diff --git a/src/gallium/drivers/ilo/ilo_3d.c b/src/gallium/drivers/ilo/ilo_3d.c index 90063761d16..48984ba006a 100644 --- a/src/gallium/drivers/ilo/ilo_3d.c +++ b/src/gallium/drivers/ilo/ilo_3d.c @@ -746,7 +746,7 @@ upload_shaders(struct ilo_3d *hw3d, struct ilo_shader_cache *shc) intel_bo_unreference(hw3d->kernel.bo); hw3d->kernel.bo = intel_winsys_alloc_buffer(hw3d->cp->winsys, - "kernel bo", new_size, INTEL_DOMAIN_CPU); + "kernel bo", new_size, true); if (!hw3d->kernel.bo) { ilo_err("failed to allocate kernel bo\n"); return false; diff --git a/src/gallium/drivers/ilo/ilo_3d_pipeline.c b/src/gallium/drivers/ilo/ilo_3d_pipeline.c index 1f18bbbedfe..1152a7ffc3e 100644 --- a/src/gallium/drivers/ilo/ilo_3d_pipeline.c +++ b/src/gallium/drivers/ilo/ilo_3d_pipeline.c @@ -95,7 +95,7 @@ ilo_3d_pipeline_create(struct ilo_cp *cp, const struct ilo_dev_info *dev) p->invalidate_flags = ILO_3D_PIPELINE_INVALIDATE_ALL; p->workaround_bo = intel_winsys_alloc_buffer(p->cp->winsys, - "PIPE_CONTROL workaround", 4096, INTEL_DOMAIN_INSTRUCTION); + "PIPE_CONTROL workaround", 4096, false); if (!p->workaround_bo) { ilo_warn("failed to allocate PIPE_CONTROL workaround bo\n"); FREE(p); diff --git a/src/gallium/drivers/ilo/ilo_blitter_blt.c b/src/gallium/drivers/ilo/ilo_blitter_blt.c index 1dd00af6c9c..5c2d2d49821 100644 --- a/src/gallium/drivers/ilo/ilo_blitter_blt.c +++ b/src/gallium/drivers/ilo/ilo_blitter_blt.c @@ -153,8 +153,7 @@ gen6_emit_COLOR_BLT(struct ilo_dev_info *dev, ilo_cp_write(cp, dw0); ilo_cp_write(cp, dw1); ilo_cp_write(cp, height << 16 | width); - ilo_cp_write_bo(cp, dst_offset, dst_bo, INTEL_DOMAIN_RENDER, - INTEL_DOMAIN_RENDER); + ilo_cp_write_bo(cp, dst_offset, dst_bo, INTEL_RELOC_WRITE); ilo_cp_write(cp, pattern); ilo_cp_end(cp); } @@ -204,8 +203,7 @@ gen6_emit_XY_COLOR_BLT(struct ilo_dev_info *dev, ilo_cp_write(cp, dw1); ilo_cp_write(cp, y1 << 16 | x1); ilo_cp_write(cp, y2 << 16 | x2); - ilo_cp_write_bo(cp, dst_offset, dst_bo, - INTEL_DOMAIN_RENDER, INTEL_DOMAIN_RENDER); + ilo_cp_write_bo(cp, dst_offset, dst_bo, INTEL_RELOC_WRITE); ilo_cp_write(cp, pattern); ilo_cp_end(cp); } @@ -247,10 +245,9 @@ gen6_emit_SRC_COPY_BLT(struct ilo_dev_info *dev, ilo_cp_write(cp, dw0); ilo_cp_write(cp, dw1); ilo_cp_write(cp, height << 16 | width); - ilo_cp_write_bo(cp, dst_offset, dst_bo, INTEL_DOMAIN_RENDER, - INTEL_DOMAIN_RENDER); + ilo_cp_write_bo(cp, dst_offset, dst_bo, INTEL_RELOC_WRITE); ilo_cp_write(cp, src_pitch); - ilo_cp_write_bo(cp, src_offset, src_bo, INTEL_DOMAIN_RENDER, 0); + ilo_cp_write_bo(cp, src_offset, src_bo, 0); ilo_cp_end(cp); } @@ -316,11 +313,10 @@ gen6_emit_XY_SRC_COPY_BLT(struct ilo_dev_info *dev, ilo_cp_write(cp, dw1); ilo_cp_write(cp, y1 << 16 | x1); ilo_cp_write(cp, y2 << 16 | x2); - ilo_cp_write_bo(cp, dst_offset, dst_bo, INTEL_DOMAIN_RENDER, - INTEL_DOMAIN_RENDER); + ilo_cp_write_bo(cp, dst_offset, dst_bo, INTEL_RELOC_WRITE); ilo_cp_write(cp, src_y << 16 | src_x); ilo_cp_write(cp, src_pitch >> src_pitch_shift); - ilo_cp_write_bo(cp, src_offset, src_bo, INTEL_DOMAIN_RENDER, 0); + ilo_cp_write_bo(cp, src_offset, src_bo, 0); ilo_cp_end(cp); } diff --git a/src/gallium/drivers/ilo/ilo_cp.c b/src/gallium/drivers/ilo/ilo_cp.c index ec99e0fb02e..0b2bbfacf4c 100644 --- a/src/gallium/drivers/ilo/ilo_cp.c +++ b/src/gallium/drivers/ilo/ilo_cp.c @@ -163,7 +163,7 @@ ilo_cp_realloc_bo(struct ilo_cp *cp) * won't point at the same address, which is needed for jmpbuf */ bo = intel_winsys_alloc_buffer(cp->winsys, - "batch buffer", cp->bo_size * 4, INTEL_DOMAIN_CPU); + "batch buffer", cp->bo_size * 4, true); if (unlikely(!bo)) { /* reuse the old one */ bo = cp->bo; diff --git a/src/gallium/drivers/ilo/ilo_cp.h b/src/gallium/drivers/ilo/ilo_cp.h index 15b66040f6e..b4963022b00 100644 --- a/src/gallium/drivers/ilo/ilo_cp.h +++ b/src/gallium/drivers/ilo/ilo_cp.h @@ -320,14 +320,14 @@ ilo_cp_write_multi(struct ilo_cp *cp, const void *vals, int num_vals) * bo to the buffer, it also emits a relocation. */ static inline void -ilo_cp_write_bo(struct ilo_cp *cp, uint32_t val, struct intel_bo *bo, - uint32_t read_domains, uint32_t write_domain) +ilo_cp_write_bo(struct ilo_cp *cp, uint32_t val, + struct intel_bo *bo, uint32_t flags) { uint64_t presumed_offset; if (bo) { - intel_bo_add_reloc(cp->bo, cp->cmd_cur * 4, bo, val, - read_domains, write_domain, &presumed_offset); + intel_bo_add_reloc(cp->bo, cp->cmd_cur * 4, bo, val, flags, + &presumed_offset); } else { presumed_offset = 0; diff --git a/src/gallium/drivers/ilo/ilo_gpe_gen6.h b/src/gallium/drivers/ilo/ilo_gpe_gen6.h index 8e394ded95f..efd928d2ed9 100644 --- a/src/gallium/drivers/ilo/ilo_gpe_gen6.h +++ b/src/gallium/drivers/ilo/ilo_gpe_gen6.h @@ -253,8 +253,6 @@ gen6_emit_MI_STORE_DATA_IMM(const struct ilo_dev_info *dev, const uint8_t cmd_len = (store_qword) ? 5 : 4; /* must use GGTT on GEN6 as in PIPE_CONTROL */ const uint32_t cmd_flags = (dev->gen == ILO_GEN(6)) ? (1 << 22) : 0; - const uint32_t read_domains = INTEL_DOMAIN_INSTRUCTION; - const uint32_t write_domain = INTEL_DOMAIN_INSTRUCTION; ILO_GPE_VALID_GEN(dev, 6, 7.5); @@ -263,7 +261,7 @@ gen6_emit_MI_STORE_DATA_IMM(const struct ilo_dev_info *dev, ilo_cp_begin(cp, cmd_len); ilo_cp_write(cp, cmd | cmd_flags | (cmd_len - 2)); ilo_cp_write(cp, 0); - ilo_cp_write_bo(cp, bo_offset, bo, read_domains, write_domain); + ilo_cp_write_bo(cp, bo_offset, bo, INTEL_RELOC_WRITE | INTEL_RELOC_GGTT); ilo_cp_write(cp, (uint32_t) val); if (store_qword) @@ -302,8 +300,6 @@ gen6_emit_MI_STORE_REGISTER_MEM(const struct ilo_dev_info *dev, const uint8_t cmd_len = 3; /* must use GGTT on GEN6 as in PIPE_CONTROL */ const uint32_t cmd_flags = (dev->gen == ILO_GEN(6)) ? (1 << 22) : 0; - const uint32_t read_domains = INTEL_DOMAIN_INSTRUCTION; - const uint32_t write_domain = INTEL_DOMAIN_INSTRUCTION; ILO_GPE_VALID_GEN(dev, 6, 7.5); @@ -312,7 +308,7 @@ gen6_emit_MI_STORE_REGISTER_MEM(const struct ilo_dev_info *dev, ilo_cp_begin(cp, cmd_len); ilo_cp_write(cp, cmd | cmd_flags | (cmd_len - 2)); ilo_cp_write(cp, reg); - ilo_cp_write_bo(cp, bo_offset, bo, read_domains, write_domain); + ilo_cp_write_bo(cp, bo_offset, bo, INTEL_RELOC_WRITE | INTEL_RELOC_GGTT); ilo_cp_end(cp); } @@ -323,8 +319,6 @@ gen6_emit_MI_REPORT_PERF_COUNT(const struct ilo_dev_info *dev, { const uint32_t cmd = ILO_GPE_MI(0x28); const uint8_t cmd_len = 3; - const uint32_t read_domains = INTEL_DOMAIN_INSTRUCTION; - const uint32_t write_domain = INTEL_DOMAIN_INSTRUCTION; ILO_GPE_VALID_GEN(dev, 6, 7.5); @@ -336,7 +330,7 @@ gen6_emit_MI_REPORT_PERF_COUNT(const struct ilo_dev_info *dev, ilo_cp_begin(cp, cmd_len); ilo_cp_write(cp, cmd | (cmd_len - 2)); - ilo_cp_write_bo(cp, bo_offset, bo, read_domains, write_domain); + ilo_cp_write_bo(cp, bo_offset, bo, INTEL_RELOC_WRITE | INTEL_RELOC_GGTT); ilo_cp_write(cp, report_id); ilo_cp_end(cp); } @@ -366,26 +360,14 @@ gen6_emit_STATE_BASE_ADDRESS(const struct ilo_dev_info *dev, ilo_cp_begin(cp, cmd_len); ilo_cp_write(cp, cmd | (cmd_len - 2)); - ilo_cp_write_bo(cp, 1, general_state_bo, - INTEL_DOMAIN_RENDER, - 0); - ilo_cp_write_bo(cp, 1, surface_state_bo, - INTEL_DOMAIN_SAMPLER, - 0); - ilo_cp_write_bo(cp, 1, dynamic_state_bo, - INTEL_DOMAIN_RENDER | INTEL_DOMAIN_INSTRUCTION, - 0); - ilo_cp_write_bo(cp, 1, indirect_object_bo, - 0, - 0); - ilo_cp_write_bo(cp, 1, instruction_bo, - INTEL_DOMAIN_INSTRUCTION, - 0); + ilo_cp_write_bo(cp, 1, general_state_bo, 0); + ilo_cp_write_bo(cp, 1, surface_state_bo, 0); + ilo_cp_write_bo(cp, 1, dynamic_state_bo, 0); + ilo_cp_write_bo(cp, 1, indirect_object_bo, 0); + ilo_cp_write_bo(cp, 1, instruction_bo, 0); if (general_state_size) { - ilo_cp_write_bo(cp, general_state_size | 1, general_state_bo, - INTEL_DOMAIN_RENDER, - 0); + ilo_cp_write_bo(cp, general_state_size | 1, general_state_bo, 0); } else { /* skip range check */ @@ -393,9 +375,7 @@ gen6_emit_STATE_BASE_ADDRESS(const struct ilo_dev_info *dev, } if (dynamic_state_size) { - ilo_cp_write_bo(cp, dynamic_state_size | 1, dynamic_state_bo, - INTEL_DOMAIN_RENDER | INTEL_DOMAIN_INSTRUCTION, - 0); + ilo_cp_write_bo(cp, dynamic_state_size | 1, dynamic_state_bo, 0); } else { /* skip range check */ @@ -403,9 +383,7 @@ gen6_emit_STATE_BASE_ADDRESS(const struct ilo_dev_info *dev, } if (indirect_object_size) { - ilo_cp_write_bo(cp, indirect_object_size | 1, indirect_object_bo, - 0, - 0); + ilo_cp_write_bo(cp, indirect_object_size | 1, indirect_object_bo, 0); } else { /* skip range check */ @@ -413,9 +391,7 @@ gen6_emit_STATE_BASE_ADDRESS(const struct ilo_dev_info *dev, } if (instruction_size) { - ilo_cp_write_bo(cp, instruction_size | 1, instruction_bo, - INTEL_DOMAIN_INSTRUCTION, - 0); + ilo_cp_write_bo(cp, instruction_size | 1, instruction_bo, 0); } else { /* skip range check */ @@ -741,8 +717,8 @@ gen6_emit_3DSTATE_VERTEX_BUFFERS(const struct ilo_dev_info *dev, dw |= cso->stride << GEN6_VB_STATE_DW0_PITCH__SHIFT; ilo_cp_write(cp, dw); - ilo_cp_write_bo(cp, start_offset, buf->bo, INTEL_DOMAIN_VERTEX, 