workshopsoft/ludic
2
0
Fork 0
Code Issues Pull requests Projects Releases 26 Packages Wiki Activity Actions

render3d(vulkan): MSAA - multisampled renderbuffers, resolved in a draw-less pass

Browse source 
gpu_rb_storage makes the image with the samples asked for; a pass takes its attachments' count and
every pipeline in it matches; a blit from a multisampled source resolves on the end of an empty
dynamic-rendering pass (colour averaged, depth from sample zero - vkCmdResolveImage cannot do depth).
gpu_msaa_max reads the device's colour-and-depth sample limits, and post_set_msaa clamps to it, so
the Anti-aliasing setting is live on Vulkan. The pipeline cache's pass key no longer packs samples
into three bits.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
Orkun ÇAKILKAYA 2026-09-15 16:21:52 +03:00
parent 97d75e1d5e
commit bda53f759e
4 changed files with 74 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

52
packages/ludic.render3d/gpu_vk_draw.ludic
View file

@ -654,6 +654,7 @@ var gvk_ring_off: int = 0
var gvk_ring_align: int = 256
var gvk_dpool: long = 0
var gvk_white: int = 0 # a 1x1 white texture for a sampler nothing was bound to
var gvk_msaa_max: int = 0 # samples the scene may use (gvk_frame_init reads the device's limits)
function gvk_frame_init() -> bool {
let props = bytes(VkPhysicalDeviceProperties_sizeof)
@ -661,6 +662,12 @@ function gvk_frame_init() -> bool {
let al = Vk.get_i64(props, VkPhysicalDeviceProperties_limits + VkPhysicalDeviceLimits_minUniformBufferOffsetAlignment)
gvk_ring_align = Text.to_int(string(al))
if gvk_ring_align < 16 { gvk_ring_align = 16 }
# MSAA: the most samples (up to the 4 the scene asks for) both a colour and a depth target can take
let lim = VkPhysicalDeviceProperties_limits
let counts = Vk.get_i32(props, lim + VkPhysicalDeviceLimits_framebufferColorSampleCounts) & Vk.get_i32(props, lim + VkPhysicalDeviceLimits_framebufferDepthSampleCounts)
gvk_msaa_max = 1
if (counts & VK_SAMPLE_COUNT_2_BIT) != 0 { gvk_msaa_max = 2 }
if (counts & VK_SAMPLE_COUNT_4_BIT) != 0 { gvk_msaa_max = 4 }
gvk_ring_buf = gvk_buf_new()
if not gvk_buf_reserve(gvk_ring_buf, GVK_RING_BYTES) { return false }
let sizes = bytes(VkDescriptorPoolSize_sizeof * 3)
@ -921,6 +928,7 @@ var gvk_pass_h: int = 0
var gvk_pass_ncolor: int = 0
var gvk_pass_cfmt: int = 0
var gvk_pass_dfmt: int = 0
var gvk_pass_samples: int = 1 # the attachments' sample count, which every pipeline in the pass must match
var gvk_pass_col: words = null # the pass's colour attachments: texture, layer + 1
var gvk_pass_dep: words = null # its depth attachment: texture, layer + 1
var gvk_clear_bits: int = 0 # GL_COLOR_BUFFER_BIT / GL_DEPTH_BUFFER_BIT waiting for the pass
@ -1043,6 +1051,7 @@ function gvk_pass_begin(rec: words, rec_o: int) -> void {
gvk_pass_h = 0
gvk_pass_cfmt = VK_FORMAT_UNDEFINED
gvk_pass_dfmt = VK_FORMAT_UNDEFINED
gvk_pass_samples = 1
for c in 0 .. gvk_pass_ncolor {
let tex = gvk_pass_col[c * 2]
let layer1 = gvk_pass_col[c * 2 + 1]
@ -1056,6 +1065,7 @@ function gvk_pass_begin(rec: words, rec_o: int) -> void {
for k in 0 .. 4 { Vk.put_i32(catt, c * aw + VkRenderingAttachmentInfo_clearValue + k * 4, gvk_clear_rgba[k]) }
} else { Vk.put_i32(catt, c * aw + VkRenderingAttachmentInfo_loadOp, VK_ATTACHMENT_LOAD_OP_LOAD) }
gvk_pass_cfmt = gvk_tex_vkfmt[tex]
gvk_pass_samples = gvk_tex_samples_of(tex)
if gvk_pass_w == 0 { gvk_pass_w = gvk_tex_w(tex); gvk_pass_h = gvk_tex_h(tex) }
}
let datt = bytes(aw)
@ -1072,6 +1082,7 @@ function gvk_pass_begin(rec: words, rec_o: int) -> void {
Vk.put_i32(datt, VkRenderingAttachmentInfo_clearValue, 0x3F800000)
} else { Vk.put_i32(datt, VkRenderingAttachmentInfo_loadOp, VK_ATTACHMENT_LOAD_OP_LOAD) }
gvk_pass_dfmt = VK_FORMAT_D32_SFLOAT
gvk_pass_samples = gvk_tex_samples_of(dtex)
if gvk_pass_w == 0 { gvk_pass_w = gvk_tex_w(dtex); gvk_pass_h = gvk_tex_h(dtex) }
}
gvk_clear_bits = 0
@ -1168,7 +1179,7 @@ function gvk_draw(p: int, m: Mesh, st: GvkState, first: int, count: int, instanc
if prof { t0 = gl_now_us() }
gvk_pass_begin(rec, rec_o)
let cb = gvk_cb
let samples = 1
let samples = gvk_pass_samples
