GL: Add subpass input emulation #112051

Clément Foucault · 2023-09-06T20:14:30+02:00

Clément Foucault commented

2023-09-06 20:14:30 +02:00

This adds basic emulation of the subpass input feature
of vulkan and to a lower extend Raster Order Group on Metal.

This help test paths that might use this feature in the future
(like shadow rendering) on all platform and or simplify higher
level code for supporting older hardware.

This add clear description to the load/store ops and to the
new GPUAttachementState.

The OpenGL backend will correctly mask un-writable
attachments and will bind as texture readable attachments.

Even if possible by the vulkan standard, the GPU API prohibit
the read and write to the same attachment inside the same
subpass.

In the GL backend, this is implemented using glTextureBarrier
and texelFetch as it is described in the ARB_texture_barrier
extension.
https://registry.khronos.org/OpenGL/extensions/ARB/ARB_texture_barrier.txt

This adds basic emulation of the subpass input feature of vulkan and to a lower extend Raster Order Group on Metal. This help test paths that might use this feature in the future (like shadow rendering) on all platform and or simplify higher level code for supporting older hardware. This add clear description to the load/store ops and to the new `GPUAttachementState`. The OpenGL backend will correctly mask un-writable attachments and will bind as texture readable attachments. Even if possible by the vulkan standard, the GPU API prohibit the read and write to the same attachment inside the same subpass. In the GL backend, this is implemented using `glTextureBarrier` and `texelFetch` as it is described in the ARB_texture_barrier extension. https://registry.khronos.org/OpenGL/extensions/ARB/ARB_texture_barrier.txt

Clément Foucault added 1 commit 2023-09-06 20:14:43 +02:00

5f84b8e767 WIP: GL: Add subpass emulation

Clément Foucault requested review from Jeroen Bakker 2023-09-06 20:16:04 +02:00

Jeroen Bakker reviewed 2023-09-14 14:49:16 +02:00

source/blender/gpu/GPU_common_types.h Outdated

						
				@ -19,2 +19,4 @@

				} eGPULoadOp;

				typedef enum GPUAttachmentLayout {

				  GPU_LAYOUT_GENERAL = 0,

Jeroen Bakker commented

2023-09-14 14:49:16 +02:00

Different name, layout tells more about the order of data on the GPU inside the image. Perhaps GPUAttachmentUsage

fclem marked this conversation as resolved

Clément Foucault added 2 commits 2023-09-28 23:21:33 +02:00

4b5dca3c26 Merge branch 'main' into gl-subpass-emulation

# Conflicts:
#	source/blender/gpu/opengl/gl_shader.cc

54f6a36a31 Remove read/write from same attachment, implement other attachement states

Clément Foucault added 1 commit 2023-09-28 23:55:58 +02:00

662f0c4529 Add framebuffer fetch extension support

Clément Foucault changed title from ~~WIP: GL: Add subpass input emulation~~ to GL: Add subpass input emulation

2023-09-28 23:58:38 +02:00

Clément Foucault added 1 commit 2023-09-29 00:10:23 +02:00

fe79842119 Add documentation

Jeroen Bakker reviewed 2023-09-29 11:51:04 +02:00

source/blender/gpu/GPU_common_types.h Outdated

						
				@ -22,0 +39,4 @@

				 */

				typedef enum eGPUStoreOp {

				  /**

				   * Do not care about the content of the attachment when the render pass starts.

Jeroen Bakker commented

2023-09-29 11:51:04 +02:00

when the render pass ends

`when the render pass ends`

fclem marked this conversation as resolved

Jeroen Bakker reviewed 2023-09-29 11:53:44 +02:00

source/blender/gpu/GPU_common_types.h Outdated

						
				@ -22,0 +64,4 @@

				typedef enum GPUAttachmentState {

				  /** Attachment will not be written during rendering. */

				  GPU_ATTACHEMENT_IGNORE = 0,

				  /** Attachment will be written during geometry passes. This also works with blending. */

Jeroen Bakker commented

2023-09-29 11:53:44 +02:00

geometry passes?

fclem marked this conversation as resolved

Clément Foucault reviewed 2023-09-29 11:59:27 +02:00

source/blender/gpu/opengl/gl_shader.cc

						
				@ -116,0 +238,4 @@

				    case Type::SHORT3:

				    case Type::SHORT4:

				    case Type::VEC3_101010I2:

				      /* TODO(fclem): This could add support for lower bit depth formats. */

Clément Foucault commented

2023-09-29 11:59:14 +02:00

This need to be supported for more versatility. However, this will not work with imageLoad because we need to know the format of the image upfront. So I'll change the workaround to use sampler instead of image.