0); - ilo_cp_write_bo(cp, end_offset, buf->bo, INTEL_DOMAIN_VERTEX, 0); + ilo_cp_write_bo(cp, start_offset, buf->bo, 0); + ilo_cp_write_bo(cp, end_offset, buf->bo, 0); ilo_cp_write(cp, instance_divisor); } else { @@ -951,8 +927,8 @@ gen6_emit_3DSTATE_INDEX_BUFFER(const struct ilo_dev_info *dev, ilo_cp_write(cp, cmd | (cmd_len - 2) | ((enable_cut_index) ? GEN6_IB_DW0_CUT_INDEX_ENABLE : 0) | format); - ilo_cp_write_bo(cp, start_offset, buf->bo, INTEL_DOMAIN_VERTEX, 0); - ilo_cp_write_bo(cp, end_offset, buf->bo, INTEL_DOMAIN_VERTEX, 0); + ilo_cp_write_bo(cp, start_offset, buf->bo, 0); + ilo_cp_write_bo(cp, end_offset, buf->bo, 0); ilo_cp_end(cp); } @@ -1527,8 +1503,7 @@ gen6_emit_3DSTATE_DEPTH_BUFFER(const struct ilo_dev_info *dev, ilo_cp_begin(cp, cmd_len); ilo_cp_write(cp, cmd | (cmd_len - 2)); ilo_cp_write(cp, zs->payload[0]); - ilo_cp_write_bo(cp, zs->payload[1], zs->bo, - INTEL_DOMAIN_RENDER, INTEL_DOMAIN_RENDER); + ilo_cp_write_bo(cp, zs->payload[1], zs->bo, INTEL_RELOC_WRITE); ilo_cp_write(cp, zs->payload[2]); ilo_cp_write(cp, zs->payload[3]); ilo_cp_write(cp, zs->payload[4]); @@ -1713,8 +1688,7 @@ gen6_emit_3DSTATE_STENCIL_BUFFER(const struct ilo_dev_info *dev, ilo_cp_write(cp, cmd | (cmd_len - 2)); /* see ilo_gpe_init_zs_surface() */ ilo_cp_write(cp, zs->payload[6]); - ilo_cp_write_bo(cp, zs->payload[7], zs->separate_s8_bo, - INTEL_DOMAIN_RENDER, INTEL_DOMAIN_RENDER); + ilo_cp_write_bo(cp, zs->payload[7], zs->separate_s8_bo, INTEL_RELOC_WRITE); ilo_cp_end(cp); } @@ -1734,8 +1708,7 @@ gen6_emit_3DSTATE_HIER_DEPTH_BUFFER(const struct ilo_dev_info *dev, ilo_cp_write(cp, cmd | (cmd_len - 2)); /* see ilo_gpe_init_zs_surface() */ ilo_cp_write(cp, zs->payload[8]); - ilo_cp_write_bo(cp, zs->payload[9], zs->hiz_bo, - INTEL_DOMAIN_RENDER, INTEL_DOMAIN_RENDER); + ilo_cp_write_bo(cp, zs->payload[9], zs->hiz_bo, INTEL_RELOC_WRITE); ilo_cp_end(cp); } @@ -1765,8 +1738,6 @@ gen6_emit_PIPE_CONTROL(const struct ilo_dev_info *dev, { const uint32_t cmd = ILO_GPE_CMD(0x3, 0x2, 0x00); const uint8_t cmd_len = (write_qword) ? 5 : 4; - const uint32_t read_domains = INTEL_DOMAIN_INSTRUCTION; - const uint32_t write_domain = INTEL_DOMAIN_INSTRUCTION; ILO_GPE_VALID_GEN(dev, 6, 7.5); @@ -1839,7 +1810,7 @@ gen6_emit_PIPE_CONTROL(const struct ilo_dev_info *dev, ilo_cp_begin(cp, cmd_len); ilo_cp_write(cp, cmd | (cmd_len - 2)); ilo_cp_write(cp, dw1); - ilo_cp_write_bo(cp, bo_offset, bo, read_domains, write_domain); + ilo_cp_write_bo(cp, bo_offset, bo, INTEL_RELOC_WRITE | INTEL_RELOC_GGTT); ilo_cp_write(cp, 0); if (write_qword) ilo_cp_write(cp, 0); @@ -2286,26 +2257,16 @@ gen6_emit_SURFACE_STATE(const struct ilo_dev_info *dev, const int state_align = 32 / 4; const int state_len = (dev->gen >= ILO_GEN(7)) ? 