let pipe = gvk_pipeline_fast(p, m, st, gvk_pass_ncolor, gvk_pass_cfmt, gvk_pass_dfmt, samples)
if pipe == 0 {
# Say so, once per program: a driver that refuses a pipeline other drivers accept (MoltenVK
@ -1431,8 +1442,45 @@ function gvk_fb_att(fb: int, depth: bool) -> int {
if depth { return gpu_fb[o + 2] }
return gpu_fb[o]
}
function gvk_tex_samples_of(tex: int) -> int {
if tex <= 0 or gvk_tex_samples == null or tex >= len(gvk_tex_samples) or gvk_tex_samples[tex] < 1 { return 1 }
return gvk_tex_samples[tex]
}
# A multisampled image into a single-sampled one: a pass that draws nothing and resolves on its end -
# colour averaged, depth from sample zero. vkCmdResolveImage cannot resolve depth; a pass can.
function gvk_resolve(cb: pointer, src: int, dst: int, depth: bool, w: int, h: int) -> void {
var layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL
var mode = VK_RESOLVE_MODE_AVERAGE_BIT
if depth { layout = VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_OPTIMAL; mode = VK_RESOLVE_MODE_SAMPLE_ZERO_BIT }
gvk_att_barrier(cb, src, 0, depth, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, layout)
gvk_att_barrier(cb, dst, 0, depth, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, layout)
let aw = VkRenderingAttachmentInfo_sizeof
let att = bytes(aw)
Vk.zero(att, aw)
Vk.put_i32(att, VkRenderingAttachmentInfo_sType, VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO)
Vk.put_i64(att, VkRenderingAttachmentInfo_imageView, gvk_view_of(src, 0))
Vk.put_i32(att, VkRenderingAttachmentInfo_imageLayout, layout)
Vk.put_i32(att, VkRenderingAttachmentInfo_resolveMode, mode)
Vk.put_i64(att, VkRenderingAttachmentInfo_resolveImageView, gvk_view_of(dst, 0))
Vk.put_i32(att, VkRenderingAttachmentInfo_resolveImageLayout, layout)
Vk.put_i32(att, VkRenderingAttachmentInfo_loadOp, VK_ATTACHMENT_LOAD_OP_LOAD)
Vk.put_i32(att, VkRenderingAttachmentInfo_storeOp, VK_ATTACHMENT_STORE_OP_STORE)
let ri = bytes(VkRenderingInfo_sizeof)
Vk.zero(ri, VkRenderingInfo_sizeof)
Vk.put_i32(ri, VkRenderingInfo_sType, VK_STRUCTURE_TYPE_RENDERING_INFO)
Vk.put_i32(ri, VkRenderingInfo_renderArea + VkRect2D_extent + VkExtent2D_width, w)
Vk.put_i32(ri, VkRenderingInfo_renderArea + VkRect2D_extent + VkExtent2D_height, h)
Vk.put_i32(ri, VkRenderingInfo_layerCount, 1)
if depth { Vk.put_ptr(ri, VkRenderingInfo_pDepthAttachment, att) }
else { Vk.put_i32(ri, VkRenderingInfo_colorAttachmentCount, 1); Vk.put_ptr(ri, VkRenderingInfo_pColorAttachments, att) }
Vk.cmd_begin_rendering(cb, ri)
Vk.cmd_end_rendering(cb)
gvk_att_barrier(cb, src, 0, depth, layout, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL)
gvk_att_barrier(cb, dst, 0, depth, layout, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL)
}
function gvk_copy(cb: pointer, src: int, dst: int, depth: bool, w: int, h: int) -> void {
if src <= 0 or dst <= 0 or gvk_tex_image[src] == 0 or gvk_tex_image[dst] == 0 { return }
if gvk_tex_samples_of(src) > 1 and gvk_tex_samples_of(dst) == 1 { gvk_resolve(cb, src, dst, depth, w, h); return }
gvk_barrier(cb, gvk_tex_image[src], depth, 0, 1, gvk_tex_layers[src], VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL)
gvk_barrier(cb, gvk_tex_image[dst], depth, 0, 1, gvk_tex_layers[dst], VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL)
let ic = bytes(VkImageCopy_sizeof)
@ -1737,7 +1785,7 @@ function gvk_pipeline_fast(p: int, m: Mesh, st: GvkState, n_color: int, color_fm
var state = st.depth_test | (st.depth_write << 1) | (st.blend << 2) | (st.cull << 3) | (st.color_write << 4) | (st.a2c << 5) | (st.bias << 6) | (st.wireframe << 7)
state = state | ((st.depth_func & 15) << 8) | ((st.cull_face & 15) << 12) | (gvk_blend_index(st.blend_src) << 16) | (gvk_blend_index(st.blend_dst) << 20)
let bias = st.bias_factor * 31 + st.bias_units
let pass = ((n_color * 1000 + color_fmt) * 1000 + depth_fmt) * 8 + samples
let pass = ((n_color * 1000 + color_fmt) * 1000 + depth_fmt) * 64 + samples
if gvk_pc_prog == null {
gvk_pc_prog = new []int; gvk_pc_layout = new []int; gvk_pc_state = new []int; gvk_pc_bias = new []int
gvk_pc_pass = new []int; gvk_pc_pipe = new []long