From cf3778bae03460325960d2aec0cd4cdf325299eb Mon Sep 17 00:00:00 2001 From: Marek Olšák Date: Mon, 14 Jun 2010 05:18:47 +0200 Subject: r300g: rewrite occlusion queries The previous implementation had issues with queries spanning over several command streams as well as using a very large number of queries. This fixes flickering in Enemy Territory: Quake Wars. The driver now renders everything correctly in this game and the graphics is awesome. --- src/gallium/drivers/r300/r300_blit.c | 130 +++++++++++++++----------- src/gallium/drivers/r300/r300_context.c | 8 +- src/gallium/drivers/r300/r300_context.h | 29 ++++-- src/gallium/drivers/r300/r300_defines.h | 6 -- src/gallium/drivers/r300/r300_emit.c | 41 ++++---- src/gallium/drivers/r300/r300_query.c | 109 +++++++++------------ src/gallium/drivers/r300/r300_screen_buffer.c | 4 - 7 files changed, 168 insertions(+), 159 deletions(-) (limited to 'src/gallium/drivers') diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c index 3bb1caef40..2a47701291 100644 --- a/src/gallium/drivers/r300/r300_blit.c +++ b/src/gallium/drivers/r300/r300_blit.c @@ -25,8 +25,23 @@ #include "util/u_format.h" -static void r300_blitter_save_states(struct r300_context* r300) +enum r300_blitter_op { + R300_CLEAR, + R300_CLEAR_SURFACE, + R300_COPY +}; + +static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op op) +{ + if (r300->query_current) { + r300->blitter_saved_query = r300->query_current; + r300_stop_query(r300); + } + + /* Yeah we have to save all those states to ensure the blitter operation + * is really transparent. The states will be restored by the blitter once + * copying is done. */ util_blitter_save_blend(r300->blitter, r300->blend_state.state); util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state); util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref)); @@ -38,6 +53,30 @@ static void r300_blitter_save_states(struct r300_context* r300) util_blitter_save_vertex_elements(r300->blitter, r300->velems); util_blitter_save_vertex_buffers(r300->blitter, r300->vertex_buffer_count, r300->vertex_buffer); + + if (op & (R300_CLEAR_SURFACE | R300_COPY)) + util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state); + + if (op & R300_COPY) { + struct r300_textures_state* state = + (struct r300_textures_state*)r300->textures_state.state; + + util_blitter_save_fragment_sampler_states( + r300->blitter, state->sampler_state_count, + (void**)state->sampler_states); + + util_blitter_save_fragment_sampler_views( + r300->blitter, state->sampler_view_count, + (struct pipe_sampler_view**)state->sampler_views); + } +} + +static void r300_blitter_end(struct r300_context *r300) +{ + if (r300->blitter_saved_query) { + r300_resume_query(r300, r300->blitter_saved_query); + r300->blitter_saved_query = NULL; + } } /* Clear currently bound buffers. */ @@ -73,13 +112,45 @@ static void r300_clear(struct pipe_context* pipe, struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)r300->fb_state.state; - r300_blitter_save_states(r300); - + r300_blitter_begin(r300, R300_CLEAR); util_blitter_clear(r300->blitter, fb->width, fb->height, fb->nr_cbufs, buffers, rgba, depth, stencil); + r300_blitter_end(r300); +} + +/* Clear a region of a color surface to a constant value. */ +static void r300_clear_render_target(struct pipe_context *pipe, + struct pipe_surface *dst, + const float *rgba, + unsigned dstx, unsigned dsty, + unsigned width, unsigned height) +{ + struct r300_context *r300 = r300_context(pipe); + + r300_blitter_begin(r300, R300_CLEAR_SURFACE); + util_blitter_clear_render_target(r300->blitter, dst, rgba, + dstx, dsty, width, height); + r300_blitter_end(r300); +} + +/* Clear a region of a depth stencil surface. */ +static void r300_clear_depth_stencil(struct pipe_context *pipe, + struct pipe_surface *dst, + unsigned clear_flags, + double depth, + unsigned stencil, + unsigned dstx, unsigned dsty, + unsigned width, unsigned height) +{ + struct r300_context *r300 = r300_context(pipe); + + r300_blitter_begin(r300, R300_CLEAR_SURFACE); + util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil, + dstx, dsty, width, height); + r300_blitter_end(r300); } /* Copy a block of pixels from one surface to another using HW. */ @@ -93,27 +164,12 @@ static void r300_hw_copy_region(struct pipe_context* pipe, unsigned width, unsigned height) { struct r300_context* r300 = r300_context(pipe); - struct r300_textures_state* state = - (struct r300_textures_state*)r300->textures_state.state; - - /* Yeah we have to save all those states to ensure this blitter operation - * is really transparent. The states will be restored by the blitter once - * copying is done. */ - r300_blitter_save_states(r300); - util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state); - - util_blitter_save_fragment_sampler_states( - r300->blitter, state->sampler_state_count, - (void**)state->sampler_states); - util_blitter_save_fragment_sampler_views( - r300->blitter, state->sampler_view_count, - (struct pipe_sampler_view**)state->sampler_views); - - /* Do a copy */ + r300_blitter_begin(r300, R300_COPY); util_blitter_copy_region(r300->blitter, dst, subdst, dstx, dsty, dstz, src, subsrc, srcx, srcy, srcz, width, height, TRUE); + r300_blitter_end(r300); } /* Copy a block of pixels from one surface to another. */ @@ -187,40 +243,6 @@ static void r300_resource_copy_region(struct pipe_context *pipe, } } -/* Clear a region of a color surface to a constant value. */ -static void r300_clear_render_target(struct pipe_context *pipe, - struct pipe_surface *dst, - const float *rgba, - unsigned dstx, unsigned dsty, - unsigned width, unsigned height) -{ - struct r300_context *r300 = r300_context(pipe); - - r300_blitter_save_states(r300); - util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state); - - util_blitter_clear_render_target(r300->blitter, dst, rgba, - dstx, dsty, width, height); -} - -/* Clear a region of a depth stencil surface. */ -static void r300_clear_depth_stencil(struct pipe_context *pipe, - struct pipe_surface *dst, - unsigned clear_flags, - double depth, - unsigned stencil, - unsigned dstx, unsigned dsty, - unsigned width, unsigned height) -{ - struct r300_context *r300 = r300_context(pipe); - - r300_blitter_save_states(r300); - util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state); - - util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil, - dstx, dsty, width, height); -} - void r300_init_blit_functions(struct r300_context *r300) { r300->context.clear = r300_clear; diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c index 57cf68f013..113942644c 100644 --- a/src/gallium/drivers/r300/r300_context.c +++ b/src/gallium/drivers/r300/r300_context.c @@ -39,7 +39,7 @@ static void r300_destroy_context(struct pipe_context* context) { struct r300_context* r300 = r300_context(context); - struct r300_query* query, * temp; + struct r300_query *query, *temp; struct r300_atom *atom; util_blitter_destroy(r300->blitter); @@ -55,9 +55,6 @@ static void r300_destroy_context(struct pipe_context* context) } } - /* Free the OQ BO. */ - context->screen->resource_destroy(context->screen, r300->oqbo); - /* If there are any queries pending or not destroyed, remove them now. */ foreach_s(query, temp, &r300->query_list) { remove_from_list(query); @@ -221,9 +218,6 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, r300_setup_atoms(r300); - /* Open up the OQ BO. */ - r300->oqbo = pipe_buffer_create(screen, - R300_BIND_OQBO, 4096); make_empty_list(&r300->query_list); r300_init_blit_functions(r300); diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h index 4869083837..90de53e2df 100644 --- a/src/gallium/drivers/r300/r300_context.h +++ b/src/gallium/drivers/r300/r300_context.h @@ -241,14 +241,23 @@ struct r300_constant_buffer { struct r300_query { /* The kind of query. Currently only OQ is supported. */ unsigned type; - /* The current count of this query. Required to be at least 32 bits. */ - unsigned int count; - /* The offset of