From e64633e8c3a5498998a45ab721bf80edca101cf5 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Marek=20Ol=C5=A1=C3=A1k?= Date: Sun, 22 Sep 2013 13:06:27 +0200 Subject: [PATCH] r600g,radeonsi: share r600_texture.c The function r600_choose_tiling is new and needs a review. The only change in functionality is that it enables 2D tiling for compressed textures on SI. It was probably accidentally turned off. v2: don't make scanout buffers linear --- src/gallium/drivers/r600/Makefile.sources | 1 - src/gallium/drivers/r600/r600_blit.c | 15 +- src/gallium/drivers/r600/r600_buffer.c | 58 +- src/gallium/drivers/r600/r600_pipe.h | 16 +- src/gallium/drivers/r600/r600_resource.h | 20 - src/gallium/drivers/r600/r600_state.c | 4 +- src/gallium/drivers/radeon/Makefile.sources | 1 + src/gallium/drivers/radeon/r600_pipe_common.c | 55 ++ src/gallium/drivers/radeon/r600_pipe_common.h | 36 + .../drivers/{r600 => radeon}/r600_texture.c | 436 ++++++----- src/gallium/drivers/radeonsi/Makefile.sources | 1 - src/gallium/drivers/radeonsi/r600.h | 3 - src/gallium/drivers/radeonsi/r600_blit.c | 13 +- src/gallium/drivers/radeonsi/r600_buffer.c | 46 +- src/gallium/drivers/radeonsi/r600_resource.c | 4 +- src/gallium/drivers/radeonsi/r600_resource.h | 12 - src/gallium/drivers/radeonsi/r600_texture.c | 864 --------------------- src/gallium/drivers/radeonsi/radeonsi_pipe.h | 10 - 18 files changed, 367 insertions(+), 1228 deletions(-) rename src/gallium/drivers/{r600 => radeon}/r600_texture.c (72%) delete mode 100644 src/gallium/drivers/radeonsi/r600_texture.c diff --git a/src/gallium/drivers/r600/Makefile.sources b/src/gallium/drivers/r600/Makefile.sources index df083d7..76fd164 100644 --- a/src/gallium/drivers/r600/Makefile.sources +++ b/src/gallium/drivers/r600/Makefile.sources @@ -9,7 +9,6 @@ C_SOURCES = \ r600_resource.c \ r600_shader.c \ r600_state.c \ - r600_texture.c \ r700_asm.c \ evergreen_hw_context.c \ evergreen_state.c \ diff --git a/src/gallium/drivers/r600/r600_blit.c b/src/gallium/drivers/r600/r600_blit.c index 20faabd..5000eae 100644 --- a/src/gallium/drivers/r600/r600_blit.c +++ b/src/gallium/drivers/r600/r600_blit.c @@ -105,12 +105,12 @@ static unsigned u_max_sample(struct pipe_resource *r) return r->nr_samples ? r->nr_samples - 1 : 0; } -void r600_blit_decompress_depth(struct pipe_context *ctx, - struct r600_texture *texture, - struct r600_texture *staging, - unsigned first_level, unsigned last_level, - unsigned first_layer, unsigned last_layer, - unsigned first_sample, unsigned last_sample) +static void r600_blit_decompress_depth(struct pipe_context *ctx, + struct r600_texture *texture, + struct r600_texture *staging, + unsigned first_level, unsigned last_level, + unsigned first_layer, unsigned last_layer, + unsigned first_sample, unsigned last_sample) { struct r600_context *rctx = (struct r600_context *)ctx; unsigned layer, level, sample, checked_last_layer, max_layer, max_sample; @@ -444,7 +444,7 @@ static void evergreen_check_alloc_cmask(struct pipe_context *ctx, if (tex->cmask_buffer) return; - r600_texture_init_cmask(rctx->screen, tex); + r600_texture_init_cmask(&rctx->screen->b, tex); /* update colorbuffer state bits */ if (tex->cmask_buffer != NULL) { @@ -955,4 +955,5 @@ void r600_init_blit_functions(struct r600_context *rctx) rctx->b.b.blit = r600_blit; rctx->b.b.flush_resource = r600_flush_resource; rctx->b.clear_buffer = r600_clear_buffer; + rctx->b.blit_decompress_depth = r600_blit_decompress_depth; } diff --git a/src/gallium/drivers/r600/r600_buffer.c b/src/gallium/drivers/r600/r600_buffer.c index 8fe78c3..107538a 100644 --- a/src/gallium/drivers/r600/r600_buffer.c +++ b/src/gallium/drivers/r600/r600_buffer.c @@ -121,7 +121,7 @@ static void *r600_buffer_transfer_map(struct pipe_context *ctx, /* Create a new one in the same pipe_resource. */ /* XXX We probably want a different alignment for buffers and textures. */ - r600_init_resource(rctx->screen, rbuffer, rbuffer->b.b.width0, 4096, + r600_init_resource(&rctx->screen->b, rbuffer, rbuffer->b.b.width0, 4096, TRUE, rbuffer->b.b.usage); /* We changed the buffer, now we need to bind it where the old one was bound. */ @@ -235,60 +235,6 @@ static const struct u_resource_vtbl r600_buffer_vtbl = NULL /* transfer_inline_write */ }; -bool r600_init_resource(struct r600_screen *rscreen, - struct r600_resource *res, - unsigned size, unsigned alignment, - bool use_reusable_pool, unsigned usage) -{ - uint32_t initial_domain, domains; - - switch(usage) { - case PIPE_USAGE_STAGING: - /* Staging resources participate in transfers, i.e. are used - * for uploads and downloads from regular resources. - * We generate them internally for some transfers. - */ - initial_domain = RADEON_DOMAIN_GTT; - domains = RADEON_DOMAIN_GTT; - break; - case PIPE_USAGE_DYNAMIC: - case PIPE_USAGE_STREAM: - /* Default to GTT, but allow the memory manager to move it to VRAM. */ - initial_domain = RADEON_DOMAIN_GTT; - domains = RADEON_DOMAIN_GTT | RADEON_DOMAIN_VRAM; - break; - case PIPE_USAGE_DEFAULT: - case PIPE_USAGE_STATIC: - case PIPE_USAGE_IMMUTABLE: - default: - /* Don't list GTT here, because the memory manager would put some - * resources to GTT no matter what the initial domain is. - * Not listing GTT in the domains improves performance a lot. */ - initial_domain = RADEON_DOMAIN_VRAM; - domains = RADEON_DOMAIN_VRAM; - break; - } - - res->buf = rscreen->b.ws->buffer_create(rscreen->b.ws, size, alignment, - use_reusable_pool, - initial_domain); - if (!res->buf) { - return false; - } - - res->cs_buf = rscreen->b.ws->buffer_get_cs_handle(res->buf); - res->domains = domains; - util_range_set_empty(&res->valid_buffer_range); - - if (rscreen->b.debug_flags & DBG_VM && res->b.b.target == PIPE_BUFFER) { - fprintf(stderr, "VM start=0x%llX end=0x%llX | Buffer %u bytes\n", - r600_resource_va(&rscreen->b.b, &res->b.b), - r600_resource_va(&rscreen->b.b, &res->b.b) + res->buf->size, - res->buf->size); - } - return true; -} - struct pipe_resource *r600_buffer_create(struct pipe_screen *screen, const struct pipe_resource *templ, unsigned alignment) @@ -304,7 +250,7 @@ struct pipe_resource *r600_buffer_create(struct pipe_screen *screen, rbuffer->b.vtbl = &r600_buffer_vtbl; util_range_init(&rbuffer->valid_buffer_range); - if (!r600_init_resource(rscreen, rbuffer, templ->width0, alignment, TRUE, templ->usage)) { + if (!r600_init_resource(&rscreen->b, rbuffer, templ->width0, alignment, TRUE, templ->usage)) { FREE(rbuffer); return NULL; } diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h index dc8274b..43fe080 100644 --- a/src/gallium/drivers/r600/r600_pipe.h +++ b/src/gallium/drivers/r600/r600_pipe.h @@ -65,9 +65,6 @@ #define R600_MAP_BUFFER_ALIGNMENT 64 -#define R600_ERR(fmt, args...) \ - fprintf(stderr, "EE %s:%d %s - "fmt, __FILE__, __LINE__, __func__, ##args) - #define R600_QUERY_DRAW_CALLS (PIPE_QUERY_DRIVER_SPECIFIC + 0) #define R600_QUERY_REQUESTED_VRAM (PIPE_QUERY_DRIVER_SPECIFIC + 1) #define R600_QUERY_REQUESTED_GTT (PIPE_QUERY_DRIVER_SPECIFIC + 2) @@ -201,9 +198,8 @@ struct r600_pipe_fences { pipe_mutex mutex; }; -/* features */ /* This must start from 16. */ -#define DBG_NO_HYPERZ (1 << 16) +/* features */ #define DBG_NO_LLVM (1 << 17) #define DBG_NO_CP_DMA (1 << 18) #define DBG_NO_ASYNC_DMA (1 << 19) @@ -616,22 +612,12 @@ void evergreen_update_db_shader_control(struct r600_context * rctx); void r600_copy_buffer(struct pipe_context *ctx, struct pipe_resource *dst, unsigned dstx, struct pipe_resource *src, const struct pipe_box *src_box); void r600_init_blit_functions(struct r600_context *rctx); -void r600_blit_decompress_depth(struct pipe_context *ctx, - struct r600_texture *texture, - struct r600_texture *staging, - unsigned first_level, unsigned last_level, - unsigned first_layer, unsigned last_layer, - unsigned first_sample, unsigned last_sample); void r600_decompress_depth_textures(struct r600_context *rctx, struct r600_samplerview_state *textures); void r600_decompress_color_textures(struct r600_context *rctx, struct r600_samplerview_state *textures); /* r600_buffer.c */ -bool r600_init_resource(struct r600_screen *rscreen, - struct r600_resource *res, - unsigned size, unsigned alignment, - bool use_reusable_pool, unsigned usage); struct pipe_resource *r600_buffer_create(struct pipe_screen *screen, const struct pipe_resource *templ, unsigned alignment); diff --git a/src/gallium/drivers/r600/r600_resource.h b/src/gallium/drivers/r600/r600_resource.h index a3ccdb0..96fea9a 100644 --- a/src/gallium/drivers/r600/r600_resource.h +++ b/src/gallium/drivers/r600/r600_resource.h @@ -89,24 +89,4 @@ static INLINE bool r600_can_read_depth(struct r600_texture *rtex) void r600_resource_destroy(struct pipe_screen *screen, struct pipe_resource *res); void r600_init_screen_resource_functions(struct pipe_screen *screen); -/* r600_texture */ -void r600_texture_get_fmask_info(struct r600_screen *rscreen, - struct r600_texture *rtex, - unsigned nr_samples, - struct r600_fmask_info *out); -void r600_texture_get_cmask_info(struct r600_screen *rscreen, - struct r600_texture *rtex, - struct r600_cmask_info *out); -void r600_texture_init_cmask(struct r600_screen *rscreen, - struct r600_texture *rtex); -struct pipe_resource *r600_texture_create(struct pipe_screen *screen, - const struct pipe_resource *templ); -struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen, - const struct pipe_resource *base, - struct winsys_handle *whandle); - -bool r600_init_flushed_depth_texture(struct pipe_context *ctx, - struct pipe_resource *texture, - struct r600_texture **staging); - #endif diff --git a/src/gallium/drivers/r600/r600_state.c b/src/gallium/drivers/r600/r600_state.c index 36ceb83..b01ab9c 100644 --- a/src/gallium/drivers/r600/r600_state.c +++ b/src/gallium/drivers/r600/r600_state.c @@ -1420,8 +1420,8 @@ static void r600_init_color_surface(struct