diff --git a/changes/render3d-small-window.md b/changes/render3d-small-window.md new file mode 100644 index 00000000..c412dccd --- /dev/null +++ b/changes/render3d-small-window.md @@ -0,0 +1,7 @@ +bump: patch +type: feature +**`r3d_small`: a renderer sized for a small UI window.** Set before `r3d_open`, it gives the frame's +uniform ring 2 x 4 MB (was 2 x 64), memory blocks of 8 MB with anything over 4 MB on its own (was 64 and +16) and descriptor pools an eighth of a world's. Maroon Lake's launcher window, which draws a picture and +text and never becomes the game, went from 817 MB to 446 MB with it and its own trims; the rest is the +font atlas, the backdrop, Metal's command-buffer storage while it draws, and the program's states. diff --git a/packages/ludic.render3d/env.ludic b/packages/ludic.render3d/env.ludic index 046ff6b1..be8e2417 100644 --- a/packages/ludic.render3d/env.ludic +++ b/packages/ludic.render3d/env.ludic @@ -587,6 +587,7 @@ export state Render3dState { r3d_time: float = 0.0 r3d_ready: bool = false r3d_plate: bool = false + r3d_small: bool = false # set before r3d_open: a small UI window (a launcher) - the ring, the memory blocks and the descriptor pools sized for it r3d_sky_path: string = null r3d_debug: bool = false r3d_debug_shadow: bool = false diff --git a/packages/ludic.render3d/gpu_vk.ludic b/packages/ludic.render3d/gpu_vk.ludic index 21d4c5f7..179f6582 100644 --- a/packages/ludic.render3d/gpu_vk.ludic +++ b/packages/ludic.render3d/gpu_vk.ludic @@ -340,6 +340,16 @@ function gvk_alloc(render3d_st: mut Render3dState, req: bytes, want: int) -> lon const GVK_BLOCK_DEVICE: int = 67108864 const GVK_BLOCK_HOST: int = 67108864 const GVK_OWN_OVER: int = 16777216 +# a small window's (r3d_small): a UI's few sheets, not a world's hundreds of textures +const GVK_BLOCK_SMALL: int = 8388608 +function gvk_block_bytes(render3d_st: Render3dState) -> int { + if render3d_st.r3d_small { return GVK_BLOCK_SMALL } + return GVK_BLOCK_DEVICE +} +function gvk_own_over(render3d_st: Render3dState) -> int { + if render3d_st.r3d_small { return GVK_BLOCK_SMALL / 2 } + return GVK_OWN_OVER +} @alloc_ok("made once per resource and kept for its life (a texture, program, sampler, view, layout or memory block is created when first asked for)") function gvk_mem_init(render3d_st: mut Render3dState) -> void { @@ -414,7 +424,7 @@ function gvk_mem_new(render3d_st: mut Render3dState, req: bytes, want: int, imag } let mp = render3d_st.gvk_map_slot mp[0] = null - if size <= GVK_OWN_OVER { + if size <= gvk_own_over(render3d_st) { var kind = t * 2 if image { kind += 1 } var i = 0 @@ -434,8 +444,8 @@ function gvk_mem_new(render3d_st: mut Render3dState, req: bytes, want: int, imag i += 1 } if blk < 0 { - var bsize = GVK_BLOCK_DEVICE - if host { bsize = GVK_BLOCK_HOST } + var bsize = gvk_block_bytes(render3d_st) + if host and not render3d_st.r3d_small { bsize = GVK_BLOCK_HOST } let mem = gvk_mem_raw(render3d_st, t, bsize, host, mp) if mem != 0 { # into the record of a block given back, when there is one: a buffer made again every few diff --git a/packages/ludic.render3d/gpu_vk_draw.ludic b/packages/ludic.render3d/gpu_vk_draw.ludic index 18fdf054..628c51bf 100644 --- a/packages/ludic.render3d/gpu_vk_draw.ludic +++ b/packages/ludic.render3d/gpu_vk_draw.ludic @@ -716,6 +716,11 @@ function gvk_bind_texture(render3d_st: mut Render3dState, p: int, name: string, # its descriptor set comes from a pool that is reset with the frame. Both are rewound at # gvk_frame_reset. const GVK_RING_BYTES: int = 64 * 1024 * 1024 +const GVK_RING_SMALL: int = 4 * 1024 * 1024 # a UI's frame: a few hundred small blocks +function gvk_ring_bytes(render3d_st: Render3dState) -> int { + if render3d_st.r3d_small { return GVK_RING_SMALL } + return GVK_RING_BYTES +} @alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play") function gvk_frame_init(render3d_st: mut Render3dState) -> bool { @@ -732,18 +737,18 @@ function gvk_frame_init(render3d_st: mut Render3dState) -> bool { if (counts & VK_SAMPLE_COUNT_4_BIT) != 0 { render3d_st.gvk_msaa_max = 4 } render3d_st.gvk_ring_buf = gvk_buf_new(render3d_st) # one half per frame slot: the cached sets bind this one buffer, and each draw's offset says the half - if not