this query into the query buffer, in bytes. */ - unsigned offset; + /* The number of pipes where query results are stored. */ + unsigned num_pipes; + /* How many results have been written, in dwords. It's incremented + * after end_query and flush. */ + unsigned num_results; /* if we've flushed the query */ boolean flushed; /* if begin has been emitted */ boolean begin_emitted; + + /* The buffer where query results are stored. */ + struct r300_winsys_buffer *buffer; + /* The size of the buffer. */ + unsigned buffer_size; + /* The domain of the buffer. */ + enum r300_buffer_domain domain; + /* Linked list members. */ struct r300_query* prev; struct r300_query* next; @@ -388,10 +397,11 @@ struct r300_context { /* Offset into the VBO. */ size_t vbo_offset; - /* Occlusion query buffer. */ - struct pipe_resource* oqbo; - /* Query list. */ + /* The currently active query. */ struct r300_query *query_current; + /* The saved query for blitter operations. */ + struct r300_query *blitter_saved_query; + /* Query list. */ struct r300_query query_list; /* Various CSO state objects. */ @@ -522,6 +532,11 @@ void r300_init_render_functions(struct r300_context *r300); void r300_init_state_functions(struct r300_context* r300); void r300_init_resource_functions(struct r300_context* r300); +/* r300_query.c */ +void r300_resume_query(struct r300_context *r300, + struct r300_query *query); +void r300_stop_query(struct r300_context *r300); + /* r300_render_translate.c */ void r300_begin_vertex_translate(struct r300_context *r300); void r300_end_vertex_translate(struct r300_context *r300); diff --git a/src/gallium/drivers/r300/r300_defines.h b/src/gallium/drivers/r300/r300_defines.h index 565a2f3e0e..d510d80a7b 100644 --- a/src/gallium/drivers/r300/r300_defines.h +++ b/src/gallium/drivers/r300/r300_defines.h @@ -32,12 +32,6 @@ #define R300_INVALID_FORMAT 0xffff -/* XXX: this is just a bandaid on larger problems in - * r300_screen_buffer.h which doesn't seem to be fully ported to - * gallium-resources. - */ -#define R300_BIND_OQBO (1<<21) - /* Tiling flags. */ enum r300_buffer_tiling { R300_BUFFER_LINEAR = 0, diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c index 96caf46148..46b958bdd9 100644 --- a/src/gallium/drivers/r300/r300_emit.c +++ b/src/gallium/drivers/r300/r300_emit.c @@ -353,13 +353,14 @@ void r300_emit_query_start(struct r300_context *r300, unsigned size, void*state) OUT_CS_REG(R300_ZB_ZPASS_DATA, 0); END_CS; query->begin_emitted = TRUE; + query->flushed = FALSE; } - static void r300_emit_query_end_frag_pipes(struct r300_context *r300, struct r300_query *query) { struct r300_capabilities* caps = &r300->screen->caps; + struct r300_winsys_buffer *buf = r300->query_current->buffer; CS_LOCALS(r300); assert(caps->num_frag_pipes); @@ -378,28 +379,28 @@ static void r300_emit_query_end_frag_pipes(struct r300_context *r300, /* pipe 3 only */ OUT_CS_REG(R300_SU_REG_DEST, 1 << 3); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 3), - 0, r300_buffer(r300->oqbo)->domain, 0); + OUT_CS_RELOC(buf, (query->num_results + 3) * 4, + 0, query->domain, 0); case 3: /* pipe 2 only */ OUT_CS_REG(R300_SU_REG_DEST, 1 << 2); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 2), - 0, r300_buffer(r300->oqbo)->domain, 0); + OUT_CS_RELOC(buf, (query->num_results + 2) * 4, + 0, query->domain, 0); case 2: /* pipe 1 only */ /* As mentioned above, accomodate RV380 and older. */ OUT_CS_REG(R300_SU_REG_DEST, 1 << (caps->high_second_pipe ? 