Rework vulkan optimizer #1658

bartosz-muszarski-arm · 2024-08-08T13:34:47Z

Current implementation does not allow to implement different kind of optimizations in a modular and scalable way.
This commit established base infrastructure allowing implementation of modules that can process vulkan calls and apply modifications to them.
Gfxrecon-optimize for vulkan uses 2 pass approach:

first pass to generate optimization data
second pass to generate modified trace
Modifications are performed in predefined order for each call separately.
Each modifier is given a pointer to same modifiable parameter buffer, a method to insert new calls before current call and a flag to request removal of current call from trace.

ci-tester-lunarg · 2024-08-08T13:34:50Z

CI gfxreconstruct build queued with queue ID 231913.

ci-tester-lunarg · 2024-08-08T18:13:22Z

CI gfxreconstruct build # 4572 running.

ci-tester-lunarg · 2024-08-08T19:25:01Z

CI gfxreconstruct build # 4572 passed.

ci-tester-lunarg · 2024-08-27T10:27:30Z

CI gfxreconstruct build queued with queue ID 243984.

ci-tester-lunarg · 2024-08-27T10:30:01Z

CI gfxreconstruct build # 4695 running.

ci-tester-lunarg · 2024-08-27T11:02:52Z

CI gfxreconstruct build # 4695 passed.

ci-tester-lunarg · 2024-08-27T15:02:38Z

CI gfxreconstruct build queued with queue ID 244265.

ci-tester-lunarg · 2024-08-27T15:17:34Z

CI gfxreconstruct build # 4699 running.

ci-tester-lunarg · 2024-08-27T15:53:00Z

CI gfxreconstruct build # 4699 passed.

ci-tester-lunarg · 2025-01-09T14:10:16Z

CI gfxreconstruct build queued with queue ID 341189.

ci-tester-lunarg · 2025-01-09T14:17:24Z

CI gfxreconstruct build # 5747 running.

ci-tester-lunarg · 2025-01-09T14:51:24Z

CI gfxreconstruct build # 5747 failed.

fabian-lunarg · 2024-09-10T09:04:17Z

framework/decode/vulkan_example_modifier.h

+// To test this example you need any trace that performs an acceleration structure copy with
+// VK_COPY_ACCELERATION_STRUCTURE_MODE_COMPACT_KHR mode, for example:
+// https://github.com/ARM-software/tracetooltests/blob/main/src/vulkan_as_3.cpp
+class VulkanExampleModifier : public util::VulkanModifierBase


why is the class called VulkanExampleModifier?

meaning it's not used as test-case, but in production code

it actually claims it 'Adds vkSetDebugUtilsObjectNameEXT calls after relevant vkCreateAccelerationStructureKHR'

I think it should have another name to better express what it is/does

It's meant for demonstration only, to showcase how the optimization interface was intended to be used, thus the name. Has no real use other than that, can be removed completely.

I'm looking into moving this class into a test-case. it would serve the purpose to demonstrate usage, but not live in the production-codebase

but then, I see another issue. usage of VulkanExampleModifier has been hard-coded into 'tools/optimize/main.cpp'

tools/optimize/vulkan_file_optimizer.cpp

fabian-lunarg · 2025-01-16T11:36:20Z

I'm currently investigating the failed CI-run from last week. it looks many tests related to dump_resources feature fail, due to different json out. this might require adjusting our ground-truth, let me come back for this
was a red herring

ci-tester-lunarg · 2025-01-20T10:18:17Z

CI gfxreconstruct build queued with queue ID 350987.

ci-tester-lunarg · 2025-01-20T11:58:31Z

CI gfxreconstruct build # 5879 running.

ci-tester-lunarg · 2025-01-20T12:27:39Z

CI gfxreconstruct build # 5879 failed.

ci-tester-lunarg · 2025-01-22T09:39:49Z

CI gfxreconstruct build queued with queue ID 352651.

ci-tester-lunarg · 2025-01-22T09:40:02Z

CI gfxreconstruct build # 5894 running.

ci-tester-lunarg · 2025-01-22T11:16:14Z

CI gfxreconstruct build # 5894 passed.

ci-tester-lunarg · 2025-01-29T12:09:34Z

CI gfxreconstruct build queued with queue ID 359365.

ci-tester-lunarg · 2025-01-29T12:31:00Z

CI gfxreconstruct build # 5964 running.

ci-tester-lunarg · 2025-01-29T13:04:15Z

CI gfxreconstruct build # 5964 passed.

ci-tester-lunarg · 2025-01-29T16:59:03Z

CI gfxreconstruct build queued with queue ID 359680.

ci-tester-lunarg · 2025-01-29T17:06:06Z

CI gfxreconstruct build # 5966 running.

ci-tester-lunarg · 2025-01-29T17:40:07Z

CI gfxreconstruct build # 5966 passed.