8 : 6; uint32_t state_offset; - uint32_t read_domains, write_domain; ILO_GPE_VALID_GEN(dev, 6, 7.5); - if (for_render) { - read_domains = INTEL_DOMAIN_RENDER; - write_domain = INTEL_DOMAIN_RENDER; - } - else { - read_domains = INTEL_DOMAIN_SAMPLER; - write_domain = 0; - } - ilo_cp_steal(cp, "SURFACE_STATE", state_len, state_align, &state_offset); STATIC_ASSERT(Elements(surf->payload) >= 8); ilo_cp_write(cp, surf->payload[0]); - ilo_cp_write_bo(cp, surf->payload[1], - surf->bo, read_domains, write_domain); + ilo_cp_write_bo(cp, surf->payload[1], surf->bo, + (for_render) ? INTEL_RELOC_WRITE : 0); ilo_cp_write(cp, surf->payload[2]); ilo_cp_write(cp, surf->payload[3]); ilo_cp_write(cp, surf->payload[4]); diff --git a/src/gallium/drivers/ilo/ilo_gpe_gen7.h b/src/gallium/drivers/ilo/ilo_gpe_gen7.h index 0816fd67477..bbba9cd6b01 100644 --- a/src/gallium/drivers/ilo/ilo_gpe_gen7.h +++ b/src/gallium/drivers/ilo/ilo_gpe_gen7.h @@ -1036,8 +1036,8 @@ gen7_emit_3DSTATE_SO_BUFFER(const struct ilo_dev_info *dev, ilo_cp_write(cp, cmd | (cmd_len - 2)); ilo_cp_write(cp, index << GEN7_SO_BUF_DW1_INDEX__SHIFT | stride); - ilo_cp_write_bo(cp, base, buf->bo, INTEL_DOMAIN_RENDER, INTEL_DOMAIN_RENDER); - ilo_cp_write_bo(cp, end, buf->bo, INTEL_DOMAIN_RENDER, INTEL_DOMAIN_RENDER); + ilo_cp_write_bo(cp, base, buf->bo, INTEL_RELOC_WRITE); + ilo_cp_write_bo(cp, end, buf->bo, INTEL_RELOC_WRITE); ilo_cp_end(cp); } diff --git a/src/gallium/drivers/ilo/ilo_query.c b/src/gallium/drivers/ilo/ilo_query.c index 46e6bf766e1..72fc1c70a70 100644 --- a/src/gallium/drivers/ilo/ilo_query.c +++ b/src/gallium/drivers/ilo/ilo_query.c @@ -229,7 +229,7 @@ ilo_query_alloc_bo(struct ilo_query *q, int reg_count, int repeat_count, intel_bo_unreference(q->bo); q->bo = intel_winsys_alloc_buffer(winsys, - name, size, INTEL_DOMAIN_INSTRUCTION); + name, size, false); q->reg_total = (q->bo) ? reg_total : 0; } diff --git a/src/gallium/drivers/ilo/ilo_resource.c b/src/gallium/drivers/ilo/ilo_resource.c index 54d6c368ae9..d312a4dffdd 100644 --- a/src/gallium/drivers/ilo/ilo_resource.c +++ b/src/gallium/drivers/ilo/ilo_resource.c @@ -76,13 +76,12 @@ resource_get_bo_name(const struct pipe_resource *templ) return name; } -static enum intel_domain_flag -resource_get_bo_initial_domain(const struct pipe_resource *templ) +static bool +resource_get_cpu_init(const struct pipe_resource *templ) { return (templ->bind & (PIPE_BIND_DEPTH_STENCIL | PIPE_BIND_RENDER_TARGET | - PIPE_BIND_STREAM_OUTPUT)) ? - INTEL_DOMAIN_RENDER : 0; + PIPE_BIND_STREAM_OUTPUT)) ? false : true; } static void @@ -152,11 +151,10 @@ tex_create_bo(struct ilo_texture *tex) { struct ilo_screen *is = ilo_screen(tex->base.screen); const char *name = resource_get_bo_name(&tex->base); - const enum intel_domain_flag initial_domain = - resource_get_bo_initial_domain(&tex->base); + const bool cpu_init = resource_get_cpu_init(&tex->base); tex->bo = intel_winsys_alloc_bo(is->winsys, name, tex->layout.tiling, - tex->layout.bo_stride, tex->layout.bo_height, initial_domain); + tex->layout.bo_stride, tex->layout.bo_height, cpu_init); return (tex->bo != NULL); } @@ -194,7 +192,7 @@ tex_create_hiz(struct ilo_texture *tex) tex->aux_bo = intel_winsys_alloc_bo(is->winsys, "hiz