3 : 1)); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 1), - 0, r300_buffer(r300->oqbo)->domain, 0); + OUT_CS_RELOC(buf, (query->num_results + 1) * 4, + 0, query->domain, 0); case 1: /* pipe 0 only */ OUT_CS_REG(R300_SU_REG_DEST, 1 << 0); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 0), - 0, r300_buffer(r300->oqbo)->domain, 0); + OUT_CS_RELOC(buf, (query->num_results + 0) * 4, + 0, query->domain, 0); break; default: fprintf(stderr, "r300: Implementation error: Chipset reports %d" @@ -415,12 +416,13 @@ static void r300_emit_query_end_frag_pipes(struct r300_context *r300, static void rv530_emit_query_end_single_z(struct r300_context *r300, struct r300_query *query) { + struct r300_winsys_buffer *buf = r300->query_current->buffer; CS_LOCALS(r300); BEGIN_CS(8); OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset, 0, r300_buffer(r300->oqbo)->domain, 0); + OUT_CS_RELOC(buf, query->num_results * 4, 0, query->domain, 0); OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL); END_CS; } @@ -428,15 +430,16 @@ static void rv530_emit_query_end_single_z(struct r300_context *r300, static void rv530_emit_query_end_double_z(struct r300_context *r300, struct r300_query *query) { + struct r300_winsys_buffer *buf = r300->query_current->buffer; CS_LOCALS(r300); BEGIN_CS(14); OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset, 0, r300_buffer(r300->oqbo)->domain, 0); + OUT_CS_RELOC(buf, (query->num_results + 0) * 4, 0, query->domain, 0); OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_1); OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1); - OUT_CS_BUF_RELOC(r300->oqbo, query->offset + sizeof(uint32_t), 0, r300_buffer(r300->oqbo)->domain, 0); + OUT_CS_RELOC(buf, (query->num_results + 1) * 4, 0, query->domain, 0); OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL); END_CS; } @@ -461,6 +464,13 @@ void r300_emit_query_end(struct r300_context* r300) r300_emit_query_end_frag_pipes(r300, query); query->begin_emitted = FALSE; + query->num_results += query->num_pipes; + + /* XXX grab all the results and reset the counter. */ + if (query->num_results >= query->buffer_size / 4 - 4) { + query->num_results = (query->buffer_size / 4) / 2; + fprintf(stderr, "r300: Rewinding OQBO...\n"); + } } void r300_emit_rs_state(struct r300_context* r300, unsigned size, void* state) @@ -895,10 +905,9 @@ validate: } } /* ...occlusion query buffer... */ - if (r300->query_start.dirty || - (r300->query_current && r300->query_current->begin_emitted)) { - if (!r300_add_buffer(r300->rws, r300->oqbo, - 0, r300_buffer(r300->oqbo)->domain)) { + if (r300->query_current) { + if (!r300->rws->add_buffer(r300->rws, r300->query_current->buffer, + 0, r300->query_current->domain)) { r300->context.flush(&r300->context, 0, NULL); goto validate; } diff --git a/src/gallium/drivers/r300/r300_query.c b/src/gallium/drivers/r300/r300_query.c index 7c08806368..42cb6ade76 100644 --- a/src/gallium/drivers/r300/r300_query.c +++ b/src/gallium/drivers/r300/r300_query.c @@ -26,6 +26,7 @@ #include "r300_context.h" #include "r300_screen.h" #include "r300_emit.h" +#include "r300_winsys.h" #include @@ -34,30 +35,27 @@ static struct pipe_query *r300_create_query(struct pipe_context *pipe, { struct r300_context *r300 = r300_context(pipe); struct r300_screen *r300screen = r300->screen; - unsigned query_size; - struct r300_query *q, *qptr; + struct r300_query *q; + + assert(query_type == PIPE_QUERY_OCCLUSION_COUNTER); q = CALLOC_STRUCT(r300_query); + if (!q) + return NULL; q->type = query_type; - assert(q->type == PIPE_QUERY_OCCLUSION_COUNTER); + q->domain = R300_DOMAIN_GTT; + q->buffer_size = 4096; if (r300screen->caps.family == CHIP_FAMILY_RV530) - query_size = r300screen->caps.num_z_pipes * sizeof(uint32_t); + q->num_pipes = r300screen->caps.num_z_pipes; else - query_size = r300screen->caps.num_frag_pipes * sizeof(uint32_t); + q->num_pipes = r300screen->caps.num_frag_pipes; - if (!is_empty_list(&r300->query_list)) { - qptr = last_elem(&r300->query_list); - q->offset = qptr->offset + query_size; - } insert_at_tail(&r300->query_list, q); - /* XXX */ - if (q->offset >= 4096) { - q->offset = 0; - fprintf(stderr, "r300: Rewinding OQBO...