#include "spu_per_fragment_op.h"
+#define LINEAR_QUAD_LAYOUT 1
+
+
+static INLINE vector float
+spu_min(vector float a, vector float b)
+{
+ vector unsigned int m;
+ m = spu_cmpgt(a, b); /* m = a > b ? ~0 : 0 */
+ return spu_sel(a, b, m);
+}
+
+
+static INLINE vector float
+spu_max(vector float a, vector float b)
+{
+ vector unsigned int m;
+ m = spu_cmpgt(a, b); /* m = a > b ? ~0 : 0 */
+ return spu_sel(b, a, m);
+}
+
/**
* Called by rasterizer for each quad after the shader has run. Do
vector unsigned int mask)
{
vector float frag_aos[4];
- unsigned int c0, c1, c2, c3;
+ unsigned int fbc0, fbc1, fbc2, fbc3 ; /* framebuffer/tile colors */
+ unsigned int fragc0, fragc1, fragc2, fragc3; /* fragment colors */
- /* do alpha test */
+ /*
+ * Do alpha test
+ */
if (spu.depth_stencil_alpha.alpha.enabled) {
- vector float ref = spu_splats(spu.depth_stencil_alpha.alpha.ref);
+ vector float ref = spu_splats(spu.depth_stencil_alpha.alpha.ref_value);
vector unsigned int amask;
switch (spu.depth_stencil_alpha.alpha.func) {
mask = spu_and(mask, amask);
}
- /* Z and/or stencil testing... */
+
+ /*
+ * Z and/or stencil testing...
+ */
if (spu.depth_stencil_alpha.depth.enabled ||
spu.depth_stencil_alpha.stencil[0].enabled) {
}
}
+
+ /*
+ * If we'll need the current framebuffer/tile colors for blending
+ * or logicop or colormask, fetch them now.
+ */
+ if (spu.blend.blend_enable ||
+ spu.blend.logicop_enable ||
+ spu.blend.colormask != 0xf) {
+
+#if LINEAR_QUAD_LAYOUT /* See comments/diagram below */
+ fbc0 = colorTile->ui[y][x*2+0];
+ fbc1 = colorTile->ui[y][x*2+1];
+ fbc2 = colorTile->ui[y][x*2+2];
+ fbc3 = colorTile->ui[y][x*2+3];
+#else
+ fbc0 = colorTile->ui[y+0][x+0];
+ fbc1 = colorTile->ui[y+0][x+1];
+ fbc2 = colorTile->ui[y+1][x+0];
+ fbc3 = colorTile->ui[y+1][x+1];
+#endif
+ }
+
+
+ /*
+ * Do blending
+ */
if (spu.blend.blend_enable) {
+ /* blending terms, misc regs */
vector float term1r, term1g, term1b, term1a;
vector float term2r, term2g, term2b, term2a;
-
- vector float fbRGBA[4];
-
vector float one, tmp;
- /* get colors from framebuffer */
+ vector float fbRGBA[4]; /* current framebuffer colors */
+
+ /* convert framebuffer colors from packed int to vector float */
{
- vector float fc[4];
- uint c0, c1, c2, c3;
-#if 0
- c0 = colorTile->ui[y+0][x+0];
- c1 = colorTile->ui[y+0][x+1];
- c2 = colorTile->ui[y+1][x+0];
- c3 = colorTile->ui[y+1][x+1];
-#else
- c0 = colorTile->ui[y][x*2+0];
- c1 = colorTile->ui[y][x*2+1];
- c2 = colorTile->ui[y][x*2+2];
- c3 = colorTile->ui[y][x*2+3];
-#endif
+ vector float temp[4]; /* float colors in AOS form */
switch (spu.fb.color_format) {
case PIPE_FORMAT_B8G8R8A8_UNORM:
- fc[0] = spu_unpack_B8G8R8A8(c0);
- fc[1] = spu_unpack_B8G8R8A8(c1);
- fc[2] = spu_unpack_B8G8R8A8(c2);
- fc[3] = spu_unpack_B8G8R8A8(c3);
