diff options
Diffstat (limited to 'source/blender/gpu/metal')
-rw-r--r-- | source/blender/gpu/metal/mtl_backend.hh | 5 | ||||
-rw-r--r-- | source/blender/gpu/metal/mtl_backend.mm | 8 | ||||
-rw-r--r-- | source/blender/gpu/metal/mtl_command_buffer.mm | 6 | ||||
-rw-r--r-- | source/blender/gpu/metal/mtl_context.hh | 24 | ||||
-rw-r--r-- | source/blender/gpu/metal/mtl_context.mm | 42 | ||||
-rw-r--r-- | source/blender/gpu/metal/mtl_memory.hh | 18 | ||||
-rw-r--r-- | source/blender/gpu/metal/mtl_memory.mm | 48 | ||||
-rw-r--r-- | source/blender/gpu/metal/mtl_query.hh | 41 | ||||
-rw-r--r-- | source/blender/gpu/metal/mtl_query.mm | 122 | ||||
-rw-r--r-- | source/blender/gpu/metal/mtl_texture.hh | 2 | ||||
-rw-r--r-- | source/blender/gpu/metal/mtl_uniform_buffer.hh | 50 | ||||
-rw-r--r-- | source/blender/gpu/metal/mtl_uniform_buffer.mm | 162 |
12 files changed, 498 insertions, 30 deletions
diff --git a/source/blender/gpu/metal/mtl_backend.hh b/source/blender/gpu/metal/mtl_backend.hh index 7228a5f7596..3e09408e43e 100644 --- a/source/blender/gpu/metal/mtl_backend.hh +++ b/source/blender/gpu/metal/mtl_backend.hh @@ -40,6 +40,11 @@ class MTLBackend : public GPUBackend { MTLBackend::platform_exit(); } + void delete_resources() + { + /* Delete any resources with context active. */ + } + static bool metal_is_supported(); static MTLBackend *get() { diff --git a/source/blender/gpu/metal/mtl_backend.mm b/source/blender/gpu/metal/mtl_backend.mm index d2524cc16e3..3328855bbf8 100644 --- a/source/blender/gpu/metal/mtl_backend.mm +++ b/source/blender/gpu/metal/mtl_backend.mm @@ -10,6 +10,8 @@ #include "mtl_backend.hh" #include "mtl_context.hh" #include "mtl_framebuffer.hh" +#include "mtl_uniform_buffer.hh" +#include "mtl_query.hh" #include "gpu_capabilities_private.hh" #include "gpu_platform_private.hh" @@ -64,8 +66,7 @@ IndexBuf *MTLBackend::indexbuf_alloc() QueryPool *MTLBackend::querypool_alloc() { - /* TODO(Metal): Implement MTLQueryPool. */ - return nullptr; + return new MTLQueryPool(); }; Shader *MTLBackend::shader_alloc(const char *name) @@ -81,8 +82,7 @@ Texture *MTLBackend::texture_alloc(const char *name) UniformBuf *MTLBackend::uniformbuf_alloc(int size, const char *name) { - /* TODO(Metal): Implement MTLUniformBuf. */ - return nullptr; + return new MTLUniformBuf(size, name); }; StorageBuf *MTLBackend::storagebuf_alloc(int size, GPUUsageType usage, const char *name) diff --git a/source/blender/gpu/metal/mtl_command_buffer.mm b/source/blender/gpu/metal/mtl_command_buffer.mm index 7b6066b3d86..9a9a2d55103 100644 --- a/source/blender/gpu/metal/mtl_command_buffer.mm +++ b/source/blender/gpu/metal/mtl_command_buffer.mm @@ -308,6 +308,12 @@ id<MTLRenderCommandEncoder> MTLCommandBufferManager::ensure_begin_render_command active_pass_descriptor_ = active_frame_buffer_->bake_render_pass_descriptor( is_rebind && (!active_frame_buffer_->get_pending_clear())); + /* Determine if there is a visibility buffer assigned to the context. */ + gpu::MTLBuffer *visibility_buffer = context_.get_visibility_buffer(); + this->active_pass_descriptor_.visibilityResultBuffer = + (visibility_buffer) ? visibility_buffer->get_metal_buffer() : nil; + context_.clear_visibility_dirty(); + /* Ensure we have already cleaned up our previous render command encoder. */ BLI_assert(active_render_command_encoder_ == nil); diff --git a/source/blender/gpu/metal/mtl_context.hh b/source/blender/gpu/metal/mtl_context.hh index 4b87b994a3d..0db87bf5da5 100644 --- a/source/blender/gpu/metal/mtl_context.hh +++ b/source/blender/gpu/metal/mtl_context.hh @@ -3,6 +3,9 @@ /** \file * \ingroup gpu */ + +#pragma once + #include "MEM_guardedalloc.h" #include "gpu_context_private.hh" @@ -588,6 +591,10 @@ class MTLContext : public Context { bool is_inside_frame_ = false; uint current_frame_index_; + /* Visibility buffer for MTLQuery results. */ + gpu::MTLBuffer *visibility_buffer_ = nullptr; + bool visibility_is_dirty_ = false; + public: /* Shaders and Pipeline state. */ MTLContextGlobalShaderPipelineState pipeline_state; @@ -619,6 +626,11 @@ class MTLContext : public Context { void memory_statistics_get(int *total_mem, int *free_mem) override; + static MTLContext *get() + { + return static_cast<MTLContext *>(Context::get()); + } + void debug_group_begin(const char *name, int index) override; void debug_group_end() override; @@ -660,6 +672,18 @@ class MTLContext : public Context { void set_scissor(int scissor_x, int scissor_y, int scissor_width, int scissor_height); void set_scissor_enabled(bool scissor_enabled); + /* Visibility buffer control. */ + void set_visibility_buffer(gpu::MTLBuffer *buffer); + gpu::MTLBuffer *get_visibility_buffer() const; + + /* Flag whether the visibility buffer for query results + * has changed. This requires a new RenderPass in order + * to update.*/ + bool is_visibility_dirty() const; + + /* Reset dirty flag state for visibility buffer. */ + void clear_visibility_dirty(); + /* Texture utilities. */ MTLContextTextureUtils &get_texture_utils() { diff --git a/source/blender/gpu/metal/mtl_context.mm b/source/blender/gpu/metal/mtl_context.mm index 2ab2e20158a..26cfe6632ef 100644 --- a/source/blender/gpu/metal/mtl_context.mm +++ b/source/blender/gpu/metal/mtl_context.mm @@ -188,7 +188,8 @@ id<MTLRenderCommandEncoder> MTLContext::ensure_begin_render_pass() * framebuffer state has been modified (is_dirty). */ if (!this->main_command_buffer.is_inside_render_pass() || this->active_fb != this->main_command_buffer.get_active_framebuffer() || - this->main_command_buffer.get_active_framebuffer()->get_dirty()) { + this->main_command_buffer.get_active_framebuffer()->get_dirty() || + this->is_visibility_dirty()) { /* Validate bound framebuffer before beginning render pass. */ if (!static_cast<MTLFrameBuffer *>(this->active_fb)->validate_render_pass()) { @@ -372,6 +373,45 @@ void MTLContext::set_scissor_enabled(bool scissor_enabled) /** \} */ /* -------------------------------------------------------------------- */ +/** \name Visibility buffer control for MTLQueryPool. + * \{ */ + +void MTLContext::set_visibility_buffer(gpu::MTLBuffer *buffer) +{ + /* Flag visibility buffer as dirty if the buffer being used for visibility has changed -- + * This is required by the render pass, and we will break the pass if the results destination + * buffer is modified. */ + if (buffer) { + visibility_is_dirty_ = (buffer != visibility_buffer_) || visibility_is_dirty_; + visibility_buffer_ = buffer; + visibility_buffer_->debug_ensure_used(); + } + else { + /* If buffer is null, reset visibility state, mark dirty to break render pass if results are no + * longer needed. */ + visibility_is_dirty_ = (visibility_buffer_ != nullptr) || visibility_is_dirty_; + visibility_buffer_ = nullptr; + } +} + +gpu::MTLBuffer *MTLContext::get_visibility_buffer() const +{ + return visibility_buffer_; +} + +void MTLContext::clear_visibility_dirty() +{ + visibility_is_dirty_ = false; +} + +bool MTLContext::is_visibility_dirty() const +{ + return visibility_is_dirty_; +} + +/** \} */ + +/* -------------------------------------------------------------------- */ /** \name Texture State Management * \{ */ diff --git a/source/blender/gpu/metal/mtl_memory.hh b/source/blender/gpu/metal/mtl_memory.hh index daa049e78af..dc5417dc11a 100644 --- a/source/blender/gpu/metal/mtl_memory.hh +++ b/source/blender/gpu/metal/mtl_memory.hh @@ -78,8 +78,10 @@ * Usage: * MTLContext::get_global_memory_manager(); - static routine to fetch global memory manager. * - * gpu::MTLBuffer *allocate_buffer(size, is_cpu_visibile, bytes=nullptr) - * gpu::MTLBuffer *allocate_buffer_aligned(size, alignment, is_cpu_visibile, bytes=nullptr) + * gpu::MTLBuffer *allocate(size, is_cpu_visibile) + * gpu::MTLBuffer *allocate_aligned(size, alignment, is_cpu_visibile) + * gpu::MTLBuffer *allocate_with_data(size, is_cpu_visibile, data_ptr) + * gpu::MTLBuffer *allocate_aligned_with_data(size, alignment, is_cpu_visibile, data_ptr) */ /* Debug memory statistics: Disabled by Macro rather than guarded for @@ -389,11 +391,13 @@ class MTLBufferPool { void init(id<MTLDevice> device); ~MTLBufferPool(); - gpu::MTLBuffer *allocate_buffer(uint64_t size, bool cpu_visible, const void *bytes = nullptr); - gpu::MTLBuffer *allocate_buffer_aligned(uint64_t size, - uint alignment, - bool cpu_visible, - const void *bytes = nullptr); + gpu::MTLBuffer *allocate(uint64_t size, bool cpu_visible); + gpu::MTLBuffer *allocate_aligned(uint64_t size, uint alignment, bool cpu_visible); + gpu::MTLBuffer *allocate_with_data(uint64_t size, bool cpu_visible, const void *data = nullptr); + gpu::MTLBuffer *allocate_aligned_with_data(uint64_t size, + uint alignment, + bool cpu_visible, + const void *data = nullptr); bool free_buffer(gpu::MTLBuffer *buffer); /* Flush MTLSafeFreeList buffers, for completed lists in `completed_safelist_queue_`, diff --git a/source/blender/gpu/metal/mtl_memory.mm b/source/blender/gpu/metal/mtl_memory.mm index e5db32ed1b1..48e27dd2bb6 100644 --- a/source/blender/gpu/metal/mtl_memory.mm +++ b/source/blender/gpu/metal/mtl_memory.mm @@ -57,17 +57,23 @@ void MTLBufferPool::free() buffer_pools_.clear(); } -gpu::MTLBuffer *MTLBufferPool::allocate_buffer(uint64_t size, bool cpu_visible, const void *bytes) +gpu::MTLBuffer *MTLBufferPool::allocate(uint64_t size, bool cpu_visible) { /* Allocate buffer with default HW-compatible alignment of 256 bytes. * See https://developer.apple.com/metal/Metal-Feature-Set-Tables.pdf for more. */ - return this->allocate_buffer_aligned(size, 256, cpu_visible, bytes); + return this->allocate_aligned(size, 256, cpu_visible); } -gpu::MTLBuffer *MTLBufferPool::allocate_buffer_aligned(uint64_t size, - uint alignment, - bool cpu_visible, - const void *bytes) +gpu::MTLBuffer *MTLBufferPool::allocate_with_data(uint64_t size, + bool cpu_visible, + const void *data) +{ + /* Allocate buffer with default HW-compatible alignemnt of 256 bytes. + * See https://developer.apple.com/metal/Metal-Feature-Set-Tables.pdf for more. */ + return this->allocate_aligned_with_data(size, 