\n"); - } + /* Open up the occlusion query buffer. */ + q->buffer = r300->rws->buffer_create(r300->rws, 4096, 0, q->domain, q->buffer_size); return (struct pipe_query*)q; } @@ -65,18 +63,26 @@ static struct pipe_query *r300_create_query(struct pipe_context *pipe, static void r300_destroy_query(struct pipe_context* pipe, struct pipe_query* query) { - struct r300_query* q = (struct r300_query*)query; + struct r300_context *r300 = r300_context(pipe); + struct r300_query* q = r300_query(query); + r300->rws->buffer_reference(r300->rws, &q->buffer, NULL); remove_from_list(q); FREE(query); } +void r300_resume_query(struct r300_context *r300, + struct r300_query *query) +{ + r300->query_current = query; + r300->query_start.dirty = TRUE; +} + static void r300_begin_query(struct pipe_context* pipe, struct pipe_query* query) { - uint32_t value = ~0U; struct r300_context* r300 = r300_context(pipe); - struct r300_query* q = (struct r300_query*)query; + struct r300_query* q = r300_query(query); if (r300->query_current != NULL) { fprintf(stderr, "r300: begin_query: " @@ -85,30 +91,29 @@ static void r300_begin_query(struct pipe_context* pipe, return; } - pipe_buffer_write(pipe, - r300->oqbo, - q->offset, - sizeof value, - &value); + q->num_results = 0; + r300_resume_query(r300, q); +} - q->flushed = FALSE; - r300->query_current = q; - r300->query_start.dirty = TRUE; +void r300_stop_query(struct r300_context *r300) +{ + r300_emit_query_end(r300); + r300->query_current = NULL; } static void r300_end_query(struct pipe_context* pipe, struct pipe_query* query) { struct r300_context* r300 = r300_context(pipe); + struct r300_query *q = r300_query(query); - if ((struct r300_query*)query != r300->query_current) { + if (q != r300->query_current) { fprintf(stderr, "r300: end_query: Got invalid query.\n"); assert(0); return; } - r300_emit_query_end(r300); - r300->query_current = NULL; + r300_stop_query(r300); } static boolean r300_get_query_result(struct pipe_context* pipe, @@ -117,54 +122,28 @@ static boolean r300_get_query_result(struct pipe_context* pipe, void* vresult) { struct r300_context* r300 = r300_context(pipe); - struct r300_screen* r300screen = r300->screen; - struct r300_query *q = (struct r300_query*)query; - struct pipe_transfer *transfer; - unsigned flags = PIPE_TRANSFER_READ; - uint32_t* map; - uint32_t temp = 0; - unsigned i, num_results; + struct r300_query *q = r300_query(query); + unsigned flags, i; + uint32_t temp, *map; uint64_t *result = (uint64_t*)vresult; - if (q->flushed == FALSE) + if (!q->flushed) pipe->flush(pipe, 0, NULL); - if (!wait) { - flags |= PIPE_TRANSFER_DONTBLOCK; - } - map = pipe_buffer_map(pipe, r300->oqbo, flags, &transfer); + flags = PIPE_TRANSFER_READ | (!wait ? PIPE_TRANSFER_DONTBLOCK : 0); + + map = r300->rws->buffer_map(r300->rws, q->buffer, flags); if (!map) return FALSE; - map += q->offset / 4; - if (r300screen->caps.family == CHIP_FAMILY_RV530) - num_results = r300screen->caps.num_z_pipes; - else - num_results = r300screen->caps.num_frag_pipes; - - for (i = 0; i < num_results; i++) { - if (*map == ~0U) { - /* Looks like our results aren't ready yet. */ - if (wait) { - fprintf(stderr, "r300: Despite waiting, OQ results haven't " - "come in yet. This is a driver bug.\n" - "r300: Returning bogus results to avoid " - "a possible infinite loop...\n"); - temp = 987654321; - } else { - temp = ~0U; - } - break; - } + /* Sum up the results. */ + temp = 0; + for (i = 0; i < q->num_results; i++) { temp += *map; map++; } - pipe_buffer_unmap(pipe, r300->oqbo, transfer); - if (temp == ~0U) { - /* Our results haven't been written yet... */ - return FALSE; - } + r300->rws->buffer_unmap(r300->rws, q->buffer); *result = temp; return TRUE; diff --git a/src/gallium/drivers/r300/r300_screen_buffer.c b/src/gallium/drivers/r300/r300_screen_buffer.c index fe53f30b30..7959e6a2f9 100644 --- a/src/gallium/drivers/r300/r300_screen_buffer.c +++ b/src/gallium/drivers/r300/r300_screen_buffer.c @@ -264,10 +264,6 @@ struct pipe_resource *r300_buffer_create(struct pipe_screen *screen, rbuf->b.vtbl = &r300_buffer_vtbl; pipe_reference_init(&rbuf->b.b.reference, 1); rbuf->b.b.screen = screen; - - if (rbuf->b.b.bind & R300_BIND_OQBO) - alignment = 4096; - rbuf->domain = R300_DOMAIN_GTT; rbuf->buf = r300screen->rws->buffer_create(r300screen->rws, -- cgit v1.2.3