+ temp[0] = spu_unpack_B8G8R8A8(fbc0);
+ temp[1] = spu_unpack_B8G8R8A8(fbc1);
+ temp[2] = spu_unpack_B8G8R8A8(fbc2);
+ temp[3] = spu_unpack_B8G8R8A8(fbc3);
break;
case PIPE_FORMAT_A8R8G8B8_UNORM:
- fc[0] = spu_unpack_A8R8G8B8(c0);
- fc[1] = spu_unpack_A8R8G8B8(c1);
- fc[2] = spu_unpack_A8R8G8B8(c2);
- fc[3] = spu_unpack_A8R8G8B8(c3);
+ temp[0] = spu_unpack_A8R8G8B8(fbc0);
+ temp[1] = spu_unpack_A8R8G8B8(fbc1);
+ temp[2] = spu_unpack_A8R8G8B8(fbc2);
+ temp[3] = spu_unpack_A8R8G8B8(fbc3);
break;
default:
ASSERT(0);
}
- _transpose_matrix4x4(fbRGBA, fc);
+ _transpose_matrix4x4(fbRGBA, temp); /* fbRGBA = transpose(temp) */
}
/*
- * Compute Src RGB terms
+ * Compute Src RGB terms (fragment color * factor)
*/
switch (spu.blend.rgb_src_factor) {
case PIPE_BLENDFACTOR_ONE:
term1g = spu_mul(fragG, fragA);
term1b = spu_mul(fragB, fragA);
break;
+ case PIPE_BLENDFACTOR_DST_COLOR:
+ term1r = spu_mul(fragR, fbRGBA[0]);
+ term1g = spu_mul(fragG, fbRGBA[1]);
+ term1b = spu_mul(fragB, fbRGBA[1]);
+ break;
+ case PIPE_BLENDFACTOR_DST_ALPHA:
+ term1r = spu_mul(fragR, fbRGBA[3]);
+ term1g = spu_mul(fragG, fbRGBA[3]);
+ term1b = spu_mul(fragB, fbRGBA[3]);
+ break;
+ case PIPE_BLENDFACTOR_CONST_COLOR:
+ term1r = spu_mul(fragR, spu_splats(spu.blend_color.color[0]));
+ term1g = spu_mul(fragG, spu_splats(spu.blend_color.color[1]));
+ term1b = spu_mul(fragB, spu_splats(spu.blend_color.color[2]));
+ break;
+ case PIPE_BLENDFACTOR_CONST_ALPHA:
+ term1r = spu_mul(fragR, spu_splats(spu.blend_color.color[3]));
+ term1g = spu_mul(fragG, spu_splats(spu.blend_color.color[3]));
+ term1b = spu_mul(fragB, spu_splats(spu.blend_color.color[3]));
+ break;
/* XXX more cases */
default:
ASSERT(0);
}
/*
- * Compute Src Alpha term
+ * Compute Src Alpha term (fragment alpha * factor)
*/
switch (spu.blend.alpha_src_factor) {
case PIPE_BLENDFACTOR_ONE:
case PIPE_BLENDFACTOR_SRC_ALPHA:
term1a = spu_mul(fragA, fragA);
break;
+ case PIPE_BLENDFACTOR_DST_COLOR:
+ /* fall-through */
+ case PIPE_BLENDFACTOR_DST_ALPHA:
+ term1a = spu_mul(fragA, fbRGBA[3]);
+ break;
+ case PIPE_BLENDFACTOR_CONST_COLOR:
+ /* fall-through */
+ case PIPE_BLENDFACTOR_CONST_ALPHA:
+ term1a = spu_mul(fragR, spu_splats(spu.blend_color.color[3]));
+ break;
/* XXX more cases */
default:
ASSERT(0);
}
/*
- * Compute Dest RGB terms
+ * Compute Dest RGB terms (framebuffer color * factor)
*/
switch (spu.blend.rgb_dst_factor) {
case PIPE_BLENDFACTOR_ONE:
- term2r = fragR;
- term2g = fragG;
- term2b = fragB;
+ term2r = fbRGBA[0];
+ term2g = fbRGBA[1];
+ term2b = fbRGBA[2];
break;
case PIPE_BLENDFACTOR_ZERO:
term2r =
term2g = spu_mul(fbRGBA[1], tmp);
term2b = spu_mul(fbRGBA[2], tmp);
break;
- /* XXX more cases */
+ case PIPE_BLENDFACTOR_DST_COLOR:
+ term2r = spu_mul(fbRGBA[0], fbRGBA[0]);
+ term2g = spu_mul(fbRGBA[1], fbRGBA[1]);
