From ef919d8dcb9272ad7b23f5dbd8b7fb2f83393b42 Mon Sep 17 00:00:00 2001 From: =?utf8?q?Marek=20Ol=C5=A1=C3=A1k?= Date: Wed, 28 Aug 2019 21:27:05 -0400 Subject: [PATCH] radeonsi: remove redundant si_texture offset and size fields Acked-by: Pierre-Eric Pelloux-Prayer --- src/gallium/drivers/radeonsi/si_blit.c | 16 +-- src/gallium/drivers/radeonsi/si_clear.c | 8 +- .../drivers/radeonsi/si_compute_blit.c | 12 +- src/gallium/drivers/radeonsi/si_descriptors.c | 14 +- src/gallium/drivers/radeonsi/si_pipe.h | 13 +- src/gallium/drivers/radeonsi/si_state.c | 38 +++--- src/gallium/drivers/radeonsi/si_texture.c | 128 ++++++++---------- 7 files changed, 106 insertions(+), 123 deletions(-) diff --git a/src/gallium/drivers/radeonsi/si_blit.c b/src/gallium/drivers/radeonsi/si_blit.c index 68ed4cc5002..f5d7e2eb6ea 100644 --- a/src/gallium/drivers/radeonsi/si_blit.c +++ b/src/gallium/drivers/radeonsi/si_blit.c @@ -453,7 +453,7 @@ static void si_blit_decompress_color(struct si_context *sctx, if (need_dcc_decompress) { custom_blend = sctx->custom_blend_dcc_decompress; - assert(tex->dcc_offset); + assert(tex->surface.dcc_offset); /* disable levels without DCC */ for (int i = first_level; i <= last_level; i++) { @@ -519,7 +519,7 @@ si_decompress_color_texture(struct si_context *sctx, struct si_texture *tex, unsigned first_level, unsigned last_level) { /* CMASK or DCC can be discarded and we can still end up here. */ - if (!tex->cmask_buffer && !tex->surface.fmask_size && !tex->dcc_offset) + if (!tex->cmask_buffer && !tex->surface.fmask_size && !tex->surface.dcc_offset) return; si_blit_decompress_color(sctx, tex, first_level, last_level, 0, @@ -582,7 +582,7 @@ static void si_check_render_feedback_texture(struct si_context *sctx, { bool render_feedback = false; - if (!tex->dcc_offset) + if (!tex->surface.dcc_offset) return; for (unsigned j = 0; j < sctx->framebuffer.state.nr_cbufs; ++j) { @@ -840,7 +840,7 @@ static void si_decompress_subresource(struct pipe_context *ctx, si_decompress_depth(sctx, stex, planes, level, level, first_layer, last_layer); - } else if (stex->surface.fmask_size || stex->cmask_buffer || stex->dcc_offset) { + } else if (stex->surface.fmask_size || stex->cmask_buffer || stex->surface.dcc_offset) { /* If we've rendered into the framebuffer and it's a blitting * source, make sure the decompression pass is invoked * by dirtying the framebuffer. @@ -896,7 +896,7 @@ void si_resource_copy_region(struct pipe_context *ctx, !util_format_is_compressed(dst->format) && !util_format_is_depth_or_stencil(src->format) && src->nr_samples <= 1 && - !sdst->dcc_offset && + !sdst->surface.dcc_offset && !(dst->target != src->target && (src->target == PIPE_TEXTURE_1D_ARRAY || dst->target == PIPE_TEXTURE_1D_ARRAY))) { si_compute_copy_image(sctx, dst, dst_level, src, src_level, dstx, dsty, dstz, src_box); @@ -1288,12 +1288,12 @@ static void si_flush_resource(struct pipe_context *ctx, if (tex->dcc_separate_buffer && !tex->separate_dcc_dirty) return; - if (!tex->is_depth && (tex->cmask_buffer || tex->dcc_offset)) { + if (!tex->is_depth && (tex->cmask_buffer || tex->surface.dcc_offset)) { si_blit_decompress_color(sctx, tex, 0, res->last_level, 0, util_max_layer(res, 0), tex->dcc_separate_buffer != NULL); - if (tex->display_dcc_offset) + if (tex->surface.display_dcc_offset) si_retile_dcc(sctx, tex); } @@ -1333,7 +1333,7 @@ void si_decompress_dcc(struct si_context *sctx, struct si_texture *tex) /* If graphics is disabled, we can't decompress DCC, but it shouldn't * be compressed either. The caller should simply discard it. */ - if (!tex->dcc_offset || !sctx->has_graphics) + if (!tex->surface.dcc_offset || !sctx->has_graphics) return; si_blit_decompress_color(sctx, tex, 0, tex->buffer.b.b.last_level, diff --git a/src/gallium/drivers/radeonsi/si_clear.c b/src/gallium/drivers/radeonsi/si_clear.c index 34a39a0edae..68a0eeb36b4 100644 --- a/src/gallium/drivers/radeonsi/si_clear.c +++ b/src/gallium/drivers/radeonsi/si_clear.c @@ -245,7 +245,7 @@ bool vi_dcc_clear_level(struct si_context *sctx, dcc_offset = 0; } else { dcc_buffer = &tex->buffer.b.b; - dcc_offset = tex->dcc_offset; + dcc_offset = tex->surface.dcc_offset; } if (sctx->chip_class >= GFX9) { @@ -515,7 +515,7 @@ static void si_do_fast_color_clear(struct si_context *sctx, if (tex->buffer.b.b.nr_samples >= 2 && tex->cmask_buffer) { uint32_t clear_value = 0xCCCCCCCC; si_clear_buffer(sctx, &tex->cmask_buffer->b.b, - tex->cmask_offset, tex->surface.cmask_size, + tex->surface.cmask_offset, tex->surface.cmask_size, &clear_value, 4, SI_COHERENCY_CB_META, false); fmask_decompress_needed = true; } @@ -540,7 +540,7 @@ static void si_do_fast_color_clear(struct si_context *sctx, /* Do the fast clear. */ uint32_t clear_value = 0; si_clear_buffer(sctx, &tex->cmask_buffer->b.b, - tex->cmask_offset, tex->surface.cmask_size, + tex->surface.cmask_offset, tex->surface.cmask_size, &clear_value, 4, SI_COHERENCY_CB_META, false); eliminate_needed = true; } @@ -692,7 +692,7 @@ static void si_clear_render_target(struct pipe_context *ctx, struct si_context *sctx = (struct si_context *)ctx; struct si_texture *sdst = (struct si_texture*)dst->texture; - if (dst->texture->nr_samples <= 1 && !sdst->dcc_offset) { + if (dst->texture->nr_samples <= 1 && !sdst->surface.dcc_offset) { si_compute_clear_render_target(ctx, dst, color, dstx, dsty, width, height, render_condition_enabled); return; diff --git a/src/gallium/drivers/radeonsi/si_compute_blit.c b/src/gallium/drivers/radeonsi/si_compute_blit.c index 3704debd60b..4b3d1df935c 100644 --- a/src/gallium/drivers/radeonsi/si_compute_blit.c +++ b/src/gallium/drivers/radeonsi/si_compute_blit.c @@ -451,9 +451,9 @@ void si_retile_dcc(struct si_context *sctx, struct si_texture *tex) unsigned num_elements = tex->surface.u.gfx9.dcc_retile_num_elements; struct pipe_image_view img[3]; - assert(tex->dcc_retile_map_offset && tex->dcc_retile_map_offset <= UINT_MAX); - assert(tex->dcc_offset && tex->dcc_offset <= UINT_MAX); - assert(tex->display_dcc_offset && tex->display_dcc_offset <= UINT_MAX); + assert(tex->surface.dcc_retile_map_offset && tex->surface.dcc_retile_map_offset <= UINT_MAX); + assert(tex->surface.dcc_offset && tex->surface.dcc_offset <= UINT_MAX); + assert(tex->surface.display_dcc_offset && tex->surface.display_dcc_offset <= UINT_MAX); for (unsigned i = 0; i < 3; i++) { img[i].resource = &tex->buffer.b.b; @@ -463,15 +463,15 @@ void si_retile_dcc(struct si_context *sctx, struct si_texture *tex) img[0].format = use_uint16 ? PIPE_FORMAT_R16G16B16A16_UINT : PIPE_FORMAT_R32G32B32A32_UINT; - img[0].u.buf.offset = tex->dcc_retile_map_offset; + img[0].u.buf.offset = tex->surface.dcc_retile_map_offset; img[0].u.buf.size = num_elements * (use_uint16 ? 2 : 4); img[1].format = PIPE_FORMAT_R8_UINT; - img[1].u.buf.offset = tex->dcc_offset; + img[1].u.buf.offset = tex->surface.dcc_offset; img[1].u.buf.size = tex->surface.dcc_size; img[2].format = PIPE_FORMAT_R8_UINT; - img[2].u.buf.offset = tex->display_dcc_offset; + img[2].u.buf.offset = tex->surface.display_dcc_offset; img[2].u.buf.size = tex->surface.u.gfx9.display_dcc_size; ctx->set_shader_images(ctx, PIPE_SHADER_COMPUTE, 0, 3, img); diff --git a/src/gallium/drivers/radeonsi/si_descriptors.c b/src/gallium/drivers/radeonsi/si_descriptors.c index 38b9b11a83c..a5769df28cb 100644 --- a/src/gallium/drivers/radeonsi/si_descriptors.c +++ b/src/gallium/drivers/radeonsi/si_descriptors.c @@ -344,7 +344,7 @@ void si_set_mutable_tex_desc_fields(struct si_screen *sscreen, if (vi_dcc_enabled(tex, first_level)) { meta_va = (!tex->dcc_separate_buffer ? tex->buffer.gpu_address : 0) + - tex->dcc_offset; + tex->surface.dcc_offset; if (sscreen->info.chip_class == GFX8) { meta_va += base_level_info->dcc_offset; @@ -356,7 +356,7 @@ void si_set_mutable_tex_desc_fields(struct si_screen *sscreen, meta_va |= dcc_tile_swizzle; } else if (vi_tc_compat_htile_enabled(tex, first_level, is_stencil ? PIPE_MASK_S : PIPE_MASK_Z)) { - meta_va = tex->buffer.gpu_address + tex->htile_offset; + meta_va = tex->buffer.gpu_address + tex->surface.htile_offset; } if (meta_va) @@ -381,7 +381,7 @@ void si_set_mutable_tex_desc_fields(struct si_screen *sscreen, if (meta_va) { struct gfx9_surf_meta_flags meta; - if (tex->dcc_offset) + if (tex->surface.dcc_offset) meta = tex->surface.u.gfx9.dcc; else meta = tex->surface.u.gfx9.htile; @@ -409,7 +409,7 @@ void si_set_mutable_tex_desc_fields(struct si_screen *sscreen, if (meta_va) { struct gfx9_surf_meta_flags meta; - if (tex->dcc_offset) + if (tex->surface.dcc_offset) meta = tex->surface.u.gfx9.dcc; else meta = tex->surface.u.gfx9.htile; @@ -498,7 +498,7 @@ static bool color_needs_decompression(struct si_texture *tex) { return tex->surface.fmask_size || (tex->dirty_level_mask && - (tex->cmask_buffer || tex->dcc_offset)); + (tex->cmask_buffer || tex->surface.dcc_offset)); } static bool depth_needs_decompression(struct si_texture *tex) @@ -547,7 +547,7 @@ static void si_set_sampler_view(struct si_context *sctx, samplers->needs_color_decompress_mask &= ~(1u << slot); } - if (tex->dcc_offset && + if (tex->surface.dcc_offset && p_atomic_read(&tex->framebuffers_bound)) sctx->need_check_render_feedback = true; } @@ -2532,7 +2532,7 @@ static void si_make_texture_handle_resident(struct pipe_context *ctx, tex_handle); } - if (tex->dcc_offset && + if (tex->surface.dcc_offset && p_atomic_read(&tex->framebuffers_bound)) sctx->need_check_render_feedback = true; diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h index 86aee2e0824..2c3471de9ea 100644 --- a/src/gallium/drivers/radeonsi/si_pipe.h +++ b/src/gallium/drivers/radeonsi/si_pipe.h @@ -288,7 +288,6 @@ struct si_texture { struct si_resource buffer; struct radeon_surf surface; - uint64_t size; struct si_texture *flushed_depth_texture; /* One texture allocation can contain these buffers: @@ -300,20 +299,14 @@ struct si_texture { * - displayable DCC buffer (if the DCC buffer is not displayable) * - DCC retile mapping buffer (if the DCC buffer is not displayable) */ - uint64_t fmask_offset; - uint64_t cmask_offset; uint64_t cmask_base_address_reg; struct si_resource *cmask_buffer; - uint64_t dcc_offset; /* 0 = disabled */ - uint64_t display_dcc_offset; - uint64_t dcc_retile_map_offset; unsigned cb_color_info; /* fast clear enable bit */ unsigned color_clear_value[2]; unsigned last_msaa_resolve_target_micro_mode; unsigned num_level0_transfers; /* Depth buffer compression and fast clear. */ - uint64_t htile_offset; float depth_clear_value; uint16_t dirty_level_mask; /* each bit says if that mipmap is compressed */ uint16_t stencil_dirty_level_mask; /* each bit says if that mipmap is compressed */ @@ -1550,7 +1543,7 @@ si_texture_reference(struct si_texture **ptr, struct si_texture *res) static inline bool vi_dcc_enabled(struct si_texture *tex, unsigned level) { - return tex->dcc_offset && level < tex->surface.num_dcc_levels; + return tex->surface.dcc_offset && level < tex->surface.num_dcc_levels; } static inline unsigned @@ -1747,13 +1740,13 @@ si_htile_enabled(struct si_texture *tex, unsigned level, unsigned zs_mask) if (zs_mask == PIPE_MASK_S && tex->htile_stencil_disabled) return false; - return tex->htile_offset && level == 0; + return tex->surface.htile_offset && level == 0; } static inline bool vi_tc_compat_htile_enabled(struct si_texture *tex, unsigned level, unsigned zs_mask) { - assert(!tex->tc_compatible_htile || tex->htile_offset); + assert(!tex->tc_compatible_htile || tex->surface.htile_offset); return tex->tc_compatible_htile && si_htile_enabled(tex, level, zs_mask); } diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c index 1d086b65fa4..893968d1997 100644 --- a/src/gallium/drivers/radeonsi/si_state.c +++ b/src/gallium/drivers/radeonsi/si_state.c @@ -2535,7 +2535,7 @@ static void si_initialize_color_surface(struct si_context *sctx, color_attrib |= S_028C74_NUM_SAMPLES(log_samples) | S_028C74_NUM_FRAGMENTS(log_fragments); - if (tex->fmask_offset) { + if (tex->surface.fmask_offset) { color_info |= S_028C70_COMPRESSION(1); unsigned fmask_bankh = util_logbase2(tex->surface.u.legacy.fmask.bankh); @@ -2703,7 +2703,7 @@ static void si_init_depth_surface(struct si_context *sctx, } surf->db_htile_data_base = (tex->buffer.gpu_address + - tex->htile_offset) >> 8; + tex->surface.htile_offset) >> 8; surf->db_htile_surface = S_028ABC_FULL_CACHE(1) | S_028ABC_PIPE_ALIGNED(tex->surface.u.gfx9.htile.pipe_aligned); if (sctx->chip_class == GFX9) { @@ -2784,7 +2784,7 @@ static void si_init_depth_surface(struct si_context *sctx, } surf->db_htile_data_base = (tex->buffer.gpu_address + - tex->htile_offset) >> 8; + tex->surface.htile_offset) >> 8; surf->db_htile_surface = S_028ABC_FULL_CACHE(1); if (tex->tc_compatible_htile) { @@ -2828,7 +2828,7 @@ void si_update_fb_dirtiness_after_rendering(struct si_context *sctx) struct pipe_surface *surf = sctx->framebuffer.state.cbufs[i]; struct si_texture *tex = (struct si_texture*)surf->texture; - if (tex->fmask_offset) + if (tex->surface.fmask_offset) tex->dirty_level_mask |= 1 << surf->u.tex.level; if (tex->dcc_gather_statistics) tex->separate_dcc_dirty = true; @@ -3013,7 +3013,7 @@ static void si_set_framebuffer_state(struct pipe_context *ctx, if (surf->color_is_int10) sctx->framebuffer.color_is_int10 |= 1 << i; - if (tex->fmask_offset) + if (tex->surface.fmask_offset) sctx->framebuffer.compressed_cb_mask |= 1 << i; else sctx->framebuffer.uncompressed_cb_mask |= 1 << i; @@ -3206,8 +3206,8 @@ static void si_emit_framebuffer_state(struct si_context *sctx) if (cb->base.u.tex.level > 0) cb_color_info &= C_028C70_FAST_CLEAR; - if (tex->fmask_offset) { - cb_color_fmask = (tex->buffer.gpu_address + tex->fmask_offset) >> 8; + if (tex->surface.fmask_offset) { + cb_color_fmask = (tex->buffer.gpu_address + tex->surface.fmask_offset) >> 8; cb_color_fmask |= tex->surface.fmask_tile_swizzle; } @@ -3222,7 +3222,7 @@ static void si_emit_framebuffer_state(struct si_context *sctx) cb_color_info |= S_028C70_DCC_ENABLE(1); cb_dcc_base = ((!tex->dcc_separate_buffer ? tex->buffer.gpu_address : 0) + - tex->dcc_offset) >> 8; + tex->surface.dcc_offset) >> 8; unsigned dcc_tile_swizzle = tex->surface.tile_swizzle; dcc_tile_swizzle &= (tex->surface.dcc_alignment - 1) >> 8; @@ -3235,7 +3235,7 @@ static void si_emit_framebuffer_state(struct si_context *sctx) /* Set mutable surface parameters. */ cb_color_base += tex->surface.u.gfx9.surf_offset >> 8; cb_color_base |= tex->surface.tile_swizzle; - if (!tex->fmask_offset) + if (!tex->surface.fmask_offset) cb_color_fmask = cb_color_base; if (cb->base.u.tex.level > 0) cb_color_cmask = cb_color_base; @@ -3277,7 +3277,7 @@ static void si_emit_framebuffer_state(struct si_context *sctx) } else if (sctx->chip_class == GFX9) { struct gfx9_surf_meta_flags meta; - if (tex->dcc_offset) + if (tex->surface.dcc_offset) meta = tex->surface.u.gfx9.dcc; else meta = tex->surface.u.gfx9.cmask; @@ -3285,7 +3285,7 @@ static void si_emit_framebuffer_state(struct si_context *sctx) /* Set mutable surface parameters. */ cb_color_base += tex->surface.u.gfx9.surf_offset >> 8; cb_color_base |= tex->surface.tile_swizzle; - if (!tex->fmask_offset) + if (!tex->surface.fmask_offset) cb_color_fmask = cb_color_base; if (cb->base.u.tex.level > 0) cb_color_cmask = cb_color_base; @@ -3325,7 +3325,7 @@ static void si_emit_framebuffer_state(struct si_context *sctx) if (level_info->mode == RADEON_SURF_MODE_2D) cb_color_base |= tex->surface.tile_swizzle; - if (!tex->fmask_offset) + if (!tex->surface.fmask_offset) cb_color_fmask = cb_color_base; if (cb->base.u.tex.level > 0) cb_color_cmask = cb_color_base; @@ -3341,7 +3341,7 @@ static void si_emit_framebuffer_state(struct si_context *sctx) cb_color_pitch = S_028C64_TILE_MAX(pitch_tile_max); cb_color_slice = S_028C68_TILE_MAX(slice_tile_max); - if (tex->fmask_offset) { + if (tex->surface.fmask_offset) { if (sctx->chip_class >= GFX7) cb_color_pitch |= S_028C64_FMASK_TILE_MAX(tex->surface.u.legacy.fmask.pitch_in_pixels / 8 - 1); cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(tex->surface.u.legacy.fmask.tiling_index); @@ -3996,17 +3996,17 @@ gfx10_make_texture_descriptor(struct si_screen *screen, state[6] = 0; state[7] = 0; - if (tex->dcc_offset) { + if (tex->surface.dcc_offset) { state[6] |= S_00A018_MAX_UNCOMPRESSED_BLOCK_SIZE(V_028C78_MAX_BLOCK_SIZE_256B) | S_00A018_MAX_COMPRESSED_BLOCK_SIZE(V_028C78_MAX_BLOCK_SIZE_128B) | S_00A018_ALPHA_IS_ON_MSB(vi_alpha_is_on_msb(screen, pipe_format)); } /* Initialize the sampler view for FMASK. */ - if (tex->fmask_offset) { + if (tex->surface.fmask_offset) { uint32_t format; - va = tex->buffer.gpu_address + tex->fmask_offset; + va = tex->buffer.gpu_address + tex->surface.fmask_offset; #define FMASK(s,f) (((unsigned)(MAX2(1, s)) * 16) + (MAX2(1, f))) switch (FMASK(res->nr_samples, res->nr_storage_samples)) { @@ -4279,7 +4279,7 @@ si_make_texture_descriptor(struct si_screen *screen, state[5] |= S_008F24_LAST_ARRAY(last_layer); } - if (tex->dcc_offset) { + if (tex->surface.dcc_offset) { state[6] = S_008F28_ALPHA_IS_ON_MSB(vi_alpha_is_on_msb(screen, pipe_format)); } else { /* The last dword is unused by hw. The shader uses it to clear @@ -4294,10 +4294,10 @@ si_make_texture_descriptor(struct si_screen *screen, } /* Initialize the sampler view for FMASK. */ - if (tex->fmask_offset) { + if (tex->surface.fmask_offset) { uint32_t data_format, num_format; - va = tex->buffer.gpu_address + tex->fmask_offset; + va = tex->buffer.gpu_address + tex->surface.fmask_offset; #define FMASK(s,f) (((unsigned)(MAX2(1, s)) * 16) + (MAX2(1, f))) if (screen->info.chip_class == GFX9) { diff --git a/src/gallium/drivers/radeonsi/si_texture.c b/src/gallium/drivers/radeonsi/si_texture.c index ddef8da4e14..f2d0a10a843 100644 --- a/src/gallium/drivers/radeonsi/si_texture.c +++ b/src/gallium/drivers/radeonsi/si_texture.c @@ -449,23 +449,23 @@ void si_texture_discard_cmask(struct si_screen *sscreen, static bool si_can_disable_dcc(struct si_texture *tex) { /* We can't disable DCC if it can be written by another process. */ - return tex->dcc_offset && + return tex->surface.dcc_offset && (!tex->buffer.b.is_shared || !(tex->buffer.external_usage & PIPE_HANDLE_USAGE_FRAMEBUFFER_WRITE)); } static void si_texture_zero_dcc_fields(struct si_texture *tex) { - tex->dcc_offset = 0; - tex->display_dcc_offset = 0; - tex->dcc_retile_map_offset = 0; + tex->surface.dcc_offset = 0; + tex->surface.display_dcc_offset = 0; + tex->surface.dcc_retile_map_offset = 0; } static bool si_texture_discard_dcc(struct si_screen *sscreen, struct si_texture *tex) { if (!si_can_disable_dcc(tex)) { - assert(tex->display_dcc_offset == 0); + assert(tex->surface.display_dcc_offset == 0); return false; } @@ -584,12 +584,11 @@ static void si_reallocate_texture_inplace(struct si_context *sctx, tex->buffer.flags = new_tex->buffer.flags; tex->surface = new_tex->surface; - tex->size = new_tex->size; si_texture_reference(&tex->flushed_depth_texture, new_tex->flushed_depth_texture); - tex->fmask_offset = new_tex->fmask_offset; - tex->cmask_offset = new_tex->cmask_offset; + tex->surface.fmask_offset = new_tex->surface.fmask_offset; + tex->surface.cmask_offset = new_tex->surface.cmask_offset; tex->cmask_base_address_reg = new_tex->cmask_base_address_reg; if (tex->cmask_buffer == &tex->buffer) @@ -602,13 +601,13 @@ static void si_reallocate_texture_inplace(struct si_context *sctx, else si_resource_reference(&tex->cmask_buffer, new_tex->cmask_buffer); - tex->dcc_offset = new_tex->dcc_offset; + tex->surface.dcc_offset = new_tex->surface.dcc_offset; tex->cb_color_info = new_tex->cb_color_info; memcpy(tex->color_clear_value, new_tex->color_clear_value, sizeof(tex->color_clear_value)); tex->last_msaa_resolve_target_micro_mode = new_tex->last_msaa_resolve_target_micro_mode; - tex->htile_offset = new_tex->htile_offset; + tex->surface.htile_offset = new_tex->surface.htile_offset; tex->depth_clear_value = new_tex->depth_clear_value; tex->dirty_level_mask = new_tex->dirty_level_mask; tex->stencil_dirty_level_mask = new_tex->stencil_dirty_level_mask; @@ -630,10 +629,10 @@ static void si_reallocate_texture_inplace(struct si_context *sctx, new_tex->last_dcc_separate_buffer); if (new_bind_flag == PIPE_BIND_LINEAR) { - assert(!tex->htile_offset); + assert(!tex->surface.htile_offset); assert(!tex->cmask_buffer); assert(!tex->surface.fmask_size); - assert(!tex->dcc_offset); + assert(!tex->surface.dcc_offset); assert(!tex->is_depth); } @@ -659,10 +658,10 @@ static void si_set_tex_bo_metadata(struct si_screen *sscreen, if (sscreen->info.chip_class >= GFX9) { md.u.gfx9.swizzle_mode = surface->u.gfx9.surf.swizzle_mode; - if (tex->dcc_offset && !tex->dcc_separate_buffer) { + if (tex->surface.dcc_offset && !tex->dcc_separate_buffer) { uint64_t dcc_offset = - tex->display_dcc_offset ? tex->display_dcc_offset - : tex->dcc_offset; + tex->surface.display_dcc_offset ? tex->surface.display_dcc_offset + : tex->surface.dcc_offset; assert((dcc_offset >> 8) != 0 && (dcc_offset >> 8) < (1 << 24)); md.u.gfx9.dcc_offset_256B = dcc_offset >> 8; @@ -730,17 +729,17 @@ static void si_set_tex_bo_metadata(struct si_screen *sscreen, case GFX7: break; case GFX8: - desc[7] = tex->dcc_offset >> 8; + desc[7] = tex->surface.dcc_offset >> 8; break; case GFX9: - desc[7] = tex->dcc_offset >> 8; + desc[7] = tex->surface.dcc_offset >> 8; desc[5] &= C_008F24_META_DATA_ADDRESS; - desc[5] |= S_008F24_META_DATA_ADDRESS(tex->dcc_offset >> 40); + desc[5] |= S_008F24_META_DATA_ADDRESS(tex->surface.dcc_offset >> 40); break; case GFX10: desc[6] &= C_00A018_META_DATA_ADDRESS_LO; - desc[6] |= S_00A018_META_DATA_ADDRESS_LO(tex->dcc_offset >> 8); - desc[7] = tex->dcc_offset >> 16; + desc[6] |= S_00A018_META_DATA_ADDRESS_LO(tex->surface.dcc_offset >> 8); + desc[7] = tex->surface.dcc_offset >> 16; break; default: assert(0); @@ -814,11 +813,11 @@ static bool si_read_tex_bo_metadata(struct si_screen *sscreen, */ switch (sscreen->info.chip_class) { case GFX8: - tex->dcc_offset = (uint64_t)desc[7] << 8; + tex->surface.dcc_offset = (uint64_t)desc[7] << 8; break; case GFX9: - tex->dcc_offset = + tex->surface.dcc_offset = ((uint64_t)desc[7] << 8) | ((uint64_t)G_008F24_META_DATA_ADDRESS(desc[5]) << 40); tex->surface.u.gfx9.dcc.pipe_aligned = @@ -833,7 +832,7 @@ static bool si_read_tex_bo_metadata(struct si_screen *sscreen, break; case GFX10: - tex->dcc_offset = + tex->surface.dcc_offset = ((uint64_t)G_00A018_META_DATA_ADDRESS_LO(desc[6]) << 8) | ((uint64_t)desc[7] << 16); tex->surface.u.gfx9.dcc.pipe_aligned = @@ -865,14 +864,14 @@ static bool si_has_displayable_dcc(struct si_texture *tex) * (it can't be scanned out and rendered to simultaneously) */ if (sscreen->info.use_display_dcc_unaligned && - tex->dcc_offset && + tex->surface.dcc_offset && !tex->surface.u.gfx9.dcc.pipe_aligned && !tex->surface.u.gfx9.dcc.rb_aligned) return true; /* This needs an explicit flush (flush_resource). */ if (sscreen->info.use_display_dcc_with_retile_blit && - tex->display_dcc_offset) + tex->surface.display_dcc_offset) return true; return false; @@ -953,7 +952,7 @@ static bool si_texture_get_handle(struct pipe_screen* screen, * disable it for external clients that want write * access. */ - if ((usage & PIPE_HANDLE_USAGE_SHADER_WRITE && tex->dcc_offset) || + if ((usage & PIPE_HANDLE_USAGE_SHADER_WRITE && tex->surface.dcc_offset) || /* Displayable DCC requires an explicit flush. */ (!(usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH) && si_has_displayable_dcc(tex))) { @@ -965,7 +964,7 @@ static bool si_texture_get_handle(struct pipe_screen* screen, } if (!(usage & PIPE_HANDLE_USAGE_EXPLICIT_FLUSH) && - (tex->cmask_buffer || tex->dcc_offset)) { + (tex->cmask_buffer || tex->surface.dcc_offset)) { /* Eliminate fast clear (both CMASK and DCC) */ si_eliminate_fast_color_clear(sctx, tex); /* eliminate_fast_color_clear flushes the context */ @@ -1095,10 +1094,10 @@ void si_print_texture_info(struct si_screen *sscreen, tex->surface.u.gfx9.surf.epitch, tex->surface.u.gfx9.surf_pitch); - if (tex->fmask_offset) { + if (tex->surface.fmask_offset) { u_log_printf(log, " FMASK: offset=%"PRIu64", size=%"PRIu64", " "alignment=%u, swmode=%u, epitch=%u\n", - tex->fmask_offset, + tex->surface.fmask_offset, tex->surface.fmask_size, tex->surface.fmask_alignment, tex->surface.u.gfx9.fmask.swizzle_mode, @@ -1108,27 +1107,27 @@ void si_print_texture_info(struct si_screen *sscreen, if (tex->cmask_buffer) { u_log_printf(log, " CMask: offset=%"PRIu64", size=%u, " "alignment=%u, rb_aligned=%u, pipe_aligned=%u\n", - tex->cmask_offset, + tex->surface.cmask_offset, tex->surface.cmask_size, tex->surface.cmask_alignment, tex->surface.u.gfx9.cmask.rb_aligned, tex->surface.u.gfx9.cmask.pipe_aligned); } - if (tex->htile_offset) { + if (tex->surface.htile_offset) { u_log_printf(log, " HTile: offset=%"PRIu64", size=%u, alignment=%u, " "rb_aligned=%u, pipe_aligned=%u\n", - tex->htile_offset, + tex->surface.htile_offset, tex->surface.htile_size, tex->surface.htile_alignment, tex->surface.u.gfx9.htile.rb_aligned, tex->surface.u.gfx9.htile.pipe_aligned); } - if (tex->dcc_offset) { + if (tex->surface.dcc_offset) { u_log_printf(log, " DCC: offset=%"PRIu64", size=%u, " "alignment=%u, pitch_max=%u, num_dcc_levels=%u\n", - tex->dcc_offset, tex->surface.dcc_size, + tex->surface.dcc_offset, tex->surface.dcc_size, tex->surface.dcc_alignment, tex->surface.u.gfx9.display_dcc_pitch_max, tex->surface.num_dcc_levels); @@ -1150,10 +1149,10 @@ void si_print_texture_info(struct si_screen *sscreen, tex->surface.u.legacy.tile_split, tex->surface.u.legacy.pipe_config, (tex->surface.flags & RADEON_SURF_SCANOUT) != 0); - if (tex->fmask_offset) + if (tex->surface.fmask_offset) u_log_printf(log, " FMask: offset=%"PRIu64", size=%"PRIu64", alignment=%u, pitch_in_pixels=%u, " "bankh=%u, slice_tile_max=%u, tile_mode_index=%u\n", - tex->fmask_offset, tex->surface.fmask_size, tex->surface.fmask_alignment, + tex->surface.fmask_offset, tex->surface.fmask_size, tex->surface.fmask_alignment, tex->surface.u.legacy.fmask.pitch_in_pixels, tex->surface.u.legacy.fmask.bankh, tex->surface.u.legacy.fmask.slice_tile_max, @@ -1162,19 +1161,19 @@ void si_print_texture_info(struct si_screen *sscreen, if (tex->cmask_buffer) u_log_printf(log, " CMask: offset=%"PRIu64", size=%u, alignment=%u, " "slice_tile_max=%u\n", - tex->cmask_offset, tex->surface.cmask_size, tex->surface.cmask_alignment, + tex->surface.cmask_offset, tex->surface.cmask_size, tex->surface.cmask_alignment, tex->surface.u.legacy.cmask_slice_tile_max); - if (tex->htile_offset) + if (tex->surface.htile_offset) u_log_printf(log, " HTile: offset=%"PRIu64", size=%u, " "alignment=%u, TC_compatible = %u\n", - tex->htile_offset, tex->surface.htile_size, + tex->surface.htile_offset, tex->surface.htile_size, tex->surface.htile_alignment, tex->tc_compatible_htile); - if (tex->dcc_offset) { + if (tex->surface.dcc_offset) { u_log_printf(log, " DCC: offset=%"PRIu64", size=%u, alignment=%u\n", - tex->dcc_offset, tex->surface.dcc_size, + tex->surface.dcc_offset, tex->surface.dcc_size, tex->surface.dcc_alignment); for (i = 0; i <= tex->buffer.b.b.last_level; i++) u_log_printf(log, " DCCLevel[%i]: enabled=%u, offset=%u, " @@ -1273,15 +1272,6 @@ si_texture_create_object(struct pipe_screen *screen, */ tex->ps_draw_ratio = 0; - /* TODO: remove these */ - tex->fmask_offset = tex->surface.fmask_offset; - tex->cmask_offset = tex->surface.cmask_offset; - tex->htile_offset = tex->surface.htile_offset; - tex->dcc_offset = tex->surface.dcc_offset; - tex->display_dcc_offset = tex->surface.display_dcc_offset; - tex->dcc_retile_map_offset = tex->surface.dcc_retile_map_offset; - tex->size = tex->surface.total_size; - if (tex->is_depth) { if (sscreen->info.chip_class >= GFX9) { tex->can_sample_z = true; @@ -1309,7 +1299,7 @@ si_texture_create_object(struct pipe_screen *screen, /* Now create the backing buffer. */ if (!buf) { - si_init_resource_fields(sscreen, resource, tex->size, + si_init_resource_fields(sscreen, resource, tex->surface.total_size, tex->surface.surf_alignment); if (!si_alloc_resource(sscreen, resource)) @@ -1329,23 +1319,23 @@ si_texture_create_object(struct pipe_screen *screen, if (tex->cmask_buffer) { /* Initialize the cmask to 0xCC (= compressed state). */ si_screen_clear_buffer(sscreen, &tex->cmask_buffer->b.b, - tex->cmask_offset, tex->surface.cmask_size, + tex->surface.cmask_offset, tex->surface.cmask_size, 0xCCCCCCCC); } - if (tex->htile_offset) { + if (tex->surface.htile_offset) { uint32_t clear_value = 0; if (sscreen->info.chip_class >= GFX9 || tex->tc_compatible_htile) clear_value = 0x0000030F; si_screen_clear_buffer(sscreen, &tex->buffer.b.b, - tex->htile_offset, + tex->surface.htile_offset, tex->surface.htile_size, clear_value); } /* Initialize DCC only if the texture is not being imported. */ - if (!buf && tex->dcc_offset) { + if (!buf && tex->surface.dcc_offset) { /* Clear DCC to black for all tiles with DCC enabled. * * This fixes corruption