256, cpu_visible, data); +} + +gpu::MTLBuffer *MTLBufferPool::allocate_aligned(uint64_t size, uint alignment, bool cpu_visible) { /* Check not required. Main GPU module usage considered thread-safe. */ // BLI_assert(BLI_thread_is_main()); @@ -153,15 +159,6 @@ gpu::MTLBuffer *MTLBufferPool::allocate_buffer_aligned(uint64_t size, /* Flag buffer as actively in-use. */ new_buffer->flag_in_use(true); - /* Upload initial data if provided -- Size based on original size param, not aligned size*/ - if (bytes) { - BLI_assert(!(options & MTLResourceStorageModePrivate)); - BLI_assert(size <= aligned_alloc_size); - BLI_assert(size <= [new_buffer->get_metal_buffer() length]); - memcpy(new_buffer->get_host_ptr(), bytes, size); - new_buffer->flush_range(0, size); - } - #if MTL_DEBUG_MEMORY_STATISTICS == 1 this->per_frame_allocation_count++; #endif @@ -169,6 +166,23 @@ gpu::MTLBuffer *MTLBufferPool::allocate_buffer_aligned(uint64_t size, return new_buffer; } +gpu::MTLBuffer *MTLBufferPool::allocate_aligned_with_data(uint64_t size, + uint alignment, + bool cpu_visible, + const void *data) +{ + gpu::MTLBuffer *buf = this->allocate_aligned(size, 256, cpu_visible); + + /* Upload initial data. */ + BLI_assert(data != nullptr); + BLI_assert(!(buf->get_resource_options() & MTLResourceStorageModePrivate)); + BLI_assert(size <= buf->get_size()); + BLI_assert(size <= [buf->get_metal_buffer() length]); + memcpy(buf->get_host_ptr(), data, size); + buf->flush_range(0, size); + return buf; +} + bool MTLBufferPool::free_buffer(gpu::MTLBuffer *buffer) { /* Ensure buffer is flagged as in-use. I.e. has not already been returned to memory pools. */ @@ -356,7 +370,7 @@ void MTLBufferPool::insert_buffer_into_pool(MTLResourceOptions options, gpu::MTL #if MTL_DEBUG_MEMORY_STATISTICS == 1 /* Debug statistics. */ - allocations_in_pool_ += buffer->size; + allocations_in_pool_ += buffer->get_size(); buffers_in_pool_++; #endif } @@ -413,7 +427,7 @@ void MTLSafeFreeList::decrement_reference() { lock_.lock(); BLI_assert(in_free_queue_ == false); - int ref_count = reference_count_--; + int ref_count = --reference_count_; if (ref_count == 0) { MTLContext::get_global_memory_manager().push_completed_safe_list(this); diff --git a/source/blender/gpu/metal/mtl_query.hh b/source/blender/gpu/metal/mtl_query.hh new file mode 100644 index 00000000000..c1ec9a2a0f5 --- /dev/null +++ b/source/blender/gpu/metal/mtl_query.hh @@ -0,0 +1,41 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ + +/** \file + * \ingroup gpu + */ + +#pragma once + +#include "BLI_vector.hh" + +#include "gpu_query.hh" +#include "mtl_context.hh" + +namespace blender::gpu { + +class MTLQueryPool : public QueryPool { + private: + /** Number of queries that have been issued since last initialization. + * Should be equal to query_ids_.size(). */ + uint32_t query_issued_; + /** Type of this query pool. */ + GPUQueryType type_; + /** Can only be initialized once. */ + bool initialized_ = false; + MTLVisibilityResultMode mtl_type_; + Vector<gpu::MTLBuffer *> buffer_; + + void allocate_buffer(); + + public: + MTLQueryPool(); + ~MTLQueryPool(); + + void init(GPUQueryType type) override; + + void begin_query() override; + void end_query() override; + + void get_occlusion_result(MutableSpan<uint32_t> r_values) override; +}; +} // namespace blender::gpu diff --git