From 4486012245c5f526059d3872ac3561f53705d1cf Mon Sep 17 00:00:00 2001 From: Keith Whitwell Date: Fri, 24 Jul 2009 16:12:48 +0100 Subject: [PATCH] llvmpipe: rip out old mulithread support --- src/gallium/drivers/llvmpipe/lp_context.c | 48 ++-- src/gallium/drivers/llvmpipe/lp_context.h | 13 +- src/gallium/drivers/llvmpipe/lp_quad_pipe.c | 48 ++-- src/gallium/drivers/llvmpipe/lp_setup.c | 253 ++------------------ 4 files changed, 57 insertions(+), 305 deletions(-) diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c index aa2159b8d5e..edd409dc949 100644 --- a/src/gallium/drivers/llvmpipe/lp_context.c +++ b/src/gallium/drivers/llvmpipe/lp_context.c @@ -88,19 +88,17 @@ static void llvmpipe_destroy( struct pipe_context *pipe ) if (llvmpipe->draw) draw_destroy( llvmpipe->draw ); - for (i = 0; i < LP_NUM_QUAD_THREADS; i++) { - llvmpipe->quad[i].polygon_stipple->destroy( llvmpipe->quad[i].polygon_stipple ); - llvmpipe->quad[i].earlyz->destroy( llvmpipe->quad[i].earlyz ); - llvmpipe->quad[i].shade->destroy( llvmpipe->quad[i].shade ); - llvmpipe->quad[i].alpha_test->destroy( llvmpipe->quad[i].alpha_test ); - llvmpipe->quad[i].depth_test->destroy( llvmpipe->quad[i].depth_test ); - llvmpipe->quad[i].stencil_test->destroy( llvmpipe->quad[i].stencil_test ); - llvmpipe->quad[i].occlusion->destroy( llvmpipe->quad[i].occlusion ); - llvmpipe->quad[i].coverage->destroy( llvmpipe->quad[i].coverage ); - llvmpipe->quad[i].blend->destroy( llvmpipe->quad[i].blend ); - llvmpipe->quad[i].colormask->destroy( llvmpipe->quad[i].colormask ); - llvmpipe->quad[i].output->destroy( llvmpipe->quad[i].output ); - } + llvmpipe->quad.polygon_stipple->destroy( llvmpipe->quad.polygon_stipple ); + llvmpipe->quad.earlyz->destroy( llvmpipe->quad.earlyz ); + llvmpipe->quad.shade->destroy( llvmpipe->quad.shade ); + llvmpipe->quad.alpha_test->destroy( llvmpipe->quad.alpha_test ); + llvmpipe->quad.depth_test->destroy( llvmpipe->quad.depth_test ); + llvmpipe->quad.stencil_test->destroy( llvmpipe->quad.stencil_test ); + llvmpipe->quad.occlusion->destroy( llvmpipe->quad.occlusion ); + llvmpipe->quad.coverage->destroy( llvmpipe->quad.coverage ); + llvmpipe->quad.blend->destroy( llvmpipe->quad.blend ); + llvmpipe->quad.colormask->destroy( llvmpipe->quad.colormask ); + llvmpipe->quad.output->destroy( llvmpipe->quad.output ); for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) lp_destroy_tile_cache(llvmpipe->cbuf_cache[i]); @@ -234,19 +232,17 @@ llvmpipe_create( struct pipe_screen *screen ) /* setup quad rendering stages */ - for (i = 0; i < LP_NUM_QUAD_THREADS; i++) { - llvmpipe->quad[i].polygon_stipple = lp_quad_polygon_stipple_stage(llvmpipe); - llvmpipe->quad[i].earlyz = lp_quad_earlyz_stage(llvmpipe); - llvmpipe->quad[i].shade = lp_quad_shade_stage(llvmpipe); - llvmpipe->quad[i].alpha_test = lp_quad_alpha_test_stage(llvmpipe); - llvmpipe->quad[i].depth_test = lp_quad_depth_test_stage(llvmpipe); - llvmpipe->quad[i].stencil_test = lp_quad_stencil_test_stage(llvmpipe); - llvmpipe->quad[i].occlusion = lp_quad_occlusion_stage(llvmpipe); - llvmpipe->quad[i].coverage = lp_quad_coverage_stage(llvmpipe); - llvmpipe->quad[i].blend = lp_quad_blend_stage(llvmpipe); - llvmpipe->quad[i].colormask = lp_quad_colormask_stage(llvmpipe); - llvmpipe->quad[i].output = lp_quad_output_stage(llvmpipe); - } + llvmpipe->quad.polygon_stipple = lp_quad_polygon_stipple_stage(llvmpipe); + llvmpipe->quad.earlyz = lp_quad_earlyz_stage(llvmpipe); + llvmpipe->quad.shade = lp_quad_shade_stage(llvmpipe); + llvmpipe->quad.alpha_test = lp_quad_alpha_test_stage(llvmpipe); + llvmpipe->quad.depth_test = lp_quad_depth_test_stage(llvmpipe); + llvmpipe->quad.stencil_test = lp_quad_stencil_test_stage(llvmpipe); + llvmpipe->quad.occlusion = lp_quad_occlusion_stage(llvmpipe); + llvmpipe->quad.coverage = lp_quad_coverage_stage(llvmpipe); + llvmpipe->quad.blend = lp_quad_blend_stage(llvmpipe); + llvmpipe->quad.colormask = lp_quad_colormask_stage(llvmpipe); + llvmpipe->quad.output = lp_quad_output_stage(llvmpipe); /* vertex shader samplers */ for (i = 0; i < PIPE_MAX_SAMPLERS; i++) { diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h index ab88401c605..c43ea8bab62 100644 --- a/src/gallium/drivers/llvmpipe/lp_context.h +++ b/src/gallium/drivers/llvmpipe/lp_context.h @@ -39,17 +39,6 @@ #include "lp_tex_sample.h" -/** - * This is a temporary variable for testing draw-stage polygon stipple. - * If zero, do stipple in lp_quad_stipple.c - */ -#define USE_DRAW_STAGE_PSTIPPLE 1 - -/* Number of threads working on individual quads. - * Setting to 1 disables this feature. - */ -#define LP_NUM_QUAD_THREADS 1 - struct llvmpipe_vbuf_render; struct draw_context; struct draw_stage; @@ -129,7 +118,7 @@ struct llvmpipe_context { struct quad_stage *output; struct quad_stage *first; /**< points to one of the above stages */ - } quad[LP_NUM_QUAD_THREADS]; + } quad; /** TGSI exec things */ struct { diff --git a/src/gallium/drivers/llvmpipe/lp_quad_pipe.c b/src/gallium/drivers/llvmpipe/lp_quad_pipe.c index ab202635238..d738d08d9e2 100644 --- a/src/gallium/drivers/llvmpipe/lp_quad_pipe.c +++ b/src/gallium/drivers/llvmpipe/lp_quad_pipe.c @@ -31,35 +31,29 @@ #include "pipe/p_shader_tokens.h" static void -lp_push_quad_first( - struct llvmpipe_context *lp, - struct quad_stage *quad, - uint i ) +lp_push_quad_first( struct llvmpipe_context *lp, + struct quad_stage *quad ) { - quad->next = lp->quad[i].first; - lp->quad[i].first = quad; + quad->next = lp->quad.first; + lp->quad.first = quad; } static void -lp_build_depth_stencil( - struct llvmpipe_context *lp, - uint i ) +lp_build_depth_stencil( struct llvmpipe_context *lp ) { if (lp->depth_stencil->stencil[0].enabled || lp->depth_stencil->stencil[1].enabled) { - lp_push_quad_first( lp, lp->quad[i].stencil_test, i ); + lp_push_quad_first( lp, lp->quad.stencil_test ); } else if (lp->depth_stencil->depth.enabled && lp->framebuffer.zsbuf) { - lp_push_quad_first( lp, lp->quad[i].depth_test, i ); + lp_push_quad_first( lp, lp->quad.depth_test ); } } void lp_build_quad_pipeline(struct llvmpipe_context *lp) { - uint i; - boolean early_depth_test = lp->depth_stencil->depth.enabled && lp->framebuffer.zsbuf && @@ -68,51 +62,43 @@ lp_build_quad_pipeline(struct llvmpipe_context *lp) !lp->fs->info.writes_z; /* build up the pipeline in reverse order... */ - for (i = 0; i < LP_NUM_QUAD_THREADS; i++) { - lp->quad[i].first = lp->quad[i].output; + lp->quad.first = lp->quad.output; if (lp->blend->colormask != 0xf) { - lp_push_quad_first( lp, lp->quad[i].colormask, i ); + lp_push_quad_first( lp, lp->quad.colormask ); } if (lp->blend->blend_enable || lp->blend->logicop_enable) { - lp_push_quad_first( lp, lp->quad[i].blend, i ); + lp_push_quad_first( lp, lp->quad.blend ); } if (lp->active_query_count) { - lp_push_quad_first( lp, lp->quad[i].occlusion, i ); + lp_push_quad_first( lp, lp->quad.occlusion ); } if (lp->rasterizer->poly_smooth || lp->rasterizer->line_smooth || lp->rasterizer->point_smooth) { - lp_push_quad_first( lp, lp->quad[i].coverage, i ); + lp_push_quad_first( lp, lp->quad.coverage ); } if (!early_depth_test) { - lp_build_depth_stencil( lp, i ); + lp_build_depth_stencil( lp ); } if (lp->depth_stencil->alpha.enabled) { - lp_push_quad_first( lp, lp->quad[i].alpha_test, i ); + lp_push_quad_first( lp, lp->quad.alpha_test ); } /* XXX always enable shader? */ if (1) { - lp_push_quad_first( lp, lp->quad[i].shade, i ); + lp_push_quad_first( lp, lp->quad.shade ); } if (early_depth_test) { - lp_build_depth_stencil( lp, i ); - lp_push_quad_first( lp, lp->quad[i].earlyz, i ); - } - -#if !USE_DRAW_STAGE_PSTIPPLE - if (lp->rasterizer->poly_stipple_enable) { - lp_push_quad_first( lp, lp->quad[i].polygon_stipple, i ); + lp_build_depth_stencil( lp ); + lp_push_quad_first( lp, lp->quad.earlyz ); } -#endif - } } diff --git a/src/gallium/drivers/llvmpipe/lp_setup.c b/src/gallium/drivers/llvmpipe/lp_setup.c index 21e769931a9..783f36bc7f0 100644 --- a/src/gallium/drivers/llvmpipe/lp_setup.c +++ b/src/gallium/drivers/llvmpipe/lp_setup.c @@ -61,87 +61,7 @@ struct edge { int lines; /**< number of lines on this edge */ }; -#if LP_NUM_QUAD_THREADS > 1 -/* Set to 1 if you want other threads to be instantly - * notified of pending jobs. - */ -#define INSTANT_NOTEMPTY_NOTIFY 0 - -struct thread_info -{ - struct setup_context *setup; - uint id; - pipe_thread handle; -}; - -struct quad_job; - -typedef void (* quad_job_routine)( struct setup_context *setup, uint thread, struct quad_job *job ); - -struct quad_job -{ - struct quad_header_input input; - struct quad_header_inout inout; - quad_job_routine routine; -}; - -#define NUM_QUAD_JOBS 64 - -struct quad_job_que -{ - struct quad_job jobs[NUM_QUAD_JOBS]; - uint first; - uint last; - pipe_mutex que_mutex; - pipe_condvar que_notfull_condvar; - pipe_condvar que_notempty_condvar; - uint jobs_added; - uint jobs_done; - pipe_condvar que_done_condvar; -}; - -static void -add_quad_job( struct quad_job_que *que, struct quad_header *quad, quad_job_routine routine ) -{ -#if INSTANT_NOTEMPTY_NOTIFY - boolean empty; -#endif - - /* Wait for empty slot, see if the que is empty. - */ - pipe_mutex_lock( que->que_mutex ); - while ((que->last + 1) % NUM_QUAD_JOBS == que->first) { -#if !INSTANT_NOTEMPTY_NOTIFY - pipe_condvar_broadcast( que->que_notempty_condvar ); -#endif - pipe_condvar_wait( que->que_notfull_condvar, que->que_mutex ); - } -#if INSTANT_NOTEMPTY_NOTIFY - empty = que->last == que->first; -#endif - que->jobs_added++; - pipe_mutex_unlock( que->que_mutex ); - - /* Submit new job. - */ - que->jobs[que->last].input = quad->input; - que->jobs[que->last].inout = quad->inout; - que->jobs[que->last].routine = routine; - que->last = (que->last + 1) % NUM_QUAD_JOBS; - -#if INSTANT_NOTEMPTY_NOTIFY - /* If the que was empty, notify consumers there's a job to be done. - */ - if (empty) { - pipe_mutex_lock( que->que_mutex ); - pipe_condvar_broadcast( que->que_notempty_condvar ); - pipe_mutex_unlock( que->que_mutex ); - } -#endif -} - -#endif /** * Triangle setup info (derived from draw_stage). @@ -169,11 +89,6 @@ struct setup_context { struct tgsi_interp_coef posCoef; /* For Z, W */ struct quad_header quad; -#if LP_NUM_QUAD_THREADS > 1 - struct quad_job_que que; - struct thread_info threads[LP_NUM_QUAD_THREADS]; -#endif - struct { int left[2]; /**< [0] = row0, [1] = row1 */ int right[2]; @@ -188,67 +103,6 @@ struct setup_context { unsigned winding; /* which winding to cull */ }; -#if LP_NUM_QUAD_THREADS > 1 - -static PIPE_THREAD_ROUTINE( quad_thread, param ) -{ - struct thread_info *info = (struct thread_info *) param; - struct quad_job_que *que = &info->setup->que; - - for (;;) { - struct