diff options
Diffstat (limited to 'src/gallium/auxiliary/draw')
20 files changed, 570 insertions, 403 deletions
diff --git a/src/gallium/auxiliary/draw/Makefile b/src/gallium/auxiliary/draw/Makefile index 28262a92c6..5ab3cfe5ce 100644 --- a/src/gallium/auxiliary/draw/Makefile +++ b/src/gallium/auxiliary/draw/Makefile @@ -19,6 +19,7 @@ C_SOURCES = \ draw_pt_vcache.c \ draw_pt_fetch_emit.c \ draw_pt_fetch_pipeline.c \ + draw_pt_fetch_shade_pipeline.c \ draw_pt_pipeline.c \ draw_pt_elts.c \ draw_prim.c \ diff --git a/src/gallium/auxiliary/draw/SConscript b/src/gallium/auxiliary/draw/SConscript index 52107912f5..a7fb5dbd61 100644 --- a/src/gallium/auxiliary/draw/SConscript +++ b/src/gallium/auxiliary/draw/SConscript @@ -18,6 +18,7 @@ draw = env.ConvenienceLibrary( 'draw_pt_vcache.c', 'draw_pt_fetch_emit.c', 'draw_pt_fetch_pipeline.c', + 'draw_pt_fetch_shade_pipeline.c', 'draw_pt_pipeline.c', 'draw_pt_elts.c', 'draw_prim.c', diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c index b3c65c90d6..0c314f6e1d 100644 --- a/src/gallium/auxiliary/draw/draw_context.c +++ b/src/gallium/auxiliary/draw/draw_context.c @@ -49,6 +49,8 @@ struct draw_context *draw_create( void ) draw->use_sse = FALSE; #endif + draw->use_pt_shaders = GETENV( "GALLIUM_PT_SHADERS" ) != NULL; + /* create pipeline stages */ draw->pipeline.wide_line = draw_wide_line_stage( draw ); draw->pipeline.wide_point = draw_wide_point_stage( draw ); @@ -86,14 +88,11 @@ struct draw_context *draw_create( void ) /* Statically allocate maximum sized vertices for the cache - could be cleverer... */ { - uint i; - const unsigned size = (MAX_VERTEX_SIZE + 0x0f) & ~0x0f; - char *tmp = align_malloc(Elements(draw->vs.queue) * size, 16); + char *tmp = align_malloc(VS_QUEUE_LENGTH * MAX_VERTEX_ALLOCATION, 16); if (!tmp) goto fail; - for (i = 0; i < Elements(draw->vs.queue); i++) - draw->vs.queue[i].vertex = (struct vertex_header *)(tmp + i * size); + draw->vs.vertex_cache = tmp; } draw->shader_queue_flush = draw_vertex_shader_queue_flush; @@ -156,8 +155,8 @@ void draw_destroy( struct draw_context *draw ) tgsi_exec_machine_free_data(&draw->machine); - if (draw->vs.queue[0].vertex) - align_free( draw->vs.queue[0].vertex ); /* Frees all the vertices. */ + if (draw->vs.vertex_cache) + align_free( draw->vs.vertex_cache ); /* Frees all the vertices. */ /* Not so fast -- we're just borrowing this at the moment. * diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h index 4d056f6dba..c8cb96c8ba 100644 --- a/src/gallium/auxiliary/draw/draw_private.h +++ b/src/gallium/auxiliary/draw/draw_private.h @@ -56,6 +56,8 @@ struct gallivm_cpu_engine; struct draw_pt_middle_end; struct draw_pt_front_end; +#define MAX_SHADER_VERTICES 128 + /** * Basic vertex info. * Carry some useful information around with the vertices in the prim pipe. @@ -76,6 +78,7 @@ struct vertex_header { /* XXX This is too large */ #define MAX_VERTEX_SIZE ((2 + PIPE_MAX_SHADER_OUTPUTS) * 4 * sizeof(float)) +#define MAX_VERTEX_ALLOCATION ((MAX_VERTEX_SIZE + 0x0f) & ~0x0f) @@ -146,12 +149,13 @@ struct draw_vertex_shader { /* Run the shader - this interface will get cleaned up in the * future: */ - void (*run)( struct draw_vertex_shader *shader, - struct draw_context *draw, - const unsigned *elts, - unsigned count, - struct vertex_header *vOut[] ); - + boolean (*run)( struct draw_vertex_shader *shader, + struct draw_context *draw, + const unsigned *elts, + unsigned count, + void *out, + unsigned vertex_size); + void (*delete)( struct draw_vertex_shader * ); }; @@ -274,6 +278,7 @@ struct draw_context boolean line_stipple; /**< do line stipple? */ boolean point_sprite; /**< convert points to quads for sprites? */ boolean use_sse; + boolean use_pt_shaders; /* temporary flag to switch on pt shader paths */ /* If a prim stage introduces new vertex attributes, they'll be stored here */ @@ -319,10 +324,8 @@ struct draw_context /* Vertex shader queue: */ struct { - struct { - unsigned elt; /**< index into the user's vertex arrays */ - struct vertex_header *vertex; - } queue[VS_QUEUE_LENGTH]; + unsigned elts[VS_QUEUE_LENGTH]; /**< index into the user's vertex arrays */ + char *vertex_cache; unsigned queue_nr; unsigned post_nr; } vs; @@ -448,4 +451,10 @@ dot4(const float *a, const float *b) return result; } +static INLINE struct vertex_header * +draw_header_from_block(char *block, int size, int num) +{ + return (struct vertex_header*)(block + num * size); +} + #endif /* DRAW_PRIVATE_H */ diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c index f59fb86f78..3d2e7bf7b8 100644 --- a/src/gallium/auxiliary/draw/draw_pt.c +++ b/src/gallium/auxiliary/draw/draw_pt.c @@ -36,21 +36,25 @@ #include "draw/draw_pt.h" -/* XXX: Shouldn't those two functions below use the '>' operator??? - */ - -static boolean too_many_verts( struct draw_context *draw, - unsigned verts ) +#if 0 +static boolean too_many_elts( struct draw_context *draw, + unsigned elts ) { - return verts < 1024; + return elts > (8 * 1024); } +#endif -static boolean too_many_elts( struct draw_context *draw, - unsigned elts ) +static INLINE unsigned reduced_prim(unsigned prim) { - return elts < (16 * 1024); + /*FIXME*/ + return prim; } +static INLINE boolean good_prim(unsigned prim) +{ + /*FIXME*/ + return FALSE; +} boolean draw_pt_arrays(struct draw_context *draw, @@ -64,6 +68,9 @@ draw_pt_arrays(struct draw_context *draw, struct draw_pt_front_end *frontend = NULL; struct draw_pt_middle_end *middle = NULL; + if (!draw->render) + return FALSE; + /*debug_printf("XXXXXXXXXX needs_pipeline = %d\n", pipeline);*/ /* Overall we do: * - frontend -- prepare fetch_elts, draw_elts - eg vcache @@ -72,6 +79,9 @@ draw_pt_arrays(struct draw_context *draw, * - backend -- the vbuf_render provided by the driver. */ + if (shading && !draw->use_pt_shaders) + return FALSE; + if (!cliptest && !pipeline && !shading) { /* This is the 'passthrough' path: @@ -87,7 +97,6 @@ draw_pt_arrays(struct draw_context *draw, */ middle = draw->pt.middle.fetch_pipeline; } -#if 0 else if (!cliptest && !pipeline) { /* Fetch user verts, run vertex shader, emit hw verts: */ @@ -111,23 +120,15 @@ draw_pt_arrays(struct draw_context *draw, */ middle = draw->pt.middle.fetch_shade_cliptest_pipeline_or_emit; } - else if (!cliptest) { - /* Fetch user verts, run vertex shader, run pipeline: - */ - middle = draw->pt.middle.fetch_shade_pipeline; - } else { /* This is what we're currently always doing: */ - /* Fetch user verts, run vertex shader, cliptest, run pipeline: + /* Fetch user verts, run vertex shader, cliptest, run pipeline + * or + * Fetch user verts, run vertex shader, run pipeline */ - middle = draw->pt.middle.fetch_shade_cliptest_pipeline; - } -#else - else { - return FALSE; + middle = draw->pt.middle.fetch_shade_cliptest_pipeline_or_emit; } -#endif /* If !pipeline, need to make sure we respect the driver's limited @@ -143,7 +144,7 @@ draw_pt_arrays(struct draw_context *draw, frontend = draw->pt.front.vcache; hw_prim = reduced_prim(prim); } - +#if 0 if (too_many_verts(nr_verts)) { /* if (is_verts(draw) && can_split(prim)) { draw = draw_arrays_split; @@ -153,6 +154,7 @@ draw_pt_arrays(struct draw_context *draw, hw_prim = reduced_prim(prim); } } +#endif if (too_many_elts(count)) { @@ -166,17 +168,13 @@ draw_pt_arrays(struct draw_context *draw, } if (!good_prim(hw_prim)) { - draw = draw->pt.front.vcache; + frontend = draw->pt.front.vcache; } } #else frontend = draw->pt.front.vcache; #endif - /* XXX: need to flush to get prim_vbuf.c to release its allocation?? - */ - draw_do_flush( draw, DRAW_FLUSH_BACKEND ); - frontend->prepare( frontend, prim, middle ); frontend->run( frontend, @@ -200,6 +198,11 @@ boolean draw_pt_init( struct draw_context *draw ) if (!draw->pt.middle.fetch_pipeline) return FALSE; + draw->pt.middle.fetch_shade_cliptest_pipeline_or_emit = + draw_pt_fetch_pipeline_or_emit( draw ); + if (!draw->pt.middle.fetch_shade_cliptest_pipeline_or_emit) + return FALSE; + draw->pt.front.vcache = draw_pt_vcache( draw ); if (!draw->pt.front.vcache) return FALSE; @@ -220,6 +223,12 @@ void draw_pt_destroy( struct draw_context *draw ) draw->pt.middle.fetch_pipeline = NULL; } + if (draw->pt.middle.fetch_shade_cliptest_pipeline_or_emit) { + draw->pt.middle.fetch_shade_cliptest_pipeline_or_emit->destroy( + draw->pt.middle.fetch_shade_cliptest_pipeline_or_emit ); + draw->pt.middle.fetch_shade_cliptest_pipeline_or_emit = NULL; + } + if (draw->pt.front.vcache) { draw->pt.front.vcache->destroy( draw->pt.front.vcache ); draw->pt.front.vcache = NULL; diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h index 590823fd33..48413b648a 100644 --- a/src/gallium/auxiliary/draw/draw_pt.h +++ b/src/gallium/auxiliary/draw/draw_pt.h @@ -123,7 +123,7 @@ const void *draw_pt_elt_ptr( struct draw_context *draw, struct draw_pt_front_end *draw_pt_vcache( struct draw_context *draw ); struct draw_pt_middle_end *draw_pt_fetch_emit( struct draw_context *draw ); struct draw_pt_middle_end *draw_pt_fetch_pipeline( struct draw_context *draw ); - +struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit(struct draw_context *draw); #endif diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c index 0806076956..3a26a5d712 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c @@ -145,7 +145,6 @@ fetch_store_general( struct fetch_emit_middle_end *feme, unsigned count ) { float *out = (float *)out_ptr; - struct vbuf_render *render = feme->draw->render; uint i, j; for (i = 0; i < count; i++) { @@ -167,7 +166,6 @@ fetch_store_general( struct fetch_emit_middle_end *feme, static void fetch_emit_prepare( struct draw_pt_middle_end *middle, unsigned prim ) { - static const float zero = 0; struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle; struct draw_context *draw = feme->draw; const struct vertex_info *vinfo; @@ -212,12 +210,6 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle, case EMIT_1F: feme->fetch[i].emit = emit_R32_FLOAT; break; - case EMIT_HEADER: - feme->fetch[i].ptr = (const ubyte *)&zero; - feme->fetch[i].pitch = 0; - feme->fetch[i].fetch = fetch_R32_FLOAT; - feme->fetch[i].emit = emit_R32_FLOAT; - break; case EMIT_1F_PSIZE: feme->fetch[i].ptr = (const ubyte *)&feme->draw->rasterizer->point_size; feme->fetch[i].pitch = 0; @@ -249,6 +241,10 @@ static void fetch_emit_run( struct draw_pt_middle_end *middle, struct draw_context *draw = feme->draw; 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)feme->hw_vertex_size, (ushort)fetch_count ); diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_pipeline.c index 4c2a281b29..a70d129c93 100644 --- a/src/gallium/auxiliary/draw/draw_pt_fetch_pipeline.c +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_pipeline.c @@ -87,7 +87,7 @@ struct fetch_pipeline_middle_end { }; - +#if 0 static void emit_R32_FLOAT( const float *attrib, float **out ) { @@ -111,7 +111,7 @@ static void emit_R32G32B32_FLOAT( const float *attrib, (*out)[2] = attrib[2]; (*out) += 3; } - +#endif static void emit_R32G32B32A32_FLOAT( const float *attrib, float **out ) { @@ -211,7 +211,6 @@ fetch_store_general( struct fetch_pipeline_middle_end *fpme, static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle, unsigned prim ) { - static const float zero = 0; struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle; struct draw_context *draw = fpme->draw; unsigned i, nr = 0; @@ -264,7 +263,6 @@ static void fetch_pipeline_run( struct draw_pt_middle_end *middle, unsigned draw_count ) { struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle; - struct draw_context *draw = fpme->draw; char *pipeline_verts; pipeline_verts = MALLOC( fpme->pipeline_vertex_size * diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c new file mode 100644 index 0000000000..04b3d2c4cf --- /dev/null +++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c @@ -0,0 +1,243 @@ +/************************************************************************** + * + * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas. + * All Rights Reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the + * "Software"), to deal in the Software without restriction, including + * without limitation the rights to use, copy, modify, merge, publish, + * distribute, sub license, and/or sell copies of the Software, and to + * permit persons to whom the Software is furnished to do so, subject to + * the following conditions: + * + * The above copyright notice and this permission notice (including the + * next paragraph) shall be included in all copies or substantial portions + * of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + * + **************************************************************************/ + +#include "pipe/p_util.h" +#include "draw/draw_context.h" +#include "draw/draw_private.h" +#include "draw/draw_vbuf.h" +#include "draw/draw_vertex.h" +#include "draw/draw_pt.h" + +struct fetch_pipeline_middle_end { + struct draw_pt_middle_end base; + struct draw_context *draw; + + struct { + const ubyte *ptr; + unsigned pitch; + void (*fetch)( const void *from, float *attrib); + void (*emit)( const float *attrib, float **out ); + } fetch[PIPE_MAX_ATTRIBS]; + + unsigned nr_fetch; + unsigned pipeline_vertex_size; + unsigned hw_vertex_size; + unsigned prim; +}; + +#if 0 +static void emit_R32_FLOAT( const float *attrib, + float **out ) +{ + (*out)[0] = attrib[0]; + (*out) += 1; +} + +static void emit_R32G32_FLOAT( const float *attrib, + float **out ) +{ + (*out)[0] = attrib[0]; + (*out)[1] = attrib[1]; + (*out) += 2; +} + +static void emit_R32G32B32_FLOAT( const float *attrib, + float **out ) +{ + (*out)[0] = attrib[0]; + (*out)[1] = attrib[1]; + (*out)[2] = attrib[2]; + (*out) += 3; +} +#endif +static void emit_R32G32B32A32_FLOAT( const float *attrib, + float **out ) +{ + (*out)[0] = attrib[0]; + (*out)[1] = attrib[1]; + (*out)[2] = attrib[2]; + (*out)[3] = attrib[3]; + (*out) += 4; +} + +static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle, + unsigned prim ) +{ + struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle; + struct draw_context *draw = fpme->draw; + unsigned i, nr = 0; + boolean ok; + const struct vertex_info *vinfo; + + fpme->prim = prim; + + ok = draw->render->set_primitive(draw->render, prim); + if (!ok) { + assert(0); + return; + } + /* Must do this after set_primitive() above: + */ + vinfo = draw->render->get_vertex_info(draw->render); + + /* Need to look at vertex shader inputs (we know it is a + * passthrough shader, so these define the outputs too). If we + * were running a shader, we'd still be looking at the inputs at + * this point. + */ + for (i = 0; i < draw->vertex_shader->info.num_inputs; i++) { + unsigned buf = draw->vertex_element[i].vertex_buffer_index; + enum pipe_format format = draw->vertex_element[i].src_format; + + fpme->fetch[nr].ptr = ((const ubyte *) draw->user.vbuffer[buf] + + draw->vertex_buffer[buf].buffer_offset + + draw->vertex_element[i].src_offset); + + fpme->fetch[nr].pitch = draw->vertex_buffer[buf].pitch; + fpme->fetch[nr].fetch = draw_get_fetch_func( format ); + + /* Always do this -- somewhat redundant... + */ + fpme->fetch[nr].emit = emit_R32G32B32A32_FLOAT; + nr++; + } + + fpme->nr_fetch = nr; + //fpme->pipeline_vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float); + fpme->pipeline_vertex_size = MAX_VERTEX_ALLOCATION; + fpme->hw_vertex_size = vinfo->size * 4; +} + + + + +static void fetch_pipeline_run( struct draw_pt_middle_end *middle, + const unsigned *fetch_elts, + unsigned fetch_count, + const ushort *draw_elts, + unsigned draw_count ) +{ + struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle; + struct draw_context *draw = fpme->draw; + struct draw_vertex_shader *shader = draw->vertex_shader; + char *pipeline_verts; + + pipeline_verts = MALLOC(fpme->pipeline_vertex_size * + fetch_count); + + if (!pipeline_verts) { + assert(0); + return; + } + + + /* Shade + */ + shader->prepare(shader, draw); + if (shader->run(shader, draw, fetch_elts, fetch_count, pipeline_verts, + fpme->pipeline_vertex_size)) { + /* Run the pipeline */ + draw_pt_run_pipeline( fpme->draw, + fpme->prim, + pipeline_verts, + fpme->pipeline_vertex_size, + fetch_count, + draw_elts, + draw_count ); + } else { + unsigned i, j; + void *hw_verts; + float *out; + + /* 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)fpme->hw_vertex_size, + (ushort)fetch_count); + if (!hw_verts) { + assert(0); + return; + } + + out = (float *)hw_verts; + for (i = 0; i < fetch_count; i++) { + struct vertex_header *header = + (struct vertex_header*)(pipeline_verts + (fpme->pipeline_vertex_size * i)); + + for (j = 0; j < fpme->nr_fetch; j++) { + float *attrib = header->data[j]; + /*debug_printf("emiting [%f, %f, %f, %f]\n", + attrib[0], attrib[1], + attrib[2], attrib[3]);*/ + fpme->fetch[j].emit(attrib, &out); + } + } + /* XXX: Draw arrays path to avoid re-emitting index list again and + * again. + */ + draw->render->draw(draw->render, + draw_elts, + draw_count); + + draw->render->release_vertices(draw->render, + hw_verts, + fpme->hw_vertex_size, + fetch_count); + } + + + FREE(pipeline_verts); +} + + + +static void fetch_pipeline_finish( struct draw_pt_middle_end *middle ) +{ + /* nothing to do */ +} + +static void fetch_pipeline_destroy( struct draw_pt_middle_end *middle ) +{ + FREE(middle); +} + + +struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit( struct draw_context *draw ) +{ + struct fetch_pipeline_middle_end *fetch_pipeline = CALLOC_STRUCT( fetch_pipeline_middle_end ); + + fetch_pipeline->base.prepare = fetch_pipeline_prepare; + fetch_pipeline->base.run = fetch_pipeline_run; + fetch_pipeline->base.finish = fetch_pipeline_finish; + fetch_pipeline->base.destroy = fetch_pipeline_destroy; + + fetch_pipeline->draw = draw; + + return &fetch_pipeline->base; +} diff --git a/src/gallium/auxiliary/draw/draw_vbuf.c b/src/gallium/auxiliary/draw/draw_vbuf.c index f83b441e93..e3216ff711 100644 --- a/src/gallium/auxiliary/draw/draw_vbuf.c +++ b/src/gallium/auxiliary/draw/draw_vbuf.c @@ -126,13 +126,6 @@ dump_emitted_vertex(const struct vertex_info *vinfo, const uint8_t *data) case EMIT_OMIT: debug_printf("EMIT_OMIT:"); break; - case EMIT_ALL: - assert(i == 0); - assert(j == 0); - debug_printf("EMIT_ALL:\t"); - for(k = 0; k < vinfo->size*4; ++k) - debug_printf("%02x ", *data++); - break; case EMIT_1F: debug_printf("EMIT_1F:\t"); debug_printf("%f ", *(float *)data); data += sizeof(float); @@ -217,19 +210,6 @@ emit_vertex( struct vbuf_stage *vbuf, case EMIT_OMIT: /* no-op */ break; - case EMIT_ALL: - /* just copy the whole vertex as-is to the vbuf */ - assert(i == 0); - assert(j == 0); - memcpy(vbuf->vertex_ptr, vertex, vinfo->size * 4); - vbuf->vertex_ptr += vinfo->size; - count += vinfo->size; - break; - case EMIT_HEADER: - memcpy(vbuf->vertex_ptr, vertex, sizeof(*vertex)); - *vbuf->vertex_ptr += sizeof(*vertex) / 4; - count += sizeof(*vertex) / 4; - break; case EMIT_1F: *vbuf->vertex_ptr++ = fui(vertex->data[j][0]); count++; diff --git a/src/gallium/auxiliary/draw/draw_vertex.c b/src/gallium/auxiliary/draw/draw_vertex.c index 970adc95e7..168036eee8 100644 --- a/src/gallium/auxiliary/draw/draw_vertex.c +++ b/src/gallium/auxiliary/draw/draw_vertex.c @@ -52,9 +52,6 @@ draw_compute_vertex_size(struct vertex_info *vinfo) switch (vinfo->emit[i]) { case EMIT_OMIT: break; - case EMIT_HEADER: - vinfo->size += sizeof(struct vertex_header) / 4; - break; case EMIT_4UB: /* fall-through */ case EMIT_1F_PSIZE: @@ -71,8 +68,6 @@ draw_compute_vertex_size(struct vertex_info *vinfo) case EMIT_4F: vinfo->size += 4; break; - case EMIT_ALL: - /* fall-through */ default: assert(0); } diff --git a/src/gallium/auxiliary/draw/draw_vertex.h b/src/gallium/auxiliary/draw/draw_vertex.h index abd2017ed3..65818463ca 100644 --- a/src/gallium/auxiliary/draw/draw_vertex.h +++ b/src/gallium/auxiliary/draw/draw_vertex.h @@ -47,8 +47,6 @@ */ enum attrib_emit { EMIT_OMIT, /**< don't emit the attribute */ - EMIT_ALL, /**< emit whole post-xform vertex, w/ header */ - EMIT_HEADER, /**< emit vertex_header struct (XXX temp?) */ EMIT_1F, EMIT_1F_PSIZE, /**< insert constant point size */ EMIT_2F, diff --git a/src/gallium/auxiliary/draw/draw_vertex_cache.c b/src/gallium/auxiliary/draw/draw_vertex_cache.c index c0248979e2..730c18bcb3 100644 --- a/src/gallium/auxiliary/draw/draw_vertex_cache.c +++ b/src/gallium/auxiliary/draw/draw_vertex_cache.c @@ -71,25 +71,28 @@ static struct vertex_header *get_vertex( struct draw_context *draw, /* Cache hit? */ if (draw->vcache.idx[slot].in == i) { -// _mesa_printf("HIT %d %d\n", slot, i); + /*debug_printf("HIT %d %d\n", slot, i);*/ assert(draw->vcache.idx[slot].out < draw->vs.queue_nr); - return draw->vs.queue[draw->vcache.idx[slot].out].vertex; + return draw_header_from_block(draw->vs.vertex_cache, + MAX_VERTEX_ALLOCATION, + draw->vcache.idx[slot].out); } /* Otherwise a collision */ slot = VCACHE_SIZE + draw->vcache.overflow++; -// _mesa_printf("XXX %d --> %d\n", i, slot); + /*debug_printf("XXX %d --> %d\n", i, slot);*/ } /* Deal with the cache miss: */ { unsigned out; - + struct vertex_header *header; + assert(slot < Elements(draw->vcache.idx)); -// _mesa_printf("NEW %d %d\n", slot, i); + /*debug_printf("NEW %d %d\n", slot, i);*/ draw->vcache.idx[slot].in = i; draw->vcache.idx[slot].out = out = draw->vs.queue_nr++; draw->vcache.referenced |= (1 << slot); @@ -99,17 +102,21 @@ static struct vertex_header *get_vertex( struct draw_context *draw, */ assert(draw->vs.queue_nr < VS_QUEUE_LENGTH); - draw->vs.queue[out].elt = i; - draw->vs.queue[out].vertex->clipmask = 0; - draw->vs.queue[out].vertex->edgeflag = draw_get_edgeflag(draw, i); - draw->vs.queue[out].vertex->pad = 0; - draw->vs.queue[out].vertex->vertex_id = UNDEFINED_VERTEX_ID; + header = draw_header_from_block(draw->vs.vertex_cache, MAX_VERTEX_ALLOCATION, + out); + draw->vs.elts[out] = i; + header->clipmask = 0; + header->edgeflag = draw_get_edgeflag(draw, i); + header->pad = 0; + header->vertex_id = UNDEFINED_VERTEX_ID; /* Need to set the vertex's edge flag here. If we're being called * by do_ef_triangle(), that function needs edge flag info! */ - return draw->vs.queue[draw->vcache.idx[slot].out].vertex; + return draw_header_from_block(draw->vs.vertex_cache, + MAX_VERTEX_ALLOCATION, + draw->vcache.idx[slot].out); } } @@ -142,8 +149,12 @@ void draw_vertex_cache_reset_vertex_ids( struct draw_context *draw ) { unsigned i; - for (i = 0; i < draw->vs.post_nr; i++) - draw->vs.queue[i].vertex->vertex_id = UNDEFINED_VERTEX_ID; + for (i = 0; i < draw->vs.post_nr; i++) { + struct vertex_header * header = + draw_header_from_block(draw->vs.vertex_cache, + MAX_VERTEX_ALLOCATION, i); + header->vertex_id = UNDEFINED_VERTEX_ID; + } } diff --git a/src/gallium/auxiliary/draw/draw_vertex_shader.c b/src/gallium/auxiliary/draw/draw_vertex_shader.c index 133418baca..8572a6d40c 100644 --- a/src/gallium/auxiliary/draw/draw_vertex_shader.c +++ b/src/gallium/auxiliary/draw/draw_vertex_shader.c @@ -37,8 +37,6 @@ #include "draw_context.h" #include "draw_vs.h" - - /** * Run the vertex shader on all vertices in the vertex queue. * Called by the draw module when the vertx cache needs to be flushed. @@ -58,25 +56,25 @@ draw_vertex_shader_queue_flush(struct draw_context *draw) // fprintf(stderr, "%s %d\n", __FUNCTION__, draw->vs.queue_nr ); /* run vertex shader on vertex cache entries, four per invokation */ - for (i = 0; i < draw->vs.queue_nr; i += 4) { - struct vertex_header *dests[4]; - unsigned elts[4]; - int j, n = MIN2(4, draw->vs.queue_nr - i); + for (i = 0; i < draw->vs.queue_nr; i += MAX_SHADER_VERTICES) { + unsigned elts[MAX_SHADER_VERTICES]; + int j, n = MIN2(MAX_SHADER_VERTICES, draw->vs.queue_nr - i); + struct vertex_header *dests = + draw_header_from_block(draw->vs.vertex_cache, + MAX_VERTEX_ALLOCATION, i); for (j = 0; j < n; j++) { - elts[j] = draw->vs.queue[i + j].elt; - dests[j] = draw->vs.queue[i + j].vertex; + elts[j] = draw->vs.elts[i + j]; } - for ( ; j < 4; j++) { + for ( ; j < MAX_SHADER_VERTICES; j++) { elts[j] = elts[0]; - dests[j] = draw->vs.queue[i + j].vertex; } assert(n > 0); - assert(n <= 4); + assert(n <= MAX_SHADER_VERTICES); - shader->run(shader, draw, elts, n, dests); + shader->run(shader, draw, elts, n, dests, MAX_VERTEX_ALLOCATION); } draw->vs.post_nr = draw->vs.queue_nr; diff --git a/src/gallium/auxiliary/draw/draw_vf.c b/src/gallium/auxiliary/draw/draw_vf.c index 7bb34ace7a..9d0154c50d 100644 --- a/src/gallium/auxiliary/draw/draw_vf.c +++ b/src/gallium/auxiliary/draw/draw_vf.c @@ -205,7 +205,7 @@ void draw_vf_set_vertex_info( struct draw_vertex_fetch *vf, const struct vertex_info *vinfo, float point_size ) { - unsigned i, j, k; + unsigned i, j; struct draw_vf_attr *a = vf->attr; struct draw_vf_attr_map attrs[PIPE_MAX_SHADER_INPUTS]; unsigned count = 0; /* for debug/sanity */ @@ -217,60 +217,6 @@ void draw_vf_set_vertex_info( struct draw_vertex_fetch *vf, case EMIT_OMIT: /* no-op */ break; - case EMIT_ALL: { - /* just copy the whole vertex as-is to the vbuf */ - unsigned s = vinfo->size; - assert(i == 0); - assert(j == 0); - /* copy the vertex header */ - /* XXX: we actually don't copy the header, just pad it */ - attrs[nr_attrs].attrib = 0; - attrs[nr_attrs].format = DRAW_EMIT_PAD; - attrs[nr_attrs].offset = offsetof(struct vertex_header, data); - s -= offsetof(struct vertex_header, data)/4; - count += offsetof(struct vertex_header, data)/4; - nr_attrs++; - /* copy the vertex data */ - for(k = 0; k < (s & ~0x3); k += 4) { - attrs[nr_attrs].attrib = k/4; - attrs[nr_attrs].format = DRAW_EMIT_4F; - attrs[nr_attrs].offset = 0; - nr_attrs++; - count += 4; - } - /* tail */ - /* XXX: actually, this shouldn't be needed */ - attrs[nr_attrs].attrib = k/4; - attrs[nr_attrs].offset = 0; - switch(s & 0x3) { - case 0: - break; - case 1: - attrs[nr_attrs].format = DRAW_EMIT_1F; - nr_attrs++; - count += 1; - break; - case 2: - attrs[nr_attrs].format = DRAW_EMIT_2F; - nr_attrs++; - count += 2; - break; - case 3: - attrs[nr_attrs].format = DRAW_EMIT_3F; - nr_attrs++; - count += 3; - break; - } - break; - } - case EMIT_HEADER: - /* XXX emit new DRAW_EMIT_HEADER attribute??? */ - attrs[nr_attrs].attrib = 0; - attrs[nr_attrs].format = DRAW_EMIT_PAD; - attrs[nr_attrs].offset = offsetof(struct vertex_header, data); - count += offsetof(struct vertex_header, data)/4; - nr_attrs++; - break; case EMIT_1F: attrs[nr_attrs].attrib = j; attrs[nr_attrs].format = DRAW_EMIT_1F; diff --git a/src/gallium/auxiliary/draw/draw_vf.h b/src/gallium/auxiliary/draw/draw_vf.h index 7555d1bd58..0ef98d6257 100644 --- a/src/gallium/auxiliary/draw/draw_vf.h +++ b/src/gallium/auxiliary/draw/draw_vf.h @@ -128,9 +128,6 @@ draw_vf_destroy( struct draw_vertex_fetch *vf ); struct draw_vf_attr; -typedef void (*draw_vf_extract_func)( const struct draw_vf_attr *a, - float *out, - const uint8_t *v ); typedef void (*draw_vf_insert_func)( const struct draw_vf_attr *a, uint8_t *v, @@ -164,7 +161,6 @@ struct draw_vf_attr uint8_t *inputptr; const draw_vf_insert_func *insert; draw_vf_insert_func do_insert; - draw_vf_extract_func extract; }; struct draw_vertex_fetch diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h index 4ee7e705e9..33ce1e335e 100644 --- a/src/gallium/auxiliary/draw/draw_vs.h +++ b/src/gallium/auxiliary/draw/draw_vs.h @@ -31,6 +31,10 @@ #ifndef DRAW_VS_H #define DRAW_VS_H +#include "draw_context.h" +#include "draw_private.h" + + struct draw_vertex_shader; struct draw_context; struct pipe_shader_state; @@ -47,4 +51,33 @@ struct draw_vertex_shader * draw_create_vs_llvm(struct draw_context *draw, const struct pipe_shader_state *templ); + +/* Should be part of the generated shader: + */ +static INLINE unsigned +compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr) +{ + unsigned mask = 0x0; + unsigned i; + + /* Do the hardwired planes first: + */ + if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT; + if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT; + if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT; + if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT; + if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT; + if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT; + + /* Followed by any remaining ones: + */ + for (i = 6; i < nr; i++) { + if (dot4(clip, plane[i]) < 0) + mask |= (1<<i); + } + + return mask; +} + + #endif diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c index c6e503686a..5c88c2e24e 100644 --- a/src/gallium/auxiliary/draw/draw_vs_exec.c +++ b/src/gallium/auxiliary/draw/draw_vs_exec.c @@ -40,32 +40,7 @@ #include "tgsi/util/tgsi_parse.h" - -static INLINE unsigned -compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr) -{ - unsigned mask = 0; - unsigned i; - - /* Do the hardwired planes first: - */ - if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT; - if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT; - if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT; - if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT; - if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT; - if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT; - - /* Followed by any remaining ones: - */ - for (i = 6; i < nr; i++) { - if (dot4(clip, plane[i]) < 0) - mask |= (1<<i); - } - - return mask; -} - +#define MAX_TGSI_VERTICES 4 static void vs_exec_prepare( struct draw_vertex_shader *shader, @@ -89,22 +64,23 @@ vs_exec_prepare( struct draw_vertex_shader *shader, * \param count number of vertices to shade [1..4] * \param vOut array of pointers to four output vertices */ -static void +static boolean vs_exec_run( struct draw_vertex_shader *shader, struct draw_context *draw, const unsigned *elts, unsigned count, - struct vertex_header *vOut[] ) + void *vOut, + unsigned vertex_size) { struct tgsi_exec_machine *machine = &draw->machine; - unsigned int j; + unsigned int i, j; + unsigned int clipped = 0; ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_MAX_ATTRIBS); ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_MAX_ATTRIBS); const float *scale = draw->viewport.scale; const float *trans = draw->viewport.translate; - assert(count <= 4); assert(draw->vertex_shader->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION); @@ -118,80 +94,87 @@ vs_exec_run( struct draw_vertex_shader *shader, machine->Outputs = ALIGN16_ASSIGN(outputs); } - draw->vertex_fetch.fetch_func( draw, machine, elts, count ); - - if (!draw->rasterizer->bypass_vs) { - /* run interpreter */ - tgsi_exec_machine_run( machine ); - } + for (i = 0; i < count; i += MAX_TGSI_VERTICES) { + unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i); + draw->vertex_fetch.fetch_func( draw, machine, &elts[i], max_vertices ); - /* store machine results */ - for (j = 0; j < count; j++) { - unsigned slot; - float x, y, z, w; - - /* Handle attr[0] (position) specially: - * - * XXX: Computing the clipmask should be done in the vertex - * program as a set of DP4 instructions appended to the - * user-provided code. - */ - x = vOut[j]->clip[0] = machine->Outputs[0].xyzw[0].f[j]; - y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j]; - z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j]; - w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j]; - - if (!draw->rasterizer->bypass_clipping) { - vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes); - - /* divide by w */ - w = 1.0f / w; - x *= w; - y *= w; - z *= w; - } - else { - vOut[j]->clipmask = 0; - } - vOut[j]->edgeflag = 1; - - if (!draw->identity_viewport) { - /* Viewport mapping */ - vOut[j]->data[0][0] = x * scale[0] + trans[0]; - vOut[j]->data[0][1] = y * scale[1] + trans[1]; - vOut[j]->data[0][2] = z * scale[2] + trans[2]; - vOut[j]->data[0][3] = w; - } - else { - vOut[j]->data[0][0] = x; - vOut[j]->data[0][1] = y; - vOut[j]->data[0][2] = z; - vOut[j]->data[0][3] = w; + if (!draw->rasterizer->bypass_vs) { + /* run interpreter */ + tgsi_exec_machine_run( machine ); } - /* Remaining attributes are packed into sequential post-transform - * vertex attrib slots. - */ - for (slot = 1; slot < draw->num_vs_outputs; slot++) { - vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j]; - vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j]; - vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j]; - vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j]; - } + /* store machine results */ + for (j = 0; j < max_vertices; j++) { + unsigned slot; + float x, y, z, w; + struct vertex_header *out = + draw_header_from_block(vOut, vertex_size, i + j); + + /* Handle attr[0] (position) specially: + * + * XXX: Computing the clipmask should be done in the vertex + * program as a set of DP4 instructions appended to the + * user-provided code. + */ + x = out->clip[0] = machine->Outputs[0].xyzw[0].f[j]; + y = out->clip[1] = machine->Outputs[0].xyzw[1].f[j]; + z = out->clip[2] = machine->Outputs[0].xyzw[2].f[j]; + w = out->clip[3] = machine->Outputs[0].xyzw[3].f[j]; + + if (!draw->rasterizer->bypass_clipping) { + out->clipmask = compute_clipmask(out->clip, draw->plane, + draw->nr_planes); + clipped += out->clipmask; + + /* divide by w */ + w = 1.0f / w; + x *= w; + y *= w; + z *= w; + } + else { + out->clipmask = 0; + } + out->edgeflag = 1; + out->vertex_id = UNDEFINED_VERTEX_ID; + + if (!draw->identity_viewport) { + /* Viewport mapping */ + out->data[0][0] = x * scale[0] + trans[0]; + out->data[0][1] = y * scale[1] + trans[1]; + out->data[0][2] = z * scale[2] + trans[2]; + out->data[0][3] = w; + } + else { + out->data[0][0] = x; + out->data[0][1] = y; + out->data[0][2] = z; + out->data[0][3] = w; + } + + /* Remaining attributes are packed into sequential post-transform + * vertex attrib slots. + */ + for (slot = 1; slot < draw->num_vs_outputs; slot++) { + out->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j]; + out->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j]; + out->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j]; + out->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j]; + } #if 0 /*DEBUG*/ - printf("Post xform vert:\n"); - for (slot = 0; slot < draw->num_vs_outputs; slot++) { - printf("%d: %f %f %f %f\n", slot, - vOut[j]->data[slot][0], - vOut[j]->data[slot][1], - vOut[j]->data[slot][2], - vOut[j]->data[slot][3]); - } -#endif - - - } /* loop over vertices */ + printf("%d) Post xform vert:\n", i + j); + for (slot = 0; slot < draw->num_vs_outputs; slot++) { + printf("\t%d: %f %f %f %f\n", slot, + out->data[slot][0], + out->data[slot][1], + out->data[slot][2], + out->data[slot][3]); + } +#endif + } /* loop over vertices */ + } + return clipped != 0; } diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c index c8268317ef..73076d2467 100644 --- a/src/gallium/auxiliary/draw/draw_vs_llvm.c +++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c @@ -50,33 +50,6 @@ struct draw_llvm_vertex_shader { }; -static INLINE unsigned -compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr) -{ - unsigned mask = 0; - unsigned i; - - /* Do the hardwired planes first: - */ - if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT; - if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT; - if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT; - if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT; - if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT; - if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT; - - /* Followed by any remaining ones: - */ - for (i = 6; i < nr; i++) { - if (dot4(clip, plane[i]) < 0) - mask |= (1<<i); - } - - return mask; -} - - - static void vs_llvm_prepare( struct draw_vertex_shader *base, struct draw_context *draw ) @@ -94,18 +67,19 @@ vs_llvm_prepare( struct draw_vertex_shader *base, * \param count number of vertices to shade [1..4] * \param vOut array of pointers to four output vertices */ -static void +static boolean vs_llvm_run( struct draw_vertex_shader *base, - struct draw_context *draw, - const unsigned *elts, + struct draw_context *draw, + const unsigned *elts, unsigned count, - struct vertex_header *vOut[] ) + void *vOut ) { - struct draw_llvm_vertex_shader *shader = + struct draw_llvm_vertex_shader *shader = (struct draw_llvm_vertex_shader *)base; struct tgsi_exec_machine *machine = &draw->machine; unsigned int j; + unsigned int clipped = 0; ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_MAX_ATTRIBS); ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_MAX_ATTRIBS); @@ -152,19 +126,22 @@ vs_llvm_run( struct draw_vertex_shader *base, w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j]; if (!draw->rasterizer->bypass_clipping) { - vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes); + vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, + draw->nr_planes); + clipped += vOut[j]->clipmask; /* divide by w */ w = 1.0f / w; x *= w; y *= w; - z *= w; + z *= w; } else { vOut[j]->clipmask = 0; } vOut[j]->edgeflag = 1; - + vOut[j]->vertex_id = UNDEFINED_VERTEX_ID; + if (!draw->identity_viewport) { /* Viewport mapping */ vOut[j]->data[0][0] = x * scale[0] + trans[0]; @@ -189,6 +166,7 @@ vs_llvm_run( struct draw_vertex_shader *base, vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j]; } } /* loop over vertices */ + return clipped != 0; } static void diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c index f40d65df08..ee0a3105b9 100644 --- a/src/gallium/auxiliary/draw/draw_vs_sse.c +++ b/src/gallium/auxiliary/draw/draw_vs_sse.c @@ -45,49 +45,24 @@ #include "tgsi/exec/tgsi_sse2.h" #include "tgsi/util/tgsi_parse.h" +#define SSE_MAX_VERTICES 4 typedef void (XSTDCALL *codegen_function) ( const struct tgsi_exec_vector *input, struct tgsi_exec_vector *output, float (*constant)[4], - struct tgsi_exec_vector *temporary ); + struct tgsi_exec_vector *temporary, + float (*immediates)[4] ); struct draw_sse_vertex_shader { struct draw_vertex_shader base; struct x86_function sse2_program; codegen_function func; + float immediates[TGSI_EXEC_NUM_IMMEDIATES][4]; }; -/* Should be part of the generated shader: - */ -static INLINE unsigned -compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr) -{ - unsigned mask = 0; - unsigned i; - - /* Do the hardwired planes first: - */ - if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT; - if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT; - if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT; - if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT; - if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT; - if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT; - - /* Followed by any remaining ones: - */ - for (i = 6; i < nr; i++) { - if (dot4(clip, plane[i]) < 0) - mask |= (1<<i); - } - - return mask; -} - - static void vs_sse_prepare( struct draw_vertex_shader *base, struct draw_context *draw ) @@ -103,23 +78,24 @@ vs_sse_prepare( struct draw_vertex_shader *base, * \param count number of vertices to shade [1..4] * \param vOut array of pointers to four output vertices */ -static void +static boolean vs_sse_run( struct