r600_context *rctx, struct r600_cmask_info cmask; struct r600_fmask_info fmask; - r600_texture_get_cmask_info(rscreen, rtex, &cmask); - r600_texture_get_fmask_info(rscreen, rtex, 8, &fmask); + r600_texture_get_cmask_info(&rscreen->b, rtex, &cmask); + r600_texture_get_fmask_info(&rscreen->b, rtex, 8, &fmask); /* CMASK. */ if (!rctx->dummy_cmask || diff --git a/src/gallium/drivers/radeon/Makefile.sources b/src/gallium/drivers/radeon/Makefile.sources index 740dbc7..894f22a 100644 --- a/src/gallium/drivers/radeon/Makefile.sources +++ b/src/gallium/drivers/radeon/Makefile.sources @@ -1,6 +1,7 @@ C_SOURCES := \ r600_pipe_common.c \ r600_streamout.c \ + r600_texture.c \ radeon_uvd.c LLVM_C_FILES := \ diff --git a/src/gallium/drivers/radeon/r600_pipe_common.c b/src/gallium/drivers/radeon/r600_pipe_common.c index 8f0798a..190896b 100644 --- a/src/gallium/drivers/radeon/r600_pipe_common.c +++ b/src/gallium/drivers/radeon/r600_pipe_common.c @@ -25,6 +25,7 @@ */ #include "r600_pipe_common.h" +#include "r600_cs.h" #include "tgsi/tgsi_parse.h" #include "util/u_format_s3tc.h" @@ -337,3 +338,57 @@ void *r600_buffer_map_sync_with_rings(struct r600_common_context *ctx, return ctx->ws->buffer_map(resource->cs_buf, NULL, usage); } + +bool r600_init_resource(struct r600_common_screen *rscreen, + struct r600_resource *res, + unsigned size, unsigned alignment, + bool use_reusable_pool, unsigned usage) +{ + uint32_t initial_domain, domains; + + switch(usage) { + case PIPE_USAGE_STAGING: + /* Staging resources participate in transfers, i.e. are used + * for uploads and downloads from regular resources. + * We generate them internally for some transfers. + */ + initial_domain = RADEON_DOMAIN_GTT; + domains = RADEON_DOMAIN_GTT; + break; + case PIPE_USAGE_DYNAMIC: + case PIPE_USAGE_STREAM: + /* Default to GTT, but allow the memory manager to move it to VRAM. */ + initial_domain = RADEON_DOMAIN_GTT; + domains = RADEON_DOMAIN_GTT | RADEON_DOMAIN_VRAM; + break; + case PIPE_USAGE_DEFAULT: + case PIPE_USAGE_STATIC: + case PIPE_USAGE_IMMUTABLE: + default: + /* Don't list GTT here, because the memory manager would put some + * resources to GTT no matter what the initial domain is. + * Not listing GTT in the domains improves performance a lot. */ + initial_domain = RADEON_DOMAIN_VRAM; + domains = RADEON_DOMAIN_VRAM; + break; + } + + res->buf = rscreen->ws->buffer_create(rscreen->ws, size, alignment, + use_reusable_pool, + initial_domain); + if (!res->buf) { + return false; + } + + res->cs_buf = rscreen->ws->buffer_get_cs_handle(res->buf); + res->domains = domains; + util_range_set_empty(&res->valid_buffer_range); + + if (rscreen->debug_flags & DBG_VM && res->b.b.target == PIPE_BUFFER) { + fprintf(stderr, "VM start=0x%llX end=0x%llX | Buffer %u bytes\n", + r600_resource_va(&rscreen->b, &res->b.b), + r600_resource_va(&rscreen->b, &res->b.b) + res->buf->size, + res->buf->size); + } + return true; +} diff --git a/src/gallium/drivers/radeon/r600_pipe_common.h b/src/gallium/drivers/radeon/r600_pipe_common.h index 84fdff1..f71b712 100644 --- a/src/gallium/drivers/radeon/r600_pipe_common.h +++ b/src/gallium/drivers/radeon/r600_pipe_common.h @@ -71,6 +71,8 @@ #define DBG_GS (1 << 10) #define DBG_PS (1 << 11) #define DBG_CS (1 << 12) +/* features */ +#define DBG_NO_HYPERZ (1 << 13) /* The maximum allowed bit is 15. */ struct r600_common_context; @@ -243,6 +245,13 @@ struct r600_common_context { void (*clear_buffer)(struct pipe_context *ctx, struct pipe_resource *dst, unsigned offset, unsigned size, unsigned value); + + void (*blit_decompress_depth)(struct pipe_context *ctx, + struct r600_texture *texture, + struct r600_texture *staging, + unsigned first_level, unsigned last_level, + unsigned first_layer, unsigned last_layer, + unsigned first_sample, unsigned last_sample); }; /* r600_common_pipe.c */ @@ -263,6 +272,10 @@ boolean r600_rings_is_buffer_referenced(struct r600_common_context *ctx, void *r600_buffer_map_sync_with_rings(struct r600_common_context *ctx, struct r600_resource *resource, unsigned usage); +bool r600_init_resource(struct r600_common_screen *rscreen, + struct r600_resource *res, + unsigned size, unsigned alignment, + bool use_reusable_pool, unsigned usage); /* r600_streamout.c */ void r600_streamout_buffers_dirty(struct r600_common_context *rctx); @@ -273,6 +286,26 @@ void r600_set_streamout_targets(struct pipe_context *ctx, void r600_emit_streamout_end(struct r600_common_context *rctx); void r600_streamout_init(struct r600_common_context *rctx); +/* r600_texture.c */ +void r600_texture_get_fmask_info(struct r600_common_screen *rscreen, + struct r600_texture *rtex, + unsigned nr_samples, + struct r600_fmask_info *out); +void r600_texture_get_cmask_info(struct r600_common_screen *rscreen, + struct r600_texture *rtex, + struct r600_cmask_info *out); +void r600_texture_init_cmask(struct r600_common_screen *rscreen, + struct r600_texture *rtex); +bool r600_init_flushed_depth_texture(struct pipe_context *ctx, + struct pipe_resource *texture, + struct r600_texture **staging); +struct pipe_resource *r600_texture_create(struct pipe_screen *screen, + const struct pipe_resource *templ); +struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen, + const struct pipe_resource *base, + struct winsys_handle *whandle); + + /* Inline helpers. */ static INLINE struct r600_resource *r600_resource(struct pipe_resource *r) @@ -287,4 +320,7 @@ r600_resource_reference(struct r600_resource **ptr, struct r600_resource *res) (struct pipe_resource *)res); } +#define R600_ERR(fmt, args...) \ + fprintf(stderr, "EE %s:%d %s - "fmt, __FILE__, __LINE__, __func__, ##args) + #endif diff --git a/src/gallium/drivers/r600/r600_texture.c b/src/gallium/drivers/radeon/r600_texture.c similarity index 72% rename from src/gallium/drivers/r600/r600_texture.c rename to src/gallium/drivers/radeon/r600_texture.c index dcceda3..26eee12 100644 --- a/src/gallium/drivers/r600/r600_texture.c +++ b/src/gallium/drivers/radeon/r600_texture.c @@ -24,12 +24,11 @@ * Jerome Glisse * Corbin Simpson */ -#include "r600_formats.h" -#include "r600d.h" - -#include +#include "r600_pipe_common.h" +#include "r600_cs.h" +#include "util/u_format.h" #include "util/u_memory.h" - +#include /* Same as resource_copy_region, except that both upsampling and downsampling are allowed. */ static void r600_copy_region_with_blit(struct pipe_context *pipe, @@ -68,7 +67,7 @@ static void r600_copy_region_with_blit(struct pipe_context *pipe, /* Copy from a full GPU texture to a transfer's staging one. */ static void r600_copy_to_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer) { - struct r600_context *rctx = (struct r600_context*)ctx; + struct r600_common_context *rctx = (struct r600_common_context*)ctx; struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer; struct pipe_resource *dst = &rtransfer->staging->b.b; struct pipe_resource *src = transfer->resource; @@ -79,7 +78,7 @@ static void r600_copy_to_staging_texture(struct pipe_context *ctx, struct r600_t return; } - if (!rctx->b.dma_copy(ctx, dst, 0, 0, 0, 0, + if (!rctx->dma_copy(ctx, dst, 0, 0, 0, 0, src, transfer->level, &transfer->box)) { ctx->resource_copy_region(ctx, dst, 0, 0, 0, 0, @@ -90,7 +89,7 @@ static void r600_copy_to_staging_texture(struct pipe_context *ctx, struct r600_t /* Copy from a transfer's staging texture to a full GPU one. */ static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer) { - struct r600_context *rctx = (struct r600_context*)ctx; + struct r600_common_context *rctx = (struct r600_common_context*)ctx; struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer; struct pipe_resource *dst = transfer->resource; struct pipe_resource *src = &rtransfer->staging->b.b; @@ -105,7 +104,7 @@ static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600 return; } - if (!rctx->b.dma_copy(ctx, dst, transfer->level, + if (!rctx->dma_copy(ctx, dst, transfer->level, transfer->box.x, transfer->box.y, transfer->box.z, src, 0, &sbox)) { ctx->resource_copy_region(ctx, dst, transfer->level, @@ -125,7 +124,7 @@ static unsigned r600_texture_get_offset(struct r600_texture *rtex, unsigned leve box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format); } -static int r600_init_surface(struct r600_screen *rscreen, +static int r600_init_surface(struct r600_common_screen *rscreen, struct radeon_surface *surface, const struct pipe_resource *ptex, unsigned array_mode, @@ -147,7 +146,7 @@ static int r600_init_surface(struct r600_screen *rscreen, surface->array_size = 1; surface->last_level = ptex->last_level; - if (rscreen->b.chip_class >= EVERGREEN && !is_flushed_depth && + if (rscreen->chip_class >= EVERGREEN && !is_flushed_depth && ptex->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT) { surface->bpe = 4; /* stencil is allocated separately on evergreen */ } else { @@ -159,23 +158,8 @@ static int r600_init_surface(struct r600_screen *rscreen, } surface->nsamples = ptex->nr_samples ? ptex->nr_samples : 1; - surface->flags = 0; + surface->flags = RADEON_SURF_SET(array_mode, MODE); - switch (array_mode) { - case V_038000_ARRAY_1D_TILED_THIN1: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_1D, MODE); - break; - case V_038000_ARRAY_2D_TILED_THIN1: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_2D, MODE); - break; - case V_038000_ARRAY_LINEAR_ALIGNED: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_LINEAR_ALIGNED, MODE); - break; - case V_038000_ARRAY_LINEAR_GENERAL: - default: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_LINEAR, MODE); - break; - } switch (ptex->target) { case PIPE_TEXTURE_1D: surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_1D, TYPE); @@ -192,7 +176,7 @@ static int r600_init_surface(struct r600_screen *rscreen, surface->array_size = ptex->array_size; break; case PIPE_TEXTURE_2D_ARRAY: - case PIPE_TEXTURE_CUBE_ARRAY: /* cube array