in 3DMark Slingshot Extreme, which @@ -1355,13 +1345,13 @@ si_texture_create_object(struct pipe_screen *screen, tex->buffer.b.b.nr_samples <= 2) { /* Simple case - all tiles have DCC enabled. */ si_screen_clear_buffer(sscreen, &tex->buffer.b.b, - tex->dcc_offset, + tex->surface.dcc_offset, tex->surface.dcc_size, DCC_CLEAR_COLOR_0000); } else if (sscreen->info.chip_class >= GFX9) { /* Clear to uncompressed. Clearing this to black is complicated. */ si_screen_clear_buffer(sscreen, &tex->buffer.b.b, - tex->dcc_offset, + tex->surface.dcc_offset, tex->surface.dcc_size, DCC_UNCOMPRESSED); } else { @@ -1369,7 +1359,7 @@ si_texture_create_object(struct pipe_screen *screen, if (tex->buffer.b.b.nr_samples >= 2) { /* Clearing this to black is complicated. */ si_screen_clear_buffer(sscreen, &tex->buffer.b.b, - tex->dcc_offset, + tex->surface.dcc_offset, tex->surface.dcc_size, DCC_UNCOMPRESSED); } else { @@ -1387,13 +1377,13 @@ si_texture_create_object(struct pipe_screen *screen, /* Mipmap levels with DCC. */ if (size) { si_screen_clear_buffer(sscreen, &tex->buffer.b.b, - tex->dcc_offset, size, + tex->surface.dcc_offset, size, DCC_CLEAR_COLOR_0000); } /* Mipmap levels without DCC. */ if (size != tex->surface.dcc_size) { si_screen_clear_buffer(sscreen, &tex->buffer.b.b, - tex->dcc_offset + size, + tex->surface.dcc_offset + size, tex->surface.dcc_size - size, DCC_UNCOMPRESSED); } @@ -1401,7 +1391,7 @@ si_texture_create_object(struct pipe_screen *screen, } /* Upload the DCC retile map. */ - if (tex->dcc_retile_map_offset) { + if (tex->surface.dcc_retile_map_offset) { /* Use a staging buffer for the upload, because * the buffer backing the texture is unmappable. */ @@ -1429,10 +1419,10 @@ si_texture_create_object(struct pipe_screen *screen, struct pipe_box box; u_box_1d(0, buf->b.b.width0, &box); - assert(tex->dcc_retile_map_offset <= UINT_MAX); + assert(tex->surface.dcc_retile_map_offset <= UINT_MAX); mtx_lock(&sscreen->aux_context_lock); sctx->dma_copy(&sctx->b, &tex->buffer.b.b, 0, - tex->dcc_retile_map_offset, 0, 0, + tex->surface.dcc_retile_map_offset, 0, 0, &buf->b.b, 0, &box); sscreen->aux_context->flush(sscreen->aux_context, NULL, 0); mtx_unlock(&sscreen->aux_context_lock); @@ -1443,7 +1433,7 @@ si_texture_create_object(struct pipe_screen *screen, /* Initialize the CMASK base register value. */ tex->cmask_base_address_reg = - (tex->buffer.gpu_address + tex->cmask_offset) >> 8; + (tex->buffer.gpu_address + tex->surface.cmask_offset) >> 8; if (sscreen->debug_flags & DBG(VM)) { fprintf(stderr, "VM start=0x%"PRIX64" end=0x%"PRIX64" | Texture %ix%ix%i, %i levels, %i samples, %s\n", @@ -1829,11 +1819,11 @@ static void si_texture_invalidate_storage(struct si_context *sctx, /* Initialize the CMASK base address (needed even without CMASK). */ tex->cmask_base_address_reg = - (tex->buffer.gpu_address + tex->cmask_offset) >> 8; + (tex->buffer.gpu_address + tex->surface.cmask_offset) >> 8; p_atomic_inc(&sscreen->dirty_tex_counter); - sctx->num_alloc_tex_transfer_bytes += tex->size; + sctx->num_alloc_tex_transfer_bytes += tex->surface.total_size; } static void *si_texture_transfer_map(struct pipe_context *ctx, @@ -2391,7 +2381,7 @@ void vi_separate_dcc_try_enable(struct si_context *sctx, assert(sctx->chip_class >= GFX8); - if (tex->dcc_offset) + if (tex->surface.dcc_offset) return; /* already enabled */ /* Enable the DCC stat gathering. */ @@ -2426,7 +2416,7 @@ void vi_separate_dcc_try_enable(struct si_context *sctx, } /* dcc_offset is the absolute GPUVM address. */ - tex->dcc_offset = tex->dcc_separate_buffer->gpu_address; + tex->surface.dcc_offset = tex->dcc_separate_buffer->gpu_address; /* no need to flag anything since this is called by fast clear that * flags framebuffer state @@ -2485,7 +2475,7 @@ void vi_separate_dcc_process_and_reset_stats(struct pipe_context *ctx, assert(!tex->last_dcc_separate_buffer); tex->last_dcc_separate_buffer = tex->dcc_separate_buffer; tex->dcc_separate_buffer = NULL; - tex->dcc_offset = 0; + tex->surface.dcc_offset = 0; /* no need to flag anything since this is called after * decompression that re-sets framebuffer state */ -- 2.30.2