Current implementation does not allow to implement different kind of optimizations in a modular and scalable way. This commit established base infrastructure allowing imlementation of modules that can process vulkan calls and apply modifications to them. Gfxrecon-optimize uses 2 pass approach - first pass to generate optimization data and a second pass to generate modified trace. Modifications are performed in predefined order for each call separately. Each modifier is given a pointer to same modifiable parameter buffer, a method to insert new calls before current call and a flag to request removal of current call from trace. Change-Id: Ic1f396d70cd58eb0d416273bf0fec4e7c9c02f58

Change-Id: Ibf8290f90a73aaa7aac0442a16f0bad70bc3e3dc

- add required 'override' statements - tested on macOS14.5, clang-1500.3.9.4

- DRY: break out a shared utility, avoiding code-duplication

ci-tester-lunarg · 2025-02-05T10:34:58Z

CI gfxreconstruct build queued with queue ID 365144.

ci-tester-lunarg · 2025-02-05T10:40:04Z

CI gfxreconstruct build # 6014 running.

ci-tester-lunarg · 2025-02-05T11:08:57Z

CI gfxreconstruct build # 6014 passed.

fabian-lunarg · 2025-02-05T11:33:55Z

tools/optimize/main.cpp

-        else if ((file_processor.GetCurrentFrameNumber() > 0) &&
-                 (file_processor.GetErrorState() == gfxrecon::decode::FileProcessor::kErrorNone))
+
+        // TODO: resref_consumer should derive from VulkanModifierBase and be handled like other


we should add a clear condition to the TODO (or remove it). when will it be removed? like a specific date or best a referenced issue/PR-#

Will remove it, this no longer applies

fabian-lunarg · 2025-02-05T11:39:55Z

tools/optimize/main.cpp


    gfxrecon::decode::FileProcessor file_processor;
    if (file_processor.Initialize(input_filename))
    {
        gfxrecon::decode::VulkanDecoder                    decoder;
        gfxrecon::decode::VulkanReferencedResourceConsumer resref_consumer;
+        auto example_modifier = std::make_unique<gfxrecon::decode::VulkanExampleModifier>();


usage of VulkanExampleModifier is hard-coded here and should not be, since it's a production codebase/tool.

when we remove that addition I do not see a clear way how modifiers can be added.
maybe I'm missing something, but what is the concept here, like e.g. will there be cmd-line parameters to gfxrecon-optimize to add/select modifiers?

In this iteration modifiers would be added similarly to this example, hardcoded without optimizer parameters.
There are kinds of modifications we typically need for any vulkan trace - removing unused blocks, unused extensions/features, corrupted blocks at the end of the file - to name a few.
Then we have modifications that user may need - like handling portable raytracing via postprocessing, general device address processing, possibly bindless, etc. - that would need to be parametrized.
Lastly there are very specific modifications that require user input, like remove calls by thread or device id, those will have to be parametrized and handled separately.
I'd like to keep things modular and scalable, have a separate class for every kind of optimization we might need to do in the future. While the user could choose the modifications he needs and their order, I think as this solution scales we'll need to group them instead, creating predefined optimization profile and have the user target that via parameter.
That would be the general idea, but the goal of this PR is to establish a base scalable architecture that we can build on top of in next PRs.

fabian-lunarg · 2025-02-06T12:41:25Z

tools/optimize/vulkan_file_optimizer.cpp

+    size_t              parameter_buffer_size = static_cast<size_t>(block_header.size) - sizeof(call_id);
+    uint64_t            uncompressed_size     = 0;
+    decode::ApiCallInfo call_info{ GetCurrentBlockIndex() };
+    bool                success = ReadBytes(&call_info.thread_id, sizeof(call_info.thread_id));


success is not checked after reading here

fabian-lunarg · 2025-02-06T12:41:59Z

tools/optimize/vulkan_file_optimizer.cpp

+
+bool VulkanFileOptimizer::ProcessFunctionCall(const format::BlockHeader& block_header, format::ApiCallId call_id)
+{
+    // Regular funciton call processing/compression handling - duplicated in multiple places, TODO: extract, reuse


typo in comment and unclear TODO. either remove or add more information?

fabian-lunarg · 2025-02-06T12:53:09Z

tools/optimize/vulkan_file_optimizer.h

+        std::vector<std::unique_ptr<util::VulkanModifierBase>> modifiers;
+    };
+
+    explicit VulkanFileOptimizer(VulkanOptimizationData* optimization_data) :


VulkanOptimizationData* is passed as raw-pointer, which is then stored as member.
better pass by value and move into place?

