4) && (dstfmt->BytesPerPixel == 4) && SDL_HasAltiVec()) { >- return Blit32to32KeyAltivec; >- } else >-#endif >- > if(srcfmt->Amask && dstfmt->Amask) > return BlitNtoNKeyCopyAlpha; > else >@@ -2232,20 +1561,20 @@ > } > } else { > /* Now the meat, choose the blitter we want */ >- int a_need = 0; >+ int a_need = 0; > if(dstfmt->Amask) > a_need = srcfmt->Amask ? COPY_ALPHA : SET_ALPHA; > table = normal_blit[srcfmt->BytesPerPixel-1]; >- for ( which=0; table[which].dstbpp; ++which ) { >- if ( MASKOK(srcfmt->Rmask, table[which].srcR) && >- MASKOK(srcfmt->Gmask, table[which].srcG) && >- MASKOK(srcfmt->Bmask, table[which].srcB) && >- MASKOK(dstfmt->Rmask, table[which].dstR) && >- MASKOK(dstfmt->Gmask, table[which].dstG) && >- MASKOK(dstfmt->Bmask, table[which].dstB) && >- dstfmt->BytesPerPixel == table[which].dstbpp && >- (a_need & table[which].alpha) == a_need && >- ((table[which].blit_features & GetBlitFeatures()) == table[which].blit_features) ) >+ for ( which=0; table[which].srcR; ++which ) { >+ if ( srcfmt->Rmask == table[which].srcR && >+ srcfmt->Gmask == table[which].srcG && >+ srcfmt->Bmask == table[which].srcB && >+ dstfmt->BytesPerPixel == table[which].dstbpp && >+ dstfmt->Rmask == table[which].dstR && >+ dstfmt->Gmask == table[which].dstG && >+ dstfmt->Bmask == table[which].dstB && >+ (a_need & table[which].alpha) == a_need && >+ (table[which].cpu_mmx == SDL_HasMMX())) > break; > } > sdata->aux_data = table[which].aux_data;
You cannot view the attachment while viewing its details because your browser does not support IFRAMEs.
View the attachment on a separate page
.
View Attachment As Diff
View Attachment As Raw
Actions:
View
|
Diff
Attachments on
bug 104533
:
67787
|
67789
|
67790