layout like 2d layout for now */ + case PIPE_TEXTURE_CUBE_ARRAY: /* cube array layout like 2d array */ surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_2D_ARRAY, TYPE); surface->array_size = ptex->array_size; break; @@ -212,9 +196,12 @@ static int r600_init_surface(struct r600_screen *rscreen, if (is_stencil) { surface->flags |= RADEON_SURF_SBUFFER | - RADEON_SURF_HAS_SBUFFER_MIPTREE; + RADEON_SURF_HAS_SBUFFER_MIPTREE; } } + if (rscreen->chip_class >= SI) { + surface->flags |= RADEON_SURF_HAS_TILE_MODE_INDEX; + } return 0; } @@ -222,14 +209,16 @@ static int r600_setup_surface(struct pipe_screen *screen, struct r600_texture *rtex, unsigned pitch_in_bytes_override) { - struct r600_screen *rscreen = (struct r600_screen*)screen; + struct r600_common_screen *rscreen = (struct r600_common_screen*)screen; int r; - r = rscreen->b.ws->surface_init(rscreen->b.ws, &rtex->surface); + r = rscreen->ws->surface_init(rscreen->ws, &rtex->surface); if (r) { return r; } + rtex->size = rtex->surface.bo_size; + if (pitch_in_bytes_override && pitch_in_bytes_override != rtex->surface.level[0].pitch_bytes) { /* old ddx on evergreen over estimate alignment for 1d, only 1 level * for those @@ -252,9 +241,9 @@ static boolean r600_texture_get_handle(struct pipe_screen* screen, struct r600_texture *rtex = (struct r600_texture*)ptex; struct r600_resource *resource = &rtex->resource; struct radeon_surface *surface = &rtex->surface; - struct r600_screen *rscreen = (struct r600_screen*)screen; + struct r600_common_screen *rscreen = (struct r600_common_screen*)screen; - rscreen->b.ws->buffer_set_tiling(resource->buf, + rscreen->ws->buffer_set_tiling(resource->buf, NULL, surface->level[0].mode >= RADEON_SURF_MODE_1D ? RADEON_LAYOUT_TILED : RADEON_LAYOUT_LINEAR, @@ -264,10 +253,10 @@ static boolean r600_texture_get_handle(struct pipe_screen* screen, surface->tile_split, surface->stencil_tile_split, surface->mtilea, - rtex->surface.level[0].pitch_bytes); + surface->level[0].pitch_bytes); - return rscreen->b.ws->buffer_get_handle(resource->buf, - rtex->surface.level[0].pitch_bytes, whandle); + return rscreen->ws->buffer_get_handle(resource->buf, + surface->level[0].pitch_bytes, whandle); } static void r600_texture_destroy(struct pipe_screen *screen, @@ -279,7 +268,7 @@ static void r600_texture_destroy(struct pipe_screen *screen, if (rtex->flushed_depth_texture) pipe_resource_reference((struct pipe_resource **)&rtex->flushed_depth_texture, NULL); - pipe_resource_reference((struct pipe_resource**)&rtex->htile, NULL); + pipe_resource_reference((struct pipe_resource**)&rtex->htile, NULL); if (rtex->cmask_buffer != &rtex->resource) { pipe_resource_reference((struct pipe_resource**)&rtex->cmask_buffer, NULL); } @@ -290,7 +279,7 @@ static void r600_texture_destroy(struct pipe_screen *screen, static const struct u_resource_vtbl r600_texture_vtbl; /* The number of samples can be specified independently of the texture. */ -void r600_texture_get_fmask_info(struct r600_screen *rscreen, +void r600_texture_get_fmask_info(struct r600_common_screen *rscreen, struct r600_texture *rtex, unsigned nr_samples, struct r600_fmask_info *out) @@ -305,11 +294,17 @@ void r600_texture_get_fmask_info(struct r600_screen *rscreen, fmask.nsamples = 1; fmask.flags |= RADEON_SURF_FMASK; + if (rscreen->chip_class >= SI) { + fmask.flags |= RADEON_SURF_HAS_TILE_MODE_INDEX; + } + switch (nr_samples) { case 2: case 4: fmask.bpe = 1; - fmask.bankh = 4; + if (rscreen->chip_class <= CAYMAN) { + fmask.bankh = 4; + } break; case 8: fmask.bpe = 4; @@ -322,11 +317,11 @@ void r600_texture_get_fmask_info(struct r600_screen *rscreen, /* Overallocate FMASK on R600-R700 to fix colorbuffer corruption. * This can be fixed by writing a separate FMASK allocator specifically * for R600-R700 asics. */ - if (rscreen->b.chip_class <= R700) { + if (rscreen->chip_class <= R700) { fmask.bpe *= 2; } - if (rscreen->b.ws->surface_init(rscreen->b.ws, &fmask)) { + if (rscreen->ws->surface_init(rscreen->ws, &fmask)) { R600_ERR("Got error in surface_init while allocating FMASK.\n"); return; } @@ -337,31 +332,24 @@ void r600_texture_get_fmask_info(struct r600_screen *rscreen, if (out->slice_tile_max) out->slice_tile_max -= 1; + out->tile_mode_index = fmask.tiling_index[0]; + out->pitch = fmask.level[0].nblk_x; out->bank_height = fmask.bankh; out->alignment = MAX2(256, fmask.bo_alignment); out->size = fmask.bo_size; } -static void r600_texture_allocate_fmask(struct r600_screen *rscreen, +static void r600_texture_allocate_fmask(struct r600_common_screen *rscreen, struct r600_texture *rtex) { - struct r600_fmask_info fmask; - r600_texture_get_fmask_info(rscreen, rtex, - rtex->resource.b.b.nr_samples, &fmask); + rtex->resource.b.b.nr_samples, &rtex->fmask); - rtex->fmask.bank_height = fmask.bank_height; - rtex->fmask.slice_tile_max = fmask.slice_tile_max; - rtex->fmask.offset = align(rtex->size, fmask.alignment); - rtex->fmask.size = fmask.size; + rtex->fmask.offset = align(rtex->size, rtex->fmask.alignment); rtex->size = rtex->fmask.offset + rtex->fmask.size; -#if 0 - printf("FMASK width=%u, height=%i, bits=%u, size=%u\n", - fmask.npix_x, fmask.npix_y, fmask.bpe * fmask.nsamples, rtex->fmask_size); -#endif } -void r600_texture_get_cmask_info(struct r600_screen *rscreen, +void r600_texture_get_cmask_info(struct r600_common_screen *rscreen, struct r600_texture *rtex, struct r600_cmask_info *out) { @@ -370,8 +358,8 @@ void r600_texture_get_cmask_info(struct r600_screen *rscreen, unsigned cmask_tile_elements = cmask_tile_width * cmask_tile_height; unsigned element_bits = 4; unsigned cmask_cache_bits = 1024; - unsigned num_pipes = rscreen->b.tiling_info.num_channels; - unsigned pipe_interleave_bytes = rscreen->b.tiling_info.group_bytes; + unsigned num_pipes = rscreen->tiling_info.num_channels; + unsigned pipe_interleave_bytes = rscreen->tiling_info.group_bytes; unsigned elements_per_macro_tile = (cmask_cache_bits / element_bits) * num_pipes; unsigned pixels_per_macro_tile = elements_per_macro_tile * cmask_tile_elements; @@ -394,41 +382,111 @@ void r600_texture_get_cmask_info(struct r600_screen *rscreen, out->size = rtex->surface.array_size * align(slice_bytes, base_align); } -static void r600_texture_allocate_cmask(struct r600_screen *rscreen, - struct r600_texture *rtex) +static void si_texture_get_cmask_info(struct r600_common_screen *rscreen, + struct r600_texture *rtex, + struct r600_cmask_info *out) { - struct r600_cmask_info cmask; + unsigned pipe_interleave_bytes = rscreen->tiling_info.group_bytes; + unsigned num_pipes = rscreen->tiling_info.num_channels; + unsigned cl_width, cl_height; - r600_texture_get_cmask_info(rscreen, rtex, &cmask); + switch (num_pipes) { + case 2: + cl_width = 32; + cl_height = 16; + break; + case 4: + cl_width = 32; + cl_height = 32; + break; + case 8: + cl_width = 64; + cl_height = 32; + break; + default: + assert(0); + return; + } + + unsigned base_align = num_pipes * pipe_interleave_bytes; + + unsigned width = align(rtex->surface.npix_x, cl_width*8); + unsigned height = align(rtex->surface.npix_y, cl_height*8); + unsigned slice_elements = (width * height) / (8*8); - rtex->cmask.slice_tile_max = cmask.slice_tile_max; - rtex->cmask.offset = align(rtex->size, cmask.alignment); - rtex->cmask.size = cmask.size; + /* Each element of CMASK is a nibble. */ + unsigned slice_bytes = slice_elements / 2; + + out->slice_tile_max = (width * height) / (128*128); + if (out->slice_tile_max) + out->slice_tile_max -= 1; + + out->alignment = MAX2(256, base_align); + out->size = rtex->surface.array_size * align(slice_bytes, base_align); +} + +static void r600_texture_allocate_cmask(struct r600_common_screen *rscreen, + struct r600_texture *rtex) +{ + if (rscreen->chip_class >= SI) { + si_texture_get_cmask_info(rscreen, rtex, &rtex->cmask); + } else { + r600_texture_get_cmask_info(rscreen, rtex, &rtex->cmask); + } + + rtex->cmask.offset = align(rtex->size, rtex->cmask.alignment); rtex->size = rtex->cmask.offset + rtex->cmask.size; -#if 0 - printf("CMASK: macro tile width = %u, macro tile height = %u, " - "pitch elements = %u, height = %u, slice tile max = %u\n", - macro_tile_width, macro_tile_height, pitch_elements, height, - rtex->cmask_slice_tile_max); -#endif } -void r600_texture_init_cmask(struct r600_screen *rscreen, - struct r600_texture *rtex) { - struct r600_cmask_info cmask; +void r600_texture_init_cmask(struct r600_common_screen *rscreen, + struct r600_texture *rtex) +{ + assert(rtex->cmask.size == 0); - assert(rtex->cmask.size == 0); + r600_texture_get_cmask_info(rscreen, rtex, &rtex->cmask); - r600_texture_get_cmask_info(rscreen, rtex, &cmask); - rtex->cmask.slice_tile_max = cmask.slice_tile_max; - rtex->cmask.offset = 0; - rtex->cmask.size = cmask.size; - rtex->cmask_buffer = (struct r600_resource *)pipe_buffer_create(&rscreen->b.b, - PIPE_BIND_CUSTOM, PIPE_USAGE_STATIC, rtex->cmask.size); + rtex->cmask_buffer = (struct r600_resource *) + pipe_buffer_create(&rscreen->b, PIPE_BIND_CUSTOM, + PIPE_USAGE_STATIC, rtex->cmask.size); + if (rtex->cmask_buffer == NULL) { + rtex->cmask.size = 0; + } +} - if (rtex->cmask_buffer == NULL) { - rtex->cmask.size = 0; - } +static void r600_texture_allocate_htile(struct r600_common_screen *rscreen, + struct r600_texture *rtex) +{ + unsigned sw = rtex->surface.level[0].nblk_x * rtex->surface.blk_w; + unsigned sh = rtex->surface.level[0].nblk_y * rtex->surface.blk_h; + unsigned htile_size; + unsigned npipes = rscreen->info.r600_num_tile_pipes; + + /* XXX also use it for other texture targets */ + if (rscreen->info.drm_minor < 26 || + rtex->resource.b.b.target != PIPE_TEXTURE_2D || + rtex->surface.level[0].nblk_x < 32 || + rtex->surface.level[0].nblk_y < 32) { + return; + } + + /* this alignment and htile size only apply to linear htile buffer */ + sw = align(sw, 16 << 3); + sh = align(sh, npipes << 3); + htile_size = (sw >> 3) * (sh >> 3) * 4; + /* must be aligned with 2K * npipes */ + htile_size = align(htile_size, (2 << 10) * npipes); + + /* XXX don't allocate it separately */ + rtex->htile = (struct r600_resource*)pipe_buffer_create(&rscreen->b, PIPE_BIND_CUSTOM, + PIPE_USAGE_STATIC, htile_size); + if (rtex->htile == NULL) { + /* this is not a fatal error as we can still keep rendering + * without htile buffer + */ + R600_ERR("r600: failed to create bo for htile buffers\n"); + } else { + r600_screen_clear_buffer(rscreen, &rtex->htile->b.b, 0, htile_size, 0); + } } static struct r600_texture * @@ -440,7 +498,7 @@ r600_texture_create_object(struct pipe_screen *screen, { struct r600_texture *rtex; struct r600_resource *resource; - struct r600_screen *rscreen = (struct r600_screen*)screen; + struct r600_common_screen *rscreen = (struct r600_common_screen*)screen; int r; rtex = CALLOC_STRUCT(r600_texture); @@ -457,6 +515,10 @@ r600_texture_create_object(struct pipe_screen *screen, /* don't include stencil-only formats which we don't support for rendering */ rtex->is_depth = util_format_has_depth(util_format_description(rtex->resource.b.b.format)); + /* Tiled depth textures utilize the non-displayable tile order. + * Applies to R600-Cayman. */ + rtex->non_disp_tiling = rtex->is_depth && rtex->surface.level[0].mode >= RADEON_SURF_MODE_1D; + rtex->surface = *surface; r = r600_setup_surface(screen, rtex, pitch_in_bytes_override); if (r) { @@ -464,7 +526,6 @@ r600_texture_create_object(struct pipe_screen *screen, return NULL; } - rtex->cmask_buffer = NULL; if (base->nr_samples > 1 && !rtex->is_depth && !buf) { r600_texture_allocate_fmask(rscreen, rtex); r600_texture_allocate_cmask(rscreen, rtex); @@ -477,42 +538,15 @@ r600_texture_create_object(struct pipe_screen *screen, return NULL; } - /* Tiled depth textures utilize the non-displayable tile order. */ - rtex->non_disp_tiling = rtex->is_depth && rtex->surface.level[0].mode >= RADEON_SURF_MODE_1D; - - /* only enable hyperz for PIPE_TEXTURE_2D not for PIPE_TEXTURE_2D_ARRAY - * Thought it might still be interessting to use hyperz for texture - * array without using fast clear features - */ - rtex->htile = NULL; - if (!(base->flags & (R600_RESOURCE_FLAG_TRANSFER | R600_RESOURCE_FLAG_FLUSHED_DEPTH)) && - util_format_is_depth_or_stencil(base->format) && - rscreen->b.info.drm_minor >= 26 && - !(rscreen->b.debug_flags & DBG_NO_HYPERZ) && - base->target == PIPE_TEXTURE_2D && - rtex->surface.level[0].nblk_x >= 32 && - rtex->surface.level[0].nblk_y >= 32) { - unsigned sw = rtex->surface.level[0].nblk_x * rtex->surface.blk_w; - unsigned sh = rtex->surface.level[0].nblk_y * rtex->surface.blk_h; - unsigned htile_size; - unsigned npipes = rscreen->b.info.r600_num_tile_pipes; - - /* this alignment and htile size only apply to linear htile buffer */ - sw = align(sw, 16 << 3); - sh = align(sh, npipes << 3); - htile_size = (sw >> 3) * (sh >> 3) * 4; - /* must be aligned with 2K * npipes */ - htile_size = align(htile_size, (2 << 10) * npipes); - - rtex->htile = (struct r600_resource*)pipe_buffer_create(&rscreen->b.b, PIPE_BIND_CUSTOM, - PIPE_USAGE_STATIC, htile_size); - if (rtex->htile == NULL) { - /* this is not a fatal error as we can still keep rendering - * without htile buffer - */ - R600_ERR("r600: failed to create bo for htile buffers\n"); + if (rtex->is_depth && + !(base->flags & (R600_RESOURCE_FLAG_TRANSFER | + R600_RESOURCE_FLAG_FLUSHED_DEPTH)) && + !(rscreen->debug_flags & DBG_NO_HYPERZ)) { + if (rscreen->chip_class >= SI) { + /* XXX implement Hyper-Z for SI. + * Reuse the CMASK allocator, which is almost the same as HTILE. */ } else { - r600_screen_clear_buffer(&rscreen->b, &rtex->htile->b.b, 0, htile_size, 0); + r600_texture_allocate_htile(rscreen, rtex); } } @@ -527,19 +561,18 @@ r600_texture_create_object(struct pipe_screen *screen, return NULL; } } else { - /* This is usually the window framebuffer. We want it in VRAM, always. */ resource->buf = buf; - resource->cs_buf = rscreen->b.ws->buffer_get_cs_handle(buf); - resource->domains = RADEON_DOMAIN_VRAM; + resource->cs_buf = rscreen->ws->buffer_get_cs_handle(buf); + resource->domains = RADEON_DOMAIN_GTT | RADEON_DOMAIN_VRAM; } if (rtex->cmask.size) { /* Initialize the cmask to 0xCC (= compressed state). */ - r600_screen_clear_buffer(&rscreen->b, &rtex->cmask_buffer->b.b, + r600_screen_clear_buffer(rscreen, &rtex->cmask_buffer->b.b, rtex->cmask.offset, rtex->cmask.size, 0xCCCCCCCC); } - if (rscreen->b.debug_flags & DBG_VM) { + if (rscreen->debug_flags & DBG_VM) { fprintf(stderr, "VM start=0x%llX end=0x%llX | Texture %ix%ix%i, %i levels, %i samples, %s\n", r600_resource_va(screen, &rtex->resource.b.b), r600_resource_va(screen, &rtex->resource.b.b) + rtex->resource.buf->size, @@ -547,7 +580,7 @@ r600_texture_create_object(struct pipe_screen *screen, base->nr_samples ? base->nr_samples : 1, util_format_short_name(base->format)); } - if (rscreen->b.debug_flags & DBG_TEX_DEPTH && rtex->is_depth && rtex->non_disp_tiling) { + if (rscreen->debug_flags & DBG_TEX_DEPTH && rtex->is_depth) { printf("Texture: npix_x=%u, npix_y=%u, npix_z=%u, blk_w=%u, " "blk_h=%u, blk_d=%u, array_size=%u, last_level=%u, " "bpe=%u, nsamples=%u, flags=%u\n", @@ -562,8 +595,8 @@ r600_texture_create_object(struct pipe_screen *screen, printf(" Z %i: offset=%llu, slice_size=%llu, npix_x=%u, " "npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, " "nblk_z=%u, pitch_bytes=%u, mode=%u\n", - i, (unsigned long long)rtex->surface.level[i].offset, - (unsigned long long)rtex->surface.level[i].slice_size, + i, rtex->surface.level[i].offset, + rtex->surface.level[i].slice_size, u_minify(rtex->resource.b.b.width0, i), u_minify(rtex->resource.b.b.height0, i), u_minify(rtex->resource.b.b.depth0, i), @@ -579,9 +612,9 @@ r600_texture_create_object(struct pipe_screen *screen, printf(" S %i: offset=%llu, slice_size=%llu, npix_x=%u, " "npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, " "nblk_z=%u, pitch_bytes=%u, mode=%u\n", - i, (unsigned long long)rtex->surface.stencil_level[i].offset, - (unsigned long long)rtex->surface.stencil_level[i].slice_size, - u_minify(rtex->resource.b.b.width0, i), + i, rtex->surface.stencil_level[i].offset, + rtex->surface.stencil_level[i].slice_size, + u_minify(rtex->resource.b.b.width0, i), u_minify(rtex->resource.b.b.height0, i), u_minify(rtex->resource.b.b.depth0, i), rtex->surface.stencil_level[i].nblk_x, @@ -595,44 +628,75 @@ r600_texture_create_object(struct pipe_screen *screen, return rtex; } -struct pipe_resource *r600_texture_create(struct pipe_screen *screen, - const struct pipe_resource *templ) +static unsigned r600_choose_tiling(struct r600_common_screen *rscreen, + const struct pipe_resource *templ) { - struct r600_screen *rscreen = (struct r600_screen*)screen; - struct radeon_surface surface; const struct util_format_description *desc = util_format_description(templ->format); - unsigned array_mode; - int r; - /* Default tiling mode for staging textures. */ - array_mode = V_038000_ARRAY_LINEAR_ALIGNED; - - /* Tiling doesn't work with the 422 (SUBSAMPLED) formats. That's not an issue, - * because 422 formats are used for videos, which prefer linear buffers - * for fast uploads anyway. */ - if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) && - (desc->layout != UTIL_FORMAT_LAYOUT_SUBSAMPLED) && - !(templ->bind & PIPE_BIND_LINEAR)) { - if (templ->flags & R600_RESOURCE_FLAG_FORCE_TILING) { - array_mode = V_038000_ARRAY_2D_TILED_THIN1; - } else if (!(templ->bind & PIPE_BIND_SCANOUT) && - templ->usage != PIPE_USAGE_STAGING && - templ->usage != PIPE_USAGE_STREAM && - templ->target != PIPE_TEXTURE_1D && - templ->target != PIPE_TEXTURE_1D_ARRAY && - templ->height0 > 3) { - array_mode = V_038000_ARRAY_2D_TILED_THIN1; - } else if (util_format_is_compressed(templ->format)) { - array_mode = V_038000_ARRAY_1D_TILED_THIN1; - } - } + /* MSAA resources must be 2D tiled. */ + if (templ->nr_samples > 1) + return RADEON_SURF_MODE_2D; + + /* Transfer resources should be linear. */ + if (templ->flags & R600_RESOURCE_FLAG_TRANSFER) + return RADEON_SURF_MODE_LINEAR_ALIGNED; + + /* Handle common candidates for the linear mode. + * Compressed textures must always be tiled. */ + if (!(templ->flags & R600_RESOURCE_FLAG_FORCE_TILING) && + !util_format_is_compressed(templ->format)) { + /* Tiling doesn't work with the 422 (SUBSAMPLED) formats on R600-Cayman. */ + if (rscreen->chip_class <= CAYMAN && + desc->layout == UTIL_FORMAT_LAYOUT_SUBSAMPLED) + return RADEON_SURF_MODE_LINEAR_ALIGNED; + + /* Cursors are linear on SI. + * (XXX double-check, maybe also use RADEON_SURF_SCANOUT) */ + if (rscreen->chip_class >= SI && + (templ->bind & PIPE_BIND_CURSOR)) + return RADEON_SURF_MODE_LINEAR_ALIGNED; + + if (templ->bind & PIPE_BIND_LINEAR) + return RADEON_SURF_MODE_LINEAR_ALIGNED; + + /* Textures with a very small height are recommended to be linear. */ + if (templ->target == PIPE_TEXTURE_1D || + templ->target == PIPE_TEXTURE_1D_ARRAY || + templ->height0 <= 4) + return RADEON_SURF_MODE_LINEAR_ALIGNED; + + /* Textures likely to be mapped often. */ + if (templ->usage == PIPE_USAGE_STAGING || + templ->usage == PIPE_USAGE_STREAM) + return RADEON_SURF_MODE_LINEAR_ALIGNED; + } + + /* Make small textures 1D tiled. */ + if (templ->width0 <= 16 || templ->height0 <= 16) + return RADEON_SURF_MODE_1D; + + /* XXX 2D tiling is currently unimplemented on CIK */ + if (rscreen->chip_class >= CIK) + return RADEON_SURF_MODE_1D; + + /* The allocator will switch to 1D if needed. */ + return RADEON_SURF_MODE_2D; +} + +struct pipe_resource *r600_texture_create(struct pipe_screen *screen, + const struct pipe_resource *templ) +{ + struct r600_common_screen *rscreen = (struct r600_common_screen*)screen; + struct radeon_surface surface = {0}; + int r; - r = r600_init_surface(rscreen, &surface, templ, array_mode, + r = r600_init_surface(rscreen, &surface, templ, + r600_choose_tiling(rscreen, templ), templ->flags & R600_RESOURCE_FLAG_FLUSHED_DEPTH); if (r) { return NULL; } - r = rscreen->b.ws->surface_best(rscreen->b.ws, &surface); + r = rscreen->ws->surface_best(rscreen->ws, &surface); if (r) { return NULL; } @@ -644,10 +708,10 @@ struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen, const struct pipe_resource *templ, struct winsys_handle *whandle) { - struct r600_screen *rscreen = (struct r600_screen*)screen; + struct r600_common_screen *rscreen = (struct r600_common_screen*)screen; struct pb_buffer *buf = NULL; unsigned stride = 0; - unsigned array_mode = 0; + unsigned array_mode; enum radeon_bo_layout micro, macro; struct radeon_surface surface; int r; @@ -657,27 +721,32 @@ struct pipe_resource *r600_texture_from_handle(struct pipe_screen *screen, templ->depth0 != 1 || templ->last_level != 0) return NULL; - buf = rscreen->b.ws->buffer_from_handle(rscreen->b.ws, whandle, &stride); + buf = rscreen->ws->buffer_from_handle(rscreen->ws, whandle, &stride); if (!buf) return NULL; - rscreen->b.ws->buffer_get_tiling(buf, µ, ¯o, + rscreen->ws->buffer_get_tiling(buf, µ, ¯o, &surface.bankw, &surface.bankh, &surface.tile_split, &surface.stencil_tile_split, &surface.mtilea); if (macro == RADEON_LAYOUT_TILED) - array_mode = V_0280A0_ARRAY_2D_TILED_THIN1; + array_mode = RADEON_SURF_MODE_2D; else if (micro == RADEON_LAYOUT_TILED) - array_mode = V_0280A0_ARRAY_1D_TILED_THIN1; + array_mode = RADEON_SURF_MODE_1D; else - array_mode = V_038000_ARRAY_LINEAR_ALIGNED; + array_mode = RADEON_SURF_MODE_LINEAR_ALIGNED; r = r600_init_surface(rscreen, &surface, templ, array_mode, false); if (r) { return NULL; } + + /* always set the scanout flags on SI */ + if (rscreen->chip_class >= SI) + surface.flags |= RADEON_SURF_SCANOUT; + return (struct pipe_resource *)r600_texture_create_object(screen, templ, stride, buf, &surface); } @@ -765,7 +834,7 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx, const struct pipe_box *box, struct pipe_transfer **ptransfer) { - struct r600_context *rctx = (struct r600_context*)ctx; + struct r600_common_context *rctx = (struct r600_common_context*)ctx; struct r600_texture *rtex = (struct r600_texture*)texture; struct r600_transfer *trans; boolean use_staging_texture = FALSE; @@ -780,14 +849,13 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx, * the CPU is much happier reading out of cached system memory * than uncached VRAM. */ - if (rtex->surface.level[level].mode >= RADEON_SURF_MODE_1D) { + if (rtex->surface.level[level].mode >= RADEON_SURF_MODE_1D) use_staging_texture = TRUE; - } /* Use a staging texture for uploads if the underlying BO is busy. */ if (!(usage & PIPE_TRANSFER_READ) && - (r600_rings_is_buffer_referenced(&rctx->b, rtex->resource.cs_buf, RADEON_USAGE_READWRITE) || - rctx->b.ws->buffer_is_busy(rtex->resource.buf, RADEON_USAGE_READWRITE))) { + (r600_rings_is_buffer_referenced(rctx, rtex->resource.cs_buf, RADEON_USAGE_READWRITE) || + rctx->ws->buffer_is_busy(rtex->resource.buf, RADEON_USAGE_READWRITE))) { use_staging_texture = TRUE; } @@ -835,8 +903,8 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx, struct pipe_resource *temp = ctx->screen->resource_create(ctx->screen, &resource); r600_copy_region_with_blit(ctx, temp, 0, 0, 0, 0, texture, level, box); - r600_blit_decompress_depth(ctx, (struct r600_texture*)temp, staging_depth, - 0, 0, 0, box->depth, 0, 0); + rctx->blit_decompress_depth(ctx, (struct r600_texture*)temp, staging_depth, + 0, 0, 0, box->depth, 0, 0); pipe_resource_reference((struct pipe_resource**)&temp, NULL); } } @@ -849,16 +917,16 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx, return NULL; } - r600_blit_decompress_depth(ctx, rtex, staging_depth, - level, level, - box->z, box->z + box->depth - 1, - 0, 0); + rctx->blit_decompress_depth(ctx, rtex, staging_depth, + level, level, + box->z, box->z + box->depth - 1, + 0, 0); offset = r600_texture_get_offset(staging_depth, level, box); } trans->transfer.stride = staging_depth->surface.level[level].pitch_bytes; - trans->transfer.layer_stride = staging_depth->surface.level[level].slice_size; + trans->transfer.layer_stride = staging_depth->surface.level[level].slice_size; trans->staging = (struct r600_resource*)staging_depth; } else if (use_staging_texture) { struct pipe_resource resource; @@ -893,7 +961,7 @@ static void *r600_texture_transfer_map(struct pipe_context *ctx, buf = &rtex->resource; } - if (!(map = r600_buffer_map_sync_with_rings(&rctx->b, buf, usage))) { + if (!(map = r600_buffer_map_sync_with_rings(rctx, buf, usage))) { pipe_resource_reference((struct pipe_resource**)&trans->staging, NULL); FREE(trans); return NULL; @@ -907,17 +975,17 @@ static void r600_texture_transfer_unmap(struct pipe_context *ctx, struct pipe_transfer* transfer) { struct r600_transfer *rtransfer = (struct r600_transfer*)transfer; - struct r600_context *rctx = (struct r600_context*)ctx; + struct r600_common_context *rctx = (struct r600_common_context*)ctx; struct radeon_winsys_cs_handle *buf; struct pipe_resource *texture = transfer->resource; struct r600_texture *rtex = (struct r600_texture*)texture; if (rtransfer->staging) { - buf = ((struct r600_resource *)rtransfer->staging)->cs_buf; + buf = rtransfer->staging->cs_buf; } else { - buf = ((struct r600_resource *)transfer->resource)->cs_buf; + buf = r600_resource(transfer->resource)->cs_buf; } - rctx->b.ws->buffer_unmap(buf); + rctx->ws->buffer_unmap(buf); if ((transfer->usage & PIPE_TRANSFER_WRITE) && rtransfer->staging) { if (rtex->is_depth && rtex->resource.b.b.nr_samples <= 1) { diff --git a/src/gallium/drivers/radeonsi/Makefile.sources b/src/gallium/drivers/radeonsi/Makefile.sources index 3b04c80..1302c6a 100644 --- a/src/gallium/drivers/radeonsi/Makefile.sources +++ b/src/gallium/drivers/radeonsi/Makefile.sources @@ -6,7 +6,6 @@ C_SOURCES := \ r600_query.c \ r600_resource.c \ radeonsi_shader.c \ - r600_texture.c \ r600_translate.c \ radeonsi_pm4.c \ radeonsi_compute.c \ diff --git a/src/gallium/drivers/radeonsi/r600.h b/src/gallium/drivers/radeonsi/r600.h index 0f5d653..0a9380b 100644 --- a/src/gallium/drivers/radeonsi/r600.h +++ b/src/gallium/drivers/radeonsi/r600.h @@ -32,9 +32,6 @@ #include "radeonsi_resource.h" -#define R600_ERR(fmt, args...) \ - fprintf(stderr, "EE %s:%d %s - "fmt, __FILE__, __LINE__, __func__, ##args) - struct winsys_handle; /* R600/R700 STATES */ diff --git a/src/gallium/drivers/radeonsi/r600_blit.c b/src/gallium/drivers/radeonsi/r600_blit.c index 483596e..851b80c 100644 --- a/src/gallium/drivers/radeonsi/r600_blit.c +++ b/src/gallium/drivers/radeonsi/r600_blit.c @@ -108,12 +108,12 @@ static unsigned u_max_sample(struct pipe_resource *r) return r->nr_samples ? r->nr_samples - 1 : 0; } -void r600_blit_decompress_depth(struct pipe_context *ctx, - struct r600_texture *texture, - struct r600_texture *staging, - unsigned first_level, unsigned last_level, - unsigned first_layer, unsigned last_layer, - unsigned first_sample, unsigned last_sample) +static void r600_blit_decompress_depth(struct pipe_context *ctx, + struct r600_texture *texture, + struct r600_texture *staging, + unsigned first_level, unsigned last_level, + unsigned first_layer, unsigned last_layer, + unsigned first_sample, unsigned last_sample) { struct r600_context *rctx = (struct r600_context *)ctx; unsigned layer, level, sample, checked_last_layer, max_layer, max_sample; @@ -749,4 +749,5 @@ void si_init_blit_functions(struct r600_context *rctx) rctx->b.b.resource_copy_region = r600_resource_copy_region; rctx->b.b.blit = si_blit; rctx->b.b.flush_resource = si_flush_resource; + rctx->b.blit_decompress_depth = r600_blit_decompress_depth; } diff --git a/src/gallium/drivers/radeonsi/r600_buffer.c b/src/gallium/drivers/radeonsi/r600_buffer.c index 12e5c4d..4c95130 100644 --- a/src/gallium/drivers/radeonsi/r600_buffer.c +++ b/src/gallium/drivers/radeonsi/r600_buffer.c @@ -96,50 +96,6 @@ static const struct u_resource_vtbl r600_buffer_vtbl = NULL /* transfer_inline_write */ }; -bool si_init_resource(struct r600_screen *rscreen, - struct r600_resource *res, - unsigned size, unsigned alignment, - boolean use_reusable_pool, unsigned usage) -{ - uint32_t initial_domain, domains; - - /* Staging resources particpate in transfers and blits only - * and are used for uploads and downloads from regular - * resources. We generate them internally for some transfers. - */ - if (usage == PIPE_USAGE_STAGING) { - domains = RADEON_DOMAIN_GTT; - initial_domain = RADEON_DOMAIN_GTT; - } else { - domains = RADEON_DOMAIN_GTT | RADEON_DOMAIN_VRAM; - - switch(usage) { - case PIPE_USAGE_DYNAMIC: - case PIPE_USAGE_STREAM: - case PIPE_USAGE_STAGING: - initial_domain = RADEON_DOMAIN_GTT; - break; - case PIPE_USAGE_DEFAULT: - case PIPE_USAGE_STATIC: - case PIPE_USAGE_IMMUTABLE: - default: - initial_domain = RADEON_DOMAIN_VRAM; - break; - } - } - - res->buf = rscreen->b.ws->buffer_create(rscreen->b.ws, size, alignment, - use_reusable_pool, - initial_domain); - if (!res->buf) { - return false; - } - - res->cs_buf = rscreen->b.ws->buffer_get_cs_handle(res->buf); - res->domains = domains; - return true; -} - struct pipe_resource *si_buffer_create(struct pipe_screen *screen, const struct pipe_resource *templ) { @@ -156,7 +112,7 @@ struct pipe_resource *si_buffer_create(struct pipe_screen *screen, rbuffer->b.vtbl = &r600_buffer_vtbl; util_range_init(&rbuffer->valid_buffer_range); - if (!si_init_resource(rscreen, rbuffer, templ->width0, alignment, TRUE, templ->usage)) { + if (!r600_init_resource(&rscreen->b, rbuffer, templ->width0, alignment, TRUE, templ->usage)) { FREE(rbuffer); return NULL; } diff --git a/src/gallium/drivers/radeonsi/r600_resource.c b/src/gallium/drivers/radeonsi/r600_resource.c index 660e279..745d3ba 100644 --- a/src/gallium/drivers/radeonsi/r600_resource.c +++ b/src/gallium/drivers/radeonsi/r600_resource.c @@ -29,7 +29,7 @@ static struct pipe_resource *r600_resource_create(struct pipe_screen *screen, if (templ->target == PIPE_BUFFER) { return si_buffer_create(screen, templ); } else { - return si_texture_create(screen, templ); + return r600_texture_create(screen, templ); } } @@ -40,7 +40,7 @@ static struct pipe_resource *r600_resource_from_handle(struct pipe_screen * scre if (templ->target == PIPE_BUFFER) { return NULL; } else { - return si_texture_from_handle(screen, templ, whandle); + return r600_texture_from_handle(screen, templ, whandle); } } diff --git a/src/gallium/drivers/radeonsi/r600_resource.h b/src/gallium/drivers/radeonsi/r600_resource.h index 116287c..be9ab33 100644 --- a/src/gallium/drivers/radeonsi/r600_resource.h +++ b/src/gallium/drivers/radeonsi/r600_resource.h @@ -31,18 +31,6 @@ struct r600_surface { void r600_init_screen_resource_functions(struct pipe_screen *screen); -/* r600_texture */ -struct pipe_resource *si_texture_create(struct pipe_screen *screen, - const struct pipe_resource *templ); -struct pipe_resource *si_texture_from_handle(struct pipe_screen *screen, - const struct pipe_resource *base, - struct winsys_handle *whandle); - -bool r600_init_flushed_depth_texture(struct pipe_context *ctx, - struct pipe_resource *texture, - struct r600_texture **staging); - - struct r600_context; void r600_upload_const_buffer(struct r600_context *rctx, struct r600_resource **rbuffer, diff --git a/src/gallium/drivers/radeonsi/r600_texture.c b/src/gallium/drivers/radeonsi/r600_texture.c deleted file mode 100644 index fd6afb9..0000000 --- a/src/gallium/drivers/radeonsi/r600_texture.c +++ /dev/null @@ -1,864 +0,0 @@ -/* - * Copyright 2010 Jerome Glisse - * - * 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 - * on 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 - * THE AUTHOR(S) AND/OR THEIR 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. - * - * Authors: - * Jerome Glisse - * Corbin Simpson - */ -#include -#include "pipe/p_screen.h" -#include "util/u_format.h" -#include "util/u_math.h" -#include "util/u_inlines.h" -#include "util/u_memory.h" -#include "pipebuffer/pb_buffer.h" -#include "radeonsi_pipe.h" -#include "r600_resource.h" -#include "sid.h" - -/* Same as resource_copy_region, except that both upsampling and downsampling are allowed. */ -static void r600_copy_region_with_blit(struct pipe_context *pipe, - struct pipe_resource *dst, - unsigned dst_level, - unsigned dstx, unsigned dsty, unsigned dstz, - struct pipe_resource *src, - unsigned src_level, - const struct pipe_box *src_box) -{ - struct pipe_blit_info blit; - - memset(&blit, 0, sizeof(blit)); - blit.src.resource = src; - blit.src.format = src->format; - blit.src.level = src_level; - blit.src.box = *src_box; - blit.dst.resource = dst; - blit.dst.format = dst->format; - blit.dst.level = dst_level; - blit.dst.box.x = dstx; - blit.dst.box.y = dsty; - blit.dst.box.z = dstz; - blit.dst.box.width = src_box->width; - blit.dst.box.height = src_box->height; - blit.dst.box.depth = src_box->depth; - blit.mask = util_format_get_mask(src->format) & - util_format_get_mask(dst->format); - blit.filter = PIPE_TEX_FILTER_NEAREST; - - if (blit.mask) { - pipe->blit(pipe, &blit); - } -} - -/* Copy from a full GPU texture to a transfer's staging one. */ -static void r600_copy_to_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer) -{ - struct r600_context *rctx = (struct r600_context*)ctx; - struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer; - struct pipe_resource *dst = &rtransfer->staging->b.b; - struct pipe_resource *src = transfer->resource; - - if (src->nr_samples > 1) { - r600_copy_region_with_blit(ctx, dst, 0, 0, 0, 0, - src, transfer->level, &transfer->box); - return; - } - - if (!rctx->b.dma_copy(ctx, dst, 0, 0, 0, 0, - src, transfer->level, - &transfer->box)) { - ctx->resource_copy_region(ctx, dst, 0, 0, 0, 0, - src, transfer->level, &transfer->box); - } -} - -/* Copy from a transfer's staging texture to a full GPU one. */ -static void r600_copy_from_staging_texture(struct pipe_context *ctx, struct r600_transfer *rtransfer) -{ - struct r600_context *rctx = (struct r600_context*)ctx; - struct pipe_transfer *transfer = (struct pipe_transfer*)rtransfer; - struct pipe_resource *dst = transfer->resource; - struct pipe_resource *src = &rtransfer->staging->b.b; - struct pipe_box sbox; - - u_box_3d(0, 0, 0, transfer->box.width, transfer->box.height, transfer->box.depth, &sbox); - - if (dst->nr_samples > 1) { - r600_copy_region_with_blit(ctx, dst, transfer->level, - transfer->box.x, transfer->box.y, transfer->box.z, - src, 0, &sbox); - return; - } - - if (!rctx->b.dma_copy(ctx, dst, transfer->level, - transfer->box.x, transfer->box.y, transfer->box.z, - src, 0, &sbox)) { - ctx->resource_copy_region(ctx, dst, transfer->level, - transfer->box.x, transfer->box.y, transfer->box.z, - src, 0, &sbox); - } -} - -static unsigned r600_texture_get_offset(struct r600_texture *rtex, unsigned level, - const struct pipe_box *box) -{ - enum pipe_format format = rtex->resource.b.b.format; - - return rtex->surface.level[level].offset + - box->z * rtex->surface.level[level].slice_size + - box->y / util_format_get_blockheight(format) * rtex->surface.level[level].pitch_bytes + - box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format); -} - -static int r600_init_surface(struct r600_screen *rscreen, - struct radeon_surface *surface, - const struct pipe_resource *ptex, - unsigned array_mode, - bool is_flushed_depth) -{ - const struct util_format_description *desc = - util_format_description(ptex->format); - bool is_depth, is_stencil; - - is_depth = util_format_has_depth(desc); - is_stencil = util_format_has_stencil(desc); - - surface->npix_x = ptex->width0; - surface->npix_y = ptex->height0; - surface->npix_z = ptex->depth0; - surface->blk_w = util_format_get_blockwidth(ptex->format); - surface->blk_h = util_format_get_blockheight(ptex->format); - surface->blk_d = 1; - surface->array_size = 1; - surface->last_level = ptex->last_level; - - if (!is_flushed_depth && - ptex->format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT) { - surface->bpe = 4; /* stencil is allocated separately on evergreen */ - } else { - surface->bpe = util_format_get_blocksize(ptex->format); - /* align byte per element on dword */ - if (surface->bpe == 3) { - surface->bpe = 4; - } - } - - surface->nsamples = ptex->nr_samples ? ptex->nr_samples : 1; - surface->flags = 0; - - switch (array_mode) { - case V_009910_ARRAY_1D_TILED_THIN1: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_1D, MODE); - break; - case V_009910_ARRAY_2D_TILED_THIN1: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_2D, MODE); - break; - case V_009910_ARRAY_LINEAR_ALIGNED: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_LINEAR_ALIGNED, MODE); - break; - case V_009910_ARRAY_LINEAR_GENERAL: - default: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_MODE_LINEAR, MODE); - break; - } - switch (ptex->target) { - case PIPE_TEXTURE_1D: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_1D, TYPE); - break; - case PIPE_TEXTURE_RECT: - case PIPE_TEXTURE_2D: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_2D, TYPE); - break; - case PIPE_TEXTURE_3D: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_3D, TYPE); - break; - case PIPE_TEXTURE_1D_ARRAY: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_1D_ARRAY, TYPE); - surface->array_size = ptex->array_size; - break; - case PIPE_TEXTURE_2D_ARRAY: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_2D_ARRAY, TYPE); - surface->array_size = ptex->array_size; - break; - case PIPE_TEXTURE_CUBE: - surface->flags |= RADEON_SURF_SET(RADEON_SURF_TYPE_CUBEMAP, TYPE); - break; - case PIPE_BUFFER: - default: - return -EINVAL; - } - if (ptex->bind & PIPE_BIND_SCANOUT) { - surface->flags |= RADEON_SURF_SCANOUT; - } - - if (!is_flushed_depth && is_depth) { - surface->flags |= RADEON_SURF_ZBUFFER; - if (is_stencil) { - surface->flags |= RADEON_SURF_SBUFFER | - RADEON_SURF_HAS_SBUFFER_MIPTREE; - } - } - surface->flags |= RADEON_SURF_HAS_TILE_MODE_INDEX; - return 0; -} - -static int r600_setup_surface(struct pipe_screen *screen, - struct r600_texture *rtex, - unsigned pitch_in_bytes_override) -{ - struct r600_screen *rscreen = (struct r600_screen*)screen; - int r; - - r = rscreen->b.ws->surface_init(rscreen->b.ws, &rtex->surface); - if (r) { - return r; - } - - rtex->size = rtex->surface.bo_size; - - if (pitch_in_bytes_override && pitch_in_bytes_override != rtex->surface.level[0].pitch_bytes) { - /* old ddx on evergreen over estimate alignment for 1d, only 1 level - * for those - */ - rtex->surface.level[0].nblk_x = pitch_in_bytes_override / rtex->surface.bpe; - rtex->surface.level[0].pitch_bytes = pitch_in_bytes_override; - rtex->surface.level[0].slice_size = pitch_in_bytes_override * rtex->surface.level[0].nblk_y; - if (rtex->surface.flags & RADEON_SURF_SBUFFER) { - rtex->surface.stencil_offset = - rtex->surface.stencil_level[0].offset = rtex->surface.level[0].slice_size; - } - } - return 0; -} - -static boolean r600_texture_get_handle(struct pipe_screen* screen, - struct pipe_resource *ptex, - struct winsys_handle *whandle) -{ - struct r600_texture *rtex = (struct r600_texture*)ptex; - struct r600_resource *resource = &rtex->resource; - struct radeon_surface *surface = &rtex->surface; - struct r600_screen *rscreen = (struct r600_screen*)screen; - - rscreen->b.ws->buffer_set_tiling(resource->buf, - NULL, - surface->level[0].mode >= RADEON_SURF_MODE_1D ? - RADEON_LAYOUT_TILED : RADEON_LAYOUT_LINEAR, - surface->level[0].mode >= RADEON_SURF_MODE_2D ? - RADEON_LAYOUT_TILED : RADEON_LAYOUT_LINEAR, - surface->bankw, surface->bankh, - surface->tile_split, - surface->stencil_tile_split, - surface->mtilea, - surface->level[0].pitch_bytes); - - return rscreen->b.ws->buffer_get_handle(resource->buf, - surface->level[0].pitch_bytes, whandle); -} - -static void r600_texture_destroy(struct pipe_screen *screen, - struct pipe_resource *ptex) -{ - struct r600_texture *rtex = (struct r600_texture*)ptex; - struct r600_resource *resource = &rtex->resource; - - if (rtex->flushed_depth_texture) - r600_resource_reference((struct r600_resource **)&rtex->flushed_depth_texture, NULL); - - pb_reference(&resource->buf, NULL); - FREE(rtex); -} - -static const struct u_resource_vtbl r600_texture_vtbl; - -/* The number of samples can be specified independently of the texture. */ -static void r600_texture_get_fmask_info(struct r600_screen *rscreen, - struct r600_texture *rtex, - unsigned nr_samples, - struct r600_fmask_info *out) -{ - /* FMASK is allocated like an ordinary texture. */ - struct radeon_surface fmask = rtex->surface; - - memset(out, 0, sizeof(*out)); - - fmask.bo_alignment = 0; - fmask.bo_size = 0; - fmask.nsamples = 1; - fmask.flags |= RADEON_SURF_FMASK | RADEON_SURF_HAS_TILE_MODE_INDEX; - - switch (nr_samples) { - case 2: - case 4: - fmask.bpe = 1; - break; - case 8: - fmask.bpe = 4; - break; - default: - R600_ERR("Invalid sample count for FMASK allocation.