// Compute kernels that translate `DrawIndirectArgs`/`DrawIndexedIndirectArgs` // sequences into Metal indirect-command-buffer (ICB) render commands on the // GPU, so `multi_draw_(indexed_)indirect` never round-trips draw arguments // through the CPU. // // The Rust side (`wgpu-hal/src/metal/command.rs`) binds the ICB through an // argument buffer (`WgpuIcbArguments`) and dispatches one thread per potential // draw. Draws with a zero-sized dimension are `reset()` so the slot becomes a // no-op, matching WebGPU's treatment of empty draws. #include #include using namespace metal; struct WgpuIcbArguments { command_buffer icb [[id(0)]]; }; // Must match the layout of `wgpu_types::DrawIndirectArgs`. struct WgpuDrawIndirectArgs { uint vertex_count; uint instance_count; uint first_vertex; uint first_instance; }; // Must match the layout of `wgpu_types::DrawIndexedIndirectArgs`. struct WgpuDrawIndexedIndirectArgs { uint index_count; uint instance_count; uint first_index; int base_vertex; uint first_instance; }; // Must match the `ICB_PRIMITIVE_*` constants in // `wgpu-hal/src/metal/command.rs`. enum WgpuIcbPrimitiveType : uint { WGPU_ICB_PRIMITIVE_POINT = 0, WGPU_ICB_PRIMITIVE_LINE = 1, WGPU_ICB_PRIMITIVE_LINE_STRIP = 2, WGPU_ICB_PRIMITIVE_TRIANGLE = 3, WGPU_ICB_PRIMITIVE_TRIANGLE_STRIP = 4, }; static primitive_type wgpu_icb_primitive_type(uint value) { switch (value) { case WGPU_ICB_PRIMITIVE_POINT: return primitive_type::point; case WGPU_ICB_PRIMITIVE_LINE: return primitive_type::line; case WGPU_ICB_PRIMITIVE_LINE_STRIP: return primitive_type::line_strip; case WGPU_ICB_PRIMITIVE_TRIANGLE: return primitive_type::triangle; case WGPU_ICB_PRIMITIVE_TRIANGLE_STRIP: return primitive_type::triangle_strip; default: return primitive_type::triangle; } } kernel void wgpu_generate_mdi_icb( device WgpuIcbArguments& icb_args [[buffer(0)]], const device WgpuDrawIndirectArgs* draw_args [[buffer(1)]], constant uint& primitive_type_value [[buffer(2)]], constant uint& draw_count [[buffer(3)]], uint tid [[thread_position_in_grid]]) { if (tid >= draw_count) { return; } const WgpuDrawIndirectArgs args = draw_args[tid]; render_command cmd(icb_args.icb, tid); if (args.vertex_count == 0 || args.instance_count == 0) { cmd.reset(); return; } const primitive_type primitive = wgpu_icb_primitive_type(primitive_type_value); cmd.draw_primitives( primitive, args.first_vertex, args.vertex_count, args.instance_count, args.first_instance); } kernel void wgpu_generate_indexed_mdi_icb_u16( device WgpuIcbArguments& icb_args [[buffer(0)]], const device WgpuDrawIndexedIndirectArgs* draw_args [[buffer(1)]], const device ushort* index_buffer [[buffer(2)]], constant uint& primitive_type_value [[buffer(3)]], constant uint& draw_count [[buffer(4)]], uint tid [[thread_position_in_grid]]) { if (tid >= draw_count) { return; } const WgpuDrawIndexedIndirectArgs args = draw_args[tid]; render_command cmd(icb_args.icb, tid); if (args.index_count == 0 || args.instance_count == 0) { cmd.reset(); return; } const primitive_type primitive = wgpu_icb_primitive_type(primitive_type_value); cmd.draw_indexed_primitives( primitive, args.index_count, index_buffer + args.first_index, args.instance_count, args.base_vertex, args.first_instance); } kernel void wgpu_generate_indexed_mdi_icb_u32( device WgpuIcbArguments& icb_args [[buffer(0)]], const device WgpuDrawIndexedIndirectArgs* draw_args [[buffer(1)]], const device uint* index_buffer [[buffer(2)]], constant uint& primitive_type_value [[buffer(3)]], constant uint& draw_count [[buffer(4)]], uint tid [[thread_position_in_grid]]) { if (tid >= draw_count) { return; } const WgpuDrawIndexedIndirectArgs args = draw_args[tid]; render_command cmd(icb_args.icb, tid); if (args.index_count == 0 || args.instance_count == 0) { cmd.reset(); return; } const primitive_type primitive = wgpu_icb_primitive_type(primitive_type_value); cmd.draw_indexed_primitives( primitive, args.index_count, index_buffer + args.first_index, args.instance_count, args.base_vertex, args.first_instance); }