quad_job job; - boolean full; - - /* Wait for an available job. - */ - pipe_mutex_lock( que->que_mutex ); - while (que->last == que->first) - pipe_condvar_wait( que->que_notempty_condvar, que->que_mutex ); - - /* See if the que is full. - */ - full = (que->last + 1) % NUM_QUAD_JOBS == que->first; - - /* Take a job and remove it from que. - */ - job = que->jobs[que->first]; - que->first = (que->first + 1) % NUM_QUAD_JOBS; - - /* Notify the producer if the que is not full. - */ - if (full) - pipe_condvar_signal( que->que_notfull_condvar ); - pipe_mutex_unlock( que->que_mutex ); - - job.routine( info->setup, info->id, &job ); - - /* Notify the producer if that's the last finished job. - */ - pipe_mutex_lock( que->que_mutex ); - que->jobs_done++; - if (que->jobs_added == que->jobs_done) - pipe_condvar_signal( que->que_done_condvar ); - pipe_mutex_unlock( que->que_mutex ); - } - - return NULL; -} - -#define WAIT_FOR_COMPLETION(setup) \ - do {\ - pipe_mutex_lock( setup->que.que_mutex );\ - if (!INSTANT_NOTEMPTY_NOTIFY)\ - pipe_condvar_broadcast( setup->que.que_notempty_condvar );\ - while (setup->que.jobs_added != setup->que.jobs_done)\ - pipe_condvar_wait( setup->que.que_done_condvar, setup->que.que_mutex );\ - pipe_mutex_unlock( setup->que.que_mutex );\ - } while (0) - -#else - -#define WAIT_FOR_COMPLETION(setup) ((void) 0) - -#endif @@ -311,39 +165,17 @@ quad_clip( struct setup_context *setup, struct quad_header *quad ) * Emit a quad (pass to next stage) with clipping. */ static INLINE void -clip_emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread ) +clip_emit_quad( struct setup_context *setup, struct quad_header *quad ) { quad_clip( setup, quad ); + if (quad->inout.mask) { struct llvmpipe_context *lp = setup->llvmpipe; - lp->quad[thread].first->run( lp->quad[thread].first, quad ); + lp->quad.first->run( lp->quad.first, quad ); } } -#if LP_NUM_QUAD_THREADS > 1 - -static void -clip_emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job ) -{ - struct quad_header quad; - - quad.input = job->input; - quad.inout = job->inout; - quad.coef = setup->quad.coef; - quad.posCoef = setup->quad.posCoef; - quad.nr_attrs = setup->quad.nr_attrs; - clip_emit_quad( setup, &quad, thread ); -} - -#define CLIP_EMIT_QUAD(setup) add_quad_job( &setup->que, &setup->quad, clip_emit_quad_job ) - -#else - -#define CLIP_EMIT_QUAD(setup) clip_emit_quad( setup, &setup->quad, 0 ) - -#endif - /** * Emit a quad (pass to next stage). No clipping is done. */ @@ -361,7 +193,7 @@ emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread ) if (mask & 4) setup->numFragsEmitted++; if (mask & 8) setup->numFragsEmitted++; #endif - lp->quad[thread].first->run( lp->quad[thread].first, quad ); + lp->quad.first->run( lp->quad.first, quad ); #if DEBUG_FRAGS mask = quad->inout.mask; if (mask & 1) setup->numFragsWritten++; @@ -371,38 +203,15 @@ emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread ) #endif } -#if LP_NUM_QUAD_THREADS > 1 - -static void -emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job ) -{ - struct quad_header quad; - - quad.input = job->input; - quad.inout = job->inout; - quad.coef = setup->quad.coef; - quad.posCoef = setup->quad.posCoef; - quad.nr_attrs = setup->quad.nr_attrs; - emit_quad( setup, &quad, thread ); -} - -#define EMIT_QUAD(setup,x,y,qmask) do {\ - setup->quad.input.x0 = x;\ - setup->quad.input.y0 = y;\ - setup->quad.inout.mask = qmask;\ - add_quad_job( &setup->que, &setup->quad, emit_quad_job );\ - } while (0) -#else +#define EMIT_QUAD(setup,x,y,qmask) \ +do { \ + setup->quad.input.x0 = x; \ + setup->quad.input.y0 = y; \ + setup->quad.inout.mask = qmask; \ + emit_quad( setup, &setup->quad, 0 ); \ +} while (0) -#define EMIT_QUAD(setup,x,y,qmask) do {\ - setup->quad.input.x0 = x;\ - setup->quad.input.y0 = y;\ - setup->quad.inout.mask = qmask;\ - emit_quad( setup, &setup->quad, 0 );\ - } while (0) - -#endif /** * Given an X or Y coordinate, return the block/quad coordinate that it @@ -956,8 +765,6 @@ void setup_tri( struct setup_context *setup, flush_spans( setup ); - WAIT_FOR_COMPLETION(setup); - #if DEBUG_FRAGS printf("Tri: %u frags emitted, %u written\n", setup->numFragsEmitted, @@ -1101,7 +908,7 @@ plot(struct setup_context *setup, int x, int y) /* flush prev quad, start new quad */ if (setup->quad.input.x0 != -1) - CLIP_EMIT_QUAD(setup); + clip_emit_quad( setup, &setup->quad ); setup->quad.input.x0 = quadX; setup->quad.input.y0 = quadY; @@ -1223,10 +1030,8 @@ setup_line(struct setup_context *setup, /* draw final quad */ if (setup->quad.inout.mask) { - CLIP_EMIT_QUAD(setup); + clip_emit_quad( setup, &setup->quad ); } - - WAIT_FOR_COMPLETION(setup); } @@ -1334,7 +1139,7 @@ setup_point( struct setup_context *setup, setup->quad.input.x0 = (int) x - ix; setup->quad.input.y0 = (int) y - iy; setup->quad.inout.mask = (1 << ix) << (2 * iy); - CLIP_EMIT_QUAD(setup); + clip_emit_quad( setup, &setup->quad ); } else { if (round) { @@ -1395,7 +1200,7 @@ setup_point( struct setup_context *setup, if (setup->quad.inout.mask) { setup->quad.input.x0 = ix; setup->quad.input.y0 = iy; - CLIP_EMIT_QUAD(setup); + clip_emit_quad( setup, &setup->quad ); } } } @@ -1442,19 +1247,16 @@ setup_point( struct setup_context *setup, setup->quad.inout.mask = mask; setup->quad.input.x0 = ix; setup->quad.input.y0 = iy; - CLIP_EMIT_QUAD(setup); + clip_emit_quad( setup, &setup->quad ); } } } } - - WAIT_FOR_COMPLETION(setup); } void setup_prepare( struct setup_context *setup ) { struct llvmpipe_context *lp = setup->llvmpipe; - unsigned i; if (lp->dirty) { llvmpipe_update_derived(lp); @@ -1463,9 +1265,7 @@ void setup_prepare( struct setup_context *setup ) /* Note: nr_attrs is only used for debugging (vertex printing) */ setup->quad.nr_attrs = draw_num_vs_outputs(lp->draw); - for (i = 0; i < LP_NUM_QUAD_THREADS; i++) { - lp->quad[i].first->begin( lp->quad[i].first ); - } + lp->quad.first->begin( lp->quad.first ); if (lp->reduced_api_prim == PIPE_PRIM_TRIANGLES && lp->rasterizer->fill_cw == PIPE_POLYGON_MODE_FILL && @@ -1493,9 +1293,6 @@ void setup_destroy_context( struct setup_context *setup ) struct setup_context *setup_create_context( struct llvmpipe_context *llvmpipe ) { struct setup_context *setup = CALLOC_STRUCT(setup_context); -#if LP_NUM_QUAD_THREADS > 1 - uint i; -#endif setup->llvmpipe = llvmpipe; @@ -1505,22 +1302,6 @@ struct setup_context *setup_create_context( struct llvmpipe_context *llvmpipe ) setup->span.left[0] = 1000000; /* greater than right[0] */ setup->span.left[1] = 1000000; /* greater than right[1] */ -#if LP_NUM_QUAD_THREADS > 1 - setup->que.first = 0; - setup->que.last = 0; - pipe_mutex_init( setup->que.que_mutex ); - pipe_condvar_init( setup->que.que_notfull_condvar ); - pipe_condvar_init( setup->que.que_notempty_condvar ); - setup->que.jobs_added = 0; - setup->que.jobs_done = 0; - pipe_condvar_init( setup->que.que_done_condvar ); - for (i = 0; i < LP_NUM_QUAD_THREADS; i++) { - setup->threads[i].setup = setup; - setup->threads[i].id = i; - setup->threads[i].handle = pipe_thread_create( quad_thread, &setup->threads[i] ); - } -#endif - return setup; } -- 2.30.2