X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fauxiliary%2Fdraw%2Fdraw_pt_fetch_shade_emit.c;h=734c05f0688ab19a1706bdb6047cfbed4497c0d3;hb=1f196b786d6bd0c6a5dbdc638574ff716cc3d4de;hp=c6249b4b41044503e25929ff86fda568fa116876;hpb=9752ebd99e16646fed3c14712fc0af2c34c9e48f;p=mesa.git diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c index c6249b4b410..734c05f0688 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c @@ -31,7 +31,8 @@ */ -#include "pipe/p_util.h" +#include "util/u_math.h" +#include "util/u_memory.h" #include "draw/draw_context.h" #include "draw/draw_private.h" #include "draw/draw_vbuf.h" @@ -60,6 +61,9 @@ struct fetch_shade_emit { struct draw_vs_varient_key key; struct draw_vs_varient *active; + + + const struct vertex_info *vinfo; }; @@ -67,15 +71,15 @@ struct fetch_shade_emit { static void fse_prepare( struct draw_pt_middle_end *middle, unsigned prim, - unsigned opt ) + unsigned opt, + unsigned *max_vertices ) { struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; struct draw_context *draw = fse->draw; unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs; - unsigned num_vs_outputs = draw->vs.vertex_shader->info.num_outputs; const struct vertex_info *vinfo; unsigned i; - boolean need_psize = 0; + unsigned nr_vbs = 0; if (!draw->render->set_primitive( draw->render, @@ -86,17 +90,21 @@ static void fse_prepare( struct draw_pt_middle_end *middle, /* Must do this after set_primitive() above: */ - vinfo = draw->render->get_vertex_info(draw->render); + fse->vinfo = vinfo = draw->render->get_vertex_info(draw->render); fse->key.output_stride = vinfo->size * 4; - fse->key.nr_elements = MAX2(num_vs_outputs, /* outputs - translate to hw format */ - num_vs_inputs); /* inputs - fetch from api format */ + fse->key.nr_outputs = vinfo->num_attribs; + fse->key.nr_inputs = num_vs_inputs; - fse->key.viewport = !draw->identity_viewport; + fse->key.nr_elements = MAX2(fse->key.nr_outputs, /* outputs - translate to hw format */ + fse->key.nr_inputs); /* inputs - fetch from api format */ + + fse->key.viewport = (!draw->rasterizer->bypass_vs_clip_and_viewport && + !draw->identity_viewport); fse->key.clip = !draw->bypass_clipping; - fse->key.pad = 0; + fse->key.const_vbuffers = 0; memset(fse->key.element, 0, fse->key.nr_elements * sizeof(fse->key.element[0])); @@ -110,42 +118,39 @@ static void fse_prepare( struct draw_pt_middle_end *middle, */ fse->key.element[i].in.buffer = src->vertex_buffer_index; fse->key.element[i].in.offset = src->src_offset; + nr_vbs = MAX2(nr_vbs, src->vertex_buffer_index + 1); } + for (i = 0; i < 5 && i < nr_vbs; i++) { + if (draw->pt.vertex_buffer[i].stride == 0) + fse->key.const_vbuffers |= (1<key.const_vbuffers); + { unsigned dst_offset = 0; for (i = 0; i < vinfo->num_attribs; i++) { unsigned emit_sz = 0; - unsigned output_format = PIPE_FORMAT_NONE; - unsigned vs_output = vinfo->src_index[i]; - switch (vinfo->emit[i]) { + switch (vinfo->attrib[i].emit) { case EMIT_4F: - output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; emit_sz = 4 * sizeof(float); break; case EMIT_3F: - output_format = PIPE_FORMAT_R32G32B32_FLOAT; emit_sz = 3 * sizeof(float); break; case EMIT_2F: - output_format = PIPE_FORMAT_R32G32_FLOAT; emit_sz = 2 * sizeof(float); break; case EMIT_1F: - output_format = PIPE_FORMAT_R32_FLOAT; emit_sz = 1 * sizeof(float); break; case EMIT_1F_PSIZE: - need_psize = 1; - output_format = PIPE_FORMAT_R32_FLOAT; emit_sz = 1 * sizeof(float); - vs_output = num_vs_outputs + 1; break; case EMIT_4UB: - output_format = PIPE_FORMAT_B8G8R8A8_UNORM; emit_sz = 4 * sizeof(ubyte); break; default: @@ -157,37 +162,16 @@ static void fse_prepare( struct draw_pt_middle_end *middle, * numbers, not to positions in the hw vertex description -- * that's handled by the output_offset field. */ - fse->key.element[vs_output].out.format = output_format; - fse->key.element[vs_output].out.offset = dst_offset; + fse->key.element[i].out.format = vinfo->attrib[i].emit; + fse->key.element[i].out.vs_output = vinfo->attrib[i].src_index; + fse->key.element[i].out.offset = dst_offset; dst_offset += emit_sz; assert(fse->key.output_stride >= dst_offset); } } - /* To make psize work, really need to tell the vertex shader to - * copy that value from input->output. For 'translate' this was - * implicit for all elements. - */ -#if 0 - if (need_psize) { - unsigned input = num_vs_inputs + 1; - const struct pipe_vertex_element *src = &draw->pt.vertex_element[i]; - fse->key.element[i].input_format = PIPE_FORMAT_R32_FLOAT; - fse->key.element[i].input_buffer = 0; //nr_buffers + 1; - fse->key.element[i].input_offset = 0; - - fse->key.nr_elements += 1; - - } -#endif - - /* Would normally look up a vertex shader and peruse its list of - * varients somehow. We omitted that step and put all the - * hardcoded "shaders" into an array. We're just making the - * assumption that this happens to be a matching shader... ie - * you're running isosurf, aren't you? - */ + fse->active = draw_vs_lookup_varient( draw->vs.vertex_shader, &fse->key ); @@ -196,56 +180,65 @@ static void fse_prepare( struct draw_pt_middle_end *middle, return ; } + if (0) debug_printf("%s: found const_vbuffers: %x\n", __FUNCTION__, + fse->active->key.const_vbuffers); + /* Now set buffer pointers: */ - for (i = 0; i < num_vs_inputs; i++) { - unsigned buf = draw->pt.vertex_element[i].vertex_buffer_index; - - fse->active->set_input( fse->active, - i, - - ((const ubyte *) draw->pt.user.vbuffer[buf] + - draw->pt.vertex_buffer[buf].buffer_offset), - - draw->pt.vertex_buffer[buf].pitch ); + for (i = 0; i < draw->pt.nr_vertex_buffers; i++) { + fse->active->set_buffer( fse->active, + i, + ((const ubyte *) draw->pt.user.vbuffer[i] + + draw->pt.vertex_buffer[i].buffer_offset), + draw->pt.vertex_buffer[i].stride ); } - fse->active->set_constants( fse->active, - (const float (*)[4])draw->pt.user.constants ); + *max_vertices = (draw->render->max_vertex_buffer_bytes / + (vinfo->size * 4)); - fse->active->set_viewport( fse->active, - &draw->viewport ); + /* Return an even number of verts. + * This prevents "parity" errors when splitting long triangle strips which + * can lead to front/back culling mix-ups. + * Every other triangle in a strip has an alternate front/back orientation + * so splitting at an odd position can cause the orientation of subsequent + * triangles to get reversed. + */ + *max_vertices = *max_vertices & ~1; - //return TRUE; + /* Probably need to do this somewhere (or fix exec shader not to + * need it): + */ + if (1) { + struct draw_vertex_shader *vs = draw->vs.vertex_shader; + vs->prepare(vs, draw); + } } - - - - static void fse_run_linear( struct draw_pt_middle_end *middle, unsigned start, unsigned count ) { struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; struct draw_context *draw = fse->draw; - unsigned alloc_count = align(count, 4); char *hw_verts; /* XXX: need to flush to get prim_vbuf.c to release its allocation?? */ draw_do_flush( draw, DRAW_FLUSH_BACKEND ); - hw_verts = draw->render->allocate_vertices( draw->render, - (ushort)fse->key.output_stride, - (ushort)alloc_count ); + if (count >= UNDEFINED_VERTEX_ID) + goto fail; - if (!hw_verts) { - assert(0); - return; - } + if (!draw->render->allocate_vertices( draw->render, + (ushort)fse->key.output_stride, + (ushort)count )) + goto fail; + + hw_verts = draw->render->map_vertices( draw->render ); + if (!hw_verts) + goto fail; /* Single routine to fetch vertices, run shader and emit HW verts. * Clipping is done elsewhere -- either by the API or on hardware, @@ -255,6 +248,21 @@ static void fse_run_linear( struct draw_pt_middle_end *middle, start, count, hw_verts ); + + if (0) { + unsigned i; + for (i = 0; i < count; i++) { + debug_printf("\n\n%s vertex %d: (stride %d, offset %d)\n", __FUNCTION__, i, + fse->key.output_stride, + fse->key.output_stride * i); + + draw_dump_emitted_vertex( fse->vinfo, + (const uint8_t *)hw_verts + fse->key.output_stride * i ); + } + } + + draw->render->unmap_vertices( draw->render, 0, (ushort)(count - 1) ); + /* Draw arrays path to avoid re-emitting index list again and * again. */ @@ -263,10 +271,13 @@ static void fse_run_linear( struct draw_pt_middle_end *middle, count ); - draw->render->release_vertices( draw->render, - hw_verts, - fse->key.output_stride, - count ); + draw->render->release_vertices( draw->render ); + + return; + +fail: + assert(0); + return; } @@ -279,20 +290,23 @@ fse_run(struct draw_pt_middle_end *middle, { struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; struct draw_context *draw = fse->draw; - unsigned alloc_count = align(fetch_count, 4); void *hw_verts; /* XXX: need to flush to get prim_vbuf.c to release its allocation?? */ draw_do_flush( draw, DRAW_FLUSH_BACKEND ); - hw_verts = draw->render->allocate_vertices( draw->render, - (ushort)fse->key.output_stride, - (ushort)alloc_count ); - if (!hw_verts) { - assert(0); - return; - } + if (fetch_count >= UNDEFINED_VERTEX_ID) + goto fail; + + if (!draw->render->allocate_vertices( draw->render, + (ushort)fse->key.output_stride, + (ushort)fetch_count )) + goto fail; + + hw_verts = draw->render->map_vertices( draw->render ); + if (!hw_verts) + goto fail; /* Single routine to fetch vertices, run shader and emit HW verts. @@ -302,18 +316,83 @@ fse_run(struct draw_pt_middle_end *middle, fetch_count, hw_verts ); + + if (0) { + unsigned i; + for (i = 0; i < fetch_count; i++) { + debug_printf("\n\n%s vertex %d:\n", __FUNCTION__, i); + draw_dump_emitted_vertex( fse->vinfo, + (const uint8_t *)hw_verts + + fse->key.output_stride * i ); + } + } + + draw->render->unmap_vertices( draw->render, 0, (ushort)(fetch_count - 1) ); + draw->render->draw( draw->render, draw_elts, draw_count ); - draw->render->release_vertices( draw->render, - hw_verts, - fse->key.output_stride, - fetch_count ); + draw->render->release_vertices( draw->render ); + return; + +fail: + assert(0); + return; } + +static boolean fse_run_linear_elts( struct draw_pt_middle_end *middle, + unsigned start, + unsigned count, + const ushort *draw_elts, + unsigned draw_count ) +{ + struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle; + struct draw_context *draw = fse->draw; + char *hw_verts; + + /* XXX: need to flush to get prim_vbuf.c to release its allocation?? + */ + draw_do_flush( draw, DRAW_FLUSH_BACKEND ); + + if (count >= UNDEFINED_VERTEX_ID) + return FALSE; + + if (!draw->render->allocate_vertices( draw->render, + (ushort)fse->key.output_stride, + (ushort)count )) + return FALSE; + + hw_verts = draw->render->map_vertices( draw->render ); + if (!hw_verts) + return FALSE; + + /* Single routine to fetch vertices, run shader and emit HW verts. + * Clipping is done elsewhere -- either by the API or on hardware, + * or for some other reason not required... + */ + fse->active->run_linear( fse->active, + start, count, + hw_verts ); + + + draw->render->draw( draw->render, + draw_elts, + draw_count ); + + + draw->render->unmap_vertices( draw->render, 0, (ushort)(count - 1) ); + + draw->render->release_vertices( draw->render ); + + return TRUE; +} + + + static void fse_finish( struct draw_pt_middle_end *middle ) { } @@ -334,6 +413,7 @@ struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw ) fse->base.prepare = fse_prepare; fse->base.run = fse_run; fse->base.run_linear = fse_run_linear; + fse->base.run_linear_elts = fse_run_linear_elts; fse->base.finish = fse_finish; fse->base.destroy = fse_destroy; fse->draw = draw;