texture", INTEL_TILING_Y, tex->layout.aux_stride, tex->layout.aux_height, - INTEL_DOMAIN_RENDER); + false); if (!tex->aux_bo) return false; @@ -224,7 +222,7 @@ tex_create_mcs(struct ilo_texture *tex) tex->aux_bo = intel_winsys_alloc_bo(is->winsys, "mcs texture", INTEL_TILING_Y, tex->layout.aux_stride, tex->layout.aux_height, - INTEL_DOMAIN_RENDER); + false); if (!tex->aux_bo) return false; @@ -357,11 +355,10 @@ buf_create_bo(struct ilo_buffer *buf) { struct ilo_screen *is = ilo_screen(buf->base.screen); const char *name = resource_get_bo_name(&buf->base); - const enum intel_domain_flag initial_domain = - resource_get_bo_initial_domain(&buf->base); + const bool cpu_init = resource_get_cpu_init(&buf->base); buf->bo = intel_winsys_alloc_buffer(is->winsys, name, - buf->bo_size, initial_domain); + buf->bo_size, cpu_init); return (buf->bo != NULL); } diff --git a/src/gallium/winsys/intel/drm/intel_drm_winsys.c b/src/gallium/winsys/intel/drm/intel_drm_winsys.c index f94a1cb9f33..c8ae6c16a8c 100644 --- a/src/gallium/winsys/intel/drm/intel_drm_winsys.c +++ b/src/gallium/winsys/intel/drm/intel_drm_winsys.c @@ -195,12 +195,7 @@ intel_winsys_create_for_fd(int fd) /* * No need to implicitly set up a fence register for each non-linear reloc - * entry. When a fence register is needed for a reloc entry, - * drm_intel_bo_emit_reloc_fence() will be called explicitly. - * - * intel_bo_add_reloc() currently lacks "bool fenced" for this to work. - * But we never need a fence register on GEN4+ so we do not need to worry - * about it yet. + * entry. INTEL_RELOC_FENCE will be set on reloc entries that need them. */ drm_intel_bufmgr_gem_enable_fenced_relocs(winsys->bufmgr); @@ -266,10 +261,8 @@ intel_winsys_alloc_bo(struct intel_winsys *winsys, enum intel_tiling_mode tiling, unsigned long pitch, unsigned long height, - uint32_t initial_domain) + bool cpu_init) { - const bool for_render = - (initial_domain & (INTEL_DOMAIN_RENDER | INTEL_DOMAIN_INSTRUCTION)); const unsigned int alignment = 4096; /* always page-aligned */ unsigned long size; drm_intel_bo *bo; @@ -292,12 +285,12 @@ intel_winsys_alloc_bo(struct intel_winsys *winsys, size = pitch * height; - if (for_render) { - bo = drm_intel_bo_alloc_for_render(winsys->bufmgr, - name, size, alignment); + if (cpu_init) { + bo = drm_intel_bo_alloc(winsys->bufmgr, name, size, alignment); } else { - bo = drm_intel_bo_alloc(winsys->bufmgr, name, size, alignment); + bo = drm_intel_bo_alloc_for_render(winsys->bufmgr, + name, size, alignment); } if (bo && tiling != INTEL_TILING_NONE) { @@ -558,14 +551,37 @@ intel_bo_pread(struct intel_bo *bo, unsigned long offset, int intel_bo_add_reloc(struct intel_bo *bo, uint32_t offset, struct intel_bo *target_bo, uint32_t target_offset, - uint32_t read_domains, uint32_t write_domain, - uint64_t *presumed_offset) + uint32_t flags, uint64_t *presumed_offset) { + uint32_t read_domains, write_domain; int err; - err = drm_intel_bo_emit_reloc(gem_bo(bo), offset, - gem_bo(target_bo), target_offset, - read_domains, write_domain); + if (flags & INTEL_RELOC_WRITE) { + /* + * Because of the translation to domains, INTEL_RELOC_GGTT should only + * be set on GEN6 when the bo is written by MI_* or PIPE_CONTROL. The + * kernel will translate it back to INTEL_RELOC_GGTT. + */ + write_domain = (flags & INTEL_RELOC_GGTT) ? + I915_GEM_DOMAIN_INSTRUCTION : I915_GEM_DOMAIN_RENDER; + read_domains = write_domain; + } else { + write_domain = 0; + read_domains = I915_GEM_DOMAIN_RENDER | + I915_GEM_DOMAIN_SAMPLER | + I915_GEM_DOMAIN_INSTRUCTION | + I915_GEM_DOMAIN_VERTEX; + } + + if (flags & INTEL_RELOC_FENCE) { + err = drm_intel_bo_emit_reloc_fence(gem_bo(bo), offset, + gem_bo(target_bo), target_offset, + read_domains, write_domain); + } else { + err = drm_intel_bo_emit_reloc(gem_bo(bo), offset, + gem_bo(target_bo), target_offset, + read_domains, write_domain); + } *presumed_offset = gem_bo(target_bo)->offset64 + target_offset; diff --git a/src/gallium/winsys/intel/intel_winsys.h b/src/gallium/winsys/intel/intel_winsys.h index 679f24817b8..bc18bb69be5 100644 --- a/src/gallium/winsys/intel/intel_winsys.h +++ b/src/gallium/winsys/intel/intel_winsys.h @@ -44,14 +44,10 @@ enum intel_exec_flag { }; /* this is compatible with i915_drm.h's definitions */ -enum intel_domain_flag { - INTEL_DOMAIN_CPU = 0x00000001, - INTEL_DOMAIN_RENDER = 0x00000002, - INTEL_DOMAIN_SAMPLER = 0x00000004, - INTEL_DOMAIN_COMMAND = 0x00000008, - INTEL_DOMAIN_INSTRUCTION = 0x00000010, - INTEL_DOMAIN_VERTEX = 0x00000020, - INTEL_DOMAIN_GTT = 0x00000040, +enum intel_reloc_flag { + INTEL_RELOC_FENCE = 1 << 0, + INTEL_RELOC_GGTT = 1 << 1, + INTEL_RELOC_WRITE = 1 << 2, }; /* this is compatible with i915_drm.h's definitions */ @@ -126,7 +122,7 @@ intel_winsys_read_reg(struct intel_winsys *winsys, * \param tiling Tiling mode. * \param pitch Pitch of the bo. * \param height Height of the bo. - * \param initial_domain Initial (write) domain. + * \param cpu_init Will be initialized by CPU. */ struct intel_bo * intel_winsys_alloc_bo(struct intel_winsys *winsys, @@ -134,7 +130,7 @@ intel_winsys_alloc_bo(struct intel_winsys *winsys, enum intel_tiling_mode tiling, unsigned long pitch, unsigned long height, - uint32_t initial_domain); + bool cpu_init); /** * Allocate a linear buffer object. @@ -143,10 +139,10 @@ static inline struct intel_bo * intel_winsys_alloc_buffer(struct intel_winsys *winsys, const char *name, unsigned long size, - uint32_t initial_domain) + bool cpu_init) { return intel_winsys_alloc_bo(winsys, name, - INTEL_TILING_NONE, size, 1, initial_domain); + INTEL_TILING_NONE, size, 1, cpu_init); } /** @@ -275,8 +271,7 @@ intel_bo_pread(struct intel_bo *bo, unsigned long offset, int intel_bo_add_reloc(struct intel_bo *bo, uint32_t offset, struct intel_bo *target_bo, uint32_t target_offset, - uint32_t read_domains, uint32_t write_domain, - uint64_t *presumed_offset); + uint32_t flags, uint64_t *presumed_offset); /** * Return the current number of relocations.