X-Git-Url: https://git.xonotic.org/?p=xonotic%2Fdarkplaces.git;a=blobdiff_plain;f=dpsoftrast.c;h=179a4d0c5fe93cae3b4668e1f4d9221c6a79082c;hp=aff3befeb399e0075ed15a031d815810d4742baf;hb=7b54444c08e4d206acaf8e302be47e6dfe1f9b7d;hpb=2da3c048ed669c2221c61fb12c108c9588104af9 diff --git a/dpsoftrast.c b/dpsoftrast.c index aff3befe..179a4d0c 100644 --- a/dpsoftrast.c +++ b/dpsoftrast.c @@ -3,15 +3,11 @@ #define _USE_MATH_DEFINES #include #include "quakedef.h" +#include "thread.h" #include "dpsoftrast.h" -#ifdef USE_SDL -#define USE_THREADS -#endif - -#ifdef USE_THREADS -#include -#include +#ifdef _MSC_VER +#pragma warning(disable : 4324) #endif #ifndef __cplusplus @@ -21,57 +17,86 @@ typedef qboolean bool; #define ALIGN_SIZE 16 #define ATOMIC_SIZE 32 -#ifdef SSE2_PRESENT - #if defined(__GNUC__) +#ifdef SSE_POSSIBLE + #if defined(__APPLE__) + #include + #define ALIGN(var) var __attribute__((__aligned__(16))) + #define ATOMIC(var) var __attribute__((__aligned__(32))) + #define MEMORY_BARRIER (_mm_sfence()) + #define ATOMIC_COUNTER volatile int32_t + #define ATOMIC_INCREMENT(counter) (OSAtomicIncrement32Barrier(&(counter))) + #define ATOMIC_DECREMENT(counter) (OSAtomicDecrement32Barrier(&(counter))) + #define ATOMIC_ADD(counter, val) ((void)OSAtomicAdd32Barrier((val), &(counter))) + #elif defined(__GNUC__) && defined(WIN32) #define ALIGN(var) var __attribute__((__aligned__(16))) #define ATOMIC(var) var __attribute__((__aligned__(32))) - #ifdef USE_THREADS - #define MEMORY_BARRIER (_mm_sfence()) - //(__sync_synchronize()) - #define ATOMIC_COUNTER volatile int - #define ATOMIC_INCREMENT(counter) (__sync_add_and_fetch(&(counter), 1)) - #define ATOMIC_DECREMENT(counter) (__sync_add_and_fetch(&(counter), -1)) - #define ATOMIC_ADD(counter, val) ((void)__sync_fetch_and_add(&(counter), (val))) - #endif + #define MEMORY_BARRIER (_mm_sfence()) + //(__sync_synchronize()) + #define ATOMIC_COUNTER volatile LONG + // this LONG * cast serves to fix an issue with broken mingw + // packages on Ubuntu; these only declare the function to take + // a LONG *, causing a compile error here. This seems to be + // error- and warn-free on platforms that DO declare + // InterlockedIncrement correctly, like mingw on Windows. + #define ATOMIC_INCREMENT(counter) (InterlockedIncrement((LONG *) &(counter))) + #define ATOMIC_DECREMENT(counter) (InterlockedDecrement((LONG *) &(counter))) + #define ATOMIC_ADD(counter, val) ((void)InterlockedExchangeAdd((LONG *) &(counter), (val))) + #elif defined(__GNUC__) + #define ALIGN(var) var __attribute__((__aligned__(16))) + #define ATOMIC(var) var __attribute__((__aligned__(32))) + #define MEMORY_BARRIER (_mm_sfence()) + //(__sync_synchronize()) + #define ATOMIC_COUNTER volatile int + #define ATOMIC_INCREMENT(counter) (__sync_add_and_fetch(&(counter), 1)) + #define ATOMIC_DECREMENT(counter) (__sync_add_and_fetch(&(counter), -1)) + #define ATOMIC_ADD(counter, val) ((void)__sync_fetch_and_add(&(counter), (val))) #elif defined(_MSC_VER) #define ALIGN(var) __declspec(align(16)) var #define ATOMIC(var) __declspec(align(32)) var - #ifdef USE_THREADS - #define MEMORY_BARRIER (_mm_sfence()) - //(MemoryBarrier()) - #define ATOMIC_COUNTER volatile LONG - #define ATOMIC_INCREMENT(counter) (InterlockedIncrement(&(counter))) - #define ATOMIC_DECREMENT(counter) (InterlockedDecrement(&(counter))) - #define ATOMIC_ADD(counter, val) (InterlockedExchangeAdd(&(counter), (val))) - #endif - #else - #undef USE_THREADS - #undef SSE2_PRESENT + #define MEMORY_BARRIER (_mm_sfence()) + //(MemoryBarrier()) + #define ATOMIC_COUNTER volatile LONG + #define ATOMIC_INCREMENT(counter) (InterlockedIncrement(&(counter))) + #define ATOMIC_DECREMENT(counter) (InterlockedDecrement(&(counter))) + #define ATOMIC_ADD(counter, val) ((void)InterlockedExchangeAdd(&(counter), (val))) #endif #endif -#ifndef SSE2_PRESENT - #define ALIGN(var) var - #define ATOMIC(var) var +#ifndef ALIGN +#define ALIGN(var) var #endif - -#ifndef USE_THREADS - #define MEMORY_BARRIER ((void)0) - #define ATOMIC_COUNTER int - #define ATOMIC_INCREMENT(counter) (++(counter)) - #define ATOMIC_DECREMENT(counter) (--(counter)) - #define ATOMIC_ADD(counter, val) ((void)((counter) += (val))) +#ifndef ATOMIC +#define ATOMIC(var) var +#endif +#ifndef MEMORY_BARRIER +#define MEMORY_BARRIER ((void)0) +#endif +#ifndef ATOMIC_COUNTER +#define ATOMIC_COUNTER int +#endif +#ifndef ATOMIC_INCREMENT +#define ATOMIC_INCREMENT(counter) (++(counter)) +#endif +#ifndef ATOMIC_DECREMENT +#define ATOMIC_DECREMENT(counter) (--(counter)) +#endif +#ifndef ATOMIC_ADD +#define ATOMIC_ADD(counter, val) ((void)((counter) += (val))) #endif -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE #include +#if defined(__GNUC__) && (__GNUC < 4 || __GNUC_MINOR__ < 6) && !defined(__clang__) + #define _mm_cvtss_f32(val) (__builtin_ia32_vec_ext_v4sf ((__v4sf)(val), 0)) +#endif + #define MM_MALLOC(size) _mm_malloc(size, ATOMIC_SIZE) static void *MM_CALLOC(size_t nmemb, size_t size) { void *ptr = _mm_malloc(nmemb*size, ATOMIC_SIZE); - if(ptr != NULL) memset(ptr, 0, nmemb*size); + if (ptr != NULL) memset(ptr, 0, nmemb*size); return ptr; } @@ -178,15 +203,17 @@ typedef ALIGN(struct DPSOFTRAST_State_Span_s int triangle; // triangle this span was generated by int x; // framebuffer x coord int y; // framebuffer y coord - int length; // pixel count int startx; // usable range (according to pixelmask) int endx; // usable range (according to pixelmask) unsigned char *pixelmask; // true for pixels that passed depth test, false for others + int depthbase; // depthbuffer value at x (add depthslope*startx to get first pixel's depthbuffer value) + int depthslope; // depthbuffer value pixel delta } DPSOFTRAST_State_Span); #define DPSOFTRAST_DRAW_MAXSPANS 1024 #define DPSOFTRAST_DRAW_MAXTRIANGLES 128 +#define DPSOFTRAST_DRAW_MAXSPANLENGTH 256 #define DPSOFTRAST_VALIDATE_FB 1 #define DPSOFTRAST_VALIDATE_DEPTHFUNC 2 @@ -204,15 +231,14 @@ typedef enum DPSOFTRAST_BLENDMODE_e DPSOFTRAST_BLENDMODE_MUL2, DPSOFTRAST_BLENDMODE_SUBALPHA, DPSOFTRAST_BLENDMODE_PSEUDOALPHA, + DPSOFTRAST_BLENDMODE_INVADD, DPSOFTRAST_BLENDMODE_TOTAL } DPSOFTRAST_BLENDMODE; typedef ATOMIC(struct DPSOFTRAST_State_Thread_s { -#ifdef USE_THREADS - SDL_Thread *thread; -#endif + void *thread; int index; int cullface; @@ -223,16 +249,16 @@ typedef ATOMIC(struct DPSOFTRAST_State_Thread_s int depthtest; int depthfunc; int scissortest; - int alphatest; - int alphafunc; - float alphavalue; int viewport[4]; int scissor[4]; float depthrange[2]; float polygonoffset[2]; + float clipplane[4]; + ALIGN(float fb_clipplane[4]); int shader_mode; int shader_permutation; + int shader_exactspecularmath; DPSOFTRAST_Texture *texbound[DPSOFTRAST_MAXTEXTUREUNITS]; @@ -244,7 +270,7 @@ typedef ATOMIC(struct DPSOFTRAST_State_Thread_s // derived values (DPSOFTRAST_VALIDATE_FB) int fb_colormask; - int fb_clearscissor[4]; + int fb_scissor[4]; ALIGN(float fb_viewportcenter[4]); ALIGN(float fb_viewportscale[4]); @@ -254,20 +280,25 @@ typedef ATOMIC(struct DPSOFTRAST_State_Thread_s // derived values (DPSOFTRAST_VALIDATE_BLENDFUNC) int fb_blendmode; + // band boundaries + int miny1; + int maxy1; + int miny2; + int maxy2; + ATOMIC(volatile int commandoffset); volatile bool waiting; volatile bool starving; -#ifdef USE_THREADS - SDL_cond *waitcond; - SDL_cond *drawcond; - SDL_mutex *drawmutex; -#endif + void *waitcond; + void *drawcond; + void *drawmutex; int numspans; int numtriangles; DPSOFTRAST_State_Span spans[DPSOFTRAST_DRAW_MAXSPANS]; DPSOFTRAST_State_Triangle triangles[DPSOFTRAST_DRAW_MAXTRIANGLES]; + unsigned char pixelmaskarray[DPSOFTRAST_DRAW_MAXSPANLENGTH+4]; // LordHavoc: padded to allow some termination bytes } DPSOFTRAST_State_Thread); @@ -306,6 +337,7 @@ typedef ATOMIC(struct DPSOFTRAST_State_s int shader_mode; int shader_permutation; + int shader_exactspecularmath; int texture_max; int texture_end; @@ -317,6 +349,8 @@ typedef ATOMIC(struct DPSOFTRAST_State_s // error reporting const char *errorstring; + bool usethreads; + int interlace; int numthreads; DPSOFTRAST_State_Thread *threads; @@ -332,7 +366,9 @@ DPSOFTRAST_State dpsoftrast; #define DPSOFTRAST_DEPTHOFFSET (128.0f) #define DPSOFTRAST_BGRA8_FROM_RGBA32F(r,g,b,a) (((int)(r * 255.0f + 0.5f) << 16) | ((int)(g * 255.0f + 0.5f) << 8) | (int)(b * 255.0f + 0.5f) | ((int)(a * 255.0f + 0.5f) << 24)) #define DPSOFTRAST_DEPTH32_FROM_DEPTH32F(d) ((int)(DPSOFTRAST_DEPTHSCALE * (1-d))) -#define DPSOFTRAST_DRAW_MAXSPANLENGTH 256 + +static void DPSOFTRAST_Draw_DepthTest(DPSOFTRAST_State_Thread *thread, DPSOFTRAST_State_Span *span); +static void DPSOFTRAST_Draw_DepthWrite(const DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Span *span); static void DPSOFTRAST_RecalcViewport(const int *viewport, float *fb_viewportcenter, float *fb_viewportscale) { @@ -346,6 +382,31 @@ static void DPSOFTRAST_RecalcViewport(const int *viewport, float *fb_viewportcen fb_viewportscale[0] = 1.0f; } +static void DPSOFTRAST_RecalcThread(DPSOFTRAST_State_Thread *thread) +{ + if (dpsoftrast.interlace) + { + thread->miny1 = (thread->index*dpsoftrast.fb_height)/(2*dpsoftrast.numthreads); + thread->maxy1 = ((thread->index+1)*dpsoftrast.fb_height)/(2*dpsoftrast.numthreads); + thread->miny2 = ((dpsoftrast.numthreads+thread->index)*dpsoftrast.fb_height)/(2*dpsoftrast.numthreads); + thread->maxy2 = ((dpsoftrast.numthreads+thread->index+1)*dpsoftrast.fb_height)/(2*dpsoftrast.numthreads); + } + else + { + thread->miny1 = thread->miny2 = (thread->index*dpsoftrast.fb_height)/dpsoftrast.numthreads; + thread->maxy1 = thread->maxy2 = ((thread->index+1)*dpsoftrast.fb_height)/dpsoftrast.numthreads; + } +} + +static void DPSOFTRAST_RecalcClipPlane(DPSOFTRAST_State_Thread *thread) +{ + thread->fb_clipplane[0] = thread->clipplane[0] / thread->fb_viewportscale[1]; + thread->fb_clipplane[1] = thread->clipplane[1] / thread->fb_viewportscale[2]; + thread->fb_clipplane[2] = thread->clipplane[2] / thread->fb_viewportscale[3]; + thread->fb_clipplane[3] = thread->clipplane[3] / thread->fb_viewportscale[0]; + thread->fb_clipplane[3] -= thread->fb_viewportcenter[1]*thread->fb_clipplane[0] + thread->fb_viewportcenter[2]*thread->fb_clipplane[1] + thread->fb_viewportcenter[3]*thread->fb_clipplane[2] + thread->fb_viewportcenter[0]*thread->fb_clipplane[3]; +} + static void DPSOFTRAST_RecalcFB(DPSOFTRAST_State_Thread *thread) { // calculate framebuffer scissor, viewport, viewport clipped by scissor, @@ -361,12 +422,14 @@ static void DPSOFTRAST_RecalcFB(DPSOFTRAST_State_Thread *thread) if (x2 > dpsoftrast.fb_width) x2 = dpsoftrast.fb_width; if (y1 < 0) y1 = 0; if (y2 > dpsoftrast.fb_height) y2 = dpsoftrast.fb_height; - thread->fb_clearscissor[0] = x1; - thread->fb_clearscissor[1] = y1; - thread->fb_clearscissor[2] = x2 - x1; - thread->fb_clearscissor[3] = y2 - y1; + thread->fb_scissor[0] = x1; + thread->fb_scissor[1] = y1; + thread->fb_scissor[2] = x2 - x1; + thread->fb_scissor[3] = y2 - y1; DPSOFTRAST_RecalcViewport(thread->viewport, thread->fb_viewportcenter, thread->fb_viewportscale); + DPSOFTRAST_RecalcClipPlane(thread); + DPSOFTRAST_RecalcThread(thread); } static void DPSOFTRAST_RecalcDepthFunc(DPSOFTRAST_State_Thread *thread) @@ -382,7 +445,7 @@ static void DPSOFTRAST_RecalcBlendFunc(DPSOFTRAST_State_Thread *thread) { #define BLENDFUNC(sfactor, dfactor, blendmode) \ case (sfactor<<16)|dfactor: thread->fb_blendmode = blendmode; break; - BLENDFUNC(GL_SRC_COLOR, GL_ONE, DPSOFTRAST_BLENDMODE_SUBALPHA) + BLENDFUNC(GL_SRC_ALPHA, GL_ONE, DPSOFTRAST_BLENDMODE_SUBALPHA) default: thread->fb_blendmode = DPSOFTRAST_BLENDMODE_OPAQUE; break; } } @@ -399,7 +462,7 @@ static void DPSOFTRAST_RecalcBlendFunc(DPSOFTRAST_State_Thread *thread) BLENDFUNC(GL_DST_COLOR, GL_ZERO, DPSOFTRAST_BLENDMODE_MUL) BLENDFUNC(GL_DST_COLOR, GL_SRC_COLOR, DPSOFTRAST_BLENDMODE_MUL2) BLENDFUNC(GL_ONE, GL_ONE_MINUS_SRC_ALPHA, DPSOFTRAST_BLENDMODE_PSEUDOALPHA) - BLENDFUNC(GL_SRC_COLOR, GL_ONE, DPSOFTRAST_BLENDMODE_SUBALPHA) + BLENDFUNC(GL_ONE_MINUS_DST_COLOR, GL_ONE, DPSOFTRAST_BLENDMODE_INVADD) default: thread->fb_blendmode = DPSOFTRAST_BLENDMODE_OPAQUE; break; } } @@ -450,13 +513,13 @@ static void DPSOFTRAST_Texture_Grow(void) dpsoftrast.texture_max *= 2; dpsoftrast.texture = (DPSOFTRAST_Texture *)realloc(dpsoftrast.texture, dpsoftrast.texture_max * sizeof(DPSOFTRAST_Texture)); for (i = 0; i < DPSOFTRAST_MAXTEXTUREUNITS; i++) - if(dpsoftrast.texbound[i]) + if (dpsoftrast.texbound[i]) dpsoftrast.texbound[i] = dpsoftrast.texture + (dpsoftrast.texbound[i] - oldtexture); for (j = 0; j < dpsoftrast.numthreads; j++) { thread = &dpsoftrast.threads[j]; for (i = 0; i < DPSOFTRAST_MAXTEXTUREUNITS; i++) - if(thread->texbound[i]) + if (thread->texbound[i]) thread->texbound[i] = dpsoftrast.texture + (thread->texbound[i] - oldtexture); } } @@ -683,13 +746,16 @@ void DPSOFTRAST_Texture_UpdatePartial(int index, int mip, const unsigned char *p texture = DPSOFTRAST_Texture_GetByIndex(index);if (!texture) return; if (texture->binds) DPSOFTRAST_Flush(); - dst = texture->bytes + (blocky * texture->mipmap[0][2] + blockx) * 4; - while (blockheight > 0) + if (pixels) { - memcpy(dst, pixels, blockwidth * 4); - pixels += blockwidth * 4; - dst += texture->mipmap[0][2] * 4; - blockheight--; + dst = texture->bytes + (blocky * texture->mipmap[0][2] + blockx) * 4; + while (blockheight > 0) + { + memcpy(dst, pixels, blockwidth * 4); + pixels += blockwidth * 4; + dst += texture->mipmap[0][2] * 4; + blockheight--; + } } DPSOFTRAST_Texture_CalculateMipmaps(index); } @@ -699,7 +765,8 @@ void DPSOFTRAST_Texture_UpdateFull(int index, const unsigned char *pixels) texture = DPSOFTRAST_Texture_GetByIndex(index);if (!texture) return; if (texture->binds) DPSOFTRAST_Flush(); - memcpy(texture->bytes, pixels, texture->mipmap[0][1]); + if (pixels) + memcpy(texture->bytes, pixels, texture->mipmap[0][1]); DPSOFTRAST_Texture_CalculateMipmaps(index); } int DPSOFTRAST_Texture_GetWidth(int index, int mip) @@ -742,32 +809,16 @@ void DPSOFTRAST_Texture_Filter(int index, DPSOFTRAST_TEXTURE_FILTER filter) texture->filter = filter; } -void DPSOFTRAST_SetRenderTargets(int width, int height, unsigned int *depthpixels, unsigned int *colorpixels0, unsigned int *colorpixels1, unsigned int *colorpixels2, unsigned int *colorpixels3) -{ - if (width != dpsoftrast.fb_width || height != dpsoftrast.fb_height || depthpixels != dpsoftrast.fb_depthpixels || - colorpixels0 != dpsoftrast.fb_colorpixels[0] || colorpixels1 != dpsoftrast.fb_colorpixels[1] || - colorpixels2 != dpsoftrast.fb_colorpixels[2] || colorpixels3 != dpsoftrast.fb_colorpixels[3]) - DPSOFTRAST_Flush(); - dpsoftrast.fb_width = width; - dpsoftrast.fb_height = height; - dpsoftrast.fb_depthpixels = depthpixels; - dpsoftrast.fb_colorpixels[0] = colorpixels0; - dpsoftrast.fb_colorpixels[1] = colorpixels1; - dpsoftrast.fb_colorpixels[2] = colorpixels2; - dpsoftrast.fb_colorpixels[3] = colorpixels3; -} - static void DPSOFTRAST_Draw_FlushThreads(void); static void DPSOFTRAST_Draw_SyncCommands(void) { - MEMORY_BARRIER; + if(dpsoftrast.usethreads) MEMORY_BARRIER; dpsoftrast.drawcommand = dpsoftrast.commandpool.freecommand; } static void DPSOFTRAST_Draw_FreeCommandPool(int space) { -#ifdef USE_THREADS DPSOFTRAST_State_Thread *thread; int i; int freecommand = dpsoftrast.commandpool.freecommand; @@ -795,20 +846,17 @@ static void DPSOFTRAST_Draw_FreeCommandPool(int space) if (usedcommands <= DPSOFTRAST_DRAW_MAXCOMMANDPOOL-space || waitindex < 0) break; thread = &dpsoftrast.threads[waitindex]; - SDL_LockMutex(thread->drawmutex); + Thread_LockMutex(thread->drawmutex); if (thread->commandoffset != dpsoftrast.drawcommand) { thread->waiting = true; - if (thread->starving) SDL_CondSignal(thread->drawcond); - SDL_CondWait(thread->waitcond, thread->drawmutex); + if (thread->starving) Thread_CondSignal(thread->drawcond); + Thread_CondWait(thread->waitcond, thread->drawmutex); thread->waiting = false; } - SDL_UnlockMutex(thread->drawmutex); + Thread_UnlockMutex(thread->drawmutex); } dpsoftrast.commandpool.usedcommands = usedcommands; -#else - DPSOFTRAST_Draw_FlushThreads(); -#endif } #define DPSOFTRAST_ALIGNCOMMAND(size) \ @@ -822,15 +870,18 @@ static void *DPSOFTRAST_AllocateCommand(int opcode, int size) int freecommand = dpsoftrast.commandpool.freecommand; int usedcommands = dpsoftrast.commandpool.usedcommands; int extra = sizeof(DPSOFTRAST_Command); - if(DPSOFTRAST_DRAW_MAXCOMMANDPOOL - freecommand < size) + if (DPSOFTRAST_DRAW_MAXCOMMANDPOOL - freecommand < size) extra += DPSOFTRAST_DRAW_MAXCOMMANDPOOL - freecommand; - if(usedcommands > DPSOFTRAST_DRAW_MAXCOMMANDPOOL - (size + extra)) + if (usedcommands > DPSOFTRAST_DRAW_MAXCOMMANDPOOL - (size + extra)) { - DPSOFTRAST_Draw_FreeCommandPool(size + extra); + if (dpsoftrast.usethreads) + DPSOFTRAST_Draw_FreeCommandPool(size + extra); + else + DPSOFTRAST_Draw_FlushThreads(); freecommand = dpsoftrast.commandpool.freecommand; usedcommands = dpsoftrast.commandpool.usedcommands; } - if(DPSOFTRAST_DRAW_MAXCOMMANDPOOL - freecommand < size) + if (DPSOFTRAST_DRAW_MAXCOMMANDPOOL - freecommand < size) { command = (DPSOFTRAST_Command *) &dpsoftrast.commandpool.commands[freecommand]; command->opcode = DPSOFTRAST_OPCODE_Reset; @@ -853,6 +904,8 @@ static void DPSOFTRAST_UndoCommand(int size) int freecommand = dpsoftrast.commandpool.freecommand; int usedcommands = dpsoftrast.commandpool.usedcommands; freecommand -= size; + if (freecommand < 0) + freecommand += DPSOFTRAST_DRAW_MAXCOMMANDPOOL; usedcommands -= size; dpsoftrast.commandpool.freecommand = freecommand; dpsoftrast.commandpool.usedcommands = usedcommands; @@ -885,18 +938,22 @@ void DPSOFTRAST_Viewport(int x, int y, int width, int height) DEFCOMMAND(2, ClearColor, float r; float g; float b; float a;) static void DPSOFTRAST_Interpret_ClearColor(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_Command_ClearColor *command) { - int i, x1, y1, x2, y2, w, h, x, y, t1, t2; + int i, x1, y1, x2, y2, w, h, x, y; + int miny1, maxy1, miny2, maxy2; + int bandy; unsigned int *p; unsigned int c; DPSOFTRAST_Validate(thread, DPSOFTRAST_VALIDATE_FB); - x1 = thread->fb_clearscissor[0]; - y1 = thread->fb_clearscissor[1]; - x2 = thread->fb_clearscissor[0] + thread->fb_clearscissor[2]; - y2 = thread->fb_clearscissor[1] + thread->fb_clearscissor[3]; - t1 = (thread->index*dpsoftrast.fb_height)/dpsoftrast.numthreads; - t2 = ((thread->index+1)*dpsoftrast.fb_height)/dpsoftrast.numthreads; - if(y1 < t1) y1 = t1; - if(y2 > t2) y2 = t2; + miny1 = thread->miny1; + maxy1 = thread->maxy1; + miny2 = thread->miny2; + maxy2 = thread->maxy2; + x1 = thread->fb_scissor[0]; + y1 = thread->fb_scissor[1]; + x2 = thread->fb_scissor[0] + thread->fb_scissor[2]; + y2 = thread->fb_scissor[1] + thread->fb_scissor[3]; + if (y1 < miny1) y1 = miny1; + if (y2 > maxy2) y2 = maxy2; w = x2 - x1; h = y2 - y1; if (w < 1 || h < 1) @@ -907,7 +964,8 @@ static void DPSOFTRAST_Interpret_ClearColor(DPSOFTRAST_State_Thread *thread, con { if (!dpsoftrast.fb_colorpixels[i]) continue; - for (y = y1;y < y2;y++) + for (y = y1, bandy = min(y2, maxy1); y < y2; bandy = min(y2, maxy2), y = max(y, miny2)) + for (;y < bandy;y++) { p = dpsoftrast.fb_colorpixels[i] + y * dpsoftrast.fb_width; for (x = x1;x < x2;x++) @@ -927,24 +985,29 @@ void DPSOFTRAST_ClearColor(float r, float g, float b, float a) DEFCOMMAND(3, ClearDepth, float depth;) static void DPSOFTRAST_Interpret_ClearDepth(DPSOFTRAST_State_Thread *thread, DPSOFTRAST_Command_ClearDepth *command) { - int x1, y1, x2, y2, w, h, x, y, t1, t2; + int x1, y1, x2, y2, w, h, x, y; + int miny1, maxy1, miny2, maxy2; + int bandy; unsigned int *p; unsigned int c; DPSOFTRAST_Validate(thread, DPSOFTRAST_VALIDATE_FB); - x1 = thread->fb_clearscissor[0]; - y1 = thread->fb_clearscissor[1]; - x2 = thread->fb_clearscissor[0] + thread->fb_clearscissor[2]; - y2 = thread->fb_clearscissor[1] + thread->fb_clearscissor[3]; - t1 = (thread->index*dpsoftrast.fb_height)/dpsoftrast.numthreads; - t2 = ((thread->index+1)*dpsoftrast.fb_height)/dpsoftrast.numthreads; - if(y1 < t1) y1 = t1; - if(y2 > t2) y2 = t2; + miny1 = thread->miny1; + maxy1 = thread->maxy1; + miny2 = thread->miny2; + maxy2 = thread->maxy2; + x1 = thread->fb_scissor[0]; + y1 = thread->fb_scissor[1]; + x2 = thread->fb_scissor[0] + thread->fb_scissor[2]; + y2 = thread->fb_scissor[1] + thread->fb_scissor[3]; + if (y1 < miny1) y1 = miny1; + if (y2 > maxy2) y2 = maxy2; w = x2 - x1; h = y2 - y1; if (w < 1 || h < 1) return; c = DPSOFTRAST_DEPTH32_FROM_DEPTH32F(command->depth); - for (y = y1;y < y2;y++) + for (y = y1, bandy = min(y2, maxy1); y < y2; bandy = min(y2, maxy2), y = max(y, miny2)) + for (;y < bandy;y++) { p = dpsoftrast.fb_depthpixels + y * dpsoftrast.fb_width; for (x = x1;x < x2;x++) @@ -1102,30 +1165,6 @@ void DPSOFTRAST_CullFace(int mode) command->mode = mode; } -DEFCOMMAND(15, AlphaTest, int enable;) -static void DPSOFTRAST_Interpret_AlphaTest(DPSOFTRAST_State_Thread *thread, DPSOFTRAST_Command_AlphaTest *command) -{ - thread->alphatest = command->enable; -} -void DPSOFTRAST_AlphaTest(int enable) -{ - DPSOFTRAST_Command_AlphaTest *command = DPSOFTRAST_ALLOCATECOMMAND(AlphaTest); - command->enable = enable; -} - -DEFCOMMAND(16, AlphaFunc, int func; float ref;) -static void DPSOFTRAST_Interpret_AlphaFunc(DPSOFTRAST_State_Thread *thread, DPSOFTRAST_Command_AlphaFunc *command) -{ - thread->alphafunc = command->func; - thread->alphavalue = command->ref; -} -void DPSOFTRAST_AlphaFunc(int func, float ref) -{ - DPSOFTRAST_Command_AlphaFunc *command = DPSOFTRAST_ALLOCATECOMMAND(AlphaFunc); - command->func = func; - command->ref = ref; -} - void DPSOFTRAST_Color4f(float r, float g, float b, float a) { dpsoftrast.color[0] = r; @@ -1143,7 +1182,6 @@ void DPSOFTRAST_GetPixelsBGRA(int blockx, int blocky, int blockwidth, int blockh int bx2 = blockx + blockwidth; int by2 = blocky + blockheight; int bw; - int bh; int x; int y; unsigned char *inpixels; @@ -1155,7 +1193,6 @@ void DPSOFTRAST_GetPixelsBGRA(int blockx, int blocky, int blockwidth, int blockh if (bx2 > dpsoftrast.fb_width) bx2 = dpsoftrast.fb_width; if (by2 > dpsoftrast.fb_height) by2 = dpsoftrast.fb_height; bw = bx2 - bx1; - bh = by2 - by1; inpixels = (unsigned char *)dpsoftrast.fb_colorpixels[0]; if (dpsoftrast.bigendian) { @@ -1209,8 +1246,7 @@ void DPSOFTRAST_CopyRectangleToTexture(int index, int mip, int tx, int ty, int s DPSOFTRAST_Texture *texture; texture = DPSOFTRAST_Texture_GetByIndex(index);if (!texture) return; if (mip < 0 || mip >= texture->mipmaps) return; - if (texture->binds) - DPSOFTRAST_Flush(); + DPSOFTRAST_Flush(); spixels = dpsoftrast.fb_colorpixels[0]; swidth = dpsoftrast.fb_width; sheight = dpsoftrast.fb_height; @@ -1233,8 +1269,9 @@ void DPSOFTRAST_CopyRectangleToTexture(int index, int mip, int tx, int ty, int s if (th > sh) th = sh; if (tw < 1 || th < 1) return; + sy1 = sheight - 1 - sy1; for (y = 0;y < th;y++) - memcpy(tpixels + ((ty1 + y) * twidth + tx1), spixels + ((sy1 + y) * swidth + sx1), tw*4); + memcpy(tpixels + ((ty1 + y) * twidth + tx1), spixels + ((sy1 - y) * swidth + sx1), tw*4); if (texture->mipmaps > 1) DPSOFTRAST_Texture_CalculateMipmaps(index); } @@ -1294,20 +1331,23 @@ void DPSOFTRAST_SetTexCoordPointer(int unitnum, int numcomponents, size_t stride dpsoftrast.stride_texcoord[unitnum] = stride; } -DEFCOMMAND(18, SetShader, int mode; int permutation;) +DEFCOMMAND(18, SetShader, int mode; int permutation; int exactspecularmath;) static void DPSOFTRAST_Interpret_SetShader(DPSOFTRAST_State_Thread *thread, DPSOFTRAST_Command_SetShader *command) { thread->shader_mode = command->mode; thread->shader_permutation = command->permutation; + thread->shader_exactspecularmath = command->exactspecularmath; } -void DPSOFTRAST_SetShader(int mode, int permutation) +void DPSOFTRAST_SetShader(int mode, int permutation, int exactspecularmath) { DPSOFTRAST_Command_SetShader *command = DPSOFTRAST_ALLOCATECOMMAND(SetShader); command->mode = mode; command->permutation = permutation; + command->exactspecularmath = exactspecularmath; dpsoftrast.shader_mode = mode; dpsoftrast.shader_permutation = permutation; + dpsoftrast.shader_exactspecularmath = exactspecularmath; } DEFCOMMAND(19, Uniform4f, DPSOFTRAST_UNIFORM index; float val[4];) @@ -1345,13 +1385,13 @@ static void DPSOFTRAST_Interpret_UniformMatrix4f(DPSOFTRAST_State_Thread *thread } void DPSOFTRAST_UniformMatrix4fv(DPSOFTRAST_UNIFORM uniform, int arraysize, int transpose, const float *v) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE int i, index; for (i = 0, index = (int)uniform;i < arraysize;i++, index += 4, v += 16) { __m128 m0, m1, m2, m3; DPSOFTRAST_Command_UniformMatrix4f *command = DPSOFTRAST_ALLOCATECOMMAND(UniformMatrix4f); - command->index = index; + command->index = (DPSOFTRAST_UNIFORM)index; if (((size_t)v)&(ALIGN_SIZE-1)) { m0 = _mm_loadu_ps(v); @@ -1404,7 +1444,22 @@ void DPSOFTRAST_Uniform1i(DPSOFTRAST_UNIFORM index, int i0) dpsoftrast.uniform1i[command->index] = i0; } -#ifdef SSE2_PRESENT +DEFCOMMAND(24, ClipPlane, float clipplane[4];) +static void DPSOFTRAST_Interpret_ClipPlane(DPSOFTRAST_State_Thread *thread, DPSOFTRAST_Command_ClipPlane *command) +{ + memcpy(thread->clipplane, command->clipplane, 4*sizeof(float)); + thread->validate |= DPSOFTRAST_VALIDATE_FB; +} +void DPSOFTRAST_ClipPlane(float x, float y, float z, float w) +{ + DPSOFTRAST_Command_ClipPlane *command = DPSOFTRAST_ALLOCATECOMMAND(ClipPlane); + command->clipplane[0] = x; + command->clipplane[1] = y; + command->clipplane[2] = z; + command->clipplane[3] = w; +} + +#ifdef SSE_POSSIBLE static void DPSOFTRAST_Load4fTo4f(float *dst, const unsigned char *src, int size, int stride) { float *end = dst + size*4; @@ -1600,7 +1655,7 @@ static void DPSOFTRAST_Fill4f(float *dst, const float *src, int size) void DPSOFTRAST_Vertex_Transform(float *out4f, const float *in4f, int numitems, const float *inmatrix16f) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE static const float identitymatrix[4][4] = {{1,0,0,0},{0,1,0,0},{0,0,1,0},{0,0,0,1}}; __m128 m0, m1, m2, m3; float *end; @@ -1651,7 +1706,7 @@ void DPSOFTRAST_Vertex_Copy(float *out4f, const float *in4f, int numitems) memcpy(out4f, in4f, numitems * sizeof(float[4])); } -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE #define DPSOFTRAST_PROJECTVERTEX(out, in, viewportcenter, viewportscale) \ { \ __m128 p = (in), w = _mm_shuffle_ps(p, p, _MM_SHUFFLE(3, 3, 3, 3)); \ @@ -1677,10 +1732,13 @@ void DPSOFTRAST_Vertex_Copy(float *out4f, const float *in4f, int numitems) _mm_mul_ps(_mm_shuffle_ps(p, p, _MM_SHUFFLE(3, 3, 3, 3)), m3)))); \ } -static int DPSOFTRAST_Vertex_BoundY(int *starty, int *endy, __m128 minpos, __m128 maxpos, __m128 viewportcenter, __m128 viewportscale, __m128 m0, __m128 m1, __m128 m2, __m128 m3) +static int DPSOFTRAST_Vertex_BoundY(int *starty, int *endy, const float *minposf, const float *maxposf, const float *inmatrix16f) { int clipmask = 0xFF; + __m128 viewportcenter = _mm_load_ps(dpsoftrast.fb_viewportcenter), viewportscale = _mm_load_ps(dpsoftrast.fb_viewportscale); __m128 bb[8], clipdist[8], minproj = _mm_set_ss(2.0f), maxproj = _mm_set_ss(-2.0f); + __m128 m0 = _mm_loadu_ps(inmatrix16f), m1 = _mm_loadu_ps(inmatrix16f + 4), m2 = _mm_loadu_ps(inmatrix16f + 8), m3 = _mm_loadu_ps(inmatrix16f + 12); + __m128 minpos = _mm_load_ps(minposf), maxpos = _mm_load_ps(maxposf); m0 = _mm_shuffle_ps(m0, m0, _MM_SHUFFLE(3, 2, 0, 1)); m1 = _mm_shuffle_ps(m1, m1, _MM_SHUFFLE(3, 2, 0, 1)); m2 = _mm_shuffle_ps(m2, m2, _MM_SHUFFLE(3, 2, 0, 1)); @@ -1747,11 +1805,10 @@ static int DPSOFTRAST_Vertex_BoundY(int *starty, int *endy, __m128 minpos, __m12 *endy = _mm_cvttss_si32(minproj)+1; return clipmask; } -#endif static int DPSOFTRAST_Vertex_Project(float *out4f, float *screen4f, int *starty, int *endy, const float *in4f, int numitems) { -#ifdef SSE2_PRESENT + static const float identitymatrix[16] = {1,0,0,0, 0,1,0,0, 0,0,1,0, 0,0,0,1}; float *end = out4f + numitems*4; __m128 viewportcenter = _mm_load_ps(dpsoftrast.fb_viewportcenter), viewportscale = _mm_load_ps(dpsoftrast.fb_viewportscale); __m128 minpos, maxpos; @@ -1788,19 +1845,19 @@ static int DPSOFTRAST_Vertex_Project(float *out4f, float *screen4f, int *starty, } } if (starty && endy) - return DPSOFTRAST_Vertex_BoundY(starty, endy, minpos, maxpos, viewportcenter, viewportscale, - _mm_setr_ps(1.0f, 0.0f, 0.0f, 0.0f), - _mm_setr_ps(0.0f, 1.0f, 0.0f, 0.0f), - _mm_setr_ps(0.0f, 0.0f, 1.0f, 0.0f), - _mm_setr_ps(0.0f, 0.0f, 0.0f, 1.0f)); + { + ALIGN(float minposf[4]); + ALIGN(float maxposf[4]); + _mm_store_ps(minposf, minpos); + _mm_store_ps(maxposf, maxpos); + return DPSOFTRAST_Vertex_BoundY(starty, endy, minposf, maxposf, identitymatrix); + } return 0; -#endif } static int DPSOFTRAST_Vertex_TransformProject(float *out4f, float *screen4f, int *starty, int *endy, const float *in4f, int numitems, const float *inmatrix16f) { -#ifdef SSE2_PRESENT - static const float identitymatrix[4][4] = {{1,0,0,0},{0,1,0,0},{0,0,1,0},{0,0,0,1}}; + static const float identitymatrix[16] = {1,0,0,0, 0,1,0,0, 0,0,1,0, 0,0,0,1}; __m128 m0, m1, m2, m3, viewportcenter, viewportscale, minpos, maxpos; float *end; if (!memcmp(identitymatrix, inmatrix16f, sizeof(float[16]))) @@ -1847,13 +1904,20 @@ static int DPSOFTRAST_Vertex_TransformProject(float *out4f, float *screen4f, int } } if (starty && endy) - return DPSOFTRAST_Vertex_BoundY(starty, endy, minpos, maxpos, viewportcenter, viewportscale, m0, m1, m2, m3); + { + ALIGN(float minposf[4]); + ALIGN(float maxposf[4]); + _mm_store_ps(minposf, minpos); + _mm_store_ps(maxposf, maxpos); + return DPSOFTRAST_Vertex_BoundY(starty, endy, minposf, maxposf, inmatrix16f); + } return 0; -#endif } +#endif static float *DPSOFTRAST_Array_Load(int outarray, int inarray) { +#ifdef SSE_POSSIBLE float *outf = dpsoftrast.post_array4f[outarray]; const unsigned char *inb; int firstvertex = dpsoftrast.firstvertex; @@ -1905,6 +1969,9 @@ static float *DPSOFTRAST_Array_Load(int outarray, int inarray) break; } return outf; +#else + return NULL; +#endif } static float *DPSOFTRAST_Array_Transform(int outarray, int inarray, const float *inmatrix16f) @@ -1917,17 +1984,25 @@ static float *DPSOFTRAST_Array_Transform(int outarray, int inarray, const float #if 0 static float *DPSOFTRAST_Array_Project(int outarray, int inarray) { +#ifdef SSE_POSSIBLE float *data = inarray >= 0 ? DPSOFTRAST_Array_Load(outarray, inarray) : dpsoftrast.post_array4f[outarray]; dpsoftrast.drawclipped = DPSOFTRAST_Vertex_Project(data, dpsoftrast.screencoord4f, &dpsoftrast.drawstarty, &dpsoftrast.drawendy, data, dpsoftrast.numvertices); return data; +#else + return NULL; +#endif } #endif static float *DPSOFTRAST_Array_TransformProject(int outarray, int inarray, const float *inmatrix16f) { +#ifdef SSE_POSSIBLE float *data = inarray >= 0 ? DPSOFTRAST_Array_Load(outarray, inarray) : dpsoftrast.post_array4f[outarray]; dpsoftrast.drawclipped = DPSOFTRAST_Vertex_TransformProject(data, dpsoftrast.screencoord4f, &dpsoftrast.drawstarty, &dpsoftrast.drawendy, data, dpsoftrast.numvertices, inmatrix16f); return data; +#else + return NULL; +#endif } void DPSOFTRAST_Draw_Span_Begin(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, float *zf) @@ -1938,11 +2013,18 @@ void DPSOFTRAST_Draw_Span_Begin(DPSOFTRAST_State_Thread *thread, const DPSOFTRAS float wslope = triangle->w[0]; float w = triangle->w[2] + span->x*wslope + span->y*triangle->w[1]; float endz = 1.0f / (w + wslope * startx); + if (triangle->w[0] == 0) + { + // LordHavoc: fast flat polygons (HUD/menu) + for (x = startx;x < endx;x++) + zf[x] = endz; + return; + } for (x = startx;x < endx;) { int nextsub = x + DPSOFTRAST_DRAW_MAXSUBSPAN, endsub = nextsub - 1; float z = endz, dz; - if(nextsub >= endx) nextsub = endsub = endx-1; + if (nextsub >= endx) nextsub = endsub = endx-1; endz = 1.0f / (w + wslope * nextsub); dz = x < nextsub ? (endz - z) / (nextsub - x) : 0.0f; for (; x <= endsub; x++, z += dz) @@ -1950,303 +2032,316 @@ void DPSOFTRAST_Draw_Span_Begin(DPSOFTRAST_State_Thread *thread, const DPSOFTRAS } } -void DPSOFTRAST_Draw_Span_Finish(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, const float * RESTRICT in4f) +void DPSOFTRAST_Draw_Span_FinishBGRA8(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, const unsigned char* RESTRICT in4ub) { +#ifdef SSE_POSSIBLE int x; int startx = span->startx; int endx = span->endx; - int d[4]; - float a, b; + int maskx; + int subx; + const unsigned int * RESTRICT ini = (const unsigned int *)in4ub; unsigned char * RESTRICT pixelmask = span->pixelmask; unsigned char * RESTRICT pixel = (unsigned char *)dpsoftrast.fb_colorpixels[0]; + unsigned int * RESTRICT pixeli = (unsigned int *)dpsoftrast.fb_colorpixels[0]; if (!pixel) return; pixel += (span->y * dpsoftrast.fb_width + span->x) * 4; + pixeli += span->y * dpsoftrast.fb_width + span->x; // handle alphatest now (this affects depth writes too) - if (thread->alphatest) + if (thread->shader_permutation & SHADERPERMUTATION_ALPHAKILL) for (x = startx;x < endx;x++) - if (in4f[x*4+3] < 0.5f) + if (in4ub[x*4+3] < 128) pixelmask[x] = false; - // FIXME: this does not handle bigendian + // LordHavoc: clear pixelmask for some pixels in alphablend cases, this + // helps sprites, text and hud artwork switch(thread->fb_blendmode) { - case DPSOFTRAST_BLENDMODE_OPAQUE: - for (x = startx;x < endx;x++) - { - if (!pixelmask[x]) - continue; - d[0] = (int)(in4f[x*4+2]*255.0f);if (d[0] > 255) d[0] = 255; - d[1] = (int)(in4f[x*4+1]*255.0f);if (d[1] > 255) d[1] = 255; - d[2] = (int)(in4f[x*4+0]*255.0f);if (d[2] > 255) d[2] = 255; - d[3] = (int)(in4f[x*4+3]*255.0f);if (d[3] > 255) d[3] = 255; - pixel[x*4+0] = d[0]; - pixel[x*4+1] = d[1]; - pixel[x*4+2] = d[2]; - pixel[x*4+3] = d[3]; - } - break; case DPSOFTRAST_BLENDMODE_ALPHA: - for (x = startx;x < endx;x++) - { - if (!pixelmask[x]) - continue; - a = in4f[x*4+3] * 255.0f; - b = 1.0f - in4f[x*4+3]; - d[0] = (int)(in4f[x*4+2]*a+pixel[x*4+0]*b);if (d[0] > 255) d[0] = 255; - d[1] = (int)(in4f[x*4+1]*a+pixel[x*4+1]*b);if (d[1] > 255) d[1] = 255; - d[2] = (int)(in4f[x*4+0]*a+pixel[x*4+2]*b);if (d[2] > 255) d[2] = 255; - d[3] = (int)(in4f[x*4+3]*a+pixel[x*4+3]*b);if (d[3] > 255) d[3] = 255; - pixel[x*4+0] = d[0]; - pixel[x*4+1] = d[1]; - pixel[x*4+2] = d[2]; - pixel[x*4+3] = d[3]; - } - break; case DPSOFTRAST_BLENDMODE_ADDALPHA: + case DPSOFTRAST_BLENDMODE_SUBALPHA: + maskx = startx; for (x = startx;x < endx;x++) { - if (!pixelmask[x]) - continue; - a = in4f[x*4+3] * 255.0f; - d[0] = (int)(in4f[x*4+2]*a+pixel[x*4+0]);if (d[0] > 255) d[0] = 255; - d[1] = (int)(in4f[x*4+1]*a+pixel[x*4+1]);if (d[1] > 255) d[1] = 255; - d[2] = (int)(in4f[x*4+0]*a+pixel[x*4+2]);if (d[2] > 255) d[2] = 255; - d[3] = (int)(in4f[x*4+3]*a+pixel[x*4+3]);if (d[3] > 255) d[3] = 255; - pixel[x*4+0] = d[0]; - pixel[x*4+1] = d[1]; - pixel[x*4+2] = d[2]; - pixel[x*4+3] = d[3]; + if (in4ub[x*4+3] >= 1) + { + startx = x; + for (;;) + { + while (++x < endx && in4ub[x*4+3] >= 1) ; + maskx = x; + if (x >= endx) break; + ++x; + while (++x < endx && in4ub[x*4+3] < 1) pixelmask[x] = false; + if (x >= endx) break; + } + break; + } } + endx = maskx; break; + case DPSOFTRAST_BLENDMODE_OPAQUE: case DPSOFTRAST_BLENDMODE_ADD: - for (x = startx;x < endx;x++) - { - if (!pixelmask[x]) - continue; - d[0] = (int)(in4f[x*4+2]*255.0f+pixel[x*4+0]);if (d[0] > 255) d[0] = 255; - d[1] = (int)(in4f[x*4+1]*255.0f+pixel[x*4+1]);if (d[1] > 255) d[1] = 255; - d[2] = (int)(in4f[x*4+0]*255.0f+pixel[x*4+2]);if (d[2] > 255) d[2] = 255; - d[3] = (int)(in4f[x*4+3]*255.0f+pixel[x*4+3]);if (d[3] > 255) d[3] = 255; - pixel[x*4+0] = d[0]; - pixel[x*4+1] = d[1]; - pixel[x*4+2] = d[2]; - pixel[x*4+3] = d[3]; - } - break; case DPSOFTRAST_BLENDMODE_INVMOD: - for (x = startx;x < endx;x++) - { - if (!pixelmask[x]) - continue; - d[0] = (int)((1.0f-in4f[x*4+2])*pixel[x*4+0]);if (d[0] > 255) d[0] = 255; - d[1] = (int)((1.0f-in4f[x*4+1])*pixel[x*4+1]);if (d[1] > 255) d[1] = 255; - d[2] = (int)((1.0f-in4f[x*4+0])*pixel[x*4+2]);if (d[2] > 255) d[2] = 255; - d[3] = (int)((1.0f-in4f[x*4+3])*pixel[x*4+3]);if (d[3] > 255) d[3] = 255; - pixel[x*4+0] = d[0]; - pixel[x*4+1] = d[1]; - pixel[x*4+2] = d[2]; - pixel[x*4+3] = d[3]; - } - break; case DPSOFTRAST_BLENDMODE_MUL: - for (x = startx;x < endx;x++) - { - if (!pixelmask[x]) - continue; - d[0] = (int)(in4f[x*4+2]*pixel[x*4+0]);if (d[0] > 255) d[0] = 255; - d[1] = (int)(in4f[x*4+1]*pixel[x*4+1]);if (d[1] > 255) d[1] = 255; - d[2] = (int)(in4f[x*4+0]*pixel[x*4+2]);if (d[2] > 255) d[2] = 255; - d[3] = (int)(in4f[x*4+3]*pixel[x*4+3]);if (d[3] > 255) d[3] = 255; - pixel[x*4+0] = d[0]; - pixel[x*4+1] = d[1]; - pixel[x*4+2] = d[2]; - pixel[x*4+3] = d[3]; - } - break; case DPSOFTRAST_BLENDMODE_MUL2: - for (x = startx;x < endx;x++) - { - if (!pixelmask[x]) - continue; - d[0] = (int)(in4f[x*4+2]*pixel[x*4+0]*2.0f);if (d[0] > 255) d[0] = 255; - d[1] = (int)(in4f[x*4+1]*pixel[x*4+1]*2.0f);if (d[1] > 255) d[1] = 255; - d[2] = (int)(in4f[x*4+0]*pixel[x*4+2]*2.0f);if (d[2] > 255) d[2] = 255; - d[3] = (int)(in4f[x*4+3]*pixel[x*4+3]*2.0f);if (d[3] > 255) d[3] = 255; - pixel[x*4+0] = d[0]; - pixel[x*4+1] = d[1]; - pixel[x*4+2] = d[2]; - pixel[x*4+3] = d[3]; - } + case DPSOFTRAST_BLENDMODE_PSEUDOALPHA: + case DPSOFTRAST_BLENDMODE_INVADD: break; - case DPSOFTRAST_BLENDMODE_SUBALPHA: - for (x = startx;x < endx;x++) + } + // put some special values at the end of the mask to ensure the loops end + pixelmask[endx] = 1; + pixelmask[endx+1] = 0; + // LordHavoc: use a double loop to identify subspans, this helps the + // optimized copy/blend loops to