+ term2b = spu_mul(fbRGBA[2], fbRGBA[2]);
+ break;
+ case PIPE_BLENDFACTOR_DST_ALPHA:
+ term2r = spu_mul(fbRGBA[0], fbRGBA[3]);
+ term2g = spu_mul(fbRGBA[1], fbRGBA[3]);
+ term2b = spu_mul(fbRGBA[2], fbRGBA[3]);
+ break;
+ case PIPE_BLENDFACTOR_CONST_COLOR:
+ term2r = spu_mul(fbRGBA[0], spu_splats(spu.blend_color.color[0]));
+ term2g = spu_mul(fbRGBA[1], spu_splats(spu.blend_color.color[1]));
+ term2b = spu_mul(fbRGBA[2], spu_splats(spu.blend_color.color[2]));
+ break;
+ case PIPE_BLENDFACTOR_CONST_ALPHA:
+ term2r = spu_mul(fbRGBA[0], spu_splats(spu.blend_color.color[3]));
+ term2g = spu_mul(fbRGBA[1], spu_splats(spu.blend_color.color[3]));
+ term2b = spu_mul(fbRGBA[2], spu_splats(spu.blend_color.color[3]));
+ break;
+ /* XXX more cases */
default:
ASSERT(0);
}
/*
- * Compute Dest Alpha term
+ * Compute Dest Alpha term (framebuffer alpha * factor)
*/
switch (spu.blend.alpha_dst_factor) {
case PIPE_BLENDFACTOR_ONE:
- term2a = fragA;
+ term2a = fbRGBA[3];
break;
case PIPE_BLENDFACTOR_SRC_COLOR:
term2a = spu_splats(0.0f);
tmp = spu_sub(one, fragA);
term2a = spu_mul(fbRGBA[3], tmp);
break;
+ case PIPE_BLENDFACTOR_DST_COLOR:
+ /* fall-through */
+ case PIPE_BLENDFACTOR_DST_ALPHA:
+ term2a = spu_mul(fbRGBA[3], fbRGBA[3]);
+ break;
+ case PIPE_BLENDFACTOR_CONST_COLOR:
+ /* fall-through */
+ case PIPE_BLENDFACTOR_CONST_ALPHA:
+ term2a = spu_mul(fbRGBA[3], spu_splats(spu.blend_color.color[3]));
+ break;
/* XXX more cases */
default:
ASSERT(0);
fragG = spu_sub(term1g, term2g);
fragB = spu_sub(term1b, term2b);
break;
- /* XXX more cases */
+ case PIPE_BLEND_REVERSE_SUBTRACT:
+ fragR = spu_sub(term2r, term1r);
+ fragG = spu_sub(term2g, term1g);
+ fragB = spu_sub(term2b, term1b);
+ break;
+ case PIPE_BLEND_MIN:
+ fragR = spu_min(term1r, term2r);
+ fragG = spu_min(term1g, term2g);
+ fragB = spu_min(term1b, term2b);
+ break;
+ case PIPE_BLEND_MAX:
+ fragR = spu_max(term1r, term2r);
+ fragG = spu_max(term1g, term2g);
+ fragB = spu_max(term1b, term2b);
+ break;
default:
ASSERT(0);
}
case PIPE_BLEND_SUBTRACT:
fragA = spu_sub(term1a, term2a);
break;
- /* XXX more cases */
+ case PIPE_BLEND_REVERSE_SUBTRACT:
+ fragA = spu_sub(term2a, term1a);
+ break;
+ case PIPE_BLEND_MIN:
+ fragA = spu_min(term1a, term2a);
+ break;
+ case PIPE_BLEND_MAX:
+ fragA = spu_max(term1a, term2a);
+ break;
default:
ASSERT(0);
}
}
- /* XXX do colormask test here */
-
-
- if (spu_extract(spu_orx(mask), 0)) {
- spu.cur_ctile_status = TILE_STATUS_DIRTY;
- }
- else {
- return;
- }
-
- /* convert RRRR,GGGG,BBBB,AAAA to RGBA,RGBA,RGBA,RGBA */
+ /*
+ * Convert RRRR,GGGG,BBBB,AAAA to RGBA,RGBA,RGBA,RGBA.
+ */
#if 0
+ /* original code */
{
vector float frag_soa[4];
frag_soa[0] = fragR;
(void) fragB;
#endif
+ /*
+ * Pack fragment float colors into 32-bit RGBA words.