You mean transfer the ownership of optimization_data inside VulkanFileOptimizer? I don't mind, but also don't see the benefit. The purpose here was to avoid copies that may become costly as the optimization data grows. Caller maintains ownership, consumer gets the input as raw ptr.

std::unique_ptr<gfxrecon::VulkanFileOptimizer::VulkanOptimizationData> GetVulkanOptimizationData(...)

returns a unique-pointer.
so yes, I think it would be better to transfer ownership of that instead of using a raw-pointer.
there would be no copy and the benefit would be to avoid a potentially dangling pointer in optimization_data_

Returning unique ptr here does not generate a copy

bartosz-muszarski-arm marked this pull request as ready for review August 27, 2024 15:22

fabian-lunarg self-assigned this Sep 9, 2024

fabian-lunarg mentioned this pull request Sep 9, 2024

Fix compilation on macOS bartosz-muszarski-arm/gfxreconstruct#2

Merged

fabian-lunarg requested review from fabian-lunarg and MarkY-LunarG September 10, 2024 09:12

bradgrantham-lunarg added P2 A high-priority code maintenance issue or a functional problem that is recoverable or not a crash. optimize Issue with optimize vulkan labels Oct 10, 2024

fabian-lunarg reviewed Jan 9, 2025

View reviewed changes

fabian-lunarg force-pushed the vulkan-optimizer branch from dfc7788 to 84bf02f Compare January 22, 2025 09:39

fabian-lunarg force-pushed the vulkan-optimizer branch from 84bf02f to 814527f Compare January 29, 2025 12:09

bartosz-muszarski-arm and others added 4 commits February 5, 2025 11:25

Enable adding new calls in post, add example

325eca7

Change-Id: Ibf8290f90a73aaa7aac0442a16f0bad70bc3e3dc

Fix compilation on macOS

5ace53b

- add required 'override' statements - tested on macOS14.5, clang-1500.3.9.4

Consolidate duplication of WriteFunctionCall routines

ef32975

- DRY: break out a shared utility, avoiding code-duplication

fabian-lunarg force-pushed the vulkan-optimizer branch from be1bd25 to ef32975 Compare February 5, 2025 10:34

fabian-lunarg reviewed Feb 5, 2025

View reviewed changes

fabian-lunarg reviewed Feb 6, 2025

View reviewed changes

Rework vulkan optimizer #1658

Are you sure you want to change the base?

Rework vulkan optimizer #1658

Conversation

bartosz-muszarski-arm commented Aug 8, 2024

ci-tester-lunarg commented Aug 8, 2024

ci-tester-lunarg commented Aug 8, 2024

ci-tester-lunarg commented Aug 8, 2024

ci-tester-lunarg commented Aug 27, 2024

ci-tester-lunarg commented Aug 27, 2024

ci-tester-lunarg commented Aug 27, 2024

ci-tester-lunarg commented Aug 27, 2024

ci-tester-lunarg commented Aug 27, 2024

ci-tester-lunarg commented Aug 27, 2024

ci-tester-lunarg commented Jan 9, 2025

ci-tester-lunarg commented Jan 9, 2025

ci-tester-lunarg commented Jan 9, 2025

Choose a reason for hiding this comment

bartosz-muszarski-arm Jan 23, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

fabian-lunarg commented Jan 16, 2025 • edited Loading

ci-tester-lunarg commented Jan 20, 2025

ci-tester-lunarg commented Jan 20, 2025

ci-tester-lunarg commented Jan 20, 2025

ci-tester-lunarg commented Jan 22, 2025

ci-tester-lunarg commented Jan 22, 2025

ci-tester-lunarg commented Jan 22, 2025

ci-tester-lunarg commented Jan 29, 2025

ci-tester-lunarg commented Jan 29, 2025

ci-tester-lunarg commented Jan 29, 2025

ci-tester-lunarg commented Jan 29, 2025

ci-tester-lunarg commented Jan 29, 2025

ci-tester-lunarg commented Jan 29, 2025

ci-tester-lunarg commented Feb 5, 2025

ci-tester-lunarg commented Feb 5, 2025

ci-tester-lunarg commented Feb 5, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

bartosz-muszarski-arm Jan 23, 2025 •

edited

Loading

fabian-lunarg commented Jan 16, 2025 •

edited

Loading