perform at their best, most triangles + // have only one run of pixels, and do the search using wide reads... + x = startx; + while (x < endx) + { + // if this pixel is masked off, it's probably not alone... + if (!pixelmask[x]) { - if (!pixelmask[x]) - continue; - a = in4f[x*4+3] * -255.0f; - d[0] = (int)(in4f[x*4+2]*a+pixel[x*4+0]);if (d[0] > 255) d[0] = 255;if (d[0] < 0) d[0] = 0; - d[1] = (int)(in4f[x*4+1]*a+pixel[x*4+1]);if (d[1] > 255) d[1] = 255;if (d[1] < 0) d[1] = 0; - d[2] = (int)(in4f[x*4+0]*a+pixel[x*4+2]);if (d[2] > 255) d[2] = 255;if (d[2] < 0) d[2] = 0; - d[3] = (int)(in4f[x*4+3]*a+pixel[x*4+3]);if (d[3] > 255) d[3] = 255;if (d[3] < 0) d[3] = 0; - pixel[x*4+0] = d[0]; - pixel[x*4+1] = d[1]; - pixel[x*4+2] = d[2]; - pixel[x*4+3] = d[3]; + x++; +#if 1 + if (x + 8 < endx) + { + // the 4-item search must be aligned or else it stalls badly + if ((x & 3) && !pixelmask[x]) + { + if(pixelmask[x]) goto endmasked; + x++; + if (x & 3) + { + if(pixelmask[x]) goto endmasked; + x++; + if (x & 3) + { + if(pixelmask[x]) goto endmasked; + x++; + } + } + } + while (*(unsigned int *)&pixelmask[x] == 0x00000000) + x += 4; + } +#endif + for (;!pixelmask[x];x++) + ; + // rather than continue the loop, just check the end variable + if (x >= endx) + break; } - break; - case DPSOFTRAST_BLENDMODE_PSEUDOALPHA: - for (x = startx;x < endx;x++) + endmasked: + // find length of subspan + subx = x + 1; +#if 1 + if (subx + 8 < endx) { - if (!pixelmask[x]) - continue; - a = 255.0f; - b = 1.0f - in4f[x*4+3]; - d[0] = (int)(in4f[x*4+2]*a+pixel[x*4+0]*b);if (d[0] > 255) d[0] = 255; - d[1] = (int)(in4f[x*4+1]*a+pixel[x*4+1]*b);if (d[1] > 255) d[1] = 255; - d[2] = (int)(in4f[x*4+0]*a+pixel[x*4+2]*b);if (d[2] > 255) d[2] = 255; - d[3] = (int)(in4f[x*4+3]*a+pixel[x*4+3]*b);if (d[3] > 255) d[3] = 255; - pixel[x*4+0] = d[0]; - pixel[x*4+1] = d[1]; - pixel[x*4+2] = d[2]; - pixel[x*4+3] = d[3]; + if (subx & 3) + { + if(!pixelmask[subx]) goto endunmasked; + subx++; + if (subx & 3) + { + if(!pixelmask[subx]) goto endunmasked; + subx++; + if (subx & 3) + { + if(!pixelmask[subx]) goto endunmasked; + subx++; + } + } + } + while (*(unsigned int *)&pixelmask[subx] == 0x01010101) + subx += 4; } - break; - } -} - -void DPSOFTRAST_Draw_Span_FinishBGRA8(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, const unsigned char* RESTRICT in4ub) -{ -#ifdef SSE2_PRESENT - int x; - int startx = span->startx; - int endx = span->endx; - const unsigned int * RESTRICT ini = (const unsigned int *)in4ub; - unsigned char * RESTRICT pixelmask = span->pixelmask; - unsigned char * RESTRICT pixel = (unsigned char *)dpsoftrast.fb_colorpixels[0]; - unsigned int * RESTRICT pixeli = (unsigned int *)dpsoftrast.fb_colorpixels[0]; - if (!pixel) - return; - pixel += (span->y * dpsoftrast.fb_width + span->x) * 4; - pixeli += span->y * dpsoftrast.fb_width + span->x; - // handle alphatest now (this affects depth writes too) - if (thread->alphatest) - for (x = startx;x < endx;x++) - if (in4ub[x*4+3] < 0.5f) - pixelmask[x] = false; - // FIXME: this does not handle bigendian - switch(thread->fb_blendmode) - { - case DPSOFTRAST_BLENDMODE_OPAQUE: - for (x = startx;x + 4 <= endx;) +#endif + for (;pixelmask[subx];subx++) + ; + // the checks can overshoot, so make sure to clip it... + if (subx > endx) + subx = endx; + endunmasked: + // now that we know the subspan length... process! + switch(thread->fb_blendmode) { - if (*(const unsigned int *)&pixelmask[x] == 0x01010101) + case DPSOFTRAST_BLENDMODE_OPAQUE: +#if 0 + if (subx - x >= 16) { - _mm_storeu_si128((__m128i *)&pixeli[x], _mm_loadu_si128((const __m128i *)&ini[x])); - x += 4; + memcpy(pixeli + x, ini + x, (subx - x) * sizeof(pixeli[x])); + x = subx; } else +#elif 1 + while (x + 16 <= subx) + { + _mm_storeu_si128((__m128i *)&pixeli[x], _mm_loadu_si128((const __m128i *)&ini[x])); + _mm_storeu_si128((__m128i *)&pixeli[x+4], _mm_loadu_si128((const __m128i *)&ini[x+4])); + _mm_storeu_si128((__m128i *)&pixeli[x+8], _mm_loadu_si128((const __m128i *)&ini[x+8])); + _mm_storeu_si128((__m128i *)&pixeli[x+12], _mm_loadu_si128((const __m128i *)&ini[x+12])); + x += 16; + } +#endif { - if (pixelmask[x]) + while (x + 4 <= subx) + { + _mm_storeu_si128((__m128i *)&pixeli[x], _mm_loadu_si128((const __m128i *)&ini[x])); + x += 4; + } + if (x + 2 <= subx) + { + pixeli[x] = ini[x]; + pixeli[x+1] = ini[x+1]; + x += 2; + } + if (x < subx) + { pixeli[x] = ini[x]; - x++; + x++; + } } - } - for (;x < endx;x++) - if (pixelmask[x]) - pixeli[x] = ini[x]; - break; - case DPSOFTRAST_BLENDMODE_ALPHA: - #define FINISHBLEND(blend2, blend1) \ - for (x = startx;x + 2 <= endx;x += 2) \ - { \ - __m128i src, dst; \ - switch (*(const unsigned short*)&pixelmask[x]) \ + break; + case DPSOFTRAST_BLENDMODE_ALPHA: + #define FINISHBLEND(blend2, blend1) \ + for (;x + 1 < subx;x += 2) \ { \ - case 0x0101: \ + __m128i src, dst; \ src = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)&ini[x]), _mm_setzero_si128()); \ dst = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)&pixeli[x]), _mm_setzero_si128()); \ blend2; \ _mm_storel_epi64((__m128i *)&pixeli[x], _mm_packus_epi16(dst, dst)); \ - continue; \ - case 0x0100: \ - src = _mm_unpacklo_epi8(_mm_cvtsi32_si128(ini[x+1]), _mm_setzero_si128()); \ - dst = _mm_unpacklo_epi8(_mm_cvtsi32_si128(pixeli[x+1]), _mm_setzero_si128()); \ - blend1; \ - pixeli[x+1] = _mm_cvtsi128_si32(_mm_packus_epi16(dst, dst)); \ - continue; \ - case 0x0001: \ + } \ + if (x < subx) \ + { \ + __m128i src, dst; \ src = _mm_unpacklo_epi8(_mm_cvtsi32_si128(ini[x]), _mm_setzero_si128()); \ dst = _mm_unpacklo_epi8(_mm_cvtsi32_si128(pixeli[x]), _mm_setzero_si128()); \ blend1; \ pixeli[x] = _mm_cvtsi128_si32(_mm_packus_epi16(dst, dst)); \ - continue; \ - } \ - break; \ - } \ - for(;x < endx; x++) \ - { \ - __m128i src, dst; \ - if (!pixelmask[x]) \ - continue; \ - src = _mm_unpacklo_epi8(_mm_cvtsi32_si128(ini[x]), _mm_setzero_si128()); \ - dst = _mm_unpacklo_epi8(_mm_cvtsi32_si128(pixeli[x]), _mm_setzero_si128()); \ - blend1; \ - pixeli[x] = _mm_cvtsi128_si32(_mm_packus_epi16(dst, dst)); \ + x++; \ + } + FINISHBLEND({ + __m128i blend = _mm_shufflehi_epi16(_mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)), _MM_SHUFFLE(3, 3, 3, 3)); + dst = _mm_add_epi16(dst, _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(src, dst), 4), _mm_slli_epi16(blend, 4))); + }, { + __m128i blend = _mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)); + dst = _mm_add_epi16(dst, _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(src, dst), 4), _mm_slli_epi16(blend, 4))); + }); + break; + case DPSOFTRAST_BLENDMODE_ADDALPHA: + FINISHBLEND({ + __m128i blend = _mm_shufflehi_epi16(_mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)), _MM_SHUFFLE(3, 3, 3, 3)); + dst = _mm_add_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(src, blend), 8)); + }, { + __m128i blend = _mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)); + dst = _mm_add_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(src, blend), 8)); + }); + break; + case DPSOFTRAST_BLENDMODE_ADD: + FINISHBLEND({ dst = _mm_add_epi16(src, dst); }, { dst = _mm_add_epi16(src, dst); }); + break; + case DPSOFTRAST_BLENDMODE_INVMOD: + FINISHBLEND({ + dst = _mm_sub_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(dst, src), 8)); + }, { + dst = _mm_sub_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(dst, src), 8)); + }); + break; + case DPSOFTRAST_BLENDMODE_MUL: + FINISHBLEND({ dst = _mm_srli_epi16(_mm_mullo_epi16(src, dst), 8); }, { dst = _mm_srli_epi16(_mm_mullo_epi16(src, dst), 8); }); + break; + case DPSOFTRAST_BLENDMODE_MUL2: + FINISHBLEND({ dst = _mm_srli_epi16(_mm_mullo_epi16(src, dst), 7); }, { dst = _mm_srli_epi16(_mm_mullo_epi16(src, dst), 7); }); + break; + case DPSOFTRAST_BLENDMODE_SUBALPHA: + FINISHBLEND({ + __m128i blend = _mm_shufflehi_epi16(_mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)), _MM_SHUFFLE(3, 3, 3, 3)); + dst = _mm_sub_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(src, blend), 8)); + }, { + __m128i blend = _mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)); + dst = _mm_sub_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(src, blend), 8)); + }); + break; + case DPSOFTRAST_BLENDMODE_PSEUDOALPHA: + FINISHBLEND({ + __m128i blend = _mm_shufflehi_epi16(_mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)), _MM_SHUFFLE(3, 3, 3, 3)); + dst = _mm_add_epi16(src, _mm_sub_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(dst, blend), 8))); + }, { + __m128i blend = _mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)); + dst = _mm_add_epi16(src, _mm_sub_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(dst, blend), 8))); + }); + break; + case DPSOFTRAST_BLENDMODE_INVADD: + FINISHBLEND({ + dst = _mm_add_epi16(dst, _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(_mm_set1_epi16(255), dst), 4), _mm_slli_epi16(src, 4))); + }, { + dst = _mm_add_epi16(dst, _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(_mm_set1_epi16(255), dst), 4), _mm_slli_epi16(src, 4))); + }); + break; } + } +#endif +} - FINISHBLEND({ - __m128i blend = _mm_shufflehi_epi16(_mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)), _MM_SHUFFLE(3, 3, 3, 3)); - dst = _mm_add_epi16(dst, _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(src, dst), 4), _mm_slli_epi16(blend, 4))); - }, { - __m128i blend = _mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)); - dst = _mm_add_epi16(dst, _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(src, dst), 4), _mm_slli_epi16(blend, 4))); - }); - break; - case DPSOFTRAST_BLENDMODE_ADDALPHA: - FINISHBLEND({ - __m128i blend = _mm_shufflehi_epi16(_mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)), _MM_SHUFFLE(3, 3, 3, 3)); - dst = _mm_add_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(src, blend), 8)); - }, { - __m128i blend = _mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)); - dst = _mm_add_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(src, blend), 8)); - }); - break; - case DPSOFTRAST_BLENDMODE_ADD: - FINISHBLEND({ dst = _mm_add_epi16(src, dst); }, { dst = _mm_add_epi16(src, dst); }); - break; - case DPSOFTRAST_BLENDMODE_INVMOD: - FINISHBLEND({ - dst = _mm_sub_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(dst, src), 8)); - }, { - dst = _mm_sub_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(dst, src), 8)); - }); - break; - case DPSOFTRAST_BLENDMODE_MUL: - FINISHBLEND({ dst = _mm_srli_epi16(_mm_mullo_epi16(src, dst), 8); }, { dst = _mm_srli_epi16(_mm_mullo_epi16(src, dst), 8); }); - break; - case DPSOFTRAST_BLENDMODE_MUL2: - FINISHBLEND({ dst = _mm_srli_epi16(_mm_mullo_epi16(src, dst), 7); }, { dst = _mm_srli_epi16(_mm_mullo_epi16(src, dst), 7); }); - break; - case DPSOFTRAST_BLENDMODE_SUBALPHA: - FINISHBLEND({ - __m128i blend = _mm_shufflehi_epi16(_mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)), _MM_SHUFFLE(3, 3, 3, 3)); - dst = _mm_sub_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(src, blend), 8)); - }, { - __m128i blend = _mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)); - dst = _mm_sub_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(src, blend), 8)); - }); - break; - case DPSOFTRAST_BLENDMODE_PSEUDOALPHA: - FINISHBLEND({ - __m128i blend = _mm_shufflehi_epi16(_mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)), _MM_SHUFFLE(3, 3, 3, 3)); - dst = _mm_add_epi16(src, _mm_sub_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(dst, blend), 8))); - }, { - __m128i blend = _mm_shufflelo_epi16(src, _MM_SHUFFLE(3, 3, 3, 3)); - dst = _mm_add_epi16(src, _mm_sub_epi16(dst, _mm_srli_epi16(_mm_mullo_epi16(dst, blend), 8))); - }); - break; +static void DPSOFTRAST_Texture2DBGRA8(DPSOFTRAST_Texture *texture, int mip, float x, float y, unsigned char c[4]) + // warning: this is SLOW, only use if the optimized per-span functions won't do +{ + const unsigned char * RESTRICT pixelbase; + const unsigned char * RESTRICT pixel[4]; + int width = texture->mipmap[mip][2], height = texture->mipmap[mip][3]; + int wrapmask[2] = { width-1, height-1 }; + pixelbase = (unsigned char *)texture->bytes + texture->mipmap[mip][0]; + if(texture->filter & DPSOFTRAST_TEXTURE_FILTER_LINEAR) + { + unsigned int tc[2] = { x * (width<<12) - 2048, y * (height<<12) - 2048}; + unsigned int frac[2] = { tc[0]&0xFFF, tc[1]&0xFFF }; + unsigned int ifrac[2] = { 0x1000 - frac[0], 0x1000 - frac[1] }; + unsigned int lerp[4] = { ifrac[0]*ifrac[1], frac[0]*ifrac[1], ifrac[0]*frac[1], frac[0]*frac[1] }; + int tci[2] = { tc[0]>>12, tc[1]>>12 }; + int tci1[2] = { tci[0] + 1, tci[1] + 1 }; + if (texture->flags & DPSOFTRAST_TEXTURE_FLAG_CLAMPTOEDGE) + { + tci[0] = tci[0] >= 0 ? (tci[0] <= wrapmask[0] ? tci[0] : wrapmask[0]) : 0; + tci[1] = tci[1] >= 0 ? (tci[1] <= wrapmask[1] ? tci[1] : wrapmask[1]) : 0; + tci1[0] = tci1[0] >= 0 ? (tci1[0] <= wrapmask[0] ? tci1[0] : wrapmask[0]) : 0; + tci1[1] = tci1[1] >= 0 ? (tci1[1] <= wrapmask[1] ? tci1[1] : wrapmask[1]) : 0; + } + else + { + tci[0] &= wrapmask[0]; + tci[1] &= wrapmask[1]; + tci1[0] &= wrapmask[0]; + tci1[1] &= wrapmask[1]; + } + pixel[0] = pixelbase + 4 * (tci[1]*width+tci[0]); + pixel[1] = pixelbase + 4 * (tci[1]*width+tci1[0]); + pixel[2] = pixelbase + 4 * (tci1[1]*width+tci[0]); + pixel[3] = pixelbase + 4 * (tci1[1]*width+tci1[0]); + c[0] = (pixel[0][0]*lerp[0]+pixel[1][0]*lerp[1]+pixel[2][0]*lerp[2]+pixel[3][0]*lerp[3])>>24; + c[1] = (pixel[0][1]*lerp[0]+pixel[1][1]*lerp[1]+pixel[2][1]*lerp[2]+pixel[3][1]*lerp[3])>>24; + c[2] = (pixel[0][2]*lerp[0]+pixel[1][2]*lerp[1]+pixel[2][2]*lerp[2]+pixel[3][2]*lerp[3])>>24; + c[3] = (pixel[0][3]*lerp[0]+pixel[1][3]*lerp[1]+pixel[2][3]*lerp[2]+pixel[3][3]*lerp[3])>>24; + } + else + { + int tci[2] = { x * width, y * height }; + if (texture->flags & DPSOFTRAST_TEXTURE_FLAG_CLAMPTOEDGE) + { + tci[0] = tci[0] >= 0 ? (tci[0] <= wrapmask[0] ? tci[0] : wrapmask[0]) : 0; + tci[1] = tci[1] >= 0 ? (tci[1] <= wrapmask[1] ? tci[1] : wrapmask[1]) : 0; + } + else + { + tci[0] &= wrapmask[0]; + tci[1] &= wrapmask[1]; + } + pixel[0] = pixelbase + 4 * (tci[1]*width+tci[0]); + c[0] = pixel[0][0]; + c[1] = pixel[0][1]; + c[2] = pixel[0][2]; + c[3] = pixel[0][3]; } -#endif } void DPSOFTRAST_Draw_Span_Texture2DVarying(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, float * RESTRICT out4f, int texunitindex, int arrayindex, const float * RESTRICT zf) @@ -2313,28 +2408,38 @@ void DPSOFTRAST_Draw_Span_Texture2DVarying(DPSOFTRAST_State_Thread *thread, cons tcimax[1] = texture->mipmap[mip][3]-1; tciwrapmask[0] = texture->mipmap[mip][2]-1; tciwrapmask[1] = texture->mipmap[mip][3]-1; - endtc[0] = (data[0] + slope[0]*startx) * zf[startx] * tcscale[0] - 0.5f; - endtc[1] = (data[1] + slope[1]*startx) * zf[startx] * tcscale[1] - 0.5f; + endtc[0] = (data[0] + slope[0]*startx) * zf[startx] * tcscale[0]; + endtc[1] = (data[1] + slope[1]*startx) * zf[startx] * tcscale[1]; + if (filter) + { + endtc[0] -= 0.5f; + endtc[1] -= 0.5f; + } for (x = startx;x < endx;) { unsigned int subtc[2]; unsigned int substep[2]; - float subscale = 65536.0f/DPSOFTRAST_DRAW_MAXSUBSPAN; + float subscale = 4096.0f/DPSOFTRAST_DRAW_MAXSUBSPAN; int nextsub = x + DPSOFTRAST_DRAW_MAXSUBSPAN, endsub = nextsub - 1; - if(nextsub >= endx) + if (nextsub >= endx) { nextsub = endsub = endx-1; - if(x < nextsub) subscale = 65536.0f / (nextsub - x); + if (x < nextsub) subscale = 4096.0f / (nextsub - x); } tc[0] = endtc[0]; tc[1] = endtc[1]; - endtc[0] = (data[0] + slope[0]*nextsub) * zf[nextsub] * tcscale[0] - 0.5f; - endtc[1] = (data[1] + slope[1]*nextsub) * zf[nextsub] * tcscale[1] - 0.5f; + endtc[0] = (data[0] + slope[0]*nextsub) * zf[nextsub] * tcscale[0]; + endtc[1] = (data[1] + slope[1]*nextsub) * zf[nextsub] * tcscale[1]; + if (filter) + { + endtc[0] -= 0.5f; + endtc[1] -= 0.5f; + } substep[0] = (endtc[0] - tc[0]) * subscale; substep[1] = (endtc[1] - tc[1]) * subscale; - subtc[0] = tc[0] * (1<<16); - subtc[1] = tc[1] * (1<<16); - if(filter) + subtc[0] = tc[0] * (1<<12); + subtc[1] = tc[1] * (1<<12); + if (filter) { if (flags & DPSOFTRAST_TEXTURE_FLAG_CLAMPTOEDGE) { @@ -2343,8 +2448,8 @@ void DPSOFTRAST_Draw_Span_Texture2DVarying(DPSOFTRAST_State_Thread *thread, cons unsigned int frac[2] = { subtc[0]&0xFFF, subtc[1]&0xFFF }; unsigned int ifrac[2] = { 0x1000 - frac[0], 0x1000 - frac[1] }; unsigned int lerp[4] = { ifrac[0]*ifrac[1], frac[0]*ifrac[1], ifrac[0]*frac[1], frac[0]*frac[1] }; - tci[0] = subtc[0]>>16; - tci[1] = subtc[1]>>16; + tci[0] = subtc[0]>>12; + tci[1] = subtc[1]>>12; tci1[0] = tci[0] + 1; tci1[1] = tci[1] + 1; tci[0] = tci[0] >= tcimin[0] ? (tci[0] <= tcimax[0] ? tci[0] : tcimax[0]) : tcimin[0]; @@ -2372,8 +2477,8 @@ void DPSOFTRAST_Draw_Span_Texture2DVarying(DPSOFTRAST_State_Thread *thread, cons unsigned int frac[2] = { subtc[0]&0xFFF, subtc[1]&0xFFF }; unsigned int ifrac[2] = { 0x1000 - frac[0], 0x1000 - frac[1] }; unsigned int lerp[4] = { ifrac[0]*ifrac[1], frac[0]*ifrac[1], ifrac[0]*frac[1], frac[0]*frac[1] }; - tci[0] = subtc[0]>>16; - tci[1] = subtc[1]>>16; + tci[0] = subtc[0]>>12; + tci[1] = subtc[1]>>12; tci1[0] = tci[0] + 1; tci1[1] = tci[1] + 1; tci[0] &= tciwrapmask[0]; @@ -2399,8 +2504,8 @@ void DPSOFTRAST_Draw_Span_Texture2DVarying(DPSOFTRAST_State_Thread *thread, cons { for (; x <= endsub; x++, subtc[0] += substep[0], subtc[1] += substep[1]) { - tci[0] = subtc[0]>>16; - tci[1] = subtc[1]>>16; + tci[0] = subtc[0]>>12; + tci[1] = subtc[1]>>12; tci[0] = tci[0] >= tcimin[0] ? (tci[0] <= tcimax[0] ? tci[0] : tcimax[0]) : tcimin[0]; tci[1] = tci[1] >= tcimin[1] ? (tci[1] <= tcimax[1] ? tci[1] : tcimax[1]) : tcimin[1]; pixel[0] = pixelbase + 4 * (tci[1]*tciwidth+tci[0]); @@ -2418,8 +2523,8 @@ void DPSOFTRAST_Draw_Span_Texture2DVarying(DPSOFTRAST_State_Thread *thread, cons { for (; x <= endsub; x++, subtc[0] += substep[0], subtc[1] += substep[1]) { - tci[0] = subtc[0]>>16; - tci[1] = subtc[1]>>16; + tci[0] = subtc[0]>>12; + tci[1] = subtc[1]>>12; tci[0] &= tciwrapmask[0]; tci[1] &= tciwrapmask[1]; pixel[0] = pixelbase + 4 * (tci[1]*tciwidth+tci[0]); @@ -2438,7 +2543,7 @@ void DPSOFTRAST_Draw_Span_Texture2DVarying(DPSOFTRAST_State_Thread *thread, cons void DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, unsigned char * RESTRICT out4ub, int texunitindex, int arrayindex, const float * RESTRICT zf) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE int x; int startx = span->startx; int endx = span->endx; @@ -2449,13 +2554,14 @@ void DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(DPSOFTRAST_State_Thread *thread, __m128i subtc, substep, endsubtc; int filter; int mip; + int affine; // LordHavoc: optimized affine texturing case unsigned int * RESTRICT outi = (unsigned int *)out4ub; const unsigned char * RESTRICT pixelbase; DPSOFTRAST_Texture *texture = thread->texbound[texunitindex]; // if no texture is bound, just fill it with white if (!texture) { - memset(out4ub + startx*4, 255, span->length*4); + memset(out4ub + startx*4, 255, (span->endx - span->startx)*4); return; } mip = triangle->mip[texunitindex]; @@ -2468,6 +2574,7 @@ void DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(DPSOFTRAST_State_Thread *thread, outi[x] = k; return; } + affine = zf[startx] == zf[endx-1]; filter = texture->filter & DPSOFTRAST_TEXTURE_FILTER_LINEAR; DPSOFTRAST_CALCATTRIB(triangle, span, data, slope, arrayindex); flags = texture->flags; @@ -2476,7 +2583,9 @@ void DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(DPSOFTRAST_State_Thread *thread, tcscale = _mm_cvtepi32_ps(tcsize); data = _mm_mul_ps(_mm_movelh_ps(data, data), tcscale); slope = _mm_mul_ps(_mm_movelh_ps(slope, slope), tcscale); - endtc = _mm_sub_ps(_mm_mul_ps(_mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(startx))), _mm_load1_ps(&zf[startx])), _mm_set1_ps(0.5f)); + endtc = _mm_mul_ps(_mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(startx))), _mm_load1_ps(&zf[startx])); + if (filter) + endtc = _mm_sub_ps(endtc, _mm_set1_ps(0.5f)); endsubtc = _mm_cvtps_epi32(_mm_mul_ps(endtc, _mm_set1_ps(65536.0f))); tcoffset = _mm_add_epi32(_mm_slli_epi32(_mm_shuffle_epi32(tcsize, _MM_SHUFFLE(0, 0, 0, 0)), 18), _mm_set1_epi32(4)); tcmax = _mm_packs_epi32(tcmask, tcmask); @@ -2484,14 +2593,16 @@ void DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(DPSOFTRAST_State_Thread *thread, { int nextsub = x + DPSOFTRAST_DRAW_MAXSUBSPAN, endsub = nextsub - 1; __m128 subscale = _mm_set1_ps(65536.0f/DPSOFTRAST_DRAW_MAXSUBSPAN); - if(nextsub >= endx) + if (nextsub >= endx || affine) { nextsub = endsub = endx-1; - if(x < nextsub) subscale = _mm_set1_ps(65536.0f / (nextsub - x)); + if (x < nextsub) subscale = _mm_set1_ps(65536.0f / (nextsub - x)); } tc = endtc; subtc = endsubtc; - endtc = _mm_sub_ps(_mm_mul_ps(_mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(nextsub))), _mm_load1_ps(&zf[nextsub])), _mm_set1_ps(0.5f)); + endtc = _mm_mul_ps(_mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(nextsub))), _mm_load1_ps(&zf[nextsub])); + if (filter) + endtc = _mm_sub_ps(endtc, _mm_set1_ps(0.5f)); substep = _mm_cvtps_epi32(_mm_mul_ps(_mm_sub_ps(endtc, tc), subscale)); endsubtc = _mm_cvtps_epi32(_mm_mul_ps(endtc, _mm_set1_ps(65536.0f))); subtc = _mm_unpacklo_epi64(subtc, _mm_add_epi32(subtc, substep)); @@ -2714,7 +2825,7 @@ void DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(DPSOFTRAST_State_Thread *thread, void DPSOFTRAST_Draw_Span_TextureCubeVaryingBGRA8(const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, unsigned char * RESTRICT out4ub, int texunitindex, int arrayindex, const float * RESTRICT zf) { // TODO: IMPLEMENT - memset(out4ub, 255, span->length*4); + memset(out4ub + span->startx*4, 255, (span->startx - span->endx)*4); } float DPSOFTRAST_SampleShadowmap(const float *vector) @@ -2854,86 +2965,114 @@ void DPSOFTRAST_Draw_Span_MixUniformColor(const DPSOFTRAST_State_Triangle * REST void DPSOFTRAST_Draw_Span_MultiplyVaryingBGRA8(const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, unsigned char *out4ub, const unsigned char *in4ub, int arrayindex, const float *zf) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE int x; int startx = span->startx; int endx = span->endx; __m128 data, slope; + __m128 mod, endmod; + __m128i submod, substep, endsubmod; DPSOFTRAST_CALCATTRIB(triangle, span, data, slope, arrayindex); data = _mm_shuffle_ps(data, data, _MM_SHUFFLE(3, 0, 1, 2)); slope = _mm_shuffle_ps(slope, slope, _MM_SHUFFLE(3, 0, 1, 2)); - data = _mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(startx))); - data = _mm_mul_ps(data, _mm_set1_ps(256.0f)); - slope = _mm_mul_ps(slope, _mm_set1_ps(256.0f)); - for (x = startx;x+2 <= endx;x += 2, data = _mm_add_ps(data, slope)) - { - __m128i pix = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_loadl_epi64((const __m128i *)&in4ub[x*4])); - __m128i mod = _mm_cvtps_epi32(_mm_mul_ps(data, _mm_load1_ps(&zf[x]))), mod2; - data = _mm_add_ps(data, slope); - mod2 = _mm_cvtps_epi32(_mm_mul_ps(data, _mm_load1_ps(&zf[x+1]))); - mod = _mm_unpacklo_epi64(_mm_packs_epi32(mod, mod), _mm_packs_epi32(mod2, mod2)); - pix = _mm_mulhi_epu16(pix, mod); - _mm_storel_epi64((__m128i *)&out4ub[x*4], _mm_packus_epi16(pix, pix)); - } - for (;x < endx;x++, data = _mm_add_ps(data, slope)) + endmod = _mm_mul_ps(_mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(startx))), _mm_load1_ps(&zf[startx])); + endsubmod = _mm_cvtps_epi32(_mm_mul_ps(endmod, _mm_set1_ps(256.0f))); + for (x = startx; x < endx;) { - __m128i pix = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_cvtsi32_si128(*(const int *)&in4ub[x*4])); - __m128i mod = _mm_cvtps_epi32(_mm_mul_ps(data, _mm_load1_ps(&zf[x]))); - mod = _mm_packs_epi32(mod, mod); - pix = _mm_mulhi_epu16(pix, mod); - *(int *)&out4ub[x*4] = _mm_cvtsi128_si32(_mm_packus_epi16(pix, pix)); + int nextsub = x + DPSOFTRAST_DRAW_MAXSUBSPAN, endsub = nextsub - 1; + __m128 subscale = _mm_set1_ps(256.0f/DPSOFTRAST_DRAW_MAXSUBSPAN); + if (nextsub >= endx) + { + nextsub = endsub = endx-1; + if (x < nextsub) subscale = _mm_set1_ps(256.0f / (nextsub - x)); + } + mod = endmod; + submod = endsubmod; + endmod = _mm_mul_ps(_mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(nextsub))), _mm_load1_ps(&zf[nextsub])); + substep = _mm_cvtps_epi32(_mm_mul_ps(_mm_sub_ps(endmod, mod), subscale)); + endsubmod = _mm_cvtps_epi32(_mm_mul_ps(endmod, _mm_set1_ps(256.0f))); + submod = _mm_packs_epi32(submod, _mm_add_epi32(submod, substep)); + substep = _mm_packs_epi32(substep, substep); + for (; x + 1 <= endsub; x += 2, submod = _mm_add_epi16(submod, substep)) + { + __m128i pix = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_loadl_epi64((const __m128i *)&in4ub[x*4])); + pix = _mm_mulhi_epu16(pix, submod); + _mm_storel_epi64((__m128i *)&out4ub[x*4], _mm_packus_epi16(pix, pix)); + } + if (x <= endsub) + { + __m128i pix = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_cvtsi32_si128(*(const int *)&in4ub[x*4])); + pix = _mm_mulhi_epu16(pix, submod); + *(int *)&out4ub[x*4] = _mm_cvtsi128_si32(_mm_packus_epi16(pix, pix)); + x++; + } } #endif } void DPSOFTRAST_Draw_Span_VaryingBGRA8(const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, unsigned char *out4ub, int arrayindex, const float *zf) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE int x; int startx = span->startx; int endx = span->endx; __m128 data, slope; + __m128 mod, endmod; + __m128i submod, substep, endsubmod; DPSOFTRAST_CALCATTRIB(triangle, span, data, slope, arrayindex); data = _mm_shuffle_ps(data, data, _MM_SHUFFLE(3, 0, 1, 2)); slope = _mm_shuffle_ps(slope, slope, _MM_SHUFFLE(3, 0, 1, 2)); - data = _mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(startx))); - data = _mm_mul_ps(data, _mm_set1_ps(255.0f)); - slope = _mm_mul_ps(slope, _mm_set1_ps(255.0f)); - for (x = startx;x+2 <= endx;x += 2, data = _mm_add_ps(data, slope)) - { - __m128i pix = _mm_cvtps_epi32(_mm_mul_ps(data, _mm_load1_ps(&zf[x]))), pix2; - data = _mm_add_ps(data, slope); - pix2 = _mm_cvtps_epi32(_mm_mul_ps(data, _mm_load1_ps(&zf[x+1]))); - pix = _mm_unpacklo_epi64(_mm_packs_epi32(pix, pix), _mm_packs_epi32(pix2, pix2)); - _mm_storel_epi64((__m128i *)&out4ub[x*4], _mm_packus_epi16(pix, pix)); - } - for (;x < endx;x++, data = _mm_add_ps(data, slope)) + endmod = _mm_mul_ps(_mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(startx))), _mm_load1_ps(&zf[startx])); + endsubmod = _mm_cvtps_epi32(_mm_mul_ps(endmod, _mm_set1_ps(4095.0f))); + for (x = startx; x < endx;) { - __m128i pix = _mm_cvtps_epi32(_mm_mul_ps(data, _mm_load1_ps(&zf[x]))); - pix = _mm_packs_epi32(pix, pix); - *(int *)&out4ub[x*4] = _mm_cvtsi128_si32(_mm_packus_epi16(pix, pix)); + int nextsub = x + DPSOFTRAST_DRAW_MAXSUBSPAN, endsub = nextsub - 1; + __m128 subscale = _mm_set1_ps(4095.0f/DPSOFTRAST_DRAW_MAXSUBSPAN); + if (nextsub >= endx) + { + nextsub = endsub = endx-1; + if (x < nextsub) subscale = _mm_set1_ps(4095.0f / (nextsub - x)); + } + mod = endmod; + submod = endsubmod; + endmod = _mm_mul_ps(_mm_add_ps(data, _mm_mul_ps(slope, _mm_set1_ps(nextsub))), _mm_load1_ps(&zf[nextsub])); + substep = _mm_cvtps_epi32(_mm_mul_ps(_mm_sub_ps(endmod, mod), subscale)); + endsubmod = _mm_cvtps_epi32(_mm_mul_ps(endmod, _mm_set1_ps(4095.0f))); + submod = _mm_packs_epi32(submod, _mm_add_epi32(submod, substep)); + substep = _mm_packs_epi32(substep, substep); + for (; x + 1 <= endsub; x += 2, submod = _mm_add_epi16(submod, substep)) + { + __m128i pix = _mm_srai_epi16(submod, 4); + _mm_storel_epi64((__m128i *)&out4ub[x*4], _mm_packus_epi16(pix, pix)); + } + if (x <= endsub) + { + __m128i pix = _mm_srai_epi16(submod, 4); + *(int *)&out4ub[x*4] = _mm_cvtsi128_si32(_mm_packus_epi16(pix, pix)); + x++; + } } #endif } void DPSOFTRAST_Draw_Span_AddBloomBGRA8(const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, unsigned char *out4ub, const unsigned char *ina4ub, const unsigned char *inb4ub, const float *subcolor) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE int x, startx = span->startx, endx = span->endx; __m128i localcolor = _mm_shuffle_epi32(_mm_cvtps_epi32(_mm_mul_ps(_mm_loadu_ps(subcolor), _mm_set1_ps(255.0f))), _MM_SHUFFLE(3, 0, 1, 2)); - localcolor = _mm_shuffle_epi32(_mm_packs_epi32(localcolor, localcolor), _MM_SHUFFLE(1, 0, 1, 0)); + localcolor = _mm_packs_epi32(localcolor, localcolor); for (x = startx;x+2 <= endx;x+=2) { __m128i pix1 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)&ina4ub[x*4]), _mm_setzero_si128()); __m128i pix2 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)&inb4ub[x*4]), _mm_setzero_si128()); - pix1 = _mm_add_epi16(pix1, _mm_sub_epi16(pix2, localcolor)); + pix1 = _mm_add_epi16(pix1, _mm_subs_epu16(pix2, localcolor)); _mm_storel_epi64((__m128i *)&out4ub[x*4], _mm_packus_epi16(pix1, pix1)); } - if(x < endx) + if (x < endx) { __m128i pix1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const int *)&ina4ub[x*4]), _mm_setzero_si128()); __m128i pix2 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const int *)&inb4ub[x*4]), _mm_setzero_si128()); - pix1 = _mm_add_epi16(pix1, _mm_sub_epi16(pix2, localcolor)); + pix1 = _mm_add_epi16(pix1, _mm_subs_epu16(pix2, localcolor)); *(int *)&out4ub[x*4] = _mm_cvtsi128_si32(_mm_packus_epi16(pix1, pix1)); } #endif @@ -2941,7 +3080,7 @@ void DPSOFTRAST_Draw_Span_AddBloomBGRA8(const DPSOFTRAST_State_Triangle * RESTRI void DPSOFTRAST_Draw_Span_MultiplyBuffersBGRA8(const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, unsigned char *out4ub, const unsigned char *ina4ub, const unsigned char *inb4ub) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE int x, startx = span->startx, endx = span->endx; for (x = startx;x+2 <= endx;x+=2) { @@ -2950,7 +3089,7 @@ void DPSOFTRAST_Draw_Span_MultiplyBuffersBGRA8(const DPSOFTRAST_State_Triangle * pix1 = _mm_mulhi_epu16(pix1, pix2); _mm_storel_epi64((__m128i *)&out4ub[x*4], _mm_packus_epi16(pix1, pix1)); } - if(x < endx) + if (x < endx) { __m128i pix1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const int *)&ina4ub[x*4]), _mm_setzero_si128()); __m128i pix2 = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_cvtsi32_si128(*(const int *)&inb4ub[x*4])); @@ -2962,7 +3101,7 @@ void DPSOFTRAST_Draw_Span_MultiplyBuffersBGRA8(const DPSOFTRAST_State_Triangle * void DPSOFTRAST_Draw_Span_AddBuffersBGRA8(const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, unsigned char *out4ub, const unsigned char *ina4ub, const unsigned char *inb4ub) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE int x, startx = span->startx, endx = span->endx; for (x = startx;x+2 <= endx;x+=2) { @@ -2971,7 +3110,7 @@ void DPSOFTRAST_Draw_Span_AddBuffersBGRA8(const DPSOFTRAST_State_Triangle * REST pix1 = _mm_add_epi16(pix1, pix2); _mm_storel_epi64((__m128i *)&out4ub[x*4], _mm_packus_epi16(pix1, pix1)); } - if(x < endx) + if (x < endx) { __m128i pix1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const int *)&ina4ub[x*4]), _mm_setzero_si128()); __m128i pix2 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const int *)&inb4ub[x*4]), _mm_setzero_si128()); @@ -2983,10 +3122,10 @@ void DPSOFTRAST_Draw_Span_AddBuffersBGRA8(const DPSOFTRAST_State_Triangle * REST void DPSOFTRAST_Draw_Span_TintedAddBuffersBGRA8(const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, unsigned char *out4ub, const unsigned char *ina4ub, const unsigned char *inb4ub, const float *inbtintbgra) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE int x, startx = span->startx, endx = span->endx; __m128i tint = _mm_cvtps_epi32(_mm_mul_ps(_mm_loadu_ps(inbtintbgra), _mm_set1_ps(256.0f))); - tint = _mm_shuffle_epi32(_mm_packs_epi32(tint, tint), _MM_SHUFFLE(1, 0, 1, 0)); + tint = _mm_packs_epi32(tint, tint); for (x = startx;x+2 <= endx;x+=2) { __m128i pix1 = _mm_unpacklo_epi8(_mm_loadl_epi64((const __m128i *)&ina4ub[x*4]), _mm_setzero_si128()); @@ -2994,7 +3133,7 @@ void DPSOFTRAST_Draw_Span_TintedAddBuffersBGRA8(const DPSOFTRAST_State_Triangle pix1 = _mm_add_epi16(pix1, _mm_mulhi_epu16(tint, pix2)); _mm_storel_epi64((__m128i *)&out4ub[x*4], _mm_packus_epi16(pix1, pix1)); } - if(x < endx) + if (x < endx) { __m128i pix1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const int *)&ina4ub[x*4]), _mm_setzero_si128()); __m128i pix2 = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_cvtsi32_si128(*(const int *)&inb4ub[x*4])); @@ -3006,7 +3145,7 @@ void DPSOFTRAST_Draw_Span_TintedAddBuffersBGRA8(const DPSOFTRAST_State_Triangle void DPSOFTRAST_Draw_Span_MixBuffersBGRA8(const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, unsigned char *out4ub, const unsigned char *ina4ub, const unsigned char *inb4ub) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE int x, startx = span->startx, endx = span->endx; for (x = startx;x+2 <= endx;x+=2) { @@ -3016,7 +3155,7 @@ void DPSOFTRAST_Draw_Span_MixBuffersBGRA8(const DPSOFTRAST_State_Triangle * REST pix1 = _mm_add_epi16(pix1, _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(pix2, pix1), 4), _mm_slli_epi16(blend, 4))); _mm_storel_epi64((__m128i *)&out4ub[x*4], _mm_packus_epi16(pix1, pix1)); } - if(x < endx) + if (x < endx) { __m128i pix1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const int *)&ina4ub[x*4]), _mm_setzero_si128()); __m128i pix2 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const int *)&inb4ub[x*4]), _mm_setzero_si128()); @@ -3029,10 +3168,10 @@ void DPSOFTRAST_Draw_Span_MixBuffersBGRA8(const DPSOFTRAST_State_Triangle * REST void DPSOFTRAST_Draw_Span_MixUniformColorBGRA8(const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span, unsigned char *out4ub, const unsigned char *in4ub, const float *color) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE int x, startx = span->startx, endx = span->endx; __m128i localcolor = _mm_shuffle_epi32(_mm_cvtps_epi32(_mm_mul_ps(_mm_loadu_ps(color), _mm_set1_ps(255.0f))), _MM_SHUFFLE(3, 0, 1, 2)), blend; - localcolor = _mm_shuffle_epi32(_mm_packs_epi32(localcolor, localcolor), _MM_SHUFFLE(1, 0, 1, 0)); + localcolor = _mm_packs_epi32(localcolor, localcolor); blend = _mm_slli_epi16(_mm_shufflehi_epi16(_mm_shufflelo_epi16(localcolor, _MM_SHUFFLE(3, 3, 3, 3)), _MM_SHUFFLE(3, 3, 3, 3)), 4); for (x = startx;x+2 <= endx;x+=2) { @@ -3040,7 +3179,7 @@ void DPSOFTRAST_Draw_Span_MixUniformColorBGRA8(const DPSOFTRAST_State_Triangle * pix = _mm_add_epi16(pix, _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(localcolor, pix), 4), blend)); _mm_storel_epi64((__m128i *)&out4ub[x*4], _mm_packus_epi16(pix, pix)); } - if(x < endx) + if (x < endx) { __m128i pix = _mm_unpacklo_epi8(_mm_cvtsi32_si128(*(const int *)&in4ub[x*4]), _mm_setzero_si128()); pix = _mm_add_epi16(pix, _mm_mulhi_epi16(_mm_slli_epi16(_mm_sub_epi16(localcolor, pix), 4), blend)); @@ -3102,7 +3241,7 @@ void DPSOFTRAST_VertexShader_PostProcess(void) { DPSOFTRAST_Array_TransformProject(DPSOFTRAST_ARRAY_POSITION, DPSOFTRAST_ARRAY_POSITION, dpsoftrast.uniform4f + 4*DPSOFTRAST_UNIFORM_ModelViewProjectionMatrixM1); DPSOFTRAST_Array_Load(DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD0); - DPSOFTRAST_Array_Load(DPSOFTRAST_ARRAY_TEXCOORD1, DPSOFTRAST_ARRAY_TEXCOORD1); + DPSOFTRAST_Array_Load(DPSOFTRAST_ARRAY_TEXCOORD1, DPSOFTRAST_ARRAY_TEXCOORD4); } void DPSOFTRAST_PixelShader_PostProcess(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span) @@ -3143,7 +3282,7 @@ void DPSOFTRAST_PixelShader_Depth_Or_Shadow(DPSOFTRAST_State_Thread *thread, con float buffer_z[DPSOFTRAST_DRAW_MAXSPANLENGTH]; unsigned char buffer_FragColorbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; DPSOFTRAST_Draw_Span_Begin(thread, triangle, span, buffer_z); - memset(buffer_FragColorbgra8, 0, span->length*4); + memset(buffer_FragColorbgra8 + span->startx*4, 0, (span->endx - span->startx)*4); DPSOFTRAST_Draw_Span_FinishBGRA8(thread, triangle, span, buffer_FragColorbgra8); } @@ -3157,25 +3296,44 @@ void DPSOFTRAST_VertexShader_FlatColor(void) void DPSOFTRAST_PixelShader_FlatColor(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span) { +#ifdef SSE_POSSIBLE + unsigned char * RESTRICT pixelmask = span->pixelmask; + unsigned char * RESTRICT pixel = (unsigned char *)dpsoftrast.fb_colorpixels[0] + (span->y * dpsoftrast.fb_width + span->x) * 4; int x, startx = span->startx, endx = span->endx; - int Color_Ambienti[4]; + __m128i Color_Ambientm; float buffer_z[DPSOFTRAST_DRAW_MAXSPANLENGTH]; unsigned char buffer_texture_colorbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; unsigned char buffer_FragColorbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; - Color_Ambienti[2] = (int)(thread->uniform4f[DPSOFTRAST_UNIFORM_Color_Ambient*4+0]*256.0f); - Color_Ambienti[1] = (int)(thread->uniform4f[DPSOFTRAST_UNIFORM_Color_Ambient*4+1]*256.0f); - Color_Ambienti[0] = (int)(thread->uniform4f[DPSOFTRAST_UNIFORM_Color_Ambient*4+2]*256.0f); - Color_Ambienti[3] = (int)(thread->uniform4f[DPSOFTRAST_UNIFORM_Alpha*4+0] *256.0f); DPSOFTRAST_Draw_Span_Begin(thread, triangle, span, buffer_z); DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_colorbgra8, GL20TU_COLOR, 2, buffer_z); + if ((thread->shader_permutation & SHADERPERMUTATION_ALPHAKILL) || thread->fb_blendmode != DPSOFTRAST_BLENDMODE_OPAQUE) + pixel = buffer_FragColorbgra8; + Color_Ambientm = _mm_shuffle_epi32(_mm_cvtps_epi32(_mm_mul_ps(_mm_load_ps(&thread->uniform4f[DPSOFTRAST_UNIFORM_Color_Ambient*4]), _mm_set1_ps(256.0f))), _MM_SHUFFLE(3, 0, 1, 2)); + Color_Ambientm = _mm_and_si128(Color_Ambientm, _mm_setr_epi32(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0)); + Color_Ambientm = _mm_or_si128(Color_Ambientm, _mm_setr_epi32(0, 0, 0, (int)(thread->uniform4f[DPSOFTRAST_UNIFORM_Alpha*4+0]*255.0f))); + Color_Ambientm = _mm_packs_epi32(Color_Ambientm, Color_Ambientm); for (x = startx;x < endx;x++) { - buffer_FragColorbgra8[x*4+0] = (buffer_texture_colorbgra8[x*4+0] * Color_Ambienti[0])>>8; - buffer_FragColorbgra8[x*4+1] = (buffer_texture_colorbgra8[x*4+1] * Color_Ambienti[1])>>8; - buffer_FragColorbgra8[x*4+2] = (buffer_texture_colorbgra8[x*4+2] * Color_Ambienti[2])>>8; - buffer_FragColorbgra8[x*4+3] = (buffer_texture_colorbgra8[x*4+3] * Color_Ambienti[3])>>8; + __m128i color, pix; + if (x + 4 <= endx && *(const unsigned int *)&pixelmask[x] == 0x01010101) + { + __m128i pix2; + color = _mm_loadu_si128((const __m128i *)&buffer_texture_colorbgra8[x*4]); + pix = _mm_mulhi_epu16(Color_Ambientm, _mm_unpacklo_epi8(_mm_setzero_si128(), color)); + pix2 = _mm_mulhi_epu16(Color_Ambientm, _mm_unpackhi_epi8(_mm_setzero_si128(), color)); + _mm_storeu_si128((__m128i *)&pixel[x*4], _mm_packus_epi16(pix, pix2)); + x += 3; + continue; + } + if (!pixelmask[x]) + continue; + color = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_cvtsi32_si128(*(const int *)&buffer_texture_colorbgra8[x*4])); + pix = _mm_mulhi_epu16(Color_Ambientm, color); + *(int *)&pixel[x*4] = _mm_cvtsi128_si32(_mm_packus_epi16(pix, pix)); } - DPSOFTRAST_Draw_Span_FinishBGRA8(thread, triangle, span, buffer_FragColorbgra8); + if (pixel == buffer_FragColorbgra8) + DPSOFTRAST_Draw_Span_FinishBGRA8(thread, triangle, span, buffer_FragColorbgra8); +#endif } @@ -3189,7 +3347,7 @@ void DPSOFTRAST_VertexShader_VertexColor(void) void DPSOFTRAST_PixelShader_VertexColor(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE unsigned char * RESTRICT pixelmask = span->pixelmask; unsigned char * RESTRICT pixel = (unsigned char *)dpsoftrast.fb_colorpixels[0] + (span->y * dpsoftrast.fb_width + span->x) * 4; int x, startx = span->startx, endx = span->endx; @@ -3201,7 +3359,7 @@ void DPSOFTRAST_PixelShader_VertexColor(DPSOFTRAST_State_Thread *thread, const D int arrayindex = DPSOFTRAST_ARRAY_COLOR; DPSOFTRAST_Draw_Span_Begin(thread, triangle, span, buffer_z); DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_colorbgra8, GL20TU_COLOR, 2, buffer_z); - if (thread->alphatest || thread->fb_blendmode != DPSOFTRAST_BLENDMODE_OPAQUE) + if ((thread->shader_permutation & SHADERPERMUTATION_ALPHAKILL) || thread->fb_blendmode != DPSOFTRAST_BLENDMODE_OPAQUE) pixel = buffer_FragColorbgra8; Color_Ambientm = _mm_shuffle_epi32(_mm_cvtps_epi32(_mm_mul_ps(_mm_load_ps(&thread->uniform4f[DPSOFTRAST_UNIFORM_Color_Ambient*4]), _mm_set1_ps(256.0f))), _MM_SHUFFLE(3, 0, 1, 2)); Color_Ambientm = _mm_and_si128(Color_Ambientm, _mm_setr_epi32(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0)); @@ -3239,7 +3397,7 @@ void DPSOFTRAST_PixelShader_VertexColor(DPSOFTRAST_State_Thread *thread, const D x += 3; continue; } - if(!pixelmask[x]) + if (!pixelmask[x]) continue; color = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_cvtsi32_si128(*(const int *)&buffer_texture_colorbgra8[x*4])); mod = _mm_cvtps_epi32(_mm_mul_ps(data, _mm_load1_ps(&buffer_z[x]))); @@ -3247,7 +3405,7 @@ void DPSOFTRAST_PixelShader_VertexColor(DPSOFTRAST_State_Thread *thread, const D pix = _mm_mulhi_epu16(_mm_add_epi16(_mm_mulhi_epu16(mod, Color_Diffusem), Color_Ambientm), color); *(int *)&pixel[x*4] = _mm_cvtsi128_si32(_mm_packus_epi16(pix, pix)); } - if(pixel == buffer_FragColorbgra8) + if (pixel == buffer_FragColorbgra8) DPSOFTRAST_Draw_Span_FinishBGRA8(thread, triangle, span, buffer_FragColorbgra8); #endif } @@ -3263,7 +3421,7 @@ void DPSOFTRAST_VertexShader_Lightmap(void) void DPSOFTRAST_PixelShader_Lightmap(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE unsigned char * RESTRICT pixelmask = span->pixelmask; unsigned char * RESTRICT pixel = (unsigned char *)dpsoftrast.fb_colorpixels[0] + (span->y * dpsoftrast.fb_width + span->x) * 4; int x, startx = span->startx, endx = span->endx; @@ -3276,7 +3434,7 @@ void DPSOFTRAST_PixelShader_Lightmap(DPSOFTRAST_State_Thread *thread, const DPSO DPSOFTRAST_Draw_Span_Begin(thread, triangle, span, buffer_z); DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_colorbgra8, GL20TU_COLOR, DPSOFTRAST_ARRAY_TEXCOORD0, buffer_z); DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_lightmapbgra8, GL20TU_LIGHTMAP, DPSOFTRAST_ARRAY_TEXCOORD4, buffer_z); - if (thread->alphatest || thread->fb_blendmode != DPSOFTRAST_BLENDMODE_OPAQUE) + if ((thread->shader_permutation & SHADERPERMUTATION_ALPHAKILL) || thread->fb_blendmode != DPSOFTRAST_BLENDMODE_OPAQUE) pixel = buffer_FragColorbgra8; Color_Ambientm = _mm_shuffle_epi32(_mm_cvtps_epi32(_mm_mul_ps(_mm_load_ps(&thread->uniform4f[DPSOFTRAST_UNIFORM_Color_Ambient*4]), _mm_set1_ps(256.0f))), _MM_SHUFFLE(3, 0, 1, 2)); Color_Ambientm = _mm_and_si128(Color_Ambientm, _mm_setr_epi32(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0)); @@ -3311,7 +3469,7 @@ void DPSOFTRAST_PixelShader_Lightmap(DPSOFTRAST_State_Thread *thread, const DPSO x += 3; continue; } - if(!pixelmask[x]) + if (!pixelmask[x]) continue; color = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_cvtsi32_si128(*(const int *)&buffer_texture_colorbgra8[x*4])); lightmap = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_cvtsi32_si128(*(const int *)&buffer_texture_lightmapbgra8[x*4])); @@ -3339,7 +3497,7 @@ void DPSOFTRAST_PixelShader_Lightmap(DPSOFTRAST_State_Thread *thread, const DPSO x += 3; continue; } - if(!pixelmask[x]) + if (!pixelmask[x]) continue; color = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_cvtsi32_si128(*(const int *)&buffer_texture_colorbgra8[x*4])); lightmap = _mm_unpacklo_epi8(_mm_setzero_si128(), _mm_cvtsi32_si128(*(const int *)&buffer_texture_lightmapbgra8[x*4])); @@ -3347,52 +3505,49 @@ void DPSOFTRAST_PixelShader_Lightmap(DPSOFTRAST_State_Thread *thread, const DPSO *(int *)&pixel[x*4] = _mm_cvtsi128_si32(_mm_packus_epi16(pix, pix)); } } - if(pixel == buffer_FragColorbgra8) + if (pixel == buffer_FragColorbgra8) DPSOFTRAST_Draw_Span_FinishBGRA8(thread, triangle, span, buffer_FragColorbgra8); #endif } +void DPSOFTRAST_VertexShader_LightDirection(void); +void DPSOFTRAST_PixelShader_LightDirection(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span); void DPSOFTRAST_VertexShader_FakeLight(void) { - DPSOFTRAST_Array_TransformProject(DPSOFTRAST_ARRAY_POSITION, DPSOFTRAST_ARRAY_POSITION, dpsoftrast.uniform4f + 4*DPSOFTRAST_UNIFORM_ModelViewProjectionMatrixM1); + DPSOFTRAST_VertexShader_LightDirection(); } void DPSOFTRAST_PixelShader_FakeLight(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span) { - // TODO: IMPLEMENT - float buffer_z[DPSOFTRAST_DRAW_MAXSPANLENGTH]; - unsigned char buffer_FragColorbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; - DPSOFTRAST_Draw_Span_Begin(thread, triangle, span, buffer_z); - memset(buffer_FragColorbgra8, 0, span->length*4); - DPSOFTRAST_Draw_Span_FinishBGRA8(thread, triangle, span, buffer_FragColorbgra8); + DPSOFTRAST_PixelShader_LightDirection(thread, triangle, span); } void DPSOFTRAST_VertexShader_LightDirectionMap_ModelSpace(void) { - DPSOFTRAST_VertexShader_Lightmap(); + DPSOFTRAST_VertexShader_LightDirection(); + DPSOFTRAST_Array_Load(DPSOFTRAST_ARRAY_TEXCOORD4, DPSOFTRAST_ARRAY_TEXCOORD4); } void DPSOFTRAST_PixelShader_LightDirectionMap_ModelSpace(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span) { - DPSOFTRAST_PixelShader_Lightmap(thread, triangle, span); - // TODO: IMPLEMENT + DPSOFTRAST_PixelShader_LightDirection(thread, triangle, span); } void DPSOFTRAST_VertexShader_LightDirectionMap_TangentSpace(void) { - DPSOFTRAST_VertexShader_Lightmap(); + DPSOFTRAST_VertexShader_LightDirection(); + DPSOFTRAST_Array_Load(DPSOFTRAST_ARRAY_TEXCOORD4, DPSOFTRAST_ARRAY_TEXCOORD4); } void DPSOFTRAST_PixelShader_LightDirectionMap_TangentSpace(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span) { - DPSOFTRAST_PixelShader_Lightmap(thread, triangle, span); - // TODO: IMPLEMENT + DPSOFTRAST_PixelShader_LightDirection(thread, triangle, span); } @@ -3440,20 +3595,20 @@ void DPSOFTRAST_VertexShader_LightDirection(void) LightVector[0] = svector[0] * LightDir[0] + svector[1] * LightDir[1] + svector[2] * LightDir[2]; LightVector[1] = tvector[0] * LightDir[0] + tvector[1] * LightDir[1] + tvector[2] * LightDir[2]; LightVector[2] = normal[0] * LightDir[0] + normal[1] * LightDir[1] + normal[2] * LightDir[2]; - dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD1][i*4+0] = LightVector[0]; - dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD1][i*4+1] = LightVector[1]; - dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD1][i*4+2] = LightVector[2]; - dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD1][i*4+3] = 0.0f; + dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD5][i*4+0] = LightVector[0]; + dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD5][i*4+1] = LightVector[1]; + dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD5][i*4+2] = LightVector[2]; + dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD5][i*4+3] = 0.0f; EyeVectorModelSpace[0] = EyePosition[0] - position[0]; EyeVectorModelSpace[1] = EyePosition[1] - position[1]; EyeVectorModelSpace[2] = EyePosition[2] - position[2]; EyeVector[0] = svector[0] * EyeVectorModelSpace[0] + svector[1] * EyeVectorModelSpace[1] + svector[2] * EyeVectorModelSpace[2]; EyeVector[1] = tvector[0] * EyeVectorModelSpace[0] + tvector[1] * EyeVectorModelSpace[1] + tvector[2] * EyeVectorModelSpace[2]; EyeVector[2] = normal[0] * EyeVectorModelSpace[0] + normal[1] * EyeVectorModelSpace[1] + normal[2] * EyeVectorModelSpace[2]; - dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD2][i*4+0] = EyeVector[0]; - dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD2][i*4+1] = EyeVector[1]; - dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD2][i*4+2] = EyeVector[2]; - dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD2][i*4+3] = 0.0f; + dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD6][i*4+0] = EyeVector[0]; + dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD6][i*4+1] = EyeVector[1]; + dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD6][i*4+2] = EyeVector[2]; + dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD6][i*4+3] = 0.0f; } DPSOFTRAST_Array_TransformProject(DPSOFTRAST_ARRAY_POSITION, -1, dpsoftrast.uniform4f + 4*DPSOFTRAST_UNIFORM_ModelViewProjectionMatrixM1); } @@ -3486,6 +3641,8 @@ void DPSOFTRAST_PixelShader_LightDirection(DPSOFTRAST_State_Thread *thread, cons unsigned char buffer_texture_glowbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; unsigned char buffer_texture_pantsbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; unsigned char buffer_texture_shirtbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; + unsigned char buffer_texture_deluxemapbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; + unsigned char buffer_texture_lightmapbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; unsigned char buffer_FragColorbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; int x, startx = span->startx, endx = span->endx; float Color_Ambient[4], Color_Diffuse[4], Color_Specular[4], Color_Glow[4], Color_Pants[4], Color_Shirt[4], LightColor[4]; @@ -3493,11 +3650,18 @@ void DPSOFTRAST_PixelShader_LightDirection(DPSOFTRAST_State_Thread *thread, cons float LightVectorslope[4]; float EyeVectordata[4]; float EyeVectorslope[4]; + float VectorSdata[4]; + float VectorSslope[4]; + float VectorTdata[4]; + float VectorTslope[4]; + float VectorRdata[4]; + float VectorRslope[4]; float z; float diffusetex[4]; float glosstex[4]; float surfacenormal[4]; float lightnormal[4]; + float lightnormal_modelspace[4]; float eyenormal[4]; float specularnormal[4]; float diffuse; @@ -3541,15 +3705,37 @@ void DPSOFTRAST_PixelShader_LightDirection(DPSOFTRAST_State_Thread *thread, cons LightColor[1] = thread->uniform4f[DPSOFTRAST_UNIFORM_LightColor*4+1]; LightColor[0] = thread->uniform4f[DPSOFTRAST_UNIFORM_LightColor*4+2]; LightColor[3] = 0.0f; - DPSOFTRAST_CALCATTRIB4F(triangle, span, LightVectordata, LightVectorslope, DPSOFTRAST_ARRAY_TEXCOORD1); DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_normalbgra8, GL20TU_NORMAL, DPSOFTRAST_ARRAY_TEXCOORD0, buffer_z); Color_Specular[2] = thread->uniform4f[DPSOFTRAST_UNIFORM_Color_Specular*4+0]; Color_Specular[1] = thread->uniform4f[DPSOFTRAST_UNIFORM_Color_Specular*4+1]; Color_Specular[0] = thread->uniform4f[DPSOFTRAST_UNIFORM_Color_Specular*4+2]; Color_Specular[3] = 0.0f; SpecularPower = thread->uniform4f[DPSOFTRAST_UNIFORM_SpecularPower*4+0] * (1.0f / 255.0f); - DPSOFTRAST_CALCATTRIB4F(triangle, span, EyeVectordata, EyeVectorslope, DPSOFTRAST_ARRAY_TEXCOORD2); + DPSOFTRAST_CALCATTRIB4F(triangle, span, EyeVectordata, EyeVectorslope, DPSOFTRAST_ARRAY_TEXCOORD6); DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_glossbgra8, GL20TU_GLOSS, DPSOFTRAST_ARRAY_TEXCOORD0, buffer_z); + + if(thread->shader_mode == SHADERMODE_LIGHTDIRECTIONMAP_MODELSPACE) + { + DPSOFTRAST_CALCATTRIB4F(triangle, span, VectorSdata, VectorSslope, DPSOFTRAST_ARRAY_TEXCOORD1); + DPSOFTRAST_CALCATTRIB4F(triangle, span, VectorTdata, VectorTslope, DPSOFTRAST_ARRAY_TEXCOORD2); + DPSOFTRAST_CALCATTRIB4F(triangle, span, VectorRdata, VectorRslope, DPSOFTRAST_ARRAY_TEXCOORD3); + DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_lightmapbgra8, GL20TU_LIGHTMAP, DPSOFTRAST_ARRAY_TEXCOORD4, buffer_z); + DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_deluxemapbgra8, GL20TU_DELUXEMAP, DPSOFTRAST_ARRAY_TEXCOORD4, buffer_z); + } + else if(thread->shader_mode == SHADERMODE_LIGHTDIRECTIONMAP_TANGENTSPACE) + { + DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_lightmapbgra8, GL20TU_LIGHTMAP, DPSOFTRAST_ARRAY_TEXCOORD4, buffer_z); + DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_deluxemapbgra8, GL20TU_DELUXEMAP, DPSOFTRAST_ARRAY_TEXCOORD4, buffer_z); + } + else if(thread->shader_mode == SHADERMODE_FAKELIGHT) + { + // nothing of this needed + } + else + { + DPSOFTRAST_CALCATTRIB4F(triangle, span, LightVectordata, LightVectorslope, DPSOFTRAST_ARRAY_TEXCOORD5); + } + for (x = startx;x < endx;x++) { z = buffer_z[x]; @@ -3573,23 +3759,105 @@ void DPSOFTRAST_PixelShader_LightDirection(DPSOFTRAST_State_Thread *thread, cons surfacenormal[2] = buffer_texture_normalbgra8[x*4+0] * (1.0f / 128.0f) - 1.0f; DPSOFTRAST_Vector3Normalize(surfacenormal); - lightnormal[0] = (LightVectordata[0] + LightVectorslope[0]*x) * z; - lightnormal[1] = (LightVectordata[1] + LightVectorslope[1]*x) * z; - lightnormal[2] = (LightVectordata[2] + LightVectorslope[2]*x) * z; - DPSOFTRAST_Vector3Normalize(lightnormal); + if(thread->shader_mode == SHADERMODE_LIGHTDIRECTIONMAP_MODELSPACE) + { + // myhalf3 lightnormal_modelspace = myhalf3(dp_texture2D(Texture_Deluxemap, TexCoordSurfaceLightmap.zw)) * 2.0 + myhalf3(-1.0, -1.0, -1.0);\n"; + lightnormal_modelspace[0] = buffer_texture_deluxemapbgra8[x*4+2] * (1.0f / 128.0f) - 1.0f; + lightnormal_modelspace[1] = buffer_texture_deluxemapbgra8[x*4+1] * (1.0f / 128.0f) - 1.0f; + lightnormal_modelspace[2] = buffer_texture_deluxemapbgra8[x*4+0] * (1.0f / 128.0f) - 1.0f; + + // lightnormal.x = dot(lightnormal_modelspace, myhalf3(VectorS));\n" + lightnormal[0] = lightnormal_modelspace[0] * (VectorSdata[0] + VectorSslope[0] * x) + + lightnormal_modelspace[1] * (VectorSdata[1] + VectorSslope[1] * x) + + lightnormal_modelspace[2] * (VectorSdata[2] + VectorSslope[2] * x); + + // lightnormal.y = dot(lightnormal_modelspace, myhalf3(VectorT));\n" + lightnormal[1] = lightnormal_modelspace[0] * (VectorTdata[0] + VectorTslope[0] * x) + + lightnormal_modelspace[1] * (VectorTdata[1] + VectorTslope[1] * x) + + lightnormal_modelspace[2] * (VectorTdata[2] + VectorTslope[2] * x); + + // lightnormal.z = dot(lightnormal_modelspace, myhalf3(VectorR));\n" + lightnormal[2] = lightnormal_modelspace[0] * (VectorRdata[0] + VectorRslope[0] * x) + + lightnormal_modelspace[1] * (VectorRdata[1] + VectorRslope[1] * x) + + lightnormal_modelspace[2] * (VectorRdata[2] + VectorRslope[2] * x); + + // lightnormal = normalize(lightnormal); // VectorS/T/R are not always perfectly normalized, and EXACTSPECULARMATH is very picky about this\n" + DPSOFTRAST_Vector3Normalize(lightnormal); + + // myhalf3 lightcolor = myhalf3(dp_texture2D(Texture_Lightmap, TexCoordSurfaceLightmap.zw));\n"; + { + float f = 1.0f / (256.0f * max(0.25f, lightnormal[2])); + LightColor[0] = buffer_texture_lightmapbgra8[x*4+0] * f; + LightColor[1] = buffer_texture_lightmapbgra8[x*4+1] * f; + LightColor[2] = buffer_texture_lightmapbgra8[x*4+2] * f; + } + } + else if(thread->shader_mode == SHADERMODE_LIGHTDIRECTIONMAP_TANGENTSPACE) + { + lightnormal[0] = buffer_texture_deluxemapbgra8[x*4+2] * (1.0f / 128.0f) - 1.0f; + lightnormal[1] = buffer_texture_deluxemapbgra8[x*4+1] * (1.0f / 128.0f) - 1.0f; + lightnormal[2] = buffer_texture_deluxemapbgra8[x*4+0] * (1.0f / 128.0f) - 1.0f; + { + float f = 1.0f / 256.0f; + LightColor[0] = buffer_texture_lightmapbgra8[x*4+0] * f; + LightColor[1] = buffer_texture_lightmapbgra8[x*4+1] * f; + LightColor[2] = buffer_texture_lightmapbgra8[x*4+2] * f; + } + } + else if(thread->shader_mode == SHADERMODE_FAKELIGHT) + { + lightnormal[0] = (EyeVectordata[0] + EyeVectorslope[0]*x) * z; + lightnormal[1] = (EyeVectordata[1] + EyeVectorslope[1]*x) * z; + lightnormal[2] = (EyeVectordata[2] + EyeVectorslope[2]*x) * z; + DPSOFTRAST_Vector3Normalize(lightnormal); + + LightColor[0] = 1.0; + LightColor[1] = 1.0; + LightColor[2] = 1.0; + } + else + { + lightnormal[0] = (LightVectordata[0] + LightVectorslope[0]*x) * z; + lightnormal[1] = (LightVectordata[1] + LightVectorslope[1]*x) * z; + lightnormal[2] = (LightVectordata[2] + LightVectorslope[2]*x) * z; + DPSOFTRAST_Vector3Normalize(lightnormal); + } - eyenormal[0] = (EyeVectordata[0] + EyeVectorslope[0]*x) * z; - eyenormal[1] = (EyeVectordata[1] + EyeVectorslope[1]*x) * z; - eyenormal[2] = (EyeVectordata[2] + EyeVectorslope[2]*x) * z; - DPSOFTRAST_Vector3Normalize(eyenormal); + diffuse = DPSOFTRAST_Vector3Dot(surfacenormal, lightnormal);if (diffuse < 0.0f) diffuse = 0.0f; - specularnormal[0] = lightnormal[0] + eyenormal[0]; - specularnormal[1] = lightnormal[1] + eyenormal[1]; - specularnormal[2] = lightnormal[2] + eyenormal[2]; - DPSOFTRAST_Vector3Normalize(specularnormal); + if(thread->shader_exactspecularmath) + { + // reflect lightnormal at surfacenormal, take the negative of that + // i.e. we want (2*dot(N, i) * N - I) for N=surfacenormal, I=lightnormal + float f; + f = DPSOFTRAST_Vector3Dot(lightnormal, surfacenormal); + specularnormal[0] = 2*f*surfacenormal[0] - lightnormal[0]; + specularnormal[1] = 2*f*surfacenormal[1] - lightnormal[1]; + specularnormal[2] = 2*f*surfacenormal[2] - lightnormal[2]; + + // dot of this and