a/source/blender/gpu/metal/mtl_query.mm b/source/blender/gpu/metal/mtl_query.mm new file mode 100644 index 00000000000..33ae8f554c3 --- /dev/null +++ b/source/blender/gpu/metal/mtl_query.mm @@ -0,0 +1,122 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ + +/** \file + * \ingroup gpu + */ + +#include "mtl_query.hh" + +namespace blender::gpu { + +static const size_t VISIBILITY_COUNT_PER_BUFFER = 512; +/* defined in the documentation but not queryable programmatically: + * https://developer.apple.com/documentation/metal/mtlvisibilityresultmode/mtlvisibilityresultmodeboolean?language=objc + */ +static const size_t VISIBILITY_RESULT_SIZE_IN_BYTES = 8; + +MTLQueryPool::MTLQueryPool() +{ + allocate_buffer(); +} +MTLQueryPool::~MTLQueryPool() +{ + for (gpu::MTLBuffer *buf : buffer_) { + BLI_assert(buf); + buf->free(); + } +} + +void MTLQueryPool::allocate_buffer() +{ + /* Allocate Metal buffer for visibility results. */ + size_t buffer_size_in_bytes = VISIBILITY_COUNT_PER_BUFFER * VISIBILITY_RESULT_SIZE_IN_BYTES; + gpu::MTLBuffer *buffer = MTLContext::get_global_memory_manager().allocate_buffer( + buffer_size_in_bytes, true); + BLI_assert(buffer); + buffer_.append(buffer); +} + +static inline MTLVisibilityResultMode to_mtl_type(GPUQueryType type) +{ + if (type == GPU_QUERY_OCCLUSION) { + return MTLVisibilityResultModeBoolean; + } + BLI_assert(0); + return MTLVisibilityResultModeBoolean; +} + +void MTLQueryPool::init(GPUQueryType type) +{ + BLI_assert(initialized_ == false); + initialized_ = true; + type_ = type; + mtl_type_ = to_mtl_type(type); + query_issued_ = 0; +} + +void MTLQueryPool::begin_query() +{ + MTLContext *ctx = reinterpret_cast<MTLContext *>(GPU_context_active_get()); + + /* Ensure our allocated buffer pool has enough space for the current queries. */ + int query_id = query_issued_; + int requested_buffer = query_id / VISIBILITY_COUNT_PER_BUFFER; + if (requested_buffer >= buffer_.size()) { + allocate_buffer(); + } + + BLI_assert(requested_buffer < buffer_.size()); + gpu::MTLBuffer *buffer = buffer_[requested_buffer]; + + /* Ensure visibility buffer is set on the context. If visibility buffer changes, + * we need to begin a new render pass with an updated reference in the + * MTLRenderPassDescriptor. */ + ctx->set_visibility_buffer(buffer); + + ctx->ensure_begin_render_pass(); + id<MTLRenderCommandEncoder> rec = ctx->main_command_buffer.get_active_render_command_encoder(); + [rec setVisibilityResultMode:mtl_type_ + offset:(query_id % VISIBILITY_COUNT_PER_BUFFER) * + VISIBILITY_RESULT_SIZE_IN_BYTES]; + query_issued_ += 1; +} + +void MTLQueryPool::end_query() +{ + MTLContext *ctx = reinterpret_cast<MTLContext *>(GPU_context_active_get()); + + id<MTLRenderCommandEncoder> rec = ctx->main_command_buffer.get_active_render_command_encoder(); + [rec setVisibilityResultMode:MTLVisibilityResultModeDisabled offset:0]; +} + +void MTLQueryPool::get_occlusion_result(MutableSpan<uint32_t> r_values) +{ + MTLContext *ctx = reinterpret_cast<MTLContext *>(GPU_context_active_get()); + + /* Create a blit encoder to synchronize the query buffer results between + * GPU and CPU when not using shared-memory. */ + if ([ctx->device hasUnifiedMemory] == false) { + id<MTLBlitCommandEncoder> blit_encoder = ctx->main_command_buffer.ensure_begin_blit_encoder(); + BLI_assert(blit_encoder); + for (gpu::MTLBuffer *buf : buffer_) { + [blit_encoder synchronizeResource:buf->get_metal_buffer()]; + } + BLI_assert(ctx->get_inside_frame()); + } + + /* Wait for GPU operatiosn to complete and for query buffer contents + * to be synchronised back to host memory. */ + GPU_finish(); + + /* Iterate through all possible visibility buffers and copy results into provided + * container. */ + for (const int i : IndexRange(query_issued_)) { + int requested_buffer = i / VISIBILITY_COUNT_PER_BUFFER; + const uint64_t *queries = static_cast<const uint64_t *>( + buffer_[requested_buffer]->get_host_ptr()); + r_values[i] = static_cast<uint32_t>(queries[i % VISIBILITY_COUNT_PER_BUFFER]); + } + ctx->set_visibility_buffer(nullptr); +} + +} // namespace blender::gpu diff --git a/source/blender/gpu/metal/mtl_texture.hh b/source/blender/gpu/metal/mtl_texture.hh index 9387d5af814..82a7a20a310 100644 --- a/source/blender/gpu/metal/mtl_texture.hh +++ b/source/blender/gpu/metal/mtl_texture.hh @@ -93,7 +93,7 @@ struct TextureReadRoutineSpecialisation { * 0 = Not a Depth format, * 1 = FLOAT DEPTH, * 2 = 24Bit Integer Depth, - * 4 = 32bit uinteger Depth. */ + * 4 = 32bit Unsigned-Integer Depth. */ int depth_format_mode; bool operator==(const TextureReadRoutineSpecialisation &other) const diff --git a/source/blender/gpu/metal/mtl_uniform_buffer.hh b/source/blender/gpu/metal/mtl_uniform_buffer.hh new file mode 100644 index 00000000000..722d819cf96 --- /dev/null +++ b/source/blender/gpu/metal/mtl_uniform_buffer.hh @@ -0,0 +1,50 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ + +/** \file + * \ingroup gpu + */ + +#pragma once + +#include "MEM_guardedalloc.h" +#include "gpu_uniform_buffer_private.hh" + +#include "mtl_context.hh" + +namespace blender::gpu { + +/** + * Implementation of Uniform Buffers using Metal. + **/ +class MTLUniformBuf : public UniformBuf { + private: + /* Allocation Handle. */ + gpu::MTLBuffer *metal_buffer_ = nullptr; + + /* Whether buffer has contents, if false, no GPU buffer will + * have yet been allocated. */ + bool has_data_ = false; + + /* Bindstate tracking. */ + int bind_slot_ = -1; + MTLContext *bound_ctx_ = nullptr; + + public: + MTLUniformBuf(size_t size, const char *name); + ~MTLUniformBuf(); + + void update(const void *data) override; + void bind(int slot) override; + void unbind() override; + + id<MTLBuffer> get_metal_buffer(int *r_offset); + int get_size(); + const char *get_name() + { + return name_; + } + + MEM_CXX_CLASS_ALLOC_FUNCS("MTLUniformBuf"); +}; + +} // namespace blender::gpu diff --git a/source/blender/gpu/metal/mtl_uniform_buffer.mm b/source/blender/gpu/metal/mtl_uniform_buffer.mm new file mode 100644 index 00000000000..3415c41f2cc --- /dev/null +++ b/source/blender/gpu/metal/mtl_uniform_buffer.mm @@ -0,0 +1,162 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ + +/** \file + * \ingroup gpu + */ + +#include "BKE_global.h" + +#include "BLI_string.h" + +#include "gpu_backend.hh" +#include "gpu_context_private.hh" + +#include "mtl_backend.hh" +#include "mtl_context.hh" +#include "mtl_debug.hh" +#include "mtl_uniform_buffer.hh" + +namespace blender::gpu { + +MTLUniformBuf::MTLUniformBuf(size_t size, const char *name) : UniformBuf(size, name) +{ +} + +MTLUniformBuf::~MTLUniformBuf() +{ + if (metal_buffer_ != nullptr) { + metal_buffer_->free(); + metal_buffer_ = nullptr; + } + has_data_ = false; + + /* Ensure UBO is not bound to active CTX. + * UBO bindings are reset upon Context-switch so we do not need + * to check deactivated context's. */ + MTLContext *ctx = MTLContext::get(); + if (ctx) { + for (int i = 0; i < MTL_MAX_UNIFORM_BUFFER_BINDINGS; i++) { + MTLUniformBufferBinding &slot = ctx->pipeline_state.ubo_bindings[i]; + if (slot.bound && slot.ubo == this) { + slot.bound = false; + slot.ubo = nullptr; + } + } + } +} + +void MTLUniformBuf::update(const void *data) +{ + BLI_assert(this); + BLI_assert(size_in_bytes_ > 0); + + /* Free existing allocation. + * The previous UBO resource will be tracked by the memory manager, + * in case dependent GPU work is still executing. */ + if (metal_buffer_ != nullptr) { + metal_buffer_->free(); + metal_buffer_ = nullptr; + } + + /* Allocate MTL buffer */ + MTLContext *ctx = static_cast<MTLContext *>(unwrap(GPU_context_active_get())); + BLI_assert(ctx); + BLI_assert(ctx->device); + UNUSED_VARS_NDEBUG(ctx); + + if (data != nullptr) { + metal_buffer_ = MTLContext::get_global_memory_manager().allocate_with_data( + size_in_bytes_, true, data); + has_data_ = true; + + metal_buffer_->set_label(@"Uniform Buffer"); + BLI_assert(metal_buffer_ != nullptr); + BLI_assert(metal_buffer_->get_metal_buffer() != nil); + } + else { + /* If data is not yet present, no buffer will be allocated and MTLContext will use an empty + * null buffer, containing zeroes, if the UBO is bound. */ + metal_buffer_ = nullptr; + has_data_ = false; + } +} + +void MTLUniformBuf::bind(int slot) +{ + if (slot < 0) { + MTL_LOG_WARNING("Failed to bind UBO %p. uniform location %d invalid.\n", this, slot); + return; + } + + BLI_assert(slot < MTL_MAX_UNIFORM_BUFFER_BINDINGS); + + /* Bind current UBO to active context. */ + MTLContext *ctx = MTLContext::get(); + BLI_assert(ctx); + + MTLUniformBufferBinding &ctx_ubo_bind_slot = ctx->pipeline_state.ubo_bindings[slot]; + ctx_ubo_bind_slot.ubo = this; + ctx_ubo_bind_slot.bound = true; + + bind_slot_ = slot; + bound_ctx_ = ctx; + + /* Check if we have any deferred data to upload. */ + if (data_ != nullptr) { + this->update(data_); + MEM_SAFE_FREE(data_); + } + + /* Ensure there is atleast an empty dummy buffer. */ + if (metal_buffer_ == nullptr) { + this->update(nullptr); + } +} + +void MTLUniformBuf::unbind() +{ + /* Unbind in debug mode to validate missing binds. + * Otherwise, only perform a full unbind upon destruction + * to ensure no lingering references. */ +#ifndef NDEBUG + if (true) { +#else + if (G.debug & G_DEBUG_GPU) { +#endif + if (bound_ctx_ != nullptr && bind_slot_ > -1) { + MTLUniformBufferBinding &ctx_ubo_bind_slot = + bound_ctx_->pipeline_state.ubo_bindings[bind_slot_]; + if (ctx_ubo_bind_slot.bound && ctx_ubo_bind_slot.ubo == this) { + ctx_ubo_bind_slot.bound = false; + ctx_ubo_bind_slot.ubo = nullptr; + } + } + } + + /* Reset bind index. */ + bind_slot_ = -1; + bound_ctx_ = nullptr; +} + +id<MTLBuffer> MTLUniformBuf::get_metal_buffer(int *r_offset) +{ + BLI_assert(this); + *r_offset = 0; + if (metal_buffer_ != nullptr && has_data_) { + *r_offset = 0; + metal_buffer_->debug_ensure_used(); + return metal_buffer_->get_metal_buffer(); + } + else { + *r_offset = 0; + return nil; + } +} + +int MTLUniformBuf::get_size() +{ + BLI_assert(this); + return size_in_bytes_; +} + +} // blender::gpu |