This repository has been archived by the owner on Apr 8, 2024. It is now read-only.
Fixed lookup buffer size calculation for Vulkan implementation. #99
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
For me the behaviour is still the same on Linux at least. There is, however recently found (it is working with and without that fix) ONE working Vulkan enabled GPU |
poszu
suggested changes
Apr 24, 2023
There was a problem hiding this comment.
I ran the tests with build/test/gpu-setup-test --test --logs -N 8192 on llvmpipe and it fails:
Test LEAFS: Label size: 128, count 131072, buffer 2.0M
[2023-04-24 11:10:43] Vulkan library LOADED.
[2023-04-24 11:10:43] initCpu() finished.
CPU: 131072 hashes, 609 h/s, status: 0
WARNING: lavapipe is not a conformant vulkan implementation, testing use only.
[2023-04-24 11:14:19] 92% Max Allocation: 1971322880
[2023-04-24 11:14:19] GPU 0: selecting buffer_size of 1880
[2023-04-24 11:14:19] GPU 0: setting thread_concurrency to 7520 based on buffer size 1880 and lookup gap 4
[2023-04-24 11:14:19] 64:128 34362 -> 126892
[2023-04-24 11:14:19] SPIR-V program 64:128 126892 bytes
[2023-04-24 11:14:19] initVulkan() finished. Found llvmpipe (LLVM 12.0.0, 256 bits)
llvmpipe (LLVM 12.0.0, 256 bits): 131072 hashes, 14883 h/s
ZEROS result
WRONG result for label size 128 from provider 0 [llvmpipe (LLVM 12.0.0, 256 bits)]
| cgpu->lookup_gap = 4; | ||
|
|
||
| unsigned int bsize = 1024; | ||
| size_t ipt = (bsize / cgpu->lookup_gap + (bsize % cgpu->lookup_gap > 0)); | ||
| size_t ipt = scrypt_mem / cgpu->lookup_gap; |
There was a problem hiding this comment.
Could you please document and explain in comments the logic behind calculations of buffers/memory sizes in this function and what variables mean (for example, what "ipt" stands for)?
There was a problem hiding this comment.
Scrypt memory size in bytes scrypt_mem = 128 * r * N
Lookup gap reduces memory usage but increases computational complexity. With lookup gap memory usage is
ipt = scrypt_mem / lookup_gap
max concurrent threads = allocated_buffer_size / ipt
src/vulkan/driver-vulkan.cpp
Outdated
|
|
||
| if (!cgpu->buffer_size) { | ||
| unsigned int base_alloc = (int)(cgpu->gpu_max_alloc * 88 / 100 / 1024 / 1024 / 8) * 8 * 1024 * 1024; | ||
| cgpu->thread_concurrency = (uint32_t)(base_alloc / scrypt_mem / ipt); | ||
| unsigned int base_alloc = (int)(cgpu->gpu_max_alloc * 92 / 100 / 1024 / 1024 / 8) * 8 * 1024 * 1024; |
There was a problem hiding this comment.
A value of 88% is generally accepted, but this is a legacy of cards with 4G memory or less. 92% is performance improvement and is acceptable as long as there are no memory allocation errors. In fact, for cards with more than 4G memory, a value of 100% is acceptable.
That's from |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixed lookup buffer size calculation for Vulkan implementation.