\n"); - return; - } - - if (rscreen->b.ws->surface_init(rscreen->b.ws, &fmask)) { - R600_ERR("Got error in surface_init while allocating FMASK.\n"); - return; - } - - assert(fmask.level[0].mode == RADEON_SURF_MODE_2D); - - out->slice_tile_max = (fmask.level[0].nblk_x * fmask.level[0].nblk_y) / 64; - if (out->slice_tile_max) - out->slice_tile_max -= 1; - - out->tile_mode_index = fmask.tiling_index[0]; - out->pitch = fmask.level[0].nblk_x; - out->bank_height = fmask.bankh; - out->alignment = MAX2(256, fmask.bo_alignment); - out->size = fmask.bo_size; -} - -static void r600_texture_allocate_fmask(struct r600_screen *rscreen, - struct r600_texture *rtex) -{ - r600_texture_get_fmask_info(rscreen, rtex, - rtex->resource.b.b.nr_samples, &rtex->fmask); - - rtex->fmask.offset = align(rtex->size, rtex->fmask.alignment); - rtex->size = rtex->fmask.offset + rtex->fmask.size; -} - -static void si_texture_get_cmask_info(struct r600_screen *rscreen, - struct r600_texture *rtex, - struct r600_cmask_info *out) -{ - unsigned pipe_interleave_bytes = rscreen->b.tiling_info.group_bytes; - unsigned num_pipes = rscreen->b.tiling_info.num_channels; - unsigned cl_width, cl_height; - - switch (num_pipes) { - case 2: - cl_width = 32; - cl_height = 16; - break; - case 4: - cl_width = 32; - cl_height = 32; - break; - case 8: - cl_width = 64; - cl_height = 32; - break; - default: - assert(0); - return; - } - - unsigned base_align = num_pipes * pipe_interleave_bytes; - - unsigned width = align(rtex->surface.npix_x, cl_width*8); - unsigned height = align(rtex->surface.npix_y, cl_height*8); - unsigned slice_elements = (width * height) / (8*8); - - /* Each element of CMASK is a nibble. */ - unsigned slice_bytes = slice_elements / 2; - - out->slice_tile_max = (width * height) / (128*128); - if (out->slice_tile_max) - out->slice_tile_max -= 1; - - out->alignment = MAX2(256, base_align); - out->size = rtex->surface.array_size * align(slice_bytes, base_align); -} - -static void r600_texture_allocate_cmask(struct r600_screen *rscreen, - struct r600_texture *rtex) -{ - si_texture_get_cmask_info(rscreen, rtex, &rtex->cmask); - - if (rtex->cmask.size) { - rtex->cmask.offset = align(rtex->size, rtex->cmask.alignment); - rtex->size = rtex->cmask.offset + rtex->cmask.size; - } -} - -static struct r600_texture * -r600_texture_create_object(struct pipe_screen *screen, - const struct pipe_resource *base, - unsigned pitch_in_bytes_override, - struct pb_buffer *buf, - struct radeon_surface *surface) -{ - struct r600_texture *rtex; - struct r600_resource *resource; - struct r600_screen *rscreen = (struct r600_screen*)screen; - int r; - - rtex = CALLOC_STRUCT(r600_texture); - if (rtex == NULL) - return NULL; - - resource = &rtex->resource; - resource->b.b = *base; - resource->b.vtbl = &r600_texture_vtbl; - pipe_reference_init(&resource->b.b.reference, 1); - resource->b.b.screen = screen; - rtex->pitch_override = pitch_in_bytes_override; - - /* don't include stencil-only formats which we don't support for rendering */ - rtex->is_depth = util_format_has_depth(util_format_description(rtex->resource.b.b.format)); - - rtex->surface = *surface; - r = r600_setup_surface(screen, rtex, pitch_in_bytes_override); - if (r) { - FREE(rtex); - return NULL; - } - - if (base->nr_samples > 1 && !rtex->is_depth && !buf) { - r600_texture_allocate_fmask(rscreen, rtex); - r600_texture_allocate_cmask(rscreen, rtex); - } - - if (!rtex->is_depth && base->nr_samples > 1 && - (!rtex->fmask.size || !rtex->cmask.size)) { - FREE(rtex); - return NULL; - } - - /* Now create the backing buffer. */ - if (!buf) { - unsigned base_align = rtex->surface.bo_alignment; - - if (!si_init_resource(rscreen, resource, rtex->size, base_align, FALSE, base->usage)) { - FREE(rtex); - return NULL; - } - } else if (buf) { - resource->buf = buf; - resource->cs_buf = rscreen->b.ws->buffer_get_cs_handle(buf); - resource->domains = RADEON_DOMAIN_GTT | RADEON_DOMAIN_VRAM; - } - - if (rtex->cmask.size) { - /* Initialize the cmask to 0xCC (= compressed state). */ - r600_screen_clear_buffer(&rscreen->b, &resource->b.b, - rtex->cmask.offset, rtex->cmask.size, - 0xCCCCCCCC); - } - - if (rscreen->b.debug_flags & DBG_TEX_DEPTH && rtex->is_depth) { - printf("Texture: npix_x=%u, npix_y=%u, npix_z=%u, blk_w=%u, " - "blk_h=%u, blk_d=%u, array_size=%u, last_level=%u, " - "bpe=%u, nsamples=%u, flags=%u\n", - rtex->surface.npix_x, rtex->surface.npix_y, - rtex->surface.npix_z, rtex->surface.blk_w, - rtex->surface.blk_h, rtex->surface.blk_d, - rtex->surface.array_size, rtex->surface.last_level, - rtex->surface.bpe, rtex->surface.nsamples, - rtex->surface.flags); - if (rtex->surface.flags & RADEON_SURF_ZBUFFER) { - for (int i = 0; i <= rtex->surface.last_level; i++) { - printf(" Z %i: offset=%llu, slice_size=%llu, npix_x=%u, " - "npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, " - "nblk_z=%u, pitch_bytes=%u, mode=%u\n", - i, rtex->surface.level[i].offset, - rtex->surface.level[i].slice_size, - rtex->surface.level[i].npix_x, - rtex->surface.level[i].npix_y, - rtex->surface.level[i].npix_z, - rtex->surface.level[i].nblk_x, - rtex->surface.level[i].nblk_y, - rtex->surface.level[i].nblk_z, - rtex->surface.level[i].pitch_bytes, - rtex->surface.level[i].mode); - } - } - if (rtex->surface.flags & RADEON_SURF_SBUFFER) { - for (int i = 0; i <= rtex->surface.last_level; i++) { - printf(" S %i: offset=%llu, slice_size=%llu, npix_x=%u, " - "npix_y=%u, npix_z=%u, nblk_x=%u, nblk_y=%u, " - "nblk_z=%u, pitch_bytes=%u, mode=%u\n", - i, rtex->surface.stencil_level[i].offset, - rtex->surface.stencil_level[i].slice_size, - rtex->surface.stencil_level[i].npix_x, - rtex->surface.stencil_level[i].npix_y, - rtex->surface.stencil_level[i].npix_z, - rtex->surface.stencil_level[i].nblk_x, - rtex->surface.stencil_level[i].nblk_y, - rtex->surface.stencil_level[i].nblk_z, - rtex->surface.stencil_level[i].pitch_bytes, - rtex->surface.stencil_level[i].mode); - } - } - } - return rtex; -} - -struct pipe_resource *si_texture_create(struct pipe_screen *screen, - const struct pipe_resource *templ) -{ - struct r600_screen *rscreen = (struct r600_screen*)screen; - struct radeon_surface surface = {0}; - unsigned array_mode = V_009910_ARRAY_LINEAR_ALIGNED; - int r; - - if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) && - !(templ->bind & (PIPE_BIND_CURSOR | PIPE_BIND_LINEAR))) { - if (templ->flags & R600_RESOURCE_FLAG_FORCE_TILING || - templ->nr_samples > 1) { - array_mode = V_009910_ARRAY_2D_TILED_THIN1; - } else if (util_format_is_compressed(templ->format)) { - array_mode = V_009910_ARRAY_1D_TILED_THIN1; - } else if (templ->usage != PIPE_USAGE_STAGING && - templ->usage != PIPE_USAGE_STREAM && - templ->target != PIPE_TEXTURE_1D && - templ->target != PIPE_TEXTURE_1D_ARRAY && - templ->height0 > 3 && - rscreen->b.chip_class < CIK /* XXX fix me */) { - array_mode = V_009910_ARRAY_2D_TILED_THIN1; - } else { - array_mode = V_009910_ARRAY_1D_TILED_THIN1; - } - } - - r = r600_init_surface(rscreen, &surface, templ, array_mode, - templ->flags & R600_RESOURCE_FLAG_FLUSHED_DEPTH); - if (r) { - return NULL; - } - r = rscreen->b.ws->surface_best(rscreen->b.ws, &surface); - if (r) { - return NULL; - } - return (struct pipe_resource *)r600_texture_create_object(screen, templ, - 0, NULL, &surface); -} - -struct pipe_resource *si_texture_from_handle(struct pipe_screen *screen, - const struct pipe_resource *templ, - struct winsys_handle *whandle) -{ - struct r600_screen *rscreen = (struct r600_screen*)screen; - struct pb_buffer *buf = NULL; - unsigned stride = 0; - unsigned array_mode; - enum radeon_bo_layout micro, macro; - struct radeon_surface surface; - int r; - - /* Support only 2D textures without mipmaps */ - if ((templ->target != PIPE_TEXTURE_2D && templ->target != PIPE_TEXTURE_RECT) || - templ->depth0 != 1 || templ->last_level != 0) - return NULL; - - buf = rscreen->b.ws->buffer_from_handle(rscreen->b.ws, whandle, &stride); - if (!buf) - return NULL; - - rscreen->b.ws->buffer_get_tiling(buf, µ, ¯o, - &surface.bankw, &surface.bankh, - &surface.tile_split, - &surface.stencil_tile_split, - &surface.mtilea); - - if (macro == RADEON_LAYOUT_TILED) - array_mode = V_009910_ARRAY_2D_TILED_THIN1; - else if (micro == RADEON_LAYOUT_TILED) - array_mode = V_009910_ARRAY_1D_TILED_THIN1; - else - array_mode = V_009910_ARRAY_LINEAR_ALIGNED; - - r = r600_init_surface(rscreen, &surface, templ, array_mode, false); - if (r) { - return NULL; - } - - /* always set the scanout flags */ - surface.flags |= RADEON_SURF_SCANOUT; - - return (struct pipe_resource *)r600_texture_create_object(screen, templ, - stride, buf, &surface); -} - -bool r600_init_flushed_depth_texture(struct pipe_context *ctx, - struct pipe_resource *texture, - struct r600_texture **staging) -{ - struct r600_texture *rtex = (struct r600_texture*)texture; - struct pipe_resource resource; - struct r600_texture **flushed_depth_texture = staging ? - staging : &rtex->flushed_depth_texture; - - if (!staging && rtex->flushed_depth_texture) - return true; /* it's ready */ - - resource.target = texture->target; - resource.format = texture->format; - resource.width0 = texture->width0; - resource.height0 = texture->height0; - resource.depth0 = texture->depth0; - resource.array_size = texture->array_size; - resource.last_level = texture->last_level; - resource.nr_samples = texture->nr_samples; - resource.usage = staging ? PIPE_USAGE_DYNAMIC : PIPE_USAGE_DEFAULT; - resource.bind = texture->bind & ~PIPE_BIND_DEPTH_STENCIL; - resource.flags = texture->flags | R600_RESOURCE_FLAG_FLUSHED_DEPTH; - - if (staging) - resource.flags |= R600_RESOURCE_FLAG_TRANSFER; - - *flushed_depth_texture = (struct r600_texture *)ctx->screen->resource_create(ctx->screen, &resource); - if (*flushed_depth_texture == NULL) { - R600_ERR("failed to create temporary texture to hold flushed depth\n"); - return false; - } - - (*flushed_depth_texture)->is_flushing_texture = TRUE; - return true; -} - -/** - * Initialize the pipe_resource descriptor to be of the same size as the box, - * which is supposed to hold a subregion of the texture "orig" at the given - * mipmap level. - */ -static void r600_init_temp_resource_from_box(struct pipe_resource *res, - struct pipe_resource *orig, - const struct pipe_box *box, - unsigned level, unsigned flags) -{ - memset(res, 0, sizeof(*res)); - res->format = orig->format; - res->width0 = box->width; - res->height0 = box->height; - res->depth0 = 1; - res->array_size = 1; - res->usage = flags & R600_RESOURCE_FLAG_TRANSFER ? PIPE_USAGE_STAGING : PIPE_USAGE_STATIC; - res->flags = flags; - - /* We must set the correct texture target and dimensions for a 3D box. */ - if (box->depth > 1 && util_max_layer(orig, level) > 0) - res->target = orig->target; - else - res->target = PIPE_TEXTURE_2D; - - switch (res->target) { - case PIPE_TEXTURE_1D_ARRAY: - case PIPE_TEXTURE_2D_ARRAY: - case PIPE_TEXTURE_CUBE_ARRAY: - res->array_size = box->depth; - break; - case PIPE_TEXTURE_3D: - res->depth0 = box->depth; - break; - default:; - } -} - -static void *si_texture_transfer_map(struct pipe_context *ctx, - struct pipe_resource *texture, - unsigned level, - unsigned usage, - const struct pipe_box *box, - struct pipe_transfer **ptransfer) -{ - struct r600_context *rctx = (struct r600_context *)ctx; - struct r600_texture *rtex = (struct r600_texture*)texture; - struct r600_transfer *trans; - boolean use_staging_texture = FALSE; - struct radeon_winsys_cs_handle *buf; - unsigned offset = 0; - char *map; - - /* We cannot map a tiled texture directly because the data is - * in a different order, therefore we do detiling using a blit. - * - * Also, use a temporary in GTT memory for read transfers, as - * the CPU is much happier reading out of cached system memory - * than uncached VRAM. - */ - if (rtex->surface.level[level].mode != RADEON_SURF_MODE_LINEAR_ALIGNED && - rtex->surface.level[level].mode != RADEON_SURF_MODE_LINEAR) - use_staging_texture = TRUE; - - /* Use a staging texture for uploads if the underlying BO is busy. */ - if (!(usage & PIPE_TRANSFER_READ) && - (rctx->b.ws->cs_is_buffer_referenced(rctx->b.rings.gfx.cs, rtex->resource.cs_buf, RADEON_USAGE_READWRITE) || - rctx->b.ws->buffer_is_busy(rtex->resource.buf, RADEON_USAGE_READWRITE))) { - use_staging_texture = TRUE; - } - - if (texture->flags & R600_RESOURCE_FLAG_TRANSFER) { - use_staging_texture = FALSE; - } - - if (use_staging_texture && (usage & PIPE_TRANSFER_MAP_DIRECTLY)) { - return NULL; - } - - trans = CALLOC_STRUCT(r600_transfer); - if (trans == NULL) - return NULL; - trans->transfer.resource = texture; - trans->transfer.level = level; - trans->transfer.usage = usage; - trans->transfer.box = *box; - - if (rtex->is_depth) { - struct r600_texture *staging_depth; - - if (rtex->resource.b.b.nr_samples > 1) { - /* MSAA depth buffers need to be converted to single sample buffers. - * - * Mapping MSAA depth buffers can occur if ReadPixels is called - * with a multisample GLX visual. - * - * First downsample the depth buffer to a temporary texture, - * then decompress the temporary one to staging. - * - * Only the region being mapped is transfered. - */ - struct pipe_resource resource; - - r600_init_temp_resource_from_box(&resource, texture, box, level, 0); - - if (!r600_init_flushed_depth_texture(ctx, &resource, &staging_depth)) { - R600_ERR("failed to create temporary texture to hold untiled copy\n"); - FREE(trans); - return NULL; - } - - if (usage & PIPE_TRANSFER_READ) { - struct pipe_resource *temp = ctx->screen->resource_create(ctx->screen, &resource); - - r600_copy_region_with_blit(ctx, temp, 0, 0, 0, 0, texture, level, box); - r600_blit_decompress_depth(ctx, (struct r600_texture*)temp, staging_depth, - 0, 0, 0, box->depth, 0, 0); - pipe_resource_reference((struct pipe_resource**)&temp, NULL); - } - } - else { - /* XXX: only readback the rectangle which is being mapped? */ - /* XXX: when discard is true, no need to read back from depth texture */ - if (!r600_init_flushed_depth_texture(ctx, texture, &staging_depth)) { - R600_ERR("failed to create temporary texture to hold untiled copy\n"); - FREE(trans); - return NULL; - } - - r600_blit_decompress_depth(ctx, rtex, staging_depth, - level, level, - box->z, box->z + box->depth - 1, - 0, 0); - - offset = r600_texture_get_offset(staging_depth, level, box); - } - - trans->transfer.stride = staging_depth->surface.level[level].pitch_bytes; - trans->transfer.layer_stride = staging_depth->surface.level[level].slice_size; - trans->staging = (struct r600_resource*)staging_depth; - } else if (use_staging_texture) { - struct pipe_resource resource; - struct r600_texture *staging; - - r600_init_temp_resource_from_box(&resource, texture, box, level, - R600_RESOURCE_FLAG_TRANSFER); - - /* Create the temporary texture. */ - staging = (struct r600_texture*)ctx->screen->resource_create(ctx->screen, &resource); - if (staging == NULL) { - R600_ERR("failed to create temporary texture to hold untiled copy\n"); - FREE(trans); - return NULL; - } - trans->staging = &staging->resource; - trans->transfer.stride = staging->surface.level[0].pitch_bytes; - trans->transfer.layer_stride = staging->surface.level[0].slice_size; - if (usage & PIPE_TRANSFER_READ) { - r600_copy_to_staging_texture(ctx, trans); - } - } else { - /* the resource is mapped directly */ - trans->transfer.stride = rtex->surface.level[level].pitch_bytes; - trans->transfer.layer_stride = rtex->surface.level[level].slice_size; - offset = r600_texture_get_offset(rtex, level, box); - } - - if (trans->staging) { - buf = trans->staging->cs_buf; - } else { - buf = rtex->resource.cs_buf; - } - - if (!(map = rctx->b.ws->buffer_map(buf, rctx->b.rings.gfx.cs, usage))) { - pipe_resource_reference((struct pipe_resource**)&trans->staging, NULL); - FREE(trans); - return NULL; - } - - *ptransfer = &trans->transfer; - return map + offset; -} - -static void si_texture_transfer_unmap(struct pipe_context *ctx, - struct pipe_transfer* transfer) -{ - struct r600_transfer *rtransfer = (struct r600_transfer*)transfer; - struct r600_context *rctx = (struct r600_context*)ctx; - struct radeon_winsys_cs_handle *buf; - struct pipe_resource *texture = transfer->resource; - struct r600_texture *rtex = (struct r600_texture*)texture; - - if (rtransfer->staging) { - buf = rtransfer->staging->cs_buf; - } else { - buf = r600_resource(transfer->resource)->cs_buf; - } - rctx->b.ws->buffer_unmap(buf); - - if ((transfer->usage & PIPE_TRANSFER_WRITE) && rtransfer->staging) { - if (rtex->is_depth && rtex->resource.b.b.nr_samples <= 1) { - ctx->resource_copy_region(ctx, texture, transfer->level, - transfer->box.x, transfer->box.y, transfer->box.z, - &rtransfer->staging->b.b, transfer->level, - &transfer->box); - } else { - r600_copy_from_staging_texture(ctx, rtransfer); - } - } - - if (rtransfer->staging) - pipe_resource_reference((struct pipe_resource**)&rtransfer->staging, NULL); - - FREE(transfer); -} - -static const struct u_resource_vtbl r600_texture_vtbl = -{ - r600_texture_get_handle, /* get_handle */ - r600_texture_destroy, /* resource_destroy */ - si_texture_transfer_map, /* transfer_map */ - u_default_transfer_flush_region,/* transfer_flush_region */ - si_texture_transfer_unmap, /* transfer_unmap */ - NULL /* transfer_inline_write */ -}; diff --git a/src/gallium/drivers/radeonsi/radeonsi_pipe.h b/src/gallium/drivers/radeonsi/radeonsi_pipe.h index 3e7332d..26f7e09 100644 --- a/src/gallium/drivers/radeonsi/radeonsi_pipe.h +++ b/src/gallium/drivers/radeonsi/radeonsi_pipe.h @@ -205,22 +205,12 @@ struct r600_context { /* r600_blit.c */ void si_init_blit_functions(struct r600_context *rctx); -void r600_blit_decompress_depth(struct pipe_context *ctx, - struct r600_texture *texture, - struct r600_texture *staging, - unsigned first_level, unsigned last_level, - unsigned first_layer, unsigned last_layer, - unsigned first_sample, unsigned last_sample); void si_flush_depth_textures(struct r600_context *rctx, struct r600_textures_info *textures); void r600_decompress_color_textures(struct r600_context *rctx, struct r600_textures_info *textures); /* r600_buffer.c */ -bool si_init_resource(struct r600_screen *rscreen, - struct r600_resource *res, - unsigned size, unsigned alignment, - boolean use_reusable_pool, unsigned usage); struct pipe_resource *si_buffer_create(struct pipe_screen *screen, const struct pipe_resource *templ); void r600_upload_index_buffer(struct r600_context *rctx, -- 2.7.4