X-Git-Url: http://git.xonotic.org/?p=xonotic%2Fdarkplaces.git;a=blobdiff_plain;f=dpsoftrast.c;h=179a4d0c5fe93cae3b4668e1f4d9221c6a79082c;hp=13865e084ae5db5dee9e11bbb2f6aebca6c8dbf9;hb=e1cc58c8345fbadd75421fa7843ec19bdf0b8681;hpb=5428f67a5532b4b5a18cfd9dce04fc7c2bb936ab diff --git a/dpsoftrast.c b/dpsoftrast.c index 13865e08..179a4d0c 100644 --- a/dpsoftrast.c +++ b/dpsoftrast.c @@ -27,6 +27,20 @@ typedef qboolean bool; #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))) + #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))) @@ -73,6 +87,10 @@ typedef qboolean bool; #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) @@ -188,11 +206,14 @@ typedef ALIGN(struct DPSOFTRAST_State_Span_s 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 @@ -228,13 +249,12 @@ 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; @@ -278,6 +298,7 @@ typedef ATOMIC(struct DPSOFTRAST_State_Thread_s 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); @@ -345,7 +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) { @@ -375,6 +398,15 @@ static void DPSOFTRAST_RecalcThread(DPSOFTRAST_State_Thread *thread) } } +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, @@ -396,6 +428,7 @@ static void DPSOFTRAST_RecalcFB(DPSOFTRAST_State_Thread *thread) thread->fb_scissor[3] = y2 - y1; DPSOFTRAST_RecalcViewport(thread->viewport, thread->fb_viewportcenter, thread->fb_viewportscale); + DPSOFTRAST_RecalcClipPlane(thread); DPSOFTRAST_RecalcThread(thread); } @@ -1132,30 +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; @@ -1435,6 +1444,21 @@ void DPSOFTRAST_Uniform1i(DPSOFTRAST_UNIFORM index, int i0) dpsoftrast.uniform1i[command->index] = i0; } +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) { @@ -2008,327 +2032,318 @@ 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]; - } - break; - case DPSOFTRAST_BLENDMODE_SUBALPHA: - 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;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]; - } - break; case DPSOFTRAST_BLENDMODE_PSEUDOALPHA: - for (x = startx;x < endx;x++) - { - 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]; - } - break; case DPSOFTRAST_BLENDMODE_INVADD: - for (x = startx;x < endx;x++) - { - if (!pixelmask[x]) - continue; - d[0] = (int)((255.0f-pixel[x*4+2])*in4f[x*4+0] + pixel[x*4+2]);if (d[0] > 255) d[0] = 255; - d[1] = (int)((255.0f-pixel[x*4+1])*in4f[x*4+1] + pixel[x*4+1]);if (d[1] > 255) d[1] = 255; - d[2] = (int)((255.0f-pixel[x*4+0])*in4f[x*4+2] + pixel[x*4+0]);if (d[2] > 255) d[2] = 255; - d[3] = (int)((255.0f-pixel[x*4+3])*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; } -} - -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; - 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) + // 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) { - case DPSOFTRAST_BLENDMODE_OPAQUE: - for (x = startx;x + 4 <= endx;) + // if this pixel is masked off, it's probably not alone... + if (!pixelmask[x]) { - if (*(const unsigned int *)&pixelmask[x] == 0x01010101) + x++; +#if 1 + if (x + 8 < endx) { - _mm_storeu_si128((__m128i *)&pixeli[x], _mm_loadu_si128((const __m128i *)&ini[x])); - x += 4; + // 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; + } + endmasked: + // find length of subspan + subx = x + 1; +#if 1 + if (subx + 8 < endx) + { + 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; + } +#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) + { + case DPSOFTRAST_BLENDMODE_OPAQUE: +#if 0 + if (subx - x >= 16) + { + memcpy(pixeli + x, ini + x, (subx - x) * sizeof(pixeli[x])); + x = subx; } else +#elif 1 + while (x + 16 <= subx) { - if (pixelmask[x]) + _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 + { + 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]; - x++; + pixeli[x+1] = ini[x+1]; + x += 2; + } + if (x < subx) + { + pixeli[x] = ini[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 + 1 < 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; } - - 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 } +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]; + } +} + 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) { int x; @@ -2539,6 +2554,7 @@ 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]; @@ -2558,6 +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; @@ -2576,7 +2593,7 @@ 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)); @@ -3289,7 +3306,7 @@ void DPSOFTRAST_PixelShader_FlatColor(DPSOFTRAST_State_Thread *thread, const DPS unsigned char buffer_FragColorbgra8[DPSOFTRAST_DRAW_MAXSPANLENGTH*4]; 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)); @@ -3342,7 +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)); @@ -3417,7 +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)); @@ -4395,15 +4412,13 @@ void DPSOFTRAST_PixelShader_LightSource(DPSOFTRAST_State_Thread *thread, const D void DPSOFTRAST_VertexShader_Refraction(void) { - DPSOFTRAST_Array_Transform(DPSOFTRAST_ARRAY_TEXCOORD1, DPSOFTRAST_ARRAY_POSITION, 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); 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) { - // DIRTY TRICK: only do sideways displacement. Not correct, but cheaper and thus better for SW. - float buffer_z[DPSOFTRAST_DRAW_MAXSPANLENGTH]; float z; int x, startx = span->startx, endx = span->endx; @@ -4422,18 +4437,15 @@ void DPSOFTRAST_PixelShader_Refraction(DPSOFTRAST_State_Thread *thread, const DP float DistortScaleRefractReflect[2]; float RefractColor[4]; - const unsigned char * RESTRICT pixelbase; - const unsigned char * RESTRICT pixel[4]; DPSOFTRAST_Texture *texture = thread->texbound[GL20TU_REFRACTION]; if(!texture) return; - pixelbase = (unsigned char *)texture->bytes + texture->mipmap[0][0]; // read textures DPSOFTRAST_Draw_Span_Begin(thread, triangle, span, buffer_z); 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_TEXCOORD1); // or POSITION? + DPSOFTRAST_CALCATTRIB4F(triangle, span, ModelViewProjectionPositiondata, ModelViewProjectionPositionslope, DPSOFTRAST_ARRAY_TEXCOORD4); // read uniforms ScreenScaleRefractReflect[0] = thread->uniform4f[DPSOFTRAST_UNIFORM_ScreenScaleRefractReflect*4+0]; @@ -4460,7 +4472,7 @@ void DPSOFTRAST_PixelShader_Refraction(DPSOFTRAST_State_Thread *thread, const DP // " 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) @@ -4474,38 +4486,8 @@ void DPSOFTRAST_PixelShader_Refraction(DPSOFTRAST_State_Thread *thread, const DP ScreenTexCoord[1] = SafeScreenTexCoord[1] + v[1] * DistortScaleRefractReflect[1]; // " dp_FragColor = vec4(dp_texture2D(Texture_Refraction, ScreenTexCoord).rgb, 1.0) * RefractColor;\n" - if(texture->filter & DPSOFTRAST_TEXTURE_FILTER_LINEAR) - { - unsigned int tc[2] = { ScreenTexCoord[0] * (texture->mipmap[0][2]<<12) - 2048, ScreenTexCoord[1] * (texture->mipmap[0][3]<<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 }; - tci[0] = tci[0] >= 0 ? (tci[0] <= texture->mipmap[0][2]-1 ? tci[0] : texture->mipmap[0][2]-1) : 0; - tci[1] = tci[1] >= 0 ? (tci[1] <= texture->mipmap[0][3]-1 ? tci[1] : texture->mipmap[0][3]-1) : 0; - tci1[0] = tci1[0] >= 0 ? (tci1[0] <= texture->mipmap[0][2]-1 ? tci1[0] : texture->mipmap[0][2]-1) : 0; - tci1[1] = tci1[1] >= 0 ? (tci1[1] <= texture->mipmap[0][3]-1 ? tci1[1] : texture->mipmap[0][3]-1) : 0; - pixel[0] = pixelbase + 4 * (tci[1]*texture->mipmap[0][2]+tci[0]); - pixel[1] = pixelbase + 4 * (tci[1]*texture->mipmap[0][2]+tci1[0]); - pixel[2] = pixelbase + 4 * (tci1[1]*texture->mipmap[0][2]+tci[0]); - pixel[3] = pixelbase + 4 * (tci1[1]*texture->mipmap[0][2]+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; - } - else - { - int tci[2] = { ScreenTexCoord[0] * texture->mipmap[0][2], ScreenTexCoord[1] * texture->mipmap[0][3] }; - tci[0] = tci[0] >= 0 ? (tci[0] <= texture->mipmap[0][2]-1 ? tci[0] : texture->mipmap[0][2]-1) : 