TBDR下msaa 在metal vulkan和ogles的解决方案

https://developer.arm.com/solutions/graphics/developer-guides/understanding-render-passes/multi-sample-anti-aliasing

msaa在local mem上做很省但是带宽不处理多出8x 对于4xmsaa

计算如下

处理前

python
bytesPerFrame4x = 2560 * 1440 * 4 * 4
bytesPerFrame1x = 2560 * 1440 * 4 * 1

# Additional 4x bandwidth is doubled because the additional samples
# are written by one pass and then re-read to resolve the final color
bytesPerFrame = ((bytesPerFrame4x * 2) + bytesPerFrame1x)
bytesPerSecond = bytesPerFrame * 60
               = 7.9 GB/s

处理后


python
bytesPerFrame1x = 2560 * 1440 * 4 * 1

# All additional 4x bandwidth is kept entirely inside the tile memory
bytesPerSecond = bytesPerFrame1x * 60
               = 884 MB/s

处理的方法就是load store action选一共1x那种

vulkan和metal都可以这样处理之前有发blog

https://www.cnblogs.com/minggoddess/p/10950349.html

vulkan还要用

using VK_MEMORY_PROPERTY_LAZILY_ALLOCATED_BIT and constructing the VkImage with VK_IMAGE_USAGE_TRANSIENT_ATTACHMENT_BIT.做memoryless

metal直接设置memoryless

对于

ogles

用如下扩展

[EXT_multisampled_render_to_texture][EXT_msaa] extension

https://www.khronos.org/registry/OpenGL/extensions/EXT/EXT_multisampled_render_to_texture.txt

   This extension introduces functionality to perform multisampled 
    rendering to a color renderable texture, without requiring an 
    explicit resolve of multisample data.


Some GPU architectures - such as tile-based renderers - are
    capable of performing multisampled rendering by storing
    multisample data in internal high-speed memory and downsampling the
    data when writing out to external memory after rendering has
    finished. Since per-sample data is never written out to external
    memory, this approach saves bandwidth and storage space. In this
    case multisample data gets discarded, however this is acceptable
    in most cases.

自动resolve不用显示resovle了在tile上还可以省3x store 和footprint

FramebufferTexture2DMultisampleEXT

RenderbufferStorageMultisampleEX

还有depthstencil的

所有tbdr下这套解决方案在ogles 要用extension unity有实现之后会验下数据

memoryless其实是个metal和vulkan才有的概念

msaa情况下完全对应上面这个扩展1x store ok了

unity里面根据rendertexture的descriptor

antiAliasing会自动开这个扩展相关代码 glRenderbufferStorageMultisample

======================

unity里面对msaa自动开了 glRenderbufferStorageMultisample

这需要capabilities HasMultisample

ogles3 或者

HasMultiSampleAutoResolve 这个capa对应以下两个扩展

kGL_EXT_multisampled_render_to_texture

kGL_IMG_multisampled_render_to_texture

force-clamped是啥

kGL_EXT_multisampled_render_to_texture

glRenderbufferStorageMultisampleEXT

glFramebufferTexture2DMultisampleEXT

Mali用的这组

kGL_IMG_multisampled_render_to_texture

glRenderbufferStorageMultisampleIMG

glFramebufferTexture2DMultisampleIMG

kGL_APPLE_framebuffer_multisample

glRenderbufferStorageMultisampleAPPLE

glResolveMultisampleFramebufferAPPLE

有metal这个不用管了

kGL_NV_framebuffer_multisample

kGL_NV_framebuffer_blit

glRenderbufferStorageMultisampleNV

-------

----------------------------------------------------

下面就是profiler的数据了这部分好诡异好难理解

开了msaa

read memory和write mem 大幅下降如果只是shaderbusy也说不通这个降幅

--snapdragon845

和后处理有关应该和msaa没什么关系