normalize(EyeVectorFogDepth.xyz) + eyenormal[0] = (EyeVectordata[0] + EyeVectorslope[0]*x) * z; + eyenormal[1] = (EyeVectordata[1] + EyeVectorslope[1]*x) * z; + eyenormal[2] = (EyeVectordata[2] + EyeVectorslope[2]*x) * z; + DPSOFTRAST_Vector3Normalize(eyenormal); + + specular = DPSOFTRAST_Vector3Dot(eyenormal, specularnormal);if (specular < 0.0f) specular = 0.0f; + } + else + { + eyenormal[0] = (EyeVectordata[0] + EyeVectorslope[0]*x) * z; + eyenormal[1] = (EyeVectordata[1] + EyeVectorslope[1]*x) * z; + eyenormal[2] = (EyeVectordata[2] + EyeVectorslope[2]*x) * z; + DPSOFTRAST_Vector3Normalize(eyenormal); + + specularnormal[0] = lightnormal[0] + eyenormal[0]; + specularnormal[1] = lightnormal[1] + eyenormal[1]; + specularnormal[2] = lightnormal[2] + eyenormal[2]; + DPSOFTRAST_Vector3Normalize(specularnormal); + + specular = DPSOFTRAST_Vector3Dot(surfacenormal, specularnormal);if (specular < 0.0f) specular = 0.0f; + } - diffuse = DPSOFTRAST_Vector3Dot(surfacenormal, lightnormal);if (diffuse < 0.0f) diffuse = 0.0f; - specular = DPSOFTRAST_Vector3Dot(surfacenormal, specularnormal);if (specular < 0.0f) specular = 0.0f; specular = pow(specular, SpecularPower * glosstex[3]); if (thread->shader_permutation & SHADERPERMUTATION_GLOW) { @@ -3605,6 +3873,7 @@ void DPSOFTRAST_PixelShader_LightDirection(DPSOFTRAST_State_Thread *thread, cons d[2] = (int)( diffusetex[2] * Color_Ambient[2] + (diffusetex[2] * Color_Diffuse[2] * diffuse + glosstex[2] * Color_Specular[2] * specular) * LightColor[2]);if (d[2] > 255) d[2] = 255; d[3] = (int)( diffusetex[3] * Color_Ambient[3]);if (d[3] > 255) d[3] = 255; } + buffer_FragColorbgra8[x*4+0] = d[0]; buffer_FragColorbgra8[x*4+1] = d[1]; buffer_FragColorbgra8[x*4+2] = d[2]; @@ -3621,8 +3890,30 @@ void DPSOFTRAST_PixelShader_LightDirection(DPSOFTRAST_State_Thread *thread, cons LightColor[1] = thread->uniform4f[DPSOFTRAST_UNIFORM_LightColor*4+1]; LightColor[0] = thread->uniform4f[DPSOFTRAST_UNIFORM_LightColor*4+2]; LightColor[3] = 0.0f; - DPSOFTRAST_CALCATTRIB4F(triangle, span, LightVectordata, LightVectorslope, DPSOFTRAST_ARRAY_TEXCOORD1); DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_normalbgra8, GL20TU_NORMAL, DPSOFTRAST_ARRAY_TEXCOORD0, buffer_z); + + if(thread->shader_mode == SHADERMODE_LIGHTDIRECTIONMAP_MODELSPACE) + { + DPSOFTRAST_CALCATTRIB4F(triangle, span, VectorSdata, VectorSslope, DPSOFTRAST_ARRAY_TEXCOORD1); + DPSOFTRAST_CALCATTRIB4F(triangle, span, VectorTdata, VectorTslope, DPSOFTRAST_ARRAY_TEXCOORD2); + DPSOFTRAST_CALCATTRIB4F(triangle, span, VectorRdata, VectorRslope, DPSOFTRAST_ARRAY_TEXCOORD3); + DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_lightmapbgra8, GL20TU_LIGHTMAP, DPSOFTRAST_ARRAY_TEXCOORD4, buffer_z); + DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_deluxemapbgra8, GL20TU_DELUXEMAP, DPSOFTRAST_ARRAY_TEXCOORD4, buffer_z); + } + else if(thread->shader_mode == SHADERMODE_LIGHTDIRECTIONMAP_TANGENTSPACE) + { + DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_lightmapbgra8, GL20TU_LIGHTMAP, DPSOFTRAST_ARRAY_TEXCOORD4, buffer_z); + DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_deluxemapbgra8, GL20TU_DELUXEMAP, DPSOFTRAST_ARRAY_TEXCOORD4, buffer_z); + } + else if(thread->shader_mode == SHADERMODE_FAKELIGHT) + { + DPSOFTRAST_CALCATTRIB4F(triangle, span, EyeVectordata, EyeVectorslope, DPSOFTRAST_ARRAY_TEXCOORD6); + } + else + { + DPSOFTRAST_CALCATTRIB4F(triangle, span, LightVectordata, LightVectorslope, DPSOFTRAST_ARRAY_TEXCOORD5); + } + for (x = startx;x < endx;x++) { z = buffer_z[x]; @@ -3635,10 +3926,69 @@ void DPSOFTRAST_PixelShader_LightDirection(DPSOFTRAST_State_Thread *thread, cons surfacenormal[2] = buffer_texture_normalbgra8[x*4+0] * (1.0f / 128.0f) - 1.0f; DPSOFTRAST_Vector3Normalize(surfacenormal); - lightnormal[0] = (LightVectordata[0] + LightVectorslope[0]*x) * z; - lightnormal[1] = (LightVectordata[1] + LightVectorslope[1]*x) * z; - lightnormal[2] = (LightVectordata[2] + LightVectorslope[2]*x) * z; - DPSOFTRAST_Vector3Normalize(lightnormal); + if(thread->shader_mode == SHADERMODE_LIGHTDIRECTIONMAP_MODELSPACE) + { + // myhalf3 lightnormal_modelspace = myhalf3(dp_texture2D(Texture_Deluxemap, TexCoordSurfaceLightmap.zw)) * 2.0 + myhalf3(-1.0, -1.0, -1.0);\n"; + lightnormal_modelspace[0] = buffer_texture_deluxemapbgra8[x*4+2] * (1.0f / 128.0f) - 1.0f; + lightnormal_modelspace[1] = buffer_texture_deluxemapbgra8[x*4+1] * (1.0f / 128.0f) - 1.0f; + lightnormal_modelspace[2] = buffer_texture_deluxemapbgra8[x*4+0] * (1.0f / 128.0f) - 1.0f; + + // lightnormal.x = dot(lightnormal_modelspace, myhalf3(VectorS));\n" + lightnormal[0] = lightnormal_modelspace[0] * (VectorSdata[0] + VectorSslope[0] * x) + + lightnormal_modelspace[1] * (VectorSdata[1] + VectorSslope[1] * x) + + lightnormal_modelspace[2] * (VectorSdata[2] + VectorSslope[2] * x); + + // lightnormal.y = dot(lightnormal_modelspace, myhalf3(VectorT));\n" + lightnormal[1] = lightnormal_modelspace[0] * (VectorTdata[0] + VectorTslope[0] * x) + + lightnormal_modelspace[1] * (VectorTdata[1] + VectorTslope[1] * x) + + lightnormal_modelspace[2] * (VectorTdata[2] + VectorTslope[2] * x); + + // lightnormal.z = dot(lightnormal_modelspace, myhalf3(VectorR));\n" + lightnormal[2] = lightnormal_modelspace[0] * (VectorRdata[0] + VectorRslope[0] * x) + + lightnormal_modelspace[1] * (VectorRdata[1] + VectorRslope[1] * x) + + lightnormal_modelspace[2] * (VectorRdata[2] + VectorRslope[2] * x); + + // lightnormal = normalize(lightnormal); // VectorS/T/R are not always perfectly normalized, and EXACTSPECULARMATH is very picky about this\n" + DPSOFTRAST_Vector3Normalize(lightnormal); + + // myhalf3 lightcolor = myhalf3(dp_texture2D(Texture_Lightmap, TexCoordSurfaceLightmap.zw));\n"; + { + float f = 1.0f / (256.0f * max(0.25f, lightnormal[2])); + LightColor[0] = buffer_texture_lightmapbgra8[x*4+0] * f; + LightColor[1] = buffer_texture_lightmapbgra8[x*4+1] * f; + LightColor[2] = buffer_texture_lightmapbgra8[x*4+2] * f; + } + } + else if(thread->shader_mode == SHADERMODE_LIGHTDIRECTIONMAP_TANGENTSPACE) + { + lightnormal[0] = buffer_texture_deluxemapbgra8[x*4+2] * (1.0f / 128.0f) - 1.0f; + lightnormal[1] = buffer_texture_deluxemapbgra8[x*4+1] * (1.0f / 128.0f) - 1.0f; + lightnormal[2] = buffer_texture_deluxemapbgra8[x*4+0] * (1.0f / 128.0f) - 1.0f; + { + float f = 1.0f / 256.0f; + LightColor[0] = buffer_texture_lightmapbgra8[x*4+0] * f; + LightColor[1] = buffer_texture_lightmapbgra8[x*4+1] * f; + LightColor[2] = buffer_texture_lightmapbgra8[x*4+2] * f; + } + } + else if(thread->shader_mode == SHADERMODE_FAKELIGHT) + { + lightnormal[0] = (EyeVectordata[0] + EyeVectorslope[0]*x) * z; + lightnormal[1] = (EyeVectordata[1] + EyeVectorslope[1]*x) * z; + lightnormal[2] = (EyeVectordata[2] + EyeVectorslope[2]*x) * z; + DPSOFTRAST_Vector3Normalize(lightnormal); + + LightColor[0] = 1.0; + LightColor[1] = 1.0; + LightColor[2] = 1.0; + } + else + { + lightnormal[0] = (LightVectordata[0] + LightVectorslope[0]*x) * z; + lightnormal[1] = (LightVectordata[1] + LightVectorslope[1]*x) * z; + lightnormal[2] = (LightVectordata[2] + LightVectorslope[2]*x) * z; + DPSOFTRAST_Vector3Normalize(lightnormal); + } diffuse = DPSOFTRAST_Vector3Dot(surfacenormal, lightnormal);if (diffuse < 0.0f) diffuse = 0.0f; if (thread->shader_permutation & SHADERPERMUTATION_GLOW) @@ -3765,7 +4115,7 @@ void DPSOFTRAST_VertexShader_LightSource(void) void DPSOFTRAST_PixelShader_LightSource(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE float buffer_z[DPSOFTRAST_DRAW_MAXSPANLENGTH]; unsigned char buffer_texture_colorbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; unsigned char buffer_texture_normalbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; @@ -3882,19 +4232,42 @@ void DPSOFTRAST_PixelShader_LightSource(DPSOFTRAST_State_Thread *thread, const D lightnormal[2] = (LightVectordata[2] + LightVectorslope[2]*x) * z; DPSOFTRAST_Vector3Normalize(lightnormal); - eyenormal[0] = (EyeVectordata[0] + EyeVectorslope[0]*x) * z; - eyenormal[1] = (EyeVectordata[1] + EyeVectorslope[1]*x) * z; - eyenormal[2] = (EyeVectordata[2] + EyeVectorslope[2]*x) * z; - DPSOFTRAST_Vector3Normalize(eyenormal); + diffuse = DPSOFTRAST_Vector3Dot(surfacenormal, lightnormal);if (diffuse < 0.0f) diffuse = 0.0f; - specularnormal[0] = lightnormal[0] + eyenormal[0]; - specularnormal[1] = lightnormal[1] + eyenormal[1]; - specularnormal[2] = lightnormal[2] + eyenormal[2]; - DPSOFTRAST_Vector3Normalize(specularnormal); + if(thread->shader_exactspecularmath) + { + // reflect lightnormal at surfacenormal, take the negative of that + // i.e. we want (2*dot(N, i) * N - I) for N=surfacenormal, I=lightnormal + float f; + f = DPSOFTRAST_Vector3Dot(lightnormal, surfacenormal); + specularnormal[0] = 2*f*surfacenormal[0] - lightnormal[0]; + specularnormal[1] = 2*f*surfacenormal[1] - lightnormal[1]; + specularnormal[2] = 2*f*surfacenormal[2] - lightnormal[2]; + + // dot of this and normalize(EyeVectorFogDepth.xyz) + eyenormal[0] = (EyeVectordata[0] + EyeVectorslope[0]*x) * z; + eyenormal[1] = (EyeVectordata[1] + EyeVectorslope[1]*x) * z; + eyenormal[2] = (EyeVectordata[2] + EyeVectorslope[2]*x) * z; + DPSOFTRAST_Vector3Normalize(eyenormal); + + specular = DPSOFTRAST_Vector3Dot(eyenormal, specularnormal);if (specular < 0.0f) specular = 0.0f; + } + else + { + eyenormal[0] = (EyeVectordata[0] + EyeVectorslope[0]*x) * z; + eyenormal[1] = (EyeVectordata[1] + EyeVectorslope[1]*x) * z; + eyenormal[2] = (EyeVectordata[2] + EyeVectorslope[2]*x) * z; + DPSOFTRAST_Vector3Normalize(eyenormal); - diffuse = DPSOFTRAST_Vector3Dot(surfacenormal, lightnormal);if (diffuse < 0.0f) diffuse = 0.0f; - specular = DPSOFTRAST_Vector3Dot(surfacenormal, specularnormal);if (specular < 0.0f) specular = 0.0f; + specularnormal[0] = lightnormal[0] + eyenormal[0]; + specularnormal[1] = lightnormal[1] + eyenormal[1]; + specularnormal[2] = lightnormal[2] + eyenormal[2]; + DPSOFTRAST_Vector3Normalize(specularnormal); + + specular = DPSOFTRAST_Vector3Dot(surfacenormal, specularnormal);if (specular < 0.0f) specular = 0.0f; + } specular = pow(specular, SpecularPower * glosstex[3]); + if (thread->shader_permutation & SHADERPERMUTATION_CUBEFILTER) { // scale down the attenuation to account for the cubefilter multiplying everything by 255 @@ -4039,16 +4412,88 @@ void DPSOFTRAST_PixelShader_LightSource(DPSOFTRAST_State_Thread *thread, const D void DPSOFTRAST_VertexShader_Refraction(void) { + DPSOFTRAST_Array_Transform(DPSOFTRAST_ARRAY_TEXCOORD4, DPSOFTRAST_ARRAY_POSITION, dpsoftrast.uniform4f + 4*DPSOFTRAST_UNIFORM_ModelViewProjectionMatrixM1); + DPSOFTRAST_Array_Transform(DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD0, dpsoftrast.uniform4f + 4*DPSOFTRAST_UNIFORM_TexMatrixM1); DPSOFTRAST_Array_TransformProject(DPSOFTRAST_ARRAY_POSITION, DPSOFTRAST_ARRAY_POSITION, dpsoftrast.uniform4f + 4*DPSOFTRAST_UNIFORM_ModelViewProjectionMatrixM1); } void DPSOFTRAST_PixelShader_Refraction(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span) { - // TODO: IMPLEMENT float buffer_z[DPSOFTRAST_DRAW_MAXSPANLENGTH]; + float z; + int x, startx = span->startx, endx = span->endx; + + // texture reads + unsigned char buffer_texture_normalbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; unsigned char buffer_FragColorbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; + + // varyings + float ModelViewProjectionPositiondata[4]; + float ModelViewProjectionPositionslope[4]; + + // uniforms + float ScreenScaleRefractReflect[2]; + float ScreenCenterRefractReflect[2]; + float DistortScaleRefractReflect[2]; + float RefractColor[4]; + + DPSOFTRAST_Texture *texture = thread->texbound[GL20TU_REFRACTION]; + if(!texture) return; + + // read textures DPSOFTRAST_Draw_Span_Begin(thread, triangle, span, buffer_z); - memset(buffer_FragColorbgra8, 0, span->length*4); + DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_normalbgra8, GL20TU_NORMAL, DPSOFTRAST_ARRAY_TEXCOORD0, buffer_z); + + // read varyings + DPSOFTRAST_CALCATTRIB4F(triangle, span, ModelViewProjectionPositiondata, ModelViewProjectionPositionslope, DPSOFTRAST_ARRAY_TEXCOORD4); + + // read uniforms + ScreenScaleRefractReflect[0] = thread->uniform4f[DPSOFTRAST_UNIFORM_ScreenScaleRefractReflect*4+0]; + ScreenScaleRefractReflect[1] = thread->uniform4f[DPSOFTRAST_UNIFORM_ScreenScaleRefractReflect*4+1]; + ScreenCenterRefractReflect[0] = thread->uniform4f[DPSOFTRAST_UNIFORM_ScreenCenterRefractReflect*4+0]; + ScreenCenterRefractReflect[1] = thread->uniform4f[DPSOFTRAST_UNIFORM_ScreenCenterRefractReflect*4+1]; + DistortScaleRefractReflect[0] = thread->uniform4f[DPSOFTRAST_UNIFORM_DistortScaleRefractReflect*4+0]; + DistortScaleRefractReflect[1] = thread->uniform4f[DPSOFTRAST_UNIFORM_DistortScaleRefractReflect*4+1]; + RefractColor[0] = thread->uniform4f[DPSOFTRAST_UNIFORM_RefractColor*4+2]; + RefractColor[1] = thread->uniform4f[DPSOFTRAST_UNIFORM_RefractColor*4+1]; + RefractColor[2] = thread->uniform4f[DPSOFTRAST_UNIFORM_RefractColor*4+0]; + RefractColor[3] = thread->uniform4f[DPSOFTRAST_UNIFORM_RefractColor*4+3]; + + // do stuff + for (x = startx;x < endx;x++) + { + float SafeScreenTexCoord[2]; + float ScreenTexCoord[2]; + float v[3]; + float iw; + unsigned char c[4]; + + z = buffer_z[x]; + + // " vec2 ScreenScaleRefractReflectIW = ScreenScaleRefractReflect.xy * (1.0 / ModelViewProjectionPosition.w);\n" + iw = 1.0f / (ModelViewProjectionPositiondata[3] + ModelViewProjectionPositionslope[3]*x); // / z + + // " vec2 SafeScreenTexCoord = ModelViewProjectionPosition.xy * ScreenScaleRefractReflectIW + ScreenCenterRefractReflect.xy;\n" + SafeScreenTexCoord[0] = (ModelViewProjectionPositiondata[0] + ModelViewProjectionPositionslope[0]*x) * iw * ScreenScaleRefractReflect[0] + ScreenCenterRefractReflect[0]; // * z (disappears) + SafeScreenTexCoord[1] = (ModelViewProjectionPositiondata[1] + ModelViewProjectionPositionslope[1]*x) * iw * ScreenScaleRefractReflect[1] + ScreenCenterRefractReflect[1]; // * z (disappears) + + // " vec2 ScreenTexCoord = SafeScreenTexCoord + vec3(normalize(myhalf3(dp_texture2D(Texture_Normal, TexCoord)) - myhalf3(0.5))).xy * DistortScaleRefractReflect.zw;\n" + v[0] = buffer_texture_normalbgra8[x*4+2] * (1.0f / 128.0f) - 1.0f; + v[1] = buffer_texture_normalbgra8[x*4+1] * (1.0f / 128.0f) - 1.0f; + v[2] = buffer_texture_normalbgra8[x*4+0] * (1.0f / 128.0f) - 1.0f; + DPSOFTRAST_Vector3Normalize(v); + ScreenTexCoord[0] = SafeScreenTexCoord[0] + v[0] * DistortScaleRefractReflect[0]; + ScreenTexCoord[1] = SafeScreenTexCoord[1] + v[1] * DistortScaleRefractReflect[1]; + + // " dp_FragColor = vec4(dp_texture2D(Texture_Refraction, ScreenTexCoord).rgb, 1.0) * RefractColor;\n" + DPSOFTRAST_Texture2DBGRA8(texture, 0, ScreenTexCoord[0], ScreenTexCoord[1], c); + + buffer_FragColorbgra8[x*4+0] = c[0] * RefractColor[0]; + buffer_FragColorbgra8[x*4+1] = c[1] * RefractColor[1]; + buffer_FragColorbgra8[x*4+2] = c[2] * RefractColor[2]; + buffer_FragColorbgra8[x*4+3] = min(RefractColor[3] * 256, 255); + } + DPSOFTRAST_Draw_Span_FinishBGRA8(thread, triangle, span, buffer_FragColorbgra8); } @@ -4056,17 +4501,167 @@ void DPSOFTRAST_PixelShader_Refraction(DPSOFTRAST_State_Thread *thread, const DP void DPSOFTRAST_VertexShader_Water(void) { - DPSOFTRAST_Array_TransformProject(DPSOFTRAST_ARRAY_POSITION, DPSOFTRAST_ARRAY_POSITION, dpsoftrast.uniform4f + 4*DPSOFTRAST_UNIFORM_ModelViewProjectionMatrixM1); + int i; + int numvertices = dpsoftrast.numvertices; + float EyePosition[4]; + float EyeVectorModelSpace[4]; + float EyeVector[4]; + float position[4]; + float svector[4]; + float tvector[4]; + float normal[4]; + EyePosition[0] = dpsoftrast.uniform4f[DPSOFTRAST_UNIFORM_EyePosition*4+0]; + EyePosition[1] = dpsoftrast.uniform4f[DPSOFTRAST_UNIFORM_EyePosition*4+1]; + EyePosition[2] = dpsoftrast.uniform4f[DPSOFTRAST_UNIFORM_EyePosition*4+2]; + EyePosition[3] = dpsoftrast.uniform4f[DPSOFTRAST_UNIFORM_EyePosition*4+3]; + DPSOFTRAST_Array_Load(DPSOFTRAST_ARRAY_POSITION, DPSOFTRAST_ARRAY_POSITION); + DPSOFTRAST_Array_Load(DPSOFTRAST_ARRAY_TEXCOORD1, DPSOFTRAST_ARRAY_TEXCOORD1); + DPSOFTRAST_Array_Load(DPSOFTRAST_ARRAY_TEXCOORD2, DPSOFTRAST_ARRAY_TEXCOORD2); + DPSOFTRAST_Array_Load(DPSOFTRAST_ARRAY_TEXCOORD3, DPSOFTRAST_ARRAY_TEXCOORD3); + for (i = 0;i < numvertices;i++) + { + position[0] = dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_POSITION][i*4+0]; + position[1] = dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_POSITION][i*4+1]; + position[2] = dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_POSITION][i*4+2]; + svector[0] = dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD1][i*4+0]; + svector[1] = dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD1][i*4+1]; + svector[2] = dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD1][i*4+2]; + tvector[0] = dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD2][i*4+0]; + tvector[1] = dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD2][i*4+1]; + tvector[2] = dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD2][i*4+2]; + normal[0] = dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD3][i*4+0]; + normal[1] = dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD3][i*4+1]; + normal[2] = dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD3][i*4+2]; + EyeVectorModelSpace[0] = EyePosition[0] - position[0]; + EyeVectorModelSpace[1] = EyePosition[1] - position[1]; + EyeVectorModelSpace[2] = EyePosition[2] - position[2]; + EyeVector[0] = svector[0] * EyeVectorModelSpace[0] + svector[1] * EyeVectorModelSpace[1] + svector[2] * EyeVectorModelSpace[2]; + EyeVector[1] = tvector[0] * EyeVectorModelSpace[0] + tvector[1] * EyeVectorModelSpace[1] + tvector[2] * EyeVectorModelSpace[2]; + EyeVector[2] = normal[0] * EyeVectorModelSpace[0] + normal[1] * EyeVectorModelSpace[1] + normal[2] * EyeVectorModelSpace[2]; + dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD6][i*4+0] = EyeVector[0]; + dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD6][i*4+1] = EyeVector[1]; + dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD6][i*4+2] = EyeVector[2]; + dpsoftrast.post_array4f[DPSOFTRAST_ARRAY_TEXCOORD6][i*4+3] = 0.0f; + } + DPSOFTRAST_Array_TransformProject(DPSOFTRAST_ARRAY_POSITION, -1, dpsoftrast.uniform4f + 4*DPSOFTRAST_UNIFORM_ModelViewProjectionMatrixM1); + DPSOFTRAST_Array_Transform(DPSOFTRAST_ARRAY_TEXCOORD4, DPSOFTRAST_ARRAY_POSITION, dpsoftrast.uniform4f + 4*DPSOFTRAST_UNIFORM_ModelViewProjectionMatrixM1); + DPSOFTRAST_Array_Transform(DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD0, dpsoftrast.uniform4f + 4*DPSOFTRAST_UNIFORM_TexMatrixM1); } void DPSOFTRAST_PixelShader_Water(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Triangle * RESTRICT triangle, const DPSOFTRAST_State_Span * RESTRICT span) { - // TODO: IMPLEMENT float buffer_z[DPSOFTRAST_DRAW_MAXSPANLENGTH]; + float z; + int