gvk_buf_reserve(render3d_st, render3d_st.gvk_ring_buf, GVK_RING_BYTES * 2) { return false } + if not gvk_buf_reserve(render3d_st, render3d_st.gvk_ring_buf, gvk_ring_bytes(render3d_st) * 2) { return false } let sizes = bytes(VkDescriptorPoolSize_sizeof * 3) Vk.put_i32(sizes, VkDescriptorPoolSize_type, VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER) - Vk.put_i32(sizes, VkDescriptorPoolSize_descriptorCount, 32768) + Vk.put_i32(sizes, VkDescriptorPoolSize_descriptorCount, gvk_pool_n(render3d_st, 32768)) Vk.put_i32(sizes, VkDescriptorPoolSize_sizeof + VkDescriptorPoolSize_type, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER) - Vk.put_i32(sizes, VkDescriptorPoolSize_sizeof + VkDescriptorPoolSize_descriptorCount, 131072) + Vk.put_i32(sizes, VkDescriptorPoolSize_sizeof + VkDescriptorPoolSize_descriptorCount, gvk_pool_n(render3d_st, 131072)) Vk.put_i32(sizes, VkDescriptorPoolSize_sizeof * 2 + VkDescriptorPoolSize_type, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER) - Vk.put_i32(sizes, VkDescriptorPoolSize_sizeof * 2 + VkDescriptorPoolSize_descriptorCount, 16384) + Vk.put_i32(sizes, VkDescriptorPoolSize_sizeof * 2 + VkDescriptorPoolSize_descriptorCount, gvk_pool_n(render3d_st, 16384)) let dpci = bytes(VkDescriptorPoolCreateInfo_sizeof) Vk.zero(dpci, VkDescriptorPoolCreateInfo_sizeof) Vk.put_i32(dpci, VkDescriptorPoolCreateInfo_sType, VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO) - Vk.put_i32(dpci, VkDescriptorPoolCreateInfo_maxSets, 16384) + Vk.put_i32(dpci, VkDescriptorPoolCreateInfo_maxSets, gvk_pool_n(render3d_st, 16384)) Vk.put_i32(dpci, VkDescriptorPoolCreateInfo_poolSizeCount, 3) Vk.put_ptr(dpci, VkDescriptorPoolCreateInfo_pPoolSizes, sizes) let out = bytes(8) @@ -770,8 +775,8 @@ function gvk_frame_reset(render3d_st: mut Render3dState) -> void { # draw still in use makes its set again the first time it is drawn if render3d_st.gvk_sc_dead >= GVK_SC_DEAD_MAX { gvk_kpool_reset(render3d_st) } let slot = render3d_st.gvk_frame_no & 1 - render3d_st.gvk_ring_off = slot * GVK_RING_BYTES - render3d_st.gvk_ring_end = (slot + 1) * GVK_RING_BYTES + render3d_st.gvk_ring_off = slot * gvk_ring_bytes(render3d_st) + render3d_st.gvk_ring_end = (slot + 1) * gvk_ring_bytes(render3d_st) render3d_st.gvk_dpool = render3d_st.gvk_dpools[slot] Vk.reset_descriptor_pool(render3d_st.gvk_dev, render3d_st.gvk_dpool, 0) } @@ -795,18 +800,23 @@ function gvk_ring_put(render3d_st: mut Render3dState, blk: pointer, n: int) -> i # sampler, so a set is never matched against a texture it no longer describes. const GVK_KPOOL_SETS: int = 8192 const GVK_SC_DEAD_MAX: int = 256 # textures released before the kept sets start over +# a descriptor pool's counts: a small window's (r3d_small) an eighth of a world's +function gvk_pool_n(render3d_st: Render3dState, n: int) -> int { + if render3d_st.r3d_small { return n / 8 } + return n +} @alloc_ok("start-up: the device, its tables, the programs, the passes and the world's first textures are made once, before play") function gvk_kpool_make(render3d_st: mut Render3dState) -> bool { let sizes = bytes(VkDescriptorPoolSize_sizeof * 2) Vk.put_i32(sizes, VkDescriptorPoolSize_type, VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC) - Vk.put_i32(sizes, VkDescriptorPoolSize_descriptorCount, GVK_KPOOL_SETS * 2) + Vk.put_i32(sizes, VkDescriptorPoolSize_descriptorCount, gvk_pool_n(render3d_st, GVK_KPOOL_SETS) * 2) Vk.put_i32(sizes, VkDescriptorPoolSize_sizeof + VkDescriptorPoolSize_type, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER) - Vk.put_i32(sizes, VkDescriptorPoolSize_sizeof + VkDescriptorPoolSize_descriptorCount, GVK_KPOOL_SETS * 8) + Vk.put_i32(sizes, VkDescriptorPoolSize_sizeof + VkDescriptorPoolSize_descriptorCount, gvk_pool_n(render3d_st, GVK_KPOOL_SETS) * 8) let dpci = bytes(VkDescriptorPoolCreateInfo_sizeof) Vk.zero(dpci, VkDescriptorPoolCreateInfo_sizeof) Vk.put_i32(dpci, VkDescriptorPoolCreateInfo_sType, VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO) - Vk.put_i32(dpci, VkDescriptorPoolCreateInfo_maxSets, GVK_KPOOL_SETS) + Vk.put_i32(dpci, VkDescriptorPoolCreateInfo_maxSets, gvk_pool_n(render3d_st, GVK_KPOOL_SETS)) Vk.put_i32(dpci, VkDescriptorPoolCreateInfo_poolSizeCount, 2) Vk.put_ptr(dpci, VkDescriptorPoolCreateInfo_pPoolSizes, sizes) let out = bytes(8)