diff options
author | Brian <[email protected]> | 2007-07-03 17:12:11 -0600 |
---|---|---|
committer | Brian <[email protected]> | 2007-07-03 17:12:11 -0600 |
commit | f6d4f5e1e9d626356311e1868b35b79d355af739 (patch) | |
tree | d4002c103a94ae920faec6d9ac6b764bc7c7006a /src/mesa | |
parent | a813f654872380547faf67fba75fa30792f87ea6 (diff) |
Rewrite blending in terms of SIMD operations.
Diffstat (limited to 'src/mesa')
-rw-r--r-- | src/mesa/pipe/softpipe/sp_quad_blend.c | 431 |
1 files changed, 324 insertions, 107 deletions
diff --git a/src/mesa/pipe/softpipe/sp_quad_blend.c b/src/mesa/pipe/softpipe/sp_quad_blend.c index 86f9fb633cc..2599e7a2f98 100644 --- a/src/mesa/pipe/softpipe/sp_quad_blend.c +++ b/src/mesa/pipe/softpipe/sp_quad_blend.c @@ -40,130 +40,347 @@ #include "sp_quad.h" +#define VEC4_COPY(DST, SRC) \ +do { \ + DST[0] = SRC[0]; \ + DST[1] = SRC[1]; \ + DST[2] = SRC[2]; \ + DST[3] = SRC[3]; \ +} while(0) + +#define VEC4_SCALAR(DST, SRC) \ +do { \ + DST[0] = SRC; \ + DST[1] = SRC; \ + DST[2] = SRC; \ + DST[3] = SRC; \ +} while(0) + +#define VEC4_ADD(SUM, A, B) \ +do { \ + SUM[0] = A[0] + B[0]; \ + SUM[1] = A[1] + B[1]; \ + SUM[2] = A[2] + B[2]; \ + SUM[3] = A[3] + B[3]; \ +} while (0) + +#define VEC4_SUB(SUM, A, B) \ +do { \ + SUM[0] = A[0] - B[0]; \ + SUM[1] = A[1] - B[1]; \ + SUM[2] = A[2] - B[2]; \ + SUM[3] = A[3] - B[3]; \ +} while (0) + +#define VEC4_MUL(SUM, A, B) \ +do { \ + SUM[0] = A[0] * B[0]; \ + SUM[1] = A[1] * B[1]; \ + SUM[2] = A[2] * B[2]; \ + SUM[3] = A[3] * B[3]; \ +} while (0) + +#define VEC4_MIN(SUM, A, B) \ +do { \ + SUM[0] = (A[0] < B[0]) ? A[0] : B[0]; \ + SUM[1] = (A[1] < B[1]) ? A[1] : B[1]; \ + SUM[2] = (A[2] < B[2]) ? A[2] : B[2]; \ + SUM[3] = (A[3] < B[3]) ? A[3] : B[3]; \ +} while (0) + +#define VEC4_MAX(SUM, A, B) \ +do { \ + SUM[0] = (A[0] > B[0]) ? A[0] : B[0]; \ + SUM[1] = (A[1] > B[1]) ? A[1] : B[1]; \ + SUM[2] = (A[2] > B[2]) ? A[2] : B[2]; \ + SUM[3] = (A[3] > B[3]) ? A[3] : B[3]; \ +} while (0) + + static void blend_quad(struct quad_stage *qs, struct quad_header *quad) { + static const GLfloat zero[4] = { 0, 0, 0, 0 }; + static const GLfloat one[4] = { 1, 1, 1, 1 }; struct softpipe_context *softpipe = qs->softpipe; GLfloat source[4][QUAD_SIZE], dest[4][QUAD_SIZE]; - GLuint i, j; + GLuint i; /* XXX we're also looping in output_quad() !?! */ - /* copy quad's colors since we'll modify them in the loop */ - memcpy(source, quad->outputs.color, 4 * 4 * sizeof(GLfloat)); - for (i = 0; i < softpipe->framebuffer.num_cbufs; i++) { - GLfloat srcTerm[4], dstTerm[4]; struct softpipe_surface *sps = softpipe_surface(softpipe->framebuffer.cbufs[i]); sps->read_quad_f_swz(sps, quad->x0, quad->y0, dest); - /* XXX this loop could be factored out - we'd compute the src/dstTerm - * for all four pixels in the quad at once. + /* + * Compute src/first term RGB + */ + switch (softpipe->blend.rgb_src_factor) { + case PIPE_BLENDFACTOR_ONE: + VEC4_COPY(source[0], quad->outputs.color[0]); /* R */ + VEC4_COPY(source[1], quad->outputs.color[1]); /* G */ + VEC4_COPY(source[2], quad->outputs.color[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_COLOR: + VEC4_MUL(source[0], quad->outputs.color[0], quad->outputs.color[0]); /* R */ + VEC4_MUL(source[1], quad->outputs.color[1], quad->outputs.color[1]); /* G */ + VEC4_MUL(source[2], quad->outputs.color[2], quad->outputs.color[2]); /* B */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA: + { + const GLfloat *alpha = quad->outputs.color[3]; + VEC4_MUL(source[0], quad->outputs.color[0], alpha); /* R */ + VEC4_MUL(source[1], quad->outputs.color[1], alpha); /* G */ + VEC4_MUL(source[2], quad->outputs.color[2], alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_DST_COLOR: + VEC4_MUL(source[0], quad->outputs.color[0], dest[0]); /* R */ + VEC4_MUL(source[1], quad->outputs.color[1], dest[1]); /* G */ + VEC4_MUL(source[2], quad->outputs.color[2], dest[2]); /* B */ + break; + case PIPE_BLENDFACTOR_DST_ALPHA: + { + const GLfloat *alpha = dest[3]; + VEC4_MUL(source[0], quad->outputs.color[0], alpha); /* R */ + VEC4_MUL(source[1], quad->outputs.color[1], alpha); /* G */ + VEC4_MUL(source[2], quad->outputs.color[2], alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_CONST_COLOR: + { + GLfloat comp[4]; + VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */ + VEC4_MUL(source[0], quad->outputs.color[0], comp); /* R */ + VEC4_SCALAR(comp, softpipe->blend_color.color[1]); /* G */ + VEC4_MUL(source[1], quad->outputs.color[1], comp); /* G */ + VEC4_SCALAR(comp, softpipe->blend_color.color[2]); /* B */ + VEC4_MUL(source[2], quad->outputs.color[2], comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_CONST_ALPHA: + { + GLfloat alpha[4]; + VEC4_SCALAR(alpha, softpipe->blend_color.color[3]); + VEC4_MUL(source[0], quad->outputs.color[0], alpha); /* R */ + VEC4_MUL(source[1], quad->outputs.color[1], alpha); /* G */ + VEC4_MUL(source[2], quad->outputs.color[2], alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_SRC1_COLOR: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_SRC1_ALPHA: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(source[0], zero); /* R */ + VEC4_COPY(source[1], zero); /* G */ + VEC4_COPY(source[2], zero); /* B */ + break; + case PIPE_BLENDFACTOR_INV_SRC_COLOR: + { + GLfloat inv_comp[4]; + VEC4_SUB(inv_comp, one, quad->outputs.color[0]); /* R */ + VEC4_MUL(source[0], quad->outputs.color[0], inv_comp); /* R */ + VEC4_SUB(inv_comp, one, quad->outputs.color[1]); /* G */ + VEC4_MUL(source[1], quad->outputs.color[1], inv_comp); /* G */ + VEC4_SUB(inv_comp, one, quad->outputs.color[2]); /* B */ + VEC4_MUL(source[2], quad->outputs.color[2], inv_comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + { + GLfloat inv_alpha[4]; + VEC4_SUB(inv_alpha, one, quad->outputs.color[3]); + VEC4_MUL(source[0], quad->outputs.color[0], inv_alpha); /* R */ + VEC4_MUL(source[1], quad->outputs.color[1], inv_alpha); /* G */ + VEC4_MUL(source[2], quad->outputs.color[2], inv_alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_ALPHA: + { + GLfloat inv_alpha[4]; + VEC4_SUB(inv_alpha, one, dest[3]); + VEC4_MUL(source[0], quad->outputs.color[0], inv_alpha); /* R */ + VEC4_MUL(source[1], quad->outputs.color[1], inv_alpha); /* G */ + VEC4_MUL(source[2], quad->outputs.color[2], inv_alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_DST_COLOR: + { + GLfloat inv_comp[4]; + VEC4_SUB(inv_comp, one, dest[0]); /* R */ + VEC4_MUL(source[0], quad->outputs.color[0], inv_comp); /* R */ + VEC4_SUB(inv_comp, one, dest[1]); /* G */ + VEC4_MUL(source[1], quad->outputs.color[1], inv_comp); /* G */ + VEC4_SUB(inv_comp, one, dest[2]); /* B */ + VEC4_MUL(source[2], quad->outputs.color[2], inv_comp); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_CONST_COLOR: + { + GLfloat inv_comp[4]; + /* R */ + VEC4_SCALAR(inv_comp, 1.0 - softpipe->blend_color.color[0]); + VEC4_MUL(source[0], quad->outputs.color[0], inv_comp); + /* G */ + VEC4_SCALAR(inv_comp, 1.0 - softpipe->blend_color.color[1]); + VEC4_MUL(source[1], quad->outputs.color[1], inv_comp); + /* B */ + VEC4_SCALAR(inv_comp, 1.0 - softpipe->blend_color.color[2]); + VEC4_MUL(source[2], quad->outputs.color[2], inv_comp); + } + break; + case PIPE_BLENDFACTOR_INV_CONST_ALPHA: + { + GLfloat alpha[4], inv_alpha[4]; + VEC4_SCALAR(alpha, 1.0 - softpipe->blend_color.color[3]); + VEC4_MUL(source[0], quad->outputs.color[0], inv_alpha); /* R */ + VEC4_MUL(source[1], quad->outputs.color[1], inv_alpha); /* G */ + VEC4_MUL(source[2], quad->outputs.color[2], inv_alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_INV_SRC1_COLOR: + assert(0); /* to do */ + break; + case PIPE_BLENDFACTOR_INV_SRC1_ALPHA: + assert(0); /* to do */ + break; + default: + abort(); + } + + /* + * Compute src/first term A + */ + switch (softpipe->blend.alpha_src_factor) { + case PIPE_BLENDFACTOR_ONE: + VEC4_COPY(source[3], quad->outputs.color[3]); /* A */ + break; + case PIPE_BLENDFACTOR_SRC_ALPHA: + { + const GLfloat *alpha = quad->outputs.color[3]; + VEC4_MUL(source[3], quad->outputs.color[3], alpha); /* A */ + } + break; + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(source[3], zero); /* A */ + break; + /* XXX fill in remaining terms */ + default: + abort(); + } + + + /* + * Compute dest/second term RGB + */ + switch (softpipe->blend.rgb_dst_factor) { + case PIPE_BLENDFACTOR_ONE: + /* dest = dest * 1 NO-OP, leave dest as-is */ + break; + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + { + GLfloat one_minus_alpha[QUAD_SIZE]; + VEC4_SUB(one_minus_alpha, one, quad->outputs.color[3]); + VEC4_MUL(dest[0], dest[0], one_minus_alpha); /* R */ + VEC4_MUL(dest[1], dest[1], one_minus_alpha); /* G */ + VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */ + } + break; + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(dest[0], zero); /* R */ + VEC4_COPY(dest[1], zero); /* G */ + VEC4_COPY(dest[2], zero); /* B */ + break; + /* XXX fill in remaining terms */ + default: + abort(); + } + + /* + * Compute dest/second term A */ - for (j = 0; j < QUAD_SIZE; j++) { - - /* if this pixel in the quad is alive */ - if (quad->mask & (1 << j)) { - - switch (softpipe->blend.rgb_src_factor) { - case PIPE_BLENDFACTOR_ONE: - srcTerm[0] = srcTerm[1] = srcTerm[2] = 1.0; - break; - case PIPE_BLENDFACTOR_SRC_ALPHA: - srcTerm[0] = srcTerm[1] = srcTerm[2] = quad->outputs.color[3][j]; - break; - case PIPE_BLENDFACTOR_ZERO: - srcTerm[0] = srcTerm[1] = srcTerm[2] = 0.0; - break; - /* XXX fill in remaining terms */ - default: - abort(); - } - - switch (softpipe->blend.alpha_src_factor) { - case PIPE_BLENDFACTOR_ONE: - srcTerm[3] = 1.0; - break; - case PIPE_BLENDFACTOR_SRC_ALPHA: - srcTerm[3] = quad->outputs.color[3][j]; - break; - case PIPE_BLENDFACTOR_ZERO: - srcTerm[3] = 0.0; - break; - /* XXX fill in remaining terms */ - default: - abort(); - } - - switch (softpipe->blend.rgb_dst_factor) { - case PIPE_BLENDFACTOR_ONE: - dstTerm[0] = dstTerm[1] = dstTerm[2] = 1.0; - break; - case PIPE_BLENDFACTOR_INV_SRC_ALPHA: - dstTerm[0] = dstTerm[1] = dstTerm[2] = 1.0 - quad->outputs.color[3][j]; - break; - case PIPE_BLENDFACTOR_ZERO: - dstTerm[0] = dstTerm[1] = dstTerm[2] = 0.0; - break; - /* XXX fill in remaining terms */ - default: - abort(); - } - - switch (softpipe->blend.alpha_dst_factor) { - case PIPE_BLENDFACTOR_ONE: - dstTerm[3] = 1.0; - break; - case PIPE_BLENDFACTOR_INV_SRC_ALPHA: - dstTerm[3] = 1.0 - quad->outputs.color[3][j]; - break; - case PIPE_BLENDFACTOR_ZERO: - dstTerm[3] = 0.0; - break; - /* XXX fill in remaining terms */ - default: - abort(); - } - - switch (softpipe->blend.rgb_func) { - case PIPE_BLEND_ADD: - quad->outputs.color[0][j] = source[0][j] * srcTerm[0] + dest[0][j] * dstTerm[0]; - quad->outputs.color[1][j] = source[1][j] * srcTerm[1] + dest[1][j] * dstTerm[1]; - quad->outputs.color[2][j] = source[2][j] * srcTerm[2] + dest[2][j] * dstTerm[2]; - quad->outputs.color[3][j] = source[3][j] * srcTerm[3] + dest[3][j] * dstTerm[3]; - break; - case PIPE_BLEND_SUBTRACT: - quad->outputs.color[0][j] = source[0][j] * srcTerm[0] - dest[0][j] * dstTerm[0]; - quad->outputs.color[1][j] = source[1][j] * srcTerm[1] - dest[1][j] * dstTerm[1]; - quad->outputs.color[2][j] = source[2][j] * srcTerm[2] - dest[2][j] * dstTerm[2]; - quad->outputs.color[3][j] = source[3][j] * srcTerm[3] - dest[3][j] * dstTerm[3]; - break; - case PIPE_BLEND_REVERSE_SUBTRACT: - quad->outputs.color[0][j] = dest[0][j] * dstTerm[0] - source[0][j] * srcTerm[0]; - quad->outputs.color[1][j] = dest[1][j] * dstTerm[1] - source[1][j] * srcTerm[1]; - quad->outputs.color[2][j] = dest[2][j] * dstTerm[2] - source[2][j] * srcTerm[2]; - quad->outputs.color[3][j] = dest[3][j] * dstTerm[3] - source[3][j] * srcTerm[3]; - break; - case PIPE_BLEND_MIN: - quad->outputs.color[0][j] = MIN2(dest[0][j], source[0][j]); - quad->outputs.color[1][j] = MIN2(dest[1][j], source[1][j]); - quad->outputs.color[2][j] = MIN2(dest[2][j], source[2][j]); - quad->outputs.color[3][j] = MIN2(dest[3][j], source[3][j]); - break; - case PIPE_BLEND_MAX: - quad->outputs.color[0][j] = MAX2(dest[0][j], source[0][j]); - quad->outputs.color[1][j] = MAX2(dest[1][j], source[1][j]); - quad->outputs.color[2][j] = MAX2(dest[2][j], source[2][j]); - quad->outputs.color[3][j] = MAX2(dest[3][j], source[3][j]); - break; - default: - abort(); - } + switch (softpipe->blend.alpha_dst_factor) { + case PIPE_BLENDFACTOR_ONE: + /* dest = dest * 1 NO-OP, leave dest as-is */ + break; + case PIPE_BLENDFACTOR_INV_SRC_ALPHA: + { + GLfloat one_minus_alpha[QUAD_SIZE]; + VEC4_SUB(one_minus_alpha, one, quad->outputs.color[3]); + VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* A */ } - } /* loop over quad's pixels*/ + break; + case PIPE_BLENDFACTOR_ZERO: + VEC4_COPY(dest[3], zero); /* A */ + break; + /* XXX fill in remaining terms */ + default: + abort(); + } + + /* + * Combine RGB terms + */ + switch (softpipe->blend.rgb_func) { + case PIPE_BLEND_ADD: + VEC4_ADD(quad->outputs.color[0], source[0], dest[0]); /* R */ + VEC4_ADD(quad->outputs.color[1], source[1], dest[1]); /* G */ + VEC4_ADD(quad->outputs.color[2], source[2], dest[2]); /* B */ + break; + case PIPE_BLEND_SUBTRACT: + VEC4_SUB(quad->outputs.color[0], source[0], dest[0]); /* R */ + VEC4_SUB(quad->outputs.color[1], source[1], dest[1]); /* G */ + VEC4_SUB(quad->outputs.color[2], source[2], dest[2]); /* B */ + break; + case PIPE_BLEND_REVERSE_SUBTRACT: + VEC4_SUB(quad->outputs.color[0], dest[0], source[0]); /* R */ + VEC4_SUB(quad->outputs.color[1], dest[1], source[1]); /* G */ + VEC4_SUB(quad->outputs.color[2], dest[2], source[2]); /* B */ + break; + case PIPE_BLEND_MIN: + VEC4_MIN(quad->outputs.color[0], source[0], dest[0]); /* R */ + VEC4_MIN(quad->outputs.color[1], source[1], dest[1]); /* G */ + VEC4_MIN(quad->outputs.color[2], source[2], dest[2]); /* B */ + break; + case PIPE_BLEND_MAX: + VEC4_MAX(quad->outputs.color[0], source[0], dest[0]); /* R */ + VEC4_MAX(quad->outputs.color[1], source[1], dest[1]); /* G */ + VEC4_MAX(quad->outputs.color[2], source[2], dest[2]); /* B */ + break; + default: + abort(); + } + + /* + * Combine A terms + */ + switch (softpipe->blend.alpha_func) { + case PIPE_BLEND_ADD: + VEC4_ADD(quad->outputs.color[3], source[3], dest[3]); /* A */ + break; + case PIPE_BLEND_SUBTRACT: + VEC4_SUB(quad->outputs.color[3], source[3], dest[3]); /* A */ + break; + case PIPE_BLEND_REVERSE_SUBTRACT: + VEC4_SUB(quad->outputs.color[3], dest[3], source[3]); /* A */ + break; + case PIPE_BLEND_MIN: + VEC4_MIN(quad->outputs.color[3], source[3], dest[3]); /* A */ + break; + case PIPE_BLEND_MAX: + VEC4_MAX(quad->outputs.color[3], source[3], dest[3]); /* A */ + default: + abort(); + } /* pass blended quad to next stage */ qs->next->run(qs->next, quad); |