+ */
switch (spu.fb.color_format) {
case PIPE_FORMAT_A8R8G8B8_UNORM:
- c0 = spu_pack_A8R8G8B8(frag_aos[0]);
- c1 = spu_pack_A8R8G8B8(frag_aos[1]);
- c2 = spu_pack_A8R8G8B8(frag_aos[2]);
- c3 = spu_pack_A8R8G8B8(frag_aos[3]);
+ fragc0 = spu_pack_A8R8G8B8(frag_aos[0]);
+ fragc1 = spu_pack_A8R8G8B8(frag_aos[1]);
+ fragc2 = spu_pack_A8R8G8B8(frag_aos[2]);
+ fragc3 = spu_pack_A8R8G8B8(frag_aos[3]);
break;
-
case PIPE_FORMAT_B8G8R8A8_UNORM:
- c0 = spu_pack_B8G8R8A8(frag_aos[0]);
- c1 = spu_pack_B8G8R8A8(frag_aos[1]);
- c2 = spu_pack_B8G8R8A8(frag_aos[2]);
- c3 = spu_pack_B8G8R8A8(frag_aos[3]);
+ fragc0 = spu_pack_B8G8R8A8(frag_aos[0]);
+ fragc1 = spu_pack_B8G8R8A8(frag_aos[1]);
+ fragc2 = spu_pack_B8G8R8A8(frag_aos[2]);
+ fragc3 = spu_pack_B8G8R8A8(frag_aos[3]);
break;
default:
fprintf(stderr, "SPU: Bad pixel format in spu_default_fragment_ops\n");
ASSERT(0);
}
-#if 0
+
+ /*
+ * Do color masking
+ */
+ if (spu.blend.colormask != 0xf) {
+ uint cmask = 0x0; /* each byte corresponds to a color channel */
+
+ /* Form bitmask depending on color buffer format and colormask bits */
+ switch (spu.fb.color_format) {
+ case PIPE_FORMAT_A8R8G8B8_UNORM:
+ if (spu.blend.colormask & PIPE_MASK_R)
+ cmask |= 0x00ff0000; /* red */
+ if (spu.blend.colormask & PIPE_MASK_G)
+ cmask |= 0x0000ff00; /* green */
+ if (spu.blend.colormask & PIPE_MASK_B)
+ cmask |= 0x000000ff; /* blue */
+ if (spu.blend.colormask & PIPE_MASK_A)
+ cmask |= 0xff000000; /* alpha */
+ break;
+ case PIPE_FORMAT_B8G8R8A8_UNORM:
+ if (spu.blend.colormask & PIPE_MASK_R)
+ cmask |= 0x0000ff00; /* red */
+ if (spu.blend.colormask & PIPE_MASK_G)
+ cmask |= 0x00ff0000; /* green */
+ if (spu.blend.colormask & PIPE_MASK_B)
+ cmask |= 0xff000000; /* blue */
+ if (spu.blend.colormask & PIPE_MASK_A)
+ cmask |= 0x000000ff; /* alpha */
+ break;
+ default:
+ ASSERT(0);
+ }
+
+ /*
+ * Apply color mask to the 32-bit packed colors.
+ * if (cmask[i])
+ * frag color[i] = frag color[i];
+ * else
+ * frag color[i] = framebuffer color[i];
+ */
+ fragc0 = (fragc0 & cmask) | (fbc0 & ~cmask);
+ fragc1 = (fragc1 & cmask) | (fbc1 & ~cmask);
+ fragc2 = (fragc2 & cmask) | (fbc2 & ~cmask);
+ fragc3 = (fragc3 & cmask) | (fbc3 & ~cmask);
+ }
+
+
+ /*
+ * Do logic ops
+ */
+ if (spu.blend.logicop_enable) {
+ /* XXX to do */
+ /* apply logicop to 32-bit packed colors (fragcx and fbcx) */
+ }
+
+
+ /*
+ * If mask is non-zero, mark tile as dirty.
+ */
+ if (spu_extract(spu_orx(mask), 0)) {
+ spu.cur_ctile_status = TILE_STATUS_DIRTY;
+ }
+ else {
+ /* write no fragments */
+ return;
+ }
+
+
+ /*
+ * Write new fragment/quad colors to the framebuffer/tile.
+ * Only write pixels where the corresponding mask word is set.
+ */
+#if LINEAR_QUAD_LAYOUT
/*
* Quad layout:
- * +--+--+
- * |p0|p1|
- * +--+--+
- * |p2|p3|
- * +--+--+
+ * +--+--+--+--+
+ * |p0|p1|p2|p3|...
+ * +--+--+--+--+
*/
if (spu_extract(mask, 0))
- colorTile->ui[y+0][x+0] = c0;
+ colorTile->ui[y][x*2] = fragc0;
if (spu_extract(mask, 1))
- colorTile->ui[y+0][x+1] = c1;
+ colorTile->ui[y][x*2+1] = fragc1;
if (spu_extract(mask, 2))
- colorTile->ui[y+1][x+0] = c2;
+ colorTile->ui[y][x*2+2] = fragc2;
if (spu_extract(mask, 3))
- colorTile->ui[y+1][x+1] = c3;
+ colorTile->ui[y][x*2+3] = fragc3;
#else
/*
* Quad layout:
- * +--+--+--+--+
- * |p0|p1|p2|p3|
- * +--+--+--+--+
+ * +--+--+
+ * |p0|p1|...
+ * +--+--+
+ * |p2|p3|...
+ * +--+--+
*/
if (spu_extract(mask, 0))
- colorTile->ui[y][x*2] = c0;
+ colorTile->ui[y+0][x+0] = fragc0;
if (spu_extract(mask, 1))
- colorTile->ui[y][x*2+1] = c1;
+ colorTile->ui[y+0][x+1] = fragc1;
if (spu_extract(mask, 2))
- colorTile->ui[y][x*2+2] = c2;
+ colorTile->ui[y+1][x+0] = fragc2;
if (spu_extract(mask, 3))
- colorTile->ui[y][x*2+3] = c3;
+ colorTile->ui[y+1][x+1] = fragc3;
#endif
}