0; - tci[1] = tci[1] >= 0 ? (tci[1] <= texture->mipmap[0][3]-1 ? tci[1] : texture->mipmap[0][3]-1) : 0; - pixel[0] = pixelbase + 4 * (tci[1]*texture->mipmap[0][2]+tci[0]); - c[0] = pixel[0][0]; - c[1] = pixel[0][1]; - c[2] = pixel[0][2]; - } + DPSOFTRAST_Texture2DBGRA8(texture, 0, ScreenTexCoord[0], ScreenTexCoord[1], c); - //p = (int) bound(startx, x + (ScreenTexCoord[0] - SafeScreenTexCoord[0]) / (ModelViewProjectionPositionslope[0]*z), endx-1); 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]; @@ -4519,17 +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 + span->startx*4, 0, (span->endx - span->startx)*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); } @@ -4609,88 +4741,96 @@ static const DPSOFTRAST_ShaderModeInfo DPSOFTRAST_ShaderModeTable[SHADERMODE_COU {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, {DPSOFTRAST_ARRAY_TEXCOORD0, DPSOFTRAST_ARRAY_TEXCOORD1, ~0}, {GL20TU_NORMAL, GL20TU_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}}, }; -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; - startx = span->startx; - endx = span->endx; - 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; - } - //colorpixel = dpsoftrast.fb_colorpixels[0] + (span->y * dpsoftrast.fb_width + span->x) * 4;; - //for (x = startx;x < endx;x++) - // colorpixel[x] = (depthpixel[x] & 0xFF000000) ? (0x00FF0000) : (depthpixel[x] & 0x00FF0000); - // if there is no color buffer, skip pixel shader - 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 + span->startx, 1, span->endx - span->startx); - span->pixelmask = pixelmask; - 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; } @@ -4721,6 +4861,8 @@ static void DPSOFTRAST_Interpret_Draw(DPSOFTRAST_State_Thread *thread, DPSOFTRAS int numpoints; int clipcase; float clipdist[4]; + float clip0origin, clip0slope; + int clip0dir; __m128 triangleedge1, triangleedge2, trianglenormal; __m128 clipfrac[3]; __m128 screen[4]; @@ -4945,6 +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++) { @@ -5006,6 +5185,8 @@ static void DPSOFTRAST_Interpret_Draw(DPSOFTRAST_State_Thread *thread, DPSOFTRAS int yccmask = _mm_movemask_epi8(ycc); int edge0p, edge0n, edge1p, edge1n; int nexty; + float w, wslope; + float clip0; if (numpoints == 4) { switch(yccmask) @@ -5059,28 +5240,47 @@ static void DPSOFTRAST_Interpret_Draw(DPSOFTRAST_State_Thread *thread, DPSOFTRAS xcoords = _mm_shuffle_ps(xcoords, xcoords, _MM_SHUFFLE(1, 0, 3, 2)); xslope = _mm_shuffle_ps(xslope, xslope, _MM_SHUFFLE(1, 0, 3, 2)); } - for(; y <= nexty; y++, xcoords = _mm_add_ps(xcoords, xslope)) + 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 < minx) - { - if (startx < 0) startx = 0; - startx += (minx-startx)&~(DPSOFTRAST_DRAW_MAXSPANLENGTH-1); - } + if (startx < minx) startx = minx; if (endx > maxx) endx = maxx; if (startx >= endx) continue; + + 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->startx = max(minx - offset, 0); + span->startx = 0; span->endx = min(endx - offset, DPSOFTRAST_DRAW_MAXSPANLENGTH); if (span->startx >= span->endx) - continue; + 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); } @@ -5259,14 +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); @@ -5405,6 +5604,7 @@ int DPSOFTRAST_Init(int width, int height, int numthreads, int interlace, unsign 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; @@ -5414,9 +5614,6 @@ int DPSOFTRAST_Init(int width, int height, int numthreads, int interlace, unsign 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; @@ -5429,8 +5626,10 @@ int DPSOFTRAST_Init(int width, int height, int numthreads, int interlace, unsign thread->depthrange[1] = 1; thread->polygonoffset[0] = 0; thread->polygonoffset[1] = 0; - - DPSOFTRAST_RecalcThread(thread); + thread->clipplane[0] = 0; + thread->clipplane[1] = 0; + thread->clipplane[2] = 0; + thread->clipplane[3] = 1; thread->numspans = 0; thread->numtriangles = 0;