draw_vertex_shader *base, struct draw_context *draw, const unsigned *elts, unsigned count, - struct vertex_header *vOut[] ) + void *vOut, + unsigned vertex_size ) { struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base; struct tgsi_exec_machine *machine = &draw->machine; - unsigned int j; + unsigned int i, j; + unsigned int clipped = 0; ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_MAX_ATTRIBS); ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_MAX_ATTRIBS); const float *scale = draw->viewport.scale; const float *trans = draw->viewport.translate; - assert(count <= 4); assert(draw->vertex_shader->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION); @@ -134,76 +110,93 @@ vs_sse_run( struct draw_vertex_shader *base, machine->Outputs = ALIGN16_ASSIGN(outputs); } - - /* Fetch vertices. This may at some point be integrated into the - * compiled shader -- that would require a reorganization where - * multiple versions of the compiled shader might exist, - * specialized for each fetch state. - */ - draw->vertex_fetch.fetch_func( draw, machine, elts, count ); - - - if (!draw->rasterizer->bypass_vs) { - /* run compiled shader - */ - shader->func(machine->Inputs, - machine->Outputs, - machine->Consts, - machine->Temps ); - } - - - /* XXX: Computing the clipmask and emitting results should be done - * in the vertex program as a set of instructions appended to - * the user-provided code. - */ - for (j = 0; j < count; j++) { - unsigned slot; - float x, y, z, w; - - x = vOut[j]->clip[0] = machine->Outputs[0].xyzw[0].f[j]; - y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j]; - z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j]; - w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j]; - - if (!draw->rasterizer->bypass_clipping) { - vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes); - - /* divide by w */ - w = 1.0f / w; - x *= w; - y *= w; - z *= w; - } - else { - vOut[j]->clipmask = 0; - } - vOut[j]->edgeflag = 1; - - if (!draw->identity_viewport) { - /* Viewport mapping */ - vOut[j]->data[0][0] = x * scale[0] + trans[0]; - vOut[j]->data[0][1] = y * scale[1] + trans[1]; - vOut[j]->data[0][2] = z * scale[2] + trans[2]; - vOut[j]->data[0][3] = w; - } - else { - vOut[j]->data[0][0] = x; - vOut[j]->data[0][1] = y; - vOut[j]->data[0][2] = z; - vOut[j]->data[0][3] = w; + for (i = 0; i < count; i += SSE_MAX_VERTICES) { + unsigned int max_vertices = MIN2(SSE_MAX_VERTICES, count - i); + /* Fetch vertices. This may at some point be integrated into the + * compiled shader -- that would require a reorganization where + * multiple versions of the compiled shader might exist, + * specialized for each fetch state. + */ + draw->vertex_fetch.fetch_func(draw, machine, &elts[i], max_vertices); + + if (!draw->rasterizer->bypass_vs) { + /* run compiled shader + */ + shader->func(machine->Inputs, + machine->Outputs, + machine->Consts, + machine->Temps, + shader->immediates); } - /* Remaining attributes are packed into sequential post-transform - * vertex attrib slots. + /* XXX: Computing the clipmask and emitting results should be done + * in the vertex program as a set of instructions appended to + * the user-provided code. */ - for (slot = 1; slot < draw->num_vs_outputs; slot++) { - vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j]; - vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j]; - vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j]; - vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j]; + for (j = 0; j < max_vertices; j++) { + unsigned slot; + float x, y, z, w; + struct vertex_header *out = + draw_header_from_block(vOut, vertex_size, i + j); + + x = out->clip[0] = machine->Outputs[0].xyzw[0].f[j]; + y = out->clip[1] = machine->Outputs[0].xyzw[1].f[j]; + z = out->clip[2] = machine->Outputs[0].xyzw[2].f[j]; + w = out->clip[3] = machine->Outputs[0].xyzw[3].f[j]; + + if (!draw->rasterizer->bypass_clipping) { + out->clipmask = compute_clipmask(out->clip, draw->plane, + draw->nr_planes); + clipped += out->clipmask; + + /* divide by w */ + w = 1.0f / w; + x *= w; + y *= w; + z *= w; + } + else { + out->clipmask = 0; + } + out->edgeflag = 1; + out->vertex_id = UNDEFINED_VERTEX_ID; + + if (!draw->identity_viewport) { + /* Viewport mapping */ + out->data[0][0] = x * scale[0] + trans[0]; + out->data[0][1] = y * scale[1] + trans[1]; + out->data[0][2] = z * scale[2] + trans[2]; + out->data[0][3] = w; + } + else { + out->data[0][0] = x; + out->data[0][1] = y; + out->data[0][2] = z; + out->data[0][3] = w; + } + + /* Remaining attributes are packed into sequential post-transform + * vertex attrib slots. + */ + for (slot = 1; slot < draw->num_vs_outputs; slot++) { + out->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j]; + out->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j]; + out->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j]; + out->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j]; + } +#if 0 /*DEBUG*/ + printf("%d) Post xform vert:\n", i + j); + for (slot = 0; slot < draw->num_vs_outputs; slot++) { + printf("\t%d: %f %f %f %f\n", slot, + out->data[slot][0], + out->data[slot][1], + out->data[slot][2], + out->data[slot][3]); + } +#endif } - } + } + return clipped != 0; } @@ -243,7 +236,7 @@ draw_create_vs_sse(struct draw_context *draw, x86_init_func( &vs->sse2_program ); if (!tgsi_emit_sse2( (struct tgsi_token *) vs->base.state.tokens, - &vs->sse2_program )) + &vs->sse2_program, vs->immediates )) goto fail; vs->func = (codegen_function) x86_get_func( &vs->sse2_program ); |