x, startx = span->startx, endx = span->endx; + + // texture reads + unsigned char buffer_texture_normalbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; unsigned char buffer_FragColorbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; + + // varyings + float ModelViewProjectionPositiondata[4]; + float ModelViewProjectionPositionslope[4]; + float EyeVectordata[4]; + float EyeVectorslope[4]; + + // uniforms + float ScreenScaleRefractReflect[4]; + float ScreenCenterRefractReflect[4]; + float DistortScaleRefractReflect[4]; + float RefractColor[4]; + float ReflectColor[4]; + float ReflectFactor; + float ReflectOffset; + + DPSOFTRAST_Texture *texture_refraction = thread->texbound[GL20TU_REFRACTION]; + DPSOFTRAST_Texture *texture_reflection = thread->texbound[GL20TU_REFLECTION]; + if(!texture_refraction || !texture_reflection) return; + + // read textures DPSOFTRAST_Draw_Span_Begin(thread, triangle, span, buffer_z); - memset(buffer_FragColorbgra8, 0, span->length*4); + DPSOFTRAST_Draw_Span_Texture2DVaryingBGRA8(thread, triangle, span, buffer_texture_normalbgra8, GL20TU_NORMAL, DPSOFTRAST_ARRAY_TEXCOORD0, buffer_z); + + // read varyings + DPSOFTRAST_CALCATTRIB4F(triangle, span, ModelViewProjectionPositiondata, ModelViewProjectionPositionslope, DPSOFTRAST_ARRAY_TEXCOORD4); + DPSOFTRAST_CALCATTRIB4F(triangle, span, EyeVectordata, EyeVectorslope, DPSOFTRAST_ARRAY_TEXCOORD6); + + // read uniforms + ScreenScaleRefractReflect[0] = thread->uniform4f[DPSOFTRAST_UNIFORM_ScreenScaleRefractReflect*4+0]; + ScreenScaleRefractReflect[1] = thread->uniform4f[DPSOFTRAST_UNIFORM_ScreenScaleRefractReflect*4+1]; + ScreenScaleRefractReflect[2] = thread->uniform4f[DPSOFTRAST_UNIFORM_ScreenScaleRefractReflect*4+2]; + ScreenScaleRefractReflect[3] = thread->uniform4f[DPSOFTRAST_UNIFORM_ScreenScaleRefractReflect*4+3]; + ScreenCenterRefractReflect[0] = thread->uniform4f[DPSOFTRAST_UNIFORM_ScreenCenterRefractReflect*4+0]; + ScreenCenterRefractReflect[1] = thread->uniform4f[DPSOFTRAST_UNIFORM_ScreenCenterRefractReflect*4+1]; + ScreenCenterRefractReflect[2] = thread->uniform4f[DPSOFTRAST_UNIFORM_ScreenCenterRefractReflect*4+2]; + ScreenCenterRefractReflect[3] = thread->uniform4f[DPSOFTRAST_UNIFORM_ScreenCenterRefractReflect*4+3]; + DistortScaleRefractReflect[0] = thread->uniform4f[DPSOFTRAST_UNIFORM_DistortScaleRefractReflect*4+0]; + DistortScaleRefractReflect[1] = thread->uniform4f[DPSOFTRAST_UNIFORM_DistortScaleRefractReflect*4+1]; + DistortScaleRefractReflect[2] = thread->uniform4f[DPSOFTRAST_UNIFORM_DistortScaleRefractReflect*4+2]; + DistortScaleRefractReflect[3] = thread->uniform4f[DPSOFTRAST_UNIFORM_DistortScaleRefractReflect*4+3]; + RefractColor[0] = thread->uniform4f[DPSOFTRAST_UNIFORM_RefractColor*4+2]; + RefractColor[1] = thread->uniform4f[DPSOFTRAST_UNIFORM_RefractColor*4+1]; + RefractColor[2] = thread->uniform4f[DPSOFTRAST_UNIFORM_RefractColor*4+0]; + RefractColor[3] = thread->uniform4f[DPSOFTRAST_UNIFORM_RefractColor*4+3]; + ReflectColor[0] = thread->uniform4f[DPSOFTRAST_UNIFORM_ReflectColor*4+2]; + ReflectColor[1] = thread->uniform4f[DPSOFTRAST_UNIFORM_ReflectColor*4+1]; + ReflectColor[2] = thread->uniform4f[DPSOFTRAST_UNIFORM_ReflectColor*4+0]; + ReflectColor[3] = thread->uniform4f[DPSOFTRAST_UNIFORM_ReflectColor*4+3]; + ReflectFactor = thread->uniform4f[DPSOFTRAST_UNIFORM_ReflectFactor*4+0]; + ReflectOffset = thread->uniform4f[DPSOFTRAST_UNIFORM_ReflectOffset*4+0]; + + // do stuff + for (x = startx;x < endx;x++) + { + float SafeScreenTexCoord[4]; + float ScreenTexCoord[4]; + float v[3]; + float iw; + unsigned char c1[4]; + unsigned char c2[4]; + float Fresnel; + + z = buffer_z[x]; + + // " vec4 ScreenScaleRefractReflectIW = ScreenScaleRefractReflect * (1.0 / ModelViewProjectionPosition.w);\n" + iw = 1.0f / (ModelViewProjectionPositiondata[3] + ModelViewProjectionPositionslope[3]*x); // / z + + // " vec4 SafeScreenTexCoord = ModelViewProjectionPosition.xyxy * ScreenScaleRefractReflectIW + ScreenCenterRefractReflect;\n" + SafeScreenTexCoord[0] = (ModelViewProjectionPositiondata[0] + ModelViewProjectionPositionslope[0]*x) * iw * ScreenScaleRefractReflect[0] + ScreenCenterRefractReflect[0]; // * z (disappears) + SafeScreenTexCoord[1] = (ModelViewProjectionPositiondata[1] + ModelViewProjectionPositionslope[1]*x) * iw * ScreenScaleRefractReflect[1] + ScreenCenterRefractReflect[1]; // * z (disappears) + SafeScreenTexCoord[2] = (ModelViewProjectionPositiondata[0] + ModelViewProjectionPositionslope[0]*x) * iw * ScreenScaleRefractReflect[2] + ScreenCenterRefractReflect[2]; // * z (disappears) + SafeScreenTexCoord[3] = (ModelViewProjectionPositiondata[1] + ModelViewProjectionPositionslope[1]*x) * iw * ScreenScaleRefractReflect[3] + ScreenCenterRefractReflect[3]; // * z (disappears) + + // " vec4 ScreenTexCoord = SafeScreenTexCoord + vec2(normalize(vec3(dp_texture2D(Texture_Normal, TexCoord)) - vec3(0.5))).xyxy * DistortScaleRefractReflect;\n" + v[0] = buffer_texture_normalbgra8[x*4+2] * (1.0f / 128.0f) - 1.0f; + v[1] = buffer_texture_normalbgra8[x*4+1] * (1.0f / 128.0f) - 1.0f; + v[2] = buffer_texture_normalbgra8[x*4+0] * (1.0f / 128.0f) - 1.0f; + DPSOFTRAST_Vector3Normalize(v); + ScreenTexCoord[0] = SafeScreenTexCoord[0] + v[0] * DistortScaleRefractReflect[0]; + ScreenTexCoord[1] = SafeScreenTexCoord[1] + v[1] * DistortScaleRefractReflect[1]; + ScreenTexCoord[2] = SafeScreenTexCoord[2] + v[0] * DistortScaleRefractReflect[2]; + ScreenTexCoord[3] = SafeScreenTexCoord[3] + v[1] * DistortScaleRefractReflect[3]; + + // " float Fresnel = pow(min(1.0, 1.0 - float(normalize(EyeVector).z)), 2.0) * ReflectFactor + ReflectOffset;\n" + v[0] = (EyeVectordata[0] + EyeVectorslope[0] * x); // * z (disappears) + v[1] = (EyeVectordata[1] + EyeVectorslope[1] * x); // * z (disappears) + v[2] = (EyeVectordata[2] + EyeVectorslope[2] * x); // * z (disappears) + DPSOFTRAST_Vector3Normalize(v); + Fresnel = 1.0f - v[2]; + Fresnel = min(1.0f, Fresnel); + Fresnel = Fresnel * Fresnel * ReflectFactor + ReflectOffset; + + // " dp_FragColor = vec4(dp_texture2D(Texture_Refraction, ScreenTexCoord).rgb, 1.0) * RefractColor;\n" + // " dp_FragColor = mix(vec4(dp_texture2D(Texture_Refraction, ScreenTexCoord.xy).rgb, 1) * RefractColor, vec4(dp_texture2D(Texture_Reflection, ScreenTexCoord.zw).rgb, 1) * ReflectColor, Fresnel);\n" + DPSOFTRAST_Texture2DBGRA8(texture_refraction, 0, ScreenTexCoord[0], ScreenTexCoord[1], c1); + DPSOFTRAST_Texture2DBGRA8(texture_reflection, 0, ScreenTexCoord[2], ScreenTexCoord[3], c2); + + buffer_FragColorbgra8[x*4+0] = (c1[0] * RefractColor[0]) * (1.0f - Fresnel) + (c2[0] * ReflectColor[0]) * Fresnel; + buffer_FragColorbgra8[x*4+1] = (c1[1] * RefractColor[1]) * (1.0f - Fresnel) + (c2[1] * ReflectColor[1]) * Fresnel; + buffer_FragColorbgra8[x*4+2] = (c1[2] * RefractColor[2]) * (1.0f - Fresnel) + (c2[2] * ReflectColor[2]) * Fresnel; + buffer_FragColorbgra8[x*4+3] = min(( RefractColor[3] * (1.0f - Fresnel) + ReflectColor[3] * Fresnel) * 256, 255); + } + DPSOFTRAST_Draw_Span_FinishBGRA8(thread, triangle, span, buffer_FragColorbgra8); } @@ -4083,7 +4678,7 @@ void DPSOFTRAST_PixelShader_ShowDepth(DPSOFTRAST_State_Thread *thread, const DPS float buffer_z[DPSOFTRAST_DRAW_MAXSPANLENGTH]; unsigned char buffer_FragColorbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; DPSOFTRAST_Draw_Span_Begin(thread, triangle, span, buffer_z); - memset(buffer_FragColorbgra8, 0, span->length*4); + memset(buffer_FragColorbgra8 + span->startx*4, 0, (span->endx - span->startx)*4); DPSOFTRAST_Draw_Span_FinishBGRA8(thread, triangle, span, buffer_FragColorbgra8); } @@ -4100,7 +4695,7 @@ void DPSOFTRAST_PixelShader_DeferredGeometry(DPSOFTRAST_State_Thread *thread, co float buffer_z[DPSOFTRAST_DRAW_MAXSPANLENGTH]; unsigned char buffer_FragColorbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; DPSOFTRAST_Draw_Span_Begin(thread, triangle, span, buffer_z); - memset(buffer_FragColorbgra8, 0, span->length*4); + memset(buffer_FragColorbgra8 + span->startx*4, 0, (span->endx - span->startx)*4); DPSOFTRAST_Draw_Span_FinishBGRA8(thread, triangle, span, buffer_FragColorbgra8); } @@ -4117,7 +4712,7 @@ void DPSOFTRAST_PixelShader_DeferredLightSource(DPSOFTRAST_State_Thread *thread, float buffer_z[DPSOFTRAST_DRAW_MAXSPANLENGTH]; unsigned char buffer_FragColorbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; DPSOFTRAST_Draw_Span_Begin(thread, triangle, span, buffer_z); - memset(buffer_FragColorbgra8, 0, span->length*4); + memset(buffer_FragColorbgra8 + span->startx*4, 0, (span->endx - span->startx)*4); DPSOFTRAST_Draw_Span_FinishBGRA8(thread, triangle, span, buffer_FragColorbgra8); } @@ -4141,95 +4736,101 @@ static const DPSOFTRAST_ShaderModeInfo DPSOFTRAST_ShaderModeTable[SHADERMODE_COU {2, DPSOFTRAST_VertexShader_FlatColor, DPSOFTRAST_PixelShader_FlatColor, {DPSOFTRAST_ARRAY_TEXCOORD0, ~0}, {GL20TU_COLOR, ~0}}, {2, DPSOFTRAST_VertexShader_VertexColor, DPSOFTRAST_PixelShader_VertexColor, {DPSOFTRAST_ARRAY_COLOR, DPSOFTRAST_ARRAY_TEXCOORD0, ~0}, {GL20TU_COLOR, ~0}}, {2, DPSOFTRAST_VertexShader_Lightmap, DPSOFTRAST_PixelShader_Lightmap, {DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD4, ~0}, {GL20TU_COLOR, GL20TU_LIGHTMAP, GL20TU_GLOW, ~0}}, - {2, DPSOFTRAST_VertexShader_FakeLight, DPSOFTRAST_PixelShader_FakeLight, {~0}, {~0}}, - {2, DPSOFTRAST_VertexShader_LightDirectionMap_ModelSpace, DPSOFTRAST_PixelShader_LightDirectionMap_ModelSpace, {DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD4, ~0}, {GL20TU_COLOR, GL20TU_LIGHTMAP, GL20TU_GLOW, ~0}}, - {2, DPSOFTRAST_VertexShader_LightDirectionMap_TangentSpace, DPSOFTRAST_PixelShader_LightDirectionMap_TangentSpace, {DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD4, ~0}, {GL20TU_COLOR, GL20TU_LIGHTMAP, GL20TU_GLOW, ~0}}, - {2, DPSOFTRAST_VertexShader_LightDirection, DPSOFTRAST_PixelShader_LightDirection, {DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD1, DPSOFTRAST_ARRAY_TEXCOORD2, DPSOFTRAST_ARRAY_TEXCOORD3, ~0}, {GL20TU_COLOR, GL20TU_PANTS, GL20TU_SHIRT, GL20TU_GLOW, GL20TU_NORMAL, GL20TU_GLOSS, ~0}}, + {2, DPSOFTRAST_VertexShader_FakeLight, DPSOFTRAST_PixelShader_FakeLight, {DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD1, DPSOFTRAST_ARRAY_TEXCOORD2, DPSOFTRAST_ARRAY_TEXCOORD3, DPSOFTRAST_ARRAY_TEXCOORD5, DPSOFTRAST_ARRAY_TEXCOORD6, ~0}, {GL20TU_COLOR, GL20TU_PANTS, GL20TU_SHIRT, GL20TU_GLOW, GL20TU_NORMAL, GL20TU_GLOSS, ~0}}, + {2, DPSOFTRAST_VertexShader_LightDirectionMap_ModelSpace, DPSOFTRAST_PixelShader_LightDirectionMap_ModelSpace, {DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD1, DPSOFTRAST_ARRAY_TEXCOORD2, DPSOFTRAST_ARRAY_TEXCOORD3, DPSOFTRAST_ARRAY_TEXCOORD4, DPSOFTRAST_ARRAY_TEXCOORD5, DPSOFTRAST_ARRAY_TEXCOORD6, ~0}, {GL20TU_COLOR, GL20TU_PANTS, GL20TU_SHIRT, GL20TU_GLOW, GL20TU_NORMAL, GL20TU_GLOSS, GL20TU_LIGHTMAP, GL20TU_DELUXEMAP, ~0}}, + {2, DPSOFTRAST_VertexShader_LightDirectionMap_TangentSpace, DPSOFTRAST_PixelShader_LightDirectionMap_TangentSpace, {DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD1, DPSOFTRAST_ARRAY_TEXCOORD2, DPSOFTRAST_ARRAY_TEXCOORD3, DPSOFTRAST_ARRAY_TEXCOORD4, DPSOFTRAST_ARRAY_TEXCOORD5, DPSOFTRAST_ARRAY_TEXCOORD6, ~0}, {GL20TU_COLOR, GL20TU_PANTS, GL20TU_SHIRT, GL20TU_GLOW, GL20TU_NORMAL, GL20TU_GLOSS, GL20TU_LIGHTMAP, GL20TU_DELUXEMAP, ~0}}, + {2, DPSOFTRAST_VertexShader_LightDirection, DPSOFTRAST_PixelShader_LightDirection, {DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD1, DPSOFTRAST_ARRAY_TEXCOORD2, DPSOFTRAST_ARRAY_TEXCOORD3, DPSOFTRAST_ARRAY_TEXCOORD5, DPSOFTRAST_ARRAY_TEXCOORD6, ~0}, {GL20TU_COLOR, GL20TU_PANTS, GL20TU_SHIRT, GL20TU_GLOW, GL20TU_NORMAL, GL20TU_GLOSS, ~0}}, {2, DPSOFTRAST_VertexShader_LightSource, DPSOFTRAST_PixelShader_LightSource, {DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD1, DPSOFTRAST_ARRAY_TEXCOORD2, DPSOFTRAST_ARRAY_TEXCOORD3, DPSOFTRAST_ARRAY_TEXCOORD4, ~0}, {GL20TU_COLOR, GL20TU_PANTS, GL20TU_SHIRT, GL20TU_GLOW, GL20TU_NORMAL, GL20TU_GLOSS, GL20TU_CUBE, ~0}}, - {2, DPSOFTRAST_VertexShader_Refraction, DPSOFTRAST_PixelShader_Refraction, {~0}}, - {2, DPSOFTRAST_VertexShader_Water, DPSOFTRAST_PixelShader_Water, {~0}}, + {2, DPSOFTRAST_VertexShader_Refraction, DPSOFTRAST_PixelShader_Refraction, {DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD4, ~0}, {GL20TU_NORMAL, GL20TU_REFRACTION, ~0}}, + {2, DPSOFTRAST_VertexShader_Water, DPSOFTRAST_PixelShader_Water, {DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD1, DPSOFTRAST_ARRAY_TEXCOORD2, DPSOFTRAST_ARRAY_TEXCOORD3, DPSOFTRAST_ARRAY_TEXCOORD4, DPSOFTRAST_ARRAY_TEXCOORD6, ~0}, {GL20TU_NORMAL, GL20TU_REFLECTION, GL20TU_REFRACTION, ~0}}, {2, DPSOFTRAST_VertexShader_ShowDepth, DPSOFTRAST_PixelShader_ShowDepth, {~0}}, {2, DPSOFTRAST_VertexShader_DeferredGeometry, DPSOFTRAST_PixelShader_DeferredGeometry, {~0}}, - {2, DPSOFTRAST_VertexShader_DeferredLightSource, DPSOFTRAST_PixelShader_DeferredLightSource, {~0}} + {2, DPSOFTRAST_VertexShader_DeferredLightSource, DPSOFTRAST_PixelShader_DeferredLightSource, {~0}}, }; -void DPSOFTRAST_Draw_ProcessSpans(DPSOFTRAST_State_Thread *thread) +static void DPSOFTRAST_Draw_DepthTest(DPSOFTRAST_State_Thread *thread, DPSOFTRAST_State_Span *span) { - int i; int x; int startx; int endx; -// unsigned int c; -// unsigned int *colorpixel; unsigned int *depthpixel; - float w; - float wslope; int depth; int depthslope; unsigned int d; + unsigned char *pixelmask; + DPSOFTRAST_State_Triangle *triangle; + triangle = &thread->triangles[span->triangle]; + depthpixel = dpsoftrast.fb_depthpixels + span->y * dpsoftrast.fb_width + span->x; + startx = span->startx; + endx = span->endx; + depth = span->depthbase; + depthslope = span->depthslope; + pixelmask = thread->pixelmaskarray; + if (thread->depthtest && dpsoftrast.fb_depthpixels) + { + switch(thread->fb_depthfunc) + { + default: + case GL_ALWAYS: for (x = startx, d = depth + depthslope*startx;x < endx;x++, d += depthslope) pixelmask[x] = true; break; + case GL_LESS: for (x = startx, d = depth + depthslope*startx;x < endx;x++, d += depthslope) pixelmask[x] = depthpixel[x] < d; break; + case GL_LEQUAL: for (x = startx, d = depth + depthslope*startx;x < endx;x++, d += depthslope) pixelmask[x] = depthpixel[x] <= d; break; + case GL_EQUAL: for (x = startx, d = depth + depthslope*startx;x < endx;x++, d += depthslope) pixelmask[x] = depthpixel[x] == d; break; + case GL_GEQUAL: for (x = startx, d = depth + depthslope*startx;x < endx;x++, d += depthslope) pixelmask[x] = depthpixel[x] >= d; break; + case GL_GREATER: for (x = startx, d = depth + depthslope*startx;x < endx;x++, d += depthslope) pixelmask[x] = depthpixel[x] > d; break; + case GL_NEVER: for (x = startx, d = depth + depthslope*startx;x < endx;x++, d += depthslope) pixelmask[x] = false; break; + } + while (startx < endx && !pixelmask[startx]) + startx++; + while (endx > startx && !pixelmask[endx-1]) + endx--; + } + else + { + // no depth testing means we're just dealing with color... + memset(pixelmask + startx, 1, endx - startx); + } + span->pixelmask = pixelmask; + span->startx = startx; + span->endx = endx; +} + +static void DPSOFTRAST_Draw_DepthWrite(const DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_State_Span *span) +{ + int x, d, depth, depthslope, startx, endx; + const unsigned char *pixelmask; + unsigned int *depthpixel; + if (thread->depthmask && thread->depthtest && dpsoftrast.fb_depthpixels) + { + depth = span->depthbase; + depthslope = span->depthslope; + pixelmask = span->pixelmask; + startx = span->startx; + endx = span->endx; + depthpixel = dpsoftrast.fb_depthpixels + span->y * dpsoftrast.fb_width + span->x; + for (x = startx, d = depth + depthslope*startx;x < endx;x++, d += depthslope) + if (pixelmask[x]) + depthpixel[x] = d; + } +} + +void DPSOFTRAST_Draw_ProcessSpans(DPSOFTRAST_State_Thread *thread) +{ + int i; DPSOFTRAST_State_Triangle *triangle; DPSOFTRAST_State_Span *span; - unsigned char pixelmask[DPSOFTRAST_DRAW_MAXSPANLENGTH]; for (i = 0; i < thread->numspans; i++) { span = &thread->spans[i]; triangle = &thread->triangles[span->triangle]; - if (thread->depthtest && dpsoftrast.fb_depthpixels) - { - wslope = triangle->w[0]; - w = triangle->w[2] + span->x*wslope + span->y*triangle->w[1]; - depthslope = (int)(wslope*DPSOFTRAST_DEPTHSCALE); - depth = (int)(w*DPSOFTRAST_DEPTHSCALE - DPSOFTRAST_DEPTHOFFSET*(thread->polygonoffset[1] + fabs(wslope)*thread->polygonoffset[0])); - depthpixel = dpsoftrast.fb_depthpixels + span->y * dpsoftrast.fb_width + span->x; - switch(thread->fb_depthfunc) - { - default: - case GL_ALWAYS: for (x = 0, d = depth;x < span->length;x++, d += depthslope) pixelmask[x] = true; break; - case GL_LESS: for (x = 0, d = depth;x < span->length;x++, d += depthslope) pixelmask[x] = depthpixel[x] < d; break; - case GL_LEQUAL: for (x = 0, d = depth;x < span->length;x++, d += depthslope) pixelmask[x] = depthpixel[x] <= d; break; - case GL_EQUAL: for (x = 0, d = depth;x < span->length;x++, d += depthslope) pixelmask[x] = depthpixel[x] == d; break; - case GL_GEQUAL: for (x = 0, d = depth;x < span->length;x++, d += depthslope) pixelmask[x] = depthpixel[x] >= d; break; - case GL_GREATER: for (x = 0, d = depth;x < span->length;x++, d += depthslope) pixelmask[x] = depthpixel[x] > d; break; - case GL_NEVER: for (x = 0, d = depth;x < span->length;x++, d += depthslope) pixelmask[x] = false; break; - } - //colorpixel = dpsoftrast.fb_colorpixels[0] + (span->y * dpsoftrast.fb_width + span->x) * 4;; - //for (x = 0;x < span->length;x++) - // colorpixel[x] = (depthpixel[x] & 0xFF000000) ? (0x00FF0000) : (depthpixel[x] & 0x00FF0000); - // if there is no color buffer, skip pixel shader - startx = 0; - endx = span->length; - while (startx < endx && !pixelmask[startx]) - startx++; - while (endx > startx && !pixelmask[endx-1]) - endx--; - if (startx >= endx) - continue; // no pixels to fill - span->pixelmask = pixelmask; - span->startx = startx; - span->endx = endx; - // run pixel shader if appropriate - // do this before running depthmask code, to allow the pixelshader - // to clear pixelmask values for alpha testing - if (dpsoftrast.fb_colorpixels[0] && thread->fb_colormask) - DPSOFTRAST_ShaderModeTable[thread->shader_mode].Span(thread, triangle, span); - if (thread->depthmask) - for (x = startx, d = depth + depthslope*startx;x < endx;x++, d += depthslope) - if (pixelmask[x]) - depthpixel[x] = d; - } - else - { - // no depth testing means we're just dealing with color... - // if there is no color buffer, skip pixel shader - if (dpsoftrast.fb_colorpixels[0] && thread->fb_colormask) - { - memset(pixelmask, 1, span->length); - span->pixelmask = pixelmask; - span->startx = 0; - span->endx = span->length; - DPSOFTRAST_ShaderModeTable[thread->shader_mode].Span(thread, triangle, span); - } - } + DPSOFTRAST_Draw_DepthTest(thread, span); + if (span->startx >= span->endx) + continue; + // run pixel shader if appropriate + // do this before running depthmask code, to allow the pixelshader + // to clear pixelmask values for alpha testing + if (dpsoftrast.fb_colorpixels[0] && thread->fb_colormask) + DPSOFTRAST_ShaderModeTable[thread->shader_mode].Span(thread, triangle, span); + DPSOFTRAST_Draw_DepthWrite(thread, span); } thread->numspans = 0; } @@ -4238,11 +4839,10 @@ DEFCOMMAND(22, Draw, int datasize; int starty; int endy; ATOMIC_COUNTER refcount static void DPSOFTRAST_Interpret_Draw(DPSOFTRAST_State_Thread *thread, DPSOFTRAST_Command_Draw *command) { -#ifdef SSE2_PRESENT +#ifdef SSE_POSSIBLE int cullface = thread->cullface; - int width = dpsoftrast.fb_width; - int miny = (thread->index*dpsoftrast.fb_height)/dpsoftrast.numthreads; - int maxy = ((thread->index+1)*dpsoftrast.fb_height)/dpsoftrast.numthreads; + int minx, maxx, miny, maxy; + int miny1, maxy1, miny2, maxy2; __m128i fbmin, fbmax; __m128 viewportcenter, viewportscale; int firstvertex = command->firstvertex; @@ -4257,16 +4857,25 @@ static void DPSOFTRAST_Interpret_Draw(DPSOFTRAST_State_Thread *thread, DPSOFTRAS int y; int e[3]; __m128i screeny; - int starty, endy; + int starty, endy, bandy; int numpoints; int clipcase; float clipdist[4]; + float clip0origin, clip0slope; + int clip0dir; __m128 triangleedge1, triangleedge2, trianglenormal; __m128 clipfrac[3]; __m128 screen[4]; DPSOFTRAST_State_Triangle *triangle; DPSOFTRAST_Texture *texture; - if (command->starty >= maxy || command->endy <= miny) + DPSOFTRAST_ValidateQuick(thread, DPSOFTRAST_VALIDATE_DRAW); + miny = thread->fb_scissor[1]; + maxy = thread->fb_scissor[1] + thread->fb_scissor[3]; + miny1 = bound(miny, thread->miny1, maxy); + maxy1 = bound(miny, thread->maxy1, maxy); + miny2 = bound(miny, thread->miny2, maxy); + maxy2 = bound(miny, thread->maxy2, maxy); + if ((command->starty >= maxy1 || command->endy <= miny1) && (command->starty >= maxy2 || command->endy <= miny2)) { if (!ATOMIC_DECREMENT(command->refcount)) { @@ -4275,9 +4884,10 @@ static void DPSOFTRAST_Interpret_Draw(DPSOFTRAST_State_Thread *thread, DPSOFTRAS } return; } - DPSOFTRAST_ValidateQuick(thread, DPSOFTRAST_VALIDATE_DRAW); - fbmin = _mm_setr_epi16(0, miny, 0, miny, 0, miny, 0, miny); - fbmax = _mm_sub_epi16(_mm_setr_epi16(width, maxy, width, maxy, width, maxy, width, maxy), _mm_set1_epi16(1)); + minx = thread->fb_scissor[0]; + maxx = thread->fb_scissor[0] + thread->fb_scissor[2]; + fbmin = _mm_setr_epi16(minx, miny1, minx, miny1, minx, miny1, minx, miny1); + fbmax = _mm_sub_epi16(_mm_setr_epi16(maxx, maxy2, maxx, maxy2, maxx, maxy2, maxx, maxy2), _mm_set1_epi16(1)); viewportcenter = _mm_load_ps(thread->fb_viewportcenter); viewportscale = _mm_load_ps(thread->fb_viewportscale); screen[3] = _mm_setzero_ps(); @@ -4442,6 +5052,8 @@ static void DPSOFTRAST_Interpret_Draw(DPSOFTRAST_State_Thread *thread, DPSOFTRAS continue; starty = _mm_extract_epi16(screenmin, 1); endy = _mm_extract_epi16(screenmax, 1)+1; + if (starty >= maxy1 && endy <= miny2) + continue; screeny = _mm_srai_epi32(screeni, 16); } @@ -4475,6 +5087,43 @@ static void DPSOFTRAST_Interpret_Draw(DPSOFTRAST_State_Thread *thread, DPSOFTRAS _mm_store_ss(&triangle->w[0], attribxslope); _mm_store_ss(&triangle->w[1], attribyslope); _mm_store_ss(&triangle->w[2], attriborigin); + + clip0origin = 0; + clip0slope = 0; + clip0dir = 0; + if(thread->fb_clipplane[0] || thread->fb_clipplane[1] || thread->fb_clipplane[2]) + { + float cliporigin, clipxslope, clipyslope; + attriborigin = _mm_shuffle_ps(screen[1], screen[1], _MM_SHUFFLE(2, 2, 2, 2)); + attribedge1 = _mm_sub_ss(_mm_shuffle_ps(screen[0], screen[0], _MM_SHUFFLE(2, 2, 2, 2)), attriborigin); + attribedge2 = _mm_sub_ss(_mm_shuffle_ps(screen[2], screen[2], _MM_SHUFFLE(2, 2, 2, 2)), attriborigin); + attribxslope = _mm_sub_ss(_mm_mul_ss(attribuxslope, attribedge1), _mm_mul_ss(attribvxslope, attribedge2)); + attribyslope = _mm_sub_ss(_mm_mul_ss(attribvyslope, attribedge2), _mm_mul_ss(attribuyslope, attribedge1)); + attriborigin = _mm_sub_ss(attriborigin, _mm_add_ss(_mm_mul_ss(attribxslope, x1), _mm_mul_ss(attribyslope, y1))); + cliporigin = _mm_cvtss_f32(attriborigin)*thread->fb_clipplane[2] + thread->fb_clipplane[3]; + clipxslope = thread->fb_clipplane[0] + _mm_cvtss_f32(attribxslope)*thread->fb_clipplane[2]; + clipyslope = thread->fb_clipplane[1] + _mm_cvtss_f32(attribyslope)*thread->fb_clipplane[2]; + if(clipxslope != 0) + { + clip0origin = -cliporigin/clipxslope; + clip0slope = -clipyslope/clipxslope; + clip0dir = clipxslope > 0 ? 1 : -1; + } + else if(clipyslope > 0) + { + clip0origin = dpsoftrast.fb_width*floor(cliporigin/clipyslope); + clip0slope = dpsoftrast.fb_width; + clip0dir = -1; + } + else if(clipyslope < 0) + { + clip0origin = dpsoftrast.fb_width*ceil(cliporigin/clipyslope); + clip0slope = -dpsoftrast.fb_width; + clip0dir = -1; + } + else if(clip0origin < 0) continue; + } + mipedgescale = _mm_setzero_ps(); for (j = 0;j < DPSOFTRAST_ARRAY_TOTAL; j++) { @@ -4490,9 +5139,9 @@ static void DPSOFTRAST_Interpret_Draw(DPSOFTRAST_State_Thread *thread, DPSOFTRAS attribxslope = _mm_sub_ps(_mm_mul_ps(attribuxslope, attribedge1), _mm_mul_ps(attribvxslope, attribedge2)); attribyslope = _mm_sub_ps(_mm_mul_ps(attribvyslope, attribedge2), _mm_mul_ps(attribuyslope, attribedge1)); attriborigin = _mm_sub_ps(attriborigin, _mm_add_ps(_mm_mul_ps(attribxslope, x1), _mm_mul_ps(attribyslope, y1))); - _mm_stream_ps(triangle->attribs[k][0], attribxslope); - _mm_stream_ps(triangle->attribs[k][1], attribyslope); - _mm_stream_ps(triangle->attribs[k][2], attriborigin); + _mm_storeu_ps(triangle->attribs[k][0], attribxslope); + _mm_storeu_ps(triangle->attribs[k][1], attribyslope); + _mm_storeu_ps(triangle->attribs[k][2], attriborigin); if (k == DPSOFTRAST_ShaderModeTable[thread->shader_mode].lodarrayindex) { mipedgescale = _mm_movelh_ps(triangleedge1, triangleedge2); @@ -4527,14 +5176,17 @@ static void DPSOFTRAST_Interpret_Draw(DPSOFTRAST_State_Thread *thread, DPSOFTRAS } } } - - for (y = starty; y < endy;) + + for (y = starty, bandy = min(endy, maxy1); y < endy; bandy = min(endy, maxy2), y = max(y, miny2)) + for (; y < bandy;) { __m128 xcoords, xslope; __m128i ycc = _mm_cmpgt_epi32(_mm_set1_epi32(y), screeny); int yccmask = _mm_movemask_epi8(ycc); int edge0p, edge0n, edge1p, edge1n; int nexty; + float w, wslope; + float clip0; if (numpoints == 4) { switch(yccmask) @@ -4577,39 +5229,58 @@ static void DPSOFTRAST_Interpret_Draw(DPSOFTRAST_State_Thread *thread, DPSOFTRAS ycc = _mm_min_epi16(ycc, _mm_shuffle_epi32(ycc, _MM_SHUFFLE(1, 0, 3, 2))); ycc = _mm_min_epi16(ycc, _mm_shuffle_epi32(ycc, _MM_SHUFFLE(2, 3, 0, 1))); nexty = _mm_extract_epi16(ycc, 0); - if(nexty >= endy) nexty = endy-1; - if (_mm_ucomigt_ss(_mm_max_ss(screen[edge0n], screen[edge0p]), _mm_min_ss(screen[edge1n], screen[edge1p]))) - { - int tmp = edge0n; - edge0n = edge1n; - edge1n = tmp; - tmp = edge0p; - edge0p = edge1p; - edge1p = tmp; - } + if (nexty >= bandy) nexty = bandy-1; xslope = _mm_sub_ps(_mm_movelh_ps(screen[edge0n], screen[edge1n]), _mm_movelh_ps(screen[edge0p], screen[edge1p])); xslope = _mm_div_ps(xslope, _mm_shuffle_ps(xslope, xslope, _MM_SHUFFLE(3, 3, 1, 1))); xcoords = _mm_add_ps(_mm_movelh_ps(screen[edge0p], screen[edge1p]), _mm_mul_ps(xslope, _mm_sub_ps(_mm_set1_ps(y), _mm_shuffle_ps(screen[edge0p], screen[edge1p], _MM_SHUFFLE(1, 1, 1, 1))))); xcoords = _mm_add_ps(xcoords, _mm_set1_ps(0.5f)); - for(; y <= nexty; y++, xcoords = _mm_add_ps(xcoords, xslope)) + if (_mm_ucomigt_ss(xcoords, _mm_shuffle_ps(xcoords, xcoords, _MM_SHUFFLE(1, 0, 3, 2)))) + { + xcoords = _mm_shuffle_ps(xcoords, xcoords, _MM_SHUFFLE(1, 0, 3, 2)); + xslope = _mm_shuffle_ps(xslope, xslope, _MM_SHUFFLE(1, 0, 3, 2)); + } + clip0 = clip0origin + (y+0.5f)*clip0slope + 0.5f; + for(; y <= nexty; y++, xcoords = _mm_add_ps(xcoords, xslope), clip0 += clip0slope) { int startx, endx, offset; startx = _mm_cvtss_si32(xcoords); endx = _mm_cvtss_si32(_mm_movehl_ps(xcoords, xcoords)); - if (startx < 0) startx = 0; - if (endx > dpsoftrast.fb_width) endx = dpsoftrast.fb_width; + if (startx < minx) startx = minx; + if (endx > maxx) endx = maxx; if (startx >= endx) continue; - for (offset = startx; offset < endx;) + + if (clip0dir) + { + if (clip0dir > 0) + { + if (startx < clip0) + { + if(endx <= clip0) continue; + startx = (int)clip0; + } + } + else if (endx > clip0) + { + if(startx >= clip0) continue; + endx = (int)clip0; + } + } + + for (offset = startx; offset < endx;offset += DPSOFTRAST_DRAW_MAXSPANLENGTH) { DPSOFTRAST_State_Span *span = &thread->spans[thread->numspans]; span->triangle = thread->numtriangles; span->x = offset; span->y = y; - span->length = endx - offset; - if (span -> length > DPSOFTRAST_DRAW_MAXSPANLENGTH) - span -> length = DPSOFTRAST_DRAW_MAXSPANLENGTH; - offset += span->length; + span->startx = 0; + span->endx = min(endx - offset, DPSOFTRAST_DRAW_MAXSPANLENGTH); + if (span->startx >= span->endx) + continue; + wslope = triangle->w[0]; + w = triangle->w[2] + span->x*wslope + span->y*triangle->w[1]; + span->depthslope = (int)(wslope*DPSOFTRAST_DEPTHSCALE); + span->depthbase = (int)(w*DPSOFTRAST_DEPTHSCALE - DPSOFTRAST_DEPTHOFFSET*(thread->polygonoffset[1] + fabs(wslope)*thread->polygonoffset[0])); if (++thread->numspans >= DPSOFTRAST_DRAW_MAXSPANS) DPSOFTRAST_Draw_ProcessSpans(thread); } @@ -4717,23 +5388,46 @@ void DPSOFTRAST_DrawTriangles(int firstvertex, int numvertices, int numtriangles command->clipped = dpsoftrast.drawclipped; command->refcount = dpsoftrast.numthreads; -#ifdef USE_THREADS - DPSOFTRAST_Draw_SyncCommands(); + if (dpsoftrast.usethreads) { int i; - int nexty = 0; + DPSOFTRAST_Draw_SyncCommands(); for (i = 0; i < dpsoftrast.numthreads; i++) { DPSOFTRAST_State_Thread *thread = &dpsoftrast.threads[i]; - int y = nexty; - nexty = ((i+1)*dpsoftrast.fb_height)/dpsoftrast.numthreads; - if (command->starty < nexty && command->endy > y && thread->starving) - SDL_CondSignal(thread->drawcond); + if (((command->starty < thread->maxy1 && command->endy > thread->miny1) || (command->starty < thread->maxy2 && command->endy > thread->miny2)) && thread->starving) + Thread_CondSignal(thread->drawcond); } } -#else - DPSOFTRAST_Draw_FlushThreads(); -#endif + else + { + DPSOFTRAST_Draw_FlushThreads(); + } +} + +DEFCOMMAND(23, SetRenderTargets, int width; int height;); +static void DPSOFTRAST_Interpret_SetRenderTargets(DPSOFTRAST_State_Thread *thread, const DPSOFTRAST_Command_SetRenderTargets *command) +{ + thread->validate |= DPSOFTRAST_VALIDATE_FB; +} +void DPSOFTRAST_SetRenderTargets(int width, int height, unsigned int *depthpixels, unsigned int *colorpixels0, unsigned int *colorpixels1, unsigned int *colorpixels2, unsigned int *colorpixels3) +{ + DPSOFTRAST_Command_SetRenderTargets *command; + if (width != dpsoftrast.fb_width || height != dpsoftrast.fb_height || depthpixels != dpsoftrast.fb_depthpixels || + colorpixels0 != dpsoftrast.fb_colorpixels[0] || colorpixels1 != dpsoftrast.fb_colorpixels[1] || + colorpixels2 != dpsoftrast.fb_colorpixels[2] || colorpixels3 != dpsoftrast.fb_colorpixels[3]) + DPSOFTRAST_Flush(); + dpsoftrast.fb_width = width; + dpsoftrast.fb_height = height; + dpsoftrast.fb_depthpixels = depthpixels; + dpsoftrast.fb_colorpixels[0] = colorpixels0; + dpsoftrast.fb_colorpixels[1] = colorpixels1; + dpsoftrast.fb_colorpixels[2] = colorpixels2; + dpsoftrast.fb_colorpixels[3] = colorpixels3; + DPSOFTRAST_RecalcViewport(dpsoftrast.viewport, dpsoftrast.fb_viewportcenter, dpsoftrast.fb_viewportscale); + command = DPSOFTRAST_ALLOCATECOMMAND(SetRenderTargets); + command->width = width; + command->height = height; } static void DPSOFTRAST_Draw_InterpretCommands(DPSOFTRAST_State_Thread *thread, int endoffset) @@ -4765,13 +5459,13 @@ static void DPSOFTRAST_Draw_InterpretCommands(DPSOFTRAST_State_Thread *thread, i INTERPCOMMAND(DepthRange) INTERPCOMMAND(PolygonOffset) INTERPCOMMAND(CullFace) - INTERPCOMMAND(AlphaTest) - INTERPCOMMAND(AlphaFunc) INTERPCOMMAND(SetTexture) INTERPCOMMAND(SetShader) INTERPCOMMAND(Uniform4f) INTERPCOMMAND(UniformMatrix4f) INTERPCOMMAND(Uniform1i) + INTERPCOMMAND(SetRenderTargets) + INTERPCOMMAND(ClipPlane) case DPSOFTRAST_OPCODE_Draw: DPSOFTRAST_Interpret_Draw(thread, (DPSOFTRAST_Command_Draw *)command); @@ -4789,7 +5483,6 @@ static void DPSOFTRAST_Draw_InterpretCommands(DPSOFTRAST_State_Thread *thread, i thread->commandoffset = commandoffset; } -#ifdef USE_THREADS static int DPSOFTRAST_Draw_Thread(void *data) { DPSOFTRAST_State_Thread *thread = (DPSOFTRAST_State_Thread *)data; @@ -4801,58 +5494,62 @@ static int DPSOFTRAST_Draw_Thread(void *data) } else { - SDL_LockMutex(thread->drawmutex); + Thread_LockMutex(thread->drawmutex); if (thread->commandoffset == dpsoftrast.drawcommand && thread->index >= 0) { - if (thread->waiting) SDL_CondSignal(thread->waitcond); + if (thread->waiting) Thread_CondSignal(thread->waitcond); thread->starving = true; - SDL_CondWait(thread->drawcond, thread->drawmutex); + Thread_CondWait(thread->drawcond, thread->drawmutex); thread->starving = false; } - SDL_UnlockMutex(thread->drawmutex); + Thread_UnlockMutex(thread->drawmutex); } } return 0; } -#endif static void DPSOFTRAST_Draw_FlushThreads(void) { DPSOFTRAST_State_Thread *thread; int i; DPSOFTRAST_Draw_SyncCommands(); -#ifdef USE_THREADS - for (i = 0; i < dpsoftrast.numthreads; i++) + if (dpsoftrast.usethreads) { - thread = &dpsoftrast.threads[i]; - if (thread->commandoffset != dpsoftrast.drawcommand) + for (i = 0; i < dpsoftrast.numthreads; i++) { - SDL_LockMutex(thread->drawmutex); - if (thread->commandoffset != dpsoftrast.drawcommand && thread->starving) - SDL_CondSignal(thread->drawcond); - SDL_UnlockMutex(thread->drawmutex); + thread = &dpsoftrast.threads[i]; + if (thread->commandoffset != dpsoftrast.drawcommand) + { + Thread_LockMutex(thread->drawmutex); + if (thread->commandoffset != dpsoftrast.drawcommand && thread->starving) + Thread_CondSignal(thread->drawcond); + Thread_UnlockMutex(thread->drawmutex); + } } - } -#endif - for (i = 0; i < dpsoftrast.numthreads; i++) - { - thread = &dpsoftrast.threads[i]; -#ifdef USE_THREADS - if (thread->commandoffset != dpsoftrast.drawcommand) + for (i = 0; i < dpsoftrast.numthreads; i++) { - SDL_LockMutex(thread->drawmutex); + thread = &dpsoftrast.threads[i]; if (thread->commandoffset != dpsoftrast.drawcommand) { - thread->waiting = true; - SDL_CondWait(thread->waitcond, thread->drawmutex); - thread->waiting = false; + Thread_LockMutex(thread->drawmutex); + if (thread->commandoffset != dpsoftrast.drawcommand) + { + thread->waiting = true; + Thread_CondWait(thread->waitcond, thread->drawmutex); + thread->waiting = false; + } + Thread_UnlockMutex(thread->drawmutex); } - SDL_UnlockMutex(thread->drawmutex); } -#else - if (thread->commandoffset != dpsoftrast.drawcommand) - DPSOFTRAST_Draw_InterpretCommands(thread, dpsoftrast.drawcommand); -#endif + } + else + { + for (i = 0; i < dpsoftrast.numthreads; i++) + { + thread = &dpsoftrast.threads[i]; + if (thread->commandoffset != dpsoftrast.drawcommand) + DPSOFTRAST_Draw_InterpretCommands(thread, dpsoftrast.drawcommand); + } } dpsoftrast.commandpool.usedcommands = 0; } @@ -4867,7 +5564,7 @@ void DPSOFTRAST_Finish(void) DPSOFTRAST_Flush(); } -void DPSOFTRAST_Init(int width, int height, int numthreads, unsigned int *colorpixels, unsigned int *depthpixels) +int DPSOFTRAST_Init(int width, int height, int numthreads, int interlace, unsigned int *colorpixels, unsigned int *depthpixels) { int i; union @@ -4898,17 +5595,16 @@ void DPSOFTRAST_Init(int width, int height, int numthreads, unsigned int *colorp dpsoftrast.color[1] = 1; dpsoftrast.color[2] = 1; dpsoftrast.color[3] = 1; -#ifdef USE_THREADS - dpsoftrast.numthreads = bound(1, numthreads, 64); -#else - dpsoftrast.numthreads = 1; -#endif + dpsoftrast.usethreads = numthreads > 0 && Thread_HasThreads(); + dpsoftrast.interlace = dpsoftrast.usethreads ? bound(0, interlace, 1) : 0; + dpsoftrast.numthreads = dpsoftrast.usethreads ? bound(1, numthreads, 64) : 1; dpsoftrast.threads = (DPSOFTRAST_State_Thread *)MM_CALLOC(dpsoftrast.numthreads, sizeof(DPSOFTRAST_State_Thread)); for (i = 0; i < dpsoftrast.numthreads; i++) { DPSOFTRAST_State_Thread *thread = &dpsoftrast.threads[i]; thread->index = i; thread->cullface = GL_BACK; + thread->colormask[0] = 1; thread->colormask[1] = 1; thread->colormask[2] = 1; thread->colormask[3] = 1; @@ -4918,9 +5614,6 @@ void DPSOFTRAST_Init(int width, int height, int numthreads, unsigned int *colorp thread->depthtest = true; thread->depthfunc = GL_LEQUAL; thread->scissortest = false; - thread->alphatest = false; - thread->alphafunc = GL_GREATER; - thread->alphavalue = 0.5f; thread->viewport[0] = 0; thread->viewport[1] = 0; thread->viewport[2] = dpsoftrast.fb_width; @@ -4933,47 +5626,50 @@ void DPSOFTRAST_Init(int width, int height, int numthreads, unsigned int *colorp thread->depthrange[1] = 1; thread->polygonoffset[0] = 0; thread->polygonoffset[1] = 0; - + thread->clipplane[0] = 0; + thread->clipplane[1] = 0; + thread->clipplane[2] = 0; + thread->clipplane[3] = 1; + thread->numspans = 0; thread->numtriangles = 0; thread->commandoffset = 0; thread->waiting = false; thread->starving = false; -#ifdef USE_THREADS - thread->waitcond = SDL_CreateCond(); - thread->drawcond = SDL_CreateCond(); - thread->drawmutex = SDL_CreateMutex(); -#endif - + thread->validate = -1; DPSOFTRAST_Validate(thread, -1); -#ifdef USE_THREADS - thread->thread = SDL_CreateThread(DPSOFTRAST_Draw_Thread, thread); -#endif + + if (dpsoftrast.usethreads) + { + thread->waitcond = Thread_CreateCond(); + thread->drawcond = Thread_CreateCond(); + thread->drawmutex = Thread_CreateMutex(); + thread->thread = Thread_CreateThread(DPSOFTRAST_Draw_Thread, thread); + } } + return 0; } void DPSOFTRAST_Shutdown(void) { int i; -#ifdef USE_THREADS - if(dpsoftrast.numthreads > 0) + if (dpsoftrast.usethreads && dpsoftrast.numthreads > 0) { DPSOFTRAST_State_Thread *thread; for (i = 0; i < dpsoftrast.numthreads; i++) { thread = &dpsoftrast.threads[i]; - SDL_LockMutex(thread->drawmutex); + Thread_LockMutex(thread->drawmutex); thread->index = -1; - SDL_CondSignal(thread->drawcond); - SDL_UnlockMutex(thread->drawmutex); - SDL_WaitThread(thread->thread, NULL); - SDL_DestroyCond(thread->waitcond); - SDL_DestroyCond(thread->drawcond); - SDL_DestroyMutex(thread->drawmutex); + Thread_CondSignal(thread->drawcond); + Thread_UnlockMutex(thread->drawmutex); + Thread_WaitThread(thread->thread, 0); + Thread_DestroyCond(thread->waitcond); + Thread_DestroyCond(thread->drawcond); + Thread_DestroyMutex(thread->drawmutex); } } -#endif for (i = 0;i < dpsoftrast.texture_end;i++) if (dpsoftrast.texture[i].bytes) MM_FREE(dpsoftrast.texture[i].bytes);