This need to be supported for more versatility. However, this will not work with `imageLoad` because we need to know the format of the image upfront. So I'll change the workaround to use `sampler` instead of `image`.

fclem marked this conversation as resolved

Jeroen Bakker reviewed 2023-09-29 12:10:30 +02:00

source/blender/gpu/vulkan/vk_framebuffer.hh Outdated

						
				@ -56,2 +56,4 @@

				  void attachment_set_loadstore_op(GPUAttachmentType type, GPULoadStore /*ls*/) override;

				  void subpass_transition(const GPUAttachmentState depth_attachment_state,

				                          Span<GPUAttachmentState> color_attachment_states) override{};

Jeroen Bakker commented

2023-09-29 12:08:31 +02:00


  void subpass_transition(const GPUAttachmentState /*depth_attachment_state*/,
                          Span<GPUAttachmentState> /*color_attachment_states*/) override {
        NOT_YET_IMPLEMENTED
};

``` void subpass_transition(const GPUAttachmentState /*depth_attachment_state*/, Span<GPUAttachmentState> /*color_attachment_states*/) override { NOT_YET_IMPLEMENTED };

fclem marked this conversation as resolved

Jeroen Bakker commented

2023-09-29 12:15:05 +02:00

When run on Intel ARC/Mesa

Note: Google Test filter = GPUOpenGLTest.framebuffer_subpass_input
[==========] Running 1 test from 1 test suite.
[----------] Global test environment set-up.
[----------] 1 test from GPUOpenGLTest
[ RUN      ] GPUOpenGLTest.framebuffer_subpass_input
Failed to start GPU frame capture!
ERROR (gpu.shader):  FragShader: 
      | 
   92 |   in_value = gl_LastFragData[0].x;
      |              ^
      | Error: `gl_LastFragData' undeclared
      | Error: type mismatch

Segmentation fault (core dumped)

Might be that the extension is supported on OpenGL ES, but not on OpenGL...
Perhaps disable the extension for intel GPUs

I validated that it worked using the same Mesa, but with AMD card.

When run on Intel ARC/Mesa ``` Note: Google Test filter = GPUOpenGLTest.framebuffer_subpass_input [==========] Running 1 test from 1 test suite. [----------] Global test environment set-up. [----------] 1 test from GPUOpenGLTest [ RUN ] GPUOpenGLTest.framebuffer_subpass_input Failed to start GPU frame capture! ERROR (gpu.shader): FragShader: | 92 | in_value = gl_LastFragData[0].x; | ^ | Error: `gl_LastFragData' undeclared | Error: type mismatch Segmentation fault (core dumped) ``` Might be that the extension is supported on OpenGL ES, but not on OpenGL... Perhaps disable the extension for intel GPUs I validated that it worked using the same Mesa, but with AMD card.

Clément Foucault added 1 commit 2023-09-29 13:08:35 +02:00

c3b6b34951 Address review comments

Clément Foucault added 1 commit 2023-09-29 14:29:18 +02:00

buildbot/vexp-code-patch-coordinator Build done. Details

018fd6d279 Try opengl es 3.0 syntax

Clément Foucault commented

2023-09-29 16:24:25 +02:00

@buildbot package windows

Clément Foucault commented

2023-09-29 16:27:15 +02:00

@blender-bot package windows

Blender Bot commented

2023-09-29 16:27:18 +02:00

Package build started. Download here when ready.

Package build started. [Download here](https://builder.blender.org/download/patch/PR112051) when ready.

Clément Foucault commented

2023-09-29 23:11:52 +02:00

Note directly related to this task, but noteworthy mention that it is possible to implement the GPU_STOREACTION_DONT_CARE using ARB_invalidate_subdata in OpenGL.

Note directly related to this task, but noteworthy mention that it is possible to implement the `GPU_STOREACTION_DONT_CARE` using `ARB_invalidate_subdata` in OpenGL.

Clément Foucault commented

2023-09-30 00:13:19 +02:00

Reading a bit more and it seems that the current approach without framebuffer fetch is still undefined behavior and that the only option is either to modify the framebuffer or use glTextureBarrier (which is 4.5 core and widely implemented). That's what I'm going to implement.
https://www.khronos.org/opengl/wiki/Memory_Model#Texture_barrier

I think i'll scrap the framebuffer fetch implementation if it still doesnt work on ARC. It isn't supported by the older hardware and will not help much anyway.

Reading a bit more and it seems that the current approach without framebuffer fetch is still undefined behavior and that the only option is either to modify the framebuffer or use `glTextureBarrier` (which is 4.5 core and widely implemented). That's what I'm going to implement. https://www.khronos.org/opengl/wiki/Memory_Model#Texture_barrier I think i'll scrap the framebuffer fetch implementation if it still doesnt work on ARC. It isn't supported by the older hardware and will not help much anyway.

Clément Foucault added 1 commit 2023-09-30 09:08:46 +02:00

23e4ab5c8a Implement new, more correct workarounds with and without glTextureBarrier

Clément Foucault added 1 commit 2023-09-30 09:54:32 +02:00

e614bbb8d5 Missing Uncommited changes

Clément Foucault added